2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
4 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
5 * Copyright (c) 2008 Cisco. All rights reserved.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #define pr_fmt(fmt) "user_mad: " fmt
38 #include <linux/module.h>
39 #include <linux/init.h>
40 #include <linux/device.h>
41 #include <linux/err.h>
43 #include <linux/cdev.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/poll.h>
46 #include <linux/mutex.h>
47 #include <linux/kref.h>
48 #include <linux/compat.h>
49 #include <linux/sched.h>
50 #include <linux/semaphore.h>
51 #include <linux/slab.h>
53 #include <linux/uaccess.h>
55 #include <rdma/ib_mad.h>
56 #include <rdma/ib_user_mad.h>
58 #include "core_priv.h"
60 MODULE_AUTHOR("Roland Dreier");
61 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
62 MODULE_LICENSE("Dual BSD/GPL");
65 IB_UMAD_MAX_PORTS = RDMA_MAX_PORTS,
66 IB_UMAD_MAX_AGENTS = 32,
69 IB_UMAD_MINOR_BASE = 0,
70 IB_UMAD_NUM_FIXED_MINOR = 64,
71 IB_UMAD_NUM_DYNAMIC_MINOR = IB_UMAD_MAX_PORTS - IB_UMAD_NUM_FIXED_MINOR,
72 IB_ISSM_MINOR_BASE = IB_UMAD_NUM_FIXED_MINOR,
76 * Our lifetime rules for these structs are the following:
77 * device special file is opened, we take a reference on the
78 * ib_umad_port's struct ib_umad_device. We drop these
79 * references in the corresponding close().
81 * In addition to references coming from open character devices, there
82 * is one more reference to each ib_umad_device representing the
83 * module's reference taken when allocating the ib_umad_device in
86 * When destroying an ib_umad_device, we drop the module's reference.
94 struct semaphore sm_sem;
96 struct mutex file_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 {
107 struct ib_umad_port ports[];
110 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 send_lock;
117 wait_queue_head_t recv_wait;
118 struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
124 struct ib_umad_packet {
125 struct ib_mad_send_buf *msg;
126 struct ib_mad_recv_wc *recv_wc;
127 struct list_head list;
129 struct ib_user_mad mad;
132 static const dev_t base_umad_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
133 static const dev_t base_issm_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE) +
134 IB_UMAD_NUM_FIXED_MINOR;
135 static dev_t dynamic_umad_dev;
136 static dev_t dynamic_issm_dev;
138 static DEFINE_IDA(umad_ida);
140 static void ib_umad_add_one(struct ib_device *device);
141 static void ib_umad_remove_one(struct ib_device *device, void *client_data);
143 static void ib_umad_dev_free(struct kref *kref)
145 struct ib_umad_device *dev =
146 container_of(kref, struct ib_umad_device, kref);
151 static void ib_umad_dev_get(struct ib_umad_device *dev)
153 kref_get(&dev->kref);
156 static void ib_umad_dev_put(struct ib_umad_device *dev)
158 kref_put(&dev->kref, ib_umad_dev_free);
161 static int hdr_size(struct ib_umad_file *file)
163 return file->use_pkey_index ? sizeof (struct ib_user_mad_hdr) :
164 sizeof (struct ib_user_mad_hdr_old);
167 /* caller must hold file->mutex */
168 static struct ib_mad_agent *__get_agent(struct ib_umad_file *file, int id)
170 return file->agents_dead ? NULL : file->agent[id];
173 static int queue_packet(struct ib_umad_file *file,
174 struct ib_mad_agent *agent,
175 struct ib_umad_packet *packet)
179 mutex_lock(&file->mutex);
181 for (packet->mad.hdr.id = 0;
182 packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
183 packet->mad.hdr.id++)
184 if (agent == __get_agent(file, packet->mad.hdr.id)) {
185 list_add_tail(&packet->list, &file->recv_list);
186 wake_up_interruptible(&file->recv_wait);
191 mutex_unlock(&file->mutex);
196 static void dequeue_send(struct ib_umad_file *file,
197 struct ib_umad_packet *packet)
199 spin_lock_irq(&file->send_lock);
200 list_del(&packet->list);
201 spin_unlock_irq(&file->send_lock);
204 static void send_handler(struct ib_mad_agent *agent,
205 struct ib_mad_send_wc *send_wc)
207 struct ib_umad_file *file = agent->context;
208 struct ib_umad_packet *packet = send_wc->send_buf->context[0];
210 dequeue_send(file, packet);
211 rdma_destroy_ah(packet->msg->ah, RDMA_DESTROY_AH_SLEEPABLE);
212 ib_free_send_mad(packet->msg);
214 if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
215 packet->length = IB_MGMT_MAD_HDR;
216 packet->mad.hdr.status = ETIMEDOUT;
217 if (!queue_packet(file, agent, packet))
223 static void recv_handler(struct ib_mad_agent *agent,
224 struct ib_mad_send_buf *send_buf,
225 struct ib_mad_recv_wc *mad_recv_wc)
227 struct ib_umad_file *file = agent->context;
228 struct ib_umad_packet *packet;
230 if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
233 packet = kzalloc(sizeof *packet, GFP_KERNEL);
237 packet->length = mad_recv_wc->mad_len;
238 packet->recv_wc = mad_recv_wc;
240 packet->mad.hdr.status = 0;
241 packet->mad.hdr.length = hdr_size(file) + mad_recv_wc->mad_len;
242 packet->mad.hdr.qpn = cpu_to_be32(mad_recv_wc->wc->src_qp);
244 * On OPA devices it is okay to lose the upper 16 bits of LID as this
245 * information is obtained elsewhere. Mask off the upper 16 bits.
