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>
37 #include <linux/proc_fs.h>
40 #include <linux/init.h>
41 #include <linux/slab.h>
42 #include <linux/errno.h>
43 #include <linux/netdevice.h>
44 #include <linux/inetdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/if_vlan.h>
47 #include <linux/bitops.h>
48 #include <linux/if_ether.h>
50 #include <rdma/ib_smi.h>
51 #include <rdma/ib_user_verbs.h>
52 #include <rdma/ib_addr.h>
54 #include <linux/mlx4/driver.h>
55 #include <linux/mlx4/cmd.h>
56 #include <linux/sched.h>
61 #define DRV_NAME MLX4_IB_DRV_NAME
62 #define DRV_VERSION "1.0"
63 #define DRV_RELDATE "April 4, 2008"
65 #define MLX4_IB_DRIVER_PROC_DIR_NAME "driver/mlx4_ib"
66 #define MLX4_IB_MRS_PROC_DIR_NAME "mrs"
68 MODULE_AUTHOR("Roland Dreier");
69 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
70 MODULE_LICENSE("Dual BSD/GPL");
71 MODULE_VERSION(DRV_VERSION);
73 int mlx4_ib_sm_guid_assign = 1;
76 struct proc_dir_entry *mlx4_mrs_dir_entry;
77 static struct proc_dir_entry *mlx4_ib_driver_dir_entry;
80 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
81 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
83 static char dev_assign_str[512];
84 //module_param_string(dev_assign_str, dev_assign_str, sizeof(dev_assign_str), 0644);
85 MODULE_PARM_DESC(dev_assign_str, "Map all device function numbers to "
86 "IB device numbers following the pattern: "
87 "bb:dd.f-0,bb:dd.f-1,... (all numbers are hexadecimals)."
88 " Max supported devices - 32");
90 static const char mlx4_ib_version[] =
91 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
92 DRV_VERSION " (" DRV_RELDATE ")\n";
94 struct update_gid_work {
95 struct work_struct work;
96 union ib_gid gids[128];
97 struct mlx4_ib_dev *dev;
109 static struct dev_rec dr[MAX_DR];
111 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
113 static struct workqueue_struct *wq;
115 static void init_query_mad(struct ib_smp *mad)
117 mad->base_version = 1;
118 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
119 mad->class_version = 1;
120 mad->method = IB_MGMT_METHOD_GET;
123 static union ib_gid zgid;
125 static int mlx4_ib_query_device(struct ib_device *ibdev,
126 struct ib_device_attr *props)
128 struct mlx4_ib_dev *dev = to_mdev(ibdev);
129 struct ib_smp *in_mad = NULL;
130 struct ib_smp *out_mad = NULL;
133 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
134 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
135 if (!in_mad || !out_mad)
138 init_query_mad(in_mad);
139 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
141 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
142 1, NULL, NULL, in_mad, out_mad);
146 memset(props, 0, sizeof *props);
148 props->fw_ver = dev->dev->caps.fw_ver;
149 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
150 IB_DEVICE_PORT_ACTIVE_EVENT |
151 IB_DEVICE_SYS_IMAGE_GUID |
152 IB_DEVICE_RC_RNR_NAK_GEN |
153 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK |
156 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
157 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
158 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
159 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
160 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
161 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
162 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
163 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
164 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
165 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
166 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
167 props->device_cap_flags |= IB_DEVICE_UD_TSO;
168 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
169 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
170 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
171 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
172 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
173 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
174 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
175 props->device_cap_flags |= IB_DEVICE_XRC;
177 props->device_cap_flags |= IB_DEVICE_QPG;
178 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
179 props->device_cap_flags |= IB_DEVICE_UD_RSS;
180 props->max_rss_tbl_sz = dev->dev->caps.max_rss_tbl_sz;
182 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
184 props->vendor_part_id = dev->dev->pdev->device;
185 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
186 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
188 props->max_mr_size = ~0ull;
189 props->page_size_cap = dev->dev->caps.page_size_cap;
190 props->max_qp = dev->dev->quotas.qp;
191 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
192 props->max_sge = min(dev->dev->caps.max_sq_sg,
193 dev->dev->caps.max_rq_sg);
194 props->max_cq = dev->dev->quotas.cq;
195 props->max_cqe = dev->dev->caps.max_cqes;
196 props->max_mr = dev->dev->quotas.mpt;
197 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
198 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
199 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
200 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
201 props->max_srq = dev->dev->quotas.srq;
202 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
203 props->max_srq_sge = dev->dev->caps.max_srq_sge;
204 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
205 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
206 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
207 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
208 props->masked_atomic_cap = props->atomic_cap;
209 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
210 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
211 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
212 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
213 props->max_mcast_grp;
214 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
223 static enum rdma_link_layer
224 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
226 struct mlx4_dev *dev = to_mdev(device)->dev;
228 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
229 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
232 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
233 struct ib_port_attr *props, int netw_view)
235 struct ib_smp *in_mad = NULL;
236 struct ib_smp *out_mad = NULL;
237 int ext_active_speed;
238 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
241 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
242 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
243 if (!in_mad || !out_mad)
246 init_query_mad(in_mad);
247 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
248 in_mad->attr_mod = cpu_to_be32(port);
250 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
251 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
