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 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
59 MODULE_AUTHOR("Roland Dreier");
60 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
61 MODULE_LICENSE("Dual BSD/GPL");
62 MODULE_VERSION(DRV_VERSION);
64 int mlx4_ib_sm_guid_assign = 1;
65 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
66 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
68 static const char mlx4_ib_version[] =
69 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
70 DRV_VERSION " (" DRV_RELDATE ")\n";
72 struct update_gid_work {
73 struct work_struct work;
74 union ib_gid gids[128];
75 struct mlx4_ib_dev *dev;
79 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
81 static struct workqueue_struct *wq;
83 static void init_query_mad(struct ib_smp *mad)
85 mad->base_version = 1;
86 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
87 mad->class_version = 1;
88 mad->method = IB_MGMT_METHOD_GET;
91 static union ib_gid zgid;
93 static int check_flow_steering_support(struct mlx4_dev *dev)
98 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
101 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
102 if (ib_num_ports || mlx4_is_mfunc(dev)) {
103 pr_warn("Device managed flow steering is unavailable "
104 "for IB ports or in multifunction env.\n");
112 static int mlx4_ib_query_device(struct ib_device *ibdev,
113 struct ib_device_attr *props)
115 struct mlx4_ib_dev *dev = to_mdev(ibdev);
116 struct ib_smp *in_mad = NULL;
117 struct ib_smp *out_mad = NULL;
120 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
121 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
122 if (!in_mad || !out_mad)
125 init_query_mad(in_mad);
126 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
128 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
129 1, NULL, NULL, in_mad, out_mad);
133 memset(props, 0, sizeof *props);
135 props->fw_ver = dev->dev->caps.fw_ver;
136 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
137 IB_DEVICE_PORT_ACTIVE_EVENT |
138 IB_DEVICE_SYS_IMAGE_GUID |
139 IB_DEVICE_RC_RNR_NAK_GEN |
140 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
141 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
142 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
143 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
144 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
145 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
146 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
147 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
148 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
149 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
150 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
151 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
152 props->device_cap_flags |= IB_DEVICE_UD_TSO;
153 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
154 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
155 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
156 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
157 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
158 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
159 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
160 props->device_cap_flags |= IB_DEVICE_XRC;
161 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
162 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
163 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
164 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
165 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
167 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
168 if (check_flow_steering_support(dev->dev))
169 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
172 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
174 props->vendor_part_id = dev->dev->pdev->device;
175 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
176 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
178 props->max_mr_size = ~0ull;
179 props->page_size_cap = dev->dev->caps.page_size_cap;
180 props->max_qp = dev->dev->quotas.qp;
181 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
182 props->max_sge = min(dev->dev->caps.max_sq_sg,
183 dev->dev->caps.max_rq_sg);
184 props->max_cq = dev->dev->quotas.cq;
185 props->max_cqe = dev->dev->caps.max_cqes;
186 props->max_mr = dev->dev->quotas.mpt;
187 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
188 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
189 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
190 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
191 props->max_srq = dev->dev->quotas.srq;
192 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
193 props->max_srq_sge = dev->dev->caps.max_srq_sge;
194 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
195 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
196 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
197 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
198 props->masked_atomic_cap = props->atomic_cap;
199 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
200 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
201 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
202 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
203 props->max_mcast_grp;
204 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
213 static enum rdma_link_layer
214 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
216 struct mlx4_dev *dev = to_mdev(device)->dev;
218 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
219 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
222 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
223 struct ib_port_attr *props, int netw_view)
225 struct ib_smp *in_mad = NULL;
226 struct ib_smp *out_mad = NULL;
227 int ext_active_speed;
228 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
231 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
232 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
233 if (!in_mad || !out_mad)
236 init_query_mad(in_mad);
237 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
238 in_mad->attr_mod = cpu_to_be32(port);
240 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
241 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
243 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
249 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
250 props->lmc = out_mad->data[34] & 0x7;
251 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
252 props->sm_sl = out_mad->data[36] & 0xf;
253 props->state = out_mad->data[32] & 0xf;
