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.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * $Id: user_mad.c 5596 2006-03-03 01:00:07Z sean.hefty $
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/device.h>
40 #include <linux/err.h>
42 #include <linux/cdev.h>
43 #include <linux/dma-mapping.h>
44 #include <linux/poll.h>
45 #include <linux/rwsem.h>
46 #include <linux/kref.h>
48 #include <asm/uaccess.h>
49 #include <asm/semaphore.h>
51 #include <rdma/ib_mad.h>
52 #include <rdma/ib_user_mad.h>
54 MODULE_AUTHOR("Roland Dreier");
55 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
56 MODULE_LICENSE("Dual BSD/GPL");
59 IB_UMAD_MAX_PORTS = 64,
60 IB_UMAD_MAX_AGENTS = 32,
63 IB_UMAD_MINOR_BASE = 0
67 * Our lifetime rules for these structs are the following: each time a
68 * device special file is opened, we look up the corresponding struct
69 * ib_umad_port by minor in the umad_port[] table while holding the
70 * port_lock. If this lookup succeeds, we take a reference on the
71 * ib_umad_port's struct ib_umad_device while still holding the
72 * port_lock; if the lookup fails, we fail the open(). We drop these
73 * references in the corresponding close().
75 * In addition to references coming from open character devices, there
76 * is one more reference to each ib_umad_device representing the
77 * module's reference taken when allocating the ib_umad_device in
80 * When destroying an ib_umad_device, we clear all of its
81 * ib_umad_ports from umad_port[] while holding port_lock before
82 * dropping the module's reference to the ib_umad_device. This is
83 * always safe because any open() calls will either succeed and obtain
84 * a reference before we clear the umad_port[] entries, or fail after
85 * we clear the umad_port[] entries.
90 struct class_device *class_dev;
93 struct class_device *sm_class_dev;
94 struct semaphore sm_sem;
96 struct rw_semaphore mutex;
97 struct list_head file_list;
99 struct ib_device *ib_dev;
100 struct ib_umad_device *umad_dev;
105 struct ib_umad_device {
106 int start_port, end_port;
108 struct ib_umad_port port[0];
111 struct ib_umad_file {
112 struct ib_umad_port *port;
113 struct list_head recv_list;
114 struct list_head send_list;
115 struct list_head port_list;
116 spinlock_t recv_lock;
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_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
137 static DEFINE_SPINLOCK(port_lock);
138 static struct ib_umad_port *umad_port[IB_UMAD_MAX_PORTS];
139 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
141 static void ib_umad_add_one(struct ib_device *device);
142 static void ib_umad_remove_one(struct ib_device *device);
144 static void ib_umad_release_dev(struct kref *ref)
146 struct ib_umad_device *dev =
147 container_of(ref, struct ib_umad_device, ref);
152 static int hdr_size(struct ib_umad_file *file)
154 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
155 sizeof (struct ib_user_mad_hdr_old);
158 /* caller must hold port->mutex at least for reading */
159 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
161 return file->agents_dead ? NULL : file->agent[id];
164 static int queue_packet(struct ib_umad_file *file,
165 struct ib_mad_agent *agent,
166 struct ib_umad_packet *packet)
170 down_read(&file->port->mutex);
172 for (packet->mad.hdr.id = 0;
173 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
174 packet->mad.hdr.id++)
175 if (agent == __get_agent(file, packet->mad.hdr.id)) {
176 spin_lock_irq(&file->recv_lock);
177 list_add_tail(&packet->list, &file->recv_list);
178 spin_unlock_irq(&file->recv_lock);
179 wake_up_interruptible(&file->recv_wait);
184 up_read(&file->port->mutex);
189 static void dequeue_send(struct ib_umad_file *file,
190 struct ib_umad_packet *packet)
192 spin_lock_irq(&file->send_lock);
193 list_del(&packet->list);
194 spin_unlock_irq(&file->send_lock);
197 static void send_handler(struct ib_mad_agent *agent,
198 struct ib_mad_send_wc *send_wc)
200 struct ib_umad_file *file = agent->context;
201 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
203 dequeue_send(file, packet);
204 ib_destroy_ah(packet->msg->ah);
205 ib_free_send_mad(packet->msg);
207 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
208 packet->length = IB_MGMT_MAD_HDR;
209 packet->mad.hdr.status = ETIMEDOUT;
210 if (!queue_packet(file, agent, packet))
216 static void recv_handler(struct ib_mad_agent *agent,
217 struct ib_mad_recv_wc *mad_recv_wc)
219 struct ib_umad_file *file = agent->context;
220 struct ib_umad_packet *packet;
222 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
225 packet = kzalloc(sizeof *packet, GFP_KERNEL);
229 packet->length = mad_recv_wc->mad_len;
230 packet->recv_wc = mad_recv_wc;
232 packet->mad.hdr.status = 0;
233 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
234 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
235 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
