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 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 rdma_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_send_buf *send_buf,
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);
233 * On OPA devices it is okay to lose the upper 16 bits of LID as this
234 * information is obtained elsewhere. Mask off the upper 16 bits.
236 if (agent->device->port_immutable[agent->port_num].core_cap_flags &
237 RDMA_CORE_PORT_INTEL_OPA)
238 packet->mad.hdr.lid = ib_lid_be16(0xFFFF &
239 mad_recv_wc->wc->slid);
241 packet->mad.hdr.lid = ib_lid_be16(mad_recv_wc->wc->slid);
242 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
243 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
244 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
245 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
246 if (packet->mad.hdr.grh_present) {
247 struct rdma_ah_attr ah_attr;
248 const struct ib_global_route *grh;
250 ib_init_ah_from_wc(agent->device, agent->port_num,
251 mad_recv_wc->wc, mad_recv_wc->recv_buf.grh,
254 grh = rdma_ah_read_grh(&ah_attr);
255 packet->mad.hdr.gid_index = grh->sgid_index;
256 packet->mad.hdr.hop_limit = grh->hop_limit;
257 packet->mad.hdr.traffic_class = grh->traffic_class;
258 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
259 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
262 if (queue_packet(file, agent, packet))
269 ib_free_recv_mad(mad_recv_wc);
272 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
273 struct ib_umad_packet *packet, size_t count)
275 struct ib_mad_recv_buf *recv_buf;
276 int left, seg_payload, offset, max_seg_payload;
279 recv_buf = &packet->recv_wc->recv_buf;
280 seg_size = packet->recv_wc->mad_seg_size;
282 /* We need enough room to copy the first (or only) MAD segment. */
283 if ((packet->length <= seg_size &&
284 count < hdr_size(file) + packet->length) ||
285 (packet->length > seg_size &&
286 count < hdr_size(file) + seg_size))
289 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
292 buf += hdr_size(file);
293 seg_payload = min_t(int, packet->length, seg_size);
294 if (copy_to_user(buf, recv_buf->mad, seg_payload))
297 if (seg_payload < packet->length) {
299 * Multipacket RMPP MAD message. Copy remainder of message.
300 * Note that last segment may have a shorter payload.
302 if (count < hdr_size(file) + packet->length) {
304 * The buffer is too small, return the first RMPP segment,
305 * which includes the RMPP message length.
309 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
310 max_seg_payload = seg_size - offset;
312 for (left = packet->length - seg_payload, buf += seg_payload;
313 left; left -= seg_payload, buf += seg_payload) {
314 recv_buf = container_of(recv_buf->list.next,
315 struct ib_mad_recv_buf, list);
316 seg_payload = min(left, max_seg_payload);
317 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
322 return hdr_size(file) + packet->length;
325 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
326 struct ib_umad_packet *packet, size_t count)
328 ssize_t size = hdr_size(file) + packet->length;
333 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
336 buf += hdr_size(file);
338 if (copy_to_user(buf, packet->mad.data, packet->length))
344 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
345 size_t count, loff_t *pos)
347 struct ib_umad_file *file = filp->private_data;
348 struct ib_umad_packet *packet;
351 if (count < hdr_size(file))
354 mutex_lock(&file->mutex);
356 while (list_empty(&file->recv_list)) {
357 mutex_unlock(&file->mutex);
359 if (filp->f_flags & O_NONBLOCK)
362 if (wait_event_interruptible(file->recv_wait,
363 !list_empty(&file->recv_list)))
366 mutex_lock(&file->mutex);
369 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
370 list_del(&packet->list);
372 mutex_unlock(&file->mutex);
375 ret = copy_recv_mad(file, buf, packet, count);
377 ret = copy_send_mad(file, buf, packet, count);
381 mutex_lock(&file->mutex);
382 list_add(&packet->list, &file->recv_list);
383 mutex_unlock(&file->mutex);
386 ib_free_recv_mad(packet->recv_wc);
392 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
396 /* Copy class specific header */
397 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
398 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
399 msg->hdr_len - IB_MGMT_RMPP_HDR))
402 /* All headers are in place. Copy data segments. */
403 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
404 seg++, left -= msg->seg_size, buf += msg->seg_size) {
405 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
406 min(left, msg->seg_size)))
412 static int same_destination(struct ib_user_mad_hdr *hdr1,
413 struct ib_user_mad_hdr *hdr2)
415 if (!hdr1->grh_present && !hdr2->grh_present)
416 return (hdr1->lid == hdr2->lid);
418 if (hdr1->grh_present && hdr2->grh_present)
419 return !memcmp(hdr1->gid, hdr2->gid, 16);
424 static int is_duplicate(struct ib_umad_file *file,
425 struct ib_umad_packet *packet)
427 struct ib_umad_packet *sent_packet;
428 struct ib_mad_hdr *sent_hdr, *hdr;
430 hdr = (struct ib_mad_hdr *) packet->mad.data;
431 list_for_each_entry(sent_packet, &file->send_list, list) {
432 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
434 if ((hdr->tid != sent_hdr->tid) ||
435 (hdr->mgmt_class != sent_hdr->mgmt_class))
439 * No need to be overly clever here. If two new operations have
440 * the same TID, reject the second as a duplicate. This is more
441 * restrictive than required by the spec.
