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 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/device.h>
39 #include <linux/err.h>
41 #include <linux/cdev.h>
42 #include <linux/dma-mapping.h>
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/kref.h>
46 #include <linux/compat.h>
47 #include <linux/semaphore.h>
49 #include <asm/uaccess.h>
51 #include <rdma/ib_mad.h>
52 #include <rdma/ib_user_mad.h>
54 MODULE_AUTHOR("Roland Dreier");
55 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
56 MODULE_LICENSE("Dual BSD/GPL");
59 IB_UMAD_MAX_PORTS
= 64,
60 IB_UMAD_MAX_AGENTS
= 32,
63 IB_UMAD_MINOR_BASE
= 0
67 * Our lifetime rules for these structs are the following: each time a
68 * device special file is opened, we look up the corresponding struct
69 * ib_umad_port by minor in the umad_port[] table while holding the
70 * port_lock. If this lookup succeeds, we take a reference on the
71 * ib_umad_port's struct ib_umad_device while still holding the
72 * port_lock; if the lookup fails, we fail the open(). We drop these
73 * references in the corresponding close().
75 * In addition to references coming from open character devices, there
76 * is one more reference to each ib_umad_device representing the
77 * module's reference taken when allocating the ib_umad_device in
80 * When destroying an ib_umad_device, we clear all of its
81 * ib_umad_ports from umad_port[] while holding port_lock before
82 * dropping the module's reference to the ib_umad_device. This is
83 * always safe because any open() calls will either succeed and obtain
84 * a reference before we clear the umad_port[] entries, or fail after
85 * we clear the umad_port[] entries.
93 struct device
*sm_dev
;
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
{
106 int start_port
, end_port
;
108 struct ib_umad_port port
[0];
111 struct ib_umad_file
{
113 struct ib_umad_port
*port
;
114 struct list_head recv_list
;
115 struct list_head send_list
;
116 struct list_head port_list
;
117 spinlock_t send_lock
;
118 wait_queue_head_t recv_wait
;
119 struct ib_mad_agent
*agent
[IB_UMAD_MAX_AGENTS
];
125 struct ib_umad_packet
{
126 struct ib_mad_send_buf
*msg
;
127 struct ib_mad_recv_wc
*recv_wc
;
128 struct list_head list
;
130 struct ib_user_mad mad
;
133 static struct class *umad_class
;
135 static const dev_t base_dev
= MKDEV(IB_UMAD_MAJOR
, IB_UMAD_MINOR_BASE
);
137 static DEFINE_SPINLOCK(port_lock
);
138 static struct ib_umad_port
*umad_port
[IB_UMAD_MAX_PORTS
];
139 static DECLARE_BITMAP(dev_map
, IB_UMAD_MAX_PORTS
);
141 static void ib_umad_add_one(struct ib_device
*device
);
142 static void ib_umad_remove_one(struct ib_device
*device
);
144 static void ib_umad_release_dev(struct kref
*ref
)
146 struct ib_umad_device
*dev
=
147 container_of(ref
, struct ib_umad_device
, ref
);
152 static int hdr_size(struct ib_umad_file
*file
)
154 return file
->use_pkey_index
? sizeof (struct ib_user_mad_hdr
) :
155 sizeof (struct ib_user_mad_hdr_old
);
158 /* caller must hold file->mutex */
159 static struct ib_mad_agent
*__get_agent(struct ib_umad_file
*file
, int id
)
161 return file
->agents_dead
? NULL
: file
->agent
[id
];
164 static int queue_packet(struct ib_umad_file
*file
,
165 struct ib_mad_agent
*agent
,
166 struct ib_umad_packet
*packet
)
170 mutex_lock(&file
->mutex
);
172 for (packet
->mad
.hdr
.id
= 0;
173 packet
->mad
.hdr
.id
< IB_UMAD_MAX_AGENTS
;
174 packet
->mad
.hdr
.id
++)
175 if (agent
== __get_agent(file
, packet
->mad
.hdr
