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 <net/addrconf.h>
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
49 #include <linux/mlx4/driver.h>
50 #include <linux/mlx4/cmd.h>
51 #include <linux/mlx4/qp.h>
56 #define DRV_NAME MLX4_IB_DRV_NAME
57 #define DRV_VERSION "2.2-1"
58 #define DRV_RELDATE "Feb 2014"
60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
63 MODULE_AUTHOR("Roland Dreier");
64 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
65 MODULE_LICENSE("Dual BSD/GPL");
66 MODULE_VERSION(DRV_VERSION);
68 int mlx4_ib_sm_guid_assign = 1;
69 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
70 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
72 static const char mlx4_ib_version[] =
73 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
74 DRV_VERSION " (" DRV_RELDATE ")\n";
76 struct update_gid_work {
77 struct work_struct work;
78 union ib_gid gids[128];
79 struct mlx4_ib_dev *dev;
83 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
85 static struct workqueue_struct *wq;
87 static void init_query_mad(struct ib_smp *mad)
89 mad->base_version = 1;
90 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
91 mad->class_version = 1;
92 mad->method = IB_MGMT_METHOD_GET;
95 static union ib_gid zgid;
97 static int check_flow_steering_support(struct mlx4_dev *dev)
99 int eth_num_ports = 0;
100 int ib_num_ports = 0;
102 int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
106 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
108 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
110 dmfs &= (!ib_num_ports ||
111 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
113 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
114 if (ib_num_ports && mlx4_is_mfunc(dev)) {
115 pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
122 static int mlx4_ib_query_device(struct ib_device *ibdev,
123 struct ib_device_attr *props)
125 struct mlx4_ib_dev *dev = to_mdev(ibdev);
126 struct ib_smp *in_mad = NULL;
127 struct ib_smp *out_mad = NULL;
130 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
131 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
132 if (!in_mad || !out_mad)
135 init_query_mad(in_mad);
136 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
138 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
139 1, NULL, NULL, in_mad, out_mad);
143 memset(props, 0, sizeof *props);
145 props->fw_ver = dev->dev->caps.fw_ver;
146 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
147 IB_DEVICE_PORT_ACTIVE_EVENT |
148 IB_DEVICE_SYS_IMAGE_GUID |
149 IB_DEVICE_RC_RNR_NAK_GEN |
150 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
151 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
152 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
153 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
154 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
155 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
156 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
157 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
158 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
159 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
160 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
161 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
162 props->device_cap_flags |= IB_DEVICE_UD_TSO;
163 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
164 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
165 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
166 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
167 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
168 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
169 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
170 props->device_cap_flags |= IB_DEVICE_XRC;
171 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
172 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
173 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
174 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
175 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
177 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
178 if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
179 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
182 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
184 props->vendor_part_id = dev->dev->pdev->device;
185 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
186 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
188 props->max_mr_size = ~0ull;
189 props->page_size_cap = dev->dev->caps.page_size_cap;
190 props->max_qp = dev->dev->quotas.qp;
191 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
192 props->max_sge = min(dev->dev->caps.max_sq_sg,
193 dev->dev->caps.max_rq_sg);
194 props->max_cq = dev->dev->quotas.cq;
195 props->max_cqe = dev->dev->caps.max_cqes;
196 props->max_mr = dev->dev->quotas.mpt;
197 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
198 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
199 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
200 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
201 props->max_srq = dev->dev->quotas.srq;
202 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
203 props->max_srq_sge = dev->dev->caps.max_srq_sge;
204 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
205 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
206 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
207 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
208 props->masked_atomic_cap = props->atomic_cap;
209 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
210 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
211 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
212 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
213 props->max_mcast_grp;
214 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
223 static enum rdma_link_layer
224 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
226 struct mlx4_dev *dev = to_mdev(device)->dev;
228 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
229 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
232 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
233 struct ib_port_attr *props, int netw_view)
235 struct ib_smp *in_mad = NULL;
236 struct ib_smp *out_mad = NULL;
237 int ext_active_speed;
238 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
241 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
242 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
243 if (!in_mad || !out_mad)
246 init_query_mad(in_mad);
247 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
248 in_mad->attr_mod = cpu_to_be32(port);
250 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
251 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
253 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
259 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
260 props->lmc = out_mad->data[34] & 0x7;
261 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
262 props->sm_sl = out_mad->data[36] & 0xf;
263 props->state = out_mad->data[32] & 0xf;
264 props->phys_state = out_mad->data[33] >> 4;
265 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
267 props->gid_tbl_len = out_mad->data[50];
269 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
270 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
271 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
272 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
273 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
274 props->active_width = out_mad->data[31] & 0xf;
275 props->active_speed = out_mad->data[35] >> 4;
276 props->max_mtu = out_mad->data[41] & 0xf;
277 props->active_mtu = out_mad->data[36] >> 4;
