2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
4 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
5 * Copyright (c) 2008 Cisco. All rights reserved.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #define pr_fmt(fmt) "user_mad: " fmt
38 #include <linux/module.h>
39 #include <linux/init.h>
40 #include <linux/device.h>
41 #include <linux/err.h>
43 #include <linux/cdev.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/poll.h>
46 #include <linux/mutex.h>
47 #include <linux/kref.h>
48 #include <linux/compat.h>
49 #include <linux/sched.h>
50 #include <linux/semaphore.h>
51 #include <linux/slab.h>
53 #include <linux/uaccess.h>
55 #include <rdma/ib_mad.h>
56 #include <rdma/ib_user_mad.h>
58 #include "core_priv.h"
60 MODULE_AUTHOR("Roland Dreier");
61 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
62 MODULE_LICENSE("Dual BSD/GPL");
65 IB_UMAD_MAX_PORTS = RDMA_MAX_PORTS,
66 IB_UMAD_MAX_AGENTS = 32,
69 IB_UMAD_MINOR_BASE = 0,
70 IB_UMAD_NUM_FIXED_MINOR = 64,
71 IB_UMAD_NUM_DYNAMIC_MINOR = IB_UMAD_MAX_PORTS - IB_UMAD_NUM_FIXED_MINOR,
72 IB_ISSM_MINOR_BASE = IB_UMAD_NUM_FIXED_MINOR,
76 * Our lifetime rules for these structs are the following:
77 * device special file is opened, we take a reference on the
78 * ib_umad_port's struct ib_umad_device. We drop these
79 * references in the corresponding close().
81 * In addition to references coming from open character devices, there
82 * is one more reference to each ib_umad_device representing the
83 * module's reference taken when allocating the ib_umad_device in
86 * When destroying an ib_umad_device, we drop the module's reference.
94 struct device *sm_dev;
95 struct semaphore sm_sem;
97 struct mutex file_mutex;
98 struct list_head file_list;
100 struct ib_device *ib_dev;
101 struct ib_umad_device *umad_dev;
106 struct ib_umad_device {
108 struct ib_umad_port port[0];
111 struct ib_umad_file {
113 struct ib_umad_port *port;
114 struct list_head recv_list;
115 struct list_head send_list;
116 struct list_head port_list;
117 spinlock_t send_lock;
118 wait_queue_head_t recv_wait;
119 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
125 struct ib_umad_packet {
126 struct ib_mad_send_buf *msg;
127 struct ib_mad_recv_wc *recv_wc;
128 struct list_head list;
130 struct ib_user_mad mad;
133 static struct class *umad_class;
135 static const dev_t base_umad_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
136 static const dev_t base_issm_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE) +
137 IB_UMAD_NUM_FIXED_MINOR;
138 static dev_t dynamic_umad_dev;
139 static dev_t dynamic_issm_dev;
141 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
143 static void ib_umad_add_one(struct ib_device *device);
144 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
146 static void ib_umad_release_dev(struct kobject *kobj)
148 struct ib_umad_device *dev =
149 container_of(kobj, struct ib_umad_device, kobj);
154 static struct kobj_type ib_umad_dev_ktype = {
155 .release = ib_umad_release_dev,
158 static int hdr_size(struct ib_umad_file *file)
160 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
161 sizeof (struct ib_user_mad_hdr_old);
164 /* caller must hold file->mutex */
165 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
167 return file->agents_dead ? NULL : file->agent[id];
170 static int queue_packet(struct ib_umad_file *file,
171 struct ib_mad_agent *agent,
172 struct ib_umad_packet *packet)
176 mutex_lock(&file->mutex);
178 for (packet->mad.hdr.id = 0;
179 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
180 packet->mad.hdr.id++)
181 if (agent == __get_agent(file, packet->mad.hdr.id)) {
182 list_add_tail(&packet->list, &file->recv_list);
183 wake_up_interruptible(&file->recv_wait);
188 mutex_unlock(&file->mutex);
193 static void dequeue_send(struct ib_umad_file *file,
194 struct ib_umad_packet *packet)
196 spin_lock_irq(&file->send_lock);
197 list_del(&packet->list);
198 spin_unlock_irq(&file->send_lock);
201 static void send_handler(struct ib_mad_agent *agent,
202 struct ib_mad_send_wc *send_wc)
204 struct ib_umad_file *file = agent->context;
205 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
207 dequeue_send(file, packet);
208 rdma_destroy_ah(packet->msg->ah);
209 ib_free_send_mad(packet->msg);
211 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
212 packet->length = IB_MGMT_MAD_HDR;
213 packet->mad.hdr.status = ETIMEDOUT;
214 if (!queue_packet(file, agent, packet))
220 static void recv_handler(struct ib_mad_agent *agent,
221 struct ib_mad_send_buf *send_buf,
222 struct ib_mad_recv_wc *mad_recv_wc)
224 struct ib_umad_file *file = agent->context;
225 struct ib_umad_packet *packet;
227 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
230 packet = kzalloc(sizeof *packet, GFP_KERNEL);
234 packet->length = mad_recv_wc->mad_len;
235 packet->recv_wc = mad_recv_wc;
237 packet->mad.hdr.status = 0;
238 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
239 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
241 * On OPA devices it is okay to lose the upper 16 bits of LID as this
242 * information is obtained elsewhere. Mask off the upper 16 bits.