254 props->phys_state = out_mad->data[33] >> 4;
255 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
257 props->gid_tbl_len = out_mad->data[50];
259 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
260 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
261 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
262 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
263 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
264 props->active_width = out_mad->data[31] & 0xf;
265 props->active_speed = out_mad->data[35] >> 4;
266 props->max_mtu = out_mad->data[41] & 0xf;
267 props->active_mtu = out_mad->data[36] >> 4;
268 props->subnet_timeout = out_mad->data[51] & 0x1f;
269 props->max_vl_num = out_mad->data[37] >> 4;
270 props->init_type_reply = out_mad->data[41] >> 4;
272 /* Check if extended speeds (EDR/FDR/...) are supported */
273 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
274 ext_active_speed = out_mad->data[62] >> 4;
276 switch (ext_active_speed) {
278 props->active_speed = IB_SPEED_FDR;
281 props->active_speed = IB_SPEED_EDR;
286 /* If reported active speed is QDR, check if is FDR-10 */
287 if (props->active_speed == IB_SPEED_QDR) {
288 init_query_mad(in_mad);
289 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
290 in_mad->attr_mod = cpu_to_be32(port);
292 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
293 NULL, NULL, in_mad, out_mad);
297 /* Checking LinkSpeedActive for FDR-10 */
298 if (out_mad->data[15] & 0x1)
299 props->active_speed = IB_SPEED_FDR10;
302 /* Avoid wrong speed value returned by FW if the IB link is down. */
303 if (props->state == IB_PORT_DOWN)
304 props->active_speed = IB_SPEED_SDR;
312 static u8 state_to_phys_state(enum ib_port_state state)
314 return state == IB_PORT_ACTIVE ? 5 : 3;
317 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
318 struct ib_port_attr *props, int netw_view)
321 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
322 struct mlx4_ib_iboe *iboe = &mdev->iboe;
323 struct net_device *ndev;
325 struct mlx4_cmd_mailbox *mailbox;
328 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
330 return PTR_ERR(mailbox);
332 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
333 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
338 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
339 IB_WIDTH_4X : IB_WIDTH_1X;
340 props->active_speed = IB_SPEED_QDR;
341 props->port_cap_flags = IB_PORT_CM_SUP;
342 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
343 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
344 props->pkey_tbl_len = 1;
345 props->max_mtu = IB_MTU_4096;
346 props->max_vl_num = 2;
347 props->state = IB_PORT_DOWN;
348 props->phys_state = state_to_phys_state(props->state);
349 props->active_mtu = IB_MTU_256;
350 spin_lock(&iboe->lock);
351 ndev = iboe->netdevs[port - 1];
355 tmp = iboe_get_mtu(ndev->mtu);
356 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
358 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
359 IB_PORT_ACTIVE : IB_PORT_DOWN;
360 props->phys_state = state_to_phys_state(props->state);
362 spin_unlock(&iboe->lock);
364 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
368 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
369 struct ib_port_attr *props, int netw_view)
373 memset(props, 0, sizeof *props);
375 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
376 ib_link_query_port(ibdev, port, props, netw_view) :
377 eth_link_query_port(ibdev, port, props, netw_view);
382 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
383 struct ib_port_attr *props)
385 /* returns host view */
386 return __mlx4_ib_query_port(ibdev, port, props, 0);
389 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
390 union ib_gid *gid, int netw_view)
392 struct ib_smp *in_mad = NULL;
393 struct ib_smp *out_mad = NULL;
395 struct mlx4_ib_dev *dev = to_mdev(ibdev);
397 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
399 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
400 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
401 if (!in_mad || !out_mad)
404 init_query_mad(in_mad);
405 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
406 in_mad->attr_mod = cpu_to_be32(port);
408 if (mlx4_is_mfunc(dev->dev) && netw_view)
409 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
411 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
415 memcpy(gid->raw, out_mad->data + 8, 8);
417 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
419 /* For any index > 0, return the null guid */
426 init_query_mad(in_mad);
427 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
428 in_mad->attr_mod = cpu_to_be32(index / 8);
430 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
431 NULL, NULL, in_mad, out_mad);
435 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
439 memset(gid->raw + 8, 0, 8);
445 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
448 struct mlx4_ib_dev *dev = to_mdev(ibdev);
450 *gid = dev->iboe.gid_table[port - 1][index];
455 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
458 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
459 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
461 return iboe_query_gid(ibdev, port, index, gid);
464 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
465 u16 *pkey, int netw_view)
467 struct ib_smp *in_mad = NULL;
468 struct ib_smp *out_mad = NULL;
469 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
472 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
473 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
474 if (!in_mad || !out_mad)
477 init_query_mad(in_mad);
478 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
479 in_mad->attr_mod = cpu_to_be32(index / 32);
481 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
482 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
484 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
489 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
497 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
499 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
502 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
503 struct ib_device_modify *props)
505 struct mlx4_cmd_mailbox *mailbox;
508 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
511 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
514 if (mlx4_is_slave(to_mdev(ibdev)->dev))
517 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
518 memcpy(ibdev->node_desc, props->node_desc, 64);
519 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
522 * If possible, pass node desc to FW, so it can generate
523 * a 144 trap. If cmd fails, just ignore.