278 props->subnet_timeout = out_mad->data[51] & 0x1f;
279 props->max_vl_num = out_mad->data[37] >> 4;
280 props->init_type_reply = out_mad->data[41] >> 4;
282 /* Check if extended speeds (EDR/FDR/...) are supported */
283 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
284 ext_active_speed = out_mad->data[62] >> 4;
286 switch (ext_active_speed) {
288 props->active_speed = IB_SPEED_FDR;
291 props->active_speed = IB_SPEED_EDR;
296 /* If reported active speed is QDR, check if is FDR-10 */
297 if (props->active_speed == IB_SPEED_QDR) {
298 init_query_mad(in_mad);
299 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
300 in_mad->attr_mod = cpu_to_be32(port);
302 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
303 NULL, NULL, in_mad, out_mad);
307 /* Checking LinkSpeedActive for FDR-10 */
308 if (out_mad->data[15] & 0x1)
309 props->active_speed = IB_SPEED_FDR10;
312 /* Avoid wrong speed value returned by FW if the IB link is down. */
313 if (props->state == IB_PORT_DOWN)
314 props->active_speed = IB_SPEED_SDR;
322 static u8 state_to_phys_state(enum ib_port_state state)
324 return state == IB_PORT_ACTIVE ? 5 : 3;
327 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
328 struct ib_port_attr *props, int netw_view)
331 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
332 struct mlx4_ib_iboe *iboe = &mdev->iboe;
333 struct net_device *ndev;
335 struct mlx4_cmd_mailbox *mailbox;
338 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
340 return PTR_ERR(mailbox);
342 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
343 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
348 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
349 IB_WIDTH_4X : IB_WIDTH_1X;
350 props->active_speed = IB_SPEED_QDR;
351 props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
352 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
353 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
354 props->pkey_tbl_len = 1;
355 props->max_mtu = IB_MTU_4096;
356 props->max_vl_num = 2;
357 props->state = IB_PORT_DOWN;
358 props->phys_state = state_to_phys_state(props->state);
359 props->active_mtu = IB_MTU_256;
360 spin_lock(&iboe->lock);
361 ndev = iboe->netdevs[port - 1];
365 tmp = iboe_get_mtu(ndev->mtu);
366 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
368 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
369 IB_PORT_ACTIVE : IB_PORT_DOWN;
370 props->phys_state = state_to_phys_state(props->state);
372 spin_unlock(&iboe->lock);
374 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
378 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
379 struct ib_port_attr *props, int netw_view)
383 memset(props, 0, sizeof *props);
385 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
386 ib_link_query_port(ibdev, port, props, netw_view) :
387 eth_link_query_port(ibdev, port, props, netw_view);
392 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
393 struct ib_port_attr *props)
395 /* returns host view */
396 return __mlx4_ib_query_port(ibdev, port, props, 0);
399 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
400 union ib_gid *gid, int netw_view)
402 struct ib_smp *in_mad = NULL;
403 struct ib_smp *out_mad = NULL;
405 struct mlx4_ib_dev *dev = to_mdev(ibdev);
407 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
409 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
410 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
411 if (!in_mad || !out_mad)
414 init_query_mad(in_mad);
415 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
416 in_mad->attr_mod = cpu_to_be32(port);
418 if (mlx4_is_mfunc(dev->dev) && netw_view)
419 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
421 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
425 memcpy(gid->raw, out_mad->data + 8, 8);
427 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
429 /* For any index > 0, return the null guid */
436 init_query_mad(in_mad);
437 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
438 in_mad->attr_mod = cpu_to_be32(index / 8);
440 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
441 NULL, NULL, in_mad, out_mad);
445 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
449 memset(gid->raw + 8, 0, 8);
455 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
458 struct mlx4_ib_dev *dev = to_mdev(ibdev);
460 *gid = dev->iboe.gid_table[port - 1][index];
465 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
468 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
469 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
471 return iboe_query_gid(ibdev, port, index, gid);
474 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
475 u16 *pkey, int netw_view)
477 struct ib_smp *in_mad = NULL;
478 struct ib_smp *out_mad = NULL;
479 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
482 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
483 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
484 if (!in_mad || !out_mad)
487 init_query_mad(in_mad);
488 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
489 in_mad->attr_mod = cpu_to_be32(index / 32);
491 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
492 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
494 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
499 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
507 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
509 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
512 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
513 struct ib_device_modify *props)
515 struct mlx4_cmd_mailbox *mailbox;
518 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
521 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
524 if (mlx4_is_slave(to_mdev(ibdev)->dev))
527 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
528 memcpy(ibdev->node_desc, props->node_desc, 64);
529 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
532 * If possible, pass node desc to FW, so it can generate
533 * a 144 trap. If cmd fails, just ignore.
535 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
539 memcpy(mailbox->buf, props->node_desc, 64);
540 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
541 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
543 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
548 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
551 struct mlx4_cmd_mailbox *mailbox;
554 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
556 return PTR_ERR(mailbox);
558 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
559 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
560 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
562 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
563 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
566 err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
567 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
569 mlx4_free_cmd_mailbox(dev->dev, mailbox);
573 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
574 struct ib_port_modify *props)
576 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
577 u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
578 struct ib_port_attr attr;
582 /* return OK if this is RoCE. CM calls ib_modify_port() regardless
583 * of whether port link layer is ETH or IB. For ETH ports, qkey
584 * violations and port capabilities are not meaningful.
589 mutex_lock(&mdev->cap_mask_mutex);
591 err = mlx4_ib_query_port(ibdev, port, &attr);
595 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
596 ~props->clr_port_cap_mask;
598 err = mlx4_ib_SET_PORT(mdev, port,
599 !!(mask & IB_PORT_RESET_QKEY_CNTR),
603 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
