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 <linux/workqueue.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 static LIST_HEAD(device_list
);
56 static LIST_HEAD(client_list
);
59 * device_mutex protects access to both device_list and client_list.
60 * There's no real point to using multiple locks or something fancier
61 * like an rwsem: we always access both lists, and we're always
62 * modifying one list or the other list. In any case this is not a
63 * hot path so there's no point in trying to optimize.
65 static DEFINE_MUTEX(device_mutex
);
67 static int ib_device_check_mandatory(struct ib_device
*device
)
69 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
73 } mandatory_table
[] = {
74 IB_MANDATORY_FUNC(query_device
),
75 IB_MANDATORY_FUNC(query_port
),
76 IB_MANDATORY_FUNC(query_pkey
),
77 IB_MANDATORY_FUNC(query_gid
),
78 IB_MANDATORY_FUNC(alloc_pd
),
79 IB_MANDATORY_FUNC(dealloc_pd
),
80 IB_MANDATORY_FUNC(create_ah
),
81 IB_MANDATORY_FUNC(destroy_ah
),
82 IB_MANDATORY_FUNC(create_qp
),
83 IB_MANDATORY_FUNC(modify_qp
),
84 IB_MANDATORY_FUNC(destroy_qp
),
85 IB_MANDATORY_FUNC(post_send
),
86 IB_MANDATORY_FUNC(post_recv
),
87 IB_MANDATORY_FUNC(create_cq
),
88 IB_MANDATORY_FUNC(destroy_cq
),
89 IB_MANDATORY_FUNC(poll_cq
),
90 IB_MANDATORY_FUNC(req_notify_cq
),
91 IB_MANDATORY_FUNC(get_dma_mr
),
92 IB_MANDATORY_FUNC(dereg_mr
)
96 for (i
= 0; i
< ARRAY_SIZE(mandatory_table
); ++i
) {
97 if (!*(void **) ((void *) device
+ mandatory_table
[i
].offset
)) {
98 printk(KERN_WARNING
"Device %s is missing mandatory function %s\n",
99 device
->name
, mandatory_table
[i
].name
);
107 static struct ib_device
*__ib_device_get_by_name(const char *name
)
109 struct ib_device
*device
;
111 list_for_each_entry(device
, &device_list
, core_list
)
112 if (!strncmp(name
, device
->name
, IB_DEVICE_NAME_MAX
))
119 static int alloc_name(char *name
)
121 unsigned long *inuse
;
122 char buf
[IB_DEVICE_NAME_MAX
];
123 struct ib_device
*device
;
126 inuse
= (unsigned long *) get_zeroed_page(GFP_KERNEL
);
130 list_for_each_entry(device
, &device_list
, core_list
) {
131 if (!sscanf(device
->name
, name
, &i
))
133 if (i
< 0 || i
>= PAGE_SIZE
* 8)
135 snprintf(buf
, sizeof buf
, name
, i
);
136 if (!strncmp(buf
, device
->name
, IB_DEVICE_NAME_MAX
))
140 i
= find_first_zero_bit(inuse
, PAGE_SIZE
* 8);
141 free_page((unsigned long) inuse
);
142 snprintf(buf
, sizeof buf
, name
, i
);
144 if (__ib_device_get_by_name(buf
))
147 strlcpy(name
, buf
, IB_DEVICE_NAME_MAX
);
151 static int start_port(struct ib_device
*device
)
153 return (device
->node_type
== RDMA_NODE_IB_SWITCH
) ? 0 : 1;
157 static int end_port(struct ib_device
*device
)
159 return (device
->node_type
== RDMA_NODE_IB_SWITCH
) ?
160 0 : device
->phys_port_cnt
;
164 * ib_alloc_device - allocate an IB device struct
165 * @size:size of structure to allocate
167 * Low-level drivers should use ib_alloc_device() to allocate &struct
168 * ib_device. @size is the size of the structure to be allocated,
169 * including any private data used by the low-level driver.
170 * ib_dealloc_device() must be used to free structures allocated with
173 struct ib_device
*ib_alloc_device(size_t size
)
175 BUG_ON(size
< sizeof (struct ib_device
));
177 return kzalloc(size
, GFP_KERNEL
);
179 EXPORT_SYMBOL(ib_alloc_device
);
182 * ib_dealloc_device - free an IB device struct
183 * @device:structure to free
185 * Free a structure allocated with ib_alloc_device().