525 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
529 memcpy(mailbox->buf, props->node_desc, 64);
530 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
531 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
533 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
538 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
541 struct mlx4_cmd_mailbox *mailbox;
543 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
545 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
547 return PTR_ERR(mailbox);
549 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
550 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
551 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
553 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
554 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
557 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
558 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
560 mlx4_free_cmd_mailbox(dev->dev, mailbox);
564 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
565 struct ib_port_modify *props)
567 struct ib_port_attr attr;
571 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
573 err = mlx4_ib_query_port(ibdev, port, &attr);
577 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
578 ~props->clr_port_cap_mask;
580 err = mlx4_SET_PORT(to_mdev(ibdev), port,
581 !!(mask & IB_PORT_RESET_QKEY_CNTR),
585 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
589 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
590 struct ib_udata *udata)
592 struct mlx4_ib_dev *dev = to_mdev(ibdev);
593 struct mlx4_ib_ucontext *context;
594 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
595 struct mlx4_ib_alloc_ucontext_resp resp;
599 return ERR_PTR(-EAGAIN);
601 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
602 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
603 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
604 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
606 resp.dev_caps = dev->dev->caps.userspace_caps;
607 resp.qp_tab_size = dev->dev->caps.num_qps;
608 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
609 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
610 resp.cqe_size = dev->dev->caps.cqe_size;
613 context = kmalloc(sizeof *context, GFP_KERNEL);
615 return ERR_PTR(-ENOMEM);
617 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
623 INIT_LIST_HEAD(&context->db_page_list);
624 mutex_init(&context->db_page_mutex);
626 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
627 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
629 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
632 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
634 return ERR_PTR(-EFAULT);
637 return &context->ibucontext;
640 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
642 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
644 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
650 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
652 struct mlx4_ib_dev *dev = to_mdev(context->device);
654 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
657 if (vma->vm_pgoff == 0) {
658 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
660 if (io_remap_pfn_range(vma, vma->vm_start,
661 to_mucontext(context)->uar.pfn,
662 PAGE_SIZE, vma->vm_page_prot))
664 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
665 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
667 if (io_remap_pfn_range(vma, vma->vm_start,
668 to_mucontext(context)->uar.pfn +
669 dev->dev->caps.num_uars,
670 PAGE_SIZE, vma->vm_page_prot))
678 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
679 struct ib_ucontext *context,
680 struct ib_udata *udata)
682 struct mlx4_ib_pd *pd;
685 pd = kmalloc(sizeof *pd, GFP_KERNEL);
687 return ERR_PTR(-ENOMEM);
689 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
696 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
697 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
699 return ERR_PTR(-EFAULT);
705 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
707 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
713 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
714 struct ib_ucontext *context,
715 struct ib_udata *udata)
717 struct mlx4_ib_xrcd *xrcd;
720 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
721 return ERR_PTR(-ENOSYS);
723 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
725 return ERR_PTR(-ENOMEM);
727 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
731 xrcd->pd = ib_alloc_pd(ibdev);
732 if (IS_ERR(xrcd->pd)) {
733 err = PTR_ERR(xrcd->pd);
737 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
738 if (IS_ERR(xrcd->cq)) {
739 err = PTR_ERR(xrcd->cq);
743 return &xrcd->ibxrcd;
746 ib_dealloc_pd(xrcd->pd);
748 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
754 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
756 ib_destroy_cq(to_mxrcd(xrcd)->cq);
757 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
758 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
764 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
766 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
767 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
768 struct mlx4_ib_gid_entry *ge;
770 ge = kzalloc(sizeof *ge, GFP_KERNEL);
775 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
776 ge->port = mqp->port;
780 mutex_lock(&mqp->mutex);
781 list_add_tail(&ge->list, &mqp->gid_list);
782 mutex_unlock(&mqp->mutex);
787 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
791 struct net_device *ndev;