244 if (rdma_cap_opa_mad(agent->device, agent->port_num))
245 packet->mad.hdr.lid = ib_lid_be16(0xFFFF &
246 mad_recv_wc->wc->slid);
248 packet->mad.hdr.lid = ib_lid_be16(mad_recv_wc->wc->slid);
249 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
250 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
251 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
252 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
253 if (packet->mad.hdr.grh_present) {
254 struct rdma_ah_attr ah_attr;
255 const struct ib_global_route *grh;
258 ret = ib_init_ah_attr_from_wc(agent->device, agent->port_num,
260 mad_recv_wc->recv_buf.grh,
265 grh = rdma_ah_read_grh(&ah_attr);
266 packet->mad.hdr.gid_index = grh->sgid_index;
267 packet->mad.hdr.hop_limit = grh->hop_limit;
268 packet->mad.hdr.traffic_class = grh->traffic_class;
269 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
270 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
271 rdma_destroy_ah_attr(&ah_attr);
274 if (queue_packet(file, agent, packet))
281 ib_free_recv_mad(mad_recv_wc);
284 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
285 struct ib_umad_packet *packet, size_t count)
287 struct ib_mad_recv_buf *recv_buf;
288 int left, seg_payload, offset, max_seg_payload;
291 recv_buf = &packet->recv_wc->recv_buf;
292 seg_size = packet->recv_wc->mad_seg_size;
294 /* We need enough room to copy the first (or only) MAD segment. */
295 if ((packet->length <= seg_size &&
296 count < hdr_size(file) + packet->length) ||
297 (packet->length > seg_size &&
298 count < hdr_size(file) + seg_size))
301 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
304 buf += hdr_size(file);
305 seg_payload = min_t(int, packet->length, seg_size);
306 if (copy_to_user(buf, recv_buf->mad, seg_payload))
309 if (seg_payload < packet->length) {
311 * Multipacket RMPP MAD message. Copy remainder of message.
312 * Note that last segment may have a shorter payload.
314 if (count < hdr_size(file) + packet->length) {
316 * The buffer is too small, return the first RMPP segment,
317 * which includes the RMPP message length.
