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);
73 static spinlock_t map_lock;
74 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
76 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
77 const char __user *buf, int in_len,
79 [IB_USER_VERBS_CMD_QUERY_PARAMS] = ib_uverbs_query_params,
80 [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
81 [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
82 [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
83 [IB_USER_VERBS_CMD_QUERY_GID] = ib_uverbs_query_gid,
84 [IB_USER_VERBS_CMD_QUERY_PKEY] = ib_uverbs_query_pkey,
85 [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
86 [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
87 [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
88 [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
89 [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
90 [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
91 [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
92 [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
93 [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
94 [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
95 [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
98 static struct vfsmount *uverbs_event_mnt;
100 static void ib_uverbs_add_one(struct ib_device *device);
101 static void ib_uverbs_remove_one(struct ib_device *device);
103 static int ib_dealloc_ucontext(struct ib_ucontext *context)
105 struct ib_uobject *uobj, *tmp;
110 down(&ib_uverbs_idr_mutex);
114 list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
115 struct ib_qp *qp = idr_find(&ib_uverbs_qp_idr, uobj->id);
116 idr_remove(&ib_uverbs_qp_idr, uobj->id);
118 list_del(&uobj->list);
122 list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
123 struct ib_cq *cq = idr_find(&ib_uverbs_cq_idr, uobj->id);
124 idr_remove(&ib_uverbs_cq_idr, uobj->id);
126 list_del(&uobj->list);
133 list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
134 struct ib_mr *mr = idr_find(&ib_uverbs_mr_idr, uobj->id);
135 struct ib_device *mrdev = mr->device;
136 struct ib_umem_object *memobj;
138 idr_remove(&ib_uverbs_mr_idr, uobj->id);
141 memobj = container_of(uobj, struct ib_umem_object, uobject);
142 ib_umem_release_on_close(mrdev, &memobj->umem);
144 list_del(&uobj->list);
148 list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
149 struct ib_pd *pd = idr_find(&ib_uverbs_pd_idr, uobj->id);
150 idr_remove(&ib_uverbs_pd_idr, uobj->id);
152 list_del(&uobj->list);
156 up(&ib_uverbs_idr_mutex);
158 return context->device->dealloc_ucontext(context);
161 static void ib_uverbs_release_file(struct kref *ref)
163 struct ib_uverbs_file *file =
164 container_of(ref, struct ib_uverbs_file, ref);
166 module_put(file->device->ib_dev->owner);
170 static ssize_t ib_uverbs_event_read(struct file *filp, char __user *buf,
171 size_t count, loff_t *pos)
173 struct ib_uverbs_event_file *file = filp->private_data;
178 spin_lock_irq(&file->lock);
180 while (list_empty(&file->event_list) && file->fd >= 0) {
181 spin_unlock_irq(&file->lock);
183 if (filp->f_flags & O_NONBLOCK)
186 if (wait_event_interruptible(file->poll_wait,
187 !list_empty(&file->event_list) ||
191 spin_lock_irq(&file->lock);
195 spin_unlock_irq(&file->lock);
199 if (file->is_async) {
200 event = list_entry(file->event_list.next,
201 struct ib_uverbs_async_event, list);
202 eventsz = sizeof (struct ib_uverbs_async_event_desc);
204 event = list_entry(file->event_list.next,
205 struct ib_uverbs_comp_event, list);
206 eventsz = sizeof (struct ib_uverbs_comp_event_desc);
209 if (eventsz > count) {
213 list_del(file->event_list.next);
215 spin_unlock_irq(&file->lock);