607 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
608 struct ib_udata *udata)
610 struct mlx4_ib_dev *dev = to_mdev(ibdev);
611 struct mlx4_ib_ucontext *context;
612 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
613 struct mlx4_ib_alloc_ucontext_resp resp;
617 return ERR_PTR(-EAGAIN);
619 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
620 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
621 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
622 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
624 resp.dev_caps = dev->dev->caps.userspace_caps;
625 resp.qp_tab_size = dev->dev->caps.num_qps;
626 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
627 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
628 resp.cqe_size = dev->dev->caps.cqe_size;
631 context = kmalloc(sizeof *context, GFP_KERNEL);
633 return ERR_PTR(-ENOMEM);
635 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
641 INIT_LIST_HEAD(&context->db_page_list);
642 mutex_init(&context->db_page_mutex);
644 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
645 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
647 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
650 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
652 return ERR_PTR(-EFAULT);
655 return &context->ibucontext;
658 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
660 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
662 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
668 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
670 struct mlx4_ib_dev *dev = to_mdev(context->device);
672 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
675 if (vma->vm_pgoff == 0) {
676 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
678 if (io_remap_pfn_range(vma, vma->vm_start,
679 to_mucontext(context)->uar.pfn,
680 PAGE_SIZE, vma->vm_page_prot))
682 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
683 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
685 if (io_remap_pfn_range(vma, vma->vm_start,
686 to_mucontext(context)->uar.pfn +
687 dev->dev->caps.num_uars,
688 PAGE_SIZE, vma->vm_page_prot))
696 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
697 struct ib_ucontext *context,
698 struct ib_udata *udata)
700 struct mlx4_ib_pd *pd;
703 pd = kmalloc(sizeof *pd, GFP_KERNEL);
705 return ERR_PTR(-ENOMEM);
707 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
714 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
715 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
717 return ERR_PTR(-EFAULT);
723 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
725 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
731 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
732 struct ib_ucontext *context,
733 struct ib_udata *udata)
735 struct mlx4_ib_xrcd *xrcd;
738 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
739 return ERR_PTR(-ENOSYS);
741 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
743 return ERR_PTR(-ENOMEM);
745 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
749 xrcd->pd = ib_alloc_pd(ibdev);
750 if (IS_ERR(xrcd->pd)) {
751 err = PTR_ERR(xrcd->pd);
755 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
756 if (IS_ERR(xrcd->cq)) {
757 err = PTR_ERR(xrcd->cq);
761 return &xrcd->ibxrcd;
764 ib_dealloc_pd(xrcd->pd);
766 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
772 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
774 ib_destroy_cq(to_mxrcd(xrcd)->cq);
775 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
776 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
782 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
784 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
785 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
786 struct mlx4_ib_gid_entry *ge;
788 ge = kzalloc(sizeof *ge, GFP_KERNEL);
793 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
794 ge->port = mqp->port;
798 mutex_lock(&mqp->mutex);
799 list_add_tail(&ge->list, &mqp->gid_list);
800 mutex_unlock(&mqp->mutex);
805 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
808 struct net_device *ndev;
814 spin_lock(&mdev->iboe.lock);
815 ndev = mdev->iboe.netdevs[mqp->port - 1];
818 spin_unlock(&mdev->iboe.lock);
828 struct mlx4_ib_steering {
829 struct list_head list;
834 static int parse_flow_attr(struct mlx4_dev *dev,
836 union ib_flow_spec *ib_spec,
837 struct _rule_hw *mlx4_spec)
839 enum mlx4_net_trans_rule_id type;
841 switch (ib_spec->type) {
842 case IB_FLOW_SPEC_ETH:
843 type = MLX4_NET_TRANS_RULE_ID_ETH;
844 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
846 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
848 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
849 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
851 case IB_FLOW_SPEC_IB:
852 type = MLX4_NET_TRANS_RULE_ID_IB;
853 mlx4_spec->ib.l3_qpn =
855 mlx4_spec->ib.qpn_mask =
856 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
860 case IB_FLOW_SPEC_IPV4:
861 type = MLX4_NET_TRANS_RULE_ID_IPV4;
862 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
863 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
864 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
865 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
868 case IB_FLOW_SPEC_TCP:
869 case IB_FLOW_SPEC_UDP:
870 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
871 MLX4_NET_TRANS_RULE_ID_TCP :
872 MLX4_NET_TRANS_RULE_ID_UDP;
873 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
874 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
875 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
876 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
882 if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
883 mlx4_hw_rule_sz(dev, type) < 0)
885 mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
886 mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
887 return mlx4_hw_rule_sz(dev, type);
890 struct default_rules {
891 __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
892 __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
893 __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
896 static const struct default_rules default_table[] = {
898 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
899 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
900 .rules_create_list = {IB_FLOW_SPEC_IB},
901 .link_layer = IB_LINK_LAYER_INFINIBAND
905 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
906 struct ib_flow_attr *flow_attr)
910 const struct default_rules *pdefault_rules = default_table;
911 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
913 for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
915 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
916 memset(&field_types, 0, sizeof(field_types));
918 if (link_layer != pdefault_rules->link_layer)
921 ib_flow = flow_attr + 1;
922 /* we assume the specs are sorted */
923 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
924 j < flow_attr->num_of_specs; k++) {
925 union ib_flow_spec *current_flow =
926 (union ib_flow_spec *)ib_flow;
928 /* same layer but different type */
929 if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
930 (pdefault_rules->mandatory_fields[k] &
931 IB_FLOW_SPEC_LAYER_MASK)) &&
932 (current_flow->type !=
933 pdefault_rules->mandatory_fields[k]))
936 /* same layer, try match next one */
937 if (current_flow->type ==
938 pdefault_rules->mandatory_fields[k]) {
941 ((union ib_flow_spec *)ib_flow)->size;
945 ib_flow = flow_attr + 1;
946 for (j = 0; j < flow_attr->num_of_specs;
947 j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
948 for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
