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);
169 * ib_alloc_device - allocate an IB device struct
170 * @size:size of structure to allocate
172 * Low-level drivers should use ib_alloc_device() to allocate &struct
173 * ib_device. @size is the size of the structure to be allocated,
174 * including any private data used by the low-level driver.
175 * ib_dealloc_device() must be used to free structures allocated with
178 struct ib_device *ib_alloc_device(size_t size)
180 BUG_ON(size < sizeof (struct ib_device));
182 return kzalloc(size, GFP_KERNEL);
184 EXPORT_SYMBOL(ib_alloc_device);
187 * ib_dealloc_device - free an IB device struct
188 * @device:structure to free
190 * Free a structure allocated with ib_alloc_device().
192 void ib_dealloc_device(struct ib_device *device)
194 if (device->reg_state == IB_DEV_UNINITIALIZED) {
199 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
201 kobject_put(&device->dev.kobj);
203 EXPORT_SYMBOL(ib_dealloc_device);
205 static int add_client_context(struct ib_device *device, struct ib_client *client)
207 struct ib_client_data *context;
210 context = kmalloc(sizeof *context, GFP_KERNEL);
212 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
213 device->name, client->name);
217 context->client = client;
218 context->data = NULL;
219 context->going_down = false;
221 down_write(&lists_rwsem);
222 spin_lock_irqsave(&device->client_data_lock, flags);
223 list_add(&context->list, &device->client_data_list);
224 spin_unlock_irqrestore(&device->client_data_lock, flags);
225 up_write(&lists_rwsem);
230 static int verify_immutable(const struct ib_device *dev, u8 port)
232 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
233 rdma_max_mad_size(dev, port) != 0);
236 static int read_port_immutable(struct ib_device *device)
239 u8 start_port = rdma_start_port(device);
240 u8 end_port = rdma_end_port(device);
244 * device->port_immutable is indexed directly by the port number to make
245 * access to this data as efficient as possible.
247 * Therefore port_immutable is declared as a 1 based array with
248 * potential empty slots at the beginning.
250 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
253 if (!device->port_immutable)
256 for (port = start_port; port <= end_port; ++port) {
257 ret = device->get_port_immutable(device, port,
258 &device->port_immutable[port]);
262 if (verify_immutable(device, port)) {
271 kfree(device->port_immutable);
277 * ib_register_device - Register an IB device with IB core
278 * @device:Device to register
280 * Low-level drivers use ib_register_device() to register their
281 * devices with the IB core. All registered clients will receive a
282 * callback for each device that is added. @device must be allocated
283 * with ib_alloc_device().
285 int ib_register_device(struct ib_device *device,
286 int (*port_callback)(struct ib_device *,
287 u8, struct kobject *))
291 mutex_lock(&device_mutex);
293 if (strchr(device->name, '%')) {
294 ret = alloc_name(device->name);
299 if (ib_device_check_mandatory(device)) {
304 INIT_LIST_HEAD(&device->event_handler_list);
305 INIT_LIST_HEAD(&device->client_data_list);
306 spin_lock_init(&device->event_handler_lock);
307 spin_lock_init(&device->client_data_lock);
309 ret = read_port_immutable(device);
311 printk(KERN_WARNING "Couldn't create per port immutable data %s\n",
316 ret = ib_device_register_sysfs(device, port_callback);
318 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
320 kfree(device->port_immutable);
324 device->reg_state = IB_DEV_REGISTERED;
327 struct ib_client *client;
329 list_for_each_entry(client, &client_list, list)
330 if (client->add && !add_client_context(device, client))
334 down_write(&lists_rwsem);
335 list_add_tail(&device->core_list, &device_list);
336 up_write(&lists_rwsem);
338 mutex_unlock(&device_mutex);
341 EXPORT_SYMBOL(ib_register_device);
344 * ib_unregister_device - Unregister an IB device
345 * @device:Device to unregister
347 * Unregister an IB device. All clients will receive a remove callback.
