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 <asm/uaccess.h>
55 #include <rdma/ib_mad.h>
56 #include <rdma/ib_user_mad.h>
58 MODULE_AUTHOR("Roland Dreier");
59 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
60 MODULE_LICENSE("Dual BSD/GPL");
63 IB_UMAD_MAX_PORTS = 64,
64 IB_UMAD_MAX_AGENTS = 32,
67 IB_UMAD_MINOR_BASE = 0
71 * Our lifetime rules for these structs are the following:
72 * device special file is opened, we take a reference on the
73 * ib_umad_port's struct ib_umad_device. We drop these
74 * references in the corresponding close().
76 * In addition to references coming from open character devices, there
77 * is one more reference to each ib_umad_device representing the
78 * module's reference taken when allocating the ib_umad_device in
81 * When destroying an ib_umad_device, we drop the module's reference.
89 struct device *sm_dev;
90 struct semaphore sm_sem;
92 struct mutex file_mutex;
93 struct list_head file_list;
95 struct ib_device *ib_dev;
96 struct ib_umad_device *umad_dev;
101 struct ib_umad_device {
103 struct ib_umad_port port[0];
106 struct ib_umad_file {
108 struct ib_umad_port *port;
109 struct list_head recv_list;
110 struct list_head send_list;
111 struct list_head port_list;
112 spinlock_t send_lock;
113 wait_queue_head_t recv_wait;
114 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
120 struct ib_umad_packet {
121 struct ib_mad_send_buf *msg;
122 struct ib_mad_recv_wc *recv_wc;
123 struct list_head list;
125 struct ib_user_mad mad;
128 static struct class *umad_class;
130 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
132 static DEFINE_SPINLOCK(port_lock);
133 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
135 static void ib_umad_add_one(struct ib_device *device);
136 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
138 static void ib_umad_release_dev(struct kobject *kobj)
140 struct ib_umad_device *dev =
141 container_of(kobj, struct ib_umad_device, kobj);
146 static struct kobj_type ib_umad_dev_ktype = {
147 .release = ib_umad_release_dev,
150 static int hdr_size(struct ib_umad_file *file)
152 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
153 sizeof (struct ib_user_mad_hdr_old);
156 /* caller must hold file->mutex */
157 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
159 return file->agents_dead ? NULL : file->agent[id];
162 static int queue_packet(struct ib_umad_file *file,
163 struct ib_mad_agent *agent,
164 struct ib_umad_packet *packet)
168 mutex_lock(&file->mutex);
170 for (packet->mad.hdr.id = 0;
171 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
172 packet->mad.hdr.id++)
173 if (agent == __get_agent(file, packet->mad.hdr.id)) {
174 list_add_tail(&packet->list, &file->recv_list);
175 wake_up_interruptible(&file->recv_wait);
180 mutex_unlock(&file->mutex);
185 static void dequeue_send(struct ib_umad_file *file,
186 struct ib_umad_packet *packet)
188 spin_lock_irq(&file->send_lock);
189 list_del(&packet->list);
190 spin_unlock_irq(&file->send_lock);
193 static void send_handler(struct ib_mad_agent *agent,
194 struct ib_mad_send_wc *send_wc)
196 struct ib_umad_file *file = agent->context;
197 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
199 dequeue_send(file, packet);
200 ib_destroy_ah(packet->msg->ah);
201 ib_free_send_mad(packet->msg);
203 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
204 packet->length = IB_MGMT_MAD_HDR;
205 packet->mad.hdr.status = ETIMEDOUT;
206 if (!queue_packet(file, agent, packet))
212 static void recv_handler(struct ib_mad_agent *agent,
213 struct ib_mad_recv_wc *mad_recv_wc)
215 struct ib_umad_file *file = agent->context;
216 struct ib_umad_packet *packet;
218 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
221 packet = kzalloc(sizeof *packet, GFP_KERNEL);
225 packet->length = mad_recv_wc->mad_len;
226 packet->recv_wc = mad_recv_wc;
228 packet->mad.hdr.status = 0;
229 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
230 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
231 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
232 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
233 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
234 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
235 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
236 if (packet->mad.hdr.grh_present) {
237 struct ib_ah_attr ah_attr;
239 ib_init_ah_from_wc(agent->device, agent->port_num,
240 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
243 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
244 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
245 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
246 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
247 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
250 if (queue_packet(file, agent, packet))
257 ib_free_recv_mad(mad_recv_wc);
260 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
261 struct ib_umad_packet *packet, size_t count)
263 struct ib_mad_recv_buf *recv_buf;
264 int left, seg_payload, offset, max_seg_payload;
267 recv_buf = &packet->recv_wc->recv_buf;
268 seg_size = packet->recv_wc->mad_seg_size;
270 /* We need enough room to copy the first (or only) MAD segment. */
271 if ((packet->length <= seg_size &&
272 count < hdr_size(file) + packet->length) ||
273 (packet->length > seg_size &&
274 count < hdr_size(file) + seg_size))
277 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
280 buf += hdr_size(file);
281 seg_payload = min_t(int, packet->length, seg_size);
282 if (copy_to_user(buf, recv_buf->mad, seg_payload))
285 if (seg_payload < packet->length) {
287 * Multipacket RMPP MAD message. Copy remainder of message.