247 if (rdma_cap_opa_mad(agent->device, agent->port_num))
248 packet->mad.hdr.lid = ib_lid_be16(0xFFFF &
249 mad_recv_wc->wc->slid);
251 packet->mad.hdr.lid = ib_lid_be16(mad_recv_wc->wc->slid);
252 packet->mad.hdr.sl = mad_recv_wc->wc->sl;
253 packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
254 packet->mad.hdr.pkey_index = mad_recv_wc->wc->pkey_index;
255 packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
256 if (packet->mad.hdr.grh_present) {
257 struct rdma_ah_attr ah_attr;
258 const struct ib_global_route *grh;
261 ret = ib_init_ah_attr_from_wc(agent->device, agent->port_num,
263 mad_recv_wc->recv_buf.grh,
268 grh = rdma_ah_read_grh(&ah_attr);
269 packet->mad.hdr.gid_index = grh->sgid_index;
270 packet->mad.hdr.hop_limit = grh->hop_limit;
271 packet->mad.hdr.traffic_class = grh->traffic_class;
272 memcpy(packet->mad.hdr.gid, &grh->dgid, 16);
273 packet->mad.hdr.flow_label = cpu_to_be32(grh->flow_label);
274 rdma_destroy_ah_attr(&ah_attr);
277 if (queue_packet(file, agent, packet))
284 ib_free_recv_mad(mad_recv_wc);
287 static ssize_t copy_recv_mad(struct ib_umad_file *file, char __user *buf,
288 struct ib_umad_packet *packet, size_t count)
290 struct ib_mad_recv_buf *recv_buf;
291 int left, seg_payload, offset, max_seg_payload;
294 recv_buf = &packet->recv_wc->recv_buf;
295 seg_size = packet->recv_wc->mad_seg_size;
297 /* We need enough room to copy the first (or only) MAD segment. */
298 if ((packet->length <= seg_size &&
299 count < hdr_size(file) + packet->length) ||
300 (packet->length > seg_size &&
301 count < hdr_size(file) + seg_size))
304 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
307 buf += hdr_size(file);
308 seg_payload = min_t(int, packet->length, seg_size);
309 if (copy_to_user(buf, recv_buf->mad, seg_payload))
312 if (seg_payload < packet->length) {
314 * Multipacket RMPP MAD message. Copy remainder of message.
315 * Note that last segment may have a shorter payload.
317 if (count < hdr_size(file) + packet->length) {
319 * The buffer is too small, return the first RMPP segment,
320 * which includes the RMPP message length.
