2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 #include <linux/module.h>
35 #include <linux/string.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/mutex.h>
41 #include <linux/netdevice.h>
42 #include <rdma/rdma_netlink.h>
44 #include "core_priv.h"
46 MODULE_AUTHOR("Roland Dreier");
47 MODULE_DESCRIPTION("core kernel InfiniBand API");
48 MODULE_LICENSE("Dual BSD/GPL");
50 struct ib_client_data {
51 struct list_head list;
52 struct ib_client *client;
54 /* The device or client is going down. Do not call client or device
55 * callbacks other than remove(). */
59 struct workqueue_struct *ib_wq;
60 EXPORT_SYMBOL_GPL(ib_wq);
62 /* The device_list and client_list contain devices and clients after their
63 * registration has completed, and the devices and clients are removed
64 * during unregistration. */
65 static LIST_HEAD(device_list);
66 static LIST_HEAD(client_list);
69 * device_mutex and lists_rwsem protect access to both device_list and
70 * client_list. device_mutex protects writer access by device and client
71 * registration / de-registration. lists_rwsem protects reader access to
72 * these lists. Iterators of these lists must lock it for read, while updates
73 * to the lists must be done with a write lock. A special case is when the
74 * device_mutex is locked. In this case locking the lists for read access is
75 * not necessary as the device_mutex implies it.
77 * lists_rwsem also protects access to the client data list.
79 static DEFINE_MUTEX(device_mutex);
80 static DECLARE_RWSEM(lists_rwsem);
83 static int ib_device_check_mandatory(struct ib_device *device)
85 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
89 } mandatory_table[] = {
90 IB_MANDATORY_FUNC(query_device),
91 IB_MANDATORY_FUNC(query_port),
92 IB_MANDATORY_FUNC(query_pkey),
93 IB_MANDATORY_FUNC(query_gid),
94 IB_MANDATORY_FUNC(alloc_pd),
95 IB_MANDATORY_FUNC(dealloc_pd),
96 IB_MANDATORY_FUNC(create_ah),
97 IB_MANDATORY_FUNC(destroy_ah),
98 IB_MANDATORY_FUNC(create_qp),
99 IB_MANDATORY_FUNC(modify_qp),
100 IB_MANDATORY_FUNC(destroy_qp),
101 IB_MANDATORY_FUNC(post_send),
102 IB_MANDATORY_FUNC(post_recv),
103 IB_MANDATORY_FUNC(create_cq),
104 IB_MANDATORY_FUNC(destroy_cq),
105 IB_MANDATORY_FUNC(poll_cq),
106 IB_MANDATORY_FUNC(req_notify_cq),
107 IB_MANDATORY_FUNC(get_dma_mr),
108 IB_MANDATORY_FUNC(dereg_mr),
109 IB_MANDATORY_FUNC(get_port_immutable)
113 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
114 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
115 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
116 device->name, mandatory_table[i].name);
124 static struct ib_device *__ib_device_get_by_name(const char *name)
126 struct ib_device *device;
128 list_for_each_entry(device, &device_list, core_list)
129 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
136 static int alloc_name(char *name)
138 unsigned long *inuse;
139 char buf[IB_DEVICE_NAME_MAX];
140 struct ib_device *device;
143 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
147 list_for_each_entry(device, &device_list, core_list) {
148 if (!sscanf(device->name, name, &i))
150 if (i < 0 || i >= PAGE_SIZE * 8)
152 snprintf(buf, sizeof buf, name, i);
153 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
157 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
158 free_page((unsigned long) inuse);
159 snprintf(buf, sizeof buf, name, i);
161 if (__ib_device_get_by_name(buf))
164 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
168 static void ib_device_release(struct device *device)
170 struct ib_device *dev = container_of(device, struct ib_device, dev);
172 kfree(dev->port_immutable);
176 static int ib_device_uevent(struct device *device,
177 struct kobj_uevent_env *env)
179 struct ib_device *dev = container_of(device, struct ib_device, dev);
181 if (add_uevent_var(env, "NAME=%s", dev->name))
185 * It would be nice to pass the node GUID with the event...
