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 {
102 int start_port, end_port;
104 struct ib_umad_port port[0];
107 struct ib_umad_file {
109 struct ib_umad_port *port;
110 struct list_head recv_list;
111 struct list_head send_list;
112 struct list_head port_list;
113 spinlock_t send_lock;
114 wait_queue_head_t recv_wait;
115 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
121 struct ib_umad_packet {
122 struct ib_mad_send_buf *msg;
123 struct ib_mad_recv_wc *recv_wc;
124 struct list_head list;
126 struct ib_user_mad mad;
129 static struct class *umad_class;
131 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
133 static DEFINE_SPINLOCK(port_lock);
134 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS);
136 static void ib_umad_add_one(struct ib_device *device);
137 static void ib_umad_remove_one(struct ib_device *device);
139 static void ib_umad_release_dev(struct kobject *kobj)
141 struct ib_umad_device *dev =
142 container_of(kobj, struct ib_umad_device, kobj);
147 static struct kobj_type ib_umad_dev_ktype = {
148 .release = ib_umad_release_dev,
151 static int hdr_size(struct ib_umad_file *file)
153 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
154 sizeof (struct ib_user_mad_hdr_old);
157 /* caller must hold file->mutex */
158 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
160 return file->agents_dead ? NULL : file->agent[id];
163 static int queue_packet(struct ib_umad_file *file,
164 struct ib_mad_agent *agent,
165 struct ib_umad_packet *packet)
169 mutex_lock(&file->mutex);
171 for (packet->mad.hdr.id = 0;
172 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
173 packet->mad.hdr.id++)
174 if (agent == __get_agent(file, packet->mad.hdr.id)) {
175 list_add_tail(&packet->list, &file->recv_list);
176 wake_up_interruptible(&file->recv_wait);
181 mutex_unlock(&file->mutex);
186 static void dequeue_send(struct ib_umad_file *file,
187 struct ib_umad_packet *packet)
189 spin_lock_irq(&file->send_lock);
190 list_del(&packet->list);
191 spin_unlock_irq(&file->send_lock);
194 static void send_handler(struct ib_mad_agent *agent,
195 struct ib_mad_send_wc *send_wc)
197 struct ib_umad_file *file = agent->context;
198 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
200 dequeue_send(file, packet);
201 ib_destroy_ah(packet->msg->ah);
202 ib_free_send_mad(packet->msg);
204 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
205 packet->length = IB_MGMT_MAD_HDR;
206 packet->mad.hdr.status = ETIMEDOUT;
207 if (!queue_packet(file, agent, packet))
213 static void recv_handler(struct ib_mad_agent *agent,
214 struct ib_mad_recv_wc *mad_recv_wc)
216 struct ib_umad_file *file = agent->context;
217 struct ib_umad_packet *packet;
219 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
222 packet = kzalloc(sizeof *packet, GFP_KERNEL);
226 packet->length = mad_recv_wc->mad_len;
227 packet->recv_wc = mad_recv_wc;
229 packet->mad.hdr.status = 0;
230 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
231 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
232 packet->mad.hdr.lid = cpu_to_be16(mad_recv_wc->wc->slid);
233 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
234 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
235 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
236 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
237 if (packet->mad.hdr.grh_present) {
238 struct ib_ah_attr ah_attr;
240 ib_init_ah_from_wc(agent->device, agent->port_num,
241 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
244 packet->mad.hdr.gid_index = ah_attr.grh.sgid_index;
245 packet->mad.hdr.hop_limit = ah_attr.grh.hop_limit;
246 packet->mad.hdr.traffic_class = ah_attr.grh.traffic_class;
247 memcpy(packet->mad.hdr.gid, &ah_attr.grh.dgid, 16);
248 packet->mad.hdr.flow_label = cpu_to_be32(ah_attr.grh.flow_label);
251 if (queue_packet(file, agent, packet))
258 ib_free_recv_mad(mad_recv_wc);
261 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
262 struct ib_umad_packet *packet, size_t count)
264 struct ib_mad_recv_buf *recv_buf;
265 int left, seg_payload, offset, max_seg_payload;
267 /* We need enough room to copy the first (or only) MAD segment. */
268 recv_buf = &packet->recv_wc->recv_buf;
269 if ((packet->length <= sizeof (*recv_buf->mad) &&
270 count < hdr_size(file) + packet->length) ||
271 (packet->length > sizeof (*recv_buf->mad) &&
272 count < hdr_size(file) + sizeof (*recv_buf->mad)))
275 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
278 buf += hdr_size(file);
279 seg_payload = min_t(int, packet->length, sizeof (*recv_buf->mad));
280 if (copy_to_user(buf, recv_buf->mad, seg_payload))
283 if (seg_payload < packet->length) {
285 * Multipacket RMPP MAD message. Copy remainder of message.
