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");
50 /* workaround for a bug in hp chipset that would cause kernel
51 panic when dma resources are exhaused */
52 int dma_map_sg_hp_wa = 0;
55 struct ib_client_data {
56 struct list_head list;
57 struct ib_client *client;
61 static LIST_HEAD(device_list);
62 static LIST_HEAD(client_list);
65 * device_mutex protects access to both device_list and client_list.
66 * There's no real point to using multiple locks or something fancier
67 * like an rwsem: we always access both lists, and we're always
68 * modifying one list or the other list. In any case this is not a
69 * hot path so there's no point in trying to optimize.
71 static DEFINE_MUTEX(device_mutex);
73 static int ib_device_check_mandatory(struct ib_device *device)
75 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
79 } mandatory_table[] = {
80 IB_MANDATORY_FUNC(query_device),
81 IB_MANDATORY_FUNC(query_port),
82 IB_MANDATORY_FUNC(query_pkey),
83 IB_MANDATORY_FUNC(query_gid),
84 IB_MANDATORY_FUNC(alloc_pd),
85 IB_MANDATORY_FUNC(dealloc_pd),
86 IB_MANDATORY_FUNC(create_ah),
87 IB_MANDATORY_FUNC(destroy_ah),
88 IB_MANDATORY_FUNC(create_qp),
89 IB_MANDATORY_FUNC(modify_qp),
90 IB_MANDATORY_FUNC(destroy_qp),
91 IB_MANDATORY_FUNC(post_send),
92 IB_MANDATORY_FUNC(post_recv),
93 IB_MANDATORY_FUNC(create_cq),
94 IB_MANDATORY_FUNC(destroy_cq),
95 IB_MANDATORY_FUNC(poll_cq),
96 IB_MANDATORY_FUNC(req_notify_cq),
97 IB_MANDATORY_FUNC(get_dma_mr),
98 IB_MANDATORY_FUNC(dereg_mr)
102 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
103 if (!*(void **) ((u_char *) device + mandatory_table[i].offset)) {
104 printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
105 device->name, mandatory_table[i].name);
113 static struct ib_device *__ib_device_get_by_name(const char *name)
115 struct ib_device *device;
117 list_for_each_entry(device, &device_list, core_list)
118 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
125 static int alloc_name(char *name)
127 unsigned long *inuse;
128 char buf[IB_DEVICE_NAME_MAX];
129 struct ib_device *device;
132 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
136 list_for_each_entry(device, &device_list, core_list) {
137 if (!sscanf(device->name, name, &i))
139 if (i < 0 || i >= PAGE_SIZE * 8)
141 snprintf(buf, sizeof buf, name, i);
142 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
146 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
147 free_page((unsigned long) inuse);
148 snprintf(buf, sizeof buf, name, i);
150 if (__ib_device_get_by_name(buf))
153 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
157 static int start_port(struct ib_device *device)
159 return (device->node_type == RDMA_NODE_IB_SWITCH) ? 0 : 1;
163 static int end_port(struct ib_device *device)
165 return (device->node_type == RDMA_NODE_IB_SWITCH) ?
166 0 : device->phys_port_cnt;
170 * ib_alloc_device - allocate an IB device struct
171 * @size:size of structure to allocate
173 * Low-level drivers should use ib_alloc_device() to allocate &struct
174 * ib_device. @size is the size of the structure to be allocated,
175 * including any private data used by the low-level driver.
176 * ib_dealloc_device() must be used to free structures allocated with
179 struct ib_device *ib_alloc_device(size_t size)
181 BUG_ON(size < sizeof (struct ib_device));
183 return kzalloc(size, GFP_KERNEL);
185 EXPORT_SYMBOL(ib_alloc_device);
188 * ib_dealloc_device - free an IB device struct
189 * @device:structure to free
191 * Free a structure allocated with ib_alloc_device().
