2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_addr.h>
47 #include <linux/mlx4/driver.h>
48 #include <linux/mlx4/cmd.h>
53 #define DRV_NAME MLX4_IB_DRV_NAME
54 #define DRV_VERSION "1.0"
55 #define DRV_RELDATE "April 4, 2008"
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
62 int mlx4_ib_sm_guid_assign = 1;
63 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
64 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
66 static const char mlx4_ib_version[] =
67 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
68 DRV_VERSION " (" DRV_RELDATE ")\n";
70 struct update_gid_work {
71 struct work_struct work;
72 union ib_gid gids[128];
73 struct mlx4_ib_dev *dev;
77 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
79 static struct workqueue_struct *wq;
81 static void init_query_mad(struct ib_smp *mad)
83 mad->base_version = 1;
84 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
85 mad->class_version = 1;
86 mad->method = IB_MGMT_METHOD_GET;
89 static union ib_gid zgid;
91 static int mlx4_ib_query_device(struct ib_device *ibdev,
92 struct ib_device_attr *props)
94 struct mlx4_ib_dev *dev = to_mdev(ibdev);
95 struct ib_smp *in_mad = NULL;
96 struct ib_smp *out_mad = NULL;
99 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
100 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
101 if (!in_mad || !out_mad)
104 init_query_mad(in_mad);
105 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
107 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
108 1, NULL, NULL, in_mad, out_mad);
112 memset(props, 0, sizeof *props);
114 props->fw_ver = dev->dev->caps.fw_ver;
115 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
116 IB_DEVICE_PORT_ACTIVE_EVENT |
117 IB_DEVICE_SYS_IMAGE_GUID |
118 IB_DEVICE_RC_RNR_NAK_GEN |
119 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
120 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
121 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
122 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
123 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
124 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
125 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
126 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
127 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
128 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
129 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
130 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
131 props->device_cap_flags |= IB_DEVICE_UD_TSO;
132 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
133 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
134 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
135 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
136 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
137 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
138 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
139 props->device_cap_flags |= IB_DEVICE_XRC;
141 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
143 props->vendor_part_id = dev->dev->pdev->device;
144 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
145 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
147 props->max_mr_size = ~0ull;
148 props->page_size_cap = dev->dev->caps.page_size_cap;
149 props->max_qp = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
150 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
151 props->max_sge = min(dev->dev->caps.max_sq_sg,
152 dev->dev->caps.max_rq_sg);
153 props->max_cq = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
154 props->max_cqe = dev->dev->caps.max_cqes;
155 props->max_mr = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
156 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
157 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
158 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
159 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
160 props->max_srq = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
161 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
162 props->max_srq_sge = dev->dev->caps.max_srq_sge;
163 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
164 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
165 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
166 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
167 props->masked_atomic_cap = props->atomic_cap;
168 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
169 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
170 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
171 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
172 props->max_mcast_grp;
173 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
182 static enum rdma_link_layer
183 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
185 struct mlx4_dev *dev = to_mdev(device)->dev;
187 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
188 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
191 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
192 struct ib_port_attr *props, int netw_view)
194 struct ib_smp *in_mad = NULL;
195 struct ib_smp *out_mad = NULL;
196 int ext_active_speed;
197 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
200 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
201 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
202 if (!in_mad || !out_mad)
205 init_query_mad(in_mad);
206 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
207 in_mad->attr_mod = cpu_to_be32(port);
209 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
210 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
212 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
218 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
219 props->lmc = out_mad->data[34] & 0x7;
220 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
221 props->sm_sl = out_mad->data[36] & 0xf;
222 props->state = out_mad->data[32] & 0xf;
223 props->phys_state = out_mad->data[33] >> 4;
224 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
226 props->gid_tbl_len = out_mad->data[50];
228 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
229 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
230 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