324 offset = ib_get_mad_data_offset(recv_buf->mad->mad_hdr.mgmt_class);
325 max_seg_payload = seg_size - offset;
327 for (left = packet->length - seg_payload, buf += seg_payload;
328 left; left -= seg_payload, buf += seg_payload) {
329 recv_buf = container_of(recv_buf->list.next,
330 struct ib_mad_recv_buf, list);
331 seg_payload = min(left, max_seg_payload);
332 if (copy_to_user(buf, ((void *) recv_buf->mad) + offset,
337 return hdr_size(file) + packet->length;
340 static ssize_t copy_send_mad(struct ib_umad_file *file, char __user *buf,
341 struct ib_umad_packet *packet, size_t count)
343 ssize_t size = hdr_size(file) + packet->length;
348 if (copy_to_user(buf, &packet->mad, hdr_size(file)))
351 buf += hdr_size(file);
353 if (copy_to_user(buf, packet->mad.data, packet->length))
359 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
360 size_t count, loff_t *pos)
362 struct ib_umad_file *file = filp->private_data;
363 struct ib_umad_packet *packet;
366 if (count < hdr_size(file))
369 mutex_lock(&file->mutex);
371 while (list_empty(&file->recv_list)) {
372 mutex_unlock(&file->mutex);
374 if (filp->f_flags & O_NONBLOCK)
377 if (wait_event_interruptible(file->recv_wait,
378 !list_empty(&file->recv_list)))
381 mutex_lock(&file->mutex);
384 packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
385 list_del(&packet->list);
387 mutex_unlock(&file->mutex);
390 ret = copy_recv_mad(file, buf, packet, count);
392 ret = copy_send_mad(file, buf, packet, count);
396 mutex_lock(&file->mutex);
397 list_add(&packet->list, &file->recv_list);
398 mutex_unlock(&file->mutex);
401 ib_free_recv_mad(packet->recv_wc);
407 static int copy_rmpp_mad(struct ib_mad_send_buf *msg, const char __user *buf)
411 /* Copy class specific header */
412 if ((msg->hdr_len > IB_MGMT_RMPP_HDR) &&
413 copy_from_user(msg->mad + IB_MGMT_RMPP_HDR, buf + IB_MGMT_RMPP_HDR,
414 msg->hdr_len - IB_MGMT_RMPP_HDR))
417 /* All headers are in place. Copy data segments. */
418 for (seg = 1, left = msg->data_len, buf += msg->hdr_len; left > 0;
419 seg++, left -= msg->seg_size, buf += msg->seg_size) {
420 if (copy_from_user(ib_get_rmpp_segment(msg, seg), buf,
421 min(left, msg->seg_size)))
427 static int same_destination(struct ib_user_mad_hdr *hdr1,
428 struct ib_user_mad_hdr *hdr2)
430 if (!hdr1->grh_present && !hdr2->grh_present)
431 return (hdr1->lid == hdr2->lid);
433 if (hdr1->grh_present && hdr2->grh_present)
434 return !memcmp(hdr1->gid, hdr2->gid, 16);
439 static int is_duplicate(struct ib_umad_file *file,
440 struct ib_umad_packet *packet)
442 struct ib_umad_packet *sent_packet;
443 struct ib_mad_hdr *sent_hdr, *hdr;
445 hdr = (struct ib_mad_hdr *) packet->mad.data;
446 list_for_each_entry(sent_packet, &file->send_list, list) {
447 sent_hdr = (struct ib_mad_hdr *) sent_packet->mad.data;
449 if ((hdr->tid != sent_hdr->tid) ||
450 (hdr->mgmt_class != sent_hdr->mgmt_class))
454 * No need to be overly clever here. If two new operations have
455 * the same TID, reject the second as a duplicate. This is more
456 * restrictive than required by the spec.
