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>
55 #define DRV_NAME MLX4_IB_DRV_NAME
56 #define DRV_VERSION "1.0"
57 #define DRV_RELDATE "April 4, 2008"
59 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
60 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
62 MODULE_AUTHOR("Roland Dreier");
63 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
64 MODULE_LICENSE("Dual BSD/GPL");
65 MODULE_VERSION(DRV_VERSION);
67 int mlx4_ib_sm_guid_assign = 1;
68 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
69 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
71 static const char mlx4_ib_version[] =
72 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
73 DRV_VERSION " (" DRV_RELDATE ")\n";
75 struct update_gid_work {
76 struct work_struct work;
77 union ib_gid gids[128];
78 struct mlx4_ib_dev *dev;
82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
84 static struct workqueue_struct *wq;
86 static void init_query_mad(struct ib_smp *mad)
88 mad->base_version = 1;
89 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
90 mad->class_version = 1;
91 mad->method = IB_MGMT_METHOD_GET;
94 static union ib_gid zgid;
96 static int check_flow_steering_support(struct mlx4_dev *dev)
98 int eth_num_ports = 0;
101 int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
105 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
107 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
109 dmfs &= (!ib_num_ports ||
110 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
112 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
113 if (ib_num_ports && mlx4_is_mfunc(dev)) {
114 pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
121 static int mlx4_ib_query_device(struct ib_device *ibdev,
122 struct ib_device_attr *props)
124 struct mlx4_ib_dev *dev = to_mdev(ibdev);
125 struct ib_smp *in_mad = NULL;
126 struct ib_smp *out_mad = NULL;
129 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
130 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
131 if (!in_mad || !out_mad)
134 init_query_mad(in_mad);
135 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
137 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
138 1, NULL, NULL, in_mad, out_mad);
142 memset(props, 0, sizeof *props);
144 props->fw_ver = dev->dev->caps.fw_ver;
145 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
146 IB_DEVICE_PORT_ACTIVE_EVENT |
147 IB_DEVICE_SYS_IMAGE_GUID |
148 IB_DEVICE_RC_RNR_NAK_GEN |
149 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
150 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
151 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
152 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
153 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
154 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
155 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
156 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
157 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
158 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
159 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
160 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
161 props->device_cap_flags |= IB_DEVICE_UD_TSO;
162 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
163 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
164 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
165 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
166 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
167 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
168 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
169 props->device_cap_flags |= IB_DEVICE_XRC;
170 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
171 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
172 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
173 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
174 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
176 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
177 if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
178 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
181 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
183 props->vendor_part_id = dev->dev->pdev->device;
184 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
185 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
187 props->max_mr_size = ~0ull;
188 props->page_size_cap = dev->dev->caps.page_size_cap;
189 props->max_qp = dev->dev->quotas.qp;
190 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
191 props->max_sge = min(dev->dev->caps.max_sq_sg,
192 dev->dev->caps.max_rq_sg);
193 props->max_cq = dev->dev->quotas.cq;
194 props->max_cqe = dev->dev->caps.max_cqes;
195 props->max_mr = dev->dev->quotas.mpt;
196 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
197 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
198 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
199 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
200 props->max_srq = dev->dev->quotas.srq;
201 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
202 props->max_srq_sge = dev->dev->caps.max_srq_sge;
203 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
204 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
205 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
206 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
207 props->masked_atomic_cap = props->atomic_cap;
208 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
209 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
210 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
211 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
212 props->max_mcast_grp;
213 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
222 static enum rdma_link_layer
223 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
225 struct mlx4_dev *dev = to_mdev(device)->dev;
227 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
228 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
231 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
232 struct ib_port_attr *props, int netw_view)
234 struct ib_smp *in_mad = NULL;
235 struct ib_smp *out_mad = NULL;
236 int ext_active_speed;
237 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
240 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
241 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
242 if (!in_mad || !out_mad)
245 init_query_mad(in_mad);
246 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
247 in_mad->attr_mod = cpu_to_be32(port);
249 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
250 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
252 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
258 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
259 props->lmc = out_mad->data[34] & 0x7;
