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 /* The device_list and client_list contain devices and clients after their
59 * registration has completed, and the devices and clients are removed
60 * during unregistration. */
61 static LIST_HEAD(device_list);
62 static LIST_HEAD(client_list);
65 * device_mutex and lists_rwsem protect access to both device_list and
66 * client_list. device_mutex protects writer access by device and client
67 * registration / de-registration. lists_rwsem protects reader access to
68 * these lists. Iterators of these lists must lock it for read, while updates
69 * to the lists must be done with a write lock. A special case is when the
70 * device_mutex is locked. In this case locking the lists for read access is
71 * not necessary as the device_mutex implies it.
73 static DEFINE_MUTEX(device_mutex);
74 static DECLARE_RWSEM(lists_rwsem);
77 static int ib_device_check_mandatory(struct ib_device *device)
79 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
83 } mandatory_table[] = {
84 IB_MANDATORY_FUNC(query_device),
85 IB_MANDATORY_FUNC(query_port),
86 IB_MANDATORY_FUNC(query_pkey),
87 IB_MANDATORY_FUNC(query_gid),
88 IB_MANDATORY_FUNC(alloc_pd),
89 IB_MANDATORY_FUNC(dealloc_pd),
90 IB_MANDATORY_FUNC(create_ah),
91 IB_MANDATORY_FUNC(destroy_ah),
92 IB_MANDATORY_FUNC(create_qp),
93 IB_MANDATORY_FUNC(modify_qp),
94 IB_MANDATORY_FUNC(destroy_qp),
95 IB_MANDATORY_FUNC(post_send),
96 IB_MANDATORY_FUNC(post_recv),
97 IB_MANDATORY_FUNC(create_cq),
98 IB_MANDATORY_FUNC(destroy_cq),
99 IB_MANDATORY_FUNC(poll_cq),
100 IB_MANDATORY_FUNC(req_notify_cq),
101 IB_MANDATORY_FUNC(get_dma_mr),
102 IB_MANDATORY_FUNC(dereg_mr),
103 IB_MANDATORY_FUNC(get_port_immutable)
107 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
108 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
109 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
110 device->name, mandatory_table[i].name);
118 static struct ib_device *__ib_device_get_by_name(const char *name)
120 struct ib_device *device;
122 list_for_each_entry(device, &device_list, core_list)
123 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
130 static int alloc_name(char *name)
132 unsigned long *inuse;
133 char buf[IB_DEVICE_NAME_MAX];
134 struct ib_device *device;
137 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
141 list_for_each_entry(device, &device_list, core_list) {
142 if (!sscanf(device->name, name, &i))
144 if (i < 0 || i >= PAGE_SIZE * 8)
146 snprintf(buf, sizeof buf, name, i);
147 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
151 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
152 free_page((unsigned long) inuse);
153 snprintf(buf, sizeof buf, name, i);
155 if (__ib_device_get_by_name(buf))
158 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
163 * ib_alloc_device - allocate an IB device struct
164 * @size:size of structure to allocate
166 * Low-level drivers should use ib_alloc_device() to allocate &struct
167 * ib_device. @size is the size of the structure to be allocated,
168 * including any private data used by the low-level driver.
169 * ib_dealloc_device() must be used to free structures allocated with
172 struct ib_device *ib_alloc_device(size_t size)
174 BUG_ON(size < sizeof (struct ib_device));
176 return kzalloc(size, GFP_KERNEL);
178 EXPORT_SYMBOL(ib_alloc_device);
181 * ib_dealloc_device - free an IB device struct
182 * @device:structure to free
184 * Free a structure allocated with ib_alloc_device().
186 void ib_dealloc_device(struct ib_device *device)
188 if (device->reg_state == IB_DEV_UNINITIALIZED) {
193 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
195 kobject_put(&device->dev.kobj);
197 EXPORT_SYMBOL(ib_dealloc_device);
199 static int add_client_context(struct ib_device *device, struct ib_client *client)
201 struct ib_client_data *context;
204 context = kmalloc(sizeof *context, GFP_KERNEL);
206 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
207 device->name, client->name);
211 context->client = client;
212 context->data = NULL;
214 spin_lock_irqsave(&device->client_data_lock, flags);
215 list_add(&context->list, &device->client_data_list);
216 spin_unlock_irqrestore(&device->client_data_lock, flags);
221 static int verify_immutable(const struct ib_device *dev, u8 port)
223 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
224 rdma_max_mad_size(dev, port) != 0);
227 static int read_port_immutable(struct ib_device *device)
230 u8 start_port = rdma_start_port(device);
231 u8 end_port = rdma_end_port(device);
235 * device->port_immutable is indexed directly by the port number to make
236 * access to this data as efficient as possible.