949 /* same layer and same type */
950 if (((union ib_flow_spec *)ib_flow)->type ==
951 pdefault_rules->mandatory_not_fields[k])
960 static int __mlx4_ib_create_default_rules(
961 struct mlx4_ib_dev *mdev,
963 const struct default_rules *pdefault_rules,
964 struct _rule_hw *mlx4_spec) {
968 for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
969 sizeof(pdefault_rules->rules_create_list[0]); i++) {
971 union ib_flow_spec ib_spec;
972 switch (pdefault_rules->rules_create_list[i]) {
976 case IB_FLOW_SPEC_IB:
977 ib_spec.type = IB_FLOW_SPEC_IB;
978 ib_spec.size = sizeof(struct ib_flow_spec_ib);
985 /* We must put empty rule, qpn is being ignored */
986 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
989 pr_info("invalid parsing\n");
993 mlx4_spec = (void *)mlx4_spec + ret;
999 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1001 enum mlx4_net_trans_promisc_mode flow_type,
1007 struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1008 struct mlx4_cmd_mailbox *mailbox;
1009 struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1012 static const u16 __mlx4_domain[] = {
1013 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1014 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1015 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1016 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1019 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1020 pr_err("Invalid priority value %d\n", flow_attr->priority);
1024 if (domain >= IB_FLOW_DOMAIN_NUM) {
1025 pr_err("Invalid domain value %d\n", domain);
1029 if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1032 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1033 if (IS_ERR(mailbox))
1034 return PTR_ERR(mailbox);
1035 ctrl = mailbox->buf;
1037 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1038 flow_attr->priority);
1039 ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1040 ctrl->port = flow_attr->port;
1041 ctrl->qpn = cpu_to_be32(qp->qp_num);
1043 ib_flow = flow_attr + 1;
1044 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1045 /* Add default flows */
1046 default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1047 if (default_flow >= 0) {
1048 ret = __mlx4_ib_create_default_rules(
1049 mdev, qp, default_table + default_flow,
1050 mailbox->buf + size);
1052 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1057 for (i = 0; i < flow_attr->num_of_specs; i++) {
1058 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1059 mailbox->buf + size);
1061 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1064 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1068 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1069 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1072 pr_err("mcg table is full. Fail to register network rule.\n");
1073 else if (ret == -ENXIO)
1074 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1076 pr_err("Invalid argumant. Fail to register network rule.\n");
1078 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1082 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1085 err = mlx4_cmd(dev, reg_id, 0, 0,
1086 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1089 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1094 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1095 struct ib_flow_attr *flow_attr,
1099 struct mlx4_ib_flow *mflow;
1100 enum mlx4_net_trans_promisc_mode type[2];
1102 memset(type, 0, sizeof(type));
1104 mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1110 switch (flow_attr->type) {
1111 case IB_FLOW_ATTR_NORMAL:
1112 type[0] = MLX4_FS_REGULAR;
1115 case IB_FLOW_ATTR_ALL_DEFAULT:
1116 type[0] = MLX4_FS_ALL_DEFAULT;
1119 case IB_FLOW_ATTR_MC_DEFAULT:
1120 type[0] = MLX4_FS_MC_DEFAULT;
1123 case IB_FLOW_ATTR_SNIFFER:
1124 type[0] = MLX4_FS_UC_SNIFFER;
1125 type[1] = MLX4_FS_MC_SNIFFER;
1133 while (i < ARRAY_SIZE(type) && type[i]) {
1134 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1141 return &mflow->ibflow;
1145 return ERR_PTR(err);
1148 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1152 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1153 struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1155 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1156 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1166 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1169 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1170 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1172 struct mlx4_ib_steering *ib_steering = NULL;
1173 enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1174 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1176 if (mdev->dev->caps.steering_mode ==
1177 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1178 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1183 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1185 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1190 err = add_gid_entry(ibqp, gid);
1195 memcpy(ib_steering->gid.raw, gid->raw, 16);
1196 ib_steering->reg_id = reg_id;
1197 mutex_lock(&mqp->mutex);
1198 list_add(&ib_steering->list, &mqp->steering_rules);
1199 mutex_unlock(&mqp->mutex);
1204 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1212 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1214 struct mlx4_ib_gid_entry *ge;
1215 struct mlx4_ib_gid_entry *tmp;
1216 struct mlx4_ib_gid_entry *ret = NULL;
1218 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1219 if (!memcmp(raw, ge->gid.raw, 16)) {
1228 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1231 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1232 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1233 struct net_device *ndev;
1234 struct mlx4_ib_gid_entry *ge;
1236 enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1237 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1239 if (mdev->dev->caps.steering_mode ==
1240 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1241 struct mlx4_ib_steering *ib_steering;
1243 mutex_lock(&mqp->mutex);
1244 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1245 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1246 list_del(&ib_steering->list);
1250 mutex_unlock(&mqp->mutex);
1251 if (&ib_steering->list == &mqp->steering_rules) {
1252 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1255 reg_id = ib_steering->reg_id;
1259 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1264 mutex_lock(&mqp->mutex);
1265 ge = find_gid_entry(mqp, gid->raw);
1267 spin_lock(&mdev->iboe.lock);
1268 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1271 spin_unlock(&mdev->iboe.lock);
1274 list_del(&ge->list);
1277 pr_warn("could not find mgid entry\n");
1279 mutex_unlock(&mqp->mutex);
1284 static int init_node_data(struct mlx4_ib_dev *dev)
1286 struct ib_smp *in_mad = NULL;
1287 struct ib_smp *out_mad = NULL;
1288 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1291 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1292 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1293 if (!in_mad || !out_mad)
1296 init_query_mad(in_mad);
1297 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1298 if (mlx4_is_master(dev->dev))
1299 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1301 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1305 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1307 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1309 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1313 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1314 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1322 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1325 struct mlx4_ib_dev *dev =