260 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
261 props->sm_sl = out_mad->data[36] & 0xf;
262 props->state = out_mad->data[32] & 0xf;
263 props->phys_state = out_mad->data[33] >> 4;
264 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
266 props->gid_tbl_len = out_mad->data[50];
268 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
269 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
270 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
271 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
272 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
273 props->active_width = out_mad->data[31] & 0xf;
274 props->active_speed = out_mad->data[35] >> 4;
275 props->max_mtu = out_mad->data[41] & 0xf;
276 props->active_mtu = out_mad->data[36] >> 4;
277 props->subnet_timeout = out_mad->data[51] & 0x1f;
278 props->max_vl_num = out_mad->data[37] >> 4;
279 props->init_type_reply = out_mad->data[41] >> 4;
281 /* Check if extended speeds (EDR/FDR/...) are supported */
282 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
283 ext_active_speed = out_mad->data[62] >> 4;
285 switch (ext_active_speed) {
287 props->active_speed = IB_SPEED_FDR;
290 props->active_speed = IB_SPEED_EDR;
295 /* If reported active speed is QDR, check if is FDR-10 */
296 if (props->active_speed == IB_SPEED_QDR) {
297 init_query_mad(in_mad);
298 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
299 in_mad->attr_mod = cpu_to_be32(port);
301 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
302 NULL, NULL, in_mad, out_mad);
306 /* Checking LinkSpeedActive for FDR-10 */
307 if (out_mad->data[15] & 0x1)
308 props->active_speed = IB_SPEED_FDR10;
311 /* Avoid wrong speed value returned by FW if the IB link is down. */
312 if (props->state == IB_PORT_DOWN)
313 props->active_speed = IB_SPEED_SDR;
321 static u8 state_to_phys_state(enum ib_port_state state)
323 return state == IB_PORT_ACTIVE ? 5 : 3;
326 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
327 struct ib_port_attr *props, int netw_view)
330 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
331 struct mlx4_ib_iboe *iboe = &mdev->iboe;
332 struct net_device *ndev;
334 struct mlx4_cmd_mailbox *mailbox;
337 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
339 return PTR_ERR(mailbox);
341 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
342 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
347 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
348 IB_WIDTH_4X : IB_WIDTH_1X;
349 props->active_speed = IB_SPEED_QDR;
350 props->port_cap_flags = IB_PORT_CM_SUP;
351 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
352 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
353 props->pkey_tbl_len = 1;
354 props->max_mtu = IB_MTU_4096;
355 props->max_vl_num = 2;
356 props->state = IB_PORT_DOWN;
357 props->phys_state = state_to_phys_state(props->state);
358 props->active_mtu = IB_MTU_256;
359 spin_lock(&iboe->lock);
360 ndev = iboe->netdevs[port - 1];
364 tmp = iboe_get_mtu(ndev->mtu);
365 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
367 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
368 IB_PORT_ACTIVE : IB_PORT_DOWN;
369 props->phys_state = state_to_phys_state(props->state);
371 spin_unlock(&iboe->lock);
373 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
377 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
378 struct ib_port_attr *props, int netw_view)
382 memset(props, 0, sizeof *props);
384 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
385 ib_link_query_port(ibdev, port, props, netw_view) :
386 eth_link_query_port(ibdev, port, props, netw_view);
391 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
392 struct ib_port_attr *props)
394 /* returns host view */
395 return __mlx4_ib_query_port(ibdev, port, props, 0);
398 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
399 union ib_gid *gid, int netw_view)
401 struct ib_smp *in_mad = NULL;
402 struct ib_smp *out_mad = NULL;
404 struct mlx4_ib_dev *dev = to_mdev(ibdev);
406 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
408 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
409 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
410 if (!in_mad || !out_mad)
413 init_query_mad(in_mad);
414 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
415 in_mad->attr_mod = cpu_to_be32(port);
417 if (mlx4_is_mfunc(dev->dev) && netw_view)
418 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
420 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
424 memcpy(gid->raw, out_mad->data + 8, 8);
426 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
428 /* For any index > 0, return the null guid */
435 init_query_mad(in_mad);
436 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
437 in_mad->attr_mod = cpu_to_be32(index / 8);
439 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
440 NULL, NULL, in_mad, out_mad);
444 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
448 memset(gid->raw + 8, 0, 8);
454 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
457 struct mlx4_ib_dev *dev = to_mdev(ibdev);
459 *gid = dev->iboe.gid_table[port - 1][index];
464 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
467 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
468 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
470 return iboe_query_gid(ibdev, port, index, gid);
473 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
474 u16 *pkey, int netw_view)
476 struct ib_smp *in_mad = NULL;
477 struct ib_smp *out_mad = NULL;
478 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
481 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
482 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
483 if (!in_mad || !out_mad)
486 init_query_mad(in_mad);
487 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
488 in_mad->attr_mod = cpu_to_be32(index / 32);
490 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
491 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
493 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
498 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
506 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
508 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
511 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
512 struct ib_device_modify *props)
514 struct mlx4_cmd_mailbox *mailbox;
517 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
520 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
523 if (mlx4_is_slave(to_mdev(ibdev)->dev))
526 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
527 memcpy(ibdev->node_desc, props->node_desc, 64);
528 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
531 * If possible, pass node desc to FW, so it can generate
532 * a 144 trap. If cmd fails, just ignore.
