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)
99 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
100 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
101 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
102 device->name, mandatory_table[i].name);
110 static struct ib_device *__ib_device_get_by_name(const char *name)
112 struct ib_device *device;
114 list_for_each_entry(device, &device_list, core_list)
115 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
122 static int alloc_name(char *name)
124 unsigned long *inuse;
125 char buf[IB_DEVICE_NAME_MAX];
126 struct ib_device *device;
129 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
133 list_for_each_entry(device, &device_list, core_list) {
134 if (!sscanf(device->name, name, &i))
136 if (i < 0 || i >= PAGE_SIZE * 8)
138 snprintf(buf, sizeof buf, name, i);
139 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
143 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
144 free_page((unsigned long) inuse);
145 snprintf(buf, sizeof buf, name, i);
147 if (__ib_device_get_by_name(buf))
150 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
155 * ib_alloc_device - allocate an IB device struct
156 * @size:size of structure to allocate
158 * Low-level drivers should use ib_alloc_device() to allocate &struct
159 * ib_device. @size is the size of the structure to be allocated,
160 * including any private data used by the low-level driver.
161 * ib_dealloc_device() must be used to free structures allocated with
164 struct ib_device *ib_alloc_device(size_t size)
166 BUG_ON(size < sizeof (struct ib_device));
168 return kzalloc(size, GFP_KERNEL);
170 EXPORT_SYMBOL(ib_alloc_device);
173 * ib_dealloc_device - free an IB device struct
174 * @device:structure to free
176 * Free a structure allocated with ib_alloc_device().
178 void ib_dealloc_device(struct ib_device *device)
180 if (device->reg_state == IB_DEV_UNINITIALIZED) {
185 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
187 kobject_put(&device->dev.kobj);
189 EXPORT_SYMBOL(ib_dealloc_device);
191 static int add_client_context(struct ib_device *device, struct ib_client *client)
193 struct ib_client_data *context;
196 context = kmalloc(sizeof *context, GFP_KERNEL);
198 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
199 device->name, client->name);
203 context->client = client;
204 context->data = NULL;
206 spin_lock_irqsave(&device->client_data_lock, flags);
207 list_add(&context->list, &device->client_data_list);
208 spin_unlock_irqrestore(&device->client_data_lock, flags);
213 static int read_port_table_lengths(struct ib_device *device)
215 struct ib_port_attr *tprops = NULL;
216 int num_ports, ret = -ENOMEM;
219 tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
223 num_ports = rdma_end_port(device) - rdma_start_port(device) + 1;
225 device->pkey_tbl_len = kmalloc(sizeof *device->pkey_tbl_len * num_ports,
227 device->gid_tbl_len = kmalloc(sizeof *device->gid_tbl_len * num_ports,
229 if (!device->pkey_tbl_len || !device->gid_tbl_len)
232 for (port_index = 0; port_index < num_ports; ++port_index) {
233 ret = ib_query_port(device, port_index + rdma_start_port(device),
237 device->pkey_tbl_len[port_index] = tprops->pkey_tbl_len;
238 device->gid_tbl_len[port_index] = tprops->gid_tbl_len;
245 kfree(device->gid_tbl_len);
246 kfree(device->pkey_tbl_len);
253 * ib_register_device - Register an IB device with IB core
254 * @device:Device to register
256 * Low-level drivers use ib_register_device() to register their
257 * devices with the IB core. All registered clients will receive a
258 * callback for each device that is added. @device must be allocated
259 * with ib_alloc_device().
261 int ib_register_device(struct ib_device *device,
262 int (*port_callback)(struct ib_device *,
263 u8, struct kobject *))
267 mutex_lock(&device_mutex);
269 if (strchr(device->name, '%')) {
270 ret = alloc_name(device->name);
275 if (ib_device_check_mandatory(device)) {
280 INIT_LIST_HEAD(&device->event_handler_list);
281 INIT_LIST_HEAD(&device->client_data_list);
282 spin_lock_init(&device->event_handler_lock);
283 spin_lock_init(&device->client_data_lock);
285 ret = read_port_table_lengths(device);
287 printk(KERN_WARNING "Couldn't create table lengths cache for device %s\n",
292 ret = ib_device_register_sysfs(device, port_callback);
294 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
296 kfree(device->gid_tbl_len);
297 kfree(device->pkey_tbl_len);
301 list_add_tail(&device->core_list, &device_list);
303 device->reg_state = IB_DEV_REGISTERED;
306 struct ib_client *client;
308 list_for_each_entry(client, &client_list, list)
309 if (client->add && !add_client_context(device, client))
314 mutex_unlock(&device_mutex);
317 EXPORT_SYMBOL(ib_register_device);
320 * ib_unregister_device - Unregister an IB device
321 * @device:Device to unregister
323 * Unregister an IB device. All clients will receive a remove callback.