286 * Note that last segment may have a shorter payload.
288 if (count < hdr_size(file) + packet->length) {
290 * The buffer is too small, return the first RMPP segment,
291 * which includes the RMPP message length.
295 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
296 max_seg_payload = sizeof (struct ib_mad) - offset;
298 for (left = packet->length - seg_payload, buf += seg_payload;
299 left; left -= seg_payload, buf += seg_payload) {
300 recv_buf = container_of(recv_buf->list.next,
301 struct ib_mad_recv_buf, list);
302 seg_payload = min(left, max_seg_payload);
303 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
308 return hdr_size(file) + packet->length;
311 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
312 struct ib_umad_packet *packet, size_t count)
314 ssize_t size = hdr_size(file) + packet->length;
319 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
322 buf += hdr_size(file);
324 if (copy_to_user(buf, packet->mad.data, packet->length))
330 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
331 size_t count, loff_t *pos)
333 struct ib_umad_file *file = filp->private_data;
334 struct ib_umad_packet *packet;
337 if (count < hdr_size(file))
340 mutex_lock(&file->mutex);
342 while (list_empty(&file->recv_list)) {
343 mutex_unlock(&file->mutex);
345 if (filp->f_flags & O_NONBLOCK)
348 if (wait_event_interruptible(file->recv_wait,
349 !list_empty(&file->recv_list)))
352 mutex_lock(&file->mutex);
355 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
356 list_del(&packet->list);
358 mutex_unlock(&file->mutex);
361 ret = copy_recv_mad(file, buf, packet, count);
363 ret = copy_send_mad(file, buf, packet, count);
367 mutex_lock(&file->mutex);
368 list_add(&packet->list, &file->recv_list);
369 mutex_unlock(&file->mutex);
372 ib_free_recv_mad(packet->recv_wc);
378 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
382 /* Copy class specific header */
383 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
384 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
385 msg->hdr_len - IB_MGMT_RMPP_HDR))
388 /* All headers are in place. Copy data segments. */
389 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
390 seg++, left -= msg->seg_size, buf += msg->seg_size) {
391 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
392 min(left, msg->seg_size)))
398 static int same_destination(struct ib_user_mad_hdr *hdr1,
399 struct ib_user_mad_hdr *hdr2)
401 if (!hdr1->grh_present && !hdr2->grh_present)
402 return (hdr1->lid == hdr2->lid);
404 if (hdr1->grh_present && hdr2->grh_present)
405 return !memcmp(hdr1->gid, hdr2->gid, 16);
410 static int is_duplicate(struct ib_umad_file *file,
411 struct ib_umad_packet *packet)
413 struct ib_umad_packet *sent_packet;
414 struct ib_mad_hdr *sent_hdr, *hdr;
416 hdr = (struct ib_mad_hdr *) packet->mad.data;
417 list_for_each_entry(sent_packet, &file->send_list, list) {
418 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
420 if ((hdr->tid != sent_hdr->tid) ||
421 (hdr->mgmt_class != sent_hdr->mgmt_class))
425 * No need to be overly clever here. If two new operations have
426 * the same TID, reject the second as a duplicate. This is more
427 * restrictive than required by the spec.