288 * Note that last segment may have a shorter payload.
290 if (count < hdr_size(file) + packet->length) {
292 * The buffer is too small, return the first RMPP segment,
293 * which includes the RMPP message length.
297 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
298 max_seg_payload = seg_size - offset;
300 for (left = packet->length - seg_payload, buf += seg_payload;
301 left; left -= seg_payload, buf += seg_payload) {
302 recv_buf = container_of(recv_buf->list.next,
303 struct ib_mad_recv_buf, list);
304 seg_payload = min(left, max_seg_payload);
305 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
310 return hdr_size(file) + packet->length;
313 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
314 struct ib_umad_packet *packet, size_t count)
316 ssize_t size = hdr_size(file) + packet->length;
321 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
324 buf += hdr_size(file);
326 if (copy_to_user(buf, packet->mad.data, packet->length))
332 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
333 size_t count, loff_t *pos)
335 struct ib_umad_file *file = filp->private_data;
336 struct ib_umad_packet *packet;
339 if (count < hdr_size(file))
342 mutex_lock(&file->mutex);
344 while (list_empty(&file->recv_list)) {
345 mutex_unlock(&file->mutex);
347 if (filp->f_flags & O_NONBLOCK)
350 if (wait_event_interruptible(file->recv_wait,
351 !list_empty(&file->recv_list)))
354 mutex_lock(&file->mutex);
357 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
358 list_del(&packet->list);
360 mutex_unlock(&file->mutex);
363 ret = copy_recv_mad(file, buf, packet, count);
365 ret = copy_send_mad(file, buf, packet, count);
369 mutex_lock(&file->mutex);
370 list_add(&packet->list, &file->recv_list);
371 mutex_unlock(&file->mutex);
374 ib_free_recv_mad(packet->recv_wc);
380 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
384 /* Copy class specific header */
385 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
386 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
387 msg->hdr_len - IB_MGMT_RMPP_HDR))
390 /* All headers are in place. Copy data segments. */
391 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
392 seg++, left -= msg->seg_size, buf += msg->seg_size) {
393 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
394 min(left, msg->seg_size)))
400 static int same_destination(struct ib_user_mad_hdr *hdr1,
401 struct ib_user_mad_hdr *hdr2)
403 if (!hdr1->grh_present && !hdr2->grh_present)
404 return (hdr1->lid == hdr2->lid);
406 if (hdr1->grh_present && hdr2->grh_present)
407 return !memcmp(hdr1->gid, hdr2->gid, 16);
412 static int is_duplicate(struct ib_umad_file *file,
413 struct ib_umad_packet *packet)
415 struct ib_umad_packet *sent_packet;
416 struct ib_mad_hdr *sent_hdr, *hdr;
418 hdr = (struct ib_mad_hdr *) packet->mad.data;
419 list_for_each_entry(sent_packet, &file->send_list, list) {
420 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
422 if ((hdr->tid != sent_hdr->tid) ||
423 (hdr->mgmt_class != sent_hdr->mgmt_class))
427 * No need to be overly clever here. If two new operations have
428 * the same TID, reject the second as a duplicate. This is more
429 * restrictive than required by the spec.
