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;
55 struct workqueue_struct *ib_wq;
56 EXPORT_SYMBOL_GPL(ib_wq);
58 static LIST_HEAD(device_list);
59 static LIST_HEAD(client_list);
62 * device_mutex protects access to both device_list and client_list.
63 * There's no real point to using multiple locks or something fancier
64 * like an rwsem: we always access both lists, and we're always
65 * modifying one list or the other list. In any case this is not a
66 * hot path so there's no point in trying to optimize.
68 static DEFINE_MUTEX(device_mutex);
70 static int ib_device_check_mandatory(struct ib_device *device)
72 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
76 } mandatory_table[] = {
77 IB_MANDATORY_FUNC(query_device),
78 IB_MANDATORY_FUNC(query_port),
79 IB_MANDATORY_FUNC(query_pkey),
80 IB_MANDATORY_FUNC(query_gid),
81 IB_MANDATORY_FUNC(alloc_pd),
82 IB_MANDATORY_FUNC(dealloc_pd),
83 IB_MANDATORY_FUNC(create_ah),
84 IB_MANDATORY_FUNC(destroy_ah),
85 IB_MANDATORY_FUNC(create_qp),
86 IB_MANDATORY_FUNC(modify_qp),
87 IB_MANDATORY_FUNC(destroy_qp),
88 IB_MANDATORY_FUNC(post_send),
89 IB_MANDATORY_FUNC(post_recv),
90 IB_MANDATORY_FUNC(create_cq),
91 IB_MANDATORY_FUNC(destroy_cq),
92 IB_MANDATORY_FUNC(poll_cq),
93 IB_MANDATORY_FUNC(req_notify_cq),
94 IB_MANDATORY_FUNC(get_dma_mr),
95 IB_MANDATORY_FUNC(dereg_mr),
96 IB_MANDATORY_FUNC(get_port_immutable)
100 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
101 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
102 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
103 device->name, mandatory_table[i].name);
111 static struct ib_device *__ib_device_get_by_name(const char *name)
113 struct ib_device *device;
115 list_for_each_entry(device, &device_list, core_list)
116 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
123 static int alloc_name(char *name)
125 unsigned long *inuse;
126 char buf[IB_DEVICE_NAME_MAX];
127 struct ib_device *device;
130 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
134 list_for_each_entry(device, &device_list, core_list) {
135 if (!sscanf(device->name, name, &i))
137 if (i < 0 || i >= PAGE_SIZE * 8)
139 snprintf(buf, sizeof buf, name, i);
140 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
144 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
145 free_page((unsigned long) inuse);
146 snprintf(buf, sizeof buf, name, i);
148 if (__ib_device_get_by_name(buf))
151 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
156 * ib_alloc_device - allocate an IB device struct
157 * @size:size of structure to allocate
159 * Low-level drivers should use ib_alloc_device() to allocate &struct
160 * ib_device. @size is the size of the structure to be allocated,
161 * including any private data used by the low-level driver.
162 * ib_dealloc_device() must be used to free structures allocated with
165 struct ib_device *ib_alloc_device(size_t size)
167 BUG_ON(size < sizeof (struct ib_device));
169 return kzalloc(size, GFP_KERNEL);
171 EXPORT_SYMBOL(ib_alloc_device);
174 * ib_dealloc_device - free an IB device struct
175 * @device:structure to free
177 * Free a structure allocated with ib_alloc_device().
179 void ib_dealloc_device(struct ib_device *device)
181 if (device->reg_state == IB_DEV_UNINITIALIZED) {
186 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
188 kobject_put(&device->dev.kobj);
190 EXPORT_SYMBOL(ib_dealloc_device);
192 static int add_client_context(struct ib_device *device, struct ib_client *client)
194 struct ib_client_data *context;
197 context = kmalloc(sizeof *context, GFP_KERNEL);
199 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
200 device->name, client->name);
204 context->client = client;
205 context->data = NULL;
207 spin_lock_irqsave(&device->client_data_lock, flags);
208 list_add(&context->list, &device->client_data_list);
209 spin_unlock_irqrestore(&device->client_data_lock, flags);
214 static int verify_immutable(const struct ib_device *dev, u8 port)
216 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
217 rdma_max_mad_size(dev, port) != 0);
220 static int read_port_immutable(struct ib_device *device)
223 u8 start_port = rdma_start_port(device);
224 u8 end_port = rdma_end_port(device);
228 * device->port_immutable is indexed directly by the port number to make
229 * access to this data as efficient as possible.