231 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
232 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
233 props->active_width = out_mad->data[31] & 0xf;
234 props->active_speed = out_mad->data[35] >> 4;
235 props->max_mtu = out_mad->data[41] & 0xf;
236 props->active_mtu = out_mad->data[36] >> 4;
237 props->subnet_timeout = out_mad->data[51] & 0x1f;
238 props->max_vl_num = out_mad->data[37] >> 4;
239 props->init_type_reply = out_mad->data[41] >> 4;
241 /* Check if extended speeds (EDR/FDR/...) are supported */
242 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
243 ext_active_speed = out_mad->data[62] >> 4;
245 switch (ext_active_speed) {
247 props->active_speed = IB_SPEED_FDR;
250 props->active_speed = IB_SPEED_EDR;
255 /* If reported active speed is QDR, check if is FDR-10 */
256 if (props->active_speed == IB_SPEED_QDR) {
257 init_query_mad(in_mad);
258 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
259 in_mad->attr_mod = cpu_to_be32(port);
261 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
262 NULL, NULL, in_mad, out_mad);
266 /* Checking LinkSpeedActive for FDR-10 */
267 if (out_mad->data[15] & 0x1)
268 props->active_speed = IB_SPEED_FDR10;
271 /* Avoid wrong speed value returned by FW if the IB link is down. */
272 if (props->state == IB_PORT_DOWN)
273 props->active_speed = IB_SPEED_SDR;
281 static u8 state_to_phys_state(enum ib_port_state state)
283 return state == IB_PORT_ACTIVE ? 5 : 3;
286 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
287 struct ib_port_attr *props, int netw_view)
290 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
291 struct mlx4_ib_iboe *iboe = &mdev->iboe;
292 struct net_device *ndev;
294 struct mlx4_cmd_mailbox *mailbox;
297 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
299 return PTR_ERR(mailbox);
301 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
302 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
307 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
308 IB_WIDTH_4X : IB_WIDTH_1X;
309 props->active_speed = IB_SPEED_QDR;
310 props->port_cap_flags = IB_PORT_CM_SUP;
311 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
312 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
313 props->pkey_tbl_len = 1;
314 props->max_mtu = IB_MTU_4096;
315 props->max_vl_num = 2;
316 props->state = IB_PORT_DOWN;
317 props->phys_state = state_to_phys_state(props->state);
318 props->active_mtu = IB_MTU_256;
319 spin_lock(&iboe->lock);
320 ndev = iboe->netdevs[port - 1];
324 tmp = iboe_get_mtu(ndev->mtu);
325 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
327 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
328 IB_PORT_ACTIVE : IB_PORT_DOWN;
329 props->phys_state = state_to_phys_state(props->state);
331 spin_unlock(&iboe->lock);
333 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
337 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
338 struct ib_port_attr *props, int netw_view)
342 memset(props, 0, sizeof *props);
344 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
345 ib_link_query_port(ibdev, port, props, netw_view) :
346 eth_link_query_port(ibdev, port, props, netw_view);
351 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
352 struct ib_port_attr *props)
354 /* returns host view */
355 return __mlx4_ib_query_port(ibdev, port, props, 0);
358 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
359 union ib_gid *gid, int netw_view)
361 struct ib_smp *in_mad = NULL;
362 struct ib_smp *out_mad = NULL;
364 struct mlx4_ib_dev *dev = to_mdev(ibdev);
366 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
368 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
369 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
370 if (!in_mad || !out_mad)
373 init_query_mad(in_mad);
374 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
375 in_mad->attr_mod = cpu_to_be32(port);
377 if (mlx4_is_mfunc(dev->dev) && netw_view)
378 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
380 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
384 memcpy(gid->raw, out_mad->data + 8, 8);
386 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
388 /* For any index > 0, return the null guid */
395 init_query_mad(in_mad);
396 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
397 in_mad->attr_mod = cpu_to_be32(index / 8);
399 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
400 NULL, NULL, in_mad, out_mad);
404 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
408 memset(gid->raw + 8, 0, 8);
414 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
417 struct mlx4_ib_dev *dev = to_mdev(ibdev);
419 *gid = dev->iboe.gid_table[port - 1][index];
424 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
427 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
428 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
430 return iboe_query_gid(ibdev, port, index, gid);
433 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
434 u16 *pkey, int netw_view)
436 struct ib_smp *in_mad = NULL;
437 struct ib_smp *out_mad = NULL;
438 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
441 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
442 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
443 if (!in_mad || !out_mad)
446 init_query_mad(in_mad);
447 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
448 in_mad->attr_mod = cpu_to_be32(index / 32);
450 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
451 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
453 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
458 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
466 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
468 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
471 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
472 struct ib_device_modify *props)
474 struct mlx4_cmd_mailbox *mailbox;
477 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
480 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
483 if (mlx4_is_slave(to_mdev(ibdev)->dev))
486 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
487 memcpy(ibdev->node_desc, props->node_desc, 64);
488 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
491 * If possible, pass node desc to FW, so it can generate
492 * a 144 trap. If cmd fails, just ignore.