534 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
538 memcpy(mailbox->buf, props->node_desc, 64);
539 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
540 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
542 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
547 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
550 struct mlx4_cmd_mailbox *mailbox;
552 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
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, is_eth, MLX4_CMD_SET_PORT,
567 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
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 ib_port_attr attr;
580 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
582 err = mlx4_ib_query_port(ibdev, port, &attr);
586 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
587 ~props->clr_port_cap_mask;
589 err = mlx4_SET_PORT(to_mdev(ibdev), port,
590 !!(mask & IB_PORT_RESET_QKEY_CNTR),
594 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
598 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
599 struct ib_udata *udata)
601 struct mlx4_ib_dev *dev = to_mdev(ibdev);
602 struct mlx4_ib_ucontext *context;
603 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
604 struct mlx4_ib_alloc_ucontext_resp resp;
608 return ERR_PTR(-EAGAIN);
610 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
611 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
612 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
613 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
615 resp.dev_caps = dev->dev->caps.userspace_caps;
616 resp.qp_tab_size = dev->dev->caps.num_qps;
617 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
618 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
619 resp.cqe_size = dev->dev->caps.cqe_size;
622 context = kmalloc(sizeof *context, GFP_KERNEL);
624 return ERR_PTR(-ENOMEM);
626 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
632 INIT_LIST_HEAD(&context->db_page_list);
633 mutex_init(&context->db_page_mutex);
635 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
636 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
638 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
641 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
643 return ERR_PTR(-EFAULT);
646 return &context->ibucontext;
649 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
651 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
653 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
659 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
661 struct mlx4_ib_dev *dev = to_mdev(context->device);
663 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
666 if (vma->vm_pgoff == 0) {
667 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
669 if (io_remap_pfn_range(vma, vma->vm_start,
670 to_mucontext(context)->uar.pfn,
671 PAGE_SIZE, vma->vm_page_prot))
673 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
674 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
676 if (io_remap_pfn_range(vma, vma->vm_start,
677 to_mucontext(context)->uar.pfn +
678 dev->dev->caps.num_uars,
679 PAGE_SIZE, vma->vm_page_prot))
687 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
688 struct ib_ucontext *context,
689 struct ib_udata *udata)
691 struct mlx4_ib_pd *pd;
694 pd = kmalloc(sizeof *pd, GFP_KERNEL);
696 return ERR_PTR(-ENOMEM);
698 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
705 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
706 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
708 return ERR_PTR(-EFAULT);
714 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
716 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
722 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
723 struct ib_ucontext *context,
724 struct ib_udata *udata)
726 struct mlx4_ib_xrcd *xrcd;
729 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
730 return ERR_PTR(-ENOSYS);
732 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
734 return ERR_PTR(-ENOMEM);
736 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
740 xrcd->pd = ib_alloc_pd(ibdev);
741 if (IS_ERR(xrcd->pd)) {
742 err = PTR_ERR(xrcd->pd);
746 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
747 if (IS_ERR(xrcd->cq)) {
748 err = PTR_ERR(xrcd->cq);
752 return &xrcd->ibxrcd;
755 ib_dealloc_pd(xrcd->pd);
757 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
763 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
765 ib_destroy_cq(to_mxrcd(xrcd)->cq);
766 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
767 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
773 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
775 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
776 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
777 struct mlx4_ib_gid_entry *ge;
779 ge = kzalloc(sizeof *ge, GFP_KERNEL);
784 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
785 ge->port = mqp->port;
789 mutex_lock(&mqp->mutex);
790 list_add_tail(&ge->list, &mqp->gid_list);
791 mutex_unlock(&mqp->mutex);
796 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
799 struct net_device *ndev;
805 spin_lock(&mdev->iboe.lock);
806 ndev = mdev->iboe.netdevs[mqp->port - 1];
809 spin_unlock(&mdev->iboe.lock);
819 struct mlx4_ib_steering {
820 struct list_head list;
825 static int parse_flow_attr(struct mlx4_dev *dev,
827 union ib_flow_spec *ib_spec,
828 struct _rule_hw *mlx4_spec)
830 enum mlx4_net_trans_rule_id type;
832 switch (ib_spec->type) {
833 case IB_FLOW_SPEC_ETH:
834 type = MLX4_NET_TRANS_RULE_ID_ETH;
835 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
837 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
839 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
840 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
842 case IB_FLOW_SPEC_IB:
843 type = MLX4_NET_TRANS_RULE_ID_IB;
844 mlx4_spec->ib.l3_qpn =
846 mlx4_spec->ib.qpn_mask =
847 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
851 case IB_FLOW_SPEC_IPV4:
852 type = MLX4_NET_TRANS_RULE_ID_IPV4;
853 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
854 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
855 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
856 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
859 case IB_FLOW_SPEC_TCP:
860 case IB_FLOW_SPEC_UDP:
861 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
862 MLX4_NET_TRANS_RULE_ID_TCP :
863 MLX4_NET_TRANS_RULE_ID_UDP;
864 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
865 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
866 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
867 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
873 if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
874 mlx4_hw_rule_sz(dev, type) < 0)
876 mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
877 mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
878 return mlx4_hw_rule_sz(dev, type);
881 struct default_rules {
882 __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
883 __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
884 __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
887 static const struct default_rules default_table[] = {
889 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