458 if (!ib_response_mad(hdr)) {
459 if (!ib_response_mad(sent_hdr))
462 } else if (!ib_response_mad(sent_hdr))
465 if (same_destination(&packet->mad.hdr, &sent_packet->mad.hdr))
472 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
473 size_t count, loff_t *pos)
475 struct ib_umad_file *file = filp->private_data;
476 struct ib_umad_packet *packet;
477 struct ib_mad_agent *agent;
478 struct rdma_ah_attr ah_attr;
480 struct ib_rmpp_mad *rmpp_mad;
482 int ret, data_len, hdr_len, copy_offset, rmpp_active;
485 if (count < hdr_size(file) + IB_MGMT_RMPP_HDR)
488 packet = kzalloc(sizeof *packet + IB_MGMT_RMPP_HDR, GFP_KERNEL);
492 if (copy_from_user(&packet->mad, buf, hdr_size(file))) {
497 if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
502 buf += hdr_size(file);
504 if (copy_from_user(packet->mad.data, buf, IB_MGMT_RMPP_HDR)) {
509 mutex_lock(&file->mutex);
511 agent = __get_agent(file, packet->mad.hdr.id);
517 memset(&ah_attr, 0, sizeof ah_attr);
518 ah_attr.type = rdma_ah_find_type(agent->device,
519 file->port->port_num);
520 rdma_ah_set_dlid(&ah_attr, be16_to_cpu(packet->mad.hdr.lid));
521 rdma_ah_set_sl(&ah_attr, packet->mad.hdr.sl);
522 rdma_ah_set_path_bits(&ah_attr, packet->mad.hdr.path_bits);
523 rdma_ah_set_port_num(&ah_attr, file->port->port_num);
524 if (packet->mad.hdr.grh_present) {
525 rdma_ah_set_grh(&ah_attr, NULL,
526 be32_to_cpu(packet->mad.hdr.flow_label),
527 packet->mad.hdr.gid_index,
528 packet->mad.hdr.hop_limit,
529 packet->mad.hdr.traffic_class);
530 rdma_ah_set_dgid_raw(&ah_attr, packet->mad.hdr.gid);
533 ah = rdma_create_user_ah(agent->qp->pd, &ah_attr, NULL);
539 rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
540 hdr_len = ib_get_mad_data_offset(rmpp_mad->mad_hdr.mgmt_class);
542 if (ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
543 && ib_mad_kernel_rmpp_agent(agent)) {
544 copy_offset = IB_MGMT_RMPP_HDR;
545 rmpp_active = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
546 IB_MGMT_RMPP_FLAG_ACTIVE;
548 copy_offset = IB_MGMT_MAD_HDR;
552 base_version = ((struct ib_mad_hdr *)&packet->mad.data)->base_version;
553 data_len = count - hdr_size(file) - hdr_len;
554 packet->msg = ib_create_send_mad(agent,
555 be32_to_cpu(packet->mad.hdr.qpn),
556 packet->mad.hdr.pkey_index, rmpp_active,
557 hdr_len, data_len, GFP_KERNEL,
559 if (IS_ERR(packet->msg)) {
560 ret = PTR_ERR(packet->msg);
564 packet->msg->ah = ah;
565 packet->msg->timeout_ms = packet->mad.hdr.timeout_ms;
566 packet->msg->retries = packet->mad.hdr.retries;
567 packet->msg->context[0] = packet;
569 /* Copy MAD header. Any RMPP header is already in place. */
570 memcpy(packet->msg->mad, packet->mad.data, IB_MGMT_MAD_HDR);
573 if (copy_from_user(packet->msg->mad + copy_offset,
575 hdr_len + data_len - copy_offset)) {
580 ret = copy_rmpp_mad(packet->msg, buf);
586 * Set the high-order part of the transaction ID to make MADs from
587 * different agents unique, and allow routing responses back to the
588 * original requestor.
590 if (!ib_response_mad(packet->msg->mad)) {
591 tid = &((struct ib_mad_hdr *) packet->msg->mad)->tid;
592 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
593 (be64_to_cpup(tid) & 0xffffffff));
594 rmpp_mad->mad_hdr.tid = *tid;
597 if (!ib_mad_kernel_rmpp_agent(agent)
598 && ib_is_mad_class_rmpp(rmpp_mad->mad_hdr.mgmt_class)
599 && (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE)) {
600 spin_lock_irq(&file->send_lock);
601 list_add_tail(&packet->list, &file->send_list);
602 spin_unlock_irq(&file->send_lock);
604 spin_lock_irq(&file->send_lock);
605 ret = is_duplicate(file, packet);
607 list_add_tail(&packet->list, &file->send_list);
608 spin_unlock_irq(&file->send_lock);
615 ret = ib_post_send_mad(packet->msg, NULL);
619 mutex_unlock(&file->mutex);
623 dequeue_send(file, packet);