797 spin_lock(&mdev->iboe.lock);
798 ndev = mdev->iboe.netdevs[mqp->port - 1];
801 spin_unlock(&mdev->iboe.lock);
804 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
806 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
815 struct mlx4_ib_steering {
816 struct list_head list;
821 static int parse_flow_attr(struct mlx4_dev *dev,
822 union ib_flow_spec *ib_spec,
823 struct _rule_hw *mlx4_spec)
825 enum mlx4_net_trans_rule_id type;
827 switch (ib_spec->type) {
828 case IB_FLOW_SPEC_ETH:
829 type = MLX4_NET_TRANS_RULE_ID_ETH;
830 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
832 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
834 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
835 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
838 case IB_FLOW_SPEC_IPV4:
839 type = MLX4_NET_TRANS_RULE_ID_IPV4;
840 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
841 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
842 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
843 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
846 case IB_FLOW_SPEC_TCP:
847 case IB_FLOW_SPEC_UDP:
848 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
849 MLX4_NET_TRANS_RULE_ID_TCP :
850 MLX4_NET_TRANS_RULE_ID_UDP;
851 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
852 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
853 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
854 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
860 if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
861 mlx4_hw_rule_sz(dev, type) < 0)
863 mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
864 mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
865 return mlx4_hw_rule_sz(dev, type);
868 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
870 enum mlx4_net_trans_promisc_mode flow_type,
876 struct mlx4_ib_dev *mdev = to_mdev(qp->device);
877 struct mlx4_cmd_mailbox *mailbox;
878 struct mlx4_net_trans_rule_hw_ctrl *ctrl;
880 static const u16 __mlx4_domain[] = {
881 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
882 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
883 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
884 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
887 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
888 pr_err("Invalid priority value %d\n", flow_attr->priority);
892 if (domain >= IB_FLOW_DOMAIN_NUM) {
893 pr_err("Invalid domain value %d\n", domain);
897 if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
900 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
902 return PTR_ERR(mailbox);
905 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
906 flow_attr->priority);
907 ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
908 ctrl->port = flow_attr->port;
909 ctrl->qpn = cpu_to_be32(qp->qp_num);
911 ib_flow = flow_attr + 1;
912 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
913 for (i = 0; i < flow_attr->num_of_specs; i++) {
914 ret = parse_flow_attr(mdev->dev, ib_flow, mailbox->buf + size);
916 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
919 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
923 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
924 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
927 pr_err("mcg table is full. Fail to register network rule.\n");
928 else if (ret == -ENXIO)
929 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
931 pr_err("Invalid argumant. Fail to register network rule.\n");
933 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
937 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
940 err = mlx4_cmd(dev, reg_id, 0, 0,
941 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
944 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
949 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
950 struct ib_flow_attr *flow_attr,
954 struct mlx4_ib_flow *mflow;
955 enum mlx4_net_trans_promisc_mode type[2];
957 memset(type, 0, sizeof(type));
959 mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
965 switch (flow_attr->type) {
966 case IB_FLOW_ATTR_NORMAL:
967 type[0] = MLX4_FS_REGULAR;
970 case IB_FLOW_ATTR_ALL_DEFAULT:
971 type[0] = MLX4_FS_ALL_DEFAULT;
974 case IB_FLOW_ATTR_MC_DEFAULT:
975 type[0] = MLX4_FS_MC_DEFAULT;
978 case IB_FLOW_ATTR_SNIFFER:
979 type[0] = MLX4_FS_UC_SNIFFER;
980 type[1] = MLX4_FS_MC_SNIFFER;
988 while (i < ARRAY_SIZE(type) && type[i]) {
989 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
996 return &mflow->ibflow;
1000 return ERR_PTR(err);
1003 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1007 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1008 struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1010 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1011 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1021 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1024 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1025 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1027 struct mlx4_ib_steering *ib_steering = NULL;
1029 if (mdev->dev->caps.steering_mode ==
1030 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1031 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1036 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1038 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1039 MLX4_PROT_IB_IPV6, ®_id);
1043 err = add_gid_entry(ibqp, gid);
1048 memcpy(ib_steering->gid.raw, gid->raw, 16);
1049 ib_steering->reg_id = reg_id;
1050 mutex_lock(&mqp->mutex);
1051 list_add(&ib_steering->list, &mqp->steering_rules);
1052 mutex_unlock(&mqp->mutex);