253 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
259 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
260 props->lmc = out_mad->data[34] & 0x7;
261 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
262 props->sm_sl = out_mad->data[36] & 0xf;
263 props->state = out_mad->data[32] & 0xf;
264 props->phys_state = out_mad->data[33] >> 4;
265 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
267 props->gid_tbl_len = out_mad->data[50];
269 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
270 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
271 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
272 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
273 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
274 props->active_width = out_mad->data[31] & 0xf;
275 props->active_speed = out_mad->data[35] >> 4;
276 props->max_mtu = out_mad->data[41] & 0xf;
277 props->active_mtu = out_mad->data[36] >> 4;
278 props->subnet_timeout = out_mad->data[51] & 0x1f;
279 props->max_vl_num = out_mad->data[37] >> 4;
280 props->init_type_reply = out_mad->data[41] >> 4;
282 /* Check if extended speeds (EDR/FDR/...) are supported */
283 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
284 ext_active_speed = out_mad->data[62] >> 4;
286 switch (ext_active_speed) {
288 props->active_speed = IB_SPEED_FDR;
291 props->active_speed = IB_SPEED_EDR;
296 /* If reported active speed is QDR, check if is FDR-10 */
297 if (props->active_speed == IB_SPEED_QDR) {
298 init_query_mad(in_mad);
299 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
300 in_mad->attr_mod = cpu_to_be32(port);
302 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
303 NULL, NULL, in_mad, out_mad);
307 /* Checking LinkSpeedActive for FDR-10 */
308 if (out_mad->data[15] & 0x1)
309 props->active_speed = IB_SPEED_FDR10;
312 /* Avoid wrong speed value returned by FW if the IB link is down. */
313 if (props->state == IB_PORT_DOWN)
314 props->active_speed = IB_SPEED_SDR;
322 static u8 state_to_phys_state(enum ib_port_state state)
324 return state == IB_PORT_ACTIVE ? 5 : 3;
327 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
328 struct ib_port_attr *props, int netw_view)
331 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
332 struct mlx4_ib_iboe *iboe = &mdev->iboe;
333 struct net_device *ndev;
335 struct mlx4_cmd_mailbox *mailbox;
338 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
340 return PTR_ERR(mailbox);
342 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
343 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
348 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
349 IB_WIDTH_4X : IB_WIDTH_1X;
350 props->active_speed = IB_SPEED_QDR;
351 props->port_cap_flags = IB_PORT_CM_SUP;
353 props->gid_tbl_len = MLX4_ROCE_MAX_GIDS;
355 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
357 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
358 props->pkey_tbl_len = 1;
359 props->max_mtu = IB_MTU_4096;
360 props->max_vl_num = 2;
361 props->state = IB_PORT_DOWN;
362 props->phys_state = state_to_phys_state(props->state);
363 props->active_mtu = IB_MTU_256;
364 spin_lock(&iboe->lock);
365 ndev = iboe->netdevs[port - 1];
369 tmp = iboe_get_mtu(ndev->if_mtu);
370 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
372 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
373 IB_PORT_ACTIVE : IB_PORT_DOWN;
374 props->phys_state = state_to_phys_state(props->state);
376 spin_unlock(&iboe->lock);
378 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
382 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
383 struct ib_port_attr *props, int netw_view)
387 memset(props, 0, sizeof *props);
389 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
390 ib_link_query_port(ibdev, port, props, netw_view) :
391 eth_link_query_port(ibdev, port, props, netw_view);
396 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
397 struct ib_port_attr *props)
399 /* returns host view */
400 return __mlx4_ib_query_port(ibdev, port, props, 0);
403 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
404 union ib_gid *gid, int netw_view)
406 struct ib_smp *in_mad = NULL;
407 struct ib_smp *out_mad = NULL;
409 struct mlx4_ib_dev *dev = to_mdev(ibdev);
411 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
413 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
414 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
415 if (!in_mad || !out_mad)
418 init_query_mad(in_mad);
419 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
420 in_mad->attr_mod = cpu_to_be32(port);
422 if (mlx4_is_mfunc(dev->dev) && netw_view)
423 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
425 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
429 memcpy(gid->raw, out_mad->data + 8, 8);
431 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
433 /* For any index > 0, return the null guid */
440 init_query_mad(in_mad);
441 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
442 in_mad->attr_mod = cpu_to_be32(index / 8);
444 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
445 NULL, NULL, in_mad, out_mad);
449 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
453 memset(gid->raw + 8, 0, 8);
459 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
462 struct mlx4_ib_dev *dev = to_mdev(ibdev);
464 *gid = dev->iboe.gid_table[port - 1][index];
469 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
472 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
473 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
475 return iboe_query_gid(ibdev, port, index, gid);
478 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
479 u16 *pkey, int netw_view)
481 struct ib_smp *in_mad = NULL;
482 struct ib_smp *out_mad = NULL;
483 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
486 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
487 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
488 if (!in_mad || !out_mad)
491 init_query_mad(in_mad);
492 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
493 in_mad->attr_mod = cpu_to_be32(index / 32);
495 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
496 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
498 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
503 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
511 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
513 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
516 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
517 struct ib_device_modify *props)
519 struct mlx4_cmd_mailbox *mailbox;
522 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
525 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
528 if (mlx4_is_slave(to_mdev(ibdev)->dev))
531 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
532 memcpy(ibdev->node_desc, props->node_desc, 64);
