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_protocol),
80 IB_MANDATORY_FUNC(query_pkey),
81 IB_MANDATORY_FUNC(query_gid),
82 IB_MANDATORY_FUNC(alloc_pd),
83 IB_MANDATORY_FUNC(dealloc_pd),
84 IB_MANDATORY_FUNC(create_ah),
85 IB_MANDATORY_FUNC(destroy_ah),
86 IB_MANDATORY_FUNC(create_qp),
87 IB_MANDATORY_FUNC(modify_qp),
88 IB_MANDATORY_FUNC(destroy_qp),
89 IB_MANDATORY_FUNC(post_send),
90 IB_MANDATORY_FUNC(post_recv),
91 IB_MANDATORY_FUNC(create_cq),
92 IB_MANDATORY_FUNC(destroy_cq),
93 IB_MANDATORY_FUNC(poll_cq),
94 IB_MANDATORY_FUNC(req_notify_cq),
95 IB_MANDATORY_FUNC(get_dma_mr),
96 IB_MANDATORY_FUNC(dereg_mr),
97 IB_MANDATORY_FUNC(get_port_immutable)
101 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
102 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
103 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
104 device->name, mandatory_table[i].name);
112 static struct ib_device *__ib_device_get_by_name(const char *name)
114 struct ib_device *device;
116 list_for_each_entry(device, &device_list, core_list)
117 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
124 static int alloc_name(char *name)
126 unsigned long *inuse;
127 char buf[IB_DEVICE_NAME_MAX];
128 struct ib_device *device;
131 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
135 list_for_each_entry(device, &device_list, core_list) {
136 if (!sscanf(device->name, name, &i))
138 if (i < 0 || i >= PAGE_SIZE * 8)
140 snprintf(buf, sizeof buf, name, i);
141 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
145 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
146 free_page((unsigned long) inuse);
147 snprintf(buf, sizeof buf, name, i);
149 if (__ib_device_get_by_name(buf))
152 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
157 * ib_alloc_device - allocate an IB device struct
158 * @size:size of structure to allocate
160 * Low-level drivers should use ib_alloc_device() to allocate &struct
161 * ib_device. @size is the size of the structure to be allocated,
162 * including any private data used by the low-level driver.
163 * ib_dealloc_device() must be used to free structures allocated with
166 struct ib_device *ib_alloc_device(size_t size)
168 BUG_ON(size < sizeof (struct ib_device));
170 return kzalloc(size, GFP_KERNEL);
172 EXPORT_SYMBOL(ib_alloc_device);
175 * ib_dealloc_device - free an IB device struct
176 * @device:structure to free
178 * Free a structure allocated with ib_alloc_device().
180 void ib_dealloc_device(struct ib_device *device)
182 if (device->reg_state == IB_DEV_UNINITIALIZED) {
187 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
189 kobject_put(&device->dev.kobj);
191 EXPORT_SYMBOL(ib_dealloc_device);
193 static int add_client_context(struct ib_device *device, struct ib_client *client)
195 struct ib_client_data *context;
198 context = kmalloc(sizeof *context, GFP_KERNEL);
200 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
201 device->name, client->name);
205 context->client = client;
206 context->data = NULL;
208 spin_lock_irqsave(&device->client_data_lock, flags);
209 list_add(&context->list, &device->client_data_list);
210 spin_unlock_irqrestore(&device->client_data_lock, flags);
215 static int read_port_immutable(struct ib_device *device)
218 u8 start_port = rdma_start_port(device);
219 u8 end_port = rdma_end_port(device);
223 * device->port_immutable is indexed directly by the port number to make
224 * access to this data as efficient as possible.
226 * Therefore port_immutable is declared as a 1 based array with
227 * potential empty slots at the beginning.
229 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
232 if (!device->port_immutable)
235 for (port = start_port; port <= end_port; ++port) {
236 ret = device->get_port_immutable(device, port,
237 &device->port_immutable[port]);
245 kfree(device->port_immutable);
251 * ib_register_device - Register an IB device with IB core
252 * @device:Device to register
254 * Low-level drivers use ib_register_device() to register their
255 * devices with the IB core. All registered clients will receive a
256 * callback for each device that is added. @device must be allocated
257 * with ib_alloc_device().
259 int ib_register_device(struct ib_device *device,
260 int (*port_callback)(struct ib_device *,
261 u8, struct kobject *))
265 mutex_lock(&device_mutex);
267 if (strchr(device->name, '%')) {
268 ret = alloc_name(device->name);
273 if (ib_device_check_mandatory(device)) {
278 INIT_LIST_HEAD(&device->event_handler_list);
279 INIT_LIST_HEAD(&device->client_data_list);
280 spin_lock_init(&device->event_handler_lock);
281 spin_lock_init(&device->client_data_lock);
283 ret = read_port_immutable(device);
285 printk(KERN_WARNING "Couldn't create per port immutable data %s\n",
290 ret = ib_device_register_sysfs(device, port_callback);
292 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
294 kfree(device->port_immutable);
298 list_add_tail(&device->core_list, &device_list);
300 device->reg_state = IB_DEV_REGISTERED;
303 struct ib_client *client;
305 list_for_each_entry(client, &client_list, list)
306 if (client->add && !add_client_context(device, client))
311 mutex_unlock(&device_mutex);
314 EXPORT_SYMBOL(ib_register_device);
317 * ib_unregister_device - Unregister an IB device
318 * @device:Device to unregister
320 * Unregister an IB device. All clients will receive a remove callback.