218 if (copy_to_user(buf, event, eventsz))
229 static unsigned int ib_uverbs_event_poll(struct file *filp,
230 struct poll_table_struct *wait)
232 unsigned int pollflags = 0;
233 struct ib_uverbs_event_file *file = filp->private_data;
235 poll_wait(filp, &file->poll_wait, wait);
237 spin_lock_irq(&file->lock);
240 else if (!list_empty(&file->event_list))
241 pollflags = POLLIN | POLLRDNORM;
242 spin_unlock_irq(&file->lock);
247 static void ib_uverbs_event_release(struct ib_uverbs_event_file *file)
249 struct list_head *entry, *tmp;
251 spin_lock_irq(&file->lock);
252 if (file->fd != -1) {
254 list_for_each_safe(entry, tmp, &file->event_list)
256 kfree(list_entry(entry, struct ib_uverbs_async_event, list));
258 kfree(list_entry(entry, struct ib_uverbs_comp_event, list));
260 spin_unlock_irq(&file->lock);
263 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
265 struct ib_uverbs_event_file *file = filp->private_data;
267 return fasync_helper(fd, filp, on, &file->async_queue);
270 static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
272 struct ib_uverbs_event_file *file = filp->private_data;
274 ib_uverbs_event_release(file);
275 ib_uverbs_event_fasync(-1, filp, 0);
276 kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
281 static struct file_operations uverbs_event_fops = {
283 * No .owner field since we artificially create event files,
284 * so there is no increment to the module reference count in
285 * the open path. All event files come from a uverbs command
286 * file, which already takes a module reference, so this is OK.
288 .read = ib_uverbs_event_read,
289 .poll = ib_uverbs_event_poll,
290 .release = ib_uverbs_event_close,
291 .fasync = ib_uverbs_event_fasync
294 void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
296 struct ib_uverbs_file *file = cq_context;
297 struct ib_uverbs_comp_event *entry;
300 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
304 entry->desc.cq_handle = cq->uobject->user_handle;
306 spin_lock_irqsave(&file->comp_file[0].lock, flags);
307 list_add_tail(&entry->list, &file->comp_file[0].event_list);
308 spin_unlock_irqrestore(&file->comp_file[0].lock, flags);
310 wake_up_interruptible(&file->comp_file[0].poll_wait);
311 kill_fasync(&file->comp_file[0].async_queue, SIGIO, POLL_IN);
314 static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
315 __u64 element, __u64 event)
317 struct ib_uverbs_async_event *entry;
320 entry = kmalloc(sizeof *entry, GFP_ATOMIC);
324 entry->desc.element = element;
325 entry->desc.event_type = event;
327 spin_lock_irqsave(&file->async_file.lock, flags);
328 list_add_tail(&entry->list, &file->async_file.event_list);
329 spin_unlock_irqrestore(&file->async_file.lock, flags);
331 wake_up_interruptible(&file->async_file.poll_wait);
332 kill_fasync(&file->async_file.async_queue, SIGIO, POLL_IN);
335 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
337 ib_uverbs_async_handler(context_ptr,
338 event->element.cq->uobject->user_handle,
342 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
344 ib_uverbs_async_handler(context_ptr,
345 event->element.qp->uobject->user_handle,
349 static void ib_uverbs_event_handler(struct ib_event_handler *handler,
350 struct ib_event *event)
352 struct ib_uverbs_file *file =
353 container_of(handler, struct ib_uverbs_file, event_handler);
355 ib_uverbs_async_handler(file, event->element.port_num, event->event);
358 static int ib_uverbs_event_init(struct ib_uverbs_event_file *file,
359 struct ib_uverbs_file *uverbs_file)
363 spin_lock_init(&file->lock);
364 INIT_LIST_HEAD(&file->event_list);
365 init_waitqueue_head(&file->poll_wait);
366 file->uverbs_file = uverbs_file;