494 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
498 memset(mailbox->buf, 0, 256);
499 memcpy(mailbox->buf, props->node_desc, 64);
500 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
501 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
503 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
508 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
511 struct mlx4_cmd_mailbox *mailbox;
513 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
515 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
517 return PTR_ERR(mailbox);
519 memset(mailbox->buf, 0, 256);
521 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
522 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
523 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
525 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
526 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
529 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
530 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
532 mlx4_free_cmd_mailbox(dev->dev, mailbox);
536 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
537 struct ib_port_modify *props)
539 struct ib_port_attr attr;
543 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
545 err = mlx4_ib_query_port(ibdev, port, &attr);
549 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
550 ~props->clr_port_cap_mask;
552 err = mlx4_SET_PORT(to_mdev(ibdev), port,
553 !!(mask & IB_PORT_RESET_QKEY_CNTR),
557 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
561 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
562 struct ib_udata *udata)
564 struct mlx4_ib_dev *dev = to_mdev(ibdev);
565 struct mlx4_ib_ucontext *context;
566 struct mlx4_ib_alloc_ucontext_resp resp;
570 return ERR_PTR(-EAGAIN);
572 resp.qp_tab_size = dev->dev->caps.num_qps;
573 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
574 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
576 context = kmalloc(sizeof *context, GFP_KERNEL);
578 return ERR_PTR(-ENOMEM);
580 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
586 INIT_LIST_HEAD(&context->db_page_list);
587 mutex_init(&context->db_page_mutex);
589 err = ib_copy_to_udata(udata, &resp, sizeof resp);
591 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
593 return ERR_PTR(-EFAULT);
596 return &context->ibucontext;
599 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
601 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
603 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
609 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
611 struct mlx4_ib_dev *dev = to_mdev(context->device);
613 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
616 if (vma->vm_pgoff == 0) {
617 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
619 if (io_remap_pfn_range(vma, vma->vm_start,
620 to_mucontext(context)->uar.pfn,
621 PAGE_SIZE, vma->vm_page_prot))
623 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
624 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
626 if (io_remap_pfn_range(vma, vma->vm_start,
627 to_mucontext(context)->uar.pfn +
628 dev->dev->caps.num_uars,
629 PAGE_SIZE, vma->vm_page_prot))
637 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
638 struct ib_ucontext *context,
639 struct ib_udata *udata)
641 struct mlx4_ib_pd *pd;
644 pd = kmalloc(sizeof *pd, GFP_KERNEL);
646 return ERR_PTR(-ENOMEM);
648 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
655 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
656 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
658 return ERR_PTR(-EFAULT);
664 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
666 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
672 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
673 struct ib_ucontext *context,
674 struct ib_udata *udata)
676 struct mlx4_ib_xrcd *xrcd;
679 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
680 return ERR_PTR(-ENOSYS);
682 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
684 return ERR_PTR(-ENOMEM);
686 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
690 xrcd->pd = ib_alloc_pd(ibdev);
691 if (IS_ERR(xrcd->pd)) {
692 err = PTR_ERR(xrcd->pd);
696 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
697 if (IS_ERR(xrcd->cq)) {
698 err = PTR_ERR(xrcd->cq);
702 return &xrcd->ibxrcd;
705 ib_dealloc_pd(xrcd->pd);
707 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
713 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
715 ib_destroy_cq(to_mxrcd(xrcd)->cq);
716 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
717 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
723 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
725 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
726 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
727 struct mlx4_ib_gid_entry *ge;
729 ge = kzalloc(sizeof *ge, GFP_KERNEL);
734 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
735 ge->port = mqp->port;
739 mutex_lock(&mqp->mutex);
740 list_add_tail(&ge->list, &mqp->gid_list);
741 mutex_unlock(&mqp->mutex);
746 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
750 struct net_device *ndev;
756 spin_lock(&mdev->iboe.lock);
757 ndev = mdev->iboe.netdevs[mqp->port - 1];
760 spin_unlock(&mdev->iboe.lock);
763 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
765 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
774 struct mlx4_ib_steering {
775 struct list_head list;
780 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