238 * Therefore port_immutable is declared as a 1 based array with
239 * potential empty slots at the beginning.
241 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
244 if (!device->port_immutable)
247 for (port = start_port; port <= end_port; ++port) {
248 ret = device->get_port_immutable(device, port,
249 &device->port_immutable[port]);
253 if (verify_immutable(device, port)) {
262 kfree(device->port_immutable);
268 * ib_register_device - Register an IB device with IB core
269 * @device:Device to register
271 * Low-level drivers use ib_register_device() to register their
272 * devices with the IB core. All registered clients will receive a
273 * callback for each device that is added. @device must be allocated
274 * with ib_alloc_device().
276 int ib_register_device(struct ib_device *device,
277 int (*port_callback)(struct ib_device *,
278 u8, struct kobject *))
282 mutex_lock(&device_mutex);
284 if (strchr(device->name, '%')) {
285 ret = alloc_name(device->name);
290 if (ib_device_check_mandatory(device)) {
295 INIT_LIST_HEAD(&device->event_handler_list);
296 INIT_LIST_HEAD(&device->client_data_list);
297 spin_lock_init(&device->event_handler_lock);
298 spin_lock_init(&device->client_data_lock);
300 ret = read_port_immutable(device);
302 printk(KERN_WARNING "Couldn't create per port immutable data %s\n",
307 ret = ib_device_register_sysfs(device, port_callback);
309 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
311 kfree(device->port_immutable);
315 device->reg_state = IB_DEV_REGISTERED;
318 struct ib_client *client;
320 list_for_each_entry(client, &client_list, list)
321 if (client->add && !add_client_context(device, client))
325 down_write(&lists_rwsem);
326 list_add_tail(&device->core_list, &device_list);
327 up_write(&lists_rwsem);
329 mutex_unlock(&device_mutex);
332 EXPORT_SYMBOL(ib_register_device);
335 * ib_unregister_device - Unregister an IB device
336 * @device:Device to unregister
338 * Unregister an IB device. All clients will receive a remove callback.
340 void ib_unregister_device(struct ib_device *device)
342 struct ib_client *client;
343 struct ib_client_data *context, *tmp;
346 mutex_lock(&device_mutex);
348 down_write(&lists_rwsem);
349 list_del(&device->core_list);
350 up_write(&lists_rwsem);
352 list_for_each_entry_reverse(client, &client_list, list)
354 client->remove(device);
356 mutex_unlock(&device_mutex);
358 ib_device_unregister_sysfs(device);
360 spin_lock_irqsave(&device->client_data_lock, flags);
361 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
363 spin_unlock_irqrestore(&device->client_data_lock, flags);
365 device->reg_state = IB_DEV_UNREGISTERED;
367 EXPORT_SYMBOL(ib_unregister_device);
370 * ib_register_client - Register an IB client
371 * @client:Client to register
373 * Upper level users of the IB drivers can use ib_register_client() to
374 * register callbacks for IB device addition and removal. When an IB
375 * device is added, each registered client's add method will be called
376 * (in the order the clients were registered), and when a device is
377 * removed, each client's remove method will be called (in the reverse
378 * order that clients were registered). In addition, when
379 * ib_register_client() is called, the client will receive an add
380 * callback for all devices already registered.
382 int ib_register_client(struct ib_client *client)
384 struct ib_device *device;
386 mutex_lock(&device_mutex);
388 list_for_each_entry(device, &device_list, core_list)
389 if (client->add && !add_client_context(device, client))
392 down_write(&lists_rwsem);
393 list_add_tail(&client->list, &client_list);
394 up_write(&lists_rwsem);
396 mutex_unlock(&device_mutex);
400 EXPORT_SYMBOL(ib_register_client);
403 * ib_unregister_client - Unregister an IB client
404 * @client:Client to unregister
406 * Upper level users use ib_unregister_client() to remove their client
407 * registration. When ib_unregister_client() is called, the client
408 * will receive a remove callback for each IB device still registered.
