2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_addr.h>
47 #include <linux/mlx4/driver.h>
48 #include <linux/mlx4/cmd.h>
53 #define DRV_NAME MLX4_IB_DRV_NAME
54 #define DRV_VERSION "1.0"
55 #define DRV_RELDATE "April 4, 2008"
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
62 int mlx4_ib_sm_guid_assign = 1;
63 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
64 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
66 static const char mlx4_ib_version[] =
67 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
68 DRV_VERSION " (" DRV_RELDATE ")\n";
70 struct update_gid_work {
71 struct work_struct work;
72 union ib_gid gids[128];
73 struct mlx4_ib_dev *dev;
77 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
79 static struct workqueue_struct *wq;
81 static void init_query_mad(struct ib_smp *mad)
83 mad->base_version = 1;
84 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
85 mad->class_version = 1;
86 mad->method = IB_MGMT_METHOD_GET;
89 static union ib_gid zgid;
91 static int mlx4_ib_query_device(struct ib_device *ibdev,
92 struct ib_device_attr *props)
94 struct mlx4_ib_dev *dev = to_mdev(ibdev);
95 struct ib_smp *in_mad = NULL;
96 struct ib_smp *out_mad = NULL;
99 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
100 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
101 if (!in_mad || !out_mad)
104 init_query_mad(in_mad);
105 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
107 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
108 1, NULL, NULL, in_mad, out_mad);
112 memset(props, 0, sizeof *props);
114 props->fw_ver = dev->dev->caps.fw_ver;
115 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
116 IB_DEVICE_PORT_ACTIVE_EVENT |
117 IB_DEVICE_SYS_IMAGE_GUID |
118 IB_DEVICE_RC_RNR_NAK_GEN |
119 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
120 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
121 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
122 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
123 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
124 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
125 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
126 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
127 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
128 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
129 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
130 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
131 props->device_cap_flags |= IB_DEVICE_UD_TSO;
132 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
133 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
134 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
135 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
136 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
137 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
138 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
139 props->device_cap_flags |= IB_DEVICE_XRC;
140 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
141 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
142 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
143 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
144 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
146 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
149 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
151 props->vendor_part_id = dev->dev->pdev->device;
152 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
153 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
155 props->max_mr_size = ~0ull;
156 props->page_size_cap = dev->dev->caps.page_size_cap;
157 props->max_qp = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
158 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
159 props->max_sge = min(dev->dev->caps.max_sq_sg,
160 dev->dev->caps.max_rq_sg);
161 props->max_cq = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
162 props->max_cqe = dev->dev->caps.max_cqes;
163 props->max_mr = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
164 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
165 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
166 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
167 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
168 props->max_srq = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
169 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
170 props->max_srq_sge = dev->dev->caps.max_srq_sge;
171 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
172 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
173 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
174 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
175 props->masked_atomic_cap = props->atomic_cap;
176 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
177 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
178 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
179 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
180 props->max_mcast_grp;
181 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
190 static enum rdma_link_layer
191 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
193 struct mlx4_dev *dev = to_mdev(device)->dev;
195 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
196 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
199 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
200 struct ib_port_attr *props, int netw_view)
202 struct ib_smp *in_mad = NULL;
203 struct ib_smp *out_mad = NULL;
204 int ext_active_speed;
205 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
208 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
209 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
210 if (!in_mad || !out_mad)
213 init_query_mad(in_mad);
214 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
215 in_mad->attr_mod = cpu_to_be32(port);
217 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
218 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
220 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
226 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
227 props->lmc = out_mad->data[34] & 0x7;
228 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
229 props->sm_sl = out_mad->data[36] & 0xf;
230 props->state = out_mad->data[32] & 0xf;
231 props->phys_state = out_mad->data[33] >> 4;
232 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
234 props->gid_tbl_len = out_mad->data[50];