.id
)) {
176 list_add_tail(&packet
->list
, &file
->recv_list
);
177 wake_up_interruptible(&file
->recv_wait
);
182 mutex_unlock(&file
->mutex
);
187 static void dequeue_send(struct ib_umad_file
*file
,
188 struct ib_umad_packet
*packet
)
190 spin_lock_irq(&file
->send_lock
);
191 list_del(&packet
->list
);
192 spin_unlock_irq(&file
->send_lock
);
195 static void send_handler(struct ib_mad_agent
*agent
,
196 struct ib_mad_send_wc
*send_wc
)
198 struct ib_umad_file
*file
= agent
->context
;
199 struct ib_umad_packet
*packet
= send_wc
->send_buf
->context
[0];
201 dequeue_send(file
, packet
);
202 ib_destroy_ah(packet
->msg
->ah
);
203 ib_free_send_mad(packet
->msg
);
205 if (send_wc
->status
== IB_WC_RESP_TIMEOUT_ERR
) {
206 packet
->length
= IB_MGMT_MAD_HDR
;
207 packet
->mad
.hdr
.status
= ETIMEDOUT
;
208 if (!queue_packet(file
, agent
, packet
))
214 static void recv_handler(struct ib_mad_agent
*agent
,
215 struct ib_mad_recv_wc
*mad_recv_wc
)
217 struct ib_umad_file
*file
= agent
->context
;
218 struct ib_umad_packet
*packet
;
220 if (mad_recv_wc
->wc
->status
!= IB_WC_SUCCESS
)
223 packet
= kzalloc(sizeof *packet
, GFP_KERNEL
);
227 packet
->length
= mad_recv_wc
->mad_len
;
228 packet
->recv_wc
= mad_recv_wc
;
230 packet
->mad
.hdr
.status
= 0;
231 packet
->mad
.hdr
.length
= hdr_size(file
) + mad_recv_wc
->mad_len
;
232 packet
->mad
.hdr
.qpn
= cpu_to_be32(mad_recv_wc
->wc
->src_qp
);
233 packet
->mad
.hdr
.lid
= cpu_to_be16(mad_recv_wc
->wc
->slid
);
234 packet
->mad
.hdr
.sl
= mad_recv_wc
->wc
->sl
;
235 packet
->mad
.hdr
.path_bits
= mad_recv_wc
->wc
->dlid_path_bits
;
236 packet
->mad
.hdr
.pkey_index
= mad_recv_wc
->wc
->pkey_index
;
237 packet
->mad
.hdr
.grh_present
= !!(mad_recv_wc
->wc
->wc_flags
& IB_WC_GRH
);
238 if (packet
->mad
.hdr
.grh_present
) {
239 struct ib_ah_attr ah_attr
;
241 ib_init_ah_from_wc(agent
->device
, agent
->port_num
,
242 mad_recv_wc
->wc
, mad_recv_wc
->recv_buf
.grh
,
245 packet
->mad
.hdr
.gid_index
= ah_attr
.grh
.sgid_index
;
246 packet
->mad
.hdr
.hop_limit
= ah_attr
.grh
.hop_limit
;
247 packet
->mad
.hdr
.traffic_class
= ah_attr
.grh
.traffic_class
;
248 memcpy(packet
->mad
.hdr
.gid
, &ah_attr
.grh
.dgid
, 16);
249 packet
->mad
.hdr
.flow_label
= cpu_to_be32(ah_attr
.grh
.flow_label
);
252 if (queue_packet(file
, agent
, packet
))
259 ib_free_recv_mad(mad_recv_wc
);
262 static ssize_t
copy_recv_mad(struct ib_umad_file
*file
, char __user
*buf
,
263 struct ib_umad_packet
*packet
, size_t count
)
265 struct ib_mad_recv_buf
*recv_buf
;
266 int left
, seg_payload
, offset
, max_seg_payload
;
268 /* We need enough room to copy the first (or only) MAD segment. */
269 recv_buf
= &packet
->recv_wc
->recv_buf
;
270 if ((packet
->length
<= sizeof (*recv_buf
->mad
) &&
271 count
< hdr_size(file
) + packet
->length
) ||
272 (packet
->length
> sizeof (*recv_buf
->mad
) &&
273 count
< hdr_size(file
) + sizeof (*recv_buf
->mad
)))
276 if (copy_to_user(buf
, &packet
->mad
, hdr_size(file
)))
279 buf
+= hdr_size(file
);
280 seg_payload
= min_t(int, packet
->length
, sizeof (*recv_buf
->mad
));
281 if (copy_to_user(buf
, recv_buf
->mad
, seg_payload
))
284 if (seg_payload
< packet
->length
) {
286 * Multipacket RMPP MAD message. Copy remainder of message.
287 * Note that last segment may have a shorter payload.
289 if (count
< hdr_size(file
) + packet
->length
) {
291 * The buffer is too small, return the first RMPP segment,
292 * which includes the RMPP message length.