1326 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1327 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1330 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1333 struct mlx4_ib_dev *dev =
1334 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1335 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1336 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1337 (int) dev->dev->caps.fw_ver & 0xffff);
1340 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1343 struct mlx4_ib_dev *dev =
1344 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1345 return sprintf(buf, "%x\n", dev->dev->rev_id);
1348 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1351 struct mlx4_ib_dev *dev =
1352 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1353 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1354 dev->dev->board_id);
1357 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1358 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1359 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1360 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1362 static struct device_attribute *mlx4_class_attributes[] = {
1369 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1370 struct net_device *dev)
1372 memcpy(eui, dev->dev_addr, 3);
1373 memcpy(eui + 5, dev->dev_addr + 3, 3);
1374 if (vlan_id < 0x1000) {
1375 eui[3] = vlan_id >> 8;
1376 eui[4] = vlan_id & 0xff;
1384 static void update_gids_task(struct work_struct *work)
1386 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1387 struct mlx4_cmd_mailbox *mailbox;
1390 struct mlx4_dev *dev = gw->dev->dev;
1392 mailbox = mlx4_alloc_cmd_mailbox(dev);
1393 if (IS_ERR(mailbox)) {
1394 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1398 gids = mailbox->buf;
1399 memcpy(gids, gw->gids, sizeof gw->gids);
1401 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1402 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1405 pr_warn("set port command failed\n");
1407 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1409 mlx4_free_cmd_mailbox(dev, mailbox);
1413 static void reset_gids_task(struct work_struct *work)
1415 struct update_gid_work *gw =
1416 container_of(work, struct update_gid_work, work);
1417 struct mlx4_cmd_mailbox *mailbox;
1420 struct mlx4_dev *dev = gw->dev->dev;
1422 mailbox = mlx4_alloc_cmd_mailbox(dev);
1423 if (IS_ERR(mailbox)) {
1424 pr_warn("reset gid table failed\n");
1428 gids = mailbox->buf;
1429 memcpy(gids, gw->gids, sizeof(gw->gids));
1431 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1432 IB_LINK_LAYER_ETHERNET) {
1433 err = mlx4_cmd(dev, mailbox->dma,
1434 MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1435 1, MLX4_CMD_SET_PORT,
1436 MLX4_CMD_TIME_CLASS_B,
1439 pr_warn(KERN_WARNING
1440 "set port %d command failed\n", gw->port);
1443 mlx4_free_cmd_mailbox(dev, mailbox);
1448 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1449 union ib_gid *gid, int clear,
1452 struct update_gid_work *work;
1454 int need_update = 0;
1462 max_gids = dev->dev->caps.gid_table_len[port];
1463 for (i = 1; i < max_gids; ++i) {
1464 if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1471 dev->iboe.gid_table[port - 1][found] =
1480 !memcmp(&dev->iboe.gid_table[port - 1][i],
1481 &zgid, sizeof(*gid)))
1487 if (found == -1 && !clear && free >= 0) {
1488 dev->iboe.gid_table[port - 1][free] = *gid;
1495 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1499 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1500 INIT_WORK(&work->work, update_gids_task);
1503 queue_work(wq, &work->work);
1508 static void mlx4_make_default_gid(struct net_device *dev, union ib_gid *gid)
1510 gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1511 mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1515 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1517 struct update_gid_work *work;
1519 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1523 memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1524 memset(work->gids, 0, sizeof(work->gids));
1525 INIT_WORK(&work->work, reset_gids_task);
1528 queue_work(wq, &work->work);
1532 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1533 struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1535 struct mlx4_ib_iboe *iboe;
1537 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1538 rdma_vlan_dev_real_dev(event_netdev) :
1540 union ib_gid default_gid;
1542 mlx4_make_default_gid(real_dev, &default_gid);
1544 if (!memcmp(gid, &default_gid, sizeof(*gid)))
1547 if (event != NETDEV_DOWN && event != NETDEV_UP)
1550 if ((real_dev != event_netdev) &&
1551 (event == NETDEV_DOWN) &&
1552 rdma_link_local_addr((struct in6_addr *)gid))
1555 iboe = &ibdev->iboe;
1556 spin_lock(&iboe->lock);
1558 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1559 if ((netif_is_bond_master(real_dev) &&
1560 (real_dev == iboe->masters[port - 1])) ||
1561 (!netif_is_bond_master(real_dev) &&
1562 (real_dev == iboe->netdevs[port - 1])))
1563 update_gid_table(ibdev, port, gid,
1564 event == NETDEV_DOWN, 0);
1566 spin_unlock(&iboe->lock);
1571 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1572 struct mlx4_ib_dev *ibdev)
1575 struct mlx4_ib_iboe *iboe;
1576 struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1577 rdma_vlan_dev_real_dev(dev) : dev;
1579 iboe = &ibdev->iboe;
1581 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1582 if ((netif_is_bond_master(real_dev) &&
1583 (real_dev == iboe->masters[port - 1])) ||
1584 (!netif_is_bond_master(real_dev) &&
1585 (real_dev == iboe->netdevs[port - 1])))
1588 if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1594 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1597 struct mlx4_ib_dev *ibdev;
1598 struct in_ifaddr *ifa = ptr;
1600 struct net_device *event_netdev = ifa->ifa_dev->dev;
1602 ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1604 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1606 mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1610 #if IS_ENABLED(CONFIG_IPV6)
1611 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1614 struct mlx4_ib_dev *ibdev;
1615 struct inet6_ifaddr *ifa = ptr;
1616 union ib_gid *gid = (union ib_gid *)&ifa->addr;
1617 struct net_device *event_netdev = ifa->idev->dev;
1619 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1621 mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1626 #define MLX4_IB_INVALID_MAC ((u64)-1)
1627 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1628 struct net_device *dev,
1632 u64 release_mac = MLX4_IB_INVALID_MAC;
1633 struct mlx4_ib_qp *qp;
1635 read_lock(&dev_base_lock);
1636 new_smac = mlx4_mac_to_u64(dev->dev_addr);
1637 read_unlock(&dev_base_lock);
1639 mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1640 qp = ibdev->qp1_proxy[port - 1];
1643 u64 old_smac = qp->pri.smac;
1644 struct mlx4_update_qp_params update_params;
1646 if (new_smac == old_smac)
1649 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1651 if (new_smac_index < 0)
1654 update_params.smac_index = new_smac_index;
1655 if (mlx4_update_qp(ibdev->dev, &qp->mqp, MLX4_UPDATE_QP_SMAC,
1657 release_mac = new_smac;
1661 qp->pri.smac = new_smac;
1662 qp->pri.smac_index = new_smac_index;
1664 release_mac = old_smac;
1668 mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1669 if (release_mac != MLX4_IB_INVALID_MAC)
1670 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1673 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1674 struct mlx4_ib_dev *ibdev, u8 port)
1676 struct in_device *in_dev;
1677 #if IS_ENABLED(CONFIG_IPV6)
1678 struct inet6_dev *in6_dev;
1680 struct inet6_ifaddr *ifp;
1685 if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1689 in_dev = in_dev_get(dev);
1692 /*ifa->ifa_address;*/