236 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
237 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
238 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
239 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
240 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
241 props->active_width = out_mad->data[31] & 0xf;
242 props->active_speed = out_mad->data[35] >> 4;
243 props->max_mtu = out_mad->data[41] & 0xf;
244 props->active_mtu = out_mad->data[36] >> 4;
245 props->subnet_timeout = out_mad->data[51] & 0x1f;
246 props->max_vl_num = out_mad->data[37] >> 4;
247 props->init_type_reply = out_mad->data[41] >> 4;
249 /* Check if extended speeds (EDR/FDR/...) are supported */
250 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
251 ext_active_speed = out_mad->data[62] >> 4;
253 switch (ext_active_speed) {
255 props->active_speed = IB_SPEED_FDR;
258 props->active_speed = IB_SPEED_EDR;
263 /* If reported active speed is QDR, check if is FDR-10 */
264 if (props->active_speed == IB_SPEED_QDR) {
265 init_query_mad(in_mad);
266 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
267 in_mad->attr_mod = cpu_to_be32(port);
269 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
270 NULL, NULL, in_mad, out_mad);
274 /* Checking LinkSpeedActive for FDR-10 */
275 if (out_mad->data[15] & 0x1)
276 props->active_speed = IB_SPEED_FDR10;
279 /* Avoid wrong speed value returned by FW if the IB link is down. */
280 if (props->state == IB_PORT_DOWN)
281 props->active_speed = IB_SPEED_SDR;
289 static u8 state_to_phys_state(enum ib_port_state state)
291 return state == IB_PORT_ACTIVE ? 5 : 3;
294 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
295 struct ib_port_attr *props, int netw_view)
298 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
299 struct mlx4_ib_iboe *iboe = &mdev->iboe;
300 struct net_device *ndev;
302 struct mlx4_cmd_mailbox *mailbox;
305 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
307 return PTR_ERR(mailbox);
309 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
310 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
315 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
316 IB_WIDTH_4X : IB_WIDTH_1X;
317 props->active_speed = IB_SPEED_QDR;
318 props->port_cap_flags = IB_PORT_CM_SUP;
319 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
320 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
321 props->pkey_tbl_len = 1;
322 props->max_mtu = IB_MTU_4096;
323 props->max_vl_num = 2;
324 props->state = IB_PORT_DOWN;
325 props->phys_state = state_to_phys_state(props->state);
326 props->active_mtu = IB_MTU_256;
327 spin_lock(&iboe->lock);
328 ndev = iboe->netdevs[port - 1];
332 tmp = iboe_get_mtu(ndev->mtu);
333 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
335 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
336 IB_PORT_ACTIVE : IB_PORT_DOWN;
337 props->phys_state = state_to_phys_state(props->state);
339 spin_unlock(&iboe->lock);
341 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
345 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
346 struct ib_port_attr *props, int netw_view)
350 memset(props, 0, sizeof *props);
352 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
353 ib_link_query_port(ibdev, port, props, netw_view) :
354 eth_link_query_port(ibdev, port, props, netw_view);
359 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
360 struct ib_port_attr *props)
362 /* returns host view */
363 return __mlx4_ib_query_port(ibdev, port, props, 0);
366 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
367 union ib_gid *gid, int netw_view)
369 struct ib_smp *in_mad = NULL;
370 struct ib_smp *out_mad = NULL;
372 struct mlx4_ib_dev *dev = to_mdev(ibdev);
374 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
376 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
377 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
378 if (!in_mad || !out_mad)
381 init_query_mad(in_mad);
382 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
383 in_mad->attr_mod = cpu_to_be32(port);
385 if (mlx4_is_mfunc(dev->dev) && netw_view)
386 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
388 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
392 memcpy(gid->raw, out_mad->data + 8, 8);
394 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
396 /* For any index > 0, return the null guid */
403 init_query_mad(in_mad);
404 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
405 in_mad->attr_mod = cpu_to_be32(index / 8);
407 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
408 NULL, NULL, in_mad, out_mad);
412 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
416 memset(gid->raw + 8, 0, 8);
422 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
425 struct mlx4_ib_dev *dev = to_mdev(ibdev);
427 *gid = dev->iboe.gid_table[port - 1][index];
432 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
435 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
436 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
438 return iboe_query_gid(ibdev, port, index, gid);
441 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
442 u16 *pkey, int netw_view)
444 struct ib_smp *in_mad = NULL;
445 struct ib_smp *out_mad = NULL;
446 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
449 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
450 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
451 if (!in_mad || !out_mad)
454 init_query_mad(in_mad);
455 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
456 in_mad->attr_mod = cpu_to_be32(index / 32);
458 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
459 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
461 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
466 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
474 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
476 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
479 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
480 struct ib_device_modify *props)
482 struct mlx4_cmd_mailbox *mailbox;
485 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
488 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
491 if (mlx4_is_slave(to_mdev(ibdev)->dev))
494 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
495 memcpy(ibdev->node_desc, props->node_desc, 64);
496 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
499 * If possible, pass node desc to FW, so it can generate
500 * a 144 trap. If cmd fails, just ignore.