191 static struct class ib_class = {
192 .name = "infiniband",
193 .dev_release = ib_device_release,
194 .dev_uevent = ib_device_uevent,
198 * ib_alloc_device - allocate an IB device struct
199 * @size:size of structure to allocate
201 * Low-level drivers should use ib_alloc_device() to allocate &struct
202 * ib_device. @size is the size of the structure to be allocated,
203 * including any private data used by the low-level driver.
204 * ib_dealloc_device() must be used to free structures allocated with
207 struct ib_device *ib_alloc_device(size_t size)
209 struct ib_device *device;
211 if (WARN_ON(size < sizeof(struct ib_device)))
214 device = kzalloc(size, GFP_KERNEL);
218 device->dev.class = &ib_class;
219 device_initialize(&device->dev);
221 dev_set_drvdata(&device->dev, device);
223 INIT_LIST_HEAD(&device->event_handler_list);
224 spin_lock_init(&device->event_handler_lock);
225 spin_lock_init(&device->client_data_lock);
226 INIT_LIST_HEAD(&device->client_data_list);
227 INIT_LIST_HEAD(&device->port_list);
231 EXPORT_SYMBOL(ib_alloc_device);
234 * ib_dealloc_device - free an IB device struct
235 * @device:structure to free
237 * Free a structure allocated with ib_alloc_device().
239 void ib_dealloc_device(struct ib_device *device)
241 WARN_ON(device->reg_state != IB_DEV_UNREGISTERED &&
242 device->reg_state != IB_DEV_UNINITIALIZED);
243 kobject_put(&device->dev.kobj);
245 EXPORT_SYMBOL(ib_dealloc_device);
247 static int add_client_context(struct ib_device *device, struct ib_client *client)
249 struct ib_client_data *context;
252 context = kmalloc(sizeof *context, GFP_KERNEL);
254 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
255 device->name, client->name);
259 context->client = client;
260 context->data = NULL;
261 context->going_down = false;
263 down_write(&lists_rwsem);
264 spin_lock_irqsave(&device->client_data_lock, flags);
265 list_add(&context->list, &device->client_data_list);
266 spin_unlock_irqrestore(&device->client_data_lock, flags);
267 up_write(&lists_rwsem);
272 static int verify_immutable(const struct ib_device *dev, u8 port)
274 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
275 rdma_max_mad_size(dev, port) != 0);
278 static int read_port_immutable(struct ib_device *device)
281 u8 start_port = rdma_start_port(device);
282 u8 end_port = rdma_end_port(device);
286 * device->port_immutable is indexed directly by the port number to make
287 * access to this data as efficient as possible.
289 * Therefore port_immutable is declared as a 1 based array with
290 * potential empty slots at the beginning.
292 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
295 if (!device->port_immutable)
298 for (port = start_port; port <= end_port; ++port) {
299 ret = device->get_port_immutable(device, port,
300 &device->port_immutable[port]);
304 if (verify_immutable(device, port))
311 * ib_register_device - Register an IB device with IB core
312 * @device:Device to register
314 * Low-level drivers use ib_register_device() to register their
315 * devices with the IB core. All registered clients will receive a
316 * callback for each device that is added. @device must be allocated
317 * with ib_alloc_device().
319 int ib_register_device(struct ib_device *device,
320 int (*port_callback)(struct ib_device *,
321 u8, struct kobject *))
325 mutex_lock(&device_mutex);
327 if (strchr(device->name, '%')) {
328 ret = alloc_name(device->name);
333 if (ib_device_check_mandatory(device)) {
338 ret = read_port_immutable(device);
340 printk(KERN_WARNING "Couldn't create per port immutable data %s\n",
345 ret = ib_device_register_sysfs(device, port_callback);
347 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
352 device->reg_state = IB_DEV_REGISTERED;
355 struct ib_client *client;
357 list_for_each_entry(client, &client_list, list)
358 if (client->add && !add_client_context(device, client))
362 down_write(&lists_rwsem);
363 list_add_tail(&device->core_list, &device_list);
364 up_write(&lists_rwsem);
366 mutex_unlock(&device_mutex);
369 EXPORT_SYMBOL(ib_register_device);
372 * ib_unregister_device - Unregister an IB device
373 * @device:Device to unregister
375 * Unregister an IB device. All clients will receive a remove callback.
