2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_addr.h>
47 #include <linux/mlx4/driver.h>
48 #include <linux/mlx4/cmd.h>
53 #define DRV_NAME "mlx4_ib"
54 #define DRV_VERSION "1.0"
55 #define DRV_RELDATE "April 4, 2008"
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
62 static const char mlx4_ib_version[] =
63 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
64 DRV_VERSION " (" DRV_RELDATE ")\n";
66 struct update_gid_work {
67 struct work_struct work;
68 union ib_gid gids[128];
69 struct mlx4_ib_dev *dev;
73 static struct workqueue_struct *wq;
75 static void init_query_mad(struct ib_smp *mad)
77 mad->base_version = 1;
78 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
79 mad->class_version = 1;
80 mad->method = IB_MGMT_METHOD_GET;
83 static union ib_gid zgid;
85 static int mlx4_ib_query_device(struct ib_device *ibdev,
86 struct ib_device_attr *props)
88 struct mlx4_ib_dev *dev = to_mdev(ibdev);
89 struct ib_smp *in_mad = NULL;
90 struct ib_smp *out_mad = NULL;
93 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
94 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
95 if (!in_mad || !out_mad)
98 init_query_mad(in_mad);
99 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
101 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
105 memset(props, 0, sizeof *props);
107 props->fw_ver = dev->dev->caps.fw_ver;
108 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
109 IB_DEVICE_PORT_ACTIVE_EVENT |
110 IB_DEVICE_SYS_IMAGE_GUID |
111 IB_DEVICE_RC_RNR_NAK_GEN |
112 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
113 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
114 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
115 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
116 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
117 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
118 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
119 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
120 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
121 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
122 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
123 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
124 props->device_cap_flags |= IB_DEVICE_UD_TSO;
125 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
126 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
127 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
128 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
129 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
130 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
132 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
134 props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30));
135 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
136 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
138 props->max_mr_size = ~0ull;
139 props->page_size_cap = dev->dev->caps.page_size_cap;
140 props->max_qp = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
141 props->max_qp_wr = dev->dev->caps.max_wqes;
142 props->max_sge = min(dev->dev->caps.max_sq_sg,
143 dev->dev->caps.max_rq_sg);
144 props->max_cq = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
145 props->max_cqe = dev->dev->caps.max_cqes;
146 props->max_mr = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
147 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
148 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
149 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
150 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
151 props->max_srq = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
152 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
153 props->max_srq_sge = dev->dev->caps.max_srq_sge;
154 props->max_fast_reg_page_list_len = PAGE_SIZE / sizeof (u64);
155 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
156 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
157 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
158 props->masked_atomic_cap = IB_ATOMIC_HCA;
159 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
160 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
161 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
162 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
163 props->max_mcast_grp;
164 props->max_map_per_fmr = (1 << (32 - ilog2(dev->dev->caps.num_mpts))) - 1;
173 static enum rdma_link_layer
174 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
176 struct mlx4_dev *dev = to_mdev(device)->dev;
178 return dev->caps.port_mask & (1 << (port_num - 1)) ?
179 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
182 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
183 struct ib_port_attr *props,
184 struct ib_smp *out_mad)
186 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
187 props->lmc = out_mad->data[34] & 0x7;
188 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
189 props->sm_sl = out_mad->data[36] & 0xf;
190 props->state = out_mad->data[32] & 0xf;
191 props->phys_state = out_mad->data[33] >> 4;
192 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
193 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
194 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
195 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
196 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
197 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
198 props->active_width = out_mad->data[31] & 0xf;
199 props->active_speed = out_mad->data[35] >> 4;
200 props->max_mtu = out_mad->data[41] & 0xf;
201 props->active_mtu = out_mad->data[36] >> 4;
202 props->subnet_timeout = out_mad->data[51] & 0x1f;
203 props->max_vl_num = out_mad->data[37] >> 4;
204 props->init_type_reply = out_mad->data[41] >> 4;
209 static u8 state_to_phys_state(enum ib_port_state state)
211 return state == IB_PORT_ACTIVE ? 5 : 3;
214 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
215 struct ib_port_attr *props,
216 struct ib_smp *out_mad)
218 struct mlx4_ib_iboe *iboe = &to_mdev(ibdev)->iboe;
219 struct net_device *ndev;
222 props->active_width = IB_WIDTH_4X;
223 props->active_speed = 4;
224 props->port_cap_flags = IB_PORT_CM_SUP;
225 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
226 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
227 props->pkey_tbl_len = 1;
228 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
229 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
230 props->max_mtu = IB_MTU_2048;
231 props->subnet_timeout = 0;
232 props->max_vl_num = out_mad->data[37] >> 4;
233 props->init_type_reply = 0;
234 props->state = IB_PORT_DOWN;
235 props->phys_state = state_to_phys_state(props->state);
236 props->active_mtu = IB_MTU_256;
237 spin_lock(&iboe->lock);
238 ndev = iboe->netdevs[port - 1];
242 tmp = iboe_get_mtu(ndev->mtu);
243 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
245 props->state = netif_running(ndev) && netif_oper_up(ndev) ?