502 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
506 memset(mailbox->buf, 0, 256);
507 memcpy(mailbox->buf, props->node_desc, 64);
508 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
509 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
511 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
516 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
519 struct mlx4_cmd_mailbox *mailbox;
521 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
523 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
525 return PTR_ERR(mailbox);
527 memset(mailbox->buf, 0, 256);
529 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
530 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
531 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
533 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
534 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
537 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
538 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
540 mlx4_free_cmd_mailbox(dev->dev, mailbox);
544 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
545 struct ib_port_modify *props)
547 struct ib_port_attr attr;
551 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
553 err = mlx4_ib_query_port(ibdev, port, &attr);
557 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
558 ~props->clr_port_cap_mask;
560 err = mlx4_SET_PORT(to_mdev(ibdev), port,
561 !!(mask & IB_PORT_RESET_QKEY_CNTR),
565 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
569 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
570 struct ib_udata *udata)
572 struct mlx4_ib_dev *dev = to_mdev(ibdev);
573 struct mlx4_ib_ucontext *context;
574 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
575 struct mlx4_ib_alloc_ucontext_resp resp;
579 return ERR_PTR(-EAGAIN);
581 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
582 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
583 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
584 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
586 resp.dev_caps = dev->dev->caps.userspace_caps;
587 resp.qp_tab_size = dev->dev->caps.num_qps;
588 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
589 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
590 resp.cqe_size = dev->dev->caps.cqe_size;
593 context = kmalloc(sizeof *context, GFP_KERNEL);
595 return ERR_PTR(-ENOMEM);
597 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
603 INIT_LIST_HEAD(&context->db_page_list);
604 mutex_init(&context->db_page_mutex);
606 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
607 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
609 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
612 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
614 return ERR_PTR(-EFAULT);
617 return &context->ibucontext;
620 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
622 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
624 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
630 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
632 struct mlx4_ib_dev *dev = to_mdev(context->device);
634 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
637 if (vma->vm_pgoff == 0) {
638 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
640 if (io_remap_pfn_range(vma, vma->vm_start,
641 to_mucontext(context)->uar.pfn,
642 PAGE_SIZE, vma->vm_page_prot))
644 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
645 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
647 if (io_remap_pfn_range(vma, vma->vm_start,
648 to_mucontext(context)->uar.pfn +
649 dev->dev->caps.num_uars,
650 PAGE_SIZE, vma->vm_page_prot))
658 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
659 struct ib_ucontext *context,
660 struct ib_udata *udata)
662 struct mlx4_ib_pd *pd;
665 pd = kmalloc(sizeof *pd, GFP_KERNEL);
667 return ERR_PTR(-ENOMEM);
669 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
676 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
677 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
679 return ERR_PTR(-EFAULT);
685 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
687 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
693 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
694 struct ib_ucontext *context,
695 struct ib_udata *udata)
697 struct mlx4_ib_xrcd *xrcd;
700 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
701 return ERR_PTR(-ENOSYS);
703 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
705 return ERR_PTR(-ENOMEM);
707 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
711 xrcd->pd = ib_alloc_pd(ibdev);
712 if (IS_ERR(xrcd->pd)) {
713 err = PTR_ERR(xrcd->pd);
717 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
718 if (IS_ERR(xrcd->cq)) {
719 err = PTR_ERR(xrcd->cq);
723 return &xrcd->ibxrcd;
726 ib_dealloc_pd(xrcd->pd);
728 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
734 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
736 ib_destroy_cq(to_mxrcd(xrcd)->cq);
737 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
738 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
744 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
746 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
747 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
748 struct mlx4_ib_gid_entry *ge;
750 ge = kzalloc(sizeof *ge, GFP_KERNEL);
755 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
756 ge->port = mqp->port;
760 mutex_lock(&mqp->mutex);
761 list_add_tail(&ge->list, &mqp->gid_list);
762 mutex_unlock(&mqp->mutex);
767 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
771 struct net_device *ndev;
777 spin_lock(&mdev->iboe.lock);
778 ndev = mdev->iboe.netdevs[mqp->port - 1];