236 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
237 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
238 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
239 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
240 if (packet->mad.hdr.grh_present) {
241 struct ib_ah_attr ah_attr;
243 ib_init_ah_from_wc(agent->device, agent->port_num,
244 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
247 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
248 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
249 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
250 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
251 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
254 if (queue_packet(file, agent, packet))
261 ib_free_recv_mad(mad_recv_wc);
264 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
265 struct ib_umad_packet *packet, size_t count)
267 struct ib_mad_recv_buf *recv_buf;
268 int left, seg_payload, offset, max_seg_payload;
270 /* We need enough room to copy the first (or only) MAD segment. */
271 recv_buf = &packet->recv_wc->recv_buf;
272 if ((packet->length <= sizeof (*recv_buf->mad) &&
273 count < hdr_size(file) + packet->length) ||
274 (packet->length > sizeof (*recv_buf->mad) &&
275 count < hdr_size(file) + sizeof (*recv_buf->mad)))
278 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
281 buf += hdr_size(file);
282 seg_payload = min_t(int, packet->length, sizeof (*recv_buf->mad));
283 if (copy_to_user(buf, recv_buf->mad, seg_payload))
286 if (seg_payload < packet->length) {
288 * Multipacket RMPP MAD message. Copy remainder of message.
289 * Note that last segment may have a shorter payload.
291 if (count < hdr_size(file) + packet->length) {
293 * The buffer is too small, return the first RMPP segment,
294 * which includes the RMPP message length.
298 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
299 max_seg_payload = sizeof (struct ib_mad) - offset;
301 for (left = packet->length - seg_payload, buf += seg_payload;
302 left; left -= seg_payload, buf += seg_payload) {
303 recv_buf = container_of(recv_buf->list.next,
304 struct ib_mad_recv_buf, list);
305 seg_payload = min(left, max_seg_payload);
306 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
311 return hdr_size(file) + packet->length;
314 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
315 struct ib_umad_packet *packet, size_t count)
317 ssize_t size = hdr_size(file) + packet->length;
322 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
325 buf += hdr_size(file);
327 if (copy_to_user(buf, packet->mad.data, packet->length))
333 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
334 size_t count, loff_t *pos)
336 struct ib_umad_file *file = filp->private_data;
337 struct ib_umad_packet *packet;
340 if (count < hdr_size(file))
343 spin_lock_irq(&file->recv_lock);
345 while (list_empty(&file->recv_list)) {
346 spin_unlock_irq(&file->recv_lock);
348 if (filp->f_flags & O_NONBLOCK)
351 if (wait_event_interruptible(file->recv_wait,
352 !list_empty(&file->recv_list)))
355 spin_lock_irq(&file->recv_lock);
358 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
359 list_del(&packet->list);
361 spin_unlock_irq(&file->recv_lock);
364 ret = copy_recv_mad(file, buf, packet, count);
366 ret = copy_send_mad(file, buf, packet, count);
370 spin_lock_irq(&file->recv_lock);
371 list_add(&packet->list, &file->recv_list);
372 spin_unlock_irq(&file->recv_lock);
375 ib_free_recv_mad(packet->recv_wc);
381 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
385 /* Copy class specific header */
386 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
387 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
388 msg->hdr_len - IB_MGMT_RMPP_HDR))
391 /* All headers are in place. Copy data segments. */
392 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
393 seg++, left -= msg->seg_size, buf += msg->seg_size) {
394 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
395 min(left, msg->seg_size)))
401 static int same_destination(struct ib_user_mad_hdr *hdr1,
402 struct ib_user_mad_hdr *hdr2)
404 if (!hdr1->grh_present && !hdr2->grh_present)
405 return (hdr1->lid == hdr2->lid);
407 if (hdr1->grh_present && hdr2->grh_present)
408 return !memcmp(hdr1->gid, hdr2->gid, 16);
413 static int is_duplicate(struct ib_umad_file *file,
414 struct ib_umad_packet *packet)
416 struct ib_umad_packet *sent_packet;
417 struct ib_mad_hdr *sent_hdr, *hdr;
419 hdr = (struct ib_mad_hdr *) packet->mad.data;
420 list_for_each_entry(sent_packet, &file->send_list, list) {
421 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
423 if ((hdr->tid != sent_hdr->tid) ||
424 (hdr->mgmt_class != sent_hdr->mgmt_class))
428 * No need to be overly clever here. If two new operations have
429 * the same TID, reject the second as a duplicate. This is more
430 * restrictive than required by the spec.