296 offset
= ib_get_mad_data_offset(recv_buf
->mad
->mad_hdr
.mgmt_class
);
297 max_seg_payload
= sizeof (struct ib_mad
) - offset
;
299 for (left
= packet
->length
- seg_payload
, buf
+= seg_payload
;
300 left
; left
-= seg_payload
, buf
+= seg_payload
) {
301 recv_buf
= container_of(recv_buf
->list
.next
,
302 struct ib_mad_recv_buf
, list
);
303 seg_payload
= min(left
, max_seg_payload
);
304 if (copy_to_user(buf
, ((void *) recv_buf
->mad
) + offset
,
309 return hdr_size(file
) + packet
->length
;
312 static ssize_t
copy_send_mad(struct ib_umad_file
*file
, char __user
*buf
,
313 struct ib_umad_packet
*packet
, size_t count
)
315 ssize_t size
= hdr_size(file
) + packet
->length
;
320 if (copy_to_user(buf
, &packet
->mad
, hdr_size(file
)))
323 buf
+= hdr_size(file
);
325 if (copy_to_user(buf
, packet
->mad
.data
, packet
->length
))
331 static ssize_t
ib_umad_read(struct file
*filp
, char __user
*buf
,
332 size_t count
, loff_t
*pos
)
334 struct ib_umad_file
*file
= filp
->private_data
;
335 struct ib_umad_packet
*packet
;
338 if (count
< hdr_size(file
))
341 mutex_lock(&file
->mutex
);
343 while (list_empty(&file
->recv_list
)) {
344 mutex_unlock(&file
->mutex
);
346 if (filp
->f_flags
& O_NONBLOCK
)
349 if (wait_event_interruptible(file
->recv_wait
,
350 !list_empty(&file
->recv_list
)))
353 mutex_lock(&file
->mutex
);
356 packet
= list_entry(file
->recv_list
.next
, struct ib_umad_packet
, list
);
357 list_del(&packet
->list
);
359 mutex_unlock(&file
->mutex
);
362 ret
= copy_recv_mad(file
, buf
, packet
, count
);
364 ret
= copy_send_mad(file
, buf
, packet
, count
);
368 mutex_lock(&file
->mutex
);
369 list_add(&packet
->list
, &file
->recv_list
);
370 mutex_unlock(&file
->mutex
);
373 ib_free_recv_mad(packet
->recv_wc
);
379 static int copy_rmpp_mad(struct ib_mad_send_buf
*msg
, const char __user
*buf
)
383 /* Copy class specific header */
384 if ((msg
->hdr_len
> IB_MGMT_RMPP_HDR
) &&
385 copy_from_user(msg
->mad
+ IB_MGMT_RMPP_HDR
, buf
+ IB_MGMT_RMPP_HDR
,
386 msg
->hdr_len
- IB_MGMT_RMPP_HDR
))
389 /* All headers are in place. Copy data segments. */
390 for (seg
= 1, left
= msg
->data_len
, buf
+= msg
->hdr_len
; left
> 0;
391 seg
++, left
-= msg
->seg_size
, buf
+= msg
->seg_size
) {
392 if (copy_from_user(ib_get_rmpp_segment(msg
, seg
), buf
,
393 min(left
, msg
->seg_size
)))
399 static int same_destination(struct ib_user_mad_hdr
*hdr1
,
400 struct ib_user_mad_hdr
*hdr2
)
402 if (!hdr1
->grh_present
&& !hdr2
->grh_present
)
403 return (hdr1
->lid
== hdr2
->lid
);
405 if (hdr1
->grh_present
&& hdr2
->grh_present
)
406 return !memcmp(hdr1
->gid
, hdr2
->gid
, 16);
411 static int is_duplicate(struct ib_umad_file
*file
,
412 struct ib_umad_packet
*packet
)
414 struct ib_umad_packet
*sent_packet
;
415 struct ib_mad_hdr
*sent_hdr
, *hdr
;
417 hdr
= (struct ib_mad_hdr
*) packet
->mad
.data
;
418 list_for_each_entry(sent_packet
, &file
->send_list
, list
) {
419 sent_hdr
= (struct ib_mad_hdr
*) sent_packet
->mad
.data
;
421 if ((hdr
->tid
!= sent_hdr
->tid
) ||
422 (hdr
->mgmt_class
!= sent_hdr
->mgmt_class
))
426 * No need to be overly clever here. If two new operations have
427 * the same TID, reject the second as a duplicate. This is more
428 * restrictive than required by the spec.