443 if (!ib_response_mad(hdr)) {
444 if (!ib_response_mad(sent_hdr))
447 } else if (!ib_response_mad(sent_hdr))
450 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
457 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
458 size_t count, loff_t *pos)
460 struct ib_umad_file *file = filp->private_data;
461 struct ib_umad_packet *packet;
462 struct ib_mad_agent *agent;
463 struct rdma_ah_attr ah_attr;
465 struct ib_rmpp_mad *rmpp_mad;
467 int ret, data_len, hdr_len, copy_offset, rmpp_active;
470 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
473 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
477 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
482 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
487 buf += hdr_size(file);
489 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
494 mutex_lock(&file->mutex);
496 agent = __get_agent(file, packet->mad.hdr.id);
502 memset(&ah_attr, 0, sizeof ah_attr);
503 ah_attr.type = rdma_ah_find_type(file->port->ib_dev,
504 file->port->port_num);
505 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
506 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
507 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
508 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
509 if (packet->mad.hdr.grh_present) {
510 rdma_ah_set_grh(&ah_attr, NULL,
511 be32_to_cpu(packet->mad.hdr.flow_label),
512 packet->mad.hdr.gid_index,
513 packet->mad.hdr.hop_limit,
514 packet->mad.hdr.traffic_class);
515 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
518 ah = rdma_create_user_ah(agent->qp->pd, &ah_attr, NULL);
524 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
525 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
527 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
528 && ib_mad_kernel_rmpp_agent(agent)) {
529 copy_offset = IB_MGMT_RMPP_HDR;
530 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
531 IB_MGMT_RMPP_FLAG_ACTIVE;
533 copy_offset = IB_MGMT_MAD_HDR;
537 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
538 data_len = count - hdr_size(file) - hdr_len;
539 packet->msg = ib_create_send_mad(agent,
540 be32_to_cpu(packet->mad.hdr.qpn),
541 packet->mad.hdr.pkey_index, rmpp_active,
542 hdr_len, data_len, GFP_KERNEL,
544 if (IS_ERR(packet->msg)) {
545 ret = PTR_ERR(packet->msg);
549 packet->msg->ah = ah;
550 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
551 packet->msg->retries = packet->mad.hdr.retries;
552 packet->msg->context[0] = packet;
554 /* Copy MAD header. Any RMPP header is already in place. */
555 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
558 if (copy_from_user(packet->msg->mad + copy_offset,
560 hdr_len + data_len - copy_offset)) {
565 ret = copy_rmpp_mad(packet->msg, buf);
571 * Set the high-order part of the transaction ID to make MADs from
572 * different agents unique, and allow routing responses back to the
573 * original requestor.