377 void ib_unregister_device(struct ib_device *device)
379 struct ib_client_data *context, *tmp;
382 mutex_lock(&device_mutex);
384 down_write(&lists_rwsem);
385 list_del(&device->core_list);
386 spin_lock_irqsave(&device->client_data_lock, flags);
387 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
388 context->going_down = true;
389 spin_unlock_irqrestore(&device->client_data_lock, flags);
390 downgrade_write(&lists_rwsem);
392 list_for_each_entry_safe(context, tmp, &device->client_data_list,
394 if (context->client->remove)
395 context->client->remove(device, context->data);
397 up_read(&lists_rwsem);
399 mutex_unlock(&device_mutex);
401 ib_device_unregister_sysfs(device);
403 down_write(&lists_rwsem);
404 spin_lock_irqsave(&device->client_data_lock, flags);
405 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
407 spin_unlock_irqrestore(&device->client_data_lock, flags);
408 up_write(&lists_rwsem);
410 device->reg_state = IB_DEV_UNREGISTERED;
412 EXPORT_SYMBOL(ib_unregister_device);
415 * ib_register_client - Register an IB client
416 * @client:Client to register
418 * Upper level users of the IB drivers can use ib_register_client() to
419 * register callbacks for IB device addition and removal. When an IB
420 * device is added, each registered client's add method will be called
421 * (in the order the clients were registered), and when a device is
422 * removed, each client's remove method will be called (in the reverse
423 * order that clients were registered). In addition, when
424 * ib_register_client() is called, the client will receive an add
425 * callback for all devices already registered.
427 int ib_register_client(struct ib_client *client)
429 struct ib_device *device;
431 mutex_lock(&device_mutex);
433 list_for_each_entry(device, &device_list, core_list)
434 if (client->add && !add_client_context(device, client))
437 down_write(&lists_rwsem);
438 list_add_tail(&client->list, &client_list);
439 up_write(&lists_rwsem);
441 mutex_unlock(&device_mutex);
445 EXPORT_SYMBOL(ib_register_client);
448 * ib_unregister_client - Unregister an IB client
449 * @client:Client to unregister
451 * Upper level users use ib_unregister_client() to remove their client
452 * registration. When ib_unregister_client() is called, the client
453 * will receive a remove callback for each IB device still registered.
455 void ib_unregister_client(struct ib_client *client)
457 struct ib_client_data *context, *tmp;
458 struct ib_device *device;
461 mutex_lock(&device_mutex);
463 down_write(&lists_rwsem);
464 list_del(&client->list);
465 up_write(&lists_rwsem);
467 list_for_each_entry(device, &device_list, core_list) {
468 struct ib_client_data *found_context = NULL;
470 down_write(&lists_rwsem);
471 spin_lock_irqsave(&device->client_data_lock, flags);
472 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
473 if (context->client == client) {
474 context->going_down = true;
475 found_context = context;
478 spin_unlock_irqrestore(&device->client_data_lock, flags);
479 up_write(&lists_rwsem);
482 client->remove(device, found_context ?
483 found_context->data : NULL);
485 if (!found_context) {
486 pr_warn("No client context found for %s/%s\n",
487 device->name, client->name);
491 down_write(&lists_rwsem);
492 spin_lock_irqsave(&device->client_data_lock, flags);
493 list_del(&found_context->list);
494 kfree(found_context);
495 spin_unlock_irqrestore(&device->client_data_lock, flags);
496 up_write(&lists_rwsem);
499 mutex_unlock(&device_mutex);
501 EXPORT_SYMBOL(ib_unregister_client);
504 * ib_get_client_data - Get IB client context
505 * @device:Device to get context for
506 * @client:Client to get context for
508 * ib_get_client_data() returns client context set with
509 * ib_set_client_data().