430 if (!ib_response_mad((struct ib_mad
*) hdr
)) {
431 if (!ib_response_mad((struct ib_mad
*) sent_hdr
))
434 } else if (!ib_response_mad((struct ib_mad
*) sent_hdr
))
437 if (same_destination(&packet
->mad
.hdr
, &sent_packet
->mad
.hdr
))
444 static ssize_t
ib_umad_write(struct file
*filp
, const char __user
*buf
,
445 size_t count
, loff_t
*pos
)
447 struct ib_umad_file
*file
= filp
->private_data
;
448 struct ib_umad_packet
*packet
;
449 struct ib_mad_agent
*agent
;
450 struct ib_ah_attr ah_attr
;
452 struct ib_rmpp_mad
*rmpp_mad
;
454 int ret
, data_len
, hdr_len
, copy_offset
, rmpp_active
;
456 if (count
< hdr_size(file
) + IB_MGMT_RMPP_HDR
)
459 packet
= kzalloc(sizeof *packet
+ IB_MGMT_RMPP_HDR
, GFP_KERNEL
);
463 if (copy_from_user(&packet
->mad
, buf
, hdr_size(file
))) {
468 if (packet
->mad
.hdr
.id
< 0 ||
469 packet
->mad
.hdr
.id
>= IB_UMAD_MAX_AGENTS
) {
474 buf
+= hdr_size(file
);
476 if (copy_from_user(packet
->mad
.data
, buf
, IB_MGMT_RMPP_HDR
)) {
481 mutex_lock(&file
->mutex
);
483 agent
= __get_agent(file
, packet
->mad
.hdr
.id
);
489 memset(&ah_attr
, 0, sizeof ah_attr
);
490 ah_attr
.dlid
= be16_to_cpu(packet
->mad
.hdr
.lid
);
491 ah_attr
.sl
= packet
->mad
.hdr
.sl
;
492 ah_attr
.src_path_bits
= packet
->mad
.hdr
.path_bits
;
493 ah_attr
.port_num
= file
->port
->port_num
;
494 if (packet
->mad
.hdr
.grh_present
) {
495 ah_attr
.ah_flags
= IB_AH_GRH
;
496 memcpy(ah_attr
.grh
.dgid
.raw
, packet
->mad
.hdr
.gid
, 16);
497 ah_attr
.grh
.sgid_index
= packet
->mad
.hdr
.gid_index
;
498 ah_attr
.grh
.flow_label
= be32_to_cpu(packet
->mad
.hdr
.flow_label
);
499 ah_attr
.grh
.hop_limit
= packet
->mad
.hdr
.hop_limit
;
500 ah_attr
.grh
.traffic_class
= packet
->mad
.hdr
.traffic_class
;
503 ah
= ib_create_ah(agent
->qp
->pd
, &ah_attr
);
509 rmpp_mad
= (struct ib_rmpp_mad
*) packet
->mad
.data
;
510 hdr_len
= ib_get_mad_data_offset(rmpp_mad
->mad_hdr
.mgmt_class
);
511 if (!ib_is_mad_class_rmpp(rmpp_mad
->mad_hdr
.mgmt_class
)) {
512 copy_offset
= IB_MGMT_MAD_HDR
;
515 copy_offset
= IB_MGMT_RMPP_HDR
;
516 rmpp_active
= ib_get_rmpp_flags(&rmpp_mad
->rmpp_hdr
) &
517 IB_MGMT_RMPP_FLAG_ACTIVE
;
520 data_len
= count
- hdr_size(file
) - hdr_len
;
521 packet
->msg
= ib_create_send_mad(agent
,
522 be32_to_cpu(packet
->mad
.hdr
.qpn
),
523 packet
->mad
.hdr
.pkey_index
, rmpp_active
,
524 hdr_len
, data_len
, GFP_KERNEL
);
525 if (IS_ERR(packet
->msg
)) {
526 ret
= PTR_ERR(packet
->msg
);
530 packet
->msg
->ah
= ah
;
531 packet
->msg
->timeout_ms
= packet
->mad
.hdr
.timeout_ms
;
532 packet
->msg
->retries
= packet
->mad
.hdr
.retries
;
533 packet
->msg
->context
[0] = packet
;
535 /* Copy MAD header. Any RMPP header is already in place. */
536 memcpy(packet
->msg
->mad
, packet
->mad
.data
, IB_MGMT_MAD_HDR
);
539 if (copy_from_user(packet
->msg
->mad
+ copy_offset
,
541 hdr_len
+ data_len
- copy_offset
)) {
546 ret
= copy_rmpp_mad(packet
->msg
, buf
);
552 * Set the high-order part of the transaction ID to make MADs from
553 * different agents unique, and allow routing responses back to the
554 * original requestor.