575 if (!ib_response_mad(packet->msg->mad)) {
576 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
577 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
578 (be64_to_cpup(tid) & 0xffffffff));
579 rmpp_mad->mad_hdr.tid = *tid;
582 if (!ib_mad_kernel_rmpp_agent(agent)
583 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
584 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
585 spin_lock_irq(&file->send_lock);
586 list_add_tail(&packet->list, &file->send_list);
587 spin_unlock_irq(&file->send_lock);
589 spin_lock_irq(&file->send_lock);
590 ret = is_duplicate(file, packet);
592 list_add_tail(&packet->list, &file->send_list);
593 spin_unlock_irq(&file->send_lock);
600 ret = ib_post_send_mad(packet->msg, NULL);
604 mutex_unlock(&file->mutex);
608 dequeue_send(file, packet);
610 ib_free_send_mad(packet->msg);
614 mutex_unlock(&file->mutex);
620 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
622 struct ib_umad_file *file = filp->private_data;
624 /* we will always be able to post a MAD send */
625 unsigned int mask = POLLOUT | POLLWRNORM;
627 poll_wait(filp, &file->recv_wait, wait);
629 if (!list_empty(&file->recv_list))
630 mask |= POLLIN | POLLRDNORM;
635 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
636 int compat_method_mask)
638 struct ib_user_mad_reg_req ureq;
639 struct ib_mad_reg_req req;
640 struct ib_mad_agent *agent = NULL;
644 mutex_lock(&file->port->file_mutex);
645 mutex_lock(&file->mutex);
647 if (!file->port->ib_dev) {
648 dev_notice(file->port->dev,
649 "ib_umad_reg_agent: invalid device\n");
654 if (copy_from_user(&ureq, arg, sizeof ureq)) {
659 if (ureq.qpn != 0 && ureq.qpn != 1) {
660 dev_notice(file->port->dev,
661 "ib_umad_reg_agent: invalid QPN %d specified\n",
667 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
668 if (!__get_agent(file, agent_id))
671 dev_notice(file->port->dev,
672 "ib_umad_reg_agent: Max Agents (%u) reached\n",
678 if (ureq.mgmt_class) {
679 memset(&req, 0, sizeof(req));
680 req.mgmt_class = ureq.mgmt_class;
681 req.mgmt_class_version = ureq.mgmt_class_version;
682 memcpy(req.oui, ureq.oui, sizeof req.oui);
684 if (compat_method_mask) {
685 u32 *umm = (u32 *) ureq.method_mask;
688 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
690 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
692 memcpy(req.method_mask, ureq.method_mask,
693 sizeof req.method_mask);
696 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
697 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
698 ureq.mgmt_class ? &req : NULL,
700 send_handler, recv_handler, file, 0);
702 ret = PTR_ERR(agent);
707 if (put_user(agent_id,
708 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
713 if (!file->already_used) {
714 file->already_used = 1;
715 if (!file->use_pkey_index) {
716 dev_warn(file->port->dev,
717 "process %s did not enable P_Key index support.\n",
719 dev_warn(file->port->dev,
720 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
724 file->agent[agent_id] = agent;
728 mutex_unlock(&file->mutex);
731 ib_unregister_mad_agent(agent);
733 mutex_unlock(&file->port->file_mutex);
738 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
740 struct ib_user_mad_reg_req2 ureq;
741 struct ib_mad_reg_req req;
742 struct ib_mad_agent *agent = NULL;
746 mutex_lock(&file->port->file_mutex);
747 mutex_lock(&file->mutex);
749 if (!file->port->ib_dev) {
750 dev_notice(file->port->dev,
751 "ib_umad_reg_agent2: invalid device\n");
756 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
761 if (ureq.qpn != 0 && ureq.qpn != 1) {
762 dev_notice(file->port->dev,
763 "ib_umad_reg_agent2: invalid QPN %d specified\n",
769 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
770 dev_notice(file->port->dev,
771 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
772 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
775 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
776 (u32 __user *) (arg + offsetof(struct
777 ib_user_mad_reg_req2, flags))))
783 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
784 if (!__get_agent(file, agent_id))