625 ib_free_send_mad(packet->msg);
627 rdma_destroy_ah(ah, RDMA_DESTROY_AH_SLEEPABLE);
629 mutex_unlock(&file->mutex);
635 static __poll_t ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
637 struct ib_umad_file *file = filp->private_data;
639 /* we will always be able to post a MAD send */
640 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
642 poll_wait(filp, &file->recv_wait, wait);
644 if (!list_empty(&file->recv_list))
645 mask |= EPOLLIN | EPOLLRDNORM;
650 static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
651 int compat_method_mask)
653 struct ib_user_mad_reg_req ureq;
654 struct ib_mad_reg_req req;
655 struct ib_mad_agent *agent = NULL;
659 mutex_lock(&file->port->file_mutex);
660 mutex_lock(&file->mutex);
662 if (!file->port->ib_dev) {
663 dev_notice(&file->port->dev,
664 "ib_umad_reg_agent: invalid device\n");
669 if (copy_from_user(&ureq, arg, sizeof ureq)) {
674 if (ureq.qpn != 0 && ureq.qpn != 1) {
675 dev_notice(&file->port->dev,
676 "ib_umad_reg_agent: invalid QPN %d specified\n",
682 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
683 if (!__get_agent(file, agent_id))
686 dev_notice(&file->port->dev,
687 "ib_umad_reg_agent: Max Agents (%u) reached\n",
693 if (ureq.mgmt_class) {
694 memset(&req, 0, sizeof(req));
695 req.mgmt_class = ureq.mgmt_class;
696 req.mgmt_class_version = ureq.mgmt_class_version;
697 memcpy(req.oui, ureq.oui, sizeof req.oui);
699 if (compat_method_mask) {
700 u32 *umm = (u32 *) ureq.method_mask;
703 for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
705 umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
707 memcpy(req.method_mask, ureq.method_mask,
708 sizeof req.method_mask);
711 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
712 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
713 ureq.mgmt_class ? &req : NULL,
715 send_handler, recv_handler, file, 0);
717 ret = PTR_ERR(agent);
722 if (put_user(agent_id,
723 (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
728 if (!file->already_used) {
729 file->already_used = 1;
730 if (!file->use_pkey_index) {
731 dev_warn(&file->port->dev,
732 "process %s did not enable P_Key index support.\n",
734 dev_warn(&file->port->dev,
735 " Documentation/infiniband/user_mad.txt has info on the new ABI.\n");
739 file->agent[agent_id] = agent;
743 mutex_unlock(&file->mutex);
746 ib_unregister_mad_agent(agent);
748 mutex_unlock(&file->port->file_mutex);
753 static int ib_umad_reg_agent2(struct ib_umad_file *file, void __user *arg)
755 struct ib_user_mad_reg_req2 ureq;
756 struct ib_mad_reg_req req;
757 struct ib_mad_agent *agent = NULL;
761 mutex_lock(&file->port->file_mutex);
762 mutex_lock(&file->mutex);
764 if (!file->port->ib_dev) {
765 dev_notice(&file->port->dev,
766 "ib_umad_reg_agent2: invalid device\n");
771 if (copy_from_user(&ureq, arg, sizeof(ureq))) {
776 if (ureq.qpn != 0 && ureq.qpn != 1) {
777 dev_notice(&file->port->dev,
778 "ib_umad_reg_agent2: invalid QPN %d specified\n",
784 if (ureq.flags & ~IB_USER_MAD_REG_FLAGS_CAP) {
785 dev_notice(&file->port->dev,
786 "ib_umad_reg_agent2 failed: invalid registration flags specified 0x%x; supported 0x%x\n",
787 ureq.flags, IB_USER_MAD_REG_FLAGS_CAP);
790 if (put_user((u32)IB_USER_MAD_REG_FLAGS_CAP,
791 (u32 __user *) (arg + offsetof(struct
792 ib_user_mad_reg_req2, flags))))
798 for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
799 if (!__get_agent(file, agent_id))
802 dev_notice(&file->port->dev,
803 "ib_umad_reg_agent2: Max Agents (%u) reached\n",
809 if (ureq.mgmt_class) {
810 memset(&req, 0, sizeof(req));
811 req.mgmt_class = ureq.mgmt_class;
812 req.mgmt_class_version = ureq.mgmt_class_version;
813 if (ureq.oui & 0xff000000) {
814 dev_notice(&file->port->dev,
815 "ib_umad_reg_agent2 failed: oui invalid 0x%08x\n",
820 req.oui[2] = ureq.oui & 0x0000ff;