367 file->async_queue = NULL;
369 file->fd = get_unused_fd();
373 filp = get_empty_filp();
375 put_unused_fd(file->fd);
379 filp->f_op = &uverbs_event_fops;
380 filp->f_vfsmnt = mntget(uverbs_event_mnt);
381 filp->f_dentry = dget(uverbs_event_mnt->mnt_root);
382 filp->f_mapping = filp->f_dentry->d_inode->i_mapping;
383 filp->f_flags = O_RDONLY;
384 filp->f_mode = FMODE_READ;
385 filp->private_data = file;
387 fd_install(file->fd, filp);
392 static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
393 size_t count, loff_t *pos)
395 struct ib_uverbs_file *file = filp->private_data;
396 struct ib_uverbs_cmd_hdr hdr;
398 if (count < sizeof hdr)
401 if (copy_from_user(&hdr, buf, sizeof hdr))
404 if (hdr.in_words * 4 != count)
407 if (hdr.command < 0 || hdr.command >= ARRAY_SIZE(uverbs_cmd_table))
410 if (!file->ucontext &&
411 hdr.command != IB_USER_VERBS_CMD_QUERY_PARAMS &&
412 hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
415 return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
416 hdr.in_words * 4, hdr.out_words * 4);
419 static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
421 struct ib_uverbs_file *file = filp->private_data;
426 return file->device->ib_dev->mmap(file->ucontext, vma);
429 static int ib_uverbs_open(struct inode *inode, struct file *filp)
431 struct ib_uverbs_device *dev =
432 container_of(inode->i_cdev, struct ib_uverbs_device, dev);
433 struct ib_uverbs_file *file;
437 if (!try_module_get(dev->ib_dev->owner))
440 file = kmalloc(sizeof *file +
441 (dev->num_comp - 1) * sizeof (struct ib_uverbs_event_file),
447 kref_init(&file->ref);
449 file->ucontext = NULL;
451 ret = ib_uverbs_event_init(&file->async_file, file);
455 file->async_file.is_async = 1;
457 kref_get(&file->ref);
459 for (i = 0; i < dev->num_comp; ++i) {
460 ret = ib_uverbs_event_init(&file->comp_file[i], file);
463 kref_get(&file->ref);
464 file->comp_file[i].is_async = 0;
468 filp->private_data = file;
470 INIT_IB_EVENT_HANDLER(&file->event_handler, dev->ib_dev,
471 ib_uverbs_event_handler);
472 if (ib_register_event_handler(&file->event_handler))
479 ib_uverbs_event_release(&file->comp_file[i]);
481 ib_uverbs_event_release(&file->async_file);
484 kref_put(&file->ref, ib_uverbs_release_file);
489 static int ib_uverbs_close(struct inode *inode, struct file *filp)
491 struct ib_uverbs_file *file = filp->private_data;
494 ib_unregister_event_handler(&file->event_handler);
495 ib_uverbs_event_release(&file->async_file);
496 ib_dealloc_ucontext(file->ucontext);
498 for (i = 0; i < file->device->num_comp; ++i)
499 ib_uverbs_event_release(&file->comp_file[i]);
501 kref_put(&file->ref, ib_uverbs_release_file);
506 static struct file_operations uverbs_fops = {
507 .owner = THIS_MODULE,
508 .write = ib_uverbs_write,
509 .open = ib_uverbs_open,
510 .release = ib_uverbs_close
513 static struct file_operations uverbs_mmap_fops = {
514 .owner = THIS_MODULE,
515 .write = ib_uverbs_write,
516 .mmap = ib_uverbs_mmap,
517 .open = ib_uverbs_open,
518 .release = ib_uverbs_close
521 static struct ib_client uverbs_client = {
523 .add = ib_uverbs_add_one,
524 .remove = ib_uverbs_remove_one
527 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
529 struct ib_uverbs_device *dev =
530 container_of(class_dev, struct ib_uverbs_device, class_dev);
532 return sprintf(buf, "%s\n", dev->ib_dev->name);
534 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
536 static void ib_uverbs_release_class_dev(struct class_device *class_dev)
538 struct ib_uverbs_device *dev =
539 container_of(class_dev, struct ib_uverbs_device, class_dev);
542 clear_bit(dev->devnum, dev_map);
546 static struct class uverbs_class = {