349 void ib_unregister_device(struct ib_device *device)
351 struct ib_client_data *context, *tmp;
354 mutex_lock(&device_mutex);
356 down_write(&lists_rwsem);
357 list_del(&device->core_list);
358 spin_lock_irqsave(&device->client_data_lock, flags);
359 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
360 context->going_down = true;
361 spin_unlock_irqrestore(&device->client_data_lock, flags);
362 downgrade_write(&lists_rwsem);
364 list_for_each_entry_safe(context, tmp, &device->client_data_list,
366 if (context->client->remove)
367 context->client->remove(device, context->data);
369 up_read(&lists_rwsem);
371 mutex_unlock(&device_mutex);
373 ib_device_unregister_sysfs(device);
375 down_write(&lists_rwsem);
376 spin_lock_irqsave(&device->client_data_lock, flags);
377 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
379 spin_unlock_irqrestore(&device->client_data_lock, flags);
380 up_write(&lists_rwsem);
382 device->reg_state = IB_DEV_UNREGISTERED;
384 EXPORT_SYMBOL(ib_unregister_device);
387 * ib_register_client - Register an IB client
388 * @client:Client to register
390 * Upper level users of the IB drivers can use ib_register_client() to
391 * register callbacks for IB device addition and removal. When an IB
392 * device is added, each registered client's add method will be called
393 * (in the order the clients were registered), and when a device is
394 * removed, each client's remove method will be called (in the reverse
395 * order that clients were registered). In addition, when
396 * ib_register_client() is called, the client will receive an add
397 * callback for all devices already registered.
399 int ib_register_client(struct ib_client *client)
401 struct ib_device *device;
403 mutex_lock(&device_mutex);
405 list_for_each_entry(device, &device_list, core_list)
406 if (client->add && !add_client_context(device, client))
409 down_write(&lists_rwsem);
410 list_add_tail(&client->list, &client_list);
411 up_write(&lists_rwsem);
413 mutex_unlock(&device_mutex);
417 EXPORT_SYMBOL(ib_register_client);
420 * ib_unregister_client - Unregister an IB client
421 * @client:Client to unregister
423 * Upper level users use ib_unregister_client() to remove their client
424 * registration. When ib_unregister_client() is called, the client
425 * will receive a remove callback for each IB device still registered.
427 void ib_unregister_client(struct ib_client *client)
429 struct ib_client_data *context, *tmp;
430 struct ib_device *device;
433 mutex_lock(&device_mutex);
435 down_write(&lists_rwsem);
436 list_del(&client->list);
437 up_write(&lists_rwsem);
439 list_for_each_entry(device, &device_list, core_list) {
440 struct ib_client_data *found_context = NULL;
442 down_write(&lists_rwsem);
443 spin_lock_irqsave(&device->client_data_lock, flags);
444 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
445 if (context->client == client) {
446 context->going_down = true;
447 found_context = context;
450 spin_unlock_irqrestore(&device->client_data_lock, flags);
451 up_write(&lists_rwsem);
454 client->remove(device, found_context ?
455 found_context->data : NULL);
457 if (!found_context) {
458 pr_warn("No client context found for %s/%s\n",
459 device->name, client->name);
463 down_write(&lists_rwsem);
464 spin_lock_irqsave(&device->client_data_lock, flags);
465 list_del(&found_context->list);
466 kfree(found_context);
467 spin_unlock_irqrestore(&device->client_data_lock, flags);
468 up_write(&lists_rwsem);
471 mutex_unlock(&device_mutex);
473 EXPORT_SYMBOL(ib_unregister_client);
476 * ib_get_client_data - Get IB client context
477 * @device:Device to get context for
478 * @client:Client to get context for
480 * ib_get_client_data() returns client context set with
481 * ib_set_client_data().
483 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
485 struct ib_client_data *context;
489 spin_lock_irqsave(&device->client_data_lock, flags);
490 list_for_each_entry(context, &device->client_data_list, list)
491 if (context->client == client) {
495 spin_unlock_irqrestore(&device->client_data_lock, flags);
499 EXPORT_SYMBOL(ib_get_client_data);
502 * ib_set_client_data - Set IB client context
503 * @device:Device to set context for
504 * @client:Client to set context for
505 * @data:Context to set
507 * ib_set_client_data() sets client context that can be retrieved with
508 * ib_get_client_data().