193 void ib_dealloc_device(struct ib_device *device)
195 if (device->reg_state == IB_DEV_UNINITIALIZED) {
200 BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
202 kobject_put(&device->dev.kobj);
204 EXPORT_SYMBOL(ib_dealloc_device);
206 static int add_client_context(struct ib_device *device, struct ib_client *client)
208 struct ib_client_data *context;
211 context = kmalloc(sizeof *context, GFP_KERNEL);
213 printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
214 device->name, client->name);
218 context->client = client;
219 context->data = NULL;
221 spin_lock_irqsave(&device->client_data_lock, flags);
222 list_add(&context->list, &device->client_data_list);
223 spin_unlock_irqrestore(&device->client_data_lock, flags);
228 static int read_port_table_lengths(struct ib_device *device)
230 struct ib_port_attr *tprops = NULL;
231 int num_ports, ret = -ENOMEM;
234 tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
238 num_ports = end_port(device) - start_port(device) + 1;
240 device->pkey_tbl_len = kmalloc(sizeof *device->pkey_tbl_len * num_ports,
242 device->gid_tbl_len = kmalloc(sizeof *device->gid_tbl_len * num_ports,
244 if (!device->pkey_tbl_len || !device->gid_tbl_len)
247 for (port_index = 0; port_index < num_ports; ++port_index) {
248 ret = ib_query_port(device, port_index + start_port(device),
252 device->pkey_tbl_len[port_index] = tprops->pkey_tbl_len;
253 device->gid_tbl_len[port_index] = tprops->gid_tbl_len;
260 kfree(device->gid_tbl_len);
261 kfree(device->pkey_tbl_len);
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);
299 device->ib_uverbs_xrcd_table = RB_ROOT;
300 mutex_init(&device->xrcd_table_mutex);
302 ret = read_port_table_lengths(device);
304 printk(KERN_WARNING "Couldn't create table lengths cache for device %s\n",
309 ret = ib_device_register_sysfs(device, port_callback);
311 printk(KERN_WARNING "Couldn't register device %s with driver model\n",
313 kfree(device->gid_tbl_len);
314 kfree(device->pkey_tbl_len);
318 list_add_tail(&device->core_list, &device_list);
320 device->reg_state = IB_DEV_REGISTERED;
323 struct ib_client *client;
325 list_for_each_entry(client, &client_list, list)
326 if (client->add && !add_client_context(device, client))
331 mutex_unlock(&device_mutex);
334 EXPORT_SYMBOL(ib_register_device);
337 * ib_unregister_device - Unregister an IB device
338 * @device:Device to unregister
340 * Unregister an IB device. All clients will receive a remove callback.
342 void ib_unregister_device(struct ib_device *device)
344 struct ib_client *client;
345 struct ib_client_data *context, *tmp;
348 mutex_lock(&device_mutex);
350 list_for_each_entry_reverse(client, &client_list, list)
352 client->remove(device);
354 list_del(&device->core_list);
356 kfree(device->gid_tbl_len);
357 kfree(device->pkey_tbl_len);
359 mutex_unlock(&device_mutex);
361 ib_device_unregister_sysfs(device);
363 spin_lock_irqsave(&device->client_data_lock, flags);
364 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
366 spin_unlock_irqrestore(&device->client_data_lock, flags);
368 device->reg_state = IB_DEV_UNREGISTERED;
370 EXPORT_SYMBOL(ib_unregister_device);
373 * ib_register_client - Register an IB client
374 * @client:Client to register
376 * Upper level users of the IB drivers can use ib_register_client() to
377 * register callbacks for IB device addition and removal. When an IB
378 * device is added, each registered client's add method will be called
379 * (in the order the clients were registered), and when a device is
380 * removed, each client's remove method will be called (in the reverse
381 * order that clients were registered). In addition, when
382 * ib_register_client() is called, the client will receive an add
383 * callback for all devices already registered.
