2 * Copyright (c) 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
4 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5 * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
6 * Copyright (c) 2005 PathScale, Inc. All rights reserved.
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/device.h>
40 #include <linux/err.h>
42 #include <linux/poll.h>
43 #include <linux/sched.h>
44 #include <linux/file.h>
45 #include <linux/cdev.h>
46 #include <linux/anon_inodes.h>
47 #include <linux/slab.h>
49 #include <asm/uaccess.h>
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
55 MODULE_LICENSE("Dual BSD/GPL");
58 IB_UVERBS_MAJOR = 231,
59 IB_UVERBS_BASE_MINOR = 192,
60 IB_UVERBS_MAX_DEVICES = 32
63 #define IB_UVERBS_BASE_DEV MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
65 static struct class *uverbs_class;
67 DEFINE_SPINLOCK(ib_uverbs_idr_lock);
68 DEFINE_IDR(ib_uverbs_pd_idr);
69 DEFINE_IDR(ib_uverbs_mr_idr);
70 DEFINE_IDR(ib_uverbs_mw_idr);
71 DEFINE_IDR(ib_uverbs_ah_idr);
72 DEFINE_IDR(ib_uverbs_cq_idr);
73 DEFINE_IDR(ib_uverbs_qp_idr);
74 DEFINE_IDR(ib_uverbs_srq_idr);
75 DEFINE_IDR(ib_uverbs_xrcd_idr);
76 DEFINE_IDR(ib_uverbs_rule_idr);
78 static DEFINE_SPINLOCK(map_lock);
79 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
81 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
82 const char __user *buf, int in_len,
84 [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
85 [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
86 [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
87 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
88 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
89 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
90 [IB_USER_VERBS_CMD_REREG_MR] = ib_uverbs_rereg_mr,
91 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
92 [IB_USER_VERBS_CMD_ALLOC_MW] = ib_uverbs_alloc_mw,
93 [IB_USER_VERBS_CMD_DEALLOC_MW] = ib_uverbs_dealloc_mw,
94 [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
95 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
96 [IB_USER_VERBS_CMD_RESIZE_CQ] = ib_uverbs_resize_cq,
97 [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq,
98 [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq,
99 [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
100 [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
101 [IB_USER_VERBS_CMD_QUERY_QP] = ib_uverbs_query_qp,
102 [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
103 [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
104 [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send,
105 [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv,
106 [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv,
107 [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah,
108 [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah,
109 [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
110 [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
111 [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq,
112 [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
113 [IB_USER_VERBS_CMD_QUERY_SRQ] = ib_uverbs_query_srq,
114 [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
115 [IB_USER_VERBS_CMD_OPEN_XRCD] = ib_uverbs_open_xrcd,
116 [IB_USER_VERBS_CMD_CLOSE_XRCD] = ib_uverbs_close_xrcd,
117 [IB_USER_VERBS_CMD_CREATE_XSRQ] = ib_uverbs_create_xsrq,
118 [IB_USER_VERBS_CMD_OPEN_QP] = ib_uverbs_open_qp,
121 static int (*uverbs_ex_cmd_table[])(struct ib_uverbs_file *file,
122 struct ib_udata *ucore,
123 struct ib_udata *uhw) = {
124 [IB_USER_VERBS_EX_CMD_CREATE_FLOW] = ib_uverbs_ex_create_flow,
125 [IB_USER_VERBS_EX_CMD_DESTROY_FLOW] = ib_uverbs_ex_destroy_flow,
126 [IB_USER_VERBS_EX_CMD_QUERY_DEVICE] = ib_uverbs_ex_query_device,
129 static void ib_uverbs_add_one(struct ib_device *device);
130 static void ib_uverbs_remove_one(struct ib_device *device);
132 static void ib_uverbs_release_dev(struct kref *ref)
134 struct ib_uverbs_device *dev =
135 container_of(ref, struct ib_uverbs_device, ref);
137 complete(&dev->comp);
140 static void ib_uverbs_release_event_file(struct kref *ref)
142 struct ib_uverbs_event_file *file =
143 container_of(ref, struct ib_uverbs_event_file, ref);
148 void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