787 dev_notice(file->port->dev,
788 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
794 if (ureq.mgmt_class) {
795 memset(&req, 0, sizeof(req));
796 req.mgmt_class = ureq.mgmt_class;
797 req.mgmt_class_version = ureq.mgmt_class_version;
798 if (ureq.oui & 0xff000000) {
799 dev_notice(file->port->dev,
800 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
805 req.oui[2] = ureq.oui & 0x0000ff;
806 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
807 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
808 memcpy(req.method_mask, ureq.method_mask,
809 sizeof(req.method_mask));
812 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
813 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
814 ureq.mgmt_class ? &req : NULL,
816 send_handler, recv_handler, file,
819 ret = PTR_ERR(agent);
824 if (put_user(agent_id,
826 offsetof(struct ib_user_mad_reg_req2, id)))) {
831 if (!file->already_used) {
832 file->already_used = 1;
833 file->use_pkey_index = 1;
836 file->agent[agent_id] = agent;
840 mutex_unlock(&file->mutex);
843 ib_unregister_mad_agent(agent);
845 mutex_unlock(&file->port->file_mutex);
851 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
853 struct ib_mad_agent *agent = NULL;
857 if (get_user(id, arg))
860 mutex_lock(&file->port->file_mutex);
861 mutex_lock(&file->mutex);
863 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
868 agent = file->agent[id];
869 file->agent[id] = NULL;
872 mutex_unlock(&file->mutex);
875 ib_unregister_mad_agent(agent);
877 mutex_unlock(&file->port->file_mutex);
882 static long ib_umad_enable_pkey(struct ib_umad_file *file)
886 mutex_lock(&file->mutex);
887 if (file->already_used)
890 file->use_pkey_index = 1;
891 mutex_unlock(&file->mutex);
896 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
900 case IB_USER_MAD_REGISTER_AGENT:
901 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
902 case IB_USER_MAD_UNREGISTER_AGENT:
903 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
904 case IB_USER_MAD_ENABLE_PKEY:
905 return ib_umad_enable_pkey(filp->private_data);
906 case IB_USER_MAD_REGISTER_AGENT2:
907 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
914 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
918 case IB_USER_MAD_REGISTER_AGENT:
919 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
920 case IB_USER_MAD_UNREGISTER_AGENT:
921 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
922 case IB_USER_MAD_ENABLE_PKEY:
923 return ib_umad_enable_pkey(filp->private_data);
924 case IB_USER_MAD_REGISTER_AGENT2:
925 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
933 * ib_umad_open() does not need the BKL:
935 * - the ib_umad_port structures are properly reference counted, and
936 * everything else is purely local to the file being created, so
937 * races against other open calls are not a problem;
938 * - the ioctl method does not affect any global state outside of the
939 * file structure being operated on;
941 static int ib_umad_open(struct inode *inode, struct file *filp)
943 struct ib_umad_port *port;
944 struct ib_umad_file *file;
947 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
949 mutex_lock(&port->file_mutex);
955 file = kzalloc(sizeof *file, GFP_KERNEL);
959 mutex_init(&file->mutex);
960 spin_lock_init(&file->send_lock);
961 INIT_LIST_HEAD(&file->recv_list);
962 INIT_LIST_HEAD(&file->send_list);
963 init_waitqueue_head(&file->recv_wait);
966 filp->private_data = file;
968 list_add_tail(&file->port_list, &port->file_list);
970 ret = nonseekable_open(inode, filp);
972 list_del(&file->port_list);
977 kobject_get(&port->umad_dev->kobj);
980 mutex_unlock(&port->file_mutex);
984 static int ib_umad_close(struct inode *inode, struct file *filp)
986 struct ib_umad_file *file = filp->private_data;
987 struct ib_umad_device *dev = file->port->umad_dev;
988 struct ib_umad_packet *packet, *tmp;
992 mutex_lock(&file->port->file_mutex);
993 mutex_lock(&file->mutex);
995 already_dead = file->agents_dead;
996 file->agents_dead = 1;
998 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
