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/mutex.h>
40 #include <linux/workqueue.h>
42 #include "core_priv.h"
44 MODULE_AUTHOR("Roland Dreier");
45 MODULE_DESCRIPTION("core kernel InfiniBand API");
46 MODULE_LICENSE("Dual BSD/GPL");
49 /* workaround for a bug in hp chipset that would cause kernel
50 panic when dma resources are exhaused */
51 int dma_map_sg_hp_wa = 0;
54 struct ib_client_data {
55 struct list_head list;
56 struct ib_client *client;
60 static LIST_HEAD(device_list);
61 static LIST_HEAD(client_list);
64 * device_mutex protects access to both device_list and client_list.
65 * There's no real point to using multiple locks or something fancier
66 * like an rwsem: we always access both lists, and we're always
67 * modifying one list or the other list. In any case this is not a
68 * hot path so there's no point in trying to optimize.
70 static DEFINE_MUTEX(device_mutex);
72 static int ib_device_check_mandatory(struct ib_device *device)
74 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
78 } mandatory_table[] = {
79 IB_MANDATORY_FUNC(query_device),
80 IB_MANDATORY_FUNC(query_port),
81 IB_MANDATORY_FUNC(query_pkey),
82 IB_MANDATORY_FUNC(query_gid),
83 IB_MANDATORY_FUNC(alloc_pd),
84 IB_MANDATORY_FUNC(dealloc_pd),
85 IB_MANDATORY_FUNC(create_ah),
86 IB_MANDATORY_FUNC(destroy_ah),
87 IB_MANDATORY_FUNC(create_qp),
88 IB_MANDATORY_FUNC(modify_qp),
89 IB_MANDATORY_FUNC(destroy_qp),
90 IB_MANDATORY_FUNC(post_send),
91 IB_MANDATORY_FUNC(post_recv),
92 IB_MANDATORY_FUNC(create_cq),
93 IB_MANDATORY_FUNC(destroy_cq),
94 IB_MANDATORY_FUNC(poll_cq),
95 IB_MANDATORY_FUNC(req_notify_cq),
96 IB_MANDATORY_FUNC(get_dma_mr),
97 IB_MANDATORY_FUNC(dereg_mr)
101 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
102 if (!*(void **) ((u_char *) 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);
156 static int start_port(struct ib_device *device)
158 return (device->node_type == RDMA_NODE_IB_SWITCH) ? 0 : 1;
162 static int end_port(struct ib_device *device)
164 return (device->node_type == RDMA_NODE_IB_SWITCH) ?
165 0 : device->phys_port_cnt;
169 * ib_alloc_device - allocate an IB device struct
170 * @size:size of structure to allocate
172 * Low-level drivers should use ib_alloc_device() to allocate &struct
173 * ib_device. @size is the size of the structure to be allocated,
174 * including any private data used by the low-level driver.
175 * ib_dealloc_device() must be used to free structures allocated with
178 struct ib_device *ib_alloc_device(size_t size)
180 BUG_ON(size < sizeof (struct ib_device));
182 return kzalloc(size, GFP_KERNEL);
184 EXPORT_SYMBOL(ib_alloc_device);
187 * ib_dealloc_device - free an IB device struct
188 * @device:structure to free
190 * Free a structure allocated with ib_alloc_device().
192 void ib_dealloc_device(struct ib_device *device)
194 if (device->reg_state == IB_DEV_UNINITIALIZED) {
199 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
201 kobject_put(&device->dev.kobj);
203 EXPORT_SYMBOL(ib_dealloc_device);
205 static int add_client_context(struct ib_device *device, struct ib_client *client)
207 struct ib_client_data *context;
210 context = kmalloc(sizeof *context, GFP_KERNEL);
212 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
213 device->name, client->name);
217 context->client = client;
218 context->data = NULL;
220 spin_lock_irqsave(&device->client_data_lock, flags);
221 list_add(&context->list, &device->client_data_list);
222 spin_unlock_irqrestore(&device->client_data_lock, flags);
227 static int read_port_table_lengths(struct ib_device *device)
229 struct ib_port_attr *tprops = NULL;
230 int num_ports, ret = -ENOMEM;
233 tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
237 num_ports = end_port(device) - start_port(device) + 1;
239 device->pkey_tbl_len = kmalloc(sizeof *device->pkey_tbl_len * num_ports,
241 device->gid_tbl_len = kmalloc(sizeof *device->gid_tbl_len * num_ports,
243 if (!device->pkey_tbl_len || !device->gid_tbl_len)
246 for (port_index = 0; port_index < num_ports; ++port_index) {
247 ret = ib_query_port(device, port_index + start_port(device),
251 device->pkey_tbl_len[port_index] = tprops->pkey_tbl_len;
252 device->gid_tbl_len[port_index] = tprops->gid_tbl_len;
259 kfree(device->gid_tbl_len);
260 kfree(device->pkey_tbl_len);
267 * ib_register_device - Register an IB device with IB core
268 * @device:Device to register
270 * Low-level drivers use ib_register_device() to register their
271 * devices with the IB core. All registered clients will receive a
272 * callback for each device that is added. @device must be allocated
273 * with ib_alloc_device().