325 void ib_unregister_device(struct ib_device *device)
327 struct ib_client *client;
328 struct ib_client_data *context, *tmp;
331 mutex_lock(&device_mutex);
333 list_for_each_entry_reverse(client, &client_list, list)
335 client->remove(device);
337 list_del(&device->core_list);
339 kfree(device->gid_tbl_len);
340 kfree(device->pkey_tbl_len);
342 mutex_unlock(&device_mutex);
344 ib_device_unregister_sysfs(device);
346 spin_lock_irqsave(&device->client_data_lock, flags);
347 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
349 spin_unlock_irqrestore(&device->client_data_lock, flags);
351 device->reg_state = IB_DEV_UNREGISTERED;
353 EXPORT_SYMBOL(ib_unregister_device);
356 * ib_register_client - Register an IB client
357 * @client:Client to register
359 * Upper level users of the IB drivers can use ib_register_client() to
360 * register callbacks for IB device addition and removal. When an IB
361 * device is added, each registered client's add method will be called
362 * (in the order the clients were registered), and when a device is
363 * removed, each client's remove method will be called (in the reverse
364 * order that clients were registered). In addition, when
365 * ib_register_client() is called, the client will receive an add
366 * callback for all devices already registered.
368 int ib_register_client(struct ib_client *client)
370 struct ib_device *device;
372 mutex_lock(&device_mutex);
374 list_add_tail(&client->list, &client_list);
375 list_for_each_entry(device, &device_list, core_list)
376 if (client->add && !add_client_context(device, client))
379 mutex_unlock(&device_mutex);
383 EXPORT_SYMBOL(ib_register_client);
386 * ib_unregister_client - Unregister an IB client
387 * @client:Client to unregister
389 * Upper level users use ib_unregister_client() to remove their client
390 * registration. When ib_unregister_client() is called, the client
391 * will receive a remove callback for each IB device still registered.
393 void ib_unregister_client(struct ib_client *client)
395 struct ib_client_data *context, *tmp;
396 struct ib_device *device;
399 mutex_lock(&device_mutex);
401 list_for_each_entry(device, &device_list, core_list) {
403 client->remove(device);
405 spin_lock_irqsave(&device->client_data_lock, flags);
406 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
407 if (context->client == client) {
408 list_del(&context->list);
411 spin_unlock_irqrestore(&device->client_data_lock, flags);
413 list_del(&client->list);
415 mutex_unlock(&device_mutex);
417 EXPORT_SYMBOL(ib_unregister_client);
420 * ib_get_client_data - Get IB client context
421 * @device:Device to get context for
422 * @client:Client to get context for
424 * ib_get_client_data() returns client context set with
425 * ib_set_client_data().
427 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
429 struct ib_client_data *context;
433 spin_lock_irqsave(&device->client_data_lock, flags);
434 list_for_each_entry(context, &device->client_data_list, list)
435 if (context->client == client) {
439 spin_unlock_irqrestore(&device->client_data_lock, flags);
443 EXPORT_SYMBOL(ib_get_client_data);
446 * ib_set_client_data - Set IB client context
447 * @device:Device to set context for
448 * @client:Client to set context for
449 * @data:Context to set
451 * ib_set_client_data() sets client context that can be retrieved with
452 * ib_get_client_data().
454 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
457 struct ib_client_data *context;
460 spin_lock_irqsave(&device->client_data_lock, flags);
461 list_for_each_entry(context, &device->client_data_list, list)
462 if (context->client == client) {
463 context->data = data;
467 printk(KERN_WARNING "No client context found for %s/%s\n",
468 device->name, client->name);
471 spin_unlock_irqrestore(&device->client_data_lock, flags);
473 EXPORT_SYMBOL(ib_set_client_data);
476 * ib_register_event_handler - Register an IB event handler
477 * @event_handler:Handler to register
479 * ib_register_event_handler() registers an event handler that will be
480 * called back when asynchronous IB events occur (as defined in
481 * chapter 11 of the InfiniBand Architecture Specification). This
482 * callback may occur in interrupt context.
484 int ib_register_event_handler (struct ib_event_handler *event_handler)
488 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
489 list_add_tail(&event_handler->list,
490 &event_handler->device->event_handler_list);
491 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
495 EXPORT_SYMBOL(ib_register_event_handler);
498 * ib_unregister_event_handler - Unregister an event handler
499 * @event_handler:Handler to unregister
501 * Unregister an event handler registered with
502 * ib_register_event_handler().
504 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
508 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
509 list_del(&event_handler->list);
510 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
514 EXPORT_SYMBOL(ib_unregister_event_handler);
517 * ib_dispatch_event - Dispatch an asynchronous event
518 * @event:Event to dispatch
520 * Low-level drivers must call ib_dispatch_event() to dispatch the
521 * event to all registered event handlers when an asynchronous event
524 void ib_dispatch_event(struct ib_event *event)
527 struct ib_event_handler *handler;
529 spin_lock_irqsave(&event->device->event_handler_lock, flags);
531 list_for_each_entry(handler, &event->device->event_handler_list, list)
532 handler->handler(handler, event);
534 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
536 EXPORT_SYMBOL(ib_dispatch_event);
539 * ib_query_device - Query IB device attributes
540 * @device:Device to query
541 * @device_attr:Device attributes
543 * ib_query_device() returns the attributes of a device through the
544 * @device_attr pointer.