322 void ib_unregister_device(struct ib_device *device)
324 struct ib_client *client;
325 struct ib_client_data *context, *tmp;
328 mutex_lock(&device_mutex);
330 list_for_each_entry_reverse(client, &client_list, list)
332 client->remove(device);
334 list_del(&device->core_list);
336 mutex_unlock(&device_mutex);
338 ib_device_unregister_sysfs(device);
340 spin_lock_irqsave(&device->client_data_lock, flags);
341 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
343 spin_unlock_irqrestore(&device->client_data_lock, flags);
345 device->reg_state = IB_DEV_UNREGISTERED;
347 EXPORT_SYMBOL(ib_unregister_device);
350 * ib_register_client - Register an IB client
351 * @client:Client to register
353 * Upper level users of the IB drivers can use ib_register_client() to
354 * register callbacks for IB device addition and removal. When an IB
355 * device is added, each registered client's add method will be called
356 * (in the order the clients were registered), and when a device is
357 * removed, each client's remove method will be called (in the reverse
358 * order that clients were registered). In addition, when
359 * ib_register_client() is called, the client will receive an add
360 * callback for all devices already registered.
362 int ib_register_client(struct ib_client *client)
364 struct ib_device *device;
366 mutex_lock(&device_mutex);
368 list_add_tail(&client->list, &client_list);
369 list_for_each_entry(device, &device_list, core_list)
370 if (client->add && !add_client_context(device, client))
373 mutex_unlock(&device_mutex);
377 EXPORT_SYMBOL(ib_register_client);
380 * ib_unregister_client - Unregister an IB client
381 * @client:Client to unregister
383 * Upper level users use ib_unregister_client() to remove their client
384 * registration. When ib_unregister_client() is called, the client
385 * will receive a remove callback for each IB device still registered.
387 void ib_unregister_client(struct ib_client *client)
389 struct ib_client_data *context, *tmp;
390 struct ib_device *device;
393 mutex_lock(&device_mutex);
395 list_for_each_entry(device, &device_list, core_list) {
397 client->remove(device);
399 spin_lock_irqsave(&device->client_data_lock, flags);
400 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
401 if (context->client == client) {
402 list_del(&context->list);
405 spin_unlock_irqrestore(&device->client_data_lock, flags);
407 list_del(&client->list);
409 mutex_unlock(&device_mutex);
411 EXPORT_SYMBOL(ib_unregister_client);
414 * ib_get_client_data - Get IB client context
415 * @device:Device to get context for
416 * @client:Client to get context for
418 * ib_get_client_data() returns client context set with
419 * ib_set_client_data().
421 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
423 struct ib_client_data *context;
427 spin_lock_irqsave(&device->client_data_lock, flags);
428 list_for_each_entry(context, &device->client_data_list, list)
429 if (context->client == client) {
433 spin_unlock_irqrestore(&device->client_data_lock, flags);
437 EXPORT_SYMBOL(ib_get_client_data);
440 * ib_set_client_data - Set IB client context
441 * @device:Device to set context for
442 * @client:Client to set context for
443 * @data:Context to set
445 * ib_set_client_data() sets client context that can be retrieved with
446 * ib_get_client_data().
448 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
451 struct ib_client_data *context;
454 spin_lock_irqsave(&device->client_data_lock, flags);
455 list_for_each_entry(context, &device->client_data_list, list)
456 if (context->client == client) {
457 context->data = data;
461 printk(KERN_WARNING "No client context found for %s/%s\n",
462 device->name, client->name);
465 spin_unlock_irqrestore(&device->client_data_lock, flags);
467 EXPORT_SYMBOL(ib_set_client_data);
470 * ib_register_event_handler - Register an IB event handler
471 * @event_handler:Handler to register
473 * ib_register_event_handler() registers an event handler that will be
474 * called back when asynchronous IB events occur (as defined in
475 * chapter 11 of the InfiniBand Architecture Specification). This
476 * callback may occur in interrupt context.
478 int ib_register_event_handler (struct ib_event_handler *event_handler)
482 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
483 list_add_tail(&event_handler->list,
484 &event_handler->device->event_handler_list);
485 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
489 EXPORT_SYMBOL(ib_register_event_handler);
492 * ib_unregister_event_handler - Unregister an event handler
493 * @event_handler:Handler to unregister
495 * Unregister an event handler registered with
496 * ib_register_event_handler().