1000 ib_free_recv_mad(packet->recv_wc);
1004 list_del(&file->port_list);
1006 mutex_unlock(&file->mutex);
1009 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1011 ib_unregister_mad_agent(file->agent[i]);
1013 mutex_unlock(&file->port->file_mutex);
1016 kobject_put(&dev->kobj);
1021 static const struct file_operations umad_fops = {
1022 .owner = THIS_MODULE,
1023 .read = ib_umad_read,
1024 .write = ib_umad_write,
1025 .poll = ib_umad_poll,
1026 .unlocked_ioctl = ib_umad_ioctl,
1027 #ifdef CONFIG_COMPAT
1028 .compat_ioctl = ib_umad_compat_ioctl,
1030 .open = ib_umad_open,
1031 .release = ib_umad_close,
1032 .llseek = no_llseek,
1035 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1037 struct ib_umad_port *port;
1038 struct ib_port_modify props = {
1039 .set_port_cap_mask = IB_PORT_SM
1043 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1045 if (filp->f_flags & O_NONBLOCK) {
1046 if (down_trylock(&port->sm_sem)) {
1051 if (down_interruptible(&port->sm_sem)) {
1057 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1061 filp->private_data = port;
1063 ret = nonseekable_open(inode, filp);
1065 goto err_clr_sm_cap;
1067 kobject_get(&port->umad_dev->kobj);
1072 swap(props.set_port_cap_mask, props.clr_port_cap_mask);
1073 ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1082 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1084 struct ib_umad_port *port = filp->private_data;
1085 struct ib_port_modify props = {
1086 .clr_port_cap_mask = IB_PORT_SM
1090 mutex_lock(&port->file_mutex);
1092 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1093 mutex_unlock(&port->file_mutex);
1097 kobject_put(&port->umad_dev->kobj);
1102 static const struct file_operations umad_sm_fops = {
1103 .owner = THIS_MODULE,
1104 .open = ib_umad_sm_open,
1105 .release = ib_umad_sm_close,
1106 .llseek = no_llseek,
1109 static struct ib_client umad_client = {
1111 .add = ib_umad_add_one,
1112 .remove = ib_umad_remove_one
1115 static ssize_t show_ibdev(struct device *dev, struct device_attribute *attr,
1118 struct ib_umad_port *port = dev_get_drvdata(dev);
1123 return sprintf(buf, "%s\n", port->ib_dev->name);
1125 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
1127 static ssize_t show_port(struct device *dev, struct device_attribute *attr,
1130 struct ib_umad_port *port = dev_get_drvdata(dev);
1135 return sprintf(buf, "%d\n", port->port_num);
1137 static DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
1139 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
1140 __stringify(IB_USER_MAD_ABI_VERSION));
1142 static dev_t overflow_maj;
1143 static DECLARE_BITMAP(overflow_map, IB_UMAD_MAX_PORTS);
1144 static int find_overflow_devnum(struct ib_device *device)
1148 if (!overflow_maj) {
1149 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UMAD_MAX_PORTS * 2,
1152 dev_err(&device->dev,
1153 "couldn't register dynamic device number\n");
1158 ret = find_first_zero_bit(overflow_map, IB_UMAD_MAX_PORTS);
1159 if (ret >= IB_UMAD_MAX_PORTS)
1165 static int ib_umad_init_port(struct ib_device *device, int port_num,
1166 struct ib_umad_device *umad_dev,
1167 struct ib_umad_port *port)
1172 spin_lock(&port_lock);
1173 devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
1174 if (devnum >= IB_UMAD_MAX_PORTS) {
1175 spin_unlock(&port_lock);
1176 devnum = find_overflow_devnum(device);
1180 spin_lock(&port_lock);
1181 port->dev_num = devnum + IB_UMAD_MAX_PORTS;
1182 base = devnum + overflow_maj;
1183 set_bit(devnum, overflow_map);
1185 port->dev_num = devnum;
1186 base = devnum + base_dev;
1187 set_bit(devnum, dev_map);
1189 spin_unlock(&port_lock);
1191 port->ib_dev = device;
1192 port->port_num = port_num;
1193 sema_init(&port->sm_sem, 1);
1194 mutex_init(&port->file_mutex);
1195 INIT_LIST_HEAD(&port->file_list);
1197 cdev_init(&port->cdev, &umad_fops);
1198 port->cdev.owner = THIS_MODULE;
1199 cdev_set_parent(&port->cdev, &umad_dev->kobj);
1200 kobject_set_name(&port->cdev.kobj, "umad%d", port->dev_num);
1201 if (cdev_add(&port->cdev, base, 1))
1204 port->dev = device_create(umad_class, device->dev.parent,