890 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
891 .rules_create_list = {IB_FLOW_SPEC_IB},
892 .link_layer = IB_LINK_LAYER_INFINIBAND
896 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
897 struct ib_flow_attr *flow_attr)
901 const struct default_rules *pdefault_rules = default_table;
902 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
904 for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
906 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
907 memset(&field_types, 0, sizeof(field_types));
909 if (link_layer != pdefault_rules->link_layer)
912 ib_flow = flow_attr + 1;
913 /* we assume the specs are sorted */
914 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
915 j < flow_attr->num_of_specs; k++) {
916 union ib_flow_spec *current_flow =
917 (union ib_flow_spec *)ib_flow;
919 /* same layer but different type */
920 if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
921 (pdefault_rules->mandatory_fields[k] &
922 IB_FLOW_SPEC_LAYER_MASK)) &&
923 (current_flow->type !=
924 pdefault_rules->mandatory_fields[k]))
927 /* same layer, try match next one */
928 if (current_flow->type ==
929 pdefault_rules->mandatory_fields[k]) {
932 ((union ib_flow_spec *)ib_flow)->size;
936 ib_flow = flow_attr + 1;
937 for (j = 0; j < flow_attr->num_of_specs;
938 j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
939 for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
940 /* same layer and same type */
941 if (((union ib_flow_spec *)ib_flow)->type ==
942 pdefault_rules->mandatory_not_fields[k])
951 static int __mlx4_ib_create_default_rules(
952 struct mlx4_ib_dev *mdev,
954 const struct default_rules *pdefault_rules,
955 struct _rule_hw *mlx4_spec) {
959 for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
960 sizeof(pdefault_rules->rules_create_list[0]); i++) {
962 union ib_flow_spec ib_spec;
963 switch (pdefault_rules->rules_create_list[i]) {
967 case IB_FLOW_SPEC_IB:
968 ib_spec.type = IB_FLOW_SPEC_IB;
969 ib_spec.size = sizeof(struct ib_flow_spec_ib);
976 /* We must put empty rule, qpn is being ignored */
977 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
980 pr_info("invalid parsing\n");
984 mlx4_spec = (void *)mlx4_spec + ret;
990 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
992 enum mlx4_net_trans_promisc_mode flow_type,
998 struct mlx4_ib_dev *mdev = to_mdev(qp->device);
999 struct mlx4_cmd_mailbox *mailbox;
1000 struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1003 static const u16 __mlx4_domain[] = {
1004 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1005 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1006 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1007 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1010 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1011 pr_err("Invalid priority value %d\n", flow_attr->priority);
1015 if (domain >= IB_FLOW_DOMAIN_NUM) {
1016 pr_err("Invalid domain value %d\n", domain);
1020 if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1023 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1024 if (IS_ERR(mailbox))
1025 return PTR_ERR(mailbox);
1026 ctrl = mailbox->buf;
1028 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1029 flow_attr->priority);
1030 ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1031 ctrl->port = flow_attr->port;
1032 ctrl->qpn = cpu_to_be32(qp->qp_num);
1034 ib_flow = flow_attr + 1;
1035 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1036 /* Add default flows */
1037 default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1038 if (default_flow >= 0) {
1039 ret = __mlx4_ib_create_default_rules(
1040 mdev, qp, default_table + default_flow,
1041 mailbox->buf + size);
1043 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1048 for (i = 0; i < flow_attr->num_of_specs; i++) {
1049 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1050 mailbox->buf + size);
1052 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1055 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1059 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1060 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1063 pr_err("mcg table is full. Fail to register network rule.\n");
1064 else if (ret == -ENXIO)
1065 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1067 pr_err("Invalid argumant. Fail to register network rule.\n");
1069 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1073 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1076 err = mlx4_cmd(dev, reg_id, 0, 0,
1077 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1080 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1085 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1086 struct ib_flow_attr *flow_attr,
1090 struct mlx4_ib_flow *mflow;
1091 enum mlx4_net_trans_promisc_mode type[2];
1093 memset(type, 0, sizeof(type));
1095 mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1101 switch (flow_attr->type) {
1102 case IB_FLOW_ATTR_NORMAL:
1103 type[0] = MLX4_FS_REGULAR;
1106 case IB_FLOW_ATTR_ALL_DEFAULT:
1107 type[0] = MLX4_FS_ALL_DEFAULT;
1110 case IB_FLOW_ATTR_MC_DEFAULT:
1111 type[0] = MLX4_FS_MC_DEFAULT;
1114 case IB_FLOW_ATTR_SNIFFER:
1115 type[0] = MLX4_FS_UC_SNIFFER;
1116 type[1] = MLX4_FS_MC_SNIFFER;
1124 while (i < ARRAY_SIZE(type) && type[i]) {
1125 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1132 return &mflow->ibflow;
1136 return ERR_PTR(err);
1139 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1143 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1144 struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1146 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1147 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1157 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1160 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1161 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1163 struct mlx4_ib_steering *ib_steering = NULL;
1164 enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1165 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1167 if (mdev->dev->caps.steering_mode ==
1168 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1169 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1174 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1176 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1181 err = add_gid_entry(ibqp, gid);
1186 memcpy(ib_steering->gid.raw, gid->raw, 16);
1187 ib_steering->reg_id = reg_id;
1188 mutex_lock(&mqp->mutex);
1189 list_add(&ib_steering->list, &mqp->steering_rules);
1190 mutex_unlock(&mqp->mutex);
1195 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1203 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1205 struct mlx4_ib_gid_entry *ge;
1206 struct mlx4_ib_gid_entry *tmp;
1207 struct mlx4_ib_gid_entry *ret = NULL;
1209 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1210 if (!memcmp(raw, ge->gid.raw, 16)) {
1219 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1222 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1223 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1224 struct net_device *ndev;
1225 struct mlx4_ib_gid_entry *ge;
1227 enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1228 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1230 if (mdev->dev->caps.steering_mode ==