783 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
784 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
786 struct mlx4_ib_steering *ib_steering = NULL;
788 if (mdev->dev->caps.steering_mode ==
789 MLX4_STEERING_MODE_DEVICE_MANAGED) {
790 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
795 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
797 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
798 MLX4_PROT_IB_IPV6, ®_id);
802 err = add_gid_entry(ibqp, gid);
807 memcpy(ib_steering->gid.raw, gid->raw, 16);
808 ib_steering->reg_id = reg_id;
809 mutex_lock(&mqp->mutex);
810 list_add(&ib_steering->list, &mqp->steering_rules);
811 mutex_unlock(&mqp->mutex);
816 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
817 MLX4_PROT_IB_IPV6, reg_id);
824 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
826 struct mlx4_ib_gid_entry *ge;
827 struct mlx4_ib_gid_entry *tmp;
828 struct mlx4_ib_gid_entry *ret = NULL;
830 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
831 if (!memcmp(raw, ge->gid.raw, 16)) {
840 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
843 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
844 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
846 struct net_device *ndev;
847 struct mlx4_ib_gid_entry *ge;
850 if (mdev->dev->caps.steering_mode ==
851 MLX4_STEERING_MODE_DEVICE_MANAGED) {
852 struct mlx4_ib_steering *ib_steering;
854 mutex_lock(&mqp->mutex);
855 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
856 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
857 list_del(&ib_steering->list);
861 mutex_unlock(&mqp->mutex);
862 if (&ib_steering->list == &mqp->steering_rules) {
863 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
866 reg_id = ib_steering->reg_id;
870 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
871 MLX4_PROT_IB_IPV6, reg_id);
875 mutex_lock(&mqp->mutex);
876 ge = find_gid_entry(mqp, gid->raw);
878 spin_lock(&mdev->iboe.lock);
879 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
882 spin_unlock(&mdev->iboe.lock);
883 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
886 dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
893 pr_warn("could not find mgid entry\n");
895 mutex_unlock(&mqp->mutex);
900 static int init_node_data(struct mlx4_ib_dev *dev)
902 struct ib_smp *in_mad = NULL;
903 struct ib_smp *out_mad = NULL;
904 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
907 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
908 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
909 if (!in_mad || !out_mad)
912 init_query_mad(in_mad);
913 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
914 if (mlx4_is_master(dev->dev))
915 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
917 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
921 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
923 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
925 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
929 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
930 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
938 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
941 struct mlx4_ib_dev *dev =
942 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
943 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
946 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
949 struct mlx4_ib_dev *dev =
950 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
951 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
952 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
953 (int) dev->dev->caps.fw_ver & 0xffff);
956 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
959 struct mlx4_ib_dev *dev =
960 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
961 return sprintf(buf, "%x\n", dev->dev->rev_id);
964 static ssize_t show_board(struct device *device, struct device_attribute *attr,
967 struct mlx4_ib_dev *dev =
968 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
969 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
973 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
974 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
975 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
976 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
978 static struct device_attribute *mlx4_class_attributes[] = {
985 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
987 memcpy(eui, dev->dev_addr, 3);
988 memcpy(eui + 5, dev->dev_addr + 3, 3);
989 if (vlan_id < 0x1000) {
990 eui[3] = vlan_id >> 8;
991 eui[4] = vlan_id & 0xff;
999 static void update_gids_task(struct work_struct *work)
1001 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1002 struct mlx4_cmd_mailbox *mailbox;
1005 struct mlx4_dev *dev = gw->dev->dev;
1007 mailbox = mlx4_alloc_cmd_mailbox(dev);
1008 if (IS_ERR(mailbox)) {
1009 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1013 gids = mailbox->buf;
1014 memcpy(gids, gw->gids, sizeof gw->gids);
1016 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1017 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1020 pr_warn("set port command failed\n");
1022 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1023 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1026 mlx4_free_cmd_mailbox(dev, mailbox);
1030 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1032 struct net_device *ndev = dev->iboe.netdevs[port - 1];
1033 struct update_gid_work *work;
1034 struct net_device *tmp;
1041 int need_update = 0;
1044 work = kzalloc(sizeof *work, GFP_ATOMIC);