246 IB_PORT_ACTIVE : IB_PORT_DOWN;
247 props->phys_state = state_to_phys_state(props->state);
250 spin_unlock(&iboe->lock);
254 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
255 struct ib_port_attr *props)
257 struct ib_smp *in_mad = NULL;
258 struct ib_smp *out_mad = NULL;
261 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
262 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
263 if (!in_mad || !out_mad)
266 memset(props, 0, sizeof *props);
268 init_query_mad(in_mad);
269 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
270 in_mad->attr_mod = cpu_to_be32(port);
272 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
276 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
277 ib_link_query_port(ibdev, port, props, out_mad) :
278 eth_link_query_port(ibdev, port, props, out_mad);
287 static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
290 struct ib_smp *in_mad = NULL;
291 struct ib_smp *out_mad = NULL;
294 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
295 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
296 if (!in_mad || !out_mad)
299 init_query_mad(in_mad);
300 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
301 in_mad->attr_mod = cpu_to_be32(port);
303 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
307 memcpy(gid->raw, out_mad->data + 8, 8);
309 init_query_mad(in_mad);
310 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
311 in_mad->attr_mod = cpu_to_be32(index / 8);
313 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
317 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
325 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
328 struct mlx4_ib_dev *dev = to_mdev(ibdev);
330 *gid = dev->iboe.gid_table[port - 1][index];
335 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
338 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
339 return __mlx4_ib_query_gid(ibdev, port, index, gid);
341 return iboe_query_gid(ibdev, port, index, gid);
344 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
347 struct ib_smp *in_mad = NULL;
348 struct ib_smp *out_mad = NULL;
351 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
352 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
353 if (!in_mad || !out_mad)
356 init_query_mad(in_mad);
357 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
358 in_mad->attr_mod = cpu_to_be32(index / 32);
360 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
364 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
372 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
373 struct ib_device_modify *props)
375 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
378 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
379 spin_lock(&to_mdev(ibdev)->sm_lock);
380 memcpy(ibdev->node_desc, props->node_desc, 64);
381 spin_unlock(&to_mdev(ibdev)->sm_lock);
387 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
390 struct mlx4_cmd_mailbox *mailbox;
392 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
394 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
396 return PTR_ERR(mailbox);
398 memset(mailbox->buf, 0, 256);
400 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
401 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
402 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
404 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
405 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
408 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
409 MLX4_CMD_TIME_CLASS_B);
411 mlx4_free_cmd_mailbox(dev->dev, mailbox);
415 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
416 struct ib_port_modify *props)
418 struct ib_port_attr attr;
422 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
424 err = mlx4_ib_query_port(ibdev, port, &attr);
428 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
429 ~props->clr_port_cap_mask;
431 err = mlx4_SET_PORT(to_mdev(ibdev), port,
432 !!(mask & IB_PORT_RESET_QKEY_CNTR),
436 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
440 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
441 struct ib_udata *udata)
443 struct mlx4_ib_dev *dev = to_mdev(ibdev);
444 struct mlx4_ib_ucontext *context;
445 struct mlx4_ib_alloc_ucontext_resp resp;
449 return ERR_PTR(-EAGAIN);
451 resp.qp_tab_size = dev->dev->caps.num_qps;
452 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
453 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
455 context = kmalloc(sizeof *context, GFP_KERNEL);
457 return ERR_PTR(-ENOMEM);
459 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
465 INIT_LIST_HEAD(&context->db_page_list);
466 mutex_init(&context->db_page_mutex);
468 err = ib_copy_to_udata(udata, &resp, sizeof resp);
470 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
472 return ERR_PTR(-EFAULT);
475 return &context->ibucontext;
478 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
480 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
482 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
488 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
490 struct mlx4_ib_dev *dev = to_mdev(context->device);
492 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
495 if (vma->vm_pgoff == 0) {
496 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
498 if (io_remap_pfn_range(vma, vma->vm_start,
499 to_mucontext(context)->uar.pfn,
500 PAGE_SIZE, vma->vm_page_prot))