1231 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1232 struct mlx4_ib_steering *ib_steering;
1234 mutex_lock(&mqp->mutex);
1235 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1236 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1237 list_del(&ib_steering->list);
1241 mutex_unlock(&mqp->mutex);
1242 if (&ib_steering->list == &mqp->steering_rules) {
1243 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1246 reg_id = ib_steering->reg_id;
1250 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1255 mutex_lock(&mqp->mutex);
1256 ge = find_gid_entry(mqp, gid->raw);
1258 spin_lock(&mdev->iboe.lock);
1259 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1262 spin_unlock(&mdev->iboe.lock);
1265 list_del(&ge->list);
1268 pr_warn("could not find mgid entry\n");
1270 mutex_unlock(&mqp->mutex);
1275 static int init_node_data(struct mlx4_ib_dev *dev)
1277 struct ib_smp *in_mad = NULL;
1278 struct ib_smp *out_mad = NULL;
1279 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1282 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1283 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1284 if (!in_mad || !out_mad)
1287 init_query_mad(in_mad);
1288 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1289 if (mlx4_is_master(dev->dev))
1290 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1292 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1296 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1298 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1300 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1304 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1305 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1313 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1316 struct mlx4_ib_dev *dev =
1317 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1318 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1321 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1324 struct mlx4_ib_dev *dev =
1325 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1326 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1327 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1328 (int) dev->dev->caps.fw_ver & 0xffff);
1331 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1334 struct mlx4_ib_dev *dev =
1335 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1336 return sprintf(buf, "%x\n", dev->dev->rev_id);
1339 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1342 struct mlx4_ib_dev *dev =
1343 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1344 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1345 dev->dev->board_id);
1348 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1349 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1350 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1351 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1353 static struct device_attribute *mlx4_class_attributes[] = {
1360 static void update_gids_task(struct work_struct *work)
1362 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1363 struct mlx4_cmd_mailbox *mailbox;
1366 struct mlx4_dev *dev = gw->dev->dev;
1368 mailbox = mlx4_alloc_cmd_mailbox(dev);
1369 if (IS_ERR(mailbox)) {
1370 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1374 gids = mailbox->buf;
1375 memcpy(gids, gw->gids, sizeof gw->gids);
1377 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1378 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1381 pr_warn("set port command failed\n");
1383 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1385 mlx4_free_cmd_mailbox(dev, mailbox);
1389 static void reset_gids_task(struct work_struct *work)
1391 struct update_gid_work *gw =
1392 container_of(work, struct update_gid_work, work);
1393 struct mlx4_cmd_mailbox *mailbox;
1397 struct mlx4_dev *dev = gw->dev->dev;
1399 mailbox = mlx4_alloc_cmd_mailbox(dev);
1400 if (IS_ERR(mailbox)) {
1401 pr_warn("reset gid table failed\n");
1405 gids = mailbox->buf;
1406 memcpy(gids, gw->gids, sizeof(gw->gids));
1408 for (i = 1; i < gw->dev->num_ports + 1; i++) {
1409 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, i) ==
1410 IB_LINK_LAYER_ETHERNET) {
1411 err = mlx4_cmd(dev, mailbox->dma,
1412 MLX4_SET_PORT_GID_TABLE << 8 | i,
1413 1, MLX4_CMD_SET_PORT,
1414 MLX4_CMD_TIME_CLASS_B,
1417 pr_warn(KERN_WARNING
1418 "set port %d command failed\n", i);
1422 mlx4_free_cmd_mailbox(dev, mailbox);
1427 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1428 union ib_gid *gid, int clear)
1430 struct update_gid_work *work;
1432 int need_update = 0;
1437 max_gids = dev->dev->caps.gid_table_len[port];
1438 for (i = 0; i < max_gids; ++i) {
1439 if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1446 dev->iboe.gid_table[port - 1][found] = zgid;
1454 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid,
1460 if (found == -1 && !clear && free >= 0) {
1461 dev->iboe.gid_table[port - 1][free] = *gid;
1468 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1472 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1473 INIT_WORK(&work->work, update_gids_task);
1476 queue_work(wq, &work->work);
1481 static int reset_gid_table(struct mlx4_ib_dev *dev)
1483 struct update_gid_work *work;
1486 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1489 memset(dev->iboe.gid_table, 0, sizeof(dev->iboe.gid_table));
1490 memset(work->gids, 0, sizeof(work->gids));
1491 INIT_WORK(&work->work, reset_gids_task);
1493 queue_work(wq, &work->work);
1497 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1498 struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1500 struct mlx4_ib_iboe *iboe;
1502 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1503 rdma_vlan_dev_real_dev(event_netdev) :
1506 if (event != NETDEV_DOWN && event != NETDEV_UP)
1509 if ((real_dev != event_netdev) &&
1510 (event == NETDEV_DOWN) &&
1511 rdma_link_local_addr((struct in6_addr *)gid))
1514 iboe = &ibdev->iboe;
1515 spin_lock(&iboe->lock);
1517 for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1518 if ((netif_is_bond_master(real_dev) &&
1519 (real_dev == iboe->masters[port - 1])) ||
1520 (!netif_is_bond_master(real_dev) &&
1521 (real_dev == iboe->netdevs[port - 1])))
1522 update_gid_table(ibdev, port, gid,
1523 event == NETDEV_DOWN);
1525 spin_unlock(&iboe->lock);
1530 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1531 struct mlx4_ib_dev *ibdev)
1534 struct mlx4_ib_iboe *iboe;
1535 struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1536 rdma_vlan_dev_real_dev(dev) : dev;
1538 iboe = &ibdev->iboe;
1539 spin_lock(&iboe->lock);
1541 for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1542 if ((netif_is_bond_master(real_dev) &&
1543 (real_dev == iboe->masters[port - 1])) ||
1544 (!netif_is_bond_master(real_dev) &&
1545 (real_dev == iboe->netdevs[port - 1])))
1548 spin_unlock(&iboe->lock);
1550 if ((port == 0) || (port > MLX4_MAX_PORTS))
1556 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1559 struct mlx4_ib_dev *ibdev;
1560 struct in_ifaddr *ifa = ptr;
1562 struct net_device *event_netdev = ifa->ifa_dev->dev;
1564 ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1566 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1568 mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1572 #if IS_ENABLED(CONFIG_IPV6)