149 struct ib_uverbs_event_file *ev_file,
150 struct ib_ucq_object *uobj)
152 struct ib_uverbs_event *evt, *tmp;
155 spin_lock_irq(&ev_file->lock);
156 list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
157 list_del(&evt->list);
160 spin_unlock_irq(&ev_file->lock);
162 kref_put(&ev_file->ref, ib_uverbs_release_event_file);
165 spin_lock_irq(&file->async_file->lock);
166 list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
167 list_del(&evt->list);
170 spin_unlock_irq(&file->async_file->lock);
173 void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
174 struct ib_uevent_object *uobj)
176 struct ib_uverbs_event *evt, *tmp;
178 spin_lock_irq(&file->async_file->lock);
179 list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
180 list_del(&evt->list);
183 spin_unlock_irq(&file->async_file->lock);
186 static void ib_uverbs_detach_umcast(struct ib_qp *qp,
187 struct ib_uqp_object *uobj)
189 struct ib_uverbs_mcast_entry *mcast, *tmp;
191 list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
192 ib_detach_mcast(qp, &mcast->gid, mcast->lid);
193 list_del(&mcast->list);
198 static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
199 struct ib_ucontext *context)
201 struct ib_uobject *uobj, *tmp;
206 context->closing = 1;
208 list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
209 struct ib_ah *ah = uobj->object;
211 idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
216 /* Remove MWs before QPs, in order to support type 2A MWs. */
217 list_for_each_entry_safe(uobj, tmp, &context->mw_list, list) {
218 struct ib_mw *mw = uobj->object;
220 idr_remove_uobj(&ib_uverbs_mw_idr, uobj);
225 list_for_each_entry_safe(uobj, tmp, &context->rule_list, list) {
226 struct ib_flow *flow_id = uobj->object;
228 idr_remove_uobj(&ib_uverbs_rule_idr, uobj);
229 ib_destroy_flow(flow_id);
233 list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
234 struct ib_qp *qp = uobj->object;
235 struct ib_uqp_object *uqp =
236 container_of(uobj, struct ib_uqp_object, uevent.uobject);
238 idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
239 if (qp != qp->real_qp) {
242 ib_uverbs_detach_umcast(qp, uqp);
245 ib_uverbs_release_uevent(file, &uqp->uevent);
249 list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
250 struct ib_srq *srq = uobj->object;
251 struct ib_uevent_object *uevent =
252 container_of(uobj, struct ib_uevent_object, uobject);
254 idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
256 ib_uverbs_release_uevent(file, uevent);
260 list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
261 struct ib_cq *cq = uobj->object;
262 struct ib_uverbs_event_file *ev_file = cq->cq_context;
263 struct ib_ucq_object *ucq =
264 container_of(uobj, struct ib_ucq_object, uobject);
266 idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
268 ib_uverbs_release_ucq(file, ev_file, ucq);
272 list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
273 struct ib_mr *mr = uobj->object;
275 idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
280 mutex_lock(&file->device->xrcd_tree_mutex);
281 list_for_each_entry_safe(uobj, tmp, &context->xrcd_list, list) {
282 struct ib_xrcd *xrcd = uobj->object;
283 struct ib_uxrcd_object *uxrcd =
284 container_of(uobj, struct ib_uxrcd_object, uobject);
286 idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
287 ib_uverbs_dealloc_xrcd(file->device, xrcd);
290 mutex_unlock(&file->device->xrcd_tree_mutex);
292 list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
293 struct ib_pd *pd = uobj->object;
295 idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
300 put_pid(context->tgid);
302 return context->device->dealloc_ucontext(context);
305 static void ib_uverbs_release_file(struct kref *ref)
307 struct ib_uverbs_file *file =
308 container_of(ref, struct ib_uverbs_file, ref);
310 module_put(file->device->ib_dev->owner);
311 kref_put(&file->device->ref, ib_uverbs_release_dev);
316 static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
317 size_t count, loff_t *pos)
319 struct ib_uverbs_event_file *file = filp->private_data;
320 struct ib_uverbs_event *event;
324 spin_lock_irq(&file->lock);
326 while (list_empty(&file->event_list)) {
327 spin_unlock_irq(&file->lock);
329 if (filp->f_flags & O_NONBLOCK)
332 if (wait_event_interruptible(file->poll_wait,