410 void ib_unregister_client(struct ib_client *client)
412 struct ib_client_data *context, *tmp;
413 struct ib_device *device;
416 mutex_lock(&device_mutex);
418 down_write(&lists_rwsem);
419 list_del(&client->list);
420 up_write(&lists_rwsem);
422 list_for_each_entry(device, &device_list, core_list) {
424 client->remove(device);
426 spin_lock_irqsave(&device->client_data_lock, flags);
427 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
428 if (context->client == client) {
429 list_del(&context->list);
432 spin_unlock_irqrestore(&device->client_data_lock, flags);
435 mutex_unlock(&device_mutex);
437 EXPORT_SYMBOL(ib_unregister_client);
440 * ib_get_client_data - Get IB client context
441 * @device:Device to get context for
442 * @client:Client to get context for
444 * ib_get_client_data() returns client context set with
445 * ib_set_client_data().
447 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
449 struct ib_client_data *context;
453 spin_lock_irqsave(&device->client_data_lock, flags);
454 list_for_each_entry(context, &device->client_data_list, list)
455 if (context->client == client) {
459 spin_unlock_irqrestore(&device->client_data_lock, flags);
463 EXPORT_SYMBOL(ib_get_client_data);
466 * ib_set_client_data - Set IB client context
467 * @device:Device to set context for
468 * @client:Client to set context for
469 * @data:Context to set
471 * ib_set_client_data() sets client context that can be retrieved with
472 * ib_get_client_data().
474 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
477 struct ib_client_data *context;
480 spin_lock_irqsave(&device->client_data_lock, flags);
481 list_for_each_entry(context, &device->client_data_list, list)
482 if (context->client == client) {
483 context->data = data;
487 printk(KERN_WARNING "No client context found for %s/%s\n",
488 device->name, client->name);
491 spin_unlock_irqrestore(&device->client_data_lock, flags);
493 EXPORT_SYMBOL(ib_set_client_data);
496 * ib_register_event_handler - Register an IB event handler
497 * @event_handler:Handler to register
499 * ib_register_event_handler() registers an event handler that will be
500 * called back when asynchronous IB events occur (as defined in
501 * chapter 11 of the InfiniBand Architecture Specification). This
502 * callback may occur in interrupt context.
504 int ib_register_event_handler (struct ib_event_handler *event_handler)
508 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
509 list_add_tail(&event_handler->list,
510 &event_handler->device->event_handler_list);
511 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
515 EXPORT_SYMBOL(ib_register_event_handler);
518 * ib_unregister_event_handler - Unregister an event handler
519 * @event_handler:Handler to unregister
521 * Unregister an event handler registered with
522 * ib_register_event_handler().
524 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
528 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
529 list_del(&event_handler->list);
530 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
534 EXPORT_SYMBOL(ib_unregister_event_handler);
537 * ib_dispatch_event - Dispatch an asynchronous event
538 * @event:Event to dispatch
540 * Low-level drivers must call ib_dispatch_event() to dispatch the
541 * event to all registered event handlers when an asynchronous event
544 void ib_dispatch_event(struct ib_event *event)
547 struct ib_event_handler *handler;
549 spin_lock_irqsave(&event->device->event_handler_lock, flags);
551 list_for_each_entry(handler, &event->device->event_handler_list, list)
552 handler->handler(handler, event);
554 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
556 EXPORT_SYMBOL(ib_dispatch_event);
559 * ib_query_device - Query IB device attributes
560 * @device:Device to query
561 * @device_attr:Device attributes
563 * ib_query_device() returns the attributes of a device through the
564 * @device_attr pointer.
566 int ib_query_device(struct ib_device *device,
567 struct ib_device_attr *device_attr)
569 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
571 memset(device_attr, 0, sizeof(*device_attr));
573 return device->query_device(device, device_attr, &uhw);
575 EXPORT_SYMBOL(ib_query_device);
578 * ib_query_port - Query IB port attributes
579 * @device:Device to query
580 * @port_num:Port number to query
581 * @port_attr:Port attributes
583 * ib_query_port() returns the attributes of a port through the
584 * @port_attr pointer.