187 void ib_dealloc_device(struct ib_device
*device
)
189 if (device
->reg_state
== IB_DEV_UNINITIALIZED
) {
194 BUG_ON(device
->reg_state
!= IB_DEV_UNREGISTERED
);
196 kobject_put(&device
->dev
.kobj
);
198 EXPORT_SYMBOL(ib_dealloc_device
);
200 static int add_client_context(struct ib_device
*device
, struct ib_client
*client
)
202 struct ib_client_data
*context
;
205 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
207 printk(KERN_WARNING
"Couldn't allocate client context for %s/%s\n",
208 device
->name
, client
->name
);
212 context
->client
= client
;
213 context
->data
= NULL
;
215 spin_lock_irqsave(&device
->client_data_lock
, flags
);
216 list_add(&context
->list
, &device
->client_data_list
);
217 spin_unlock_irqrestore(&device
->client_data_lock
, flags
);
222 static int read_port_table_lengths(struct ib_device
*device
)
224 struct ib_port_attr
*tprops
= NULL
;
225 int num_ports
, ret
= -ENOMEM
;
228 tprops
= kmalloc(sizeof *tprops
, GFP_KERNEL
);
232 num_ports
= end_port(device
) - start_port(device
) + 1;
234 device
->pkey_tbl_len
= kmalloc(sizeof *device
->pkey_tbl_len
* num_ports
,
236 device
->gid_tbl_len
= kmalloc(sizeof *device
->gid_tbl_len
* num_ports
,
238 if (!device
->pkey_tbl_len
|| !device
->gid_tbl_len
)
241 for (port_index
= 0; port_index
< num_ports
; ++port_index
) {
242 ret
= ib_query_port(device
, port_index
+ start_port(device
),
246 device
->pkey_tbl_len
[port_index
] = tprops
->pkey_tbl_len
;
247 device
->gid_tbl_len
[port_index
] = tprops
->gid_tbl_len
;
254 kfree(device
->gid_tbl_len
);
255 kfree(device
->pkey_tbl_len
);
262 * ib_register_device - Register an IB device with IB core
263 * @device:Device to register
265 * Low-level drivers use ib_register_device() to register their
266 * devices with the IB core. All registered clients will receive a
267 * callback for each device that is added. @device must be allocated
268 * with ib_alloc_device().
270 int ib_register_device(struct ib_device
*device
)
274 mutex_lock(&device_mutex
);
276 if (strchr(device
->name
, '%')) {
277 ret
= alloc_name(device
->name
);
282 if (ib_device_check_mandatory(device
)) {
287 INIT_LIST_HEAD(&device
->event_handler_list
);
288 INIT_LIST_HEAD(&device
->client_data_list
);
289 spin_lock_init(&device
->event_handler_lock
);
290 spin_lock_init(&device
->client_data_lock
);
292 ret
= read_port_table_lengths(device
);
294 printk(KERN_WARNING
"Couldn't create table lengths cache for device %s\n",
299 ret
= ib_device_register_sysfs(device
);
301 printk(KERN_WARNING
"Couldn't register device %s with driver model\n",
303 kfree(device
->gid_tbl_len
);
304 kfree(device
->pkey_tbl_len
);
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 kfree(device
->gid_tbl_len
);
347 kfree(device
->pkey_tbl_len
);
349 mutex_unlock(&device_mutex
);
351 ib_device_unregister_sysfs(device
);
353 spin_lock_irqsave(&device
->client_data_lock
, flags
);
354 list_for_each_entry_safe(context
, tmp
, &device
->client_data_list
, list
)
356 spin_unlock_irqrestore(&device
->client_data_lock
, flags
);
358 device
->reg_state
= IB_DEV_UNREGISTERED
;
360 EXPORT_SYMBOL(ib_unregister_device
);
363 * ib_register_client - Register an IB client
364 * @client:Client to register
366 * Upper level users of the IB drivers can use ib_register_client() to
367 * register callbacks for IB device addition and removal. When an IB
368 * device is added, each registered client's add method will be called
369 * (in the order the clients were registered), and when a device is
370 * removed, each client's remove method will be called (in the reverse
371 * order that clients were registered). In addition, when
372 * ib_register_client() is called, the client will receive an add
373 * callback for all devices already registered.
375 int ib_register_client(struct ib_client
*client
)
377 struct ib_device
*device
;
379 mutex_lock(&device_mutex
);
381 list_add_tail(&client
->list
, &client_list
);
382 list_for_each_entry(device
, &device_list
, core_list
)
383 if (client
->add
&& !add_client_context(device
, client
))
386 mutex_unlock(&device_mutex
);
390 EXPORT_SYMBOL(ib_register_client
);
393 * ib_unregister_client - Unregister an IB client
394 * @client:Client to unregister
396 * Upper level users use ib_unregister_client() to remove their client
397 * registration. When ib_unregister_client() is called, the client
398 * will receive a remove callback for each IB device still registered.