533 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
536 * If possible, pass node desc to FW, so it can generate
537 * a 144 trap. If cmd fails, just ignore.
539 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
543 memset(mailbox->buf, 0, 256);
544 memcpy(mailbox->buf, props->node_desc, 64);
545 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
546 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
548 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
553 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
556 struct mlx4_cmd_mailbox *mailbox;
558 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
560 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
562 return PTR_ERR(mailbox);
564 memset(mailbox->buf, 0, 256);
566 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
567 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
568 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
570 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
571 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
574 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
575 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
577 mlx4_free_cmd_mailbox(dev->dev, mailbox);
581 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
582 struct ib_port_modify *props)
584 struct ib_port_attr attr;
588 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
590 err = mlx4_ib_query_port(ibdev, port, &attr);
594 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
595 ~props->clr_port_cap_mask;
597 err = mlx4_SET_PORT(to_mdev(ibdev), port,
598 !!(mask & IB_PORT_RESET_QKEY_CNTR),
602 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
606 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
607 struct ib_udata *udata)
609 struct mlx4_ib_dev *dev = to_mdev(ibdev);
610 struct mlx4_ib_ucontext *context;
611 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
612 struct mlx4_ib_alloc_ucontext_resp resp;
616 return ERR_PTR(-EAGAIN);
618 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
619 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
620 if (mlx4_wc_enabled()) {
621 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
622 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
624 resp_v3.bf_reg_size = 0;
625 resp_v3.bf_regs_per_page = 0;
628 resp.dev_caps = dev->dev->caps.userspace_caps;
629 resp.qp_tab_size = dev->dev->caps.num_qps;
630 if (mlx4_wc_enabled()) {
631 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
632 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
634 resp.bf_reg_size = 0;
635 resp.bf_regs_per_page = 0;
637 resp.cqe_size = dev->dev->caps.cqe_size;
640 context = kmalloc(sizeof *context, GFP_KERNEL);
642 return ERR_PTR(-ENOMEM);
644 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
650 INIT_LIST_HEAD(&context->db_page_list);
651 mutex_init(&context->db_page_mutex);
653 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
654 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
656 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
659 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
661 return ERR_PTR(-EFAULT);
664 return &context->ibucontext;
667 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
669 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
671 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
677 static unsigned long mlx4_ib_get_unmapped_area(struct file *file,
679 unsigned long len, unsigned long pgoff,
682 struct mm_struct *mm;
683 struct vm_area_struct *vma;
684 unsigned long start_addr;
685 unsigned long page_size_order;
686 unsigned long command;
690 return current->mm->get_unmapped_area(file, addr, len,
693 /* Last 8 bits hold the command others are data per that command */
694 command = pgoff & MLX4_IB_MMAP_CMD_MASK;
695 if (command != MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES)
696 return current->mm->get_unmapped_area(file, addr, len,
699 page_size_order = pgoff >> MLX4_IB_MMAP_CMD_BITS;
700 /* code is based on the huge-pages get_unmapped_area code */
701 start_addr = mm->free_area_cache;
703 if (len <= mm->cached_hole_size)
704 start_addr = TASK_UNMAPPED_BASE;
708 addr = ALIGN(start_addr, 1 << page_size_order);
710 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
711 /* At this point: (!vma || addr < vma->vm_end). */
712 if (TASK_SIZE - len < addr) {
714 * Start a new search - just in case we missed
717 if (start_addr != TASK_UNMAPPED_BASE) {
718 start_addr = TASK_UNMAPPED_BASE;
724 if (!vma || addr + len <= vma->vm_start)
726 addr = ALIGN(vma->vm_end, 1 << page_size_order);
731 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
733 struct mlx4_ib_dev *dev = to_mdev(context->device);
736 /* Last 8 bits hold the command others are data per that command */
737 unsigned long command = vma->vm_pgoff & MLX4_IB_MMAP_CMD_MASK;
739 if (command < MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES) {
740 /* compatability handling for commands 0 & 1*/
741 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
744 if (command == MLX4_IB_MMAP_UAR_PAGE) {
745 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
747 if (io_remap_pfn_range(vma, vma->vm_start,
748 to_mucontext(context)->uar.pfn,
749 PAGE_SIZE, vma->vm_page_prot))
751 } else if (command == MLX4_IB_MMAP_BLUE_FLAME_PAGE &&
752 dev->dev->caps.bf_reg_size != 0) {
753 vma->vm_page_prot = pgprot_wc(vma->vm_page_prot);
755 if (io_remap_pfn_range(vma, vma->vm_start,
756 to_mucontext(context)->uar.pfn +
757 dev->dev->caps.num_uars,
758 PAGE_SIZE, vma->vm_page_prot))
760 } else if (command == MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES) {
761 /* Getting contiguous physical pages */
762 unsigned long total_size = vma->vm_end - vma->vm_start;
763 unsigned long page_size_order = (vma->vm_pgoff) >>
764 MLX4_IB_MMAP_CMD_BITS;
765 struct ib_cmem *ib_cmem;
766 ib_cmem = ib_cmem_alloc_contiguous_pages(context, total_size,
768 if (IS_ERR(ib_cmem)) {
769 err = PTR_ERR(ib_cmem);
773 err = ib_cmem_map_contiguous_pages_to_vma(ib_cmem, vma);
775 ib_cmem_release_contiguous_pages(ib_cmem);
785 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
786 struct ib_ucontext *context,
787 struct ib_udata *udata)
789 struct mlx4_ib_pd *pd;
792 pd = kmalloc(sizeof *pd, GFP_KERNEL);
794 return ERR_PTR(-ENOMEM);
796 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
803 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
804 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
806 return ERR_PTR(-EFAULT);
812 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
814 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
820 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
821 struct ib_ucontext *context,
822 struct ib_udata *udata)
824 struct mlx4_ib_xrcd *xrcd;
827 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
828 return ERR_PTR(-ENOSYS);