432 if (!ib_response_mad((struct ib_mad *) hdr)) {
433 if (!ib_response_mad((struct ib_mad *) sent_hdr))
436 } else if (!ib_response_mad((struct ib_mad *) sent_hdr))
439 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
446 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
447 size_t count, loff_t *pos)
449 struct ib_umad_file *file = filp->private_data;
450 struct ib_umad_packet *packet;
451 struct ib_mad_agent *agent;
452 struct ib_ah_attr ah_attr;
454 struct ib_rmpp_mad *rmpp_mad;
456 int ret, data_len, hdr_len, copy_offset, rmpp_active;
458 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
461 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
465 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
470 if (packet->mad.hdr.id < 0 ||
471 packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
476 buf += hdr_size(file);
478 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
483 down_read(&file->port->mutex);
485 agent = __get_agent(file, packet->mad.hdr.id);
491 memset(&ah_attr, 0, sizeof ah_attr);
492 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
493 ah_attr.sl = packet->mad.hdr.sl;
494 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
495 ah_attr.port_num = file->port->port_num;
496 if (packet->mad.hdr.grh_present) {
497 ah_attr.ah_flags = IB_AH_GRH;
498 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
499 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index;
500 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label);
501 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
502 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
505 ah = ib_create_ah(agent->qp->pd, &ah_attr);
511 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
512 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
513 if (!ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)) {
514 copy_offset = IB_MGMT_MAD_HDR;
517 copy_offset = IB_MGMT_RMPP_HDR;
518 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
519 IB_MGMT_RMPP_FLAG_ACTIVE;
522 data_len = count - hdr_size(file) - hdr_len;
523 packet->msg = ib_create_send_mad(agent,
524 be32_to_cpu(packet->mad.hdr.qpn),
525 packet->mad.hdr.pkey_index, rmpp_active,
526 hdr_len, data_len, GFP_KERNEL);
527 if (IS_ERR(packet->msg)) {
528 ret = PTR_ERR(packet->msg);
532 packet->msg->ah = ah;
533 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
534 packet->msg->retries = packet->mad.hdr.retries;
535 packet->msg->context[0] = packet;
537 /* Copy MAD header. Any RMPP header is already in place. */
538 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
541 if (copy_from_user(packet->msg->mad + copy_offset,
543 hdr_len + data_len - copy_offset)) {
548 ret = copy_rmpp_mad(packet->msg, buf);
554 * Set the high-order part of the transaction ID to make MADs from
555 * different agents unique, and allow routing responses back to the
556 * original requestor.