321 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
322 max_seg_payload = seg_size - offset;
324 for (left = packet->length - seg_payload, buf += seg_payload;
325 left; left -= seg_payload, buf += seg_payload) {
326 recv_buf = container_of(recv_buf->list.next,
327 struct ib_mad_recv_buf, list);
328 seg_payload = min(left, max_seg_payload);
329 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
334 return hdr_size(file) + packet->length;
337 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
338 struct ib_umad_packet *packet, size_t count)
340 ssize_t size = hdr_size(file) + packet->length;
345 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
348 buf += hdr_size(file);
350 if (copy_to_user(buf, packet->mad.data, packet->length))
356 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
357 size_t count, loff_t *pos)
359 struct ib_umad_file *file = filp->private_data;
360 struct ib_umad_packet *packet;
363 if (count < hdr_size(file))
366 mutex_lock(&file->mutex);
368 while (list_empty(&file->recv_list)) {
369 mutex_unlock(&file->mutex);
371 if (filp->f_flags & O_NONBLOCK)
374 if (wait_event_interruptible(file->recv_wait,
375 !list_empty(&file->recv_list)))
378 mutex_lock(&file->mutex);
381 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
382 list_del(&packet->list);
384 mutex_unlock(&file->mutex);
387 ret = copy_recv_mad(file, buf, packet, count);
389 ret = copy_send_mad(file, buf, packet, count);
393 mutex_lock(&file->mutex);
394 list_add(&packet->list, &file->recv_list);
395 mutex_unlock(&file->mutex);
398 ib_free_recv_mad(packet->recv_wc);
404 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
408 /* Copy class specific header */
409 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
410 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
411 msg->hdr_len - IB_MGMT_RMPP_HDR))
414 /* All headers are in place. Copy data segments. */
415 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
416 seg++, left -= msg->seg_size, buf += msg->seg_size) {
417 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
418 min(left, msg->seg_size)))
424 static int same_destination(struct ib_user_mad_hdr *hdr1,
425 struct ib_user_mad_hdr *hdr2)
427 if (!hdr1->grh_present && !hdr2->grh_present)
428 return (hdr1->lid == hdr2->lid);
430 if (hdr1->grh_present && hdr2->grh_present)
431 return !memcmp(hdr1->gid, hdr2->gid, 16);
436 static int is_duplicate(struct ib_umad_file *file,
437 struct ib_umad_packet *packet)
439 struct ib_umad_packet *sent_packet;
440 struct ib_mad_hdr *sent_hdr, *hdr;
442 hdr = (struct ib_mad_hdr *) packet->mad.data;
443 list_for_each_entry(sent_packet, &file->send_list, list) {
444 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
446 if ((hdr->tid != sent_hdr->tid) ||
447 (hdr->mgmt_class != sent_hdr->mgmt_class))
451 * No need to be overly clever here. If two new operations have
452 * the same TID, reject the second as a duplicate. This is more
453 * restrictive than required by the spec.
455 if (!ib_response_mad(hdr)) {
456 if (!ib_response_mad(sent_hdr))
459 } else if (!ib_response_mad(sent_hdr))
462 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
469 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
470 size_t count, loff_t *pos)
472 struct ib_umad_file *file = filp->private_data;
473 struct ib_umad_packet *packet;
474 struct ib_mad_agent *agent;
475 struct rdma_ah_attr ah_attr;
477 struct ib_rmpp_mad *rmpp_mad;
479 int ret, data_len, hdr_len, copy_offset, rmpp_active;
482 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
485 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
489 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
494 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
499 buf += hdr_size(file);
501 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
506 mutex_lock(&file->mutex);
508 agent = __get_agent(file, packet->mad.hdr.id);
514 memset(&ah_attr, 0, sizeof ah_attr);
515 ah_attr.type = rdma_ah_find_type(agent->device,
516 file->port->port_num);
517 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
518 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
519 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
520 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
521 if (packet->mad.hdr.grh_present) {
522 rdma_ah_set_grh(&ah_attr, NULL,
523 be32_to_cpu(packet->mad.hdr.flow_label),
524 packet->mad.hdr.gid_index,
525 packet->mad.hdr.hop_limit,
526 packet->mad.hdr.traffic_class);
527 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
530 ah = rdma_create_user_ah(agent->qp->pd, &ah_attr, NULL);
536 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
537 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
539 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
540 && ib_mad_kernel_rmpp_agent(agent)) {
541 copy_offset = IB_MGMT_RMPP_HDR;
542 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
543 IB_MGMT_RMPP_FLAG_ACTIVE;
545 copy_offset = IB_MGMT_MAD_HDR;
549 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
550 data_len = count - hdr_size(file) - hdr_len;
551 packet->msg = ib_create_send_mad(agent,
552 be32_to_cpu(packet->mad.hdr.qpn),
553 packet->mad.hdr.pkey_index, rmpp_active,
554 hdr_len, data_len, GFP_KERNEL,
556 if (IS_ERR(packet->msg)) {
557 ret = PTR_ERR(packet->msg);
561 packet->msg->ah = ah;
562 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
563 packet->msg->retries = packet->mad.hdr.retries;
564 packet->msg->context[0] = packet;
566 /* Copy MAD header. Any RMPP header is already in place. */
567 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
570 if (copy_from_user(packet->msg->mad + copy_offset,
572 hdr_len + data_len - copy_offset)) {
577 ret = copy_rmpp_mad(packet->msg, buf);
583 * Set the high-order part of the transaction ID to make MADs from
584 * different agents unique, and allow routing responses back to the
585 * original requestor.