275 int ib_register_device(struct ib_device *device,
276 int (*port_callback)(struct ib_device *,
277 u8, struct kobject *))
281 mutex_lock(&device_mutex);
283 if (strchr(device->name, '%')) {
284 ret = alloc_name(device->name);
289 if (ib_device_check_mandatory(device)) {
294 INIT_LIST_HEAD(&device->event_handler_list);
295 INIT_LIST_HEAD(&device->client_data_list);
296 spin_lock_init(&device->event_handler_lock);
297 spin_lock_init(&device->client_data_lock);
298 device->ib_uverbs_xrcd_table = RB_ROOT;
299 mutex_init(&device->xrcd_table_mutex);
301 ret = read_port_table_lengths(device);
303 printk(KERN_WARNING "Couldn't create table lengths cache for device %s\n",
308 ret = ib_device_register_sysfs(device, port_callback);
310 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
312 kfree(device->gid_tbl_len);
313 kfree(device->pkey_tbl_len);
317 list_add_tail(&device->core_list, &device_list);
319 device->reg_state = IB_DEV_REGISTERED;
322 struct ib_client *client;
324 list_for_each_entry(client, &client_list, list)
325 if (client->add && !add_client_context(device, client))
330 mutex_unlock(&device_mutex);
333 EXPORT_SYMBOL(ib_register_device);
336 * ib_unregister_device - Unregister an IB device
337 * @device:Device to unregister
339 * Unregister an IB device. All clients will receive a remove callback.
341 void ib_unregister_device(struct ib_device *device)
343 struct ib_client *client;
344 struct ib_client_data *context, *tmp;
347 mutex_lock(&device_mutex);
349 list_for_each_entry_reverse(client, &client_list, list)
351 client->remove(device);
353 list_del(&device->core_list);
355 kfree(device->gid_tbl_len);
356 kfree(device->pkey_tbl_len);
358 mutex_unlock(&device_mutex);
360 ib_device_unregister_sysfs(device);
362 spin_lock_irqsave(&device->client_data_lock, flags);
363 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
365 spin_unlock_irqrestore(&device->client_data_lock, flags);
367 device->reg_state = IB_DEV_UNREGISTERED;
369 EXPORT_SYMBOL(ib_unregister_device);
372 * ib_register_client - Register an IB client
373 * @client:Client to register
375 * Upper level users of the IB drivers can use ib_register_client() to
376 * register callbacks for IB device addition and removal. When an IB
377 * device is added, each registered client's add method will be called
378 * (in the order the clients were registered), and when a device is
379 * removed, each client's remove method will be called (in the reverse
380 * order that clients were registered). In addition, when
381 * ib_register_client() is called, the client will receive an add
382 * callback for all devices already registered.
384 int ib_register_client(struct ib_client *client)
386 struct ib_device *device;
388 mutex_lock(&device_mutex);
390 list_add_tail(&client->list, &client_list);
391 list_for_each_entry(device, &device_list, core_list)
392 if (client->add && !add_client_context(device, client))
395 mutex_unlock(&device_mutex);
399 EXPORT_SYMBOL(ib_register_client);
402 * ib_unregister_client - Unregister an IB client
403 * @client:Client to unregister
405 * Upper level users use ib_unregister_client() to remove their client
406 * registration. When ib_unregister_client() is called, the client
407 * will receive a remove callback for each IB device still registered.