511 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
513 struct ib_client_data *context;
517 spin_lock_irqsave(&device->client_data_lock, flags);
518 list_for_each_entry(context, &device->client_data_list, list)
519 if (context->client == client) {
523 spin_unlock_irqrestore(&device->client_data_lock, flags);
527 EXPORT_SYMBOL(ib_get_client_data);
530 * ib_set_client_data - Set IB client context
531 * @device:Device to set context for
532 * @client:Client to set context for
533 * @data:Context to set
535 * ib_set_client_data() sets client context that can be retrieved with
536 * ib_get_client_data().
538 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
541 struct ib_client_data *context;
544 spin_lock_irqsave(&device->client_data_lock, flags);
545 list_for_each_entry(context, &device->client_data_list, list)
546 if (context->client == client) {
547 context->data = data;
551 printk(KERN_WARNING "No client context found for %s/%s\n",
552 device->name, client->name);
555 spin_unlock_irqrestore(&device->client_data_lock, flags);
557 EXPORT_SYMBOL(ib_set_client_data);
560 * ib_register_event_handler - Register an IB event handler
561 * @event_handler:Handler to register
563 * ib_register_event_handler() registers an event handler that will be
564 * called back when asynchronous IB events occur (as defined in
565 * chapter 11 of the InfiniBand Architecture Specification). This
566 * callback may occur in interrupt context.
568 int ib_register_event_handler (struct ib_event_handler *event_handler)
572 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
573 list_add_tail(&event_handler->list,
574 &event_handler->device->event_handler_list);
575 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
579 EXPORT_SYMBOL(ib_register_event_handler);
582 * ib_unregister_event_handler - Unregister an event handler
583 * @event_handler:Handler to unregister
585 * Unregister an event handler registered with
586 * ib_register_event_handler().
588 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
592 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
593 list_del(&event_handler->list);
594 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
598 EXPORT_SYMBOL(ib_unregister_event_handler);
601 * ib_dispatch_event - Dispatch an asynchronous event
602 * @event:Event to dispatch
604 * Low-level drivers must call ib_dispatch_event() to dispatch the
605 * event to all registered event handlers when an asynchronous event
608 void ib_dispatch_event(struct ib_event *event)
611 struct ib_event_handler *handler;
613 spin_lock_irqsave(&event->device->event_handler_lock, flags);
615 list_for_each_entry(handler, &event->device->event_handler_list, list)
616 handler->handler(handler, event);
618 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
620 EXPORT_SYMBOL(ib_dispatch_event);
623 * ib_query_device - Query IB device attributes
624 * @device:Device to query
625 * @device_attr:Device attributes
627 * ib_query_device() returns the attributes of a device through the
628 * @device_attr pointer.
630 int ib_query_device(struct ib_device *device,
631 struct ib_device_attr *device_attr)
633 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
635 memset(device_attr, 0, sizeof(*device_attr));
637 return device->query_device(device, device_attr, &uhw);
639 EXPORT_SYMBOL(ib_query_device);
642 * ib_query_port - Query IB port attributes
643 * @device:Device to query
644 * @port_num:Port number to query
645 * @port_attr:Port attributes
647 * ib_query_port() returns the attributes of a port through the
648 * @port_attr pointer.
650 int ib_query_port(struct ib_device *device,
652 struct ib_port_attr *port_attr)
654 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
657 return device->query_port(device, port_num, port_attr);
659 EXPORT_SYMBOL(ib_query_port);
662 * ib_query_gid - Get GID table entry
663 * @device:Device to query
664 * @port_num:Port number to query
665 * @index:GID table index to query
668 * ib_query_gid() fetches the specified GID table entry.
670 int ib_query_gid(struct ib_device *device,
671 u8 port_num, int index, union ib_gid *gid)
673 return device->query_gid(device, port_num, index, gid);
675 EXPORT_SYMBOL(ib_query_gid);
678 * ib_query_pkey - Get P_Key table entry
679 * @device:Device to query
680 * @port_num:Port number to query
681 * @index:P_Key table index to query
682 * @pkey:Returned P_Key
684 * ib_query_pkey() fetches the specified P_Key table entry.