1048 hits = kzalloc(128, GFP_ATOMIC);
1055 for_each_netdev_rcu(&init_net, tmp) {
1056 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1057 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1058 vid = rdma_vlan_dev_vlan_id(tmp);
1059 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1062 for (i = 0; i < 128; ++i) {
1064 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1066 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1075 (memcmp(&dev->iboe.gid_table[port - 1][0],
1076 &gid, sizeof gid) ||
1077 !memcmp(&dev->iboe.gid_table[port - 1][0],
1078 &zgid, sizeof gid))) {
1079 dev->iboe.gid_table[port - 1][0] = gid;
1082 } else if (free >= 0) {
1083 dev->iboe.gid_table[port - 1][free] = gid;
1092 for (i = 0; i < 128; ++i)
1094 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1096 dev->iboe.gid_table[port - 1][i] = zgid;
1100 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1101 INIT_WORK(&work->work, update_gids_task);
1104 queue_work(wq, &work->work);
1116 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1120 case NETDEV_CHANGEADDR:
1121 update_ipv6_gids(dev, port, 0);
1125 update_ipv6_gids(dev, port, 1);
1126 dev->iboe.netdevs[port - 1] = NULL;
1130 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1132 update_ipv6_gids(dev, port, 0);
1135 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1137 update_ipv6_gids(dev, port, 1);
1140 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1143 struct net_device *dev = ptr;
1144 struct mlx4_ib_dev *ibdev;
1145 struct net_device *oldnd;
1146 struct mlx4_ib_iboe *iboe;
1149 if (!net_eq(dev_net(dev), &init_net))
1152 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1153 iboe = &ibdev->iboe;
1155 spin_lock(&iboe->lock);
1156 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1157 oldnd = iboe->netdevs[port - 1];
1158 iboe->netdevs[port - 1] =
1159 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1160 if (oldnd != iboe->netdevs[port - 1]) {
1161 if (iboe->netdevs[port - 1])
1162 netdev_added(ibdev, port);
1164 netdev_removed(ibdev, port);
1168 if (dev == iboe->netdevs[0] ||
1169 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1170 handle_en_event(ibdev, 1, event);
1171 else if (dev == iboe->netdevs[1]
1172 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1173 handle_en_event(ibdev, 2, event);
1175 spin_unlock(&iboe->lock);
1180 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1186 if (mlx4_is_master(ibdev->dev)) {
1187 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1188 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1190 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1192 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1193 /* master has the identity virt2phys pkey mapping */
1194 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1195 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1196 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1197 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1201 /* initialize pkey cache */
1202 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1204 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1206 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1212 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1215 int eq_per_port = 0;
1220 /* Legacy mode or comp_pool is not large enough */
1221 if (dev->caps.comp_pool == 0 ||
1222 dev->caps.num_ports > dev->caps.comp_pool)
1225 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1226 dev->caps.num_ports);
1230 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1231 added_eqs += eq_per_port;
1233 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1235 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1236 if (!ibdev->eq_table)
1239 ibdev->eq_added = added_eqs;
1242 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1243 for (j = 0; j < eq_per_port; j++) {
1244 sprintf(name, "mlx4-ib-%d-%d@%s",
1245 i, j, dev->pdev->bus->name);
1246 /* Set IRQ for specific name (per ring) */
1247 if (mlx4_assign_eq(dev, name, NULL,
1248 &ibdev->eq_table[eq])) {
1249 /* Use legacy (same as mlx4_en driver) */
1250 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1251 ibdev->eq_table[eq] =
1252 (eq % dev->caps.num_comp_vectors);
1258 /* Fill the reset of the vector with legacy EQ */
1259 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1260 ibdev->eq_table[eq++] = i;
1262 /* Advertise the new number of EQs to clients */
1263 ibdev->ib_dev.num_comp_vectors = total_eqs;
1266 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1270 /* no additional eqs were added */
1271 if (!ibdev->eq_table)
1274 /* Reset the advertised EQ number */
1275 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1277 /* Free only the added eqs */
1278 for (i = 0; i < ibdev->eq_added; i++) {
1279 /* Don't free legacy eqs if used */
1280 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1282 mlx4_release_eq(dev, ibdev->eq_table[i]);
1285 kfree(ibdev->eq_table);
1288 static void *mlx4_ib_add(struct mlx4_dev *dev)
1290 struct mlx4_ib_dev *ibdev;
1294 struct mlx4_ib_iboe *iboe;
1296 pr_info_once("%s", mlx4_ib_version);
1298 mlx4_foreach_non_ib_transport_port(i, dev)
1301 if (mlx4_is_mfunc(dev) && num_ports) {
1302 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1307 mlx4_foreach_ib_transport_port(i, dev)
1310 /* No point in registering a device with no ports... */
1314 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1316 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1320 iboe = &ibdev->iboe;