556 if (!ib_response_mad(packet
->msg
->mad
)) {
557 tid
= &((struct ib_mad_hdr
*) packet
->msg
->mad
)->tid
;
558 *tid
= cpu_to_be64(((u64
) agent
->hi_tid
) << 32 |
559 (be64_to_cpup(tid
) & 0xffffffff));
560 rmpp_mad
->mad_hdr
.tid
= *tid
;
563 spin_lock_irq(&file
->send_lock
);
564 ret
= is_duplicate(file
, packet
);
566 list_add_tail(&packet
->list
, &file
->send_list
);
567 spin_unlock_irq(&file
->send_lock
);
573 ret
= ib_post_send_mad(packet
->msg
, NULL
);
577 mutex_unlock(&file
->mutex
);
581 dequeue_send(file
, packet
);
583 ib_free_send_mad(packet
->msg
);
587 mutex_unlock(&file
->mutex
);
593 static unsigned int ib_umad_poll(struct file
*filp
, struct poll_table_struct
*wait
)
595 struct ib_umad_file
*file
= filp
->private_data
;
597 /* we will always be able to post a MAD send */
598 unsigned int mask
= POLLOUT
| POLLWRNORM
;
600 poll_wait(filp
, &file
->recv_wait
, wait
);
602 if (!list_empty(&file
->recv_list
))
603 mask
|= POLLIN
| POLLRDNORM
;
608 static int ib_umad_reg_agent(struct ib_umad_file
*file
, void __user
*arg
,
609 int compat_method_mask
)
611 struct ib_user_mad_reg_req ureq
;
612 struct ib_mad_reg_req req
;
613 struct ib_mad_agent
*agent
= NULL
;
617 mutex_lock(&file
->port
->file_mutex
);
618 mutex_lock(&file
->mutex
);
620 if (!file
->port
->ib_dev
) {
625 if (copy_from_user(&ureq
, arg
, sizeof ureq
)) {
630 if (ureq
.qpn
!= 0 && ureq
.qpn
!= 1) {
635 for (agent_id
= 0; agent_id
< IB_UMAD_MAX_AGENTS
; ++agent_id
)
636 if (!__get_agent(file
, agent_id
))
643 if (ureq
.mgmt_class
) {
644 req
.mgmt_class
= ureq
.mgmt_class
;
645 req
.mgmt_class_version
= ureq
.mgmt_class_version
;
646 memcpy(req
.oui
, ureq
.oui
, sizeof req
.oui
);
648 if (compat_method_mask
) {
649 u32
*umm
= (u32
*) ureq
.method_mask
;
652 for (i
= 0; i
< BITS_TO_LONGS(IB_MGMT_MAX_METHODS
); ++i
)
654 umm
[i
* 2] | ((u64
) umm
[i
* 2 + 1] << 32);
656 memcpy(req
.method_mask
, ureq
.method_mask
,
657 sizeof req
.method_mask
);
660 agent
= ib_register_mad_agent(file
->port
->ib_dev
, file
->port
->port_num
,
661 ureq
.qpn
? IB_QPT_GSI
: IB_QPT_SMI
,
662 ureq
.mgmt_class
? &req
: NULL
,
664 send_handler
, recv_handler
, file
);
666 ret
= PTR_ERR(agent
);
671 if (put_user(agent_id
,
672 (u32 __user
*) (arg
+ offsetof(struct ib_user_mad_reg_req
, id
)))) {
677 if (!file
->already_used
) {
678 file
->already_used
= 1;
679 if (!file
->use_pkey_index
) {
680 printk(KERN_WARNING
"user_mad: process %s did not enable "
681 "P_Key index support.\n", current
->comm
);
682 printk(KERN_WARNING
"user_mad: Documentation/infiniband/user_mad.txt "
683 "has info on the new ABI.\n");
687 file
->agent
[agent_id
] = agent
;
691 mutex_unlock(&file
->mutex
);
694 ib_unregister_mad_agent(agent
);
696 mutex_unlock(&file
->port
->file_mutex
);
701 static int ib_umad_unreg_agent(struct ib_umad_file
*file
, u32 __user
*arg
)
703 struct ib_mad_agent
*agent
= NULL
;
707 if (get_user(id
, arg
))
710 mutex_lock(&file
->port
->file_mutex
);
711 mutex_lock(&file
->mutex
);
713 if (id
< 0 || id
>= IB_UMAD_MAX_AGENTS
|| !__get_agent(file
, id
)) {
718 agent
= file
->agent
[id
];
719 file
->agent
[id
] = NULL
;
722 mutex_unlock(&file
->mutex
);
725 ib_unregister_mad_agent(agent
);
727 mutex_unlock(&file
->port
->file_mutex
);
732 static long ib_umad_enable_pkey(struct ib_umad_file
*file
)
736 mutex_lock(&file
->mutex
);
737 if (file
->already_used
)
740 file
->use_pkey_index
= 1;