821 req.oui[1] = (ureq.oui & 0x00ff00) >> 8;
822 req.oui[0] = (ureq.oui & 0xff0000) >> 16;
823 memcpy(req.method_mask, ureq.method_mask,
824 sizeof(req.method_mask));
827 agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
828 ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
829 ureq.mgmt_class ? &req : NULL,
831 send_handler, recv_handler, file,
834 ret = PTR_ERR(agent);
839 if (put_user(agent_id,
841 offsetof(struct ib_user_mad_reg_req2, id)))) {
846 if (!file->already_used) {
847 file->already_used = 1;
848 file->use_pkey_index = 1;
851 file->agent[agent_id] = agent;
855 mutex_unlock(&file->mutex);
858 ib_unregister_mad_agent(agent);
860 mutex_unlock(&file->port->file_mutex);
866 static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
868 struct ib_mad_agent *agent = NULL;
872 if (get_user(id, arg))
875 mutex_lock(&file->port->file_mutex);
876 mutex_lock(&file->mutex);
878 if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
883 agent = file->agent[id];
884 file->agent[id] = NULL;
887 mutex_unlock(&file->mutex);
890 ib_unregister_mad_agent(agent);
892 mutex_unlock(&file->port->file_mutex);
897 static long ib_umad_enable_pkey(struct ib_umad_file *file)
901 mutex_lock(&file->mutex);
902 if (file->already_used)
905 file->use_pkey_index = 1;
906 mutex_unlock(&file->mutex);
911 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
915 case IB_USER_MAD_REGISTER_AGENT:
916 return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
917 case IB_USER_MAD_UNREGISTER_AGENT:
918 return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
919 case IB_USER_MAD_ENABLE_PKEY:
920 return ib_umad_enable_pkey(filp->private_data);
921 case IB_USER_MAD_REGISTER_AGENT2:
922 return ib_umad_reg_agent2(filp->private_data, (void __user *) arg);
929 static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
933 case IB_USER_MAD_REGISTER_AGENT:
934 return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
935 case IB_USER_MAD_UNREGISTER_AGENT:
936 return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
937 case IB_USER_MAD_ENABLE_PKEY:
938 return ib_umad_enable_pkey(filp->private_data);
939 case IB_USER_MAD_REGISTER_AGENT2:
940 return ib_umad_reg_agent2(filp->private_data, compat_ptr(arg));
948 * ib_umad_open() does not need the BKL:
950 * - the ib_umad_port structures are properly reference counted, and
951 * everything else is purely local to the file being created, so
952 * races against other open calls are not a problem;
953 * - the ioctl method does not affect any global state outside of the
954 * file structure being operated on;
956 static int ib_umad_open(struct inode *inode, struct file *filp)
958 struct ib_umad_port *port;
959 struct ib_umad_file *file;
962 port = container_of(inode->i_cdev, struct ib_umad_port, cdev);
964 mutex_lock(&port->file_mutex);
971 file = kzalloc(sizeof(*file), GFP_KERNEL);
977 mutex_init(&file->mutex);
978 spin_lock_init(&file->send_lock);
979 INIT_LIST_HEAD(&file->recv_list);
980 INIT_LIST_HEAD(&file->send_list);
981 init_waitqueue_head(&file->recv_wait);
984 filp->private_data = file;
986 list_add_tail(&file->port_list, &port->file_list);
988 stream_open(inode, filp);
990 mutex_unlock(&port->file_mutex);
994 static int ib_umad_close(struct inode *inode, struct file *filp)
996 struct ib_umad_file *file = filp->private_data;
997 struct ib_umad_packet *packet, *tmp;
1001 mutex_lock(&file->port->file_mutex);
1002 mutex_lock(&file->mutex);
1004 already_dead = file->agents_dead;
1005 file->agents_dead = 1;
1007 list_for_each_entry_safe(packet, tmp, &file->recv_list, list) {
1008 if (packet->recv_wc)
1009 ib_free_recv_mad(packet->recv_wc);
1013 list_del(&file->port_list);
1015 mutex_unlock(&file->mutex);
1018 for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
1020 ib_unregister_mad_agent(file->agent[i]);