547 .name = "infiniband_verbs",
548 .release = ib_uverbs_release_class_dev
551 static ssize_t show_abi_version(struct class *class, char *buf)
553 return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
555 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
557 static void ib_uverbs_add_one(struct ib_device *device)
559 struct ib_uverbs_device *uverbs_dev;
561 if (!device->alloc_ucontext)
564 uverbs_dev = kmalloc(sizeof *uverbs_dev, GFP_KERNEL);
568 memset(uverbs_dev, 0, sizeof *uverbs_dev);
570 spin_lock(&map_lock);
571 uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
572 if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
573 spin_unlock(&map_lock);
576 set_bit(uverbs_dev->devnum, dev_map);
577 spin_unlock(&map_lock);
579 uverbs_dev->ib_dev = device;
580 uverbs_dev->num_comp = 1;
583 cdev_init(&uverbs_dev->dev, &uverbs_mmap_fops);
585 cdev_init(&uverbs_dev->dev, &uverbs_fops);
586 uverbs_dev->dev.owner = THIS_MODULE;
587 kobject_set_name(&uverbs_dev->dev.kobj, "uverbs%d", uverbs_dev->devnum);
588 if (cdev_add(&uverbs_dev->dev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
591 uverbs_dev->class_dev.class = &uverbs_class;
592 uverbs_dev->class_dev.dev = device->dma_device;
593 uverbs_dev->class_dev.devt = uverbs_dev->dev.dev;
594 snprintf(uverbs_dev->class_dev.class_id, BUS_ID_SIZE, "uverbs%d", uverbs_dev->devnum);
595 if (class_device_register(&uverbs_dev->class_dev))
598 if (class_device_create_file(&uverbs_dev->class_dev, &class_device_attr_ibdev))
601 ib_set_client_data(device, &uverbs_client, uverbs_dev);
606 class_device_unregister(&uverbs_dev->class_dev);
609 cdev_del(&uverbs_dev->dev);
610 clear_bit(uverbs_dev->devnum, dev_map);
617 static void ib_uverbs_remove_one(struct ib_device *device)
619 struct ib_uverbs_device *uverbs_dev = ib_get_client_data(device, &uverbs_client);
624 class_device_unregister(&uverbs_dev->class_dev);
627 static struct super_block *uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
628 const char *dev_name, void *data)
630 return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
631 INFINIBANDEVENTFS_MAGIC);
634 static struct file_system_type uverbs_event_fs = {
635 /* No owner field so module can be unloaded */
636 .name = "infinibandeventfs",
637 .get_sb = uverbs_event_get_sb,
638 .kill_sb = kill_litter_super
641 static int __init ib_uverbs_init(void)
645 spin_lock_init(&map_lock);
647 ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
650 printk(KERN_ERR "user_verbs: couldn't register device number\n");
654 ret = class_register(&uverbs_class);
656 printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
660 ret = class_create_file(&uverbs_class, &class_attr_abi_version);
662 printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
666 ret = register_filesystem(&uverbs_event_fs);
668 printk(KERN_ERR "user_verbs: couldn't register infinibandeventfs\n");
672 uverbs_event_mnt = kern_mount(&uverbs_event_fs);
673 if (IS_ERR(uverbs_event_mnt)) {
674 ret = PTR_ERR(uverbs_event_mnt);
675 printk(KERN_ERR "user_verbs: couldn't mount infinibandeventfs\n");
679 ret = ib_register_client(&uverbs_client);
681 printk(KERN_ERR "user_verbs: couldn't register client\n");
688 mntput(uverbs_event_mnt);
691 unregister_filesystem(&uverbs_event_fs);
694 class_unregister(&uverbs_class);
697 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
703 static void __exit ib_uverbs_cleanup(void)
705 ib_unregister_client(&uverbs_client);
706 mntput(uverbs_event_mnt);
707 unregister_filesystem(&uverbs_event_fs);
708 class_unregister(&uverbs_class);
709 unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
712 module_init(ib_uverbs_init);
713 module_exit(ib_uverbs_cleanup);