781 spin_unlock(&mdev->iboe.lock);
784 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
786 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
795 struct mlx4_ib_steering {
796 struct list_head list;
801 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
804 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
805 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
807 struct mlx4_ib_steering *ib_steering = NULL;
809 if (mdev->dev->caps.steering_mode ==
810 MLX4_STEERING_MODE_DEVICE_MANAGED) {
811 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
816 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
818 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
819 MLX4_PROT_IB_IPV6, ®_id);
823 err = add_gid_entry(ibqp, gid);
828 memcpy(ib_steering->gid.raw, gid->raw, 16);
829 ib_steering->reg_id = reg_id;
830 mutex_lock(&mqp->mutex);
831 list_add(&ib_steering->list, &mqp->steering_rules);
832 mutex_unlock(&mqp->mutex);
837 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
838 MLX4_PROT_IB_IPV6, reg_id);
845 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
847 struct mlx4_ib_gid_entry *ge;
848 struct mlx4_ib_gid_entry *tmp;
849 struct mlx4_ib_gid_entry *ret = NULL;
851 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
852 if (!memcmp(raw, ge->gid.raw, 16)) {
861 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
864 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
865 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
867 struct net_device *ndev;
868 struct mlx4_ib_gid_entry *ge;
871 if (mdev->dev->caps.steering_mode ==
872 MLX4_STEERING_MODE_DEVICE_MANAGED) {
873 struct mlx4_ib_steering *ib_steering;
875 mutex_lock(&mqp->mutex);
876 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
877 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
878 list_del(&ib_steering->list);
882 mutex_unlock(&mqp->mutex);
883 if (&ib_steering->list == &mqp->steering_rules) {
884 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
887 reg_id = ib_steering->reg_id;
891 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
892 MLX4_PROT_IB_IPV6, reg_id);
896 mutex_lock(&mqp->mutex);
897 ge = find_gid_entry(mqp, gid->raw);
899 spin_lock(&mdev->iboe.lock);
900 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
903 spin_unlock(&mdev->iboe.lock);
904 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
907 dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
914 pr_warn("could not find mgid entry\n");
916 mutex_unlock(&mqp->mutex);
921 static int init_node_data(struct mlx4_ib_dev *dev)
923 struct ib_smp *in_mad = NULL;
924 struct ib_smp *out_mad = NULL;
925 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
928 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
929 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
930 if (!in_mad || !out_mad)
933 init_query_mad(in_mad);
934 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
935 if (mlx4_is_master(dev->dev))
936 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
938 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
942 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
944 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
946 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
950 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
951 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
959 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
962 struct mlx4_ib_dev *dev =
963 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
964 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
967 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
970 struct mlx4_ib_dev *dev =
971 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
972 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
973 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
974 (int) dev->dev->caps.fw_ver & 0xffff);
977 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
980 struct mlx4_ib_dev *dev =
981 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
982 return sprintf(buf, "%x\n", dev->dev->rev_id);
985 static ssize_t show_board(struct device *device, struct device_attribute *attr,
988 struct mlx4_ib_dev *dev =
989 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
990 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
994 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
995 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
996 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
997 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
999 static struct device_attribute *mlx4_class_attributes[] = {
1006 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
1008 memcpy(eui, dev->dev_addr, 3);
1009 memcpy(eui + 5, dev->dev_addr + 3, 3);
1010 if (vlan_id < 0x1000) {
1011 eui[3] = vlan_id >> 8;
1012 eui[4] = vlan_id & 0xff;
1020 static void update_gids_task(struct work_struct *work)
1022 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1023 struct mlx4_cmd_mailbox *mailbox;
1026 struct mlx4_dev *dev = gw->dev->dev;
1028 mailbox = mlx4_alloc_cmd_mailbox(dev);
1029 if (IS_ERR(mailbox)) {
1030 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1034 gids = mailbox->buf;
1035 memcpy(gids, gw->gids, sizeof gw->gids);
1037 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1038 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1041 pr_warn("set port command failed\n");
1043 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1044 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1047 mlx4_free_cmd_mailbox(dev, mailbox);
1051 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1053 struct net_device *ndev = dev->iboe.netdevs[port - 1];
1054 struct update_gid_work *work;