1057 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1058 MLX4_PROT_IB_IPV6, reg_id);
1065 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1067 struct mlx4_ib_gid_entry *ge;
1068 struct mlx4_ib_gid_entry *tmp;
1069 struct mlx4_ib_gid_entry *ret = NULL;
1071 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1072 if (!memcmp(raw, ge->gid.raw, 16)) {
1081 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1084 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1085 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1087 struct net_device *ndev;
1088 struct mlx4_ib_gid_entry *ge;
1091 if (mdev->dev->caps.steering_mode ==
1092 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1093 struct mlx4_ib_steering *ib_steering;
1095 mutex_lock(&mqp->mutex);
1096 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1097 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1098 list_del(&ib_steering->list);
1102 mutex_unlock(&mqp->mutex);
1103 if (&ib_steering->list == &mqp->steering_rules) {
1104 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1107 reg_id = ib_steering->reg_id;
1111 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1112 MLX4_PROT_IB_IPV6, reg_id);
1116 mutex_lock(&mqp->mutex);
1117 ge = find_gid_entry(mqp, gid->raw);
1119 spin_lock(&mdev->iboe.lock);
1120 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1123 spin_unlock(&mdev->iboe.lock);
1124 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
1127 dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
1131 list_del(&ge->list);
1134 pr_warn("could not find mgid entry\n");
1136 mutex_unlock(&mqp->mutex);
1141 static int init_node_data(struct mlx4_ib_dev *dev)
1143 struct ib_smp *in_mad = NULL;
1144 struct ib_smp *out_mad = NULL;
1145 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1148 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1149 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1150 if (!in_mad || !out_mad)
1153 init_query_mad(in_mad);
1154 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1155 if (mlx4_is_master(dev->dev))
1156 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1158 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1162 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1164 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1166 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1170 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1171 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1179 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1182 struct mlx4_ib_dev *dev =
1183 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1184 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1187 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1190 struct mlx4_ib_dev *dev =
1191 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1192 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1193 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1194 (int) dev->dev->caps.fw_ver & 0xffff);
1197 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1200 struct mlx4_ib_dev *dev =
1201 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1202 return sprintf(buf, "%x\n", dev->dev->rev_id);
1205 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1208 struct mlx4_ib_dev *dev =
1209 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1210 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1211 dev->dev->board_id);
1214 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1215 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1216 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1217 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1219 static struct device_attribute *mlx4_class_attributes[] = {
1226 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
1228 memcpy(eui, dev->dev_addr, 3);
1229 memcpy(eui + 5, dev->dev_addr + 3, 3);
1230 if (vlan_id < 0x1000) {
1231 eui[3] = vlan_id >> 8;
1232 eui[4] = vlan_id & 0xff;
1240 static void update_gids_task(struct work_struct *work)
1242 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1243 struct mlx4_cmd_mailbox *mailbox;
1246 struct mlx4_dev *dev = gw->dev->dev;
1248 mailbox = mlx4_alloc_cmd_mailbox(dev);
1249 if (IS_ERR(mailbox)) {
1250 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1254 gids = mailbox->buf;
1255 memcpy(gids, gw->gids, sizeof gw->gids);
1257 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1258 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1261 pr_warn("set port command failed\n");
1263 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1264 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1267 mlx4_free_cmd_mailbox(dev, mailbox);
1271 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1273 struct net_device *ndev = dev->iboe.netdevs[port - 1];
1274 struct update_gid_work *work;
1275 struct net_device *tmp;
1282 int need_update = 0;
1285 work = kzalloc(sizeof *work, GFP_ATOMIC);
1289 hits = kzalloc(128, GFP_ATOMIC);
1296 for_each_netdev_rcu(&init_net, tmp) {
1297 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1298 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1299 vid = rdma_vlan_dev_vlan_id(tmp);
1300 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1303 for (i = 0; i < 128; ++i) {
1305 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1307 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1316 (memcmp(&dev->iboe.gid_table[port - 1][0],
1317 &gid, sizeof gid) ||
1318 !memcmp(&dev->iboe.gid_table[port - 1][0],
1319 &zgid, sizeof gid))) {
1320 dev->iboe.gid_table[port - 1][0] = gid;