429 if (!ib_response_mad((struct ib_mad *) hdr)) {
430 if (!ib_response_mad((struct ib_mad *) sent_hdr))
433 } else if (!ib_response_mad((struct ib_mad *) sent_hdr))
436 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
443 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
444 size_t count, loff_t *pos)
446 struct ib_umad_file *file = filp->private_data;
447 struct ib_umad_packet *packet;
448 struct ib_mad_agent *agent;
449 struct ib_ah_attr ah_attr;
451 struct ib_rmpp_mad *rmpp_mad;
453 int ret, data_len, hdr_len, copy_offset, rmpp_active;
455 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
458 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
462 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
467 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
472 buf += hdr_size(file);
474 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
479 mutex_lock(&file->mutex);
481 agent = __get_agent(file, packet->mad.hdr.id);
487 memset(&ah_attr, 0, sizeof ah_attr);
488 ah_attr.dlid = be16_to_cpu(packet->mad.hdr.lid);
489 ah_attr.sl = packet->mad.hdr.sl;
490 ah_attr.src_path_bits = packet->mad.hdr.path_bits;
491 ah_attr.port_num = file->port->port_num;
492 if (packet->mad.hdr.grh_present) {
493 ah_attr.ah_flags = IB_AH_GRH;
494 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
495 ah_attr.grh.sgid_index = packet->mad.hdr.gid_index;
496 ah_attr.grh.flow_label = be32_to_cpu(packet->mad.hdr.flow_label);
497 ah_attr.grh.hop_limit = packet->mad.hdr.hop_limit;
498 ah_attr.grh.traffic_class = packet->mad.hdr.traffic_class;
501 ah = ib_create_ah(agent->qp->pd, &ah_attr);
507 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
508 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
509 if (!ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)) {
510 copy_offset = IB_MGMT_MAD_HDR;
513 copy_offset = IB_MGMT_RMPP_HDR;
514 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
515 IB_MGMT_RMPP_FLAG_ACTIVE;
518 data_len = count - hdr_size(file) - hdr_len;
519 packet->msg = ib_create_send_mad(agent,
520 be32_to_cpu(packet->mad.hdr.qpn),
521 packet->mad.hdr.pkey_index, rmpp_active,
522 hdr_len, data_len, GFP_KERNEL);
523 if (IS_ERR(packet->msg)) {
524 ret = PTR_ERR(packet->msg);
528 packet->msg->ah = ah;
529 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
530 packet->msg->retries = packet->mad.hdr.retries;
531 packet->msg->context[0] = packet;
533 /* Copy MAD header. Any RMPP header is already in place. */
534 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
537 if (copy_from_user(packet->msg->mad + copy_offset,
539 hdr_len + data_len - copy_offset)) {
544 ret = copy_rmpp_mad(packet->msg, buf);
550 * Set the high-order part of the transaction ID to make MADs from
551 * different agents unique, and allow routing responses back to the
552 * original requestor.