558 if (!ib_response_mad(packet->msg->mad)) {
559 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
560 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
561 (be64_to_cpup(tid) & 0xffffffff));
562 rmpp_mad->mad_hdr.tid = *tid;
565 spin_lock_irq(&file->send_lock);
566 ret = is_duplicate(file, packet);
568 list_add_tail(&packet->list, &file->send_list);
569 spin_unlock_irq(&file->send_lock);
575 ret = ib_post_send_mad(packet->msg, NULL);
579 up_read(&file->port->mutex);
583 dequeue_send(file, packet);
585 ib_free_send_mad(packet->msg);
589 up_read(&file->port->mutex);
595 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
597 struct ib_umad_file *file = filp->private_data;
599 /* we will always be able to post a MAD send */
600 unsigned int mask = POLLOUT | POLLWRNORM;
602 poll_wait(filp, &file->recv_wait, wait);
604 if (!list_empty(&file->recv_list))
605 mask |= POLLIN | POLLRDNORM;
610 static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
612 struct ib_user_mad_reg_req ureq;
613 struct ib_mad_reg_req req;
614 struct ib_mad_agent *agent;
618 down_write(&file->port->mutex);
620 if (!file->port->ib_dev) {
625 if (copy_from_user(&ureq, (void __user *) arg, sizeof ureq)) {
630 if (ureq.qpn != 0 && ureq.qpn != 1) {
635 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
636 if (!__get_agent(file, agent_id))
643 if (ureq.mgmt_class) {
644 req.mgmt_class = ureq.mgmt_class;
645 req.mgmt_class_version = ureq.mgmt_class_version;
646 memcpy(req.method_mask, ureq.method_mask, sizeof req.method_mask);
647 memcpy(req.oui, ureq.oui, sizeof req.oui);
650 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
651 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
652 ureq.mgmt_class ? &req : NULL,
654 send_handler, recv_handler, file);
656 ret = PTR_ERR(agent);
660 if (put_user(agent_id,
661 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
663 ib_unregister_mad_agent(agent);
667 if (!file->already_used) {
668 file->already_used = 1;
669 if (!file->use_pkey_index) {
670 printk(KERN_WARNING "user_mad: process %s did not enable "
671 "P_Key index support.\n", current->comm);
672 printk(KERN_WARNING "user_mad: Documentation/infiniband/user_mad.txt "
673 "has info on the new ABI.\n");
677 file->agent[agent_id] = agent;
681 up_write(&file->port->mutex);
685 static int ib_umad_unreg_agent(struct ib_umad_file *file, unsigned long arg)
687 struct ib_mad_agent *agent = NULL;
691 if (get_user(id, (u32 __user *) arg))
694 down_write(&file->port->mutex);
696 if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
701 agent = file->agent[id];
702 file->agent[id] = NULL;
705 up_write(&file->port->mutex);
708 ib_unregister_mad_agent(agent);
713 static long ib_umad_enable_pkey(struct ib_umad_file *file)
717 down_write(&file->port->mutex);
718 if (file->already_used)
721 file->use_pkey_index = 1;
722 up_write(&file->port->mutex);
727 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
731 case IB_USER_MAD_REGISTER_AGENT:
732 return ib_umad_reg_agent(filp->private_data, arg);
733 case IB_USER_MAD_UNREGISTER_AGENT:
734 return ib_umad_unreg_agent(filp->private_data, arg);
735 case IB_USER_MAD_ENABLE_PKEY:
736 return ib_umad_enable_pkey(filp->private_data);
742 static int ib_umad_open(struct inode *inode, struct file *filp)
744 struct ib_umad_port *port;
745 struct ib_umad_file *file;
748 spin_lock(&port_lock);
749 port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE];
751 kref_get(&port->umad_dev->ref);
752 spin_unlock(&port_lock);
757 down_write(&port->mutex);
764 file = kzalloc(sizeof *file, GFP_KERNEL);
766 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
771 spin_lock_init(&file->recv_lock);
772 spin_lock_init(&file->send_lock);
773 INIT_LIST_HEAD(&file->recv_list);
774 INIT_LIST_HEAD(&file->send_list);
775 init_waitqueue_head(&file->recv_wait);
778 filp->private_data = file;
780 list_add_tail(&file->port_list, &port->file_list);
783 up_write(&port->mutex);
787 static int ib_umad_close(struct inode *inode, struct file *filp)
789 struct ib_umad_file *file = filp->private_data;
790 struct ib_umad_device *dev = file->port->umad_dev;