741 mutex_unlock(&file
->mutex
);
746 static long ib_umad_ioctl(struct file
*filp
, unsigned int cmd
,
750 case IB_USER_MAD_REGISTER_AGENT
:
751 return ib_umad_reg_agent(filp
->private_data
, (void __user
*) arg
, 0);
752 case IB_USER_MAD_UNREGISTER_AGENT
:
753 return ib_umad_unreg_agent(filp
->private_data
, (__u32 __user
*) arg
);
754 case IB_USER_MAD_ENABLE_PKEY
:
755 return ib_umad_enable_pkey(filp
->private_data
);
762 static long ib_umad_compat_ioctl(struct file
*filp
, unsigned int cmd
,
766 case IB_USER_MAD_REGISTER_AGENT
:
767 return ib_umad_reg_agent(filp
->private_data
, compat_ptr(arg
), 1);
768 case IB_USER_MAD_UNREGISTER_AGENT
:
769 return ib_umad_unreg_agent(filp
->private_data
, compat_ptr(arg
));
770 case IB_USER_MAD_ENABLE_PKEY
:
771 return ib_umad_enable_pkey(filp
->private_data
);
779 * ib_umad_open() does not need the BKL:
781 * - umad_port[] accesses are protected by port_lock, the
782 * ib_umad_port structures are properly reference counted, and
783 * everything else is purely local to the file being created, so
784 * races against other open calls are not a problem;
785 * - the ioctl method does not affect any global state outside of the
786 * file structure being operated on;
787 * - the port is added to umad_port[] as the last part of module
788 * initialization so the open method will either immediately run
789 * -ENXIO, or all required initialization will be done.
791 static int ib_umad_open(struct inode
*inode
, struct file
*filp
)
793 struct ib_umad_port
*port
;
794 struct ib_umad_file
*file
;
797 spin_lock(&port_lock
);
798 port
= umad_port
[iminor(inode
) - IB_UMAD_MINOR_BASE
];
800 kref_get(&port
->umad_dev
->ref
);
801 spin_unlock(&port_lock
);
806 mutex_lock(&port
->file_mutex
);
813 file
= kzalloc(sizeof *file
, GFP_KERNEL
);
815 kref_put(&port
->umad_dev
->ref
, ib_umad_release_dev
);
820 mutex_init(&file
->mutex
);
821 spin_lock_init(&file
->send_lock
);
822 INIT_LIST_HEAD(&file
->recv_list
);
823 INIT_LIST_HEAD(&file
->send_list
);
824 init_waitqueue_head(&file
->recv_wait
);
827 filp
->private_data
= file
;
829 list_add_tail(&file
->port_list
, &port
->file_list
);
832 mutex_unlock(&port
->file_mutex
);
836 static int ib_umad_close(struct inode
*inode
, struct file
*filp
)
838 struct ib_umad_file
*file
= filp
->private_data
;
839 struct ib_umad_device
*dev
= file
->port
->umad_dev
;
840 struct ib_umad_packet
*packet
, *tmp
;
844 mutex_lock(&file
->port
->file_mutex
);
845 mutex_lock(&file
->mutex
);
847 already_dead
= file
->agents_dead
;
848 file
->agents_dead
= 1;
850 list_for_each_entry_safe(packet
, tmp
, &file
->recv_list
, list
) {
852 ib_free_recv_mad(packet
->recv_wc
);
856 list_del(&file
->port_list
);
858 mutex_unlock(&file
->mutex
);
861 for (i
= 0; i
< IB_UMAD_MAX_AGENTS
; ++i
)
863 ib_unregister_mad_agent(file
->agent
[i
]);
865 mutex_unlock(&file
->port
->file_mutex
);
868 kref_put(&dev
->ref
, ib_umad_release_dev
);
873 static const struct file_operations umad_fops
= {
874 .owner
= THIS_MODULE
,
875 .read
= ib_umad_read
,
876 .write
= ib_umad_write
,
877 .poll
= ib_umad_poll
,
878 .unlocked_ioctl
= ib_umad_ioctl
,
880 .compat_ioctl
= ib_umad_compat_ioctl
,
882 .open
= ib_umad_open
,
883 .release
= ib_umad_close
886 static int ib_umad_sm_open(struct inode
*inode
, struct file
*filp
)
888 struct ib_umad_port
*port
;
889 struct ib_port_modify props
= {
890 .set_port_cap_mask
= IB_PORT_SM
894 spin_lock(&port_lock