231 * Therefore port_immutable is declared as a 1 based array with
232 * potential empty slots at the beginning.
234 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
237 if (!device->port_immutable)
240 for (port = start_port; port <= end_port; ++port) {
241 ret = device->get_port_immutable(device, port,
242 &device->port_immutable[port]);
246 if (verify_immutable(device, port)) {
255 kfree(device->port_immutable);
261 * ib_register_device - Register an IB device with IB core
262 * @device:Device to register
264 * Low-level drivers use ib_register_device() to register their
265 * devices with the IB core. All registered clients will receive a
266 * callback for each device that is added. @device must be allocated
267 * with ib_alloc_device().
269 int ib_register_device(struct ib_device *device,
270 int (*port_callback)(struct ib_device *,
271 u8, struct kobject *))
275 mutex_lock(&device_mutex);
277 if (strchr(device->name, '%')) {
278 ret = alloc_name(device->name);
283 if (ib_device_check_mandatory(device)) {
288 INIT_LIST_HEAD(&device->event_handler_list);
289 INIT_LIST_HEAD(&device->client_data_list);
290 spin_lock_init(&device->event_handler_lock);
291 spin_lock_init(&device->client_data_lock);
293 ret = read_port_immutable(device);
295 printk(KERN_WARNING "Couldn't create per port immutable data %s\n",
300 ret = ib_device_register_sysfs(device, port_callback);
302 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
304 kfree(device->port_immutable);
308 list_add_tail(&device->core_list, &device_list);
310 device->reg_state = IB_DEV_REGISTERED;
313 struct ib_client *client;
315 list_for_each_entry(client, &client_list, list)
316 if (client->add && !add_client_context(device, client))
321 mutex_unlock(&device_mutex);
324 EXPORT_SYMBOL(ib_register_device);
327 * ib_unregister_device - Unregister an IB device
328 * @device:Device to unregister
330 * Unregister an IB device. All clients will receive a remove callback.
332 void ib_unregister_device(struct ib_device *device)
334 struct ib_client *client;
335 struct ib_client_data *context, *tmp;
338 mutex_lock(&device_mutex);
340 list_for_each_entry_reverse(client, &client_list, list)
342 client->remove(device);
344 list_del(&device->core_list);
346 mutex_unlock(&device_mutex);
348 ib_device_unregister_sysfs(device);
350 spin_lock_irqsave(&device->client_data_lock, flags);
351 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
353 spin_unlock_irqrestore(&device->client_data_lock, flags);
355 device->reg_state = IB_DEV_UNREGISTERED;
357 EXPORT_SYMBOL(ib_unregister_device);
360 * ib_register_client - Register an IB client
361 * @client:Client to register
363 * Upper level users of the IB drivers can use ib_register_client() to
364 * register callbacks for IB device addition and removal. When an IB
365 * device is added, each registered client's add method will be called
366 * (in the order the clients were registered), and when a device is
367 * removed, each client's remove method will be called (in the reverse
368 * order that clients were registered). In addition, when
369 * ib_register_client() is called, the client will receive an add
370 * callback for all devices already registered.
372 int ib_register_client(struct ib_client *client)
374 struct ib_device *device;
376 mutex_lock(&device_mutex);
378 list_add_tail(&client->list, &client_list);
379 list_for_each_entry(device, &device_list, core_list)
380 if (client->add && !add_client_context(device, client))
383 mutex_unlock(&device_mutex);
387 EXPORT_SYMBOL(ib_register_client);
390 * ib_unregister_client - Unregister an IB client
391 * @client:Client to unregister
393 * Upper level users use ib_unregister_client() to remove their client
394 * registration. When ib_unregister_client() is called, the client
395 * will receive a remove callback for each IB device still registered.
