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 <rdma/rdma_netlink.h>
43 #include "core_priv.h"
45 MODULE_AUTHOR("Roland Dreier");
46 MODULE_DESCRIPTION("core kernel InfiniBand API");
47 MODULE_LICENSE("Dual BSD/GPL");
49 struct ib_client_data {
50 struct list_head list;
51 struct ib_client *client;
53 /* The device or client is going down. Do not call client or device
54 * callbacks other than remove(). */
58 struct workqueue_struct *ib_wq;
59 EXPORT_SYMBOL_GPL(ib_wq);
61 /* The device_list and client_list contain devices and clients after their
62 * registration has completed, and the devices and clients are removed
63 * during unregistration. */
64 static LIST_HEAD(device_list);
65 static LIST_HEAD(client_list);
68 * device_mutex and lists_rwsem protect access to both device_list and
69 * client_list. device_mutex protects writer access by device and client
70 * registration / de-registration. lists_rwsem protects reader access to
71 * these lists. Iterators of these lists must lock it for read, while updates
72 * to the lists must be done with a write lock. A special case is when the
73 * device_mutex is locked. In this case locking the lists for read access is
74 * not necessary as the device_mutex implies it.
76 * lists_rwsem also protects access to the client data list.
78 static DEFINE_MUTEX(device_mutex);
79 static DECLARE_RWSEM(lists_rwsem);
82 static int ib_device_check_mandatory(struct ib_device *device)
84 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
88 } mandatory_table[] = {
89 IB_MANDATORY_FUNC(query_device),
90 IB_MANDATORY_FUNC(query_port),
91 IB_MANDATORY_FUNC(query_pkey),
92 IB_MANDATORY_FUNC(query_gid),
93 IB_MANDATORY_FUNC(alloc_pd),
94 IB_MANDATORY_FUNC(dealloc_pd),
95 IB_MANDATORY_FUNC(create_ah),
96 IB_MANDATORY_FUNC(destroy_ah),
97 IB_MANDATORY_FUNC(create_qp),
98 IB_MANDATORY_FUNC(modify_qp),
99 IB_MANDATORY_FUNC(destroy_qp),
100 IB_MANDATORY_FUNC(post_send),
101 IB_MANDATORY_FUNC(post_recv),
102 IB_MANDATORY_FUNC(create_cq),
103 IB_MANDATORY_FUNC(destroy_cq),
104 IB_MANDATORY_FUNC(poll_cq),
105 IB_MANDATORY_FUNC(req_notify_cq),
106 IB_MANDATORY_FUNC(get_dma_mr),
107 IB_MANDATORY_FUNC(dereg_mr),
108 IB_MANDATORY_FUNC(get_port_immutable)
112 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
113 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
114 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
115 device->name, mandatory_table[i].name);
123 static struct ib_device *__ib_device_get_by_name(const char *name)
125 struct ib_device *device;
127 list_for_each_entry(device, &device_list, core_list)
128 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
135 static int alloc_name(char *name)
137 unsigned long *inuse;
138 char buf[IB_DEVICE_NAME_MAX];
139 struct ib_device *device;
142 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
146 list_for_each_entry(device, &device_list, core_list) {
147 if (!sscanf(device->name, name, &i))
149 if (i < 0 || i >= PAGE_SIZE * 8)
151 snprintf(buf, sizeof buf, name, i);
152 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
156 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
157 free_page((unsigned long) inuse);
158 snprintf(buf, sizeof buf, name, i);
160 if (__ib_device_get_by_name(buf))
163 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
168 * ib_alloc_device - allocate an IB device struct
169 * @size:size of structure to allocate
171 * Low-level drivers should use ib_alloc_device() to allocate &struct
172 * ib_device. @size is the size of the structure to be allocated,
173 * including any private data used by the low-level driver.
174 * ib_dealloc_device() must be used to free structures allocated with
177 struct ib_device *ib_alloc_device(size_t size)
179 BUG_ON(size < sizeof (struct ib_device));
181 return kzalloc(size, GFP_KERNEL);
183 EXPORT_SYMBOL(ib_alloc_device);
186 * ib_dealloc_device - free an IB device struct
187 * @device:structure to free
189 * Free a structure allocated with ib_alloc_device().
