2 * Copyright (c) 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Cisco Systems. All rights reserved.
4 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5 * Copyright (c) 2005 Voltaire, Inc. 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
35 * $Id: uverbs_main.c 2733 2005-06-28 19:14:34Z roland $
38 #include <linux/module.h>
39 #include <linux/init.h>
40 #include <linux/device.h>
41 #include <linux/err.h>
43 #include <linux/poll.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
47 #include <asm/uaccess.h>
51 MODULE_AUTHOR("Roland Dreier");
52 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
53 MODULE_LICENSE("Dual BSD/GPL");
55 #define INFINIBANDEVENTFS_MAGIC 0x49426576 /* "IBev" */
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 DECLARE_MUTEX(ib_uverbs_idr_mutex);
66 DEFINE_IDR(ib_uverbs_pd_idr);
67 DEFINE_IDR(ib_uverbs_mr_idr);
68 DEFINE_IDR(ib_uverbs_mw_idr);
69 DEFINE_IDR(ib_uverbs_ah_idr);
70 DEFINE_IDR(ib_uverbs_cq_idr);
71 DEFINE_IDR(ib_uverbs_qp_idr);
72 DEFINE_IDR(ib_uverbs_srq_idr);
74 static spinlock_t map_lock;
75 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
77 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
78 const char __user *buf, int in_len,
80 [IB_USER_VERBS_CMD_QUERY_PARAMS] = ib_uverbs_query_params,
81 [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
82 [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
83 [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
84 [IB_USER_VERBS_CMD_QUERY_GID] = ib_uverbs_query_gid,
85 [IB_USER_VERBS_CMD_QUERY_PKEY] = ib_uverbs_query_pkey,
86 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
87 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
88 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
89 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
90 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
91 [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
92 [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
93 [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
94 [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
95 [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
96 [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
97 [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq,
98 [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
99 [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
102 static struct vfsmount *uverbs_event_mnt;
104 static void ib_uverbs_add_one(struct ib_device *device);
105 static void ib_uverbs_remove_one(struct ib_device *device);
107 static int ib_dealloc_ucontext(struct ib_ucontext *context)
109 struct ib_uobject *uobj, *tmp;
114 down(&ib_uverbs_idr_mutex);
118 list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
119 struct ib_qp *qp = idr_find(&ib_uverbs_qp_idr, uobj->id);
120 idr_remove(&ib_uverbs_qp_idr, uobj->id);
122 list_del(&uobj->list);
123 kfree(container_of(uobj, struct ib_uevent_object, uobject));
126 list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
127 struct ib_cq *cq = idr_find(&ib_uverbs_cq_idr, uobj->id);
128 idr_remove(&ib_uverbs_cq_idr, uobj->id);
130 list_del(&uobj->list);
131 kfree(container_of(uobj, struct ib_ucq_object, uobject));
134 list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
135 struct ib_srq *srq = idr_find(&ib_uverbs_srq_idr, uobj->id);
136 idr_remove(&ib_uverbs_srq_idr, uobj->id);
138 list_del(&uobj->list);
139 kfree(container_of(uobj, struct ib_uevent_object, uobject));
144 list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
145 struct ib_mr *mr = idr_find(&ib_uverbs_mr_idr, uobj->id);
146 struct ib_device *mrdev = mr->device;
147 struct ib_umem_object *memobj;
149 idr_remove(&ib_uverbs_mr_idr, uobj->id);
152 memobj = container_of(uobj, struct ib_umem_object, uobject);
153 ib_umem_release_on_close(mrdev, &memobj->umem);
155 list_del(&uobj->list);
159 list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
160 struct ib_pd *pd = idr_find(&ib_uverbs_pd_idr, uobj->id);
161 idr_remove(&ib_uverbs_pd_idr, uobj->id);