1055 struct net_device *tmp;
1062 int need_update = 0;
1065 work = kzalloc(sizeof *work, GFP_ATOMIC);
1069 hits = kzalloc(128, GFP_ATOMIC);
1076 for_each_netdev_rcu(&init_net, tmp) {
1077 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1078 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1079 vid = rdma_vlan_dev_vlan_id(tmp);
1080 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1083 for (i = 0; i < 128; ++i) {
1085 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1087 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1096 (memcmp(&dev->iboe.gid_table[port - 1][0],
1097 &gid, sizeof gid) ||
1098 !memcmp(&dev->iboe.gid_table[port - 1][0],
1099 &zgid, sizeof gid))) {
1100 dev->iboe.gid_table[port - 1][0] = gid;
1103 } else if (free >= 0) {
1104 dev->iboe.gid_table[port - 1][free] = gid;
1113 for (i = 0; i < 128; ++i)
1115 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1117 dev->iboe.gid_table[port - 1][i] = zgid;
1121 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1122 INIT_WORK(&work->work, update_gids_task);
1125 queue_work(wq, &work->work);
1137 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1141 case NETDEV_CHANGEADDR:
1142 update_ipv6_gids(dev, port, 0);
1146 update_ipv6_gids(dev, port, 1);
1147 dev->iboe.netdevs[port - 1] = NULL;
1151 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1153 update_ipv6_gids(dev, port, 0);
1156 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1158 update_ipv6_gids(dev, port, 1);
1161 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1164 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1165 struct mlx4_ib_dev *ibdev;
1166 struct net_device *oldnd;
1167 struct mlx4_ib_iboe *iboe;
1170 if (!net_eq(dev_net(dev), &init_net))
1173 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1174 iboe = &ibdev->iboe;
1176 spin_lock(&iboe->lock);
1177 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1178 oldnd = iboe->netdevs[port - 1];
1179 iboe->netdevs[port - 1] =
1180 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1181 if (oldnd != iboe->netdevs[port - 1]) {
1182 if (iboe->netdevs[port - 1])
1183 netdev_added(ibdev, port);
1185 netdev_removed(ibdev, port);
1189 if (dev == iboe->netdevs[0] ||
1190 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1191 handle_en_event(ibdev, 1, event);
1192 else if (dev == iboe->netdevs[1]
1193 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1194 handle_en_event(ibdev, 2, event);
1196 spin_unlock(&iboe->lock);
1201 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1207 if (mlx4_is_master(ibdev->dev)) {
1208 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1209 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1211 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1213 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1214 /* master has the identity virt2phys pkey mapping */
1215 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1216 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1217 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1218 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1222 /* initialize pkey cache */
1223 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1225 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1227 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1233 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1236 int eq_per_port = 0;
1241 /* Legacy mode or comp_pool is not large enough */
1242 if (dev->caps.comp_pool == 0 ||
1243 dev->caps.num_ports > dev->caps.comp_pool)
1246 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1247 dev->caps.num_ports);
1251 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1252 added_eqs += eq_per_port;
1254 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1256 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1257 if (!ibdev->eq_table)
1260 ibdev->eq_added = added_eqs;
1263 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1264 for (j = 0; j < eq_per_port; j++) {
1265 sprintf(name, "mlx4-ib-%d-%d@%s",
1266 i, j, dev->pdev->bus->name);
1267 /* Set IRQ for specific name (per ring) */
1268 if (mlx4_assign_eq(dev, name, NULL,
1269 &ibdev->eq_table[eq])) {
1270 /* Use legacy (same as mlx4_en driver) */
1271 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1272 ibdev->eq_table[eq] =
1273 (eq % dev->caps.num_comp_vectors);
1279 /* Fill the reset of the vector with legacy EQ */
1280 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1281 ibdev->eq_table[eq++] = i;
1283 /* Advertise the new number of EQs to clients */
1284 ibdev->ib_dev.num_comp_vectors = total_eqs;
1287 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1291 /* no additional eqs were added */
1292 if (!ibdev->eq_table)
1295 /* Reset the advertised EQ number */
1296 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1298 /* Free only the added eqs */
1299 for (i = 0; i < ibdev->eq_added; i++) {
1300 /* Don't free legacy eqs if used */
1301 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1303 mlx4_release_eq(dev, ibdev->eq_table[i]);
1306 kfree(ibdev->eq_table);
1309 static void *mlx4_ib_add(struct mlx4_dev *dev)
1311 struct mlx4_ib_dev *ibdev;
1315 struct mlx4_ib_iboe *iboe;
1317 pr_info_once("%s", mlx4_ib_version);
1319 mlx4_foreach_non_ib_transport_port(i, dev)
1322 if (mlx4_is_mfunc(dev) && num_ports) {
1323 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1328 mlx4_foreach_ib_transport_port(i, dev)
1331 /* No point in registering a device with no ports... */
1335 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1337 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1341 iboe = &ibdev->iboe;