191 void ib_dealloc_device(struct ib_device *device)
193 if (device->reg_state == IB_DEV_UNINITIALIZED) {
198 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
200 kobject_put(&device->dev.kobj);
202 EXPORT_SYMBOL(ib_dealloc_device);
204 static int add_client_context(struct ib_device *device, struct ib_client *client)
206 struct ib_client_data *context;
209 context = kmalloc(sizeof *context, GFP_KERNEL);
211 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
212 device->name, client->name);
216 context->client = client;
217 context->data = NULL;
218 context->going_down = false;
220 down_write(&lists_rwsem);
221 spin_lock_irqsave(&device->client_data_lock, flags);
222 list_add(&context->list, &device->client_data_list);
223 spin_unlock_irqrestore(&device->client_data_lock, flags);
224 up_write(&lists_rwsem);
229 static int verify_immutable(const struct ib_device *dev, u8 port)
231 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
232 rdma_max_mad_size(dev, port) != 0);
235 static int read_port_immutable(struct ib_device *device)
238 u8 start_port = rdma_start_port(device);
239 u8 end_port = rdma_end_port(device);
243 * device->port_immutable is indexed directly by the port number to make
244 * access to this data as efficient as possible.
246 * Therefore port_immutable is declared as a 1 based array with
247 * potential empty slots at the beginning.
249 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
252 if (!device->port_immutable)
255 for (port = start_port; port <= end_port; ++port) {
256 ret = device->get_port_immutable(device, port,
257 &device->port_immutable[port]);
261 if (verify_immutable(device, port)) {
270 kfree(device->port_immutable);
276 * ib_register_device - Register an IB device with IB core
277 * @device:Device to register
279 * Low-level drivers use ib_register_device() to register their
280 * devices with the IB core. All registered clients will receive a
281 * callback for each device that is added. @device must be allocated
282 * with ib_alloc_device().
284 int ib_register_device(struct ib_device *device,
285 int (*port_callback)(struct ib_device *,
286 u8, struct kobject *))
290 mutex_lock(&device_mutex);
292 if (strchr(device->name, '%')) {
293 ret = alloc_name(device->name);
298 if (ib_device_check_mandatory(device)) {
303 INIT_LIST_HEAD(&device->event_handler_list);
304 INIT_LIST_HEAD(&device->client_data_list);
305 spin_lock_init(&device->event_handler_lock);
306 spin_lock_init(&device->client_data_lock);
308 ret = read_port_immutable(device);
310 printk(KERN_WARNING "Couldn't create per port immutable data %s\n",
315 ret = ib_device_register_sysfs(device, port_callback);
317 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
319 kfree(device->port_immutable);
323 device->reg_state = IB_DEV_REGISTERED;
326 struct ib_client *client;
328 list_for_each_entry(client, &client_list, list)
329 if (client->add && !add_client_context(device, client))
333 down_write(&lists_rwsem);
334 list_add_tail(&device->core_list, &device_list);
335 up_write(&lists_rwsem);
337 mutex_unlock(&device_mutex);
340 EXPORT_SYMBOL(ib_register_device);
343 * ib_unregister_device - Unregister an IB device
344 * @device:Device to unregister
346 * Unregister an IB device. All clients will receive a remove callback.
348 void ib_unregister_device(struct ib_device *device)
350 struct ib_client_data *context, *tmp;
353 mutex_lock(&device_mutex);
355 down_write(&lists_rwsem);
356 list_del(&device->core_list);
357 spin_lock_irqsave(&device->client_data_lock, flags);
358 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
359 context->going_down = true;
360 spin_unlock_irqrestore(&device->client_data_lock, flags);
361 downgrade_write(&lists_rwsem);
363 list_for_each_entry_safe(context, tmp, &device->client_data_list,
365 if (context->client->remove)
366 context->client->remove(device, context->data);
368 up_read(&lists_rwsem);
370 mutex_unlock(&device_mutex);
372 ib_device_unregister_sysfs(device);
374 down_write(&lists_rwsem);
375 spin_lock_irqsave(&device->client_data_lock, flags);
376 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
378 spin_unlock_irqrestore(&device->client_data_lock, flags);
379 up_write(&lists_rwsem);
381 device->reg_state = IB_DEV_UNREGISTERED;
383 EXPORT_SYMBOL(ib_unregister_device);
386 * ib_register_client - Register an IB client
387 * @client:Client to register
389 * Upper level users of the IB drivers can use ib_register_client() to
390 * register callbacks for IB device addition and removal. When an IB
391 * device is added, each registered client's add method will be called
392 * (in the order the clients were registered), and when a device is
393 * removed, each client's remove method will be called (in the reverse
394 * order that clients were registered). In addition, when
395 * ib_register_client() is called, the client will receive an add
396 * callback for all devices already registered.