830 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
832 return ERR_PTR(-ENOMEM);
834 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
838 xrcd->pd = ib_alloc_pd(ibdev);
839 if (IS_ERR(xrcd->pd)) {
840 err = PTR_ERR(xrcd->pd);
844 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
845 if (IS_ERR(xrcd->cq)) {
846 err = PTR_ERR(xrcd->cq);
850 return &xrcd->ibxrcd;
853 ib_dealloc_pd(xrcd->pd);
855 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
861 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
863 ib_destroy_cq(to_mxrcd(xrcd)->cq);
864 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
865 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
871 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
873 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
874 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
875 struct mlx4_ib_gid_entry *ge;
877 ge = kzalloc(sizeof *ge, GFP_KERNEL);
882 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
883 ge->port = mqp->port;
887 mutex_lock(&mqp->mutex);
888 list_add_tail(&ge->list, &mqp->gid_list);
889 mutex_unlock(&mqp->mutex);
894 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
898 struct net_device *ndev;
904 spin_lock(&mdev->iboe.lock);
905 ndev = mdev->iboe.netdevs[mqp->port - 1];
908 spin_unlock(&mdev->iboe.lock);
911 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
913 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac, 6, 0);
922 struct mlx4_ib_steering {
923 struct list_head list;
928 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
931 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
932 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
934 struct mlx4_ib_steering *ib_steering = NULL;
936 if (mdev->dev->caps.steering_mode ==
937 MLX4_STEERING_MODE_DEVICE_MANAGED) {
938 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
943 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
945 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
946 MLX4_PROT_IB_IPV6, ®_id);
950 err = add_gid_entry(ibqp, gid);
955 memcpy(ib_steering->gid.raw, gid->raw, 16);
956 ib_steering->reg_id = reg_id;
957 mutex_lock(&mqp->mutex);
958 list_add(&ib_steering->list, &mqp->steering_rules);
959 mutex_unlock(&mqp->mutex);
964 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
965 MLX4_PROT_IB_IPV6, reg_id);
973 IBV_FLOW_L4_NONE = 0,
974 IBV_FLOW_L4_OTHER = 3,
979 struct mlx4_cm_steering {
980 struct list_head list;
982 struct ib_flow_spec spec;
985 static int flow_spec_to_net_rule(struct ib_device *dev, struct ib_flow_spec *flow_spec,
986 struct list_head *rule_list_h)
988 struct mlx4_spec_list *spec_l2, *spec_l3, *spec_l4;
989 u64 mac_msk = cpu_to_be64(MLX4_MAC_MASK << 16);
991 spec_l2 = kzalloc(sizeof *spec_l2, GFP_KERNEL);
995 switch (flow_spec->type) {
997 spec_l2->id = MLX4_NET_TRANS_RULE_ID_ETH;
998 memcpy(spec_l2->eth.dst_mac, flow_spec->l2_id.eth.mac, ETH_ALEN);
999 memcpy(spec_l2->eth.dst_mac_msk, &mac_msk, ETH_ALEN);
1000 spec_l2->eth.ether_type = flow_spec->l2_id.eth.ethertype;
1001 if (flow_spec->l2_id.eth.vlan_present) {
1002 spec_l2->eth.vlan_id = flow_spec->l2_id.eth.vlan;
1003 spec_l2->eth.vlan_id_msk = cpu_to_be16(0x0fff);
1007 spec_l2->id = MLX4_NET_TRANS_RULE_ID_IB;
1008 if(flow_spec->l2_id.ib_uc.qpn) {
1009 spec_l2->ib.r_u_qpn = cpu_to_be32(flow_spec->l2_id.ib_uc.qpn);
1010 spec_l2->ib.qpn_msk = cpu_to_be32(0xffffff);
1013 case IB_FLOW_IB_MC_IPV4:
1014 case IB_FLOW_IB_MC_IPV6:
1015 spec_l2->id = MLX4_NET_TRANS_RULE_ID_IB;
1016 memcpy(spec_l2->ib.dst_gid, flow_spec->l2_id.ib_mc.mgid, 16);
1017 memset(spec_l2->ib.dst_gid_msk, 0xff, 16);
1022 list_add_tail(&spec_l2->list, rule_list_h);
1024 if (flow_spec->l2_id.eth.ethertype == cpu_to_be16(ETH_P_IP) ||
1025 flow_spec->type != IB_FLOW_ETH) {
1026 spec_l3 = kzalloc(sizeof *spec_l3, GFP_KERNEL);
1030 spec_l3->id = MLX4_NET_TRANS_RULE_ID_IPV4;
1031 spec_l3->ipv4.src_ip = flow_spec->src_ip;
1032 if (flow_spec->type != IB_FLOW_IB_MC_IPV4 &&
1033 flow_spec->type != IB_FLOW_IB_MC_IPV6)
1034 spec_l3->ipv4.dst_ip = flow_spec->dst_ip;
1036 if (spec_l3->ipv4.src_ip)
1037 spec_l3->ipv4.src_ip_msk = MLX4_BE_WORD_MASK;
1038 if (spec_l3->ipv4.dst_ip)
1039 spec_l3->ipv4.dst_ip_msk = MLX4_BE_WORD_MASK;
1041 list_add_tail(&spec_l3->list, rule_list_h);
1044 if (flow_spec->l4_protocol) {
1045 spec_l4 = kzalloc(sizeof(*spec_l4), GFP_KERNEL);
1049 spec_l4->tcp_udp.src_port = flow_spec->src_port;
1050 spec_l4->tcp_udp.dst_port = flow_spec->dst_port;
1051 if (spec_l4->tcp_udp.src_port)
1052 spec_l4->tcp_udp.src_port_msk =
1054 if (spec_l4->tcp_udp.dst_port)
1055 spec_l4->tcp_udp.dst_port_msk =
1058 switch (flow_spec->l4_protocol) {
1059 case IBV_FLOW_L4_UDP:
1060 spec_l4->id = MLX4_NET_TRANS_RULE_ID_UDP;
1062 case IBV_FLOW_L4_TCP:
1063 spec_l4->id = MLX4_NET_TRANS_RULE_ID_TCP;
1066 dev_err(dev->dma_device,
1067 "Unsupported l4 protocol.\n");
1069 return -EPROTONOSUPPORT;
1071 list_add_tail(&spec_l4->list, rule_list_h);
1076 static int __mlx4_ib_flow_attach(struct mlx4_ib_dev *mdev,
1077 struct mlx4_ib_qp *mqp,
1078 struct ib_flow_spec *flow_spec,
1079 int priority, int lock_qp)
1083 struct mlx4_cm_steering *cm_flow;
1084 struct mlx4_spec_list *spec, *tmp_spec;
1086 struct mlx4_net_trans_rule rule =
1087 { .queue_mode = MLX4_NET_TRANS_Q_FIFO,
1091 rule.promisc_mode = flow_spec->rule_type;
1092 rule.port = mqp->port;
1093 rule.qpn = mqp->mqp.qpn;
1094 INIT_LIST_HEAD(&rule.list);
1096 cm_flow = kmalloc(sizeof(*cm_flow), GFP_KERNEL);
1100 if (rule.promisc_mode == MLX4_FS_REGULAR) {
1101 rule.allow_loopback = !flow_spec->block_mc_loopback;
1102 rule.priority = MLX4_DOMAIN_UVERBS | priority;
1103 err = flow_spec_to_net_rule(&mdev->ib_dev, flow_spec,
1109 err = mlx4_flow_attach(mdev->dev, &rule, ®_id);
1113 memcpy(&cm_flow->spec, flow_spec, sizeof(*flow_spec));
1114 cm_flow->reg_id = reg_id;
1117 mutex_lock(&mqp->mutex);
1118 list_add(&cm_flow->list, &mqp->rules_list);
1120 mutex_unlock(&mqp->mutex);
1123 list_for_each_entry_safe(spec, tmp_spec, &rule.list, list) {
1124 list_del(&spec->list);
1129 dev_err(mdev->ib_dev.dma_device,
1130 "Fail to attach flow steering rule\n");
1135 static int __mlx4_ib_flow_detach(struct mlx4_ib_dev *mdev,
1136 struct mlx4_ib_qp *mqp,
1137 struct ib_flow_spec *spec, int priority,
1140 struct mlx4_cm_steering *cm_flow;
1144 mutex_lock(&mqp->mutex);
1145 list_for_each_entry(cm_flow, &mqp->rules_list, list) {
1146 if (!memcmp(&cm_flow->spec, spec, sizeof(*spec))) {
1147 list_del(&cm_flow->list);
1152 mutex_unlock(&mqp->mutex);
1154 if (&cm_flow->list == &mqp->rules_list) {
1155 dev_err(mdev->ib_dev.dma_device, "Couldn't find reg_id for flow spec. "
1156 "Steering rule is left attached\n");
1160 ret = mlx4_flow_detach(mdev->dev, cm_flow->reg_id);
1166 static int mlx4_ib_flow_attach(struct ib_qp *qp, struct ib_flow_spec *flow_spec,