163 list_del(&uobj->list);
167 up(&ib_uverbs_idr_mutex);
169 return context->device->dealloc_ucontext(context);
172 static void ib_uverbs_release_file(struct kref *ref)
174 struct ib_uverbs_file *file =
175 container_of(ref, struct ib_uverbs_file, ref);
177 module_put(file->device->ib_dev->owner);
181 static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
182 size_t count, loff_t *pos)
184 struct ib_uverbs_event_file *file = filp->private_data;
185 struct ib_uverbs_event *event;
189 spin_lock_irq(&file->lock);
191 while (list_empty(&file->event_list) && file->fd >= 0) {
192 spin_unlock_irq(&file->lock);
194 if (filp->f_flags & O_NONBLOCK)
197 if (wait_event_interruptible(file->poll_wait,
198 !list_empty(&file->event_list) ||
202 spin_lock_irq(&file->lock);
206 spin_unlock_irq(&file->lock);
210 event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
213 eventsz = sizeof (struct ib_uverbs_async_event_desc);
215 eventsz = sizeof (struct ib_uverbs_comp_event_desc);
217 if (eventsz > count) {
221 list_del(file->event_list.next);
222 if (event->counter) {
224 list_del(&event->obj_list);
228 spin_unlock_irq(&file->lock);
231 if (copy_to_user(buf, event, eventsz))
242 static unsigned int ib_uverbs_event_poll(struct file *filp,
243 struct poll_table_struct *wait)
245 unsigned int pollflags = 0;
246 struct ib_uverbs_event_file *file = filp->private_data;
248 poll_wait(filp, &file->poll_wait, wait);
250 spin_lock_irq(&file->lock);
253 else if (!list_empty(&file->event_list))
254 pollflags = POLLIN | POLLRDNORM;
255 spin_unlock_irq(&file->lock);
260 static void ib_uverbs_event_release(struct ib_uverbs_event_file *file)
262 struct ib_uverbs_event *entry, *tmp;
264 spin_lock_irq(&file->lock);
265 if (file->fd != -1) {
267 list_for_each_entry_safe(entry, tmp, &file->event_list, list)
270 spin_unlock_irq(&file->lock);
273 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
275 struct ib_uverbs_event_file *file = filp->private_data;
277 return fasync_helper(fd, filp, on, &file->async_queue);
280 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
282 struct ib_uverbs_event_file *file = filp->private_data;
284 ib_uverbs_event_release(file);
285 ib_uverbs_event_fasync(-1, filp, 0);
286 kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
291 static struct file_operations uverbs_event_fops = {
293 * No .owner field since we artificially create event files,
294 * so there is no increment to the module reference count in
295 * the open path. All event files come from a uverbs command
296 * file, which already takes a module reference, so this is OK.
298 .read = ib_uverbs_event_read,
299 .poll = ib_uverbs_event_poll,
300 .release = ib_uverbs_event_close,
301 .fasync = ib_uverbs_event_fasync
304 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
306 struct ib_uverbs_file *file = cq_context;
307 struct ib_ucq_object *uobj;
308 struct ib_uverbs_event *entry;
311 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
315 uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
317 entry->desc.comp.cq_handle = cq->uobject->user_handle;
318 entry->counter = &uobj->comp_events_reported;
320 spin_lock_irqsave(&file->comp_file[0].lock, flags);
321 list_add_tail(&entry->list, &file->comp_file[0].event_list);
322 list_add_tail(&entry->obj_list, &uobj->comp_list);
323 spin_unlock_irqrestore(&file->comp_file[0].lock, flags);
325 wake_up_interruptible(&file->comp_file[0].poll_wait);
326 kill_fasync(&file->comp_file[0].async_queue, SIGIO, POLL_IN);
329 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
330 __u64 element, __u64 event,
331 struct list_head *obj_list,
334 struct ib_uverbs_event *entry;
337 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
341 entry->desc.async.element = element;
342 entry->desc.async.event_type = event;
343 entry->counter = counter;
345 spin_lock_irqsave(&file->async_file.lock, flags);
346 list_add_tail(&entry->list, &file->async_file.event_list);
348 list_add_tail(&entry->obj_list, obj_list);