1693 ipv6_addr_set_v4mapped(ifa->ifa_address,
1694 (struct in6_addr *)&gid);
1695 update_gid_table(ibdev, port, &gid, 0, 0);
1700 #if IS_ENABLED(CONFIG_IPV6)
1702 in6_dev = in6_dev_get(dev);
1704 read_lock_bh(&in6_dev->lock);
1705 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1706 pgid = (union ib_gid *)&ifp->addr;
1707 update_gid_table(ibdev, port, pgid, 0, 0);
1709 read_unlock_bh(&in6_dev->lock);
1710 in6_dev_put(in6_dev);
1715 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1716 struct net_device *dev, u8 port)
1719 mlx4_make_default_gid(dev, &gid);
1720 update_gid_table(ibdev, port, &gid, 0, 1);
1723 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1725 struct net_device *dev;
1726 struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1729 for (i = 1; i <= ibdev->num_ports; ++i)
1730 if (reset_gid_table(ibdev, i))
1733 read_lock(&dev_base_lock);
1734 spin_lock(&iboe->lock);
1736 for_each_netdev(&init_net, dev) {
1737 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1739 mlx4_ib_get_dev_addr(dev, ibdev, port);
1742 spin_unlock(&iboe->lock);
1743 read_unlock(&dev_base_lock);
1748 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1749 struct net_device *dev,
1750 unsigned long event)
1753 struct mlx4_ib_iboe *iboe;
1754 int update_qps_port = -1;
1757 iboe = &ibdev->iboe;
1759 spin_lock(&iboe->lock);
1760 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1761 enum ib_port_state port_state = IB_PORT_NOP;
1762 struct net_device *old_master = iboe->masters[port - 1];
1763 struct net_device *curr_netdev;
1764 struct net_device *curr_master;
1766 iboe->netdevs[port - 1] =
1767 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1768 if (iboe->netdevs[port - 1])
1769 mlx4_ib_set_default_gid(ibdev,
1770 iboe->netdevs[port - 1], port);
1771 curr_netdev = iboe->netdevs[port - 1];
1773 if (iboe->netdevs[port - 1] &&
1774 netif_is_bond_slave(iboe->netdevs[port - 1])) {
1775 iboe->masters[port - 1] = netdev_master_upper_dev_get(
1776 iboe->netdevs[port - 1]);
1778 iboe->masters[port - 1] = NULL;
1780 curr_master = iboe->masters[port - 1];
1782 if (dev == iboe->netdevs[port - 1] &&
1783 (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1784 event == NETDEV_UP || event == NETDEV_CHANGE))
1785 update_qps_port = port;
1788 port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1789 IB_PORT_ACTIVE : IB_PORT_DOWN;
1790 mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1792 reset_gid_table(ibdev, port);
1794 /* if using bonding/team and a slave port is down, we don't the bond IP
1795 * based gids in the table since flows that select port by gid may get
1798 if (curr_master && (port_state == IB_PORT_DOWN)) {
1799 reset_gid_table(ibdev, port);
1800 mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1802 /* if bonding is used it is possible that we add it to masters
1803 * only after IP address is assigned to the net bonding
1806 if (curr_master && (old_master != curr_master)) {
1807 reset_gid_table(ibdev, port);
1808 mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1809 mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1812 if (!curr_master && (old_master != curr_master)) {
1813 reset_gid_table(ibdev, port);
1814 mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1815 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1819 spin_unlock(&iboe->lock);
1821 if (update_qps_port > 0)
1822 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1825 static int mlx4_ib_netdev_event(struct notifier_block *this,
1826 unsigned long event, void *ptr)
1828 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1829 struct mlx4_ib_dev *ibdev;
1831 if (!net_eq(dev_net(dev), &init_net))
1834 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1835 mlx4_ib_scan_netdevs(ibdev, dev, event);
1840 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1846 if (mlx4_is_master(ibdev->dev)) {
1847 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1848 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1850 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1852 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1853 /* master has the identity virt2phys pkey mapping */
1854 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1855 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1856 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1857 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1861 /* initialize pkey cache */
1862 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1864 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1866 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1872 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1875 int eq_per_port = 0;
1880 /* Legacy mode or comp_pool is not large enough */
1881 if (dev->caps.comp_pool == 0 ||
1882 dev->caps.num_ports > dev->caps.comp_pool)
1885 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1886 dev->caps.num_ports);
1890 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1891 added_eqs += eq_per_port;
1893 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1895 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1896 if (!ibdev->eq_table)
1899 ibdev->eq_added = added_eqs;
1902 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1903 for (j = 0; j < eq_per_port; j++) {
1904 snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
1905 i, j, dev->pdev->bus->name);
1906 /* Set IRQ for specific name (per ring) */
1907 if (mlx4_assign_eq(dev, name, NULL,
1908 &ibdev->eq_table[eq])) {
1909 /* Use legacy (same as mlx4_en driver) */
1910 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1911 ibdev->eq_table[eq] =
1912 (eq % dev->caps.num_comp_vectors);
1918 /* Fill the reset of the vector with legacy EQ */
1919 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1920 ibdev->eq_table[eq++] = i;
1922 /* Advertise the new number of EQs to clients */
1923 ibdev->ib_dev.num_comp_vectors = total_eqs;
1926 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1930 /* no additional eqs were added */
1931 if (!ibdev->eq_table)
1934 /* Reset the advertised EQ number */
1935 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1937 /* Free only the added eqs */
1938 for (i = 0; i < ibdev->eq_added; i++) {
1939 /* Don't free legacy eqs if used */
1940 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1942 mlx4_release_eq(dev, ibdev->eq_table[i]);
1945 kfree(ibdev->eq_table);
1948 static void *mlx4_ib_add(struct mlx4_dev *dev)
1950 struct mlx4_ib_dev *ibdev;
1954 struct mlx4_ib_iboe *iboe;
1955 int ib_num_ports = 0;
1957 pr_info_once("%s", mlx4_ib_version);
1960 mlx4_foreach_ib_transport_port(i, dev)
1963 /* No point in registering a device with no ports... */
1967 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1969 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1973 iboe = &ibdev->iboe;
1975 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1978 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1981 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1983 if (!ibdev->uar_map)
1985 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1989 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1990 ibdev->ib_dev.owner = THIS_MODULE;
1991 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1992 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1993 ibdev->num_ports = num_ports;
1994 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1995 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1996 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1998 if (dev->caps.userspace_caps)
1999 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2001 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2003 ibdev->ib_dev.uverbs_cmd_mask =