1343 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1346 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1349 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1351 if (!ibdev->uar_map)
1353 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1357 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1358 ibdev->ib_dev.owner = THIS_MODULE;
1359 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1360 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1361 ibdev->num_ports = num_ports;
1362 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1363 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1364 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1366 if (dev->caps.userspace_caps)
1367 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1369 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1371 ibdev->ib_dev.uverbs_cmd_mask =
1372 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1373 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1374 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1375 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1376 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1377 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1378 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1379 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1380 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1381 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1382 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1383 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1384 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1385 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1386 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1387 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1388 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1389 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1390 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1391 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1392 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1393 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1394 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1396 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1397 ibdev->ib_dev.query_port = mlx4_ib_query_port;
1398 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
1399 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1400 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1401 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1402 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1403 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1404 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1405 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1406 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1407 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1408 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1409 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1410 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1411 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1412 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
1413 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
1414 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1415 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1416 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1417 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
1418 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
1419 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1420 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1421 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1422 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
1423 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
1424 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
1425 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1426 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1427 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1428 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1429 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1430 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
1431 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1432 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1433 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
1434 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1435 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1436 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1438 if (!mlx4_is_slave(ibdev->dev)) {
1439 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1440 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1441 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1442 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1445 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1446 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
1447 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
1448 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
1449 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
1451 ibdev->ib_dev.uverbs_cmd_mask |=
1452 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
1453 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
1456 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1457 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1458 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1459 ibdev->ib_dev.uverbs_cmd_mask |=
1460 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1461 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1464 mlx4_ib_alloc_eqs(dev, ibdev);
1466 spin_lock_init(&iboe->lock);
1468 if (init_node_data(ibdev))
1471 for (i = 0; i < ibdev->num_ports; ++i) {