349 spin_unlock_irqrestore(&file->async_file.lock, flags);
351 wake_up_interruptible(&file->async_file.poll_wait);
352 kill_fasync(&file->async_file.async_queue, SIGIO, POLL_IN);
355 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
357 struct ib_ucq_object *uobj;
359 uobj = container_of(event->element.cq->uobject,
360 struct ib_ucq_object, uobject);
362 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
363 event->event, &uobj->async_list,
364 &uobj->async_events_reported);
368 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
370 struct ib_uevent_object *uobj;
372 uobj = container_of(event->element.qp->uobject,
373 struct ib_uevent_object, uobject);
375 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
376 event->event, &uobj->event_list,
377 &uobj->events_reported);
380 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr)
382 struct ib_uevent_object *uobj;
384 uobj = container_of(event->element.srq->uobject,
385 struct ib_uevent_object, uobject);
387 ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
388 event->event, &uobj->event_list,
389 &uobj->events_reported);
392 static void ib_uverbs_event_handler(struct ib_event_handler *handler,
393 struct ib_event *event)
395 struct ib_uverbs_file *file =
396 container_of(handler, struct ib_uverbs_file, event_handler);
398 ib_uverbs_async_handler(file, event->element.port_num, event->event,
402 static int ib_uverbs_event_init(struct ib_uverbs_event_file *file,
403 struct ib_uverbs_file *uverbs_file)
407 spin_lock_init(&file->lock);
408 INIT_LIST_HEAD(&file->event_list);
409 init_waitqueue_head(&file->poll_wait);
410 file->uverbs_file = uverbs_file;
411 file->async_queue = NULL;
413 file->fd = get_unused_fd();
417 filp = get_empty_filp();
419 put_unused_fd(file->fd);
423 filp->f_op = &uverbs_event_fops;
424 filp->f_vfsmnt = mntget(uverbs_event_mnt);
425 filp->f_dentry = dget(uverbs_event_mnt->mnt_root);
426 filp->f_mapping = filp->f_dentry->d_inode->i_mapping;
427 filp->f_flags = O_RDONLY;
428 filp->f_mode = FMODE_READ;
429 filp->private_data = file;
431 fd_install(file->fd, filp);
436 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
437 size_t count, loff_t *pos)
439 struct ib_uverbs_file *file = filp->private_data;
440 struct ib_uverbs_cmd_hdr hdr;
442 if (count < sizeof hdr)
445 if (copy_from_user(&hdr, buf, sizeof hdr))
448 if (hdr.in_words * 4 != count)
451 if (hdr.command < 0 ||
452 hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
453 !uverbs_cmd_table[hdr.command])
456 if (!file->ucontext &&
457 hdr.command != IB_USER_VERBS_CMD_QUERY_PARAMS &&
458 hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
461 return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
462 hdr.in_words * 4, hdr.out_words * 4);
465 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
467 struct ib_uverbs_file *file = filp->private_data;
472 return file->device->ib_dev->mmap(file->ucontext, vma);
475 static int ib_uverbs_open(struct inode *inode, struct file *filp)
477 struct ib_uverbs_device *dev =
478 container_of(inode->i_cdev, struct ib_uverbs_device, dev);
479 struct ib_uverbs_file *file;
483 if (!try_module_get(dev->ib_dev->owner))
486 file = kmalloc(sizeof *file +
487 (dev->num_comp - 1) * sizeof (struct ib_uverbs_event_file),
495 kref_init(&file->ref);
496 init_MUTEX(&file->mutex);
498 file->ucontext = NULL;
500 kref_get(&file->ref);
501 ret = ib_uverbs_event_init(&file->async_file, file);
505 file->async_file.is_async = 1;
507 for (i = 0; i < dev->num_comp; ++i) {
508 kref_get(&file->ref);
509 ret = ib_uverbs_event_init(&file->comp_file[i], file);
512 file->comp_file[i].is_async = 0;
516 filp->private_data = file;
518 INIT_IB_EVENT_HANDLER(&file->event_handler, dev->ib_dev,
519 ib_uverbs_event_handler);
520 if (ib_register_event_handler(&file->event_handler))
527 ib_uverbs_event_release(&file->comp_file[i]);
529 ib_uverbs_event_release(&file->async_file);
533 * One extra kref_put() because we took a reference before the
534 * event file creation that failed and got us here.