398 int ib_register_client(struct ib_client *client)
400 struct ib_device *device;
402 mutex_lock(&device_mutex);
404 list_for_each_entry(device, &device_list, core_list)
405 if (client->add && !add_client_context(device, client))
408 down_write(&lists_rwsem);
409 list_add_tail(&client->list, &client_list);
410 up_write(&lists_rwsem);
412 mutex_unlock(&device_mutex);
416 EXPORT_SYMBOL(ib_register_client);
419 * ib_unregister_client - Unregister an IB client
420 * @client:Client to unregister
422 * Upper level users use ib_unregister_client() to remove their client
423 * registration. When ib_unregister_client() is called, the client
424 * will receive a remove callback for each IB device still registered.
426 void ib_unregister_client(struct ib_client *client)
428 struct ib_client_data *context, *tmp;
429 struct ib_device *device;
432 mutex_lock(&device_mutex);
434 down_write(&lists_rwsem);
435 list_del(&client->list);
436 up_write(&lists_rwsem);
438 list_for_each_entry(device, &device_list, core_list) {
439 struct ib_client_data *found_context = NULL;
441 down_write(&lists_rwsem);
442 spin_lock_irqsave(&device->client_data_lock, flags);
443 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
444 if (context->client == client) {
445 context->going_down = true;
446 found_context = context;
449 spin_unlock_irqrestore(&device->client_data_lock, flags);
450 up_write(&lists_rwsem);
453 client->remove(device, found_context ?
454 found_context->data : NULL);
456 if (!found_context) {
457 pr_warn("No client context found for %s/%s\n",
458 device->name, client->name);
462 down_write(&lists_rwsem);
463 spin_lock_irqsave(&device->client_data_lock, flags);
464 list_del(&found_context->list);
465 kfree(found_context);
466 spin_unlock_irqrestore(&device->client_data_lock, flags);
467 up_write(&lists_rwsem);
470 mutex_unlock(&device_mutex);
472 EXPORT_SYMBOL(ib_unregister_client);
475 * ib_get_client_data - Get IB client context
476 * @device:Device to get context for
477 * @client:Client to get context for
479 * ib_get_client_data() returns client context set with
480 * ib_set_client_data().
482 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
484 struct ib_client_data *context;
488 spin_lock_irqsave(&device->client_data_lock, flags);
489 list_for_each_entry(context, &device->client_data_list, list)
490 if (context->client == client) {
494 spin_unlock_irqrestore(&device->client_data_lock, flags);
498 EXPORT_SYMBOL(ib_get_client_data);
501 * ib_set_client_data - Set IB client context
502 * @device:Device to set context for
503 * @client:Client to set context for
504 * @data:Context to set
506 * ib_set_client_data() sets client context that can be retrieved with
507 * ib_get_client_data().
509 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
512 struct ib_client_data *context;
515 spin_lock_irqsave(&device->client_data_lock, flags);
516 list_for_each_entry(context, &device->client_data_list, list)
517 if (context->client == client) {
518 context->data = data;
522 printk(KERN_WARNING "No client context found for %s/%s\n",
523 device->name, client->name);
526 spin_unlock_irqrestore(&device->client_data_lock, flags);
528 EXPORT_SYMBOL(ib_set_client_data);
531 * ib_register_event_handler - Register an IB event handler
532 * @event_handler:Handler to register
534 * ib_register_event_handler() registers an event handler that will be
535 * called back when asynchronous IB events occur (as defined in
536 * chapter 11 of the InfiniBand Architecture Specification). This
537 * callback may occur in interrupt context.
539 int ib_register_event_handler (struct ib_event_handler *event_handler)
543 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
544 list_add_tail(&event_handler->list,
545 &event_handler->device->event_handler_list);
546 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
550 EXPORT_SYMBOL(ib_register_event_handler);
553 * ib_unregister_event_handler - Unregister an event handler
554 * @event_handler:Handler to unregister
556 * Unregister an event handler registered with
557 * ib_register_event_handler().
559 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
563 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
564 list_del(&event_handler->list);
565 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
569 EXPORT_SYMBOL(ib_unregister_event_handler);
572 * ib_dispatch_event - Dispatch an asynchronous event
573 * @event:Event to dispatch
575 * Low-level drivers must call ib_dispatch_event() to dispatch the
576 * event to all registered event handlers when an asynchronous event
579 void ib_dispatch_event(struct ib_event *event)
582 struct ib_event_handler *handler;
584 spin_lock_irqsave(&event->device->event_handler_lock, flags);
586 list_for_each_entry(handler, &event->device->event_handler_list, list)
587 handler->handler(handler, event);
589 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
591 EXPORT_SYMBOL(ib_dispatch_event);
594 * ib_query_device - Query IB device attributes
595 * @device:Device to query
596 * @device_attr:Device attributes
598 * ib_query_device() returns the attributes of a device through the
599 * @device_attr pointer.