1205 port->cdev.dev, port,
1206 "umad%d", port->dev_num);
1207 if (IS_ERR(port->dev))
1210 if (device_create_file(port->dev, &dev_attr_ibdev))
1212 if (device_create_file(port->dev, &dev_attr_port))
1215 base += IB_UMAD_MAX_PORTS;
1216 cdev_init(&port->sm_cdev, &umad_sm_fops);
1217 port->sm_cdev.owner = THIS_MODULE;
1218 cdev_set_parent(&port->sm_cdev, &umad_dev->kobj);
1219 kobject_set_name(&port->sm_cdev.kobj, "issm%d", port->dev_num);
1220 if (cdev_add(&port->sm_cdev, base, 1))
1223 port->sm_dev = device_create(umad_class, device->dev.parent,
1224 port->sm_cdev.dev, port,
1225 "issm%d", port->dev_num);
1226 if (IS_ERR(port->sm_dev))
1229 if (device_create_file(port->sm_dev, &dev_attr_ibdev))
1231 if (device_create_file(port->sm_dev, &dev_attr_port))
1237 device_destroy(umad_class, port->sm_cdev.dev);
1240 cdev_del(&port->sm_cdev);
1243 device_destroy(umad_class, port->cdev.dev);
1246 cdev_del(&port->cdev);
1247 if (port->dev_num < IB_UMAD_MAX_PORTS)
1248 clear_bit(devnum, dev_map);
1250 clear_bit(devnum, overflow_map);
1255 static void ib_umad_kill_port(struct ib_umad_port *port)
1257 struct ib_umad_file *file;
1260 dev_set_drvdata(port->dev, NULL);
1261 dev_set_drvdata(port->sm_dev, NULL);
1263 device_destroy(umad_class, port->cdev.dev);
1264 device_destroy(umad_class, port->sm_cdev.dev);
1266 cdev_del(&port->cdev);
1267 cdev_del(&port->sm_cdev);
1269 mutex_lock(&port->file_mutex);
1271 port->ib_dev = NULL;
1273 list_for_each_entry(file, &port->file_list, port_list) {
1274 mutex_lock(&file->mutex);
1275 file->agents_dead = 1;
1276 mutex_unlock(&file->mutex);
1278 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1279 if (file->agent[id])
1280 ib_unregister_mad_agent(file->agent[id]);
1283 mutex_unlock(&port->file_mutex);
1285 if (port->dev_num < IB_UMAD_MAX_PORTS)
1286 clear_bit(port->dev_num, dev_map);
1288 clear_bit(port->dev_num - IB_UMAD_MAX_PORTS, overflow_map);
1291 static void ib_umad_add_one(struct ib_device *device)
1293 struct ib_umad_device *umad_dev;
1297 s = rdma_start_port(device);
1298 e = rdma_end_port(device);
1300 umad_dev = kzalloc(sizeof *umad_dev +
1301 (e - s + 1) * sizeof (struct ib_umad_port),
1306 kobject_init(&umad_dev->kobj, &ib_umad_dev_ktype);
1308 for (i = s; i <= e; ++i) {
1309 if (!rdma_cap_ib_mad(device, i))
1312 umad_dev->port[i - s].umad_dev = umad_dev;
1314 if (ib_umad_init_port(device, i, umad_dev,
1315 &umad_dev->port[i - s]))
1324 ib_set_client_data(device, &umad_client, umad_dev);
1330 if (!rdma_cap_ib_mad(device, i))
1333 ib_umad_kill_port(&umad_dev->port[i - s]);
1336 kobject_put(&umad_dev->kobj);
1339 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1341 struct ib_umad_device *umad_dev = client_data;
1347 for (i = 0; i <= rdma_end_port(device) - rdma_start_port(device); ++i) {
1348 if (rdma_cap_ib_mad(device, i + rdma_start_port(device)))
1349 ib_umad_kill_port(&umad_dev->port[i]);
1352 kobject_put(&umad_dev->kobj);
1355 static char *umad_devnode(struct device *dev, umode_t *mode)
1357 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1360 static int __init ib_umad_init(void)
1364 ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
1367 pr_err("couldn't register device number\n");
1371 umad_class = class_create(THIS_MODULE, "infiniband_mad");
1372 if (IS_ERR(umad_class)) {
1373 ret = PTR_ERR(umad_class);
1374 pr_err("couldn't create class infiniband_mad\n");
1378 umad_class->devnode = umad_devnode;
1380 ret = class_create_file(umad_class, &class_attr_abi_version.attr);
1382 pr_err("couldn't create abi_version attribute\n");
1386 ret = ib_register_client(&umad_client);
1388 pr_err("couldn't register ib_umad client\n");
1395 class_destroy(umad_class);
1398 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1404 static void __exit ib_umad_cleanup(void)
1406 ib_unregister_client(&umad_client);
1407 class_destroy(umad_class);
1408 unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
1410 unregister_chrdev_region(overflow_maj, IB_UMAD_MAX_PORTS * 2);
1413 module_init(ib_umad_init);
1414 module_exit(ib_umad_cleanup);