587 if (!ib_response_mad(packet->msg->mad)) {
588 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
589 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
590 (be64_to_cpup(tid) & 0xffffffff));
591 rmpp_mad->mad_hdr.tid = *tid;
594 if (!ib_mad_kernel_rmpp_agent(agent)
595 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
596 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
597 spin_lock_irq(&file->send_lock);
598 list_add_tail(&packet->list, &file->send_list);
599 spin_unlock_irq(&file->send_lock);
601 spin_lock_irq(&file->send_lock);
602 ret = is_duplicate(file, packet);
604 list_add_tail(&packet->list, &file->send_list);
605 spin_unlock_irq(&file->send_lock);
612 ret = ib_post_send_mad(packet->msg, NULL);
616 mutex_unlock(&file->mutex);
620 dequeue_send(file, packet);
622 ib_free_send_mad(packet->msg);
626 mutex_unlock(&file->mutex);
632 static __poll_t ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
634 struct ib_umad_file *file = filp->private_data;
636 /* we will always be able to post a MAD send */
637 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
639 poll_wait(filp, &file->recv_wait, wait);
641 if (!list_empty(&file->recv_list))
642 mask |= EPOLLIN | EPOLLRDNORM;
647 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
648 int compat_method_mask)
650 struct ib_user_mad_reg_req ureq;
651 struct ib_mad_reg_req req;
652 struct ib_mad_agent *agent = NULL;
656 mutex_lock(&file->port->file_mutex);
657 mutex_lock(&file->mutex);
659 if (!file->port->ib_dev) {
660 dev_notice(file->port->dev,
661 "ib_umad_reg_agent: invalid device\n");
666 if (copy_from_user(&ureq, arg, sizeof ureq)) {
671 if (ureq.qpn != 0 && ureq.qpn != 1) {
672 dev_notice(file->port->dev,
673 "ib_umad_reg_agent: invalid QPN %d specified\n",
679 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
680 if (!__get_agent(file, agent_id))
683 dev_notice(file->port->dev,
684 "ib_umad_reg_agent: Max Agents (%u) reached\n",
690 if (ureq.mgmt_class) {
691 memset(&req, 0, sizeof(req));
692 req.mgmt_class = ureq.mgmt_class;
693 req.mgmt_class_version = ureq.mgmt_class_version;
694 memcpy(req.oui, ureq.oui, sizeof req.oui);
696 if (compat_method_mask) {
697 u32 *umm = (u32 *) ureq.method_mask;
700 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
702 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
704 memcpy(req.method_mask, ureq.method_mask,
705 sizeof req.method_mask);
708 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
709 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
710 ureq.mgmt_class ? &req : NULL,
712 send_handler, recv_handler, file, 0);
714 ret = PTR_ERR(agent);
719 if (put_user(agent_id,
720 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
725 if (!file->already_used) {
726 file->already_used = 1;
727 if (!file->use_pkey_index) {
728 dev_warn(file->port->dev,
729 "process %s did not enable P_Key index support.\n",
731 dev_warn(file->port->dev,
732 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
736 file->agent[agent_id] = agent;
740 mutex_unlock(&file->mutex);
743 ib_unregister_mad_agent(agent);
745 mutex_unlock(&file->port->file_mutex);
750 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
752 struct ib_user_mad_reg_req2 ureq;
753 struct ib_mad_reg_req req;
754 struct ib_mad_agent *agent = NULL;
758 mutex_lock(&file->port->file_mutex);
759 mutex_lock(&file->mutex);
761 if (!file->port->ib_dev) {
762 dev_notice(file->port->dev,
763 "ib_umad_reg_agent2: invalid device\n");
768 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
773 if (ureq.qpn != 0 && ureq.qpn != 1) {
774 dev_notice(file->port->dev,
775 "ib_umad_reg_agent2: invalid QPN %d specified\n",
781 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
782 dev_notice(file->port->dev,
783 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
784 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
787 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
788 (u32 __user *) (arg + offsetof(struct
789 ib_user_mad_reg_req2, flags))))