586 int ib_query_port(struct ib_device *device,
588 struct ib_port_attr *port_attr)
590 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
593 return device->query_port(device, port_num, port_attr);
595 EXPORT_SYMBOL(ib_query_port);
598 * ib_query_gid - Get GID table entry
599 * @device:Device to query
600 * @port_num:Port number to query
601 * @index:GID table index to query
604 * ib_query_gid() fetches the specified GID table entry.
606 int ib_query_gid(struct ib_device *device,
607 u8 port_num, int index, union ib_gid *gid)
609 return device->query_gid(device, port_num, index, gid);
611 EXPORT_SYMBOL(ib_query_gid);
614 * ib_query_pkey - Get P_Key table entry
615 * @device:Device to query
616 * @port_num:Port number to query
617 * @index:P_Key table index to query
618 * @pkey:Returned P_Key
620 * ib_query_pkey() fetches the specified P_Key table entry.
622 int ib_query_pkey(struct ib_device *device,
623 u8 port_num, u16 index, u16 *pkey)
625 return device->query_pkey(device, port_num, index, pkey);
627 EXPORT_SYMBOL(ib_query_pkey);
630 * ib_modify_device - Change IB device attributes
631 * @device:Device to modify
632 * @device_modify_mask:Mask of attributes to change
633 * @device_modify:New attribute values
635 * ib_modify_device() changes a device's attributes as specified by
636 * the @device_modify_mask and @device_modify structure.
638 int ib_modify_device(struct ib_device *device,
639 int device_modify_mask,
640 struct ib_device_modify *device_modify)
642 if (!device->modify_device)
645 return device->modify_device(device, device_modify_mask,
648 EXPORT_SYMBOL(ib_modify_device);
651 * ib_modify_port - Modifies the attributes for the specified port.
652 * @device: The device to modify.
653 * @port_num: The number of the port to modify.
654 * @port_modify_mask: Mask used to specify which attributes of the port
656 * @port_modify: New attribute values for the port.
658 * ib_modify_port() changes a port's attributes as specified by the
659 * @port_modify_mask and @port_modify structure.
661 int ib_modify_port(struct ib_device *device,
662 u8 port_num, int port_modify_mask,
663 struct ib_port_modify *port_modify)
665 if (!device->modify_port)
668 if (port_num < rdma_start_port(device) || port_num > rdma_end_port(device))
671 return device->modify_port(device, port_num, port_modify_mask,
674 EXPORT_SYMBOL(ib_modify_port);
677 * ib_find_gid - Returns the port number and GID table index where
678 * a specified GID value occurs.
679 * @device: The device to query.
680 * @gid: The GID value to search for.
681 * @port_num: The port number of the device where the GID value was found.
682 * @index: The index into the GID table where the GID was found. This
683 * parameter may be NULL.
685 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
686 u8 *port_num, u16 *index)
688 union ib_gid tmp_gid;
691 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
692 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
693 ret = ib_query_gid(device, port, i, &tmp_gid);
696 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
707 EXPORT_SYMBOL(ib_find_gid);
710 * ib_find_pkey - Returns the PKey table index where a specified
712 * @device: The device to query.
713 * @port_num: The port number of the device to search for the PKey.
714 * @pkey: The PKey value to search for.
715 * @index: The index into the PKey table where the PKey was found.
717 int ib_find_pkey(struct ib_device *device,
718 u8 port_num, u16 pkey, u16 *index)
724 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
725 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
728 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
729 /* if there is full-member pkey take it.*/
730 if (tmp_pkey & 0x8000) {
739 /*no full-member, if exists take the limited*/
740 if (partial_ix >= 0) {
746 EXPORT_SYMBOL(ib_find_pkey);
748 static int __init ib_core_init(void)
752 ib_wq = alloc_workqueue("infiniband", 0, 0);
756 ret = ib_sysfs_setup();
758 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
764 printk(KERN_WARNING "Couldn't init IB netlink interface\n");
768 ret = ib_cache_setup();
770 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
783 destroy_workqueue(ib_wq);
787 static void __exit ib_core_cleanup(void)
792 /* Make sure that any pending umem accounting work is done. */
793 destroy_workqueue(ib_wq);
796 module_init(ib_core_init);
797 module_exit(ib_core_cleanup);