1573 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1576 struct mlx4_ib_dev *ibdev;
1577 struct inet6_ifaddr *ifa = ptr;
1578 union ib_gid *gid = (union ib_gid *)&ifa->addr;
1579 struct net_device *event_netdev = ifa->idev->dev;
1581 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1583 mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1588 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1589 struct mlx4_ib_dev *ibdev, u8 port)
1591 struct in_device *in_dev;
1592 #if IS_ENABLED(CONFIG_IPV6)
1593 struct inet6_dev *in6_dev;
1595 struct inet6_ifaddr *ifp;
1600 if ((port == 0) || (port > MLX4_MAX_PORTS))
1604 in_dev = in_dev_get(dev);
1607 /*ifa->ifa_address;*/
1608 ipv6_addr_set_v4mapped(ifa->ifa_address,
1609 (struct in6_addr *)&gid);
1610 update_gid_table(ibdev, port, &gid, 0);
1615 #if IS_ENABLED(CONFIG_IPV6)
1617 in6_dev = in6_dev_get(dev);
1619 read_lock_bh(&in6_dev->lock);
1620 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1621 pgid = (union ib_gid *)&ifp->addr;
1622 update_gid_table(ibdev, port, pgid, 0);
1624 read_unlock_bh(&in6_dev->lock);
1625 in6_dev_put(in6_dev);
1630 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1632 struct net_device *dev;
1634 if (reset_gid_table(ibdev))
1637 read_lock(&dev_base_lock);
1639 for_each_netdev(&init_net, dev) {
1640 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1642 mlx4_ib_get_dev_addr(dev, ibdev, port);
1645 read_unlock(&dev_base_lock);
1650 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev)
1652 struct mlx4_ib_iboe *iboe;
1655 iboe = &ibdev->iboe;
1657 spin_lock(&iboe->lock);
1658 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1659 struct net_device *old_master = iboe->masters[port - 1];
1660 struct net_device *curr_master;
1661 iboe->netdevs[port - 1] =
1662 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1664 if (iboe->netdevs[port - 1] &&
1665 netif_is_bond_slave(iboe->netdevs[port - 1])) {
1667 iboe->masters[port - 1] = netdev_master_upper_dev_get(
1668 iboe->netdevs[port - 1]);
1671 curr_master = iboe->masters[port - 1];
1673 /* if bonding is used it is possible that we add it to masters
1674 only after IP address is assigned to the net bonding
1676 if (curr_master && (old_master != curr_master))
1677 mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1680 spin_unlock(&iboe->lock);
1683 static int mlx4_ib_netdev_event(struct notifier_block *this,
1684 unsigned long event, void *ptr)
1686 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1687 struct mlx4_ib_dev *ibdev;
1689 if (!net_eq(dev_net(dev), &init_net))
1692 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1693 mlx4_ib_scan_netdevs(ibdev);
1698 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1704 if (mlx4_is_master(ibdev->dev)) {
1705 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1706 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1708 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1710 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1711 /* master has the identity virt2phys pkey mapping */
1712 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1713 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1714 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1715 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1719 /* initialize pkey cache */
1720 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1722 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1724 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1730 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1733 int eq_per_port = 0;
1738 /* Legacy mode or comp_pool is not large enough */
1739 if (dev->caps.comp_pool == 0 ||
1740 dev->caps.num_ports > dev->caps.comp_pool)
1743 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1744 dev->caps.num_ports);
1748 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1749 added_eqs += eq_per_port;
1751 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1753 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1754 if (!ibdev->eq_table)
1757 ibdev->eq_added = added_eqs;
1760 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1761 for (j = 0; j < eq_per_port; j++) {
1762 sprintf(name, "mlx4-ib-%d-%d@%s",
1763 i, j, dev->pdev->bus->name);
1764 /* Set IRQ for specific name (per ring) */
1765 if (mlx4_assign_eq(dev, name, NULL,
1766 &ibdev->eq_table[eq])) {
1767 /* Use legacy (same as mlx4_en driver) */
1768 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1769 ibdev->eq_table[eq] =
1770 (eq % dev->caps.num_comp_vectors);
1776 /* Fill the reset of the vector with legacy EQ */
1777 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1778 ibdev->eq_table[eq++] = i;
1780 /* Advertise the new number of EQs to clients */
1781 ibdev->ib_dev.num_comp_vectors = total_eqs;
1784 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1788 /* no additional eqs were added */
1789 if (!ibdev->eq_table)
1792 /* Reset the advertised EQ number */
1793 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1795 /* Free only the added eqs */
1796 for (i = 0; i < ibdev->eq_added; i++) {
1797 /* Don't free legacy eqs if used */
1798 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1800 mlx4_release_eq(dev, ibdev->eq_table[i]);
1803 kfree(ibdev->eq_table);
1806 static void *mlx4_ib_add(struct mlx4_dev *dev)
1808 struct mlx4_ib_dev *ibdev;
1812 struct mlx4_ib_iboe *iboe;
1814 pr_info_once("%s", mlx4_ib_version);
1816 mlx4_foreach_non_ib_transport_port(i, dev)
1819 if (mlx4_is_mfunc(dev) && num_ports) {
1820 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1825 mlx4_foreach_ib_transport_port(i, dev)
1828 /* No point in registering a device with no ports... */
1832 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1834 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1838 iboe = &ibdev->iboe;
1840 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1843 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1846 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1848 if (!ibdev->uar_map)
1850 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1854 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1855 ibdev->ib_dev.owner = THIS_MODULE;
1856 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1857 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1858 ibdev->num_ports = num_ports;
1859 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1860 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1861 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1863 if (dev->caps.userspace_caps)
1864 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1866 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1868 ibdev->ib_dev.uverbs_cmd_mask =