686 int ib_query_pkey(struct ib_device *device,
687 u8 port_num, u16 index, u16 *pkey)
689 return device->query_pkey(device, port_num, index, pkey);
691 EXPORT_SYMBOL(ib_query_pkey);
694 * ib_modify_device - Change IB device attributes
695 * @device:Device to modify
696 * @device_modify_mask:Mask of attributes to change
697 * @device_modify:New attribute values
699 * ib_modify_device() changes a device's attributes as specified by
700 * the @device_modify_mask and @device_modify structure.
702 int ib_modify_device(struct ib_device *device,
703 int device_modify_mask,
704 struct ib_device_modify *device_modify)
706 if (!device->modify_device)
709 return device->modify_device(device, device_modify_mask,
712 EXPORT_SYMBOL(ib_modify_device);
715 * ib_modify_port - Modifies the attributes for the specified port.
716 * @device: The device to modify.
717 * @port_num: The number of the port to modify.
718 * @port_modify_mask: Mask used to specify which attributes of the port
720 * @port_modify: New attribute values for the port.
722 * ib_modify_port() changes a port's attributes as specified by the
723 * @port_modify_mask and @port_modify structure.
725 int ib_modify_port(struct ib_device *device,
726 u8 port_num, int port_modify_mask,
727 struct ib_port_modify *port_modify)
729 if (!device->modify_port)
732 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
735 return device->modify_port(device, port_num, port_modify_mask,
738 EXPORT_SYMBOL(ib_modify_port);
741 * ib_find_gid - Returns the port number and GID table index where
742 * a specified GID value occurs.
743 * @device: The device to query.
744 * @gid: The GID value to search for.
745 * @port_num: The port number of the device where the GID value was found.
746 * @index: The index into the GID table where the GID was found. This
747 * parameter may be NULL.
749 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
750 u8 *port_num, u16 *index)
752 union ib_gid tmp_gid;
755 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
756 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
757 ret = ib_query_gid(device, port, i, &tmp_gid);
760 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
771 EXPORT_SYMBOL(ib_find_gid);
774 * ib_find_pkey - Returns the PKey table index where a specified
776 * @device: The device to query.
777 * @port_num: The port number of the device to search for the PKey.
778 * @pkey: The PKey value to search for.
779 * @index: The index into the PKey table where the PKey was found.
781 int ib_find_pkey(struct ib_device *device,
782 u8 port_num, u16 pkey, u16 *index)
788 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
789 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
792 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
793 /* if there is full-member pkey take it.*/
794 if (tmp_pkey & 0x8000) {
803 /*no full-member, if exists take the limited*/
804 if (partial_ix >= 0) {
810 EXPORT_SYMBOL(ib_find_pkey);
813 * ib_get_net_dev_by_params() - Return the appropriate net_dev
814 * for a received CM request
815 * @dev: An RDMA device on which the request has been received.
816 * @port: Port number on the RDMA device.
817 * @pkey: The Pkey the request came on.
818 * @gid: A GID that the net_dev uses to communicate.
819 * @addr: Contains the IP address that the request specified as its
822 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev,
825 const union ib_gid *gid,
826 const struct sockaddr *addr)
828 struct net_device *net_dev = NULL;
829 struct ib_client_data *context;
831 if (!rdma_protocol_ib(dev, port))
834 down_read(&lists_rwsem);
836 list_for_each_entry(context, &dev->client_data_list, list) {
837 struct ib_client *client = context->client;
839 if (context->going_down)
842 if (client->get_net_dev_by_params) {
843 net_dev = client->get_net_dev_by_params(dev, port, pkey,
851 up_read(&lists_rwsem);
855 EXPORT_SYMBOL(ib_get_net_dev_by_params);
857 static int __init ib_core_init(void)
861 ib_wq = alloc_workqueue("infiniband", 0, 0);
865 ret = class_register(&ib_class);
867 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
873 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
877 ret = ib_cache_setup();
879 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
889 class_unregister(&ib_class);
892 destroy_workqueue(ib_wq);
896 static void __exit ib_core_cleanup(void)
900 class_unregister(&ib_class);
901 /* Make sure that any pending umem accounting work is done. */
902 destroy_workqueue(ib_wq);
905 module_init(ib_core_init);
906 module_exit(ib_core_cleanup);