);
895 port
= umad_port
[iminor(inode
) - IB_UMAD_MINOR_BASE
- IB_UMAD_MAX_PORTS
];
897 kref_get(&port
->umad_dev
->ref
);
898 spin_unlock(&port_lock
);
903 if (filp
->f_flags
& O_NONBLOCK
) {
904 if (down_trylock(&port
->sm_sem
)) {
909 if (down_interruptible(&port
->sm_sem
)) {
915 ret
= ib_modify_port(port
->ib_dev
, port
->port_num
, 0, &props
);
921 filp
->private_data
= port
;
926 kref_put(&port
->umad_dev
->ref
, ib_umad_release_dev
);
930 static int ib_umad_sm_close(struct inode
*inode
, struct file
*filp
)
932 struct ib_umad_port
*port
= filp
->private_data
;
933 struct ib_port_modify props
= {
934 .clr_port_cap_mask
= IB_PORT_SM
938 mutex_lock(&port
->file_mutex
);
940 ret
= ib_modify_port(port
->ib_dev
, port
->port_num
, 0, &props
);
941 mutex_unlock(&port
->file_mutex
);
945 kref_put(&port
->umad_dev
->ref
, ib_umad_release_dev
);
950 static const struct file_operations umad_sm_fops
= {
951 .owner
= THIS_MODULE
,
952 .open
= ib_umad_sm_open
,
953 .release
= ib_umad_sm_close
956 static struct ib_client umad_client
= {
958 .add
= ib_umad_add_one
,
959 .remove
= ib_umad_remove_one
962 static ssize_t
show_ibdev(struct device
*dev
, struct device_attribute
*attr
,
965 struct ib_umad_port
*port
= dev_get_drvdata(dev
);
970 return sprintf(buf
, "%s\n", port
->ib_dev
->name
);
972 static DEVICE_ATTR(ibdev
, S_IRUGO
, show_ibdev
, NULL
);
974 static ssize_t
show_port(struct device
*dev
, struct device_attribute
*attr
,
977 struct ib_umad_port
*port
= dev_get_drvdata(dev
);
982 return sprintf(buf
, "%d\n", port
->port_num
);
984 static DEVICE_ATTR(port
, S_IRUGO
, show_port
, NULL
);
986 static ssize_t
show_abi_version(struct class *class, char *buf
)
988 return sprintf(buf
, "%d\n", IB_USER_MAD_ABI_VERSION
);
990 static CLASS_ATTR(abi_version
, S_IRUGO
, show_abi_version
, NULL
);
992 static int ib_umad_init_port(struct ib_device
*device
, int port_num
,
993 struct ib_umad_port
*port
)
995 spin_lock(&port_lock
);
996 port
->dev_num
= find_first_zero_bit(dev_map
, IB_UMAD_MAX_PORTS
);
997 if (port
->dev_num
>= IB_UMAD_MAX_PORTS
) {
998 spin_unlock(&port_lock
);
1001 set_bit(port
->dev_num
, dev_map
);
1002 spin_unlock(&port_lock
);
1004 port
->ib_dev
= device
;
1005 port
->port_num
= port_num
;
1006 init_MUTEX(&port
->sm_sem
);
1007 mutex_init(&port
->file_mutex
);
1008 INIT_LIST_HEAD(&port
->file_list
);
1010 port
->cdev
= cdev_alloc();
1013 port
->cdev
->owner
= THIS_MODULE
;
1014 port
->cdev
->ops
= &umad_fops
;
1015 kobject_set_name(&port
->cdev
->kobj
, "umad%d", port
->dev_num
);
1016 if (cdev_add(port
->cdev
, base_dev
+ port
->dev_num
, 1))
1019 port
->dev
= device_create(umad_class
, device
->dma_device
,
1020 port
->cdev
->dev
, port
,
1021 "umad%d", port
->dev_num
);
1022 if (IS_ERR(port
->dev
))
1025 if (device_create_file(port
->dev
, &dev_attr_ibdev
))
1027 if (device_create_file(port
->dev
, &dev_attr_port
))
1030 port
->sm_cdev
= cdev_alloc();
1033 port
->sm_cdev
->owner
= THIS_MODULE
;
1034 port
->sm_cdev
->ops
= &umad_sm_fops
;
1035 kobject_set_name(&port
->sm_cdev
->kobj
, "issm%d", port
->dev_num
);
1036 if (cdev_add(port
->sm_cdev
, base_dev
+ port
->dev_num
+ IB_UMAD_MAX_PORTS
, 1))
1039 port
->sm_dev
= device_create(umad_class
, device
->dma_device
,
1040 port
->sm_cdev
->dev
, port
,
1041 "issm%d", port
->dev_num
);
1042 if (IS_ERR(port
->sm_dev
))
1045 if (device_create_file(port
->sm_dev
, &dev_attr_ibdev
))