536 kref_put(&file->ref, ib_uverbs_release_file);
537 kref_put(&file->ref, ib_uverbs_release_file);
540 module_put(dev->ib_dev->owner);
544 static int ib_uverbs_close(struct inode *inode, struct file *filp)
546 struct ib_uverbs_file *file = filp->private_data;
549 ib_unregister_event_handler(&file->event_handler);
550 ib_uverbs_event_release(&file->async_file);
551 ib_dealloc_ucontext(file->ucontext);
553 for (i = 0; i < file->device->num_comp; ++i)
554 ib_uverbs_event_release(&file->comp_file[i]);
556 kref_put(&file->ref, ib_uverbs_release_file);
561 static struct file_operations uverbs_fops = {
562 .owner = THIS_MODULE,
563 .write = ib_uverbs_write,
564 .open = ib_uverbs_open,
565 .release = ib_uverbs_close
568 static struct file_operations uverbs_mmap_fops = {
569 .owner = THIS_MODULE,
570 .write = ib_uverbs_write,
571 .mmap = ib_uverbs_mmap,
572 .open = ib_uverbs_open,
573 .release = ib_uverbs_close
576 static struct ib_client uverbs_client = {
578 .add = ib_uverbs_add_one,
579 .remove = ib_uverbs_remove_one
582 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
584 struct ib_uverbs_device *dev =
585 container_of(class_dev, struct ib_uverbs_device, class_dev);
587 return sprintf(buf, "%s\n", dev->ib_dev->name);
589 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
591 static void ib_uverbs_release_class_dev(struct class_device *class_dev)
593 struct ib_uverbs_device *dev =
594 container_of(class_dev, struct ib_uverbs_device, class_dev);
597 clear_bit(dev->devnum, dev_map);
601 static struct class uverbs_class = {
602 .name = "infiniband_verbs",
603 .release = ib_uverbs_release_class_dev
606 static ssize_t show_abi_version(struct class *class, char *buf)
608 return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
610 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
612 static void ib_uverbs_add_one(struct ib_device *device)
614 struct ib_uverbs_device *uverbs_dev;
616 if (!device->alloc_ucontext)
619 uverbs_dev = kmalloc(sizeof *uverbs_dev, GFP_KERNEL);
623 memset(uverbs_dev, 0, sizeof *uverbs_dev);
625 spin_lock(&map_lock);
626 uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
627 if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
628 spin_unlock(&map_lock);
631 set_bit(uverbs_dev->devnum, dev_map);
632 spin_unlock(&map_lock);
634 uverbs_dev->ib_dev = device;
635 uverbs_dev->num_comp = 1;
638 cdev_init(&uverbs_dev->dev, &uverbs_mmap_fops);
640 cdev_init(&uverbs_dev->dev, &uverbs_fops);
641 uverbs_dev->dev.owner = THIS_MODULE;
642 kobject_set_name(&uverbs_dev->dev.kobj, "uverbs%d", uverbs_dev->devnum);
643 if (cdev_add(&uverbs_dev->dev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
646 uverbs_dev->class_dev.class = &uverbs_class;
647 uverbs_dev->class_dev.dev = device->dma_device;
648 uverbs_dev->class_dev.devt = uverbs_dev->dev.dev;
649 snprintf(uverbs_dev->class_dev.class_id, BUS_ID_SIZE, "uverbs%d", uverbs_dev->devnum);
650 if (class_device_register(&uverbs_dev->class_dev))
653 if (class_device_create_file(&uverbs_dev->class_dev, &class_device_attr_ibdev))
656 ib_set_client_data(device, &uverbs_client, uverbs_dev);
661 class_device_unregister(&uverbs_dev->class_dev);
664 cdev_del(&uverbs_dev->dev);
665 clear_bit(uverbs_dev->devnum, dev_map);
672 static void ib_uverbs_remove_one(struct ib_device *device)
674 struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
679 class_device_unregister(&uverbs_dev->class_dev);
682 static struct super_block *uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
683 const char *dev_name, void *data)
685 return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
686 INFINIBANDEVENTFS_MAGIC);
689 static struct file_system_type uverbs_event_fs = {
690 /* No owner field so module can be unloaded */
691 .name = "infinibandeventfs",
692 .get_sb = uverbs_event_get_sb,
693 .kill_sb = kill_litter_super
696 static int __init ib_uverbs_init(void)
700 spin_lock_init(&map_lock);
702 ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
705 printk(KERN_ERR "user_verbs: couldn't register device number\n");
709 ret = class_register(&uverbs_class);
711 printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
715 ret = class_create_file(&uverbs_class, &class_attr_abi_version);
717 printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
721 ret = register_filesystem(&uverbs_event_fs);
723 printk(KERN_ERR "user_verbs: couldn't register infinibandeventfs\n");
727 uverbs_event_mnt = kern_mount(&uverbs_event_fs);
728 if (IS_ERR(uverbs_event_mnt)) {
729 ret = PTR_ERR(uverbs_event_mnt);
730 printk(KERN_ERR "user_verbs: couldn't mount infinibandeventfs\n");
734 ret = ib_register_client(&uverbs_client);
736 printk(KERN_ERR "user_verbs: couldn't register client\n");
743 mntput(uverbs_event_mnt);
746 unregister_filesystem(&uverbs_event_fs);
749 class_unregister(&uverbs_class);
752 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
758 static void __exit ib_uverbs_cleanup(void)
760 ib_unregister_client(&uverbs_client);
761 mntput(uverbs_event_mnt);
762 unregister_filesystem(&uverbs_event_fs);
763 class_unregister(&uverbs_class);
764 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
767 module_init(ib_uverbs_init);
768 module_exit(ib_uverbs_cleanup);