554 if (!ib_response_mad(packet->msg->mad)) {
555 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
556 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
557 (be64_to_cpup(tid) & 0xffffffff));
558 rmpp_mad->mad_hdr.tid = *tid;
561 spin_lock_irq(&file->send_lock);
562 ret = is_duplicate(file, packet);
564 list_add_tail(&packet->list, &file->send_list);
565 spin_unlock_irq(&file->send_lock);
571 ret = ib_post_send_mad(packet->msg, NULL);
575 mutex_unlock(&file->mutex);
579 dequeue_send(file, packet);
581 ib_free_send_mad(packet->msg);
585 mutex_unlock(&file->mutex);
591 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
593 struct ib_umad_file *file = filp->private_data;
595 /* we will always be able to post a MAD send */
596 unsigned int mask = POLLOUT | POLLWRNORM;
598 poll_wait(filp, &file->recv_wait, wait);
600 if (!list_empty(&file->recv_list))
601 mask |= POLLIN | POLLRDNORM;
606 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
607 int compat_method_mask)
609 struct ib_user_mad_reg_req ureq;
610 struct ib_mad_reg_req req;
611 struct ib_mad_agent *agent = NULL;
615 mutex_lock(&file->port->file_mutex);
616 mutex_lock(&file->mutex);
618 if (!file->port->ib_dev) {
619 dev_notice(file->port->dev,
620 "ib_umad_reg_agent: invalid device\n");
625 if (copy_from_user(&ureq, arg, sizeof ureq)) {
630 if (ureq.qpn != 0 && ureq.qpn != 1) {
631 dev_notice(file->port->dev,
632 "ib_umad_reg_agent: invalid QPN %d specified\n",
638 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
639 if (!__get_agent(file, agent_id))
642 dev_notice(file->port->dev,
643 "ib_umad_reg_agent: Max Agents (%u) reached\n",
649 if (ureq.mgmt_class) {
650 memset(&req, 0, sizeof(req));
651 req.mgmt_class = ureq.mgmt_class;
652 req.mgmt_class_version = ureq.mgmt_class_version;
653 memcpy(req.oui, ureq.oui, sizeof req.oui);
655 if (compat_method_mask) {
656 u32 *umm = (u32 *) ureq.method_mask;
659 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
661 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
663 memcpy(req.method_mask, ureq.method_mask,
664 sizeof req.method_mask);
667 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
668 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
669 ureq.mgmt_class ? &req : NULL,
671 send_handler, recv_handler, file, 0);
673 ret = PTR_ERR(agent);
678 if (put_user(agent_id,
679 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
684 if (!file->already_used) {
685 file->already_used = 1;
686 if (!file->use_pkey_index) {
687 dev_warn(file->port->dev,
688 "process %s did not enable P_Key index support.\n",
690 dev_warn(file->port->dev,
691 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
695 file->agent[agent_id] = agent;
699 mutex_unlock(&file->mutex);
702 ib_unregister_mad_agent(agent);
704 mutex_unlock(&file->port->file_mutex);
709 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
711 struct ib_user_mad_reg_req2 ureq;
712 struct ib_mad_reg_req req;
713 struct ib_mad_agent *agent = NULL;
717 mutex_lock(&file->port->file_mutex);
718 mutex_lock(&file->mutex);
720 if (!file->port->ib_dev) {
721 dev_notice(file->port->dev,
722 "ib_umad_reg_agent2: invalid device\n");
727 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
732 if (ureq.qpn != 0 && ureq.qpn != 1) {
733 dev_notice(file->port->dev,
734 "ib_umad_reg_agent2: invalid QPN %d specified\n",
740 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
741 dev_notice(file->port->dev,
742 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
743 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
746 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
747 (u32 __user *) (arg + offsetof(struct
748 ib_user_mad_reg_req2, flags))))
754 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
755 if (!__get_agent(file, agent_id))
758 dev_notice(file->port->dev,
759 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
765 if (ureq.mgmt_class) {
766 memset(&req, 0, sizeof(req));