1169 return __mlx4_ib_flow_attach(to_mdev(qp->device), to_mqp(qp),
1170 flow_spec, priority, 1);
1173 static int mlx4_ib_flow_detach(struct ib_qp *qp, struct ib_flow_spec *spec,
1176 return __mlx4_ib_flow_detach(to_mdev(qp->device), to_mqp(qp),
1180 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1182 struct mlx4_ib_gid_entry *ge;
1183 struct mlx4_ib_gid_entry *tmp;
1184 struct mlx4_ib_gid_entry *ret = NULL;
1186 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1187 if (!memcmp(raw, ge->gid.raw, 16)) {
1196 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1199 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1200 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1202 struct net_device *ndev;
1203 struct mlx4_ib_gid_entry *ge;
1206 if (mdev->dev->caps.steering_mode ==
1207 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1208 struct mlx4_ib_steering *ib_steering;
1210 mutex_lock(&mqp->mutex);
1211 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1212 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1213 list_del(&ib_steering->list);
1217 mutex_unlock(&mqp->mutex);
1218 if (&ib_steering->list == &mqp->steering_rules) {
1219 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1222 reg_id = ib_steering->reg_id;
1226 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1227 MLX4_PROT_IB_IPV6, reg_id);
1231 mutex_lock(&mqp->mutex);
1232 ge = find_gid_entry(mqp, gid->raw);
1234 spin_lock(&mdev->iboe.lock);
1235 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1238 spin_unlock(&mdev->iboe.lock);
1239 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
1242 dev_mc_delete(mdev->iboe.netdevs[ge->port - 1], mac, 6, 0);
1246 list_del(&ge->list);
1249 pr_warn("could not find mgid entry\n");
1251 mutex_unlock(&mqp->mutex);
1256 static int init_node_data(struct mlx4_ib_dev *dev)
1258 struct ib_smp *in_mad = NULL;
1259 struct ib_smp *out_mad = NULL;
1260 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1263 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1264 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1265 if (!in_mad || !out_mad)
1268 init_query_mad(in_mad);
1269 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1270 if (mlx4_is_master(dev->dev))
1271 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1273 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1277 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1279 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1281 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1285 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1286 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1294 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1297 struct mlx4_ib_dev *dev =
1298 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1299 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1302 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1305 struct mlx4_ib_dev *dev =
1306 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1307 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1308 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1309 (int) dev->dev->caps.fw_ver & 0xffff);
1312 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1315 struct mlx4_ib_dev *dev =
1316 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1317 return sprintf(buf, "%x\n", dev->dev->rev_id);
1320 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1323 struct mlx4_ib_dev *dev =
1324 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1325 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1326 dev->dev->board_id);
1329 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1330 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1331 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1332 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1334 static struct device_attribute *mlx4_class_attributes[] = {
1341 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
1344 memcpy(eui, dev->dev_addr, 3);
1345 memcpy(eui + 5, dev->dev_addr + 3, 3);
1347 memcpy(eui, IF_LLADDR(dev), 3);
1348 memcpy(eui + 5, IF_LLADDR(dev) + 3, 3);
1350 if (vlan_id < 0x1000) {
1351 eui[3] = vlan_id >> 8;
1352 eui[4] = vlan_id & 0xff;
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 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1384 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1387 mlx4_free_cmd_mailbox(dev, mailbox);
1391 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1393 struct net_device *ndev = dev->iboe.netdevs[port - 1];
1394 struct update_gid_work *work;
1395 struct net_device *tmp;
1401 int need_update = 0;
1405 work = kzalloc(sizeof *work, GFP_ATOMIC);
1409 hits = kzalloc(128, GFP_ATOMIC);
1415 max_gids = dev->dev->caps.gid_table_len[port];
1419 for_each_netdev_rcu(&init_net, tmp) {
1422 TAILQ_FOREACH(tmp, &V_ifnet, if_link) {
1424 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1425 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1426 vid = rdma_vlan_dev_vlan_id(tmp);
1427 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1430 for (i = 0; i < max_gids; ++i) {
1431 if (index_free < 0 &&
1432 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1434 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1443 (memcmp(&dev->iboe.gid_table[port - 1][0],
1444 &gid, sizeof gid) ||
1445 !memcmp(&dev->iboe.gid_table[port - 1][0],
1446 &zgid, sizeof gid))) {
1447 dev->iboe.gid_table[port - 1][0] = gid;
1450 } else if (index_free >= 0) {
1451 dev->iboe.gid_table[port - 1][index_free] = gid;
1452 hits[index_free] = 1;
1465 for (i = 0; i < max_gids; ++i)
1467 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1469 dev->iboe.gid_table[port - 1][i] = zgid;
1473 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1474 INIT_WORK(&work->work, update_gids_task);
1477 queue_work(wq, &work->work);
1485 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1490 case NETDEV_CHANGEADDR:
1492 update_ipv6_gids(dev, port, 0);
1496 update_ipv6_gids(dev, port, 1);
1497 dev->iboe.netdevs[port - 1] = NULL;
1501 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1503 update_ipv6_gids(dev, port, 0);
1506 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1508 update_ipv6_gids(dev, port, 1);
1511 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1514 struct net_device *dev = ptr;
1515 struct mlx4_ib_dev *ibdev;
1516 struct net_device *oldnd;
1517 struct mlx4_ib_iboe *iboe;
1521 if (!net_eq(dev_net(dev), &init_net))
1525 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1526 iboe = &ibdev->iboe;
1528 spin_lock(&iboe->lock);
1529 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1530 oldnd = iboe->netdevs[port - 1];
1531 iboe->netdevs[port - 1] =
1532 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1533 if (oldnd != iboe->netdevs[port - 1]) {
1534 if (iboe->netdevs[port - 1])