601 int ib_query_device(struct ib_device *device,
602 struct ib_device_attr *device_attr)
604 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
606 memset(device_attr, 0, sizeof(*device_attr));
608 return device->query_device(device, device_attr, &uhw);
610 EXPORT_SYMBOL(ib_query_device);
613 * ib_query_port - Query IB port attributes
614 * @device:Device to query
615 * @port_num:Port number to query
616 * @port_attr:Port attributes
618 * ib_query_port() returns the attributes of a port through the
619 * @port_attr pointer.
621 int ib_query_port(struct ib_device *device,
623 struct ib_port_attr *port_attr)
625 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
628 return device->query_port(device, port_num, port_attr);
630 EXPORT_SYMBOL(ib_query_port);
633 * ib_query_gid - Get GID table entry
634 * @device:Device to query
635 * @port_num:Port number to query
636 * @index:GID table index to query
639 * ib_query_gid() fetches the specified GID table entry.
641 int ib_query_gid(struct ib_device *device,
642 u8 port_num, int index, union ib_gid *gid)
644 return device->query_gid(device, port_num, index, gid);
646 EXPORT_SYMBOL(ib_query_gid);
649 * ib_query_pkey - Get P_Key table entry
650 * @device:Device to query
651 * @port_num:Port number to query
652 * @index:P_Key table index to query
653 * @pkey:Returned P_Key
655 * ib_query_pkey() fetches the specified P_Key table entry.
657 int ib_query_pkey(struct ib_device *device,
658 u8 port_num, u16 index, u16 *pkey)
660 return device->query_pkey(device, port_num, index, pkey);
662 EXPORT_SYMBOL(ib_query_pkey);
665 * ib_modify_device - Change IB device attributes
666 * @device:Device to modify
667 * @device_modify_mask:Mask of attributes to change
668 * @device_modify:New attribute values
670 * ib_modify_device() changes a device's attributes as specified by
671 * the @device_modify_mask and @device_modify structure.
673 int ib_modify_device(struct ib_device *device,
674 int device_modify_mask,
675 struct ib_device_modify *device_modify)
677 if (!device->modify_device)
680 return device->modify_device(device, device_modify_mask,
683 EXPORT_SYMBOL(ib_modify_device);
686 * ib_modify_port - Modifies the attributes for the specified port.
687 * @device: The device to modify.
688 * @port_num: The number of the port to modify.
689 * @port_modify_mask: Mask used to specify which attributes of the port
691 * @port_modify: New attribute values for the port.
693 * ib_modify_port() changes a port's attributes as specified by the
694 * @port_modify_mask and @port_modify structure.
696 int ib_modify_port(struct ib_device *device,
697 u8 port_num, int port_modify_mask,
698 struct ib_port_modify *port_modify)
700 if (!device->modify_port)
703 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
706 return device->modify_port(device, port_num, port_modify_mask,
709 EXPORT_SYMBOL(ib_modify_port);
712 * ib_find_gid - Returns the port number and GID table index where
713 * a specified GID value occurs.
714 * @device: The device to query.
715 * @gid: The GID value to search for.
716 * @port_num: The port number of the device where the GID value was found.
717 * @index: The index into the GID table where the GID was found. This
718 * parameter may be NULL.
720 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
721 u8 *port_num, u16 *index)
723 union ib_gid tmp_gid;
726 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
727 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
728 ret = ib_query_gid(device, port, i, &tmp_gid);
731 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
742 EXPORT_SYMBOL(ib_find_gid);
745 * ib_find_pkey - Returns the PKey table index where a specified
747 * @device: The device to query.
748 * @port_num: The port number of the device to search for the PKey.
749 * @pkey: The PKey value to search for.
750 * @index: The index into the PKey table where the PKey was found.
752 int ib_find_pkey(struct ib_device *device,
753 u8 port_num, u16 pkey, u16 *index)
759 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
760 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
763 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
764 /* if there is full-member pkey take it.*/
765 if (tmp_pkey & 0x8000) {
774 /*no full-member, if exists take the limited*/
775 if (partial_ix >= 0) {
781 EXPORT_SYMBOL(ib_find_pkey);
783 static int __init ib_core_init(void)
787 ib_wq = alloc_workqueue("infiniband", 0, 0);
791 ret = ib_sysfs_setup();
793 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
799 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
803 ret = ib_cache_setup();
805 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
818 destroy_workqueue(ib_wq);
822 static void __exit ib_core_cleanup(void)
827 /* Make sure that any pending umem accounting work is done. */
828 destroy_workqueue(ib_wq);
831 module_init(ib_core_init);
832 module_exit(ib_core_cleanup);