431 if (!ib_response_mad(hdr)) {
432 if (!ib_response_mad(sent_hdr))
435 } else if (!ib_response_mad(sent_hdr))
438 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
445 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
446 size_t count, loff_t *pos)
448 struct ib_umad_file *file = filp->private_data;
449 struct ib_umad_packet *packet;
450 struct ib_mad_agent *agent;
451 struct ib_ah_attr ah_attr;
453 struct ib_rmpp_mad *rmpp_mad;
455 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 >= IB_UMAD_MAX_AGENTS) {
475 buf += hdr_size(file);
477 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
482 mutex_lock(&file->mutex);
484 agent = __get_agent(file, packet->mad.hdr.id);
490 memset(&ah_attr, 0, sizeof ah_attr);
491 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
492 ah_attr.sl = packet->mad.hdr.sl;
493 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
494 ah_attr.port_num = file->port->port_num;
495 if (packet->mad.hdr.grh_present) {
496 ah_attr.ah_flags = IB_AH_GRH;
497 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
498 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index;
499 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label);
500 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
501 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
504 ah = ib_create_ah(agent->qp->pd, &ah_attr);
510 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
511 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 && ib_mad_kernel_rmpp_agent(agent)) {
515 copy_offset = IB_MGMT_RMPP_HDR;
516 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
517 IB_MGMT_RMPP_FLAG_ACTIVE;
519 copy_offset = IB_MGMT_MAD_HDR;
523 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
524 data_len = count - hdr_size(file) - hdr_len;
525 packet->msg = ib_create_send_mad(agent,
526 be32_to_cpu(packet->mad.hdr.qpn),
527 packet->mad.hdr.pkey_index, rmpp_active,
528 hdr_len, data_len, GFP_KERNEL,
530 if (IS_ERR(packet->msg)) {
531 ret = PTR_ERR(packet->msg);
535 packet->msg->ah = ah;
536 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
537 packet->msg->retries = packet->mad.hdr.retries;
538 packet->msg->context[0] = packet;
540 /* Copy MAD header. Any RMPP header is already in place. */
541 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
544 if (copy_from_user(packet->msg->mad + copy_offset,
546 hdr_len + data_len - copy_offset)) {
551 ret = copy_rmpp_mad(packet->msg, buf);
557 * Set the high-order part of the transaction ID to make MADs from
558 * different agents unique, and allow routing responses back to the
559 * original requestor.
561 if (!ib_response_mad(packet->msg->mad)) {
562 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
563 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
564 (be64_to_cpup(tid) & 0xffffffff));
565 rmpp_mad->mad_hdr.tid = *tid;
568 if (!ib_mad_kernel_rmpp_agent(agent)
569 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
570 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
571 spin_lock_irq(&file->send_lock);
572 list_add_tail(&packet->list, &file->send_list);
573 spin_unlock_irq(&file->send_lock);
575 spin_lock_irq(&file->send_lock);
576 ret = is_duplicate(file, packet);
578 list_add_tail(&packet->list, &file->send_list);
579 spin_unlock_irq(&file->send_lock);
586 ret = ib_post_send_mad(packet->msg, NULL);
590 mutex_unlock(&file->mutex);
594 dequeue_send(file, packet);
596 ib_free_send_mad(packet->msg);
600 mutex_unlock(&file->mutex);
606 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
608 struct ib_umad_file *file = filp->private_data;
610 /* we will always be able to post a MAD send */
611 unsigned int mask = POLLOUT | POLLWRNORM;
613 poll_wait(filp, &file->recv_wait, wait);
615 if (!list_empty(&file->recv_list))
616 mask |= POLLIN | POLLRDNORM;
621 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
622 int compat_method_mask)
624 struct ib_user_mad_reg_req ureq;
625 struct ib_mad_reg_req req;
626 struct ib_mad_agent *agent = NULL;
630 mutex_lock(&file->port->file_mutex);
631 mutex_lock(&file->mutex);
633 if (!file->port->ib_dev) {
634 dev_notice(file->port->dev,
635 "ib_umad_reg_agent: invalid device\n");
640 if (copy_from_user(&ureq, arg, sizeof ureq)) {
645 if (ureq.qpn != 0 && ureq.qpn != 1) {
646 dev_notice(file->port->dev,
647 "ib_umad_reg_agent: invalid QPN %d specified\n",
653 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
654 if (!__get_agent(file, agent_id))
657 dev_notice(file->port->dev,
658 "ib_umad_reg_agent: Max Agents (%u) reached\n",
664 if (ureq.mgmt_class) {