546 int ib_query_device(struct ib_device *device,
547 struct ib_device_attr *device_attr)
549 return device->query_device(device, device_attr);
551 EXPORT_SYMBOL(ib_query_device);
554 * ib_query_port - Query IB port attributes
555 * @device:Device to query
556 * @port_num:Port number to query
557 * @port_attr:Port attributes
559 * ib_query_port() returns the attributes of a port through the
560 * @port_attr pointer.
562 int ib_query_port(struct ib_device *device,
564 struct ib_port_attr *port_attr)
566 if (!rdma_is_port_valid(device, port_num))
569 return device->query_port(device, port_num, port_attr);
571 EXPORT_SYMBOL(ib_query_port);
574 * ib_query_gid - Get GID table entry
575 * @device:Device to query
576 * @port_num:Port number to query
577 * @index:GID table index to query
580 * ib_query_gid() fetches the specified GID table entry.
582 int ib_query_gid(struct ib_device *device,
583 u8 port_num, int index, union ib_gid *gid)
585 return device->query_gid(device, port_num, index, gid);
587 EXPORT_SYMBOL(ib_query_gid);
590 * ib_query_pkey - Get P_Key table entry
591 * @device:Device to query
592 * @port_num:Port number to query
593 * @index:P_Key table index to query
594 * @pkey:Returned P_Key
596 * ib_query_pkey() fetches the specified P_Key table entry.
598 int ib_query_pkey(struct ib_device *device,
599 u8 port_num, u16 index, u16 *pkey)
601 return device->query_pkey(device, port_num, index, pkey);
603 EXPORT_SYMBOL(ib_query_pkey);
606 * ib_modify_device - Change IB device attributes
607 * @device:Device to modify
608 * @device_modify_mask:Mask of attributes to change
609 * @device_modify:New attribute values
611 * ib_modify_device() changes a device's attributes as specified by
612 * the @device_modify_mask and @device_modify structure.
614 int ib_modify_device(struct ib_device *device,
615 int device_modify_mask,
616 struct ib_device_modify *device_modify)
618 if (!device->modify_device)
621 return device->modify_device(device, device_modify_mask,
624 EXPORT_SYMBOL(ib_modify_device);
627 * ib_modify_port - Modifies the attributes for the specified port.
628 * @device: The device to modify.
629 * @port_num: The number of the port to modify.
630 * @port_modify_mask: Mask used to specify which attributes of the port
632 * @port_modify: New attribute values for the port.
634 * ib_modify_port() changes a port's attributes as specified by the
635 * @port_modify_mask and @port_modify structure.
637 int ib_modify_port(struct ib_device *device,
638 u8 port_num, int port_modify_mask,
639 struct ib_port_modify *port_modify)
641 if (!device->modify_port)
644 if (!rdma_is_port_valid(device, port_num))
647 return device->modify_port(device, port_num, port_modify_mask,
650 EXPORT_SYMBOL(ib_modify_port);
653 * ib_find_gid - Returns the port number and GID table index where
654 * a specified GID value occurs.
655 * @device: The device to query.
656 * @gid: The GID value to search for.
657 * @port_num: The port number of the device where the GID value was found.
658 * @index: The index into the GID table where the GID was found. This
659 * parameter may be NULL.
661 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
662 u8 *port_num, u16 *index)
664 union ib_gid tmp_gid;
667 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
668 for (i = 0; i < device->gid_tbl_len[port - rdma_start_port(device)]; ++i) {
669 ret = ib_query_gid(device, port, i, &tmp_gid);
672 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
683 EXPORT_SYMBOL(ib_find_gid);
686 * ib_find_pkey - Returns the PKey table index where a specified
688 * @device: The device to query.
689 * @port_num: The port number of the device to search for the PKey.
690 * @pkey: The PKey value to search for.
691 * @index: The index into the PKey table where the PKey was found.
693 int ib_find_pkey(struct ib_device *device,
694 u8 port_num, u16 pkey, u16 *index)
699 for (i = 0; i < device->pkey_tbl_len[port_num - rdma_start_port(device)]; ++i) {
700 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
704 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
712 EXPORT_SYMBOL(ib_find_pkey);
714 static int __init ib_core_init(void)
718 ib_wq = alloc_workqueue("infiniband", 0, 0);
722 ret = ib_sysfs_setup();
724 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
730 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
734 ret = ib_cache_setup();
736 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
749 destroy_workqueue(ib_wq);
753 static void __exit ib_core_cleanup(void)
758 /* Make sure that any pending umem accounting work is done. */
759 destroy_workqueue(ib_wq);
762 module_init(ib_core_init);
763 module_exit(ib_core_cleanup);