333 !list_empty(&file->event_list)))
336 spin_lock_irq(&file->lock);
339 event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
342 eventsz = sizeof (struct ib_uverbs_async_event_desc);
344 eventsz = sizeof (struct ib_uverbs_comp_event_desc);
346 if (eventsz > count) {
350 list_del(file->event_list.next);
351 if (event->counter) {
353 list_del(&event->obj_list);
357 spin_unlock_irq(&file->lock);
360 if (copy_to_user(buf, event, eventsz))
371 static unsigned int ib_uverbs_event_poll(struct file *filp,
372 struct poll_table_struct *wait)
374 unsigned int pollflags = 0;
375 struct ib_uverbs_event_file *file = filp->private_data;
377 poll_wait(filp, &file->poll_wait, wait);
379 spin_lock_irq(&file->lock);
380 if (!list_empty(&file->event_list))
381 pollflags = POLLIN | POLLRDNORM;
382 spin_unlock_irq(&file->lock);
387 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
389 struct ib_uverbs_event_file *file = filp->private_data;
391 return fasync_helper(fd, filp, on, &file->async_queue);
394 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
396 struct ib_uverbs_event_file *file = filp->private_data;
397 struct ib_uverbs_event *entry, *tmp;
399 spin_lock_irq(&file->lock);
401 list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
403 list_del(&entry->obj_list);
406 spin_unlock_irq(&file->lock);
408 if (file->is_async) {
409 ib_unregister_event_handler(&file->uverbs_file->event_handler);
410 kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
412 kref_put(&file->ref, ib_uverbs_release_event_file);
417 static const struct file_operations uverbs_event_fops = {
418 .owner = THIS_MODULE,
419 .read = ib_uverbs_event_read,
420 .poll = ib_uverbs_event_poll,
421 .release = ib_uverbs_event_close,
422 .fasync = ib_uverbs_event_fasync,
426 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
428 struct ib_uverbs_event_file *file = cq_context;
429 struct ib_ucq_object *uobj;
430 struct ib_uverbs_event *entry;
436 spin_lock_irqsave(&file->lock, flags);
437 if (file->is_closed) {
438 spin_unlock_irqrestore(&file->lock, flags);
442 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
444 spin_unlock_irqrestore(&file->lock, flags);
448 uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
450 entry->desc.comp.cq_handle = cq->uobject->user_handle;
451 entry->counter = &uobj->comp_events_reported;
453 list_add_tail(&entry->list, &file->event_list);
454 list_add_tail(&entry->obj_list, &uobj->comp_list);
455 spin_unlock_irqrestore(&file->lock, flags);
457 wake_up_interruptible(&file->poll_wait);
458 kill_fasync(&file->async_queue, SIGIO, POLL_IN);
461 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
462 __u64 element, __u64 event,
463 struct list_head *obj_list,
466 struct ib_uverbs_event *entry;
469 spin_lock_irqsave(&file->async_file->lock, flags);
470 if (file->async_file->is_closed) {
471 spin_unlock_irqrestore(&file->async_file->lock, flags);
475 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
477 spin_unlock_irqrestore(&file->async_file->lock, flags);
481 entry->desc.async.element = element;
482 entry->desc.async.event_type = event;
483 entry->desc.async.reserved = 0;
484 entry->counter = counter;
486 list_add_tail(&entry->list, &file->async_file->event_list);
488 list_add_tail(&entry->obj_list, obj_list);
489 spin_unlock_irqrestore(&file->async_file->lock, flags);
491 wake_up_interruptible(&file->async_file->poll_wait);
492 kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
495 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
497 struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
498 struct ib_ucq_object, uobject);
500 ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
501 event->event, &uobj->async_list,
502 &uobj->async_events_reported);
505 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
507 struct ib_uevent_object *uobj;
509 /* for XRC target qp's, check that qp is live */
510 if (!event->element.qp->uobject || !event->element.qp->uobject->live)
513 uobj = container_of(event->element.qp->uobject,
514 struct ib_uevent_object, uobject);
516 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
517 event->event, &uobj->event_list,
518 &uobj->events_reported);
521 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
523 struct ib_uevent_object *uobj;