1323 } else if (free >= 0) {
1324 dev->iboe.gid_table[port - 1][free] = gid;
1333 for (i = 0; i < 128; ++i)
1335 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1337 dev->iboe.gid_table[port - 1][i] = zgid;
1341 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1342 INIT_WORK(&work->work, update_gids_task);
1345 queue_work(wq, &work->work);
1357 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1361 case NETDEV_CHANGEADDR:
1362 update_ipv6_gids(dev, port, 0);
1366 update_ipv6_gids(dev, port, 1);
1367 dev->iboe.netdevs[port - 1] = NULL;
1371 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1373 update_ipv6_gids(dev, port, 0);
1376 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1378 update_ipv6_gids(dev, port, 1);
1381 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1384 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1385 struct mlx4_ib_dev *ibdev;
1386 struct net_device *oldnd;
1387 struct mlx4_ib_iboe *iboe;
1390 if (!net_eq(dev_net(dev), &init_net))
1393 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1394 iboe = &ibdev->iboe;
1396 spin_lock(&iboe->lock);
1397 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1398 oldnd = iboe->netdevs[port - 1];
1399 iboe->netdevs[port - 1] =
1400 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1401 if (oldnd != iboe->netdevs[port - 1]) {
1402 if (iboe->netdevs[port - 1])
1403 netdev_added(ibdev, port);
1405 netdev_removed(ibdev, port);
1409 if (dev == iboe->netdevs[0] ||
1410 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1411 handle_en_event(ibdev, 1, event);
1412 else if (dev == iboe->netdevs[1]
1413 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1414 handle_en_event(ibdev, 2, event);
1416 spin_unlock(&iboe->lock);
1421 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1427 if (mlx4_is_master(ibdev->dev)) {
1428 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1429 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1431 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1433 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1434 /* master has the identity virt2phys pkey mapping */
1435 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1436 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1437 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1438 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1442 /* initialize pkey cache */
1443 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1445 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1447 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1453 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1456 int eq_per_port = 0;
1461 /* Legacy mode or comp_pool is not large enough */
1462 if (dev->caps.comp_pool == 0 ||
1463 dev->caps.num_ports > dev->caps.comp_pool)
1466 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1467 dev->caps.num_ports);
1471 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1472 added_eqs += eq_per_port;
1474 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1476 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1477 if (!ibdev->eq_table)
1480 ibdev->eq_added = added_eqs;
1483 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1484 for (j = 0; j < eq_per_port; j++) {
1485 sprintf(name, "mlx4-ib-%d-%d@%s",
1486 i, j, dev->pdev->bus->name);
1487 /* Set IRQ for specific name (per ring) */
1488 if (mlx4_assign_eq(dev, name, NULL,
1489 &ibdev->eq_table[eq])) {
1490 /* Use legacy (same as mlx4_en driver) */
1491 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1492 ibdev->eq_table[eq] =
1493 (eq % dev->caps.num_comp_vectors);
1499 /* Fill the reset of the vector with legacy EQ */
1500 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1501 ibdev->eq_table[eq++] = i;
1503 /* Advertise the new number of EQs to clients */
1504 ibdev->ib_dev.num_comp_vectors = total_eqs;
1507 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1511 /* no additional eqs were added */
1512 if (!ibdev->eq_table)
1515 /* Reset the advertised EQ number */
1516 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1518 /* Free only the added eqs */
1519 for (i = 0; i < ibdev->eq_added; i++) {
1520 /* Don't free legacy eqs if used */
1521 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1523 mlx4_release_eq(dev, ibdev->eq_table[i]);
1526 kfree(ibdev->eq_table);
1529 static void *mlx4_ib_add(struct mlx4_dev *dev)
1531 struct mlx4_ib_dev *ibdev;
1535 struct mlx4_ib_iboe *iboe;
1537 pr_info_once("%s", mlx4_ib_version);
1539 mlx4_foreach_non_ib_transport_port(i, dev)
1542 if (mlx4_is_mfunc(dev) && num_ports) {
1543 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1548 mlx4_foreach_ib_transport_port(i, dev)
1551 /* No point in registering a device with no ports... */
1555 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1557 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1561 iboe = &ibdev->iboe;
1563 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1566 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1569 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1571 if (!ibdev->uar_map)
1573 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1577 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1578 ibdev->ib_dev.owner = THIS_MODULE;
1579 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1580 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1581 ibdev->num_ports = num_ports;
1582 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1583 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1584 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1586 if (dev->caps.userspace_caps)