498 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
502 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
503 list_del(&event_handler->list);
504 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
508 EXPORT_SYMBOL(ib_unregister_event_handler);
511 * ib_dispatch_event - Dispatch an asynchronous event
512 * @event:Event to dispatch
514 * Low-level drivers must call ib_dispatch_event() to dispatch the
515 * event to all registered event handlers when an asynchronous event
518 void ib_dispatch_event(struct ib_event *event)
521 struct ib_event_handler *handler;
523 spin_lock_irqsave(&event->device->event_handler_lock, flags);
525 list_for_each_entry(handler, &event->device->event_handler_list, list)
526 handler->handler(handler, event);
528 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
530 EXPORT_SYMBOL(ib_dispatch_event);
533 * ib_query_device - Query IB device attributes
534 * @device:Device to query
535 * @device_attr:Device attributes
537 * ib_query_device() returns the attributes of a device through the
538 * @device_attr pointer.
540 int ib_query_device(struct ib_device *device,
541 struct ib_device_attr *device_attr)
543 return device->query_device(device, device_attr);
545 EXPORT_SYMBOL(ib_query_device);
548 * ib_query_port - Query IB port attributes
549 * @device:Device to query
550 * @port_num:Port number to query
551 * @port_attr:Port attributes
553 * ib_query_port() returns the attributes of a port through the
554 * @port_attr pointer.
556 int ib_query_port(struct ib_device *device,
558 struct ib_port_attr *port_attr)
560 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
563 return device->query_port(device, port_num, port_attr);
565 EXPORT_SYMBOL(ib_query_port);
568 * ib_query_gid - Get GID table entry
569 * @device:Device to query
570 * @port_num:Port number to query
571 * @index:GID table index to query
574 * ib_query_gid() fetches the specified GID table entry.
576 int ib_query_gid(struct ib_device *device,
577 u8 port_num, int index, union ib_gid *gid)
579 return device->query_gid(device, port_num, index, gid);
581 EXPORT_SYMBOL(ib_query_gid);
584 * ib_query_pkey - Get P_Key table entry
585 * @device:Device to query
586 * @port_num:Port number to query
587 * @index:P_Key table index to query
588 * @pkey:Returned P_Key
590 * ib_query_pkey() fetches the specified P_Key table entry.
592 int ib_query_pkey(struct ib_device *device,
593 u8 port_num, u16 index, u16 *pkey)
595 return device->query_pkey(device, port_num, index, pkey);
597 EXPORT_SYMBOL(ib_query_pkey);
600 * ib_modify_device - Change IB device attributes
601 * @device:Device to modify
602 * @device_modify_mask:Mask of attributes to change
603 * @device_modify:New attribute values
605 * ib_modify_device() changes a device's attributes as specified by
606 * the @device_modify_mask and @device_modify structure.
608 int ib_modify_device(struct ib_device *device,
609 int device_modify_mask,
610 struct ib_device_modify *device_modify)
612 if (!device->modify_device)
615 return device->modify_device(device, device_modify_mask,
618 EXPORT_SYMBOL(ib_modify_device);
621 * ib_modify_port - Modifies the attributes for the specified port.
622 * @device: The device to modify.
623 * @port_num: The number of the port to modify.
624 * @port_modify_mask: Mask used to specify which attributes of the port
626 * @port_modify: New attribute values for the port.
628 * ib_modify_port() changes a port's attributes as specified by the
629 * @port_modify_mask and @port_modify structure.
631 int ib_modify_port(struct ib_device *device,
632 u8 port_num, int port_modify_mask,
633 struct ib_port_modify *port_modify)
635 if (!device->modify_port)
638 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
641 return device->modify_port(device, port_num, port_modify_mask,
644 EXPORT_SYMBOL(ib_modify_port);
647 * ib_find_gid - Returns the port number and GID table index where
648 * a specified GID value occurs.
649 * @device: The device to query.
650 * @gid: The GID value to search for.
651 * @port_num: The port number of the device where the GID value was found.
652 * @index: The index into the GID table where the GID was found. This
653 * parameter may be NULL.
655 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
656 u8 *port_num, u16 *index)
658 union ib_gid tmp_gid;
661 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
662 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
663 ret = ib_query_gid(device, port, i, &tmp_gid);
666 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
677 EXPORT_SYMBOL(ib_find_gid);
680 * ib_find_pkey - Returns the PKey table index where a specified
682 * @device: The device to query.
683 * @port_num: The port number of the device to search for the PKey.
684 * @pkey: The PKey value to search for.
685 * @index: The index into the PKey table where the PKey was found.
687 int ib_find_pkey(struct ib_device *device,
688 u8 port_num, u16 pkey, u16 *index)
694 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
695 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
698 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
699 /* if there is full-member pkey take it.*/
700 if (tmp_pkey & 0x8000) {
709 /*no full-member, if exists take the limited*/
710 if (partial_ix >= 0) {
716 EXPORT_SYMBOL(ib_find_pkey);
718 static int __init ib_core_init(void)
722 ib_wq = alloc_workqueue("infiniband", 0, 0);
726 ret = ib_sysfs_setup();
728 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
734 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
738 ret = ib_cache_setup();
740 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
753 destroy_workqueue(ib_wq);
757 static void __exit ib_core_cleanup(void)
762 /* Make sure that any pending umem accounting work is done. */
763 destroy_workqueue(ib_wq);
766 module_init(ib_core_init);
767 module_exit(ib_core_cleanup);