1322 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1325 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1328 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1330 if (!ibdev->uar_map)
1332 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1336 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1337 ibdev->ib_dev.owner = THIS_MODULE;
1338 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1339 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1340 ibdev->num_ports = num_ports;
1341 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1342 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1343 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1345 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1346 ibdev->ib_dev.uverbs_cmd_mask =
1347 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1348 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1349 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1350 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1351 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1352 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1353 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1354 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1355 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1356 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1357 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1358 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1359 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1360 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1361 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1362 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1363 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1364 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1365 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1366 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1367 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1368 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1369 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1371 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1372 ibdev->ib_dev.query_port = mlx4_ib_query_port;
1373 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
1374 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1375 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1376 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1377 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1378 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1379 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1380 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1381 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1382 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1383 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1384 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1385 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1386 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1387 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
1388 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
1389 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1390 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1391 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1392 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
1393 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
1394 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1395 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1396 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1397 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
1398 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
1399 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
1400 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1401 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1402 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1403 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1404 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1405 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
1406 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1407 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1408 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
1409 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1410 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1411 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1413 if (!mlx4_is_slave(ibdev->dev)) {
1414 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1415 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1416 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1417 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1420 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1421 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1422 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1423 ibdev->ib_dev.uverbs_cmd_mask |=
1424 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1425 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1428 mlx4_ib_alloc_eqs(dev, ibdev);
1430 spin_lock_init(&iboe->lock);
1432 if (init_node_data(ibdev))
1435 for (i = 0; i < ibdev->num_ports; ++i) {
1436 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1437 IB_LINK_LAYER_ETHERNET) {
1438 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1440 ibdev->counters[i] = -1;
1442 ibdev->counters[i] = -1;