2004 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
2005 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
2006 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
2007 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
2008 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
2009 (1ull << IB_USER_VERBS_CMD_REG_MR) |
2010 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
2011 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2012 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
2013 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
2014 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
2015 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
2016 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
2017 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
2018 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
2019 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
2020 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
2021 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
2022 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
2023 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
2024 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
2025 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
2026 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2028 ibdev->ib_dev.query_device = mlx4_ib_query_device;
2029 ibdev->ib_dev.query_port = mlx4_ib_query_port;
2030 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
2031 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
2032 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
2033 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
2034 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
2035 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
2036 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
2037 ibdev->ib_dev.mmap = mlx4_ib_mmap;
2038 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
2039 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
2040 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
2041 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
2042 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
2043 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
2044 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
2045 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
2046 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
2047 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
2048 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
2049 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
2050 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
2051 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
2052 ibdev->ib_dev.post_send = mlx4_ib_post_send;
2053 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
2054 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
2055 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
2056 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
2057 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
2058 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
2059 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
2060 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
2061 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
2062 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
2063 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2064 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2065 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
2066 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
2067 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
2068 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
2070 if (!mlx4_is_slave(ibdev->dev)) {
2071 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
2072 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
2073 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
2074 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
2077 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2078 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2079 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2080 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2081 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2083 ibdev->ib_dev.uverbs_cmd_mask |=
2084 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2085 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2088 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2089 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2090 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2091 ibdev->ib_dev.uverbs_cmd_mask |=
2092 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2093 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2096 if (check_flow_steering_support(dev)) {
2097 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2098 ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
2099 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
2101 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2102 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2103 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2106 mlx4_ib_alloc_eqs(dev, ibdev);
2108 spin_lock_init(&iboe->lock);
2110 if (init_node_data(ibdev))
2113 for (i = 0; i < ibdev->num_ports; ++i) {
2114 mutex_init(&ibdev->qp1_proxy_lock[i]);
2115 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2116 IB_LINK_LAYER_ETHERNET) {
2117 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2119 ibdev->counters[i] = -1;
2121 ibdev->counters[i] = -1;
2125 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2128 spin_lock_init(&ibdev->sm_lock);
2129 mutex_init(&ibdev->cap_mask_mutex);
2131 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2133 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2134 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2135 MLX4_IB_UC_STEER_QPN_ALIGN,
2136 &ibdev->steer_qpn_base);
2140 ibdev->ib_uc_qpns_bitmap =
2141 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2144 if (!ibdev->ib_uc_qpns_bitmap) {
2145 dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2146 goto err_steer_qp_release;
2149 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2151 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2152 dev, ibdev->steer_qpn_base,
2153 ibdev->steer_qpn_base +
2154 ibdev->steer_qpn_count - 1);
2156 goto err_steer_free_bitmap;
2159 if (ib_register_device(&ibdev->ib_dev, NULL))
2160 goto err_steer_free_bitmap;
2162 if (mlx4_ib_mad_init(ibdev))
2165 if (mlx4_ib_init_sriov(ibdev))
2168 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2169 if (!iboe->nb.notifier_call) {
2170 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2171 err = register_netdevice_notifier(&iboe->nb);
2173 iboe->nb.notifier_call = NULL;
2177 if (!iboe->nb_inet.notifier_call) {
2178 iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2179 err = register_inetaddr_notifier(&iboe->nb_inet);
2181 iboe->nb_inet.notifier_call = NULL;
2185 #if IS_ENABLED(CONFIG_IPV6)
2186 if (!iboe->nb_inet6.notifier_call) {
2187 iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2188 err = register_inet6addr_notifier(&iboe->nb_inet6);
2190 iboe->nb_inet6.notifier_call = NULL;
2195 for (i = 1 ; i <= ibdev->num_ports ; ++i)
2196 reset_gid_table(ibdev, i);
2198 mlx4_ib_scan_netdevs(ibdev, NULL, 0);
2200 mlx4_ib_init_gid_table(ibdev);
2203 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2204 if (device_create_file(&ibdev->ib_dev.dev,
2205 mlx4_class_attributes[j]))
2209 ibdev->ib_active = true;
2211 if (mlx4_is_mfunc(ibdev->dev))
2214 /* create paravirt contexts for any VFs which are active */