502 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
503 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
505 if (io_remap_pfn_range(vma, vma->vm_start,
506 to_mucontext(context)->uar.pfn +
507 dev->dev->caps.num_uars,
508 PAGE_SIZE, vma->vm_page_prot))
516 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
517 struct ib_ucontext *context,
518 struct ib_udata *udata)
520 struct mlx4_ib_pd *pd;
523 pd = kmalloc(sizeof *pd, GFP_KERNEL);
525 return ERR_PTR(-ENOMEM);
527 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
534 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
535 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
537 return ERR_PTR(-EFAULT);
543 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
545 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
551 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
553 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
554 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
555 struct mlx4_ib_gid_entry *ge;
557 ge = kzalloc(sizeof *ge, GFP_KERNEL);
562 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
563 ge->port = mqp->port;
567 mutex_lock(&mqp->mutex);
568 list_add_tail(&ge->list, &mqp->gid_list);
569 mutex_unlock(&mqp->mutex);
574 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
578 struct net_device *ndev;
584 spin_lock(&mdev->iboe.lock);
585 ndev = mdev->iboe.netdevs[mqp->port - 1];
588 spin_unlock(&mdev->iboe.lock);
591 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
593 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
602 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
605 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
606 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
608 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, !!(mqp->flags &
609 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK));
613 err = add_gid_entry(ibqp, gid);
620 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw);
624 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
626 struct mlx4_ib_gid_entry *ge;
627 struct mlx4_ib_gid_entry *tmp;
628 struct mlx4_ib_gid_entry *ret = NULL;
630 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
631 if (!memcmp(raw, ge->gid.raw, 16)) {
640 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
643 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
644 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
646 struct net_device *ndev;
647 struct mlx4_ib_gid_entry *ge;
649 err = mlx4_multicast_detach(mdev->dev,
650 &mqp->mqp, gid->raw);
654 mutex_lock(&mqp->mutex);
655 ge = find_gid_entry(mqp, gid->raw);
657 spin_lock(&mdev->iboe.lock);
658 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
661 spin_unlock(&mdev->iboe.lock);
662 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
665 dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
672 printk(KERN_WARNING "could not find mgid entry\n");
674 mutex_unlock(&mqp->mutex);
679 static int init_node_data(struct mlx4_ib_dev *dev)
681 struct ib_smp *in_mad = NULL;
682 struct ib_smp *out_mad = NULL;
685 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
686 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
687 if (!in_mad || !out_mad)
690 init_query_mad(in_mad);
691 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
693 err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
697 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
699 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
701 err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
705 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
706 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
714 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
717 struct mlx4_ib_dev *dev =
718 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
719 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
722 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
725 struct mlx4_ib_dev *dev =
726 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
727 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
728 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
729 (int) dev->dev->caps.fw_ver & 0xffff);
732 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
735 struct mlx4_ib_dev *dev =
736 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
737 return sprintf(buf, "%x\n", dev->dev->rev_id);
740 static ssize_t show_board(struct device *device, struct device_attribute *attr,
743 struct mlx4_ib_dev *dev =
744 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
745 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
749 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
750 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
751 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
752 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
754 static struct device_attribute *mlx4_class_attributes[] = {
761 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
763 memcpy(eui, dev->dev_addr, 3);
764 memcpy(eui + 5, dev->dev_addr + 3, 3);
765 if (vlan_id < 0x1000) {
766 eui[3] = vlan_id >> 8;
767 eui[4] = vlan_id & 0xff;
775 static void update_gids_task(struct work_struct *work)
777 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
778 struct mlx4_cmd_mailbox *mailbox;
781 struct mlx4_dev *dev = gw->dev->dev;
782 struct ib_event event;
784 mailbox = mlx4_alloc_cmd_mailbox(dev);
785 if (IS_ERR(mailbox)) {