795 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
796 if (!__get_agent(file, agent_id))
799 dev_notice(file->port->dev,
800 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
806 if (ureq.mgmt_class) {
807 memset(&req, 0, sizeof(req));
808 req.mgmt_class = ureq.mgmt_class;
809 req.mgmt_class_version = ureq.mgmt_class_version;
810 if (ureq.oui & 0xff000000) {
811 dev_notice(file->port->dev,
812 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
817 req.oui[2] = ureq.oui & 0x0000ff;
818 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
819 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
820 memcpy(req.method_mask, ureq.method_mask,
821 sizeof(req.method_mask));
824 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
825 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
826 ureq.mgmt_class ? &req : NULL,
828 send_handler, recv_handler, file,
831 ret = PTR_ERR(agent);
836 if (put_user(agent_id,
838 offsetof(struct ib_user_mad_reg_req2, id)))) {
843 if (!file->already_used) {
844 file->already_used = 1;
845 file->use_pkey_index = 1;
848 file->agent[agent_id] = agent;
852 mutex_unlock(&file->mutex);
855 ib_unregister_mad_agent(agent);
857 mutex_unlock(&file->port->file_mutex);
863 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
865 struct ib_mad_agent *agent = NULL;
869 if (get_user(id, arg))
872 mutex_lock(&file->port->file_mutex);
873 mutex_lock(&file->mutex);
875 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
880 agent = file->agent[id];
881 file->agent[id] = NULL;
884 mutex_unlock(&file->mutex);
887 ib_unregister_mad_agent(agent);
889 mutex_unlock(&file->port->file_mutex);
894 static long ib_umad_enable_pkey(struct ib_umad_file *file)
898 mutex_lock(&file->mutex);
899 if (file->already_used)
902 file->use_pkey_index = 1;
903 mutex_unlock(&file->mutex);
908 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
912 case IB_USER_MAD_REGISTER_AGENT:
913 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
914 case IB_USER_MAD_UNREGISTER_AGENT:
915 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
916 case IB_USER_MAD_ENABLE_PKEY:
917 return ib_umad_enable_pkey(filp->private_data);
918 case IB_USER_MAD_REGISTER_AGENT2:
919 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
926 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
930 case IB_USER_MAD_REGISTER_AGENT:
931 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
932 case IB_USER_MAD_UNREGISTER_AGENT:
933 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
934 case IB_USER_MAD_ENABLE_PKEY:
935 return ib_umad_enable_pkey(filp->private_data);
936 case IB_USER_MAD_REGISTER_AGENT2:
937 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
945 * ib_umad_open() does not need the BKL:
947 * - the ib_umad_port structures are properly reference counted, and
948 * everything else is purely local to the file being created, so
949 * races against other open calls are not a problem;
950 * - the ioctl method does not affect any global state outside of the
951 * file structure being operated on;
953 static int ib_umad_open(struct inode *inode, struct file *filp)
955 struct ib_umad_port *port;
956 struct ib_umad_file *file;
959 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
961 mutex_lock(&port->file_mutex);
967 file = kzalloc(sizeof *file, GFP_KERNEL);
971 mutex_init(&file->mutex);
972 spin_lock_init(&file->send_lock);
973 INIT_LIST_HEAD(&file->recv_list);
974 INIT_LIST_HEAD(&file->send_list);
975 init_waitqueue_head(&file->recv_wait);
978 filp->private_data = file;
980 list_add_tail(&file->port_list, &port->file_list);
982 ret = nonseekable_open(inode, filp);
984 list_del(&file->port_list);
989 kobject_get(&port->umad_dev->kobj);
992 mutex_unlock(&port->file_mutex);
996 static int ib_umad_close(struct inode *inode, struct file *filp)
998 struct ib_umad_file *file = filp->private_data;
999 struct ib_umad_device *dev = file->port->umad_dev;
1000 struct ib_umad_packet *packet, *tmp;