510 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
513 struct ib_client_data *context;
516 spin_lock_irqsave(&device->client_data_lock, flags);
517 list_for_each_entry(context, &device->client_data_list, list)
518 if (context->client == client) {
519 context->data = data;
523 printk(KERN_WARNING "No client context found for %s/%s\n",
524 device->name, client->name);
527 spin_unlock_irqrestore(&device->client_data_lock, flags);
529 EXPORT_SYMBOL(ib_set_client_data);
532 * ib_register_event_handler - Register an IB event handler
533 * @event_handler:Handler to register
535 * ib_register_event_handler() registers an event handler that will be
536 * called back when asynchronous IB events occur (as defined in
537 * chapter 11 of the InfiniBand Architecture Specification). This
538 * callback may occur in interrupt context.
540 int ib_register_event_handler (struct ib_event_handler *event_handler)
544 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
545 list_add_tail(&event_handler->list,
546 &event_handler->device->event_handler_list);
547 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
551 EXPORT_SYMBOL(ib_register_event_handler);
554 * ib_unregister_event_handler - Unregister an event handler
555 * @event_handler:Handler to unregister
557 * Unregister an event handler registered with
558 * ib_register_event_handler().
560 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
564 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
565 list_del(&event_handler->list);
566 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
570 EXPORT_SYMBOL(ib_unregister_event_handler);
573 * ib_dispatch_event - Dispatch an asynchronous event
574 * @event:Event to dispatch
576 * Low-level drivers must call ib_dispatch_event() to dispatch the
577 * event to all registered event handlers when an asynchronous event
580 void ib_dispatch_event(struct ib_event *event)
583 struct ib_event_handler *handler;
585 spin_lock_irqsave(&event->device->event_handler_lock, flags);
587 list_for_each_entry(handler, &event->device->event_handler_list, list)
588 handler->handler(handler, event);
590 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
592 EXPORT_SYMBOL(ib_dispatch_event);
595 * ib_query_device - Query IB device attributes
596 * @device:Device to query
597 * @device_attr:Device attributes
599 * ib_query_device() returns the attributes of a device through the
600 * @device_attr pointer.
602 int ib_query_device(struct ib_device *device,
603 struct ib_device_attr *device_attr)
605 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
607 memset(device_attr, 0, sizeof(*device_attr));
609 return device->query_device(device, device_attr, &uhw);
611 EXPORT_SYMBOL(ib_query_device);
614 * ib_query_port - Query IB port attributes
615 * @device:Device to query
616 * @port_num:Port number to query
617 * @port_attr:Port attributes
619 * ib_query_port() returns the attributes of a port through the
620 * @port_attr pointer.
622 int ib_query_port(struct ib_device *device,
624 struct ib_port_attr *port_attr)
626 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
629 return device->query_port(device, port_num, port_attr);
631 EXPORT_SYMBOL(ib_query_port);
634 * ib_query_gid - Get GID table entry
635 * @device:Device to query
636 * @port_num:Port number to query
637 * @index:GID table index to query
640 * ib_query_gid() fetches the specified GID table entry.
642 int ib_query_gid(struct ib_device *device,
643 u8 port_num, int index, union ib_gid *gid)
645 return device->query_gid(device, port_num, index, gid);
647 EXPORT_SYMBOL(ib_query_gid);
650 * ib_query_pkey - Get P_Key table entry
651 * @device:Device to query
652 * @port_num:Port number to query
653 * @index:P_Key table index to query
654 * @pkey:Returned P_Key
656 * ib_query_pkey() fetches the specified P_Key table entry.