2215 if (mlx4_is_master(ibdev->dev)) {
2216 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2217 if (j == mlx4_master_func_num(ibdev->dev))
2219 if (mlx4_is_slave_active(ibdev->dev, j))
2220 do_slave_init(ibdev, j, 1);
2226 if (ibdev->iboe.nb.notifier_call) {
2227 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2228 pr_warn("failure unregistering notifier\n");
2229 ibdev->iboe.nb.notifier_call = NULL;
2231 if (ibdev->iboe.nb_inet.notifier_call) {
2232 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2233 pr_warn("failure unregistering notifier\n");
2234 ibdev->iboe.nb_inet.notifier_call = NULL;
2236 #if IS_ENABLED(CONFIG_IPV6)
2237 if (ibdev->iboe.nb_inet6.notifier_call) {
2238 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2239 pr_warn("failure unregistering notifier\n");
2240 ibdev->iboe.nb_inet6.notifier_call = NULL;
2243 flush_workqueue(wq);
2245 mlx4_ib_close_sriov(ibdev);
2248 mlx4_ib_mad_cleanup(ibdev);
2251 ib_unregister_device(&ibdev->ib_dev);
2253 err_steer_free_bitmap:
2254 kfree(ibdev->ib_uc_qpns_bitmap);
2256 err_steer_qp_release:
2257 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2258 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2259 ibdev->steer_qpn_count);
2262 if (ibdev->counters[i - 1] != -1)
2263 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2266 iounmap(ibdev->uar_map);
2269 mlx4_uar_free(dev, &ibdev->priv_uar);
2272 mlx4_pd_free(dev, ibdev->priv_pdn);
2275 ib_dealloc_device(&ibdev->ib_dev);
2280 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2284 WARN_ON(!dev->ib_uc_qpns_bitmap);
2286 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2287 dev->steer_qpn_count,
2288 get_count_order(count));
2292 *qpn = dev->steer_qpn_base + offset;
2296 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2299 dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2302 BUG_ON(qpn < dev->steer_qpn_base);
2304 bitmap_release_region(dev->ib_uc_qpns_bitmap,
2305 qpn - dev->steer_qpn_base,
2306 get_count_order(count));
2309 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2314 struct ib_flow_attr *flow = NULL;
2315 struct ib_flow_spec_ib *ib_spec;
2318 flow_size = sizeof(struct ib_flow_attr) +
2319 sizeof(struct ib_flow_spec_ib);
2320 flow = kzalloc(flow_size, GFP_KERNEL);
2323 flow->port = mqp->port;
2324 flow->num_of_specs = 1;
2325 flow->size = flow_size;
2326 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2327 ib_spec->type = IB_FLOW_SPEC_IB;
2328 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2329 /* Add an empty rule for IB L2 */
2330 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2332 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2337 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2343 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2345 struct mlx4_ib_dev *ibdev = ibdev_ptr;
2348 mlx4_ib_close_sriov(ibdev);
2349 mlx4_ib_mad_cleanup(ibdev);
2350 ib_unregister_device(&ibdev->ib_dev);
2351 if (ibdev->iboe.nb.notifier_call) {
2352 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2353 pr_warn("failure unregistering notifier\n");
2354 ibdev->iboe.nb.notifier_call = NULL;
2357 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2358 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2359 ibdev->steer_qpn_count);
2360 kfree(ibdev->ib_uc_qpns_bitmap);
2363 if (ibdev->iboe.nb_inet.notifier_call) {
2364 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2365 pr_warn("failure unregistering notifier\n");
2366 ibdev->iboe.nb_inet.notifier_call = NULL;
2368 #if IS_ENABLED(CONFIG_IPV6)
2369 if (ibdev->iboe.nb_inet6.notifier_call) {
2370 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2371 pr_warn("failure unregistering notifier\n");
2372 ibdev->iboe.nb_inet6.notifier_call = NULL;
2376 iounmap(ibdev->uar_map);
2377 for (p = 0; p < ibdev->num_ports; ++p)
2378 if (ibdev->counters[p] != -1)
2379 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2380 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2381 mlx4_CLOSE_PORT(dev, p);
2383 mlx4_ib_free_eqs(dev, ibdev);
2385 mlx4_uar_free(dev, &ibdev->priv_uar);
2386 mlx4_pd_free(dev, ibdev->priv_pdn);
2387 ib_dealloc_device(&ibdev->ib_dev);
2390 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2392 struct mlx4_ib_demux_work **dm = NULL;
2393 struct mlx4_dev *dev = ibdev->dev;
2395 unsigned long flags;
2396 struct mlx4_active_ports actv_ports;
2398 unsigned int first_port;
2400 if (!mlx4_is_master(dev))
2403 actv_ports = mlx4_get_active_ports(dev, slave);
2404 ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2405 first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2407 dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2409 pr_err("failed to allocate memory for tunneling qp update\n");
2413 for (i = 0; i < ports; i++) {
2414 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2416 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2417 for (i = 0; i < dev->caps.num_ports; i++) {
2424 /* initialize or tear down tunnel QPs for the slave */
2425 for (i = 0; i < ports; i++) {
2426 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2427 dm[i]->port = first_port + i + 1;
2428 dm[i]->slave = slave;
2429 dm[i]->do_init = do_init;
2431 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2432 if (!ibdev->sriov.is_going_down)
2433 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2434 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2441 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2442 enum mlx4_dev_event event, unsigned long param)
2444 struct ib_event ibev;
2445 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2446 struct mlx4_eqe *eqe = NULL;
2447 struct ib_event_work *ew;
2450 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2451 eqe = (struct mlx4_eqe *)param;
2456 case MLX4_DEV_EVENT_PORT_UP:
2457 if (p > ibdev->num_ports)
2459 if (mlx4_is_master(dev) &&
2460 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2461 IB_LINK_LAYER_INFINIBAND) {
2462 mlx4_ib_invalidate_all_guid_record(ibdev, p);
2464 ibev.event = IB_EVENT_PORT_ACTIVE;
2467 case MLX4_DEV_EVENT_PORT_DOWN:
2468 if (p > ibdev->num_ports)
2470 ibev.event = IB_EVENT_PORT_ERR;
2473 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2474 ibdev->ib_active = false;
2475 ibev.event = IB_EVENT_DEVICE_FATAL;
2478 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2479 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2481 pr_err("failed to allocate memory for events work\n");
2485 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2486 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2488 /* need to queue only for port owner, which uses GEN_EQE */
2489 if (mlx4_is_master(dev))
2490 queue_work(wq, &ew->work);
2492 handle_port_mgmt_change_event(&ew->work);
2495 case MLX4_DEV_EVENT_SLAVE_INIT:
2496 /* here, p is the slave id */
2497 do_slave_init(ibdev, p, 1);
2500 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2501 /* here, p is the slave id */
2502 do_slave_init(ibdev, p, 0);
2509 ibev.device = ibdev_ptr;
2510 ibev.element.port_num = (u8) p;
2512 ib_dispatch_event(&ibev);
2515 static struct mlx4_interface mlx4_ib_interface = {
2517 .remove = mlx4_ib_remove,
2518 .event = mlx4_ib_event,
2519 .protocol = MLX4_PROT_IB_IPV6
2522 static int __init mlx4_ib_init(void)
2526 wq = create_singlethread_workqueue("mlx4_ib");
2530 err = mlx4_ib_mcg_init();
2534 err = mlx4_register_interface(&mlx4_ib_interface);
2541 mlx4_ib_mcg_destroy();
2544 destroy_workqueue(wq);
2548 static void __exit mlx4_ib_cleanup(void)
2550 mlx4_unregister_interface(&mlx4_ib_interface);
2551 mlx4_ib_mcg_destroy();
2552 destroy_workqueue(wq);
2555 module_init(mlx4_ib_init);
2556 module_exit(mlx4_ib_cleanup);