1869 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1870 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1871 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1872 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1873 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1874 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1875 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1876 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1877 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1878 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1879 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1880 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1881 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1882 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1883 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1884 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1885 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1886 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1887 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1888 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1889 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1890 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1891 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1893 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1894 ibdev->ib_dev.query_port = mlx4_ib_query_port;
1895 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
1896 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1897 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1898 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1899 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1900 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1901 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1902 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1903 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1904 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1905 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1906 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1907 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1908 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1909 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
1910 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
1911 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1912 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1913 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1914 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
1915 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
1916 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1917 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1918 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1919 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
1920 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
1921 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
1922 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1923 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1924 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1925 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1926 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1927 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
1928 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1929 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1930 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
1931 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1932 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1933 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1935 if (!mlx4_is_slave(ibdev->dev)) {
1936 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1937 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1938 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1939 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1942 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1943 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
1944 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
1945 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
1946 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
1948 ibdev->ib_dev.uverbs_cmd_mask |=
1949 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
1950 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
1953 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1954 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1955 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1956 ibdev->ib_dev.uverbs_cmd_mask |=
1957 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1958 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1961 if (check_flow_steering_support(dev)) {
1962 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
1963 ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
1964 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
1966 ibdev->ib_dev.uverbs_ex_cmd_mask |=
1967 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
1968 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
1971 mlx4_ib_alloc_eqs(dev, ibdev);
1973 spin_lock_init(&iboe->lock);
1975 if (init_node_data(ibdev))
1978 for (i = 0; i < ibdev->num_ports; ++i) {
1979 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1980 IB_LINK_LAYER_ETHERNET) {
1981 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1983 ibdev->counters[i] = -1;
1985 ibdev->counters[i] = -1;
1988 spin_lock_init(&ibdev->sm_lock);
1989 mutex_init(&ibdev->cap_mask_mutex);
1991 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
1992 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
1993 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
1994 MLX4_IB_UC_STEER_QPN_ALIGN,
1995 &ibdev->steer_qpn_base);
1999 ibdev->ib_uc_qpns_bitmap =
2000 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2003 if (!ibdev->ib_uc_qpns_bitmap) {
2004 dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2005 goto err_steer_qp_release;
2008 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2010 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2011 dev, ibdev->steer_qpn_base,
2012 ibdev->steer_qpn_base +
2013 ibdev->steer_qpn_count - 1);
2015 goto err_steer_free_bitmap;
2018 if (ib_register_device(&ibdev->ib_dev, NULL))
2019 goto err_steer_free_bitmap;
2021 if (mlx4_ib_mad_init(ibdev))
2024 if (mlx4_ib_init_sriov(ibdev))
2027 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2028 if (!iboe->nb.notifier_call) {
2029 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2030 err = register_netdevice_notifier(&iboe->nb);
2032 iboe->nb.notifier_call = NULL;
2036 if (!iboe->nb_inet.notifier_call) {
2037 iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2038 err = register_inetaddr_notifier(&iboe->nb_inet);
2040 iboe->nb_inet.notifier_call = NULL;
2044 #if IS_ENABLED(CONFIG_IPV6)
2045 if (!iboe->nb_inet6.notifier_call) {
2046 iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2047 err = register_inet6addr_notifier(&iboe->nb_inet6);
2049 iboe->nb_inet6.notifier_call = NULL;
2054 mlx4_ib_scan_netdevs(ibdev);
2055 mlx4_ib_init_gid_table(ibdev);
2058 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2059 if (device_create_file(&ibdev->ib_dev.dev,
2060 mlx4_class_attributes[j]))
2064 ibdev->ib_active = true;
2066 if (mlx4_is_mfunc(ibdev->dev))
2069 /* create paravirt contexts for any VFs which are active */
2070 if (mlx4_is_master(ibdev->dev)) {
2071 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2072 if (j == mlx4_master_func_num(ibdev->dev))
2074 if (mlx4_is_slave_active(ibdev->dev, j))
2075 do_slave_init(ibdev, j, 1);