791 struct ib_umad_packet *packet, *tmp;
795 down_write(&file->port->mutex);
797 already_dead = file->agents_dead;
798 file->agents_dead = 1;
800 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
802 ib_free_recv_mad(packet->recv_wc);
806 list_del(&file->port_list);
808 downgrade_write(&file->port->mutex);
811 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
813 ib_unregister_mad_agent(file->agent[i]);
815 up_read(&file->port->mutex);
818 kref_put(&dev->ref, ib_umad_release_dev);
823 static const struct file_operations umad_fops = {
824 .owner = THIS_MODULE,
825 .read = ib_umad_read,
826 .write = ib_umad_write,
827 .poll = ib_umad_poll,
828 .unlocked_ioctl = ib_umad_ioctl,
829 .compat_ioctl = ib_umad_ioctl,
830 .open = ib_umad_open,
831 .release = ib_umad_close
834 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
836 struct ib_umad_port *port;
837 struct ib_port_modify props = {
838 .set_port_cap_mask = IB_PORT_SM
842 spin_lock(&port_lock);
843 port = umad_port[iminor(inode) - IB_UMAD_MINOR_BASE - IB_UMAD_MAX_PORTS];
845 kref_get(&port->umad_dev->ref);
846 spin_unlock(&port_lock);
851 if (filp->f_flags & O_NONBLOCK) {
852 if (down_trylock(&port->sm_sem)) {
857 if (down_interruptible(&port->sm_sem)) {
863 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
869 filp->private_data = port;
874 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
878 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
880 struct ib_umad_port *port = filp->private_data;
881 struct ib_port_modify props = {
882 .clr_port_cap_mask = IB_PORT_SM
886 down_write(&port->mutex);
888 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
889 up_write(&port->mutex);
893 kref_put(&port->umad_dev->ref, ib_umad_release_dev);
898 static const struct file_operations umad_sm_fops = {
899 .owner = THIS_MODULE,
900 .open = ib_umad_sm_open,
901 .release = ib_umad_sm_close
904 static struct ib_client umad_client = {
906 .add = ib_umad_add_one,
907 .remove = ib_umad_remove_one
910 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
912 struct ib_umad_port *port = class_get_devdata(class_dev);
917 return sprintf(buf, "%s\n", port->ib_dev->name);
919 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
921 static ssize_t show_port(struct class_device *class_dev, char *buf)
923 struct ib_umad_port *port = class_get_devdata(class_dev);
928 return sprintf(buf, "%d\n", port->port_num);
930 static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
932 static ssize_t show_abi_version(struct class *class, char *buf)
934 return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
936 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
938 static int ib_umad_init_port(struct ib_device *device, int port_num,
939 struct ib_umad_port *port)
941 spin_lock(&port_lock);
942 port->dev_num = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
943 if (port->dev_num >= IB_UMAD_MAX_PORTS) {
944 spin_unlock(&port_lock);
947 set_bit(port->dev_num, dev_map);
948 spin_unlock(&port_lock);
950 port->ib_dev = device;
951 port->port_num = port_num;
952 init_MUTEX(&port->sm_sem);
953 init_rwsem(&port->mutex);
954 INIT_LIST_HEAD(&port->file_list);
956 port->dev = cdev_alloc();
959 port->dev->owner = THIS_MODULE;
960 port->dev->ops = &umad_fops;
961 kobject_set_name(&port->dev->kobj, "umad%d", port->dev_num);
962 if (cdev_add(port->dev, base_dev + port->dev_num, 1))
965 port->class_dev = class_device_create(umad_class, NULL, port->dev->dev,
967 "umad%d", port->dev_num);
968 if (IS_ERR(port->class_dev))
971 if (class_device_create_file(port->class_dev, &class_device_attr_ibdev))
973 if (class_device_create_file(port->class_dev, &class_device_attr_port))
976 port->sm_dev = cdev_alloc();
979 port->sm_dev->owner = THIS_MODULE;
980 port->sm_dev->ops = &umad_sm_fops;
981 kobject_set_name(&port->sm_dev->kobj, "issm%d", port->dev_num);
982 if (cdev_add(port->sm_dev, base_dev + port->dev_num + IB_UMAD_MAX_PORTS, 1))
985 port->sm_class_dev = class_device_create(umad_class, NULL, port->sm_dev->dev,
987 "issm%d", port->dev_num);
988 if (IS_ERR(port->sm_class_dev))
991 class_set_devdata(port->class_dev, port);