397 void ib_unregister_client(struct ib_client *client)
399 struct ib_client_data *context, *tmp;
400 struct ib_device *device;
403 mutex_lock(&device_mutex);
405 list_for_each_entry(device, &device_list, core_list) {
407 client->remove(device);
409 spin_lock_irqsave(&device->client_data_lock, flags);
410 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
411 if (context->client == client) {
412 list_del(&context->list);
415 spin_unlock_irqrestore(&device->client_data_lock, flags);
417 list_del(&client->list);
419 mutex_unlock(&device_mutex);
421 EXPORT_SYMBOL(ib_unregister_client);
424 * ib_get_client_data - Get IB client context
425 * @device:Device to get context for
426 * @client:Client to get context for
428 * ib_get_client_data() returns client context set with
429 * ib_set_client_data().
431 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
433 struct ib_client_data *context;
437 spin_lock_irqsave(&device->client_data_lock, flags);
438 list_for_each_entry(context, &device->client_data_list, list)
439 if (context->client == client) {
443 spin_unlock_irqrestore(&device->client_data_lock, flags);
447 EXPORT_SYMBOL(ib_get_client_data);
450 * ib_set_client_data - Set IB client context
451 * @device:Device to set context for
452 * @client:Client to set context for
453 * @data:Context to set
455 * ib_set_client_data() sets client context that can be retrieved with
456 * ib_get_client_data().
458 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
461 struct ib_client_data *context;
464 spin_lock_irqsave(&device->client_data_lock, flags);
465 list_for_each_entry(context, &device->client_data_list, list)
466 if (context->client == client) {
467 context->data = data;
471 printk(KERN_WARNING "No client context found for %s/%s\n",
472 device->name, client->name);
475 spin_unlock_irqrestore(&device->client_data_lock, flags);
477 EXPORT_SYMBOL(ib_set_client_data);
480 * ib_register_event_handler - Register an IB event handler
481 * @event_handler:Handler to register
483 * ib_register_event_handler() registers an event handler that will be
484 * called back when asynchronous IB events occur (as defined in
485 * chapter 11 of the InfiniBand Architecture Specification). This
486 * callback may occur in interrupt context.
488 int ib_register_event_handler (struct ib_event_handler *event_handler)
492 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
493 list_add_tail(&event_handler->list,
494 &event_handler->device->event_handler_list);
495 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
499 EXPORT_SYMBOL(ib_register_event_handler);
502 * ib_unregister_event_handler - Unregister an event handler
503 * @event_handler:Handler to unregister
505 * Unregister an event handler registered with
506 * ib_register_event_handler().
508 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
512 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
513 list_del(&event_handler->list);
514 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
518 EXPORT_SYMBOL(ib_unregister_event_handler);
521 * ib_dispatch_event - Dispatch an asynchronous event
522 * @event:Event to dispatch
524 * Low-level drivers must call ib_dispatch_event() to dispatch the
525 * event to all registered event handlers when an asynchronous event
528 void ib_dispatch_event(struct ib_event *event)
531 struct ib_event_handler *handler;
533 spin_lock_irqsave(&event->device->event_handler_lock, flags);
535 list_for_each_entry(handler, &event->device->event_handler_list, list)
536 handler->handler(handler, event);
538 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
540 EXPORT_SYMBOL(ib_dispatch_event);
543 * ib_query_device - Query IB device attributes
544 * @device:Device to query
545 * @device_attr:Device attributes
547 * ib_query_device() returns the attributes of a device through the
548 * @device_attr pointer.
550 int ib_query_device(struct ib_device *device,
551 struct ib_device_attr *device_attr)
553 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
555 memset(device_attr, 0, sizeof(*device_attr));
557 return device->query_device(device, device_attr, &uhw);
559 EXPORT_SYMBOL(ib_query_device);
562 * ib_query_port - Query IB port attributes
563 * @device:Device to query
564 * @port_num:Port number to query
565 * @port_attr:Port attributes
567 * ib_query_port() returns the attributes of a port through the
568 * @port_attr pointer.