658 int ib_query_pkey(struct ib_device *device,
659 u8 port_num, u16 index, u16 *pkey)
661 return device->query_pkey(device, port_num, index, pkey);
663 EXPORT_SYMBOL(ib_query_pkey);
666 * ib_modify_device - Change IB device attributes
667 * @device:Device to modify
668 * @device_modify_mask:Mask of attributes to change
669 * @device_modify:New attribute values
671 * ib_modify_device() changes a device's attributes as specified by
672 * the @device_modify_mask and @device_modify structure.
674 int ib_modify_device(struct ib_device *device,
675 int device_modify_mask,
676 struct ib_device_modify *device_modify)
678 if (!device->modify_device)
681 return device->modify_device(device, device_modify_mask,
684 EXPORT_SYMBOL(ib_modify_device);
687 * ib_modify_port - Modifies the attributes for the specified port.
688 * @device: The device to modify.
689 * @port_num: The number of the port to modify.
690 * @port_modify_mask: Mask used to specify which attributes of the port
692 * @port_modify: New attribute values for the port.
694 * ib_modify_port() changes a port's attributes as specified by the
695 * @port_modify_mask and @port_modify structure.
697 int ib_modify_port(struct ib_device *device,
698 u8 port_num, int port_modify_mask,
699 struct ib_port_modify *port_modify)
701 if (!device->modify_port)
704 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
707 return device->modify_port(device, port_num, port_modify_mask,
710 EXPORT_SYMBOL(ib_modify_port);
713 * ib_find_gid - Returns the port number and GID table index where
714 * a specified GID value occurs.
715 * @device: The device to query.
716 * @gid: The GID value to search for.
717 * @port_num: The port number of the device where the GID value was found.
718 * @index: The index into the GID table where the GID was found. This
719 * parameter may be NULL.
721 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
722 u8 *port_num, u16 *index)
724 union ib_gid tmp_gid;
727 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
728 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
729 ret = ib_query_gid(device, port, i, &tmp_gid);
732 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
743 EXPORT_SYMBOL(ib_find_gid);
746 * ib_find_pkey - Returns the PKey table index where a specified
748 * @device: The device to query.
749 * @port_num: The port number of the device to search for the PKey.
750 * @pkey: The PKey value to search for.
751 * @index: The index into the PKey table where the PKey was found.
753 int ib_find_pkey(struct ib_device *device,
754 u8 port_num, u16 pkey, u16 *index)
760 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
761 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
764 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
765 /* if there is full-member pkey take it.*/
766 if (tmp_pkey & 0x8000) {
775 /*no full-member, if exists take the limited*/
776 if (partial_ix >= 0) {
782 EXPORT_SYMBOL(ib_find_pkey);
785 * ib_get_net_dev_by_params() - Return the appropriate net_dev
786 * for a received CM request
787 * @dev: An RDMA device on which the request has been received.
788 * @port: Port number on the RDMA device.
789 * @pkey: The Pkey the request came on.
790 * @gid: A GID that the net_dev uses to communicate.
791 * @addr: Contains the IP address that the request specified as its
794 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev,
797 const union ib_gid *gid,
798 const struct sockaddr *addr)
800 struct net_device *net_dev = NULL;
801 struct ib_client_data *context;
803 if (!rdma_protocol_ib(dev, port))
806 down_read(&lists_rwsem);
808 list_for_each_entry(context, &dev->client_data_list, list) {
809 struct ib_client *client = context->client;
811 if (context->going_down)
814 if (client->get_net_dev_by_params) {
815 net_dev = client->get_net_dev_by_params(dev, port, pkey,
823 up_read(&lists_rwsem);
827 EXPORT_SYMBOL(ib_get_net_dev_by_params);
829 static int __init ib_core_init(void)
833 ib_wq = alloc_workqueue("infiniband", 0, 0);
837 ret = ib_sysfs_setup();
839 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
845 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
849 ret = ib_cache_setup();
851 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
864 destroy_workqueue(ib_wq);
868 static void __exit ib_core_cleanup(void)
873 /* Make sure that any pending umem accounting work is done. */
874 destroy_workqueue(ib_wq);
877 module_init(ib_core_init);
878 module_exit(ib_core_cleanup);