409 void ib_unregister_client(struct ib_client *client)
411 struct ib_client_data *context, *tmp;
412 struct ib_device *device;
415 mutex_lock(&device_mutex);
417 list_for_each_entry(device, &device_list, core_list) {
419 client->remove(device);
421 spin_lock_irqsave(&device->client_data_lock, flags);
422 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
423 if (context->client == client) {
424 list_del(&context->list);
427 spin_unlock_irqrestore(&device->client_data_lock, flags);
429 list_del(&client->list);
431 mutex_unlock(&device_mutex);
433 EXPORT_SYMBOL(ib_unregister_client);
436 * ib_get_client_data - Get IB client context
437 * @device:Device to get context for
438 * @client:Client to get context for
440 * ib_get_client_data() returns client context set with
441 * ib_set_client_data().
443 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
445 struct ib_client_data *context;
449 spin_lock_irqsave(&device->client_data_lock, flags);
450 list_for_each_entry(context, &device->client_data_list, list)
451 if (context->client == client) {
455 spin_unlock_irqrestore(&device->client_data_lock, flags);
459 EXPORT_SYMBOL(ib_get_client_data);
462 * ib_set_client_data - Set IB client context
463 * @device:Device to set context for
464 * @client:Client to set context for
465 * @data:Context to set
467 * ib_set_client_data() sets client context that can be retrieved with
468 * ib_get_client_data().
470 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
473 struct ib_client_data *context;
476 spin_lock_irqsave(&device->client_data_lock, flags);
477 list_for_each_entry(context, &device->client_data_list, list)
478 if (context->client == client) {
479 context->data = data;
483 printk(KERN_WARNING "No client context found for %s/%s\n",
484 device->name, client->name);
487 spin_unlock_irqrestore(&device->client_data_lock, flags);
489 EXPORT_SYMBOL(ib_set_client_data);
492 * ib_register_event_handler - Register an IB event handler
493 * @event_handler:Handler to register
495 * ib_register_event_handler() registers an event handler that will be
496 * called back when asynchronous IB events occur (as defined in
497 * chapter 11 of the InfiniBand Architecture Specification). This
498 * callback may occur in interrupt context.
500 int ib_register_event_handler (struct ib_event_handler *event_handler)
504 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
505 list_add_tail(&event_handler->list,
506 &event_handler->device->event_handler_list);
507 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
511 EXPORT_SYMBOL(ib_register_event_handler);
514 * ib_unregister_event_handler - Unregister an event handler
515 * @event_handler:Handler to unregister
517 * Unregister an event handler registered with
518 * ib_register_event_handler().
520 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
524 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
525 list_del(&event_handler->list);
526 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
530 EXPORT_SYMBOL(ib_unregister_event_handler);
533 * ib_dispatch_event - Dispatch an asynchronous event
534 * @event:Event to dispatch
536 * Low-level drivers must call ib_dispatch_event() to dispatch the
537 * event to all registered event handlers when an asynchronous event
540 void ib_dispatch_event(struct ib_event *event)
543 struct ib_event_handler *handler;
545 spin_lock_irqsave(&event->device->event_handler_lock, flags);
547 list_for_each_entry(handler, &event->device->event_handler_list, list)
548 handler->handler(handler, event);
550 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
552 EXPORT_SYMBOL(ib_dispatch_event);
555 * ib_query_device - Query IB device attributes
556 * @device:Device to query
557 * @device_attr:Device attributes
559 * ib_query_device() returns the attributes of a device through the
560 * @device_attr pointer.
562 int ib_query_device(struct ib_device *device,
563 struct ib_device_attr *device_attr)
565 return device->query_device(device, device_attr);
567 EXPORT_SYMBOL(ib_query_device);
570 * ib_query_port - Query IB port attributes
571 * @device:Device to query
572 * @port_num:Port number to query
573 * @port_attr:Port attributes
575 * ib_query_port() returns the attributes of a port through the
576 * @port_attr pointer.
578 int ib_query_port(struct ib_device *device,
580 struct ib_port_attr *port_attr)
582 if (port_num < start_port(device) || port_num > end_port(device))
585 return device->query_port(device, port_num, port_attr);
587 EXPORT_SYMBOL(ib_query_port);
590 * ib_query_gid - Get GID table entry
591 * @device:Device to query
592 * @port_num:Port number to query
593 * @index:GID table index to query
596 * ib_query_gid() fetches the specified GID table entry.