570 int ib_query_port(struct ib_device *device,
572 struct ib_port_attr *port_attr)
574 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
577 return device->query_port(device, port_num, port_attr);
579 EXPORT_SYMBOL(ib_query_port);
582 * ib_query_gid - Get GID table entry
583 * @device:Device to query
584 * @port_num:Port number to query
585 * @index:GID table index to query
588 * ib_query_gid() fetches the specified GID table entry.
590 int ib_query_gid(struct ib_device *device,
591 u8 port_num, int index, union ib_gid *gid)
593 return device->query_gid(device, port_num, index, gid);
595 EXPORT_SYMBOL(ib_query_gid);
598 * ib_query_pkey - Get P_Key table entry
599 * @device:Device to query
600 * @port_num:Port number to query
601 * @index:P_Key table index to query
602 * @pkey:Returned P_Key
604 * ib_query_pkey() fetches the specified P_Key table entry.
606 int ib_query_pkey(struct ib_device *device,
607 u8 port_num, u16 index, u16 *pkey)
609 return device->query_pkey(device, port_num, index, pkey);
611 EXPORT_SYMBOL(ib_query_pkey);
614 * ib_modify_device - Change IB device attributes
615 * @device:Device to modify
616 * @device_modify_mask:Mask of attributes to change
617 * @device_modify:New attribute values
619 * ib_modify_device() changes a device's attributes as specified by
620 * the @device_modify_mask and @device_modify structure.
622 int ib_modify_device(struct ib_device *device,
623 int device_modify_mask,
624 struct ib_device_modify *device_modify)
626 if (!device->modify_device)
629 return device->modify_device(device, device_modify_mask,
632 EXPORT_SYMBOL(ib_modify_device);
635 * ib_modify_port - Modifies the attributes for the specified port.
636 * @device: The device to modify.
637 * @port_num: The number of the port to modify.
638 * @port_modify_mask: Mask used to specify which attributes of the port
640 * @port_modify: New attribute values for the port.
642 * ib_modify_port() changes a port's attributes as specified by the
643 * @port_modify_mask and @port_modify structure.
645 int ib_modify_port(struct ib_device *device,
646 u8 port_num, int port_modify_mask,
647 struct ib_port_modify *port_modify)
649 if (!device->modify_port)
652 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
655 return device->modify_port(device, port_num, port_modify_mask,
658 EXPORT_SYMBOL(ib_modify_port);
661 * ib_find_gid - Returns the port number and GID table index where
662 * a specified GID value occurs.
663 * @device: The device to query.
664 * @gid: The GID value to search for.
665 * @port_num: The port number of the device where the GID value was found.
666 * @index: The index into the GID table where the GID was found. This
667 * parameter may be NULL.
669 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
670 u8 *port_num, u16 *index)
672 union ib_gid tmp_gid;
675 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
676 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
677 ret = ib_query_gid(device, port, i, &tmp_gid);
680 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
691 EXPORT_SYMBOL(ib_find_gid);
694 * ib_find_pkey - Returns the PKey table index where a specified
696 * @device: The device to query.
697 * @port_num: The port number of the device to search for the PKey.
698 * @pkey: The PKey value to search for.
699 * @index: The index into the PKey table where the PKey was found.
701 int ib_find_pkey(struct ib_device *device,
702 u8 port_num, u16 pkey, u16 *index)
708 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
709 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
712 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
713 /* if there is full-member pkey take it.*/
714 if (tmp_pkey & 0x8000) {
723 /*no full-member, if exists take the limited*/
724 if (partial_ix >= 0) {
730 EXPORT_SYMBOL(ib_find_pkey);
732 static int __init ib_core_init(void)
736 ib_wq = alloc_workqueue("infiniband", 0, 0);
740 ret = ib_sysfs_setup();
742 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
748 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
752 ret = ib_cache_setup();
754 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
767 destroy_workqueue(ib_wq);
771 static void __exit ib_core_cleanup(void)
776 /* Make sure that any pending umem accounting work is done. */
777 destroy_workqueue(ib_wq);
780 module_init(ib_core_init);
781 module_exit(ib_core_cleanup);