1535 netdev_added(ibdev, port);
1537 netdev_removed(ibdev, port);
1541 if (dev == iboe->netdevs[0] ||
1542 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1543 handle_en_event(ibdev, 1, event);
1544 else if (dev == iboe->netdevs[1]
1545 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1546 handle_en_event(ibdev, 2, event);
1548 spin_unlock(&iboe->lock);
1553 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1559 if (mlx4_is_master(ibdev->dev)) {
1560 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1561 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1563 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1565 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1566 /* master has the identity virt2phys pkey mapping */
1567 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1568 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1569 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1570 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1574 /* initialize pkey cache */
1575 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1577 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1579 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1585 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1588 int eq_per_port = 0;
1593 /* Legacy mode or comp_pool is not large enough */
1594 if (dev->caps.comp_pool == 0 ||
1595 dev->caps.num_ports > dev->caps.comp_pool)
1598 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1599 dev->caps.num_ports);
1603 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1604 added_eqs += eq_per_port;
1606 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1608 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1609 if (!ibdev->eq_table)
1612 ibdev->eq_added = added_eqs;
1615 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1616 for (j = 0; j < eq_per_port; j++) {
1617 //sprintf(name, "mlx4-ib-%d-%d@%s",
1618 // i, j, dev->pdev->bus->conf.pd_name);
1619 /* Set IRQ for specific name (per ring) */
1620 if (mlx4_assign_eq(dev, name,
1621 &ibdev->eq_table[eq])) {
1622 /* Use legacy (same as mlx4_en driver) */
1623 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1624 ibdev->eq_table[eq] =
1625 (eq % dev->caps.num_comp_vectors);
1631 /* Fill the reset of the vector with legacy EQ */
1632 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1633 ibdev->eq_table[eq++] = i;
1635 /* Advertise the new number of EQs to clients */
1636 ibdev->ib_dev.num_comp_vectors = total_eqs;
1639 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1643 /* no additional eqs were added */
1644 if (!ibdev->eq_table)
1647 /* Reset the advertised EQ number */
1648 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1650 /* Free only the added eqs */
1651 for (i = 0; i < ibdev->eq_added; i++) {
1652 /* Don't free legacy eqs if used */
1653 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1655 mlx4_release_eq(dev, ibdev->eq_table[i]);
1658 kfree(ibdev->eq_table);
1662 * create show function and a device_attribute struct pointing to
1663 * the function for _name
1665 #define DEVICE_DIAG_RPRT_ATTR(_name, _offset, _op_mod) \
1666 static ssize_t show_rprt_##_name(struct device *dev, \
1667 struct device_attribute *attr, \
1669 return show_diag_rprt(dev, buf, _offset, _op_mod); \
1671 static DEVICE_ATTR(_name, S_IRUGO, show_rprt_##_name, NULL);
1673 #define MLX4_DIAG_RPRT_CLEAR_DIAGS 3
1675 static size_t show_diag_rprt(struct device *device, char *buf,
1676 u32 offset, u8 op_modifier)
1679 u32 counter_offset = offset;
1680 u32 diag_counter = 0;
1681 struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
1684 ret = mlx4_query_diag_counters(dev->dev, 1, op_modifier,
1685 &counter_offset, &diag_counter);
1689 return sprintf(buf, "%d\n", diag_counter);
1692 static ssize_t clear_diag_counters(struct device *device,
1693 struct device_attribute *attr,
1694 const char *buf, size_t length)
1697 struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
1700 ret = mlx4_query_diag_counters(dev->dev, 0, MLX4_DIAG_RPRT_CLEAR_DIAGS,
1708 DEVICE_DIAG_RPRT_ATTR(rq_num_lle , 0x00, 2);
1709 DEVICE_DIAG_RPRT_ATTR(sq_num_lle , 0x04, 2);
1710 DEVICE_DIAG_RPRT_ATTR(rq_num_lqpoe , 0x08, 2);
1711 DEVICE_DIAG_RPRT_ATTR(sq_num_lqpoe , 0x0C, 2);
1712 DEVICE_DIAG_RPRT_ATTR(rq_num_lpe , 0x18, 2);
1713 DEVICE_DIAG_RPRT_ATTR(sq_num_lpe , 0x1C, 2);
1714 DEVICE_DIAG_RPRT_ATTR(rq_num_wrfe , 0x20, 2);
1715 DEVICE_DIAG_RPRT_ATTR(sq_num_wrfe , 0x24, 2);
1716 DEVICE_DIAG_RPRT_ATTR(sq_num_mwbe , 0x2C, 2);
1717 DEVICE_DIAG_RPRT_ATTR(sq_num_bre , 0x34, 2);
1718 DEVICE_DIAG_RPRT_ATTR(rq_num_lae , 0x38, 2);
1719 DEVICE_DIAG_RPRT_ATTR(sq_num_rire , 0x44, 2);
1720 DEVICE_DIAG_RPRT_ATTR(rq_num_rire , 0x48, 2);
1721 DEVICE_DIAG_RPRT_ATTR(sq_num_rae , 0x4C, 2);
1722 DEVICE_DIAG_RPRT_ATTR(rq_num_rae , 0x50, 2);
1723 DEVICE_DIAG_RPRT_ATTR(sq_num_roe , 0x54, 2);
1724 DEVICE_DIAG_RPRT_ATTR(sq_num_tree , 0x5C, 2);
1725 DEVICE_DIAG_RPRT_ATTR(sq_num_rree , 0x64, 2);
1726 DEVICE_DIAG_RPRT_ATTR(rq_num_rnr , 0x68, 2);
1727 DEVICE_DIAG_RPRT_ATTR(sq_num_rnr , 0x6C, 2);
1728 DEVICE_DIAG_RPRT_ATTR(rq_num_oos , 0x100, 2);
1729 DEVICE_DIAG_RPRT_ATTR(sq_num_oos , 0x104, 2);
1730 DEVICE_DIAG_RPRT_ATTR(rq_num_mce , 0x108, 2);
1731 DEVICE_DIAG_RPRT_ATTR(rq_num_udsdprd , 0x118, 2);
1732 DEVICE_DIAG_RPRT_ATTR(rq_num_ucsdprd , 0x120, 2);
1733 DEVICE_DIAG_RPRT_ATTR(num_cqovf , 0x1A0, 2);
1734 DEVICE_DIAG_RPRT_ATTR(num_eqovf , 0x1A4, 2);
1735 DEVICE_DIAG_RPRT_ATTR(num_baddb , 0x1A8, 2);
1737 static DEVICE_ATTR(clear_diag, S_IWUSR, NULL, clear_diag_counters);
1739 static struct attribute *diag_rprt_attrs[] = {
1740 &dev_attr_rq_num_lle.attr,
1741 &dev_attr_sq_num_lle.attr,
1742 &dev_attr_rq_num_lqpoe.attr,
1743 &dev_attr_sq_num_lqpoe.attr,
1744 &dev_attr_rq_num_lpe.attr,
1745 &dev_attr_sq_num_lpe.attr,
1746 &dev_attr_rq_num_wrfe.attr,
1747 &dev_attr_sq_num_wrfe.attr,
1748 &dev_attr_sq_num_mwbe.attr,
1749 &dev_attr_sq_num_bre.attr,
1750 &dev_attr_rq_num_lae.attr,
1751 &dev_attr_sq_num_rire.attr,
1752 &dev_attr_rq_num_rire.attr,
1753 &dev_attr_sq_num_rae.attr,
1754 &dev_attr_rq_num_rae.attr,
1755 &dev_attr_sq_num_roe.attr,
1756 &dev_attr_sq_num_tree.attr,
1757 &dev_attr_sq_num_rree.attr,
1758 &dev_attr_rq_num_rnr.attr,
1759 &dev_attr_sq_num_rnr.attr,
1760 &dev_attr_rq_num_oos.attr,
1761 &dev_attr_sq_num_oos.attr,
1762 &dev_attr_rq_num_mce.attr,
1763 &dev_attr_rq_num_udsdprd.attr,
1764 &dev_attr_rq_num_ucsdprd.attr,
1765 &dev_attr_num_cqovf.attr,
1766 &dev_attr_num_eqovf.attr,
1767 &dev_attr_num_baddb.attr,
1768 &dev_attr_clear_diag.attr,
1772 static struct attribute_group diag_counters_group = {
1773 .name = "diag_counters",
1774 .attrs = diag_rprt_attrs
1778 static int mlx4_ib_proc_init(void)
1780 /* Creating procfs directories /proc/drivers/mlx4_ib/ &&
1781 /proc/drivers/mlx4_ib/mrs for further use by the driver.