1472 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1473 IB_LINK_LAYER_ETHERNET) {
1474 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1476 ibdev->counters[i] = -1;
1478 ibdev->counters[i] = -1;
1481 spin_lock_init(&ibdev->sm_lock);
1482 mutex_init(&ibdev->cap_mask_mutex);
1484 if (ib_register_device(&ibdev->ib_dev, NULL))
1487 if (mlx4_ib_mad_init(ibdev))
1490 if (mlx4_ib_init_sriov(ibdev))
1493 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1494 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1495 err = register_netdevice_notifier(&iboe->nb);
1500 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1501 if (device_create_file(&ibdev->ib_dev.dev,
1502 mlx4_class_attributes[j]))
1506 ibdev->ib_active = true;
1508 if (mlx4_is_mfunc(ibdev->dev))
1511 /* create paravirt contexts for any VFs which are active */
1512 if (mlx4_is_master(ibdev->dev)) {
1513 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
1514 if (j == mlx4_master_func_num(ibdev->dev))
1516 if (mlx4_is_slave_active(ibdev->dev, j))
1517 do_slave_init(ibdev, j, 1);
1523 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1524 pr_warn("failure unregistering notifier\n");
1525 flush_workqueue(wq);
1528 mlx4_ib_close_sriov(ibdev);
1531 mlx4_ib_mad_cleanup(ibdev);
1534 ib_unregister_device(&ibdev->ib_dev);
1538 if (ibdev->counters[i - 1] != -1)
1539 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1542 iounmap(ibdev->uar_map);
1545 mlx4_uar_free(dev, &ibdev->priv_uar);
1548 mlx4_pd_free(dev, ibdev->priv_pdn);
1551 ib_dealloc_device(&ibdev->ib_dev);
1556 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1558 struct mlx4_ib_dev *ibdev = ibdev_ptr;
1561 mlx4_ib_close_sriov(ibdev);
1562 mlx4_ib_mad_cleanup(ibdev);
1563 ib_unregister_device(&ibdev->ib_dev);
1564 if (ibdev->iboe.nb.notifier_call) {
1565 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1566 pr_warn("failure unregistering notifier\n");
1567 ibdev->iboe.nb.notifier_call = NULL;
1569 iounmap(ibdev->uar_map);
1570 for (p = 0; p < ibdev->num_ports; ++p)
1571 if (ibdev->counters[p] != -1)
1572 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1573 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1574 mlx4_CLOSE_PORT(dev, p);
1576 mlx4_ib_free_eqs(dev, ibdev);
1578 mlx4_uar_free(dev, &ibdev->priv_uar);
1579 mlx4_pd_free(dev, ibdev->priv_pdn);
1580 ib_dealloc_device(&ibdev->ib_dev);
1583 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1585 struct mlx4_ib_demux_work **dm = NULL;
1586 struct mlx4_dev *dev = ibdev->dev;
1588 unsigned long flags;
1590 if (!mlx4_is_master(dev))
1593 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
1595 pr_err("failed to allocate memory for tunneling qp update\n");
1599 for (i = 0; i < dev->caps.num_ports; i++) {
1600 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
1602 pr_err("failed to allocate memory for tunneling qp update work struct\n");
1603 for (i = 0; i < dev->caps.num_ports; i++) {
1610 /* initialize or tear down tunnel QPs for the slave */
1611 for (i = 0; i < dev->caps.num_ports; i++) {
1612 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
1613 dm[i]->port = i + 1;
1614 dm[i]->slave = slave;
1615 dm[i]->do_init = do_init;
1617 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
1618 if (!ibdev->sriov.is_going_down)
1619 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
1620 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1627 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1628 enum mlx4_dev_event event, unsigned long param)
1630 struct ib_event ibev;
1631 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1632 struct mlx4_eqe *eqe = NULL;
1633 struct ib_event_work *ew;
1636 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1637 eqe = (struct mlx4_eqe *)param;
1642 case MLX4_DEV_EVENT_PORT_UP:
1643 if (p > ibdev->num_ports)
1645 if (mlx4_is_master(dev) &&
1646 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
1647 IB_LINK_LAYER_INFINIBAND) {
1648 mlx4_ib_invalidate_all_guid_record(ibdev, p);
1650 ibev.event = IB_EVENT_PORT_ACTIVE;
1653 case MLX4_DEV_EVENT_PORT_DOWN:
1654 if (p > ibdev->num_ports)
1656 ibev.event = IB_EVENT_PORT_ERR;
1659 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1660 ibdev->ib_active = false;
1661 ibev.event = IB_EVENT_DEVICE_FATAL;
1664 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1665 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1667 pr_err("failed to allocate memory for events work\n");
1671 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1672 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1674 /* need to queue only for port owner, which uses GEN_EQE */
1675 if (mlx4_is_master(dev))
1676 queue_work(wq, &ew->work);
1678 handle_port_mgmt_change_event(&ew->work);
1681 case MLX4_DEV_EVENT_SLAVE_INIT:
1682 /* here, p is the slave id */
1683 do_slave_init(ibdev, p, 1);
1686 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
1687 /* here, p is the slave id */
1688 do_slave_init(ibdev, p, 0);
1695 ibev.device = ibdev_ptr;
1696 ibev.element.port_num = (u8) p;
1698 ib_dispatch_event(&ibev);
1701 static struct mlx4_interface mlx4_ib_interface = {
1703 .remove = mlx4_ib_remove,
1704 .event = mlx4_ib_event,
1705 .protocol = MLX4_PROT_IB_IPV6
1708 static int __init mlx4_ib_init(void)
1712 wq = create_singlethread_workqueue("mlx4_ib");
1716 err = mlx4_ib_mcg_init();
1720 err = mlx4_register_interface(&mlx4_ib_interface);
1727 mlx4_ib_mcg_destroy();
1730 destroy_workqueue(wq);
1734 static void __exit mlx4_ib_cleanup(void)
1736 mlx4_unregister_interface(&mlx4_ib_interface);
1737 mlx4_ib_mcg_destroy();
1738 destroy_workqueue(wq);
1741 module_init(mlx4_ib_init);
1742 module_exit(mlx4_ib_cleanup);