665 memset(&req, 0, sizeof(req));
666 req.mgmt_class = ureq.mgmt_class;
667 req.mgmt_class_version = ureq.mgmt_class_version;
668 memcpy(req.oui, ureq.oui, sizeof req.oui);
670 if (compat_method_mask) {
671 u32 *umm = (u32 *) ureq.method_mask;
674 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
676 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
678 memcpy(req.method_mask, ureq.method_mask,
679 sizeof req.method_mask);
682 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
683 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
684 ureq.mgmt_class ? &req : NULL,
686 send_handler, recv_handler, file, 0);
688 ret = PTR_ERR(agent);
693 if (put_user(agent_id,
694 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
699 if (!file->already_used) {
700 file->already_used = 1;
701 if (!file->use_pkey_index) {
702 dev_warn(file->port->dev,
703 "process %s did not enable P_Key index support.\n",
705 dev_warn(file->port->dev,
706 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
710 file->agent[agent_id] = agent;
714 mutex_unlock(&file->mutex);
717 ib_unregister_mad_agent(agent);
719 mutex_unlock(&file->port->file_mutex);
724 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
726 struct ib_user_mad_reg_req2 ureq;
727 struct ib_mad_reg_req req;
728 struct ib_mad_agent *agent = NULL;
732 mutex_lock(&file->port->file_mutex);
733 mutex_lock(&file->mutex);
735 if (!file->port->ib_dev) {
736 dev_notice(file->port->dev,
737 "ib_umad_reg_agent2: invalid device\n");
742 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
747 if (ureq.qpn != 0 && ureq.qpn != 1) {
748 dev_notice(file->port->dev,
749 "ib_umad_reg_agent2: invalid QPN %d specified\n",
755 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
756 dev_notice(file->port->dev,
757 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
758 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
761 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
762 (u32 __user *) (arg + offsetof(struct
763 ib_user_mad_reg_req2, flags))))
769 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
770 if (!__get_agent(file, agent_id))
773 dev_notice(file->port->dev,
774 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
780 if (ureq.mgmt_class) {
781 memset(&req, 0, sizeof(req));
782 req.mgmt_class = ureq.mgmt_class;
783 req.mgmt_class_version = ureq.mgmt_class_version;
784 if (ureq.oui & 0xff000000) {
785 dev_notice(file->port->dev,
786 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
791 req.oui[2] = ureq.oui & 0x0000ff;
792 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
793 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
794 memcpy(req.method_mask, ureq.method_mask,
795 sizeof(req.method_mask));
798 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
799 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
800 ureq.mgmt_class ? &req : NULL,
802 send_handler, recv_handler, file,
805 ret = PTR_ERR(agent);
810 if (put_user(agent_id,
812 offsetof(struct ib_user_mad_reg_req2, id)))) {
817 if (!file->already_used) {
818 file->already_used = 1;
819 file->use_pkey_index = 1;
822 file->agent[agent_id] = agent;
826 mutex_unlock(&file->mutex);
829 ib_unregister_mad_agent(agent);
831 mutex_unlock(&file->port->file_mutex);
837 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
839 struct ib_mad_agent *agent = NULL;
843 if (get_user(id, arg))
846 mutex_lock(&file->port->file_mutex);
847 mutex_lock(&file->mutex);
849 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
854 agent = file->agent[id];
855 file->agent[id] = NULL;
858 mutex_unlock(&file->mutex);
861 ib_unregister_mad_agent(agent);
863 mutex_unlock(&file->port->file_mutex);
868 static long ib_umad_enable_pkey(struct ib_umad_file *file)
872 mutex_lock(&file->mutex);
873 if (file->already_used)
876 file->use_pkey_index = 1;
877 mutex_unlock(&file->mutex);
882 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
886 case IB_USER_MAD_REGISTER_AGENT:
887 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
888 case IB_USER_MAD_UNREGISTER_AGENT:
889 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
890 case IB_USER_MAD_ENABLE_PKEY:
891 return ib_umad_enable_pkey(filp->private_data);
892 case IB_USER_MAD_REGISTER_AGENT2:
893 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
900 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
904 case IB_USER_MAD_REGISTER_AGENT:
905 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
906 case IB_USER_MAD_UNREGISTER_AGENT:
907 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
908 case IB_USER_MAD_ENABLE_PKEY:
909 return ib_umad_enable_pkey(filp->private_data);
910 case IB_USER_MAD_REGISTER_AGENT2:
911 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
919 * ib_umad_open() does not need the BKL:
921 * - the ib_umad_port structures are properly reference counted, and
922 * everything else is purely local to the file being created, so
923 * races against other open calls are not a problem;
924 * - the ioctl method does not affect any global state outside of the
925 * file structure being operated on;
927 static int ib_umad_open(struct inode *inode, struct file *filp)
929 struct ib_umad_port *port;
930 struct ib_umad_file *file;
933 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
935 mutex_lock(&port->file_mutex);
941 file = kzalloc(sizeof *file, GFP_KERNEL);
945 mutex_init(&file->mutex);
946 spin_lock_init(&file->send_lock);
947 INIT_LIST_HEAD(&file->recv_list);
948 INIT_LIST_HEAD(&file->send_list);
949 init_waitqueue_head(&file->recv_wait);
952 filp->private_data = file;
954 list_add_tail(&file->port_list, &port->file_list);
956 ret = nonseekable_open(inode, filp);
958 list_del(&file->port_list);
963 kobject_get(&port->umad_dev->kobj);
966 mutex_unlock(&port->file_mutex);
970 static int ib_umad_close(struct inode *inode, struct file *filp)
972 struct ib_umad_file *file = filp->private_data;
973 struct ib_umad_device *dev = file->port->umad_dev;
974 struct ib_umad_packet *packet, *tmp;
978 mutex_lock(&file->port->file_mutex);
979 mutex_lock(&file->mutex);
981 already_dead = file->agents_dead;
982 file->agents_dead = 1;
984 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
986 ib_free_recv_mad(packet->recv_wc);
990 list_del(&file->port_list);
992 mutex_unlock(&file->mutex);
995 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
997 ib_unregister_mad_agent(file->agent[i]);
999 mutex_unlock(&file->port->file_mutex);
1002 kobject_put(&dev->kobj);
1007 static const struct file_operations umad_fops = {
1008 .owner = THIS_MODULE,
1009 .read = ib_umad_read,
1010 .write = ib_umad_write,
1011 .poll = ib_umad_poll,
1012 .unlocked_ioctl = ib_umad_ioctl,
1013 #ifdef CONFIG_COMPAT
1014 .compat_ioctl = ib_umad_compat_ioctl,
1016 .open = ib_umad_open,
1017 .release = ib_umad_close,
1018 .llseek = no_llseek,
1021 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1023 struct ib_umad_port *port;
1024 struct ib_port_modify props = {
1025 .set_port_cap_mask = IB_PORT_SM
1029 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1031 if (filp->f_flags & O_NONBLOCK) {
1032 if (down_trylock(&port->sm_sem)) {
1037 if (down_interruptible(&port->sm_sem)) {
1043 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1047 filp->private_data = port;
1049 ret = nonseekable_open(inode, filp);
1051 goto err_clr_sm_cap;
1053 kobject_get(&port->umad_dev->kobj);
1058 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1059 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1068 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1070 struct ib_umad_port *port = filp->private_data;
1071 struct ib_port_modify props = {
1072 .clr_port_cap_mask = IB_PORT_SM
1076 mutex_lock(&port->file_mutex);
1078 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1079 mutex_unlock(&port->file_mutex);
1083 kobject_put(&port->umad_dev->kobj);
1088 static const struct file_operations umad_sm_fops = {
1089 .owner = THIS_MODULE,
1090 .open = ib_umad_sm_open,
1091 .release = ib_umad_sm_close,
1092 .llseek = no_llseek,
1095 static struct ib_client umad_client = {
1097 .add = ib_umad_add_one,
1098 .remove = ib_umad_remove_one
1101 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1104 struct ib_umad_port *port = dev_get_drvdata(dev);
1109 return sprintf(buf, "%s\n", port->ib_dev->name);
1111 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1113 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1116 struct ib_umad_port *port = dev_get_drvdata(dev);
1121 return sprintf(buf, "%d\n", port->port_num);
1123 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1125 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1126 __stringify(IB_USER_MAD_ABI_VERSION));
1128 static dev_t overflow_maj;
1129 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1130 static int find_overflow_devnum(struct ib_device *device)
1134 if (!overflow_maj) {
1135 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1138 dev_err(&device->dev,
1139 "couldn't register dynamic device number\n");
1144 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1145 if (ret >= IB_UMAD_MAX_PORTS)
1151 static int ib_umad_init_port(struct ib_device *device, int port_num,