1785 mlx4_ib_driver_dir_entry = proc_mkdir(MLX4_IB_DRIVER_PROC_DIR_NAME,
1787 if (!mlx4_ib_driver_dir_entry) {
1788 pr_err("mlx4_ib_proc_init has failed for %s\n",
1789 MLX4_IB_DRIVER_PROC_DIR_NAME);
1794 mlx4_mrs_dir_entry = proc_mkdir(MLX4_IB_MRS_PROC_DIR_NAME,
1795 mlx4_ib_driver_dir_entry);
1796 if (!mlx4_mrs_dir_entry) {
1797 pr_err("mlx4_ib_proc_init has failed for %s\n",
1798 MLX4_IB_MRS_PROC_DIR_NAME);
1806 remove_proc_entry(MLX4_IB_DRIVER_PROC_DIR_NAME,
1813 static void init_dev_assign(void)
1815 int bus, slot, fn, ib_idx;
1816 char *p = dev_assign_str, *t;
1817 char curr_val[32] = {0};
1821 memset(dr, 0, sizeof dr);
1823 if (dev_assign_str[0] == 0)
1827 ret = sscanf(p, "%02x:%02x.%x-%x", &bus, &slot, &fn, &ib_idx);
1828 if (ret != 4 || ib_idx < 0)
1831 for (j = 0; j < i; j++)
1832 if (dr[j].nr == ib_idx)
1841 sprintf(curr_val, "%02x:%02x.%x-%x", bus, slot, fn, ib_idx);
1842 if ((!t) && strlen(p) == strlen(curr_val))
1845 if (!t || (t + 1) >= dev_assign_str + sizeof dev_assign_str)
1857 memset(dr, 0, sizeof dr);
1858 printk(KERN_WARNING "mlx4_ib: The value of 'dev_assign_str' parameter "
1859 "is incorrect. The parameter value is discarded!");
1862 static void *mlx4_ib_add(struct mlx4_dev *dev)
1864 struct mlx4_ib_dev *ibdev;
1868 struct mlx4_ib_iboe *iboe;
1870 printk(KERN_INFO "%s", mlx4_ib_version);
1872 mlx4_foreach_ib_transport_port(i, dev)
1875 /* No point in registering a device with no ports... */
1879 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1881 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1885 iboe = &ibdev->iboe;
1887 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1890 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1893 ibdev->priv_uar.map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT,
1896 if (!ibdev->priv_uar.map)
1899 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1903 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1904 ibdev->ib_dev.owner = THIS_MODULE;
1905 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1906 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1907 ibdev->num_ports = num_ports;
1908 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1909 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1910 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1912 if (dev->caps.userspace_caps)
1913 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1915 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1917 ibdev->ib_dev.uverbs_cmd_mask =
1918 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1919 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1920 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1921 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1922 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1923 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1924 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1925 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1926 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1927 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1928 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1929 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1930 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1931 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1932 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1933 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1934 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1935 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1936 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1937 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1938 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1939 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1940 (1ull << IB_USER_VERBS_CMD_OPEN_QP) |
1941 (1ull << IB_USER_VERBS_CMD_ATTACH_FLOW) |
1942 (1ull << IB_USER_VERBS_CMD_DETACH_FLOW) |
1943 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1945 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1946 ibdev->ib_dev.query_port = mlx4_ib_query_port;
1947 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
1948 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1949 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1950 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1951 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1952 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1953 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1954 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1956 ibdev->ib_dev.get_unmapped_area = mlx4_ib_get_unmapped_area;
1958 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1959 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1960 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1961 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1962 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1963 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1964 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
1965 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
1966 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1967 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1968 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1969 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
1970 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
1971 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1972 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1973 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1974 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
1975 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
1976 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
1977 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1978 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1979 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1980 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1981 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1982 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
1983 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1984 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1985 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
1986 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1987 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1988 ibdev->ib_dev.attach_flow = mlx4_ib_flow_attach;
1989 ibdev->ib_dev.detach_flow = mlx4_ib_flow_detach;
1990 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1992 if (!mlx4_is_slave(ibdev->dev)) {
1993 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1994 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1995 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1996 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1999 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2000 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2001 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2002 ibdev->ib_dev.uverbs_cmd_mask |=
2003 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2004 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2007 mlx4_ib_alloc_eqs(dev, ibdev);
2009 spin_lock_init(&iboe->lock);
2011 if (init_node_data(ibdev))
2014 for (i = 0; i < ibdev->num_ports; ++i) {
2015 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2016 IB_LINK_LAYER_ETHERNET) {
2017 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2019 ibdev->counters[i] = -1;
2021 ibdev->counters[i] = -1;
2024 spin_lock_init(&ibdev->sm_lock);
2025 mutex_init(&ibdev->cap_mask_mutex);
2027 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2028 !mlx4_is_slave(dev)) {
2029 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2030 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2031 MLX4_IB_UC_STEER_QPN_ALIGN, &ibdev->steer_qpn_base, 0);
2035 ibdev->ib_uc_qpns_bitmap =
2036 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2039 if (!ibdev->ib_uc_qpns_bitmap) {
2040 dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2041 goto err_steer_qp_release;