525 uobj = container_of(event->element.srq->uobject,
526 struct ib_uevent_object, uobject);
528 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
529 event->event, &uobj->event_list,
530 &uobj->events_reported);
533 void ib_uverbs_event_handler(struct ib_event_handler *handler,
534 struct ib_event *event)
536 struct ib_uverbs_file *file =
537 container_of(handler, struct ib_uverbs_file, event_handler);
539 ib_uverbs_async_handler(file, event->element.port_num, event->event,
543 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
546 struct ib_uverbs_event_file *ev_file;
549 ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
551 return ERR_PTR(-ENOMEM);
553 kref_init(&ev_file->ref);
554 spin_lock_init(&ev_file->lock);
555 INIT_LIST_HEAD(&ev_file->event_list);
556 init_waitqueue_head(&ev_file->poll_wait);
557 ev_file->uverbs_file = uverbs_file;
558 ev_file->async_queue = NULL;
559 ev_file->is_async = is_async;
560 ev_file->is_closed = 0;
562 filp = anon_inode_getfile("[infinibandevent]", &uverbs_event_fops,
571 * Look up a completion event file by FD. If lookup is successful,
572 * takes a ref to the event file struct that it returns; if
573 * unsuccessful, returns NULL.
575 struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
577 struct ib_uverbs_event_file *ev_file = NULL;
578 struct fd f = fdget(fd);
583 if (f.file->f_op != &uverbs_event_fops)
586 ev_file = f.file->private_data;
587 if (ev_file->is_async) {
592 kref_get(&ev_file->ref);
599 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
600 size_t count, loff_t *pos)
602 struct ib_uverbs_file *file = filp->private_data;
603 struct ib_uverbs_cmd_hdr hdr;
606 if (count < sizeof hdr)
609 if (copy_from_user(&hdr, buf, sizeof hdr))
612 flags = (hdr.command &
613 IB_USER_VERBS_CMD_FLAGS_MASK) >> IB_USER_VERBS_CMD_FLAGS_SHIFT;
618 if (hdr.command & ~(__u32)(IB_USER_VERBS_CMD_FLAGS_MASK |
619 IB_USER_VERBS_CMD_COMMAND_MASK))
622 command = hdr.command & IB_USER_VERBS_CMD_COMMAND_MASK;
624 if (command >= ARRAY_SIZE(uverbs_cmd_table) ||
625 !uverbs_cmd_table[command])
628 if (!file->ucontext &&
629 command != IB_USER_VERBS_CMD_GET_CONTEXT)
632 if (!(file->device->ib_dev->uverbs_cmd_mask & (1ull << command)))
635 if (hdr.in_words * 4 != count)
638 return uverbs_cmd_table[command](file,
643 } else if (flags == IB_USER_VERBS_CMD_FLAG_EXTENDED) {
646 struct ib_uverbs_ex_cmd_hdr ex_hdr;
647 struct ib_udata ucore;
650 size_t written_count = count;
652 if (hdr.command & ~(__u32)(IB_USER_VERBS_CMD_FLAGS_MASK |
653 IB_USER_VERBS_CMD_COMMAND_MASK))
656 command = hdr.command & IB_USER_VERBS_CMD_COMMAND_MASK;
658 if (command >= ARRAY_SIZE(uverbs_ex_cmd_table) ||
659 !uverbs_ex_cmd_table[command])
665 if (!(file->device->ib_dev->uverbs_ex_cmd_mask & (1ull << command)))
668 if (count < (sizeof(hdr) + sizeof(ex_hdr)))
671 if (copy_from_user(&ex_hdr, buf + sizeof(hdr), sizeof(ex_hdr)))
674 count -= sizeof(hdr) + sizeof(ex_hdr);
675 buf += sizeof(hdr) + sizeof(ex_hdr);
677 if ((hdr.in_words + ex_hdr.provider_in_words) * 8 != count)
680 if (ex_hdr.cmd_hdr_reserved)
683 if (ex_hdr.response) {
684 if (!hdr.out_words && !ex_hdr.provider_out_words)
687 if (!access_ok(VERIFY_WRITE,
688 (void __user *) (unsigned long) ex_hdr.response,
689 (hdr.out_words + ex_hdr.provider_out_words) * 8))
692 if (hdr.out_words || ex_hdr.provider_out_words)
696 INIT_UDATA_BUF_OR_NULL(&ucore, buf, (unsigned long) ex_hdr.response,
697 hdr.in_words * 8, hdr.out_words * 8);
699 INIT_UDATA_BUF_OR_NULL(&uhw,
701 (unsigned long) ex_hdr.response + ucore.outlen,
702 ex_hdr.provider_in_words * 8,
703 ex_hdr.provider_out_words * 8);
705 err = uverbs_ex_cmd_table[command](file,
712 return written_count;
718 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
720 struct ib_uverbs_file *file = filp->private_data;
725 return file->device->ib_dev->mmap(file->ucontext, vma);
729 * ib_uverbs_open() does not need the BKL:
731 * - the ib_uverbs_device structures are properly reference counted and
732 * everything else is purely local to the file being created, so
733 * races against other open calls are not a problem;
734 * - there is no ioctl method to race against;
735 * - the open method will either immediately run -ENXIO, or all
736 * required initialization will be done.