992 class_set_devdata(port->sm_class_dev, port);
994 if (class_device_create_file(port->sm_class_dev, &class_device_attr_ibdev))
996 if (class_device_create_file(port->sm_class_dev, &class_device_attr_port))
999 spin_lock(&port_lock);
1000 umad_port[port->dev_num] = port;
1001 spin_unlock(&port_lock);
1006 class_device_destroy(umad_class, port->sm_dev->dev);
1009 cdev_del(port->sm_dev);
1012 class_device_destroy(umad_class, port->dev->dev);
1015 cdev_del(port->dev);
1016 clear_bit(port->dev_num, dev_map);
1021 static void ib_umad_kill_port(struct ib_umad_port *port)
1023 struct ib_umad_file *file;
1026 class_set_devdata(port->class_dev, NULL);
1027 class_set_devdata(port->sm_class_dev, NULL);
1029 class_device_destroy(umad_class, port->dev->dev);
1030 class_device_destroy(umad_class, port->sm_dev->dev);
1032 cdev_del(port->dev);
1033 cdev_del(port->sm_dev);
1035 spin_lock(&port_lock);
1036 umad_port[port->dev_num] = NULL;
1037 spin_unlock(&port_lock);
1039 down_write(&port->mutex);
1041 port->ib_dev = NULL;
1044 * Now go through the list of files attached to this port and
1045 * unregister all of their MAD agents. We need to hold
1046 * port->mutex while doing this to avoid racing with
1047 * ib_umad_close(), but we can't hold the mutex for writing
1048 * while calling ib_unregister_mad_agent(), since that might
1049 * deadlock by calling back into queue_packet(). So we
1050 * downgrade our lock to a read lock, and then drop and
1051 * reacquire the write lock for the next iteration.
1053 * We do list_del_init() on the file's list_head so that the
1054 * list_del in ib_umad_close() is still OK, even after the
1055 * file is removed from the list.
1057 while (!list_empty(&port->file_list)) {
1058 file = list_entry(port->file_list.next, struct ib_umad_file,
1061 file->agents_dead = 1;
1062 list_del_init(&file->port_list);
1064 downgrade_write(&port->mutex);
1066 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1067 if (file->agent[id])
1068 ib_unregister_mad_agent(file->agent[id]);
1070 up_read(&port->mutex);
1071 down_write(&port->mutex);
1074 up_write(&port->mutex);
1076 clear_bit(port->dev_num, dev_map);
1079 static void ib_umad_add_one(struct ib_device *device)
1081 struct ib_umad_device *umad_dev;
1084 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1087 if (device->node_type == RDMA_NODE_IB_SWITCH)
1091 e = device->phys_port_cnt;
1094 umad_dev = kzalloc(sizeof *umad_dev +
1095 (e - s + 1) * sizeof (struct ib_umad_port),
1100 kref_init(&umad_dev->ref);
1102 umad_dev->start_port = s;
1103 umad_dev->end_port = e;
1105 for (i = s; i <= e; ++i) {
1106 umad_dev->port[i - s].umad_dev = umad_dev;
1108 if (ib_umad_init_port(device, i, &umad_dev->port[i - s]))
1112 ib_set_client_data(device, &umad_client, umad_dev);
1118 ib_umad_kill_port(&umad_dev->port[i - s]);
1120 kref_put(&umad_dev->ref, ib_umad_release_dev);
1123 static void ib_umad_remove_one(struct ib_device *device)
1125 struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
1131 for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
1132 ib_umad_kill_port(&umad_dev->port[i]);
1134 kref_put(&umad_dev->ref, ib_umad_release_dev);
1137 static int __init ib_umad_init(void)
1141 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1144 printk(KERN_ERR "user_mad: couldn't register device number\n");
1148 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1149 if (IS_ERR(umad_class)) {
1150 ret = PTR_ERR(umad_class);
1151 printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n");
1155 ret = class_create_file(umad_class, &class_attr_abi_version);
1157 printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n");
1161 ret = ib_register_client(&umad_client);
1163 printk(KERN_ERR "user_mad: couldn't register ib_umad client\n");
1170 class_destroy(umad_class);
1173 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1179 static void __exit ib_umad_cleanup(void)
1181 ib_unregister_client(&umad_client);
1182 class_destroy(umad_class);
1183 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1186 module_init(ib_umad_init);
1187 module_exit(ib_umad_cleanup);