1004 mutex_lock(&file->port->file_mutex);
1005 mutex_lock(&file->mutex);
1007 already_dead = file->agents_dead;
1008 file->agents_dead = 1;
1010 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
1011 if (packet->recv_wc)
1012 ib_free_recv_mad(packet->recv_wc);
1016 list_del(&file->port_list);
1018 mutex_unlock(&file->mutex);
1021 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1023 ib_unregister_mad_agent(file->agent[i]);
1025 mutex_unlock(&file->port->file_mutex);
1028 kobject_put(&dev->kobj);
1033 static const struct file_operations umad_fops = {
1034 .owner = THIS_MODULE,
1035 .read = ib_umad_read,
1036 .write = ib_umad_write,
1037 .poll = ib_umad_poll,
1038 .unlocked_ioctl = ib_umad_ioctl,
1039 #ifdef CONFIG_COMPAT
1040 .compat_ioctl = ib_umad_compat_ioctl,
1042 .open = ib_umad_open,
1043 .release = ib_umad_close,
1044 .llseek = no_llseek,
1047 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1049 struct ib_umad_port *port;
1050 struct ib_port_modify props = {
1051 .set_port_cap_mask = IB_PORT_SM
1055 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1057 if (filp->f_flags & O_NONBLOCK) {
1058 if (down_trylock(&port->sm_sem)) {
1063 if (down_interruptible(&port->sm_sem)) {
1069 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1073 filp->private_data = port;
1075 ret = nonseekable_open(inode, filp);
1077 goto err_clr_sm_cap;
1079 kobject_get(&port->umad_dev->kobj);
1084 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1085 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1094 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1096 struct ib_umad_port *port = filp->private_data;
1097 struct ib_port_modify props = {
1098 .clr_port_cap_mask = IB_PORT_SM
1102 mutex_lock(&port->file_mutex);
1104 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1105 mutex_unlock(&port->file_mutex);
1109 kobject_put(&port->umad_dev->kobj);
1114 static const struct file_operations umad_sm_fops = {
1115 .owner = THIS_MODULE,
1116 .open = ib_umad_sm_open,
1117 .release = ib_umad_sm_close,
1118 .llseek = no_llseek,
1121 static struct ib_client umad_client = {
1123 .add = ib_umad_add_one,
1124 .remove = ib_umad_remove_one
1127 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1130 struct ib_umad_port *port = dev_get_drvdata(dev);
1135 return sprintf(buf, "%s\n", port->ib_dev->name);
1137 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1139 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1142 struct ib_umad_port *port = dev_get_drvdata(dev);
1147 return sprintf(buf, "%d\n", port->port_num);
1149 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1151 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1152 __stringify(IB_USER_MAD_ABI_VERSION));
1154 static int ib_umad_init_port(struct ib_device *device, int port_num,
1155 struct ib_umad_device *umad_dev,
1156 struct ib_umad_port *port)
1162 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1163 if (devnum >= IB_UMAD_MAX_PORTS)
1165 port->dev_num = devnum;
1166 set_bit(devnum, dev_map);
1167 if (devnum >= IB_UMAD_NUM_FIXED_MINOR) {
1168 base_umad = dynamic_umad_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1169 base_issm = dynamic_issm_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1171 base_umad = devnum + base_umad_dev;
1172 base_issm = devnum + base_issm_dev;
1175 port->ib_dev = device;
1176 port->port_num = port_num;
1177 sema_init(&port->sm_sem, 1);
1178 mutex_init(&port->file_mutex);
1179 INIT_LIST_HEAD(&port->file_list);
1181 cdev_init(&port->cdev, &umad_fops);
1182 port->cdev.owner = THIS_MODULE;
1183 cdev_set_parent(&port->cdev, &umad_dev->kobj);
1184 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1185 if (cdev_add(&port->cdev, base_umad, 1))
1188 port->dev = device_create(umad_class, device->dev.parent,
1189 port->cdev.dev, port,
1190 "umad%d", port->dev_num);
1191 if (IS_ERR(port->dev))
1194 if (device_create_file(port->dev, &dev_attr_ibdev))
1196 if (device_create_file(port->dev, &dev_attr_port))