385 int ib_register_client(struct ib_client *client)
387 struct ib_device *device;
389 mutex_lock(&device_mutex);
391 list_add_tail(&client->list, &client_list);
392 list_for_each_entry(device, &device_list, core_list)
393 if (client->add && !add_client_context(device, client))
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 list_for_each_entry(device, &device_list, core_list) {
420 client->remove(device);
422 spin_lock_irqsave(&device->client_data_lock, flags);
423 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
424 if (context->client == client) {
425 list_del(&context->list);
428 spin_unlock_irqrestore(&device->client_data_lock, flags);
430 list_del(&client->list);
432 mutex_unlock(&device_mutex);
434 EXPORT_SYMBOL(ib_unregister_client);
437 * ib_get_client_data - Get IB client context
438 * @device:Device to get context for
439 * @client:Client to get context for
441 * ib_get_client_data() returns client context set with
442 * ib_set_client_data().
444 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
446 struct ib_client_data *context;
450 spin_lock_irqsave(&device->client_data_lock, flags);
451 list_for_each_entry(context, &device->client_data_list, list)
452 if (context->client == client) {
456 spin_unlock_irqrestore(&device->client_data_lock, flags);
460 EXPORT_SYMBOL(ib_get_client_data);
463 * ib_set_client_data - Set IB client context
464 * @device:Device to set context for
465 * @client:Client to set context for
466 * @data:Context to set
468 * ib_set_client_data() sets client context that can be retrieved with
469 * ib_get_client_data().
471 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
474 struct ib_client_data *context;
477 spin_lock_irqsave(&device->client_data_lock, flags);
478 list_for_each_entry(context, &device->client_data_list, list)
479 if (context->client == client) {
480 context->data = data;
484 printk(KERN_WARNING "No client context found for %s/%s\n",
485 device->name, client->name);
488 spin_unlock_irqrestore(&device->client_data_lock, flags);
490 EXPORT_SYMBOL(ib_set_client_data);
493 * ib_register_event_handler - Register an IB event handler
494 * @event_handler:Handler to register
496 * ib_register_event_handler() registers an event handler that will be
497 * called back when asynchronous IB events occur (as defined in
498 * chapter 11 of the InfiniBand Architecture Specification). This
499 * callback may occur in interrupt context.
501 int ib_register_event_handler (struct ib_event_handler *event_handler)
505 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
506 list_add_tail(&event_handler->list,
507 &event_handler->device->event_handler_list);
508 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
512 EXPORT_SYMBOL(ib_register_event_handler);
515 * ib_unregister_event_handler - Unregister an event handler
516 * @event_handler:Handler to unregister
518 * Unregister an event handler registered with
519 * ib_register_event_handler().
521 int ib_unregister_event_handler(struct ib_event_handler *event_handler)
525 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
526 list_del(&event_handler->list);
527 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
531 EXPORT_SYMBOL(ib_unregister_event_handler);
534 * ib_dispatch_event - Dispatch an asynchronous event
535 * @event:Event to dispatch
537 * Low-level drivers must call ib_dispatch_event() to dispatch the
538 * event to all registered event handlers when an asynchronous event
541 void ib_dispatch_event(struct ib_event *event)
544 struct ib_event_handler *handler;
546 spin_lock_irqsave(&event->device->event_handler_lock, flags);
548 list_for_each_entry(handler, &event->device->event_handler_list, list)
549 handler->handler(handler, event);
551 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
553 EXPORT_SYMBOL(ib_dispatch_event);
556 * ib_query_device - Query IB device attributes
557 * @device:Device to query
558 * @device_attr:Device attributes
560 * ib_query_device() returns the attributes of a device through the
561 * @device_attr pointer.
563 int ib_query_device(struct ib_device *device,
564 struct ib_device_attr *device_attr)
566 return device->query_device(device, device_attr);
568 EXPORT_SYMBOL(ib_query_device);
571 * ib_query_port - Query IB port attributes
572 * @device:Device to query
573 * @port_num:Port number to query
574 * @port_attr:Port attributes
576 * ib_query_port() returns the attributes of a port through the
577 * @port_attr pointer.