786 printk(KERN_WARNING "update gid table failed %ld\n", PTR_ERR(mailbox));
791 memcpy(gids, gw->gids, sizeof gw->gids);
793 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
794 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B);
796 printk(KERN_WARNING "set port command failed\n");
798 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
799 event.device = &gw->dev->ib_dev;
800 event.element.port_num = gw->port;
801 event.event = IB_EVENT_LID_CHANGE;
802 ib_dispatch_event(&event);
805 mlx4_free_cmd_mailbox(dev, mailbox);
809 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
811 struct net_device *ndev = dev->iboe.netdevs[port - 1];
812 struct update_gid_work *work;
813 struct net_device *tmp;
823 work = kzalloc(sizeof *work, GFP_ATOMIC);
827 hits = kzalloc(128, GFP_ATOMIC);
833 read_lock(&dev_base_lock);
834 for_each_netdev(&init_net, tmp) {
835 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
836 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
837 vid = rdma_vlan_dev_vlan_id(tmp);
838 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
841 for (i = 0; i < 128; ++i) {
843 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
845 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
854 (memcmp(&dev->iboe.gid_table[port - 1][0],
856 !memcmp(&dev->iboe.gid_table[port - 1][0],
857 &zgid, sizeof gid))) {
858 dev->iboe.gid_table[port - 1][0] = gid;
861 } else if (free >= 0) {
862 dev->iboe.gid_table[port - 1][free] = gid;
869 read_unlock(&dev_base_lock);
871 for (i = 0; i < 128; ++i)
873 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
875 dev->iboe.gid_table[port - 1][i] = zgid;
879 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
880 INIT_WORK(&work->work, update_gids_task);
883 queue_work(wq, &work->work);
895 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
899 case NETDEV_CHANGEADDR:
900 update_ipv6_gids(dev, port, 0);
904 update_ipv6_gids(dev, port, 1);
905 dev->iboe.netdevs[port - 1] = NULL;
909 static void netdev_added(struct mlx4_ib_dev *dev, int port)
911 update_ipv6_gids(dev, port, 0);
914 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
916 update_ipv6_gids(dev, port, 1);
919 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
922 struct net_device *dev = ptr;
923 struct mlx4_ib_dev *ibdev;
924 struct net_device *oldnd;
925 struct mlx4_ib_iboe *iboe;
928 if (!net_eq(dev_net(dev), &init_net))
931 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
934 spin_lock(&iboe->lock);
935 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
936 oldnd = iboe->netdevs[port - 1];
937 iboe->netdevs[port - 1] =
938 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROTOCOL_EN, port);
939 if (oldnd != iboe->netdevs[port - 1]) {
940 if (iboe->netdevs[port - 1])
941 netdev_added(ibdev, port);
943 netdev_removed(ibdev, port);
947 if (dev == iboe->netdevs[0] ||
948 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
949 handle_en_event(ibdev, 1, event);
950 else if (dev == iboe->netdevs[1]
951 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
952 handle_en_event(ibdev, 2, event);
954 spin_unlock(&iboe->lock);
959 static void *mlx4_ib_add(struct mlx4_dev *dev)
961 struct mlx4_ib_dev *ibdev;
965 struct mlx4_ib_iboe *iboe;
967 printk_once(KERN_INFO "%s", mlx4_ib_version);
969 mlx4_foreach_ib_transport_port(i, dev)
972 /* No point in registering a device with no ports... */
976 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
978 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
984 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
987 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
990 ibdev->uar_map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
993 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
997 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
998 ibdev->ib_dev.owner = THIS_MODULE;
999 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1000 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1001 ibdev->num_ports = num_ports;
1002 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1003 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1004 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1006 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1007 ibdev->ib_dev.uverbs_cmd_mask =
1008 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1009 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1010 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1011 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1012 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1013 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1014 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1015 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1016 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1017 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1018 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1019 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1020 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1021 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1022 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1023 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1024 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1025 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1026 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1027 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1028 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1030 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1031 ibdev->ib_dev.query_port = mlx4_ib_query_port;