598 int ib_query_gid(struct ib_device *device,
599 u8 port_num, int index, union ib_gid *gid)
601 return device->query_gid(device, port_num, index, gid);
603 EXPORT_SYMBOL(ib_query_gid);
606 * ib_query_pkey - Get P_Key table entry
607 * @device:Device to query
608 * @port_num:Port number to query
609 * @index:P_Key table index to query
610 * @pkey:Returned P_Key
612 * ib_query_pkey() fetches the specified P_Key table entry.
614 int ib_query_pkey(struct ib_device *device,
615 u8 port_num, u16 index, u16 *pkey)
617 return device->query_pkey(device, port_num, index, pkey);
619 EXPORT_SYMBOL(ib_query_pkey);
622 * ib_modify_device - Change IB device attributes
623 * @device:Device to modify
624 * @device_modify_mask:Mask of attributes to change
625 * @device_modify:New attribute values
627 * ib_modify_device() changes a device's attributes as specified by
628 * the @device_modify_mask and @device_modify structure.
630 int ib_modify_device(struct ib_device *device,
631 int device_modify_mask,
632 struct ib_device_modify *device_modify)
634 return device->modify_device(device, device_modify_mask,
637 EXPORT_SYMBOL(ib_modify_device);
640 * ib_modify_port - Modifies the attributes for the specified port.
641 * @device: The device to modify.
642 * @port_num: The number of the port to modify.
643 * @port_modify_mask: Mask used to specify which attributes of the port
645 * @port_modify: New attribute values for the port.
647 * ib_modify_port() changes a port's attributes as specified by the
648 * @port_modify_mask and @port_modify structure.
650 int ib_modify_port(struct ib_device *device,
651 u8 port_num, int port_modify_mask,
652 struct ib_port_modify *port_modify)
654 if (port_num < start_port(device) || port_num > end_port(device))
657 return device->modify_port(device, port_num, port_modify_mask,
660 EXPORT_SYMBOL(ib_modify_port);
663 * ib_find_gid - Returns the port number and GID table index where
664 * a specified GID value occurs.
665 * @device: The device to query.
666 * @gid: The GID value to search for.
667 * @port_num: The port number of the device where the GID value was found.
668 * @index: The index into the GID table where the GID was found. This
669 * parameter may be NULL.
671 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
672 u8 *port_num, u16 *index)
674 union ib_gid tmp_gid;
677 for (port = start_port(device); port <= end_port(device); ++port) {
678 for (i = 0; i < device->gid_tbl_len[port - start_port(device)]; ++i) {
679 ret = ib_query_gid(device, port, i, &tmp_gid);
682 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
693 EXPORT_SYMBOL(ib_find_gid);
696 * ib_find_pkey - Returns the PKey table index where a specified
698 * @device: The device to query.
699 * @port_num: The port number of the device to search for the PKey.
700 * @pkey: The PKey value to search for.
701 * @index: The index into the PKey table where the PKey was found.
703 int ib_find_pkey(struct ib_device *device,
704 u8 port_num, u16 pkey, u16 *index)
709 for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) {
710 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
714 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
722 EXPORT_SYMBOL(ib_find_pkey);
724 static int __init ib_core_init(void)
729 if (ia64_platform_is("hpzx1"))
730 dma_map_sg_hp_wa = 1;
733 ret = ib_sysfs_setup();
735 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
737 ret = ib_cache_setup();
739 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
746 static void __exit ib_core_cleanup(void)
750 /* Make sure that any pending umem accounting work is done. */
751 flush_scheduled_work();
754 module_init(ib_core_init);
755 module_exit(ib_core_cleanup);
757 #undef MODULE_VERSION
758 #include <sys/module.h>
760 ibcore_evhand(module_t mod, int event, void *arg)
765 static moduledata_t ibcore_mod = {
767 .evhand = ibcore_evhand,
770 MODULE_VERSION(ibcore, 1);
771 DECLARE_MODULE(ibcore, ibcore_mod, SI_SUB_SMP, SI_ORDER_ANY);