1022 mutex_unlock(&file->port->file_mutex);
1028 static const struct file_operations umad_fops = {
1029 .owner = THIS_MODULE,
1030 .read = ib_umad_read,
1031 .write = ib_umad_write,
1032 .poll = ib_umad_poll,
1033 .unlocked_ioctl = ib_umad_ioctl,
1034 #ifdef CONFIG_COMPAT
1035 .compat_ioctl = ib_umad_compat_ioctl,
1037 .open = ib_umad_open,
1038 .release = ib_umad_close,
1039 .llseek = no_llseek,
1042 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
1044 struct ib_umad_port *port;
1045 struct ib_port_modify props = {
1046 .set_port_cap_mask = IB_PORT_SM
1050 port = container_of(inode->i_cdev, struct ib_umad_port, sm_cdev);
1052 if (filp->f_flags & O_NONBLOCK) {
1053 if (down_trylock(&port->sm_sem)) {
1058 if (down_interruptible(&port->sm_sem)) {
1064 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1068 filp->private_data = port;
1070 nonseekable_open(inode, filp);
1080 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
1082 struct ib_umad_port *port = filp->private_data;
1083 struct ib_port_modify props = {
1084 .clr_port_cap_mask = IB_PORT_SM
1088 mutex_lock(&port->file_mutex);
1090 ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
1091 mutex_unlock(&port->file_mutex);
1098 static const struct file_operations umad_sm_fops = {
1099 .owner = THIS_MODULE,
1100 .open = ib_umad_sm_open,
1101 .release = ib_umad_sm_close,
1102 .llseek = no_llseek,
1105 static struct ib_client umad_client = {
1107 .add = ib_umad_add_one,
1108 .remove = ib_umad_remove_one
1111 static ssize_t ibdev_show(struct device *dev, struct device_attribute *attr,
1114 struct ib_umad_port *port = dev_get_drvdata(dev);
1119 return sprintf(buf, "%s\n", dev_name(&port->ib_dev->dev));
1121 static DEVICE_ATTR_RO(ibdev);
1123 static ssize_t port_show(struct device *dev, struct device_attribute *attr,
1126 struct ib_umad_port *port = dev_get_drvdata(dev);
1131 return sprintf(buf, "%d\n", port->port_num);
1133 static DEVICE_ATTR_RO(port);
1135 static struct attribute *umad_class_dev_attrs[] = {
1136 &dev_attr_ibdev.attr,
1137 &dev_attr_port.attr,
1140 ATTRIBUTE_GROUPS(umad_class_dev);
1142 static char *umad_devnode(struct device *dev, umode_t *mode)
1144 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
1147 static ssize_t abi_version_show(struct class *class,
1148 struct class_attribute *attr, char *buf)
1150 return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
1152 static CLASS_ATTR_RO(abi_version);
1154 static struct attribute *umad_class_attrs[] = {
1155 &class_attr_abi_version.attr,
1158 ATTRIBUTE_GROUPS(umad_class);
1160 static struct class umad_class = {
1161 .name = "infiniband_mad",
1162 .devnode = umad_devnode,
1163 .class_groups = umad_class_groups,
1164 .dev_groups = umad_class_dev_groups,
1167 static void ib_umad_release_port(struct device *device)
1169 struct ib_umad_port *port = dev_get_drvdata(device);
1170 struct ib_umad_device *umad_dev = port->umad_dev;
1172 ib_umad_dev_put(umad_dev);
1175 static void ib_umad_init_port_dev(struct device *dev,
1176 struct ib_umad_port *port,
1177 const struct ib_device *device)
1179 device_initialize(dev);
1180 ib_umad_dev_get(port->umad_dev);
1181 dev->class = &umad_class;
1182 dev->parent = device->dev.parent;
1183 dev_set_drvdata(dev, port);
1184 dev->release = ib_umad_release_port;
1187 static int ib_umad_init_port(struct ib_device *device, int port_num,
1188 struct ib_umad_device *umad_dev,
1189 struct ib_umad_port *port)
1196 devnum = ida_alloc_max(&umad_ida, IB_UMAD_MAX_PORTS - 1, GFP_KERNEL);
1199 port->dev_num = devnum;
1200 if (devnum >= IB_UMAD_NUM_FIXED_MINOR) {
1201 base_umad = dynamic_umad_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1202 base_issm = dynamic_issm_dev + devnum - IB_UMAD_NUM_FIXED_MINOR;
1204 base_umad = devnum + base_umad_dev;
1205 base_issm = devnum + base_issm_dev;