400 void ib_unregister_client(struct ib_client
*client
)
402 struct ib_client_data
*context
, *tmp
;
403 struct ib_device
*device
;
406 mutex_lock(&device_mutex
);
408 list_for_each_entry(device
, &device_list
, core_list
) {
410 client
->remove(device
);
412 spin_lock_irqsave(&device
->client_data_lock
, flags
);
413 list_for_each_entry_safe(context
, tmp
, &device
->client_data_list
, list
)
414 if (context
->client
== client
) {
415 list_del(&context
->list
);
418 spin_unlock_irqrestore(&device
->client_data_lock
, flags
);
420 list_del(&client
->list
);
422 mutex_unlock(&device_mutex
);
424 EXPORT_SYMBOL(ib_unregister_client
);
427 * ib_get_client_data - Get IB client context
428 * @device:Device to get context for
429 * @client:Client to get context for
431 * ib_get_client_data() returns client context set with
432 * ib_set_client_data().
434 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
)
436 struct ib_client_data
*context
;
440 spin_lock_irqsave(&device
->client_data_lock
, flags
);
441 list_for_each_entry(context
, &device
->client_data_list
, list
)
442 if (context
->client
== client
) {
446 spin_unlock_irqrestore(&device
->client_data_lock
, flags
);
450 EXPORT_SYMBOL(ib_get_client_data
);
453 * ib_set_client_data - Set IB client context
454 * @device:Device to set context for
455 * @client:Client to set context for
456 * @data:Context to set
458 * ib_set_client_data() sets client context that can be retrieved with
459 * ib_get_client_data().
461 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
464 struct ib_client_data
*context
;
467 spin_lock_irqsave(&device
->client_data_lock
, flags
);
468 list_for_each_entry(context
, &device
->client_data_list
, list
)
469 if (context
->client
== client
) {
470 context
->data
= data
;
474 printk(KERN_WARNING
"No client context found for %s/%s\n",
475 device
->name
, client
->name
);
478 spin_unlock_irqrestore(&device
->client_data_lock
, flags
);
480 EXPORT_SYMBOL(ib_set_client_data
);
483 * ib_register_event_handler - Register an IB event handler
484 * @event_handler:Handler to register
486 * ib_register_event_handler() registers an event handler that will be
487 * called back when asynchronous IB events occur (as defined in
488 * chapter 11 of the InfiniBand Architecture Specification). This
489 * callback may occur in interrupt context.
491 int ib_register_event_handler (struct ib_event_handler
*event_handler
)
495 spin_lock_irqsave(&event_handler
->device
->event_handler_lock
, flags
);
496 list_add_tail(&event_handler
->list
,
497 &event_handler
->device
->event_handler_list
);
498 spin_unlock_irqrestore(&event_handler
->device
->event_handler_lock
, flags
);
502 EXPORT_SYMBOL(ib_register_event_handler
);
505 * ib_unregister_event_handler - Unregister an event handler
506 * @event_handler:Handler to unregister
508 * Unregister an event handler registered with
509 * ib_register_event_handler().
511 int ib_unregister_event_handler(struct ib_event_handler
*event_handler
)
515 spin_lock_irqsave(&event_handler
->device
->event_handler_lock
, flags
);
516 list_del(&event_handler
->list
);
517 spin_unlock_irqrestore(&event_handler
->device
->event_handler_lock
, flags
);
521 EXPORT_SYMBOL(ib_unregister_event_handler
);
524 * ib_dispatch_event - Dispatch an asynchronous event
525 * @event:Event to dispatch
527 * Low-level drivers must call ib_dispatch_event() to dispatch the
528 * event to all registered event handlers when an asynchronous event
531 void ib_dispatch_event(struct ib_event
*event
)
534 struct ib_event_handler
*handler
;
536 spin_lock_irqsave(&event
->device
->event_handler_lock
, flags
);
538 list_for_each_entry(handler
, &event
->device
->event_handler_list
, list
)
539 handler
->handler(handler
, event
);
541 spin_unlock_irqrestore(&event
->device
->event_handler_lock
, flags
);
543 EXPORT_SYMBOL(ib_dispatch_event
);
546 * ib_query_device - Query IB device attributes
547 * @device:Device to query
548 * @device_attr:Device attributes
550 * ib_query_device() returns the attributes of a device through the
551 * @device_attr pointer.
553 int ib_query_device(struct ib_device
*device
,
554 struct ib_device_attr
*device_attr
)
556 return device
->query_device(device
, device_attr
);
558 EXPORT_SYMBOL(ib_query_device
);
561 * ib_query_port - Query IB port attributes
562 * @device:Device to query
563 * @port_num:Port number to query
564 * @port_attr:Port attributes
566 * ib_query_port() returns the attributes of a port through the
567 * @port_attr pointer.