1152 struct ib_umad_device *umad_dev,
1153 struct ib_umad_port *port)
1158 spin_lock(&port_lock);
1159 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1160 if (devnum >= IB_UMAD_MAX_PORTS) {
1161 spin_unlock(&port_lock);
1162 devnum = find_overflow_devnum(device);
1166 spin_lock(&port_lock);
1167 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1168 base = devnum + overflow_maj;
1169 set_bit(devnum, overflow_map);
1171 port->dev_num = devnum;
1172 base = devnum + base_dev;
1173 set_bit(devnum, dev_map);
1175 spin_unlock(&port_lock);
1177 port->ib_dev = device;
1178 port->port_num = port_num;
1179 sema_init(&port->sm_sem, 1);
1180 mutex_init(&port->file_mutex);
1181 INIT_LIST_HEAD(&port->file_list);
1183 cdev_init(&port->cdev, &umad_fops);
1184 port->cdev.owner = THIS_MODULE;
1185 port->cdev.kobj.parent = &umad_dev->kobj;
1186 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1187 if (cdev_add(&port->cdev, base, 1))
1190 port->dev = device_create(umad_class, device->dma_device,
1191 port->cdev.dev, port,
1192 "umad%d", port->dev_num);
1193 if (IS_ERR(port->dev))
1196 if (device_create_file(port->dev, &dev_attr_ibdev))
1198 if (device_create_file(port->dev, &dev_attr_port))
1201 base += IB_UMAD_MAX_PORTS;
1202 cdev_init(&port->sm_cdev, &umad_sm_fops);
1203 port->sm_cdev.owner = THIS_MODULE;
1204 port->sm_cdev.kobj.parent = &umad_dev->kobj;
1205 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1206 if (cdev_add(&port->sm_cdev, base, 1))
1209 port->sm_dev = device_create(umad_class, device->dma_device,
1210 port->sm_cdev.dev, port,
1211 "issm%d", port->dev_num);
1212 if (IS_ERR(port->sm_dev))
1215 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1217 if (device_create_file(port->sm_dev, &dev_attr_port))
1223 device_destroy(umad_class, port->sm_cdev.dev);
1226 cdev_del(&port->sm_cdev);
1229 device_destroy(umad_class, port->cdev.dev);
1232 cdev_del(&port->cdev);
1233 if (port->dev_num < IB_UMAD_MAX_PORTS)
1234 clear_bit(devnum, dev_map);
1236 clear_bit(devnum, overflow_map);
1241 static void ib_umad_kill_port(struct ib_umad_port *port)
1243 struct ib_umad_file *file;
1246 dev_set_drvdata(port->dev, NULL);
1247 dev_set_drvdata(port->sm_dev, NULL);
1249 device_destroy(umad_class, port->cdev.dev);
1250 device_destroy(umad_class, port->sm_cdev.dev);
1252 cdev_del(&port->cdev);
1253 cdev_del(&port->sm_cdev);
1255 mutex_lock(&port->file_mutex);
1257 port->ib_dev = NULL;
1259 list_for_each_entry(file, &port->file_list, port_list) {
1260 mutex_lock(&file->mutex);
1261 file->agents_dead = 1;
1262 mutex_unlock(&file->mutex);
1264 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1265 if (file->agent[id])
1266 ib_unregister_mad_agent(file->agent[id]);
1269 mutex_unlock(&port->file_mutex);
1271 if (port->dev_num < IB_UMAD_MAX_PORTS)
1272 clear_bit(port->dev_num, dev_map);
1274 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1277 static void ib_umad_add_one(struct ib_device *device)
1279 struct ib_umad_device *umad_dev;
1283 s = rdma_start_port(device);
1284 e = rdma_end_port(device);
1286 umad_dev = kzalloc(sizeof *umad_dev +
1287 (e - s + 1) * sizeof (struct ib_umad_port),
1292 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1294 for (i = s; i <= e; ++i) {
1295 if (!rdma_cap_ib_mad(device, i))
1298 umad_dev->port[i - s].umad_dev = umad_dev;
1300 if (ib_umad_init_port(device, i, umad_dev,
1301 &umad_dev->port[i - s]))
1310 ib_set_client_data(device, &umad_client, umad_dev);
1316 if (!rdma_cap_ib_mad(device, i))
1319 ib_umad_kill_port(&umad_dev->port[i - s]);
1322 kobject_put(&umad_dev->kobj);
1325 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1327 struct ib_umad_device *umad_dev = client_data;
1333 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1334 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1335 ib_umad_kill_port(&umad_dev->port[i]);
1338 kobject_put(&umad_dev->kobj);
1341 static char *umad_devnode(struct device *dev, umode_t *mode)
1343 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1346 static int __init ib_umad_init(void)
1350 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1353 pr_err("couldn't register device number\n");
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(base_dev, IB_UMAD_MAX_PORTS * 2);
1390 static void __exit ib_umad_cleanup(void)
1392 ib_unregister_client(&umad_client);
1393 class_destroy(umad_class);
1394 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1396 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1399 module_init(ib_umad_init);
1400 module_exit(ib_umad_cleanup);