767 req.mgmt_class = ureq.mgmt_class;
768 req.mgmt_class_version = ureq.mgmt_class_version;
769 if (ureq.oui & 0xff000000) {
770 dev_notice(file->port->dev,
771 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
776 req.oui[2] = ureq.oui & 0x0000ff;
777 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
778 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
779 memcpy(req.method_mask, ureq.method_mask,
780 sizeof(req.method_mask));
783 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
784 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
785 ureq.mgmt_class ? &req : NULL,
787 send_handler, recv_handler, file,
790 ret = PTR_ERR(agent);
795 if (put_user(agent_id,
797 offsetof(struct ib_user_mad_reg_req2, id)))) {
802 if (!file->already_used) {
803 file->already_used = 1;
804 file->use_pkey_index = 1;
807 file->agent[agent_id] = agent;
811 mutex_unlock(&file->mutex);
814 ib_unregister_mad_agent(agent);
816 mutex_unlock(&file->port->file_mutex);
822 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
824 struct ib_mad_agent *agent = NULL;
828 if (get_user(id, arg))
831 mutex_lock(&file->port->file_mutex);
832 mutex_lock(&file->mutex);
834 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
839 agent = file->agent[id];
840 file->agent[id] = NULL;
843 mutex_unlock(&file->mutex);
846 ib_unregister_mad_agent(agent);
848 mutex_unlock(&file->port->file_mutex);
853 static long ib_umad_enable_pkey(struct ib_umad_file *file)
857 mutex_lock(&file->mutex);
858 if (file->already_used)
861 file->use_pkey_index = 1;
862 mutex_unlock(&file->mutex);
867 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
871 case IB_USER_MAD_REGISTER_AGENT:
872 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
873 case IB_USER_MAD_UNREGISTER_AGENT:
874 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
875 case IB_USER_MAD_ENABLE_PKEY:
876 return ib_umad_enable_pkey(filp->private_data);
877 case IB_USER_MAD_REGISTER_AGENT2:
878 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
885 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
889 case IB_USER_MAD_REGISTER_AGENT:
890 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
891 case IB_USER_MAD_UNREGISTER_AGENT:
892 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
893 case IB_USER_MAD_ENABLE_PKEY:
894 return ib_umad_enable_pkey(filp->private_data);
895 case IB_USER_MAD_REGISTER_AGENT2:
896 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
904 * ib_umad_open() does not need the BKL:
906 * - the ib_umad_port structures are properly reference counted, and
907 * everything else is purely local to the file being created, so
908 * races against other open calls are not a problem;
909 * - the ioctl method does not affect any global state outside of the
910 * file structure being operated on;
912 static int ib_umad_open(struct inode *inode, struct file *filp)
914 struct ib_umad_port *port;
915 struct ib_umad_file *file;
918 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
920 mutex_lock(&port->file_mutex);
926 file = kzalloc(sizeof *file, GFP_KERNEL);
930 mutex_init(&file->mutex);
931 spin_lock_init(&file->send_lock);
932 INIT_LIST_HEAD(&file->recv_list);
933 INIT_LIST_HEAD(&file->send_list);
934 init_waitqueue_head(&file->recv_wait);
937 filp->private_data = file;
939 list_add_tail(&file->port_list, &port->file_list);
941 ret = nonseekable_open(inode, filp);
943 list_del(&file->port_list);
948 kobject_get(&port->umad_dev->kobj);
951 mutex_unlock(&port->file_mutex);
955 static int ib_umad_close(struct inode *inode, struct file *filp)
957 struct ib_umad_file *file = filp->private_data;
958 struct ib_umad_device *dev = file->port->umad_dev;
959 struct ib_umad_packet *packet, *tmp;
963 mutex_lock(&file->port->file_mutex);
964 mutex_lock(&file->mutex);
966 already_dead = file->agents_dead;
967 file->agents_dead = 1;
969 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
971 ib_free_recv_mad(packet->recv_wc);