2081 if (ibdev->iboe.nb.notifier_call) {
2082 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2083 pr_warn("failure unregistering notifier\n");
2084 ibdev->iboe.nb.notifier_call = NULL;
2086 if (ibdev->iboe.nb_inet.notifier_call) {
2087 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2088 pr_warn("failure unregistering notifier\n");
2089 ibdev->iboe.nb_inet.notifier_call = NULL;
2091 #if IS_ENABLED(CONFIG_IPV6)
2092 if (ibdev->iboe.nb_inet6.notifier_call) {
2093 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2094 pr_warn("failure unregistering notifier\n");
2095 ibdev->iboe.nb_inet6.notifier_call = NULL;
2098 flush_workqueue(wq);
2100 mlx4_ib_close_sriov(ibdev);
2103 mlx4_ib_mad_cleanup(ibdev);
2106 ib_unregister_device(&ibdev->ib_dev);
2108 err_steer_free_bitmap:
2109 kfree(ibdev->ib_uc_qpns_bitmap);
2111 err_steer_qp_release:
2112 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2113 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2114 ibdev->steer_qpn_count);
2117 if (ibdev->counters[i - 1] != -1)
2118 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2121 iounmap(ibdev->uar_map);
2124 mlx4_uar_free(dev, &ibdev->priv_uar);
2127 mlx4_pd_free(dev, ibdev->priv_pdn);
2130 ib_dealloc_device(&ibdev->ib_dev);
2135 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2139 WARN_ON(!dev->ib_uc_qpns_bitmap);
2141 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2142 dev->steer_qpn_count,
2143 get_count_order(count));
2147 *qpn = dev->steer_qpn_base + offset;
2151 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2154 dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2157 BUG_ON(qpn < dev->steer_qpn_base);
2159 bitmap_release_region(dev->ib_uc_qpns_bitmap,
2160 qpn - dev->steer_qpn_base,
2161 get_count_order(count));
2164 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2169 struct ib_flow_attr *flow = NULL;
2170 struct ib_flow_spec_ib *ib_spec;
2173 flow_size = sizeof(struct ib_flow_attr) +
2174 sizeof(struct ib_flow_spec_ib);
2175 flow = kzalloc(flow_size, GFP_KERNEL);
2178 flow->port = mqp->port;
2179 flow->num_of_specs = 1;
2180 flow->size = flow_size;
2181 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2182 ib_spec->type = IB_FLOW_SPEC_IB;
2183 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2184 /* Add an empty rule for IB L2 */
2185 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2187 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2192 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2198 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2200 struct mlx4_ib_dev *ibdev = ibdev_ptr;
2203 mlx4_ib_close_sriov(ibdev);
2204 mlx4_ib_mad_cleanup(ibdev);
2205 ib_unregister_device(&ibdev->ib_dev);
2206 if (ibdev->iboe.nb.notifier_call) {
2207 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2208 pr_warn("failure unregistering notifier\n");
2209 ibdev->iboe.nb.notifier_call = NULL;
2212 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2213 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2214 ibdev->steer_qpn_count);
2215 kfree(ibdev->ib_uc_qpns_bitmap);
2218 if (ibdev->iboe.nb_inet.notifier_call) {
2219 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2220 pr_warn("failure unregistering notifier\n");
2221 ibdev->iboe.nb_inet.notifier_call = NULL;
2223 #if IS_ENABLED(CONFIG_IPV6)
2224 if (ibdev->iboe.nb_inet6.notifier_call) {
2225 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2226 pr_warn("failure unregistering notifier\n");
2227 ibdev->iboe.nb_inet6.notifier_call = NULL;
2231 iounmap(ibdev->uar_map);
2232 for (p = 0; p < ibdev->num_ports; ++p)
2233 if (ibdev->counters[p] != -1)
2234 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2235 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2236 mlx4_CLOSE_PORT(dev, p);
2238 mlx4_ib_free_eqs(dev, ibdev);
2240 mlx4_uar_free(dev, &ibdev->priv_uar);
2241 mlx4_pd_free(dev, ibdev->priv_pdn);
2242 ib_dealloc_device(&ibdev->ib_dev);
2245 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2247 struct mlx4_ib_demux_work **dm = NULL;
2248 struct mlx4_dev *dev = ibdev->dev;
2250 unsigned long flags;
2252 if (!mlx4_is_master(dev))
2255 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
2257 pr_err("failed to allocate memory for tunneling qp update\n");
2261 for (i = 0; i < dev->caps.num_ports; i++) {
2262 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2264 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2265 for (i = 0; i < dev->caps.num_ports; i++) {
2272 /* initialize or tear down tunnel QPs for the slave */
2273 for (i = 0; i < dev->caps.num_ports; i++) {
2274 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2275 dm[i]->port = i + 1;
2276 dm[i]->slave = slave;
2277 dm[i]->do_init = do_init;
2279 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2280 if (!ibdev->sriov.is_going_down)
2281 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2282 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2289 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2290 enum mlx4_dev_event event, unsigned long param)
2292 struct ib_event ibev;
2293 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2294 struct mlx4_eqe *eqe = NULL;
2295 struct ib_event_work *ew;
2298 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2299 eqe = (struct mlx4_eqe *)param;
2304 case MLX4_DEV_EVENT_PORT_UP:
2305 if (p > ibdev->num_ports)
2307 if (mlx4_is_master(dev) &&
2308 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2309 IB_LINK_LAYER_INFINIBAND) {
2310 mlx4_ib_invalidate_all_guid_record(ibdev, p);
2312 ibev.event = IB_EVENT_PORT_ACTIVE;
2315 case MLX4_DEV_EVENT_PORT_DOWN:
2316 if (p > ibdev->num_ports)
2318 ibev.event = IB_EVENT_PORT_ERR;
2321 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2322 ibdev->ib_active = false;
2323 ibev.event = IB_EVENT_DEVICE_FATAL;
2326 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2327 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2329 pr_err("failed to allocate memory for events work\n");
2333 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2334 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2336 /* need to queue only for port owner, which uses GEN_EQE */
2337 if (mlx4_is_master(dev))
2338 queue_work(wq, &ew->work);
2340 handle_port_mgmt_change_event(&ew->work);
2343 case MLX4_DEV_EVENT_SLAVE_INIT:
2344 /* here, p is the slave id */
2345 do_slave_init(ibdev, p, 1);
2348 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2349 /* here, p is the slave id */
2350 do_slave_init(ibdev, p, 0);
2357 ibev.device = ibdev_ptr;
2358 ibev.element.port_num = (u8) p;
2360 ib_dispatch_event(&ibev);
2363 static struct mlx4_interface mlx4_ib_interface = {
2365 .remove = mlx4_ib_remove,
2366 .event = mlx4_ib_event,
2367 .protocol = MLX4_PROT_IB_IPV6
2370 static int __init mlx4_ib_init(void)
2374 wq = create_singlethread_workqueue("mlx4_ib");
2378 err = mlx4_ib_mcg_init();
2382 err = mlx4_register_interface(&mlx4_ib_interface);
2389 mlx4_ib_mcg_destroy();
2392 destroy_workqueue(wq);
2396 static void __exit mlx4_ib_cleanup(void)
2398 mlx4_unregister_interface(&mlx4_ib_interface);
2399 mlx4_ib_mcg_destroy();
2400 destroy_workqueue(wq);
2403 module_init(mlx4_ib_init);
2404 module_exit(mlx4_ib_cleanup);