1587 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1589 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1591 ibdev->ib_dev.uverbs_cmd_mask =
1592 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1593 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1594 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1595 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1596 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1597 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1598 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1599 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1600 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1601 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1602 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1603 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1604 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1605 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1606 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1607 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1608 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1609 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1610 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1611 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1612 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1613 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1614 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1616 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1617 ibdev->ib_dev.query_port = mlx4_ib_query_port;
1618 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
1619 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1620 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1621 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1622 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1623 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1624 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1625 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1626 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1627 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1628 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1629 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1630 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1631 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1632 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
1633 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
1634 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1635 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1636 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1637 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
1638 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
1639 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1640 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1641 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1642 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
1643 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
1644 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
1645 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1646 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1647 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1648 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1649 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1650 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
1651 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1652 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1653 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
1654 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1655 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1656 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1658 if (!mlx4_is_slave(ibdev->dev)) {
1659 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1660 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1661 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1662 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1665 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1666 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
1667 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
1668 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
1669 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
1671 ibdev->ib_dev.uverbs_cmd_mask |=
1672 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
1673 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
1676 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1677 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1678 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1679 ibdev->ib_dev.uverbs_cmd_mask |=
1680 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1681 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1684 if (check_flow_steering_support(dev)) {
1685 ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
1686 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
1688 ibdev->ib_dev.uverbs_ex_cmd_mask |=
1689 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
1690 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
1693 mlx4_ib_alloc_eqs(dev, ibdev);
1695 spin_lock_init(&iboe->lock);
1697 if (init_node_data(ibdev))
1700 for (i = 0; i < ibdev->num_ports; ++i) {
1701 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1702 IB_LINK_LAYER_ETHERNET) {