975 list_del(&file->port_list);
977 mutex_unlock(&file->mutex);
980 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
982 ib_unregister_mad_agent(file->agent[i]);
984 mutex_unlock(&file->port->file_mutex);
987 kobject_put(&dev->kobj);
992 static const struct file_operations umad_fops = {
993 .owner = THIS_MODULE,
994 .read = ib_umad_read,
995 .write = ib_umad_write,
996 .poll = ib_umad_poll,
997 .unlocked_ioctl = ib_umad_ioctl,
999 .compat_ioctl = ib_umad_compat_ioctl,
1001 .open = ib_umad_open,
1002 .release = ib_umad_close,
1003 .llseek = no_llseek,
1006 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1008 struct ib_umad_port *port;
1009 struct ib_port_modify props = {
1010 .set_port_cap_mask = IB_PORT_SM
1014 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1016 if (filp->f_flags & O_NONBLOCK) {
1017 if (down_trylock(&port->sm_sem)) {
1022 if (down_interruptible(&port->sm_sem)) {
1028 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1032 filp->private_data = port;
1034 ret = nonseekable_open(inode, filp);
1036 goto err_clr_sm_cap;
1038 kobject_get(&port->umad_dev->kobj);
1043 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1044 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1053 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1055 struct ib_umad_port *port = filp->private_data;
1056 struct ib_port_modify props = {
1057 .clr_port_cap_mask = IB_PORT_SM
1061 mutex_lock(&port->file_mutex);
1063 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1064 mutex_unlock(&port->file_mutex);
1068 kobject_put(&port->umad_dev->kobj);
1073 static const struct file_operations umad_sm_fops = {
1074 .owner = THIS_MODULE,
1075 .open = ib_umad_sm_open,
1076 .release = ib_umad_sm_close,
1077 .llseek = no_llseek,
1080 static struct ib_client umad_client = {
1082 .add = ib_umad_add_one,
1083 .remove = ib_umad_remove_one
1086 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1089 struct ib_umad_port *port = dev_get_drvdata(dev);
1094 return sprintf(buf, "%s\n", port->ib_dev->name);
1096 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1098 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1101 struct ib_umad_port *port = dev_get_drvdata(dev);
1106 return sprintf(buf, "%d\n", port->port_num);
1108 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1110 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1111 __stringify(IB_USER_MAD_ABI_VERSION));
1113 static dev_t overflow_maj;
1114 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1115 static int find_overflow_devnum(struct ib_device *device)
1119 if (!overflow_maj) {
1120 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1123 dev_err(&device->dev,
1124 "couldn't register dynamic device number\n");
1129 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1130 if (ret >= IB_UMAD_MAX_PORTS)
1136 static int ib_umad_init_port(struct ib_device *device, int port_num,
1137 struct ib_umad_device *umad_dev,
1138 struct ib_umad_port *port)
1143 spin_lock(&port_lock);
1144 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1145 if (devnum >= IB_UMAD_MAX_PORTS) {
1146 spin_unlock(&port_lock);
1147 devnum = find_overflow_devnum(device);
1151 spin_lock(&port_lock);
1152 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1153 base = devnum + overflow_maj;
1154 set_bit(devnum, overflow_map);
1156 port->dev_num = devnum;
1157 base = devnum + base_dev;
1158 set_bit(devnum, dev_map);
1160 spin_unlock(&port_lock);
1162 port->ib_dev = device;
1163 port->port_num = port_num;
1164 sema_init(&port->sm_sem, 1);
1165 mutex_init(&port->file_mutex);
1166 INIT_LIST_HEAD(&port->file_list);
1168 cdev_init(&port->cdev, &umad_fops);
1169 port->cdev.owner = THIS_MODULE;
1170 port->cdev.kobj.parent = &umad_dev->kobj;
1171 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1172 if (cdev_add(&port->cdev, base, 1))
1175 port->dev = device_create(umad_class, device->dma_device,
1176 port->cdev.dev, port,
1177 "umad%d", port->dev_num);