1445 spin_lock_init(&ibdev->sm_lock);
1446 mutex_init(&ibdev->cap_mask_mutex);
1448 if (ib_register_device(&ibdev->ib_dev, NULL))
1451 if (mlx4_ib_mad_init(ibdev))
1454 if (mlx4_ib_init_sriov(ibdev))
1457 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1458 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1459 err = register_netdevice_notifier(&iboe->nb);
1464 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1465 if (device_create_file(&ibdev->ib_dev.dev,
1466 mlx4_class_attributes[j]))
1470 ibdev->ib_active = true;
1472 if (mlx4_is_mfunc(ibdev->dev))
1475 /* create paravirt contexts for any VFs which are active */
1476 if (mlx4_is_master(ibdev->dev)) {
1477 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
1478 if (j == mlx4_master_func_num(ibdev->dev))
1480 if (mlx4_is_slave_active(ibdev->dev, j))
1481 do_slave_init(ibdev, j, 1);
1487 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1488 pr_warn("failure unregistering notifier\n");
1489 flush_workqueue(wq);
1492 mlx4_ib_close_sriov(ibdev);
1495 mlx4_ib_mad_cleanup(ibdev);
1498 ib_unregister_device(&ibdev->ib_dev);
1502 if (ibdev->counters[i - 1] != -1)
1503 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1506 iounmap(ibdev->uar_map);
1509 mlx4_uar_free(dev, &ibdev->priv_uar);
1512 mlx4_pd_free(dev, ibdev->priv_pdn);
1515 ib_dealloc_device(&ibdev->ib_dev);
1520 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1522 struct mlx4_ib_dev *ibdev = ibdev_ptr;
1525 mlx4_ib_close_sriov(ibdev);
1526 mlx4_ib_mad_cleanup(ibdev);
1527 ib_unregister_device(&ibdev->ib_dev);
1528 if (ibdev->iboe.nb.notifier_call) {
1529 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1530 pr_warn("failure unregistering notifier\n");
1531 ibdev->iboe.nb.notifier_call = NULL;
1533 iounmap(ibdev->uar_map);
1534 for (p = 0; p < ibdev->num_ports; ++p)
1535 if (ibdev->counters[p] != -1)
1536 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1537 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1538 mlx4_CLOSE_PORT(dev, p);
1540 mlx4_ib_free_eqs(dev, ibdev);
1542 mlx4_uar_free(dev, &ibdev->priv_uar);
1543 mlx4_pd_free(dev, ibdev->priv_pdn);
1544 ib_dealloc_device(&ibdev->ib_dev);
1547 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1549 struct mlx4_ib_demux_work **dm = NULL;
1550 struct mlx4_dev *dev = ibdev->dev;
1552 unsigned long flags;
1554 if (!mlx4_is_master(dev))
1557 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
1559 pr_err("failed to allocate memory for tunneling qp update\n");
1563 for (i = 0; i < dev->caps.num_ports; i++) {
1564 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
1566 pr_err("failed to allocate memory for tunneling qp update work struct\n");
1567 for (i = 0; i < dev->caps.num_ports; i++) {
1574 /* initialize or tear down tunnel QPs for the slave */
1575 for (i = 0; i < dev->caps.num_ports; i++) {
1576 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
1577 dm[i]->port = i + 1;
1578 dm[i]->slave = slave;
1579 dm[i]->do_init = do_init;
1581 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
1582 if (!ibdev->sriov.is_going_down)
1583 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
1584 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1592 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1593 enum mlx4_dev_event event, unsigned long param)
1595 struct ib_event ibev;
1596 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1597 struct mlx4_eqe *eqe = NULL;
1598 struct ib_event_work *ew;
1601 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1602 eqe = (struct mlx4_eqe *)param;
1607 case MLX4_DEV_EVENT_PORT_UP:
1608 if (p > ibdev->num_ports)
1610 if (mlx4_is_master(dev) &&
1611 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
1612 IB_LINK_LAYER_INFINIBAND) {
1613 mlx4_ib_invalidate_all_guid_record(ibdev, p);
1615 ibev.event = IB_EVENT_PORT_ACTIVE;
1618 case MLX4_DEV_EVENT_PORT_DOWN:
1619 if (p > ibdev->num_ports)
1621 ibev.event = IB_EVENT_PORT_ERR;
1624 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1625 ibdev->ib_active = false;
1626 ibev.event = IB_EVENT_DEVICE_FATAL;
1629 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1630 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1632 pr_err("failed to allocate memory for events work\n");
1636 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1637 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1639 /* need to queue only for port owner, which uses GEN_EQE */
1640 if (mlx4_is_master(dev))
1641 queue_work(wq, &ew->work);
1643 handle_port_mgmt_change_event(&ew->work);
1646 case MLX4_DEV_EVENT_SLAVE_INIT:
1647 /* here, p is the slave id */
1648 do_slave_init(ibdev, p, 1);
1651 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
1652 /* here, p is the slave id */
1653 do_slave_init(ibdev, p, 0);
1660 ibev.device = ibdev_ptr;
1661 ibev.element.port_num = (u8) p;
1663 ib_dispatch_event(&ibev);
1666 static struct mlx4_interface mlx4_ib_interface = {
1668 .remove = mlx4_ib_remove,
1669 .event = mlx4_ib_event,
1670 .protocol = MLX4_PROT_IB_IPV6
1673 static int __init mlx4_ib_init(void)
1677 wq = create_singlethread_workqueue("mlx4_ib");
1681 err = mlx4_ib_mcg_init();
1685 err = mlx4_register_interface(&mlx4_ib_interface);
1692 mlx4_ib_mcg_destroy();
1695 destroy_workqueue(wq);
1699 static void __exit mlx4_ib_cleanup(void)
1701 mlx4_unregister_interface(&mlx4_ib_interface);
1702 mlx4_ib_mcg_destroy();
1703 destroy_workqueue(wq);
1706 module_init(mlx4_ib_init);
1707 module_exit(mlx4_ib_cleanup);