1703 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1705 ibdev->counters[i] = -1;
1707 ibdev->counters[i] = -1;
1710 spin_lock_init(&ibdev->sm_lock);
1711 mutex_init(&ibdev->cap_mask_mutex);
1713 if (ib_register_device(&ibdev->ib_dev, NULL))
1716 if (mlx4_ib_mad_init(ibdev))
1719 if (mlx4_ib_init_sriov(ibdev))
1722 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1723 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1724 err = register_netdevice_notifier(&iboe->nb);
1729 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1730 if (device_create_file(&ibdev->ib_dev.dev,
1731 mlx4_class_attributes[j]))
1735 ibdev->ib_active = true;
1737 if (mlx4_is_mfunc(ibdev->dev))
1740 /* create paravirt contexts for any VFs which are active */
1741 if (mlx4_is_master(ibdev->dev)) {
1742 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
1743 if (j == mlx4_master_func_num(ibdev->dev))
1745 if (mlx4_is_slave_active(ibdev->dev, j))
1746 do_slave_init(ibdev, j, 1);
1752 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1753 pr_warn("failure unregistering notifier\n");
1754 flush_workqueue(wq);
1757 mlx4_ib_close_sriov(ibdev);
1760 mlx4_ib_mad_cleanup(ibdev);
1763 ib_unregister_device(&ibdev->ib_dev);
1767 if (ibdev->counters[i - 1] != -1)
1768 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1771 iounmap(ibdev->uar_map);
1774 mlx4_uar_free(dev, &ibdev->priv_uar);
1777 mlx4_pd_free(dev, ibdev->priv_pdn);
1780 ib_dealloc_device(&ibdev->ib_dev);
1785 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1787 struct mlx4_ib_dev *ibdev = ibdev_ptr;
1790 mlx4_ib_close_sriov(ibdev);
1791 mlx4_ib_mad_cleanup(ibdev);
1792 ib_unregister_device(&ibdev->ib_dev);
1793 if (ibdev->iboe.nb.notifier_call) {
1794 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1795 pr_warn("failure unregistering notifier\n");
1796 ibdev->iboe.nb.notifier_call = NULL;
1798 iounmap(ibdev->uar_map);
1799 for (p = 0; p < ibdev->num_ports; ++p)
1800 if (ibdev->counters[p] != -1)
1801 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1802 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1803 mlx4_CLOSE_PORT(dev, p);
1805 mlx4_ib_free_eqs(dev, ibdev);
1807 mlx4_uar_free(dev, &ibdev->priv_uar);
1808 mlx4_pd_free(dev, ibdev->priv_pdn);
1809 ib_dealloc_device(&ibdev->ib_dev);
1812 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1814 struct mlx4_ib_demux_work **dm = NULL;
1815 struct mlx4_dev *dev = ibdev->dev;
1817 unsigned long flags;
1819 if (!mlx4_is_master(dev))
1822 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
1824 pr_err("failed to allocate memory for tunneling qp update\n");
1828 for (i = 0; i < dev->caps.num_ports; i++) {
1829 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
1831 pr_err("failed to allocate memory for tunneling qp update work struct\n");
1832 for (i = 0; i < dev->caps.num_ports; i++) {
1839 /* initialize or tear down tunnel QPs for the slave */
1840 for (i = 0; i < dev->caps.num_ports; i++) {
1841 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
1842 dm[i]->port = i + 1;
1843 dm[i]->slave = slave;
1844 dm[i]->do_init = do_init;
1846 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
1847 if (!ibdev->sriov.is_going_down)
1848 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
1849 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1856 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1857 enum mlx4_dev_event event, unsigned long param)
1859 struct ib_event ibev;
1860 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1861 struct mlx4_eqe *eqe = NULL;
1862 struct ib_event_work *ew;
1865 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1866 eqe = (struct mlx4_eqe *)param;
1871 case MLX4_DEV_EVENT_PORT_UP:
1872 if (p > ibdev->num_ports)
1874 if (mlx4_is_master(dev) &&
1875 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
1876 IB_LINK_LAYER_INFINIBAND) {
1877 mlx4_ib_invalidate_all_guid_record(ibdev, p);
1879 ibev.event = IB_EVENT_PORT_ACTIVE;
1882 case MLX4_DEV_EVENT_PORT_DOWN:
1883 if (p > ibdev->num_ports)
1885 ibev.event = IB_EVENT_PORT_ERR;
1888 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1889 ibdev->ib_active = false;
1890 ibev.event = IB_EVENT_DEVICE_FATAL;
1893 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1894 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1896 pr_err("failed to allocate memory for events work\n");
1900 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1901 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1903 /* need to queue only for port owner, which uses GEN_EQE */
1904 if (mlx4_is_master(dev))
1905 queue_work(wq, &ew->work);
1907 handle_port_mgmt_change_event(&ew->work);
1910 case MLX4_DEV_EVENT_SLAVE_INIT:
1911 /* here, p is the slave id */
1912 do_slave_init(ibdev, p, 1);
1915 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
1916 /* here, p is the slave id */
1917 do_slave_init(ibdev, p, 0);
1924 ibev.device = ibdev_ptr;
1925 ibev.element.port_num = (u8) p;
1927 ib_dispatch_event(&ibev);
1930 static struct mlx4_interface mlx4_ib_interface = {
1932 .remove = mlx4_ib_remove,
1933 .event = mlx4_ib_event,
1934 .protocol = MLX4_PROT_IB_IPV6
1937 static int __init mlx4_ib_init(void)
1941 wq = create_singlethread_workqueue("mlx4_ib");
1945 err = mlx4_ib_mcg_init();
1949 err = mlx4_register_interface(&mlx4_ib_interface);
1956 mlx4_ib_mcg_destroy();
1959 destroy_workqueue(wq);
1963 static void __exit mlx4_ib_cleanup(void)
1965 mlx4_unregister_interface(&mlx4_ib_interface);
1966 mlx4_ib_mcg_destroy();
1967 destroy_workqueue(wq);
1970 module_init(mlx4_ib_init);
1971 module_exit(mlx4_ib_cleanup);