569 int ib_query_port(struct ib_device
*device
,
571 struct ib_port_attr
*port_attr
)
573 if (port_num
< start_port(device
) || port_num
> end_port(device
))
576 return device
->query_port(device
, port_num
, port_attr
);
578 EXPORT_SYMBOL(ib_query_port
);
581 * ib_query_gid - Get GID table entry
582 * @device:Device to query
583 * @port_num:Port number to query
584 * @index:GID table index to query
587 * ib_query_gid() fetches the specified GID table entry.
589 int ib_query_gid(struct ib_device
*device
,
590 u8 port_num
, int index
, union ib_gid
*gid
)
592 return device
->query_gid(device
, port_num
, index
, gid
);
594 EXPORT_SYMBOL(ib_query_gid
);
597 * ib_query_pkey - Get P_Key table entry
598 * @device:Device to query
599 * @port_num:Port number to query
600 * @index:P_Key table index to query
601 * @pkey:Returned P_Key
603 * ib_query_pkey() fetches the specified P_Key table entry.
605 int ib_query_pkey(struct ib_device
*device
,
606 u8 port_num
, u16 index
, u16
*pkey
)
608 return device
->query_pkey(device
, port_num
, index
, pkey
);
610 EXPORT_SYMBOL(ib_query_pkey
);
613 * ib_modify_device - Change IB device attributes
614 * @device:Device to modify
615 * @device_modify_mask:Mask of attributes to change
616 * @device_modify:New attribute values
618 * ib_modify_device() changes a device's attributes as specified by
619 * the @device_modify_mask and @device_modify structure.
621 int ib_modify_device(struct ib_device
*device
,
622 int device_modify_mask
,
623 struct ib_device_modify
*device_modify
)
625 return device
->modify_device(device
, device_modify_mask
,
628 EXPORT_SYMBOL(ib_modify_device
);
631 * ib_modify_port - Modifies the attributes for the specified port.
632 * @device: The device to modify.
633 * @port_num: The number of the port to modify.
634 * @port_modify_mask: Mask used to specify which attributes of the port
636 * @port_modify: New attribute values for the port.
638 * ib_modify_port() changes a port's attributes as specified by the
639 * @port_modify_mask and @port_modify structure.
641 int ib_modify_port(struct ib_device
*device
,
642 u8 port_num
, int port_modify_mask
,
643 struct ib_port_modify
*port_modify
)
645 if (port_num
< start_port(device
) || port_num
> end_port(device
))
648 return device
->modify_port(device
, port_num
, port_modify_mask
,
651 EXPORT_SYMBOL(ib_modify_port
);
654 * ib_find_gid - Returns the port number and GID table index where
655 * a specified GID value occurs.
656 * @device: The device to query.
657 * @gid: The GID value to search for.
658 * @port_num: The port number of the device where the GID value was found.
659 * @index: The index into the GID table where the GID was found. This
660 * parameter may be NULL.
662 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
663 u8
*port_num
, u16
*index
)
665 union ib_gid tmp_gid
;
668 for (port
= start_port(device
); port
<= end_port(device
); ++port
) {
669 for (i
= 0; i
< device
->gid_tbl_len
[port
- start_port(device
)]; ++i
) {
670 ret
= ib_query_gid(device
, port
, i
, &tmp_gid
);
673 if (!memcmp(&tmp_gid
, gid
, sizeof *gid
)) {
684 EXPORT_SYMBOL(ib_find_gid
);
687 * ib_find_pkey - Returns the PKey table index where a specified
689 * @device: The device to query.
690 * @port_num: The port number of the device to search for the PKey.
691 * @pkey: The PKey value to search for.
692 * @index: The index into the PKey table where the PKey was found.
694 int ib_find_pkey(struct ib_device
*device
,
695 u8 port_num
, u16 pkey
, u16
*index
)
700 for (i
= 0; i
< device
->pkey_tbl_len
[port_num
- start_port(device
)]; ++i
) {
701 ret
= ib_query_pkey(device
, port_num
, i
, &tmp_pkey
);
705 if ((pkey
& 0x7fff) == (tmp_pkey
& 0x7fff)) {
713 EXPORT_SYMBOL(ib_find_pkey
);
715 static int __init
ib_core_init(void)
719 ret
= ib_sysfs_setup();
721 printk(KERN_WARNING
"Couldn't create InfiniBand device class\n");
723 ret
= ib_cache_setup();
725 printk(KERN_WARNING
"Couldn't set up InfiniBand P_Key/GID cache\n");
732 static void __exit
ib_core_cleanup(void)
736 /* Make sure that any pending umem accounting work is done. */
737 flush_scheduled_work();
740 module_init(ib_core_init
);
741 module_exit(ib_core_cleanup
);