738 static int ib_uverbs_open(struct inode *inode, struct file *filp)
740 struct ib_uverbs_device *dev;
741 struct ib_uverbs_file *file;
744 dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev);
750 if (!try_module_get(dev->ib_dev->owner)) {
755 file = kmalloc(sizeof *file, GFP_KERNEL);
762 file->ucontext = NULL;
763 file->async_file = NULL;
764 kref_init(&file->ref);
765 mutex_init(&file->mutex);
767 filp->private_data = file;
769 return nonseekable_open(inode, filp);
772 module_put(dev->ib_dev->owner);
775 kref_put(&dev->ref, ib_uverbs_release_dev);
779 static int ib_uverbs_close(struct inode *inode, struct file *filp)
781 struct ib_uverbs_file *file = filp->private_data;
783 ib_uverbs_cleanup_ucontext(file, file->ucontext);
785 if (file->async_file)
786 kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
788 kref_put(&file->ref, ib_uverbs_release_file);
793 static const struct file_operations uverbs_fops = {
794 .owner = THIS_MODULE,
795 .write = ib_uverbs_write,
796 .open = ib_uverbs_open,
797 .release = ib_uverbs_close,
801 static const struct file_operations uverbs_mmap_fops = {
802 .owner = THIS_MODULE,
803 .write = ib_uverbs_write,
804 .mmap = ib_uverbs_mmap,
805 .open = ib_uverbs_open,
806 .release = ib_uverbs_close,
810 static struct ib_client uverbs_client = {
812 .add = ib_uverbs_add_one,
813 .remove = ib_uverbs_remove_one
816 static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
819 struct ib_uverbs_device *dev = dev_get_drvdata(device);
824 return sprintf(buf, "%s\n", dev->ib_dev->name);
826 static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
828 static ssize_t show_dev_abi_version(struct device *device,
829 struct device_attribute *attr, char *buf)
831 struct ib_uverbs_device *dev = dev_get_drvdata(device);
836 return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
838 static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
840 static CLASS_ATTR_STRING(abi_version, S_IRUGO,
841 __stringify(IB_USER_VERBS_ABI_VERSION));
843 static dev_t overflow_maj;
844 static DECLARE_BITMAP(overflow_map, IB_UVERBS_MAX_DEVICES);
847 * If we have more than IB_UVERBS_MAX_DEVICES, dynamically overflow by
848 * requesting a new major number and doubling the number of max devices we
849 * support. It's stupid, but simple.