1032 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
1033 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1034 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1035 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1036 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1037 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1038 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1039 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1040 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1041 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1042 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1043 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1044 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1045 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1046 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
1047 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
1048 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1049 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1050 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1051 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
1052 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
1053 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1054 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1055 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1056 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
1057 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
1058 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
1059 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1060 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1061 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1062 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1063 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1064 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
1065 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1066 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1067 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
1068 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1069 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1070 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1072 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1073 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1074 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1075 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1077 spin_lock_init(&iboe->lock);
1079 if (init_node_data(ibdev))
1082 spin_lock_init(&ibdev->sm_lock);
1083 mutex_init(&ibdev->cap_mask_mutex);
1085 if (ib_register_device(&ibdev->ib_dev, NULL))
1088 if (mlx4_ib_mad_init(ibdev))
1091 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1092 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1093 err = register_netdevice_notifier(&iboe->nb);
1098 for (i = 0; i < ARRAY_SIZE(mlx4_class_attributes); ++i) {
1099 if (device_create_file(&ibdev->ib_dev.dev,
1100 mlx4_class_attributes[i]))
1104 ibdev->ib_active = true;
1109 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1110 printk(KERN_WARNING "failure unregistering notifier\n");
1111 flush_workqueue(wq);
1114 ib_unregister_device(&ibdev->ib_dev);
1117 iounmap(ibdev->uar_map);
1120 mlx4_uar_free(dev, &ibdev->priv_uar);
1123 mlx4_pd_free(dev, ibdev->priv_pdn);
1126 ib_dealloc_device(&ibdev->ib_dev);
1131 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1133 struct mlx4_ib_dev *ibdev = ibdev_ptr;
1136 mlx4_ib_mad_cleanup(ibdev);
1137 ib_unregister_device(&ibdev->ib_dev);
1138 if (ibdev->iboe.nb.notifier_call) {
1139 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1140 printk(KERN_WARNING "failure unregistering notifier\n");
1141 ibdev->iboe.nb.notifier_call = NULL;
1143 iounmap(ibdev->uar_map);
1145 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1146 mlx4_CLOSE_PORT(dev, p);
1148 mlx4_uar_free(dev, &ibdev->priv_uar);
1149 mlx4_pd_free(dev, ibdev->priv_pdn);
1150 ib_dealloc_device(&ibdev->ib_dev);
1153 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1154 enum mlx4_dev_event event, int port)
1156 struct ib_event ibev;
1157 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1159 if (port > ibdev->num_ports)
1163 case MLX4_DEV_EVENT_PORT_UP:
1164 ibev.event = IB_EVENT_PORT_ACTIVE;
1167 case MLX4_DEV_EVENT_PORT_DOWN:
1168 ibev.event = IB_EVENT_PORT_ERR;
1171 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1172 ibdev->ib_active = false;
1173 ibev.event = IB_EVENT_DEVICE_FATAL;
1180 ibev.device = ibdev_ptr;
1181 ibev.element.port_num = port;
1183 ib_dispatch_event(&ibev);
1186 static struct mlx4_interface mlx4_ib_interface = {
1188 .remove = mlx4_ib_remove,
1189 .event = mlx4_ib_event,
1190 .protocol = MLX4_PROTOCOL_IB
1193 static int __init mlx4_ib_init(void)
1197 wq = create_singlethread_workqueue("mlx4_ib");
1201 err = mlx4_register_interface(&mlx4_ib_interface);
1203 destroy_workqueue(wq);
1210 static void __exit mlx4_ib_cleanup(void)
1212 mlx4_unregister_interface(&mlx4_ib_interface);
1213 destroy_workqueue(wq);
1216 module_init(mlx4_ib_init);
1217 module_exit(mlx4_ib_cleanup);