1178 if (IS_ERR(port->dev))
1181 if (device_create_file(port->dev, &dev_attr_ibdev))
1183 if (device_create_file(port->dev, &dev_attr_port))
1186 base += IB_UMAD_MAX_PORTS;
1187 cdev_init(&port->sm_cdev, &umad_sm_fops);
1188 port->sm_cdev.owner = THIS_MODULE;
1189 port->sm_cdev.kobj.parent = &umad_dev->kobj;
1190 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1191 if (cdev_add(&port->sm_cdev, base, 1))
1194 port->sm_dev = device_create(umad_class, device->dma_device,
1195 port->sm_cdev.dev, port,
1196 "issm%d", port->dev_num);
1197 if (IS_ERR(port->sm_dev))
1200 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1202 if (device_create_file(port->sm_dev, &dev_attr_port))
1208 device_destroy(umad_class, port->sm_cdev.dev);
1211 cdev_del(&port->sm_cdev);
1214 device_destroy(umad_class, port->cdev.dev);
1217 cdev_del(&port->cdev);
1218 if (port->dev_num < IB_UMAD_MAX_PORTS)
1219 clear_bit(devnum, dev_map);
1221 clear_bit(devnum, overflow_map);
1226 static void ib_umad_kill_port(struct ib_umad_port *port)
1228 struct ib_umad_file *file;
1231 dev_set_drvdata(port->dev, NULL);
1232 dev_set_drvdata(port->sm_dev, NULL);
1234 device_destroy(umad_class, port->cdev.dev);
1235 device_destroy(umad_class, port->sm_cdev.dev);
1237 cdev_del(&port->cdev);
1238 cdev_del(&port->sm_cdev);
1240 mutex_lock(&port->file_mutex);
1242 port->ib_dev = NULL;
1244 list_for_each_entry(file, &port->file_list, port_list) {
1245 mutex_lock(&file->mutex);
1246 file->agents_dead = 1;
1247 mutex_unlock(&file->mutex);
1249 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1250 if (file->agent[id])
1251 ib_unregister_mad_agent(file->agent[id]);
1254 mutex_unlock(&port->file_mutex);
1256 if (port->dev_num < IB_UMAD_MAX_PORTS)
1257 clear_bit(port->dev_num, dev_map);
1259 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1262 static void ib_umad_add_one(struct ib_device *device)
1264 struct ib_umad_device *umad_dev;
1267 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1270 if (device->node_type == RDMA_NODE_IB_SWITCH)
1274 e = device->phys_port_cnt;
1277 umad_dev = kzalloc(sizeof *umad_dev +
1278 (e - s + 1) * sizeof (struct ib_umad_port),
1283 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1285 umad_dev->start_port = s;
1286 umad_dev->end_port = e;
1288 for (i = s; i <= e; ++i) {
1289 umad_dev->port[i - s].umad_dev = umad_dev;
1291 if (ib_umad_init_port(device, i, umad_dev,
1292 &umad_dev->port[i - s]))
1296 ib_set_client_data(device, &umad_client, umad_dev);
1302 ib_umad_kill_port(&umad_dev->port[i - s]);
1304 kobject_put(&umad_dev->kobj);
1307 static void ib_umad_remove_one(struct ib_device *device)
1309 struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
1315 for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i)
1316 ib_umad_kill_port(&umad_dev->port[i]);
1318 kobject_put(&umad_dev->kobj);
1321 static char *umad_devnode(struct device *dev, umode_t *mode)
1323 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1326 static int __init ib_umad_init(void)
1330 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1333 pr_err("couldn't register device number\n");
1337 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1338 if (IS_ERR(umad_class)) {
1339 ret = PTR_ERR(umad_class);
1340 pr_err("couldn't create class infiniband_mad\n");
1344 umad_class->devnode = umad_devnode;
1346 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1348 pr_err("couldn't create abi_version attribute\n");
1352 ret = ib_register_client(&umad_client);
1354 pr_err("couldn't register ib_umad client\n");
1361 class_destroy(umad_class);
1364 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1370 static void __exit ib_umad_cleanup(void)
1372 ib_unregister_client(&umad_client);
1373 class_destroy(umad_class);
1374 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1376 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1379 module_init(ib_umad_init);
1380 module_exit(ib_umad_cleanup);