2044 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2046 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(dev, ibdev->steer_qpn_base,
2047 ibdev->steer_qpn_base + ibdev->steer_qpn_count - 1);
2049 goto err_steer_free_bitmap;
2052 if (ib_register_device(&ibdev->ib_dev, NULL))
2053 goto err_steer_free_bitmap;
2055 if (mlx4_ib_mad_init(ibdev))
2058 if (mlx4_ib_init_sriov(ibdev))
2061 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
2062 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2063 err = register_netdevice_notifier(&iboe->nb);
2068 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2069 if (device_create_file(&ibdev->ib_dev.dev,
2070 mlx4_class_attributes[j]))
2073 if (sysfs_create_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group))
2076 ibdev->ib_active = true;
2078 if (mlx4_is_mfunc(ibdev->dev))
2081 /* create paravirt contexts for any VFs which are active */
2082 if (mlx4_is_master(ibdev->dev)) {
2083 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2084 if (j == mlx4_master_func_num(ibdev->dev))
2086 if (mlx4_is_slave_active(ibdev->dev, j))
2087 do_slave_init(ibdev, j, 1);
2093 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2094 pr_warn("failure unregistering notifier\n");
2095 flush_workqueue(wq);
2098 mlx4_ib_close_sriov(ibdev);
2101 mlx4_ib_mad_cleanup(ibdev);
2104 ib_unregister_device(&ibdev->ib_dev);
2106 err_steer_free_bitmap:
2107 kfree(ibdev->ib_uc_qpns_bitmap);
2109 err_steer_qp_release:
2110 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED)
2111 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2112 ibdev->steer_qpn_count);
2115 if (ibdev->counters[i - 1] != -1)
2116 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2119 iounmap(ibdev->priv_uar.map);
2120 mlx4_ib_free_eqs(dev, ibdev);
2123 mlx4_uar_free(dev, &ibdev->priv_uar);
2126 mlx4_pd_free(dev, ibdev->priv_pdn);
2129 ib_dealloc_device(&ibdev->ib_dev);
2134 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2138 WARN_ON(!dev->ib_uc_qpns_bitmap);
2140 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2141 dev->steer_qpn_count,
2142 get_count_order(count));
2146 *qpn = dev->steer_qpn_base + offset;
2150 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2153 dev->dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED)
2156 BUG_ON(qpn < dev->steer_qpn_base);
2158 bitmap_release_region(dev->ib_uc_qpns_bitmap,
2159 qpn - dev->steer_qpn_base, get_count_order(count));
2162 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2165 struct ib_flow_spec spec = {
2166 .type = IB_FLOW_IB_UC,
2167 .l2_id.ib_uc.qpn = mqp->ibqp.qp_num,
2171 __mlx4_ib_flow_attach(mdev, mqp, &spec, MLX4_DOMAIN_NIC, 0)
2172 : __mlx4_ib_flow_detach(mdev, mqp, &spec, MLX4_DOMAIN_NIC, 0);
2175 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2177 struct mlx4_ib_dev *ibdev = ibdev_ptr;
2180 mlx4_ib_close_sriov(ibdev);
2181 sysfs_remove_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group);
2182 mlx4_ib_mad_cleanup(ibdev);
2184 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2185 device_remove_file(&ibdev->ib_dev.dev, mlx4_class_attributes[j]);
2188 ib_unregister_device(&ibdev->ib_dev);
2190 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2191 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2192 ibdev->steer_qpn_count);
2193 kfree(ibdev->ib_uc_qpns_bitmap);
2196 if (ibdev->iboe.nb.notifier_call) {
2197 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2198 pr_warn("failure unregistering notifier\n");
2199 ibdev->iboe.nb.notifier_call = NULL;
2201 iounmap(ibdev->priv_uar.map);
2202 for (p = 0; p < ibdev->num_ports; ++p)
2203 if (ibdev->counters[p] != -1)
2204 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2205 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2206 mlx4_CLOSE_PORT(dev, p);
2208 mlx4_ib_free_eqs(dev, ibdev);
2210 mlx4_uar_free(dev, &ibdev->priv_uar);
2211 mlx4_pd_free(dev, ibdev->priv_pdn);
2212 ib_dealloc_device(&ibdev->ib_dev);
2215 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2217 struct mlx4_ib_demux_work **dm = NULL;
2218 struct mlx4_dev *dev = ibdev->dev;
2220 unsigned long flags;
2222 if (!mlx4_is_master(dev))
2225 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
2227 pr_err("failed to allocate memory for tunneling qp update\n");
2231 for (i = 0; i < dev->caps.num_ports; i++) {
2232 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2234 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2235 for (i = 0; i < dev->caps.num_ports; i++) {
2242 /* initialize or tear down tunnel QPs for the slave */
2243 for (i = 0; i < dev->caps.num_ports; i++) {
2244 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2245 dm[i]->port = i + 1;
2246 dm[i]->slave = slave;
2247 dm[i]->do_init = do_init;
2249 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2250 if (!ibdev->sriov.is_going_down)
2251 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2252 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2260 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2261 enum mlx4_dev_event event, unsigned long param)
2263 struct ib_event ibev;
2264 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2265 struct mlx4_eqe *eqe = NULL;
2266 struct ib_event_work *ew;
2269 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2270 eqe = (struct mlx4_eqe *)param;
2275 case MLX4_DEV_EVENT_PORT_UP:
2276 if (p > ibdev->num_ports)
2278 if (mlx4_is_master(dev) &&
2279 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2280 IB_LINK_LAYER_INFINIBAND) {
2281 mlx4_ib_invalidate_all_guid_record(ibdev, p);
2283 mlx4_ib_info((struct ib_device *) ibdev_ptr,
2284 "Port %d logical link is up\n", p);
2285 ibev.event = IB_EVENT_PORT_ACTIVE;
2288 case MLX4_DEV_EVENT_PORT_DOWN:
2289 if (p > ibdev->num_ports)
2291 mlx4_ib_info((struct ib_device *) ibdev_ptr,
2292 "Port %d logical link is down\n", p);
2293 ibev.event = IB_EVENT_PORT_ERR;
2296 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2297 ibdev->ib_active = false;
2298 ibev.event = IB_EVENT_DEVICE_FATAL;
2301 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2302 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2304 pr_err("failed to allocate memory for events work\n");
2308 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2309 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2311 /* need to queue only for port owner, which uses GEN_EQE */
2312 if (mlx4_is_master(dev))
2313 queue_work(wq, &ew->work);
2315 handle_port_mgmt_change_event(&ew->work);
2318 case MLX4_DEV_EVENT_SLAVE_INIT:
2319 /* here, p is the slave id */
2320 do_slave_init(ibdev, p, 1);
2323 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2324 /* here, p is the slave id */
2325 do_slave_init(ibdev, p, 0);
2332 ibev.device = ibdev_ptr;
2333 ibev.element.port_num = (u8) p;
2335 ib_dispatch_event(&ibev);
2338 static struct mlx4_interface mlx4_ib_interface = {
2340 .remove = mlx4_ib_remove,
2341 .event = mlx4_ib_event,
2342 .protocol = MLX4_PROT_IB_IPV6
2345 static int __init mlx4_ib_init(void)
2349 wq = create_singlethread_workqueue("mlx4_ib");
2354 err = mlx4_ib_proc_init();
2359 err = mlx4_ib_mcg_init();
2365 err = mlx4_register_interface(&mlx4_ib_interface);
2372 mlx4_ib_mcg_destroy();
2376 remove_proc_entry(MLX4_IB_MRS_PROC_DIR_NAME,
2377 mlx4_ib_driver_dir_entry);
2378 remove_proc_entry(MLX4_IB_DRIVER_PROC_DIR_NAME, NULL);
2382 destroy_workqueue(wq);
2386 static void __exit mlx4_ib_cleanup(void)
2388 mlx4_unregister_interface(&mlx4_ib_interface);
2389 mlx4_ib_mcg_destroy();
2390 destroy_workqueue(wq);
2392 /* Remove proc entries */
2394 remove_proc_entry(MLX4_IB_MRS_PROC_DIR_NAME,
2395 mlx4_ib_driver_dir_entry);
2396 remove_proc_entry(MLX4_IB_DRIVER_PROC_DIR_NAME, NULL);
2401 module_init_order(mlx4_ib_init, SI_ORDER_MIDDLE);
2402 module_exit(mlx4_ib_cleanup);
2404 #undef MODULE_VERSION
2405 #include <sys/module.h>
2407 mlx4ib_evhand(module_t mod, int event, void *arg)
2412 static moduledata_t mlx4ib_mod = {
2414 .evhand = mlx4ib_evhand,
2417 DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_SMP, SI_ORDER_ANY);
2418 MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1);
2419 MODULE_DEPEND(mlx4ib, ibcore, 1, 1, 1);