1047 if (device_create_file(port
->sm_dev
, &dev_attr_port
))
1050 spin_lock(&port_lock
);
1051 umad_port
[port
->dev_num
] = port
;
1052 spin_unlock(&port_lock
);
1057 device_destroy(umad_class
, port
->sm_cdev
->dev
);
1060 cdev_del(port
->sm_cdev
);
1063 device_destroy(umad_class
, port
->cdev
->dev
);
1066 cdev_del(port
->cdev
);
1067 clear_bit(port
->dev_num
, dev_map
);
1072 static void ib_umad_kill_port(struct ib_umad_port
*port
)
1074 struct ib_umad_file
*file
;
1078 dev_set_drvdata(port
->dev
, NULL
);
1079 dev_set_drvdata(port
->sm_dev
, NULL
);
1081 device_destroy(umad_class
, port
->cdev
->dev
);
1082 device_destroy(umad_class
, port
->sm_cdev
->dev
);
1084 cdev_del(port
->cdev
);
1085 cdev_del(port
->sm_cdev
);
1087 spin_lock(&port_lock
);
1088 umad_port
[port
->dev_num
] = NULL
;
1089 spin_unlock(&port_lock
);
1091 mutex_lock(&port
->file_mutex
);
1093 port
->ib_dev
= NULL
;
1095 list_for_each_entry(file
, &port
->file_list
, port_list
) {
1096 mutex_lock(&file
->mutex
);
1097 already_dead
= file
->agents_dead
;
1098 file
->agents_dead
= 1;
1099 mutex_unlock(&file
->mutex
);
1101 for (id
= 0; id
< IB_UMAD_MAX_AGENTS
; ++id
)
1102 if (file
->agent
[id
])
1103 ib_unregister_mad_agent(file
->agent
[id
]);
1106 mutex_unlock(&port
->file_mutex
);
1108 clear_bit(port
->dev_num
, dev_map
);
1111 static void ib_umad_add_one(struct ib_device
*device
)
1113 struct ib_umad_device
*umad_dev
;
1116 if (rdma_node_get_transport(device
->node_type
) != RDMA_TRANSPORT_IB
)
1119 if (device
->node_type
== RDMA_NODE_IB_SWITCH
)
1123 e
= device
->phys_port_cnt
;
1126 umad_dev
= kzalloc(sizeof *umad_dev
+
1127 (e
- s
+ 1) * sizeof (struct ib_umad_port
),
1132 kref_init(&umad_dev
->ref
);
1134 umad_dev
->start_port
= s
;
1135 umad_dev
->end_port
= e
;
1137 for (i
= s
; i
<= e
; ++i
) {
1138 umad_dev
->port
[i
- s
].umad_dev
= umad_dev
;
1140 if (ib_umad_init_port(device
, i
, &umad_dev
->port
[i
- s
]))
1144 ib_set_client_data(device
, &umad_client
, umad_dev
);
1150 ib_umad_kill_port(&umad_dev
->port
[i
- s
]);
1152 kref_put(&umad_dev
->ref
, ib_umad_release_dev
);
1155 static void ib_umad_remove_one(struct ib_device
*device
)
1157 struct ib_umad_device
*umad_dev
= ib_get_client_data(device
, &umad_client
);
1163 for (i
= 0; i
<= umad_dev
->end_port
- umad_dev
->start_port
; ++i
)
1164 ib_umad_kill_port(&umad_dev
->port
[i
]);
1166 kref_put(&umad_dev
->ref
, ib_umad_release_dev
);
1169 static int __init
ib_umad_init(void)
1173 ret
= register_chrdev_region(base_dev
, IB_UMAD_MAX_PORTS
* 2,
1176 printk(KERN_ERR
"user_mad: couldn't register device number\n");
1180 umad_class
= class_create(THIS_MODULE
, "infiniband_mad");
1181 if (IS_ERR(umad_class
)) {
1182 ret
= PTR_ERR(umad_class
);
1183 printk(KERN_ERR
"user_mad: couldn't create class infiniband_mad\n");
1187 ret
= class_create_file(umad_class
, &class_attr_abi_version
);
1189 printk(KERN_ERR
"user_mad: couldn't create abi_version attribute\n");
1193 ret
= ib_register_client(&umad_client
);
1195 printk(KERN_ERR
"user_mad: couldn't register ib_umad client\n");
1202 class_destroy(umad_class
);
1205 unregister_chrdev_region(base_dev
, IB_UMAD_MAX_PORTS
* 2);
1211 static void __exit
ib_umad_cleanup(void)
1213 ib_unregister_client(&umad_client
);
1214 class_destroy(umad_class
);
1215 unregister_chrdev_region(base_dev
, IB_UMAD_MAX_PORTS
* 2);
1218 module_init(ib_umad_init
);
1219 module_exit(ib_umad_cleanup
);