851 static int find_overflow_devnum(void)
856 ret = alloc_chrdev_region(&overflow_maj, 0, IB_UVERBS_MAX_DEVICES,
859 printk(KERN_ERR "user_verbs: couldn't register dynamic device number\n");
864 ret = find_first_zero_bit(overflow_map, IB_UVERBS_MAX_DEVICES);
865 if (ret >= IB_UVERBS_MAX_DEVICES)
871 static void ib_uverbs_add_one(struct ib_device *device)
875 struct ib_uverbs_device *uverbs_dev;
877 if (!device->alloc_ucontext)
880 uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
884 kref_init(&uverbs_dev->ref);
885 init_completion(&uverbs_dev->comp);
886 uverbs_dev->xrcd_tree = RB_ROOT;
887 mutex_init(&uverbs_dev->xrcd_tree_mutex);
889 spin_lock(&map_lock);
890 devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
891 if (devnum >= IB_UVERBS_MAX_DEVICES) {
892 spin_unlock(&map_lock);
893 devnum = find_overflow_devnum();
897 spin_lock(&map_lock);
898 uverbs_dev->devnum = devnum + IB_UVERBS_MAX_DEVICES;
899 base = devnum + overflow_maj;
900 set_bit(devnum, overflow_map);
902 uverbs_dev->devnum = devnum;
903 base = devnum + IB_UVERBS_BASE_DEV;
904 set_bit(devnum, dev_map);
906 spin_unlock(&map_lock);
908 uverbs_dev->ib_dev = device;
909 uverbs_dev->num_comp_vectors = device->num_comp_vectors;
911 cdev_init(&uverbs_dev->cdev, NULL);
912 uverbs_dev->cdev.owner = THIS_MODULE;
913 uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
914 kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum);
915 if (cdev_add(&uverbs_dev->cdev, base, 1))
918 uverbs_dev->dev = device_create(uverbs_class, device->dma_device,
919 uverbs_dev->cdev.dev, uverbs_dev,
920 "uverbs%d", uverbs_dev->devnum);
921 if (IS_ERR(uverbs_dev->dev))
924 if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
926 if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
929 ib_set_client_data(device, &uverbs_client, uverbs_dev);
934 device_destroy(uverbs_class, uverbs_dev->cdev.dev);
937 cdev_del(&uverbs_dev->cdev);
938 if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
939 clear_bit(devnum, dev_map);
941 clear_bit(devnum, overflow_map);
944 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
945 wait_for_completion(&uverbs_dev->comp);
950 static void ib_uverbs_remove_one(struct ib_device *device)
952 struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
957 dev_set_drvdata(uverbs_dev->dev, NULL);
958 device_destroy(uverbs_class, uverbs_dev->cdev.dev);
959 cdev_del(&uverbs_dev->cdev);
961 if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
962 clear_bit(uverbs_dev->devnum, dev_map);
964 clear_bit(uverbs_dev->devnum - IB_UVERBS_MAX_DEVICES, overflow_map);
966 kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
967 wait_for_completion(&uverbs_dev->comp);
971 static char *uverbs_devnode(struct device *dev, umode_t *mode)
975 return kasprintf(GFP_KERNEL, "infiniband/%s", dev_name(dev));
978 static int __init ib_uverbs_init(void)
982 ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
985 printk(KERN_ERR "user_verbs: couldn't register device number\n");
989 uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
990 if (IS_ERR(uverbs_class)) {
991 ret = PTR_ERR(uverbs_class);
992 printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
996 uverbs_class->devnode = uverbs_devnode;
998 ret = class_create_file(uverbs_class, &class_attr_abi_version.attr);
1000 printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
1004 ret = ib_register_client(&uverbs_client);
1006 printk(KERN_ERR "user_verbs: couldn't register client\n");
1013 class_destroy(uverbs_class);
1016 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
1022 static void __exit ib_uverbs_cleanup(void)
1024 ib_unregister_client(&uverbs_client);
1025 class_destroy(uverbs_class);
1026 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
1028 unregister_chrdev_region(overflow_maj, IB_UVERBS_MAX_DEVICES);
1029 idr_destroy(&ib_uverbs_pd_idr);
1030 idr_destroy(&ib_uverbs_mr_idr);
1031 idr_destroy(&ib_uverbs_mw_idr);
1032 idr_destroy(&ib_uverbs_ah_idr);
1033 idr_destroy(&ib_uverbs_cq_idr);
1034 idr_destroy(&ib_uverbs_qp_idr);
1035 idr_destroy(&ib_uverbs_srq_idr);
1038 module_init(ib_uverbs_init);
1039 module_exit(ib_uverbs_cleanup);