1199 cdev_init(&port->sm_cdev, &umad_sm_fops);
1200 port->sm_cdev.owner = THIS_MODULE;
1201 cdev_set_parent(&port->sm_cdev, &umad_dev->kobj);
1202 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1203 if (cdev_add(&port->sm_cdev, base_issm, 1))
1206 port->sm_dev = device_create(umad_class, device->dev.parent,
1207 port->sm_cdev.dev, port,
1208 "issm%d", port->dev_num);
1209 if (IS_ERR(port->sm_dev))
1212 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1214 if (device_create_file(port->sm_dev, &dev_attr_port))
1220 device_destroy(umad_class, port->sm_cdev.dev);
1223 cdev_del(&port->sm_cdev);
1226 device_destroy(umad_class, port->cdev.dev);
1229 cdev_del(&port->cdev);
1230 clear_bit(devnum, dev_map);
1235 static void ib_umad_kill_port(struct ib_umad_port *port)
1237 struct ib_umad_file *file;
1240 dev_set_drvdata(port->dev, NULL);
1241 dev_set_drvdata(port->sm_dev, NULL);
1243 device_destroy(umad_class, port->cdev.dev);
1244 device_destroy(umad_class, port->sm_cdev.dev);
1246 cdev_del(&port->cdev);
1247 cdev_del(&port->sm_cdev);
1249 mutex_lock(&port->file_mutex);
1251 port->ib_dev = NULL;
1253 list_for_each_entry(file, &port->file_list, port_list) {
1254 mutex_lock(&file->mutex);
1255 file->agents_dead = 1;
1256 mutex_unlock(&file->mutex);
1258 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1259 if (file->agent[id])
1260 ib_unregister_mad_agent(file->agent[id]);
1263 mutex_unlock(&port->file_mutex);
1264 clear_bit(port->dev_num, dev_map);
1267 static void ib_umad_add_one(struct ib_device *device)
1269 struct ib_umad_device *umad_dev;
1273 s = rdma_start_port(device);
1274 e = rdma_end_port(device);
1276 umad_dev = kzalloc(sizeof *umad_dev +
1277 (e - s + 1) * sizeof (struct ib_umad_port),
1282 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1284 for (i = s; i <= e; ++i) {
1285 if (!rdma_cap_ib_mad(device, i))
1288 umad_dev->port[i - s].umad_dev = umad_dev;
1290 if (ib_umad_init_port(device, i, umad_dev,
1291 &umad_dev->port[i - s]))
1300 ib_set_client_data(device, &umad_client, umad_dev);
1306 if (!rdma_cap_ib_mad(device, i))
1309 ib_umad_kill_port(&umad_dev->port[i - s]);
1312 kobject_put(&umad_dev->kobj);
1315 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1317 struct ib_umad_device *umad_dev = client_data;
1323 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1324 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1325 ib_umad_kill_port(&umad_dev->port[i]);
1328 kobject_put(&umad_dev->kobj);
1331 static char *umad_devnode(struct device *dev, umode_t *mode)
1333 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1336 static int __init ib_umad_init(void)
1340 ret = register_chrdev_region(base_umad_dev,
1341 IB_UMAD_NUM_FIXED_MINOR * 2,
1344 pr_err("couldn't register device number\n");
1348 ret = alloc_chrdev_region(&dynamic_umad_dev, 0,
1349 IB_UMAD_NUM_DYNAMIC_MINOR * 2,
1352 pr_err("couldn't register dynamic device number\n");
1355 dynamic_issm_dev = dynamic_umad_dev + IB_UMAD_NUM_DYNAMIC_MINOR;
1357 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1358 if (IS_ERR(umad_class)) {
1359 ret = PTR_ERR(umad_class);
1360 pr_err("couldn't create class infiniband_mad\n");
1364 umad_class->devnode = umad_devnode;
1366 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1368 pr_err("couldn't create abi_version attribute\n");
1372 ret = ib_register_client(&umad_client);
1374 pr_err("couldn't register ib_umad client\n");
1381 class_destroy(umad_class);
1384 unregister_chrdev_region(dynamic_umad_dev,
1385 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1388 unregister_chrdev_region(base_umad_dev,
1389 IB_UMAD_NUM_FIXED_MINOR * 2);
1395 static void __exit ib_umad_cleanup(void)
1397 ib_unregister_client(&umad_client);
1398 class_destroy(umad_class);
1399 unregister_chrdev_region(base_umad_dev,
1400 IB_UMAD_NUM_FIXED_MINOR * 2);
1401 unregister_chrdev_region(dynamic_umad_dev,
1402 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1405 module_init(ib_umad_init);
1406 module_exit(ib_umad_cleanup);