1208 port->ib_dev = device;
1209 port->umad_dev = umad_dev;
1210 port->port_num = port_num;
1211 sema_init(&port->sm_sem, 1);
1212 mutex_init(&port->file_mutex);
1213 INIT_LIST_HEAD(&port->file_list);
1215 ib_umad_init_port_dev(&port->dev, port, device);
1216 port->dev.devt = base_umad;
1217 dev_set_name(&port->dev, "umad%d", port->dev_num);
1218 cdev_init(&port->cdev, &umad_fops);
1219 port->cdev.owner = THIS_MODULE;
1221 ret = cdev_device_add(&port->cdev, &port->dev);
1225 ib_umad_init_port_dev(&port->sm_dev, port, device);
1226 port->sm_dev.devt = base_issm;
1227 dev_set_name(&port->sm_dev, "issm%d", port->dev_num);
1228 cdev_init(&port->sm_cdev, &umad_sm_fops);
1229 port->sm_cdev.owner = THIS_MODULE;
1231 ret = cdev_device_add(&port->sm_cdev, &port->sm_dev);
1238 put_device(&port->sm_dev);
1239 cdev_device_del(&port->cdev, &port->dev);
1241 put_device(&port->dev);
1242 ida_free(&umad_ida, devnum);
1246 static void ib_umad_kill_port(struct ib_umad_port *port)
1248 struct ib_umad_file *file;
1251 mutex_lock(&port->file_mutex);
1253 /* Mark ib_dev NULL and block ioctl or other file ops to progress
1256 port->ib_dev = NULL;
1258 list_for_each_entry(file, &port->file_list, port_list) {
1259 mutex_lock(&file->mutex);
1260 file->agents_dead = 1;
1261 mutex_unlock(&file->mutex);
1263 for (id = 0; id < IB_UMAD_MAX_AGENTS; ++id)
1264 if (file->agent[id])
1265 ib_unregister_mad_agent(file->agent[id]);
1268 mutex_unlock(&port->file_mutex);
1270 cdev_device_del(&port->sm_cdev, &port->sm_dev);
1271 cdev_device_del(&port->cdev, &port->dev);
1272 ida_free(&umad_ida, port->dev_num);
1274 /* balances device_initialize() */
1275 put_device(&port->sm_dev);
1276 put_device(&port->dev);
1279 static void ib_umad_add_one(struct ib_device *device)
1281 struct ib_umad_device *umad_dev;
1285 s = rdma_start_port(device);
1286 e = rdma_end_port(device);
1288 umad_dev = kzalloc(struct_size(umad_dev, ports, e - s + 1), GFP_KERNEL);
1292 kref_init(&umad_dev->kref);
1293 for (i = s; i <= e; ++i) {
1294 if (!rdma_cap_ib_mad(device, i))
1297 if (ib_umad_init_port(device, i, umad_dev,
1298 &umad_dev->ports[i - s]))
1307 ib_set_client_data(device, &umad_client, umad_dev);
1313 if (!rdma_cap_ib_mad(device, i))
1316 ib_umad_kill_port(&umad_dev->ports[i - s]);
1319 /* balances kref_init */
1320 ib_umad_dev_put(umad_dev);
1323 static void ib_umad_remove_one(struct ib_device *device, void *client_data)
1325 struct ib_umad_device *umad_dev = client_data;
1331 rdma_for_each_port (device, i) {
1332 if (rdma_cap_ib_mad(device, i))
1334 &umad_dev->ports[i - rdma_start_port(device)]);
1336 /* balances kref_init() */
1337 ib_umad_dev_put(umad_dev);
1340 static int __init ib_umad_init(void)
1344 ret = register_chrdev_region(base_umad_dev,
1345 IB_UMAD_NUM_FIXED_MINOR * 2,
1348 pr_err("couldn't register device number\n");
1352 ret = alloc_chrdev_region(&dynamic_umad_dev, 0,
1353 IB_UMAD_NUM_DYNAMIC_MINOR * 2,
1356 pr_err("couldn't register dynamic device number\n");
1359 dynamic_issm_dev = dynamic_umad_dev + IB_UMAD_NUM_DYNAMIC_MINOR;
1361 ret = class_register(&umad_class);
1363 pr_err("couldn't create class infiniband_mad\n");
1367 ret = ib_register_client(&umad_client);
1369 pr_err("couldn't register ib_umad client\n");
1376 class_unregister(&umad_class);
1379 unregister_chrdev_region(dynamic_umad_dev,
1380 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1383 unregister_chrdev_region(base_umad_dev,
1384 IB_UMAD_NUM_FIXED_MINOR * 2);
1390 static void __exit ib_umad_cleanup(void)
1392 ib_unregister_client(&umad_client);
1393 class_unregister(&umad_class);
1394 unregister_chrdev_region(base_umad_dev,
1395 IB_UMAD_NUM_FIXED_MINOR * 2);
1396 unregister_chrdev_region(dynamic_umad_dev,
1397 IB_UMAD_NUM_DYNAMIC_MINOR * 2);
1400 module_init(ib_umad_init);
1401 module_exit(ib_umad_cleanup);