579 int ib_query_port(struct ib_device *device,
581 struct ib_port_attr *port_attr)
583 if (port_num < start_port(device) || port_num > end_port(device))
586 return device->query_port(device, port_num, port_attr);
588 EXPORT_SYMBOL(ib_query_port);
591 * ib_query_gid - Get GID table entry
592 * @device:Device to query
593 * @port_num:Port number to query
594 * @index:GID table index to query
597 * ib_query_gid() fetches the specified GID table entry.
599 int ib_query_gid(struct ib_device *device,
600 u8 port_num, int index, union ib_gid *gid)
602 return device->query_gid(device, port_num, index, gid);
604 EXPORT_SYMBOL(ib_query_gid);
607 * ib_query_pkey - Get P_Key table entry
608 * @device:Device to query
609 * @port_num:Port number to query
610 * @index:P_Key table index to query
611 * @pkey:Returned P_Key
613 * ib_query_pkey() fetches the specified P_Key table entry.
615 int ib_query_pkey(struct ib_device *device,
616 u8 port_num, u16 index, u16 *pkey)
618 return device->query_pkey(device, port_num, index, pkey);
620 EXPORT_SYMBOL(ib_query_pkey);
623 * ib_modify_device - Change IB device attributes
624 * @device:Device to modify
625 * @device_modify_mask:Mask of attributes to change
626 * @device_modify:New attribute values
628 * ib_modify_device() changes a device's attributes as specified by
629 * the @device_modify_mask and @device_modify structure.
631 int ib_modify_device(struct ib_device *device,
632 int device_modify_mask,
633 struct ib_device_modify *device_modify)
635 return device->modify_device(device, device_modify_mask,
638 EXPORT_SYMBOL(ib_modify_device);
641 * ib_modify_port - Modifies the attributes for the specified port.
642 * @device: The device to modify.
643 * @port_num: The number of the port to modify.
644 * @port_modify_mask: Mask used to specify which attributes of the port
646 * @port_modify: New attribute values for the port.
648 * ib_modify_port() changes a port's attributes as specified by the
649 * @port_modify_mask and @port_modify structure.
651 int ib_modify_port(struct ib_device *device,
652 u8 port_num, int port_modify_mask,
653 struct ib_port_modify *port_modify)
655 if (port_num < start_port(device) || port_num > end_port(device))
658 return device->modify_port(device, port_num, port_modify_mask,
661 EXPORT_SYMBOL(ib_modify_port);
664 * ib_find_gid - Returns the port number and GID table index where
665 * a specified GID value occurs.
666 * @device: The device to query.
667 * @gid: The GID value to search for.
668 * @port_num: The port number of the device where the GID value was found.
669 * @index: The index into the GID table where the GID was found. This
670 * parameter may be NULL.
672 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
673 u8 *port_num, u16 *index)
675 union ib_gid tmp_gid;
678 for (port = start_port(device); port <= end_port(device); ++port) {
679 for (i = 0; i < device->gid_tbl_len[port - start_port(device)]; ++i) {
680 ret = ib_query_gid(device, port, i, &tmp_gid);
683 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
694 EXPORT_SYMBOL(ib_find_gid);
697 * ib_find_pkey - Returns the PKey table index where a specified
699 * @device: The device to query.
700 * @port_num: The port number of the device to search for the PKey.
701 * @pkey: The PKey value to search for.
702 * @index: The index into the PKey table where the PKey was found.
704 int ib_find_pkey(struct ib_device *device,
705 u8 port_num, u16 pkey, u16 *index)
710 for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) {
711 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
715 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
723 EXPORT_SYMBOL(ib_find_pkey);
725 static int __init ib_core_init(void)
730 if (ia64_platform_is("hpzx1"))
731 dma_map_sg_hp_wa = 1;
734 ret = ib_sysfs_setup();
736 printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
738 ret = ib_cache_setup();
740 printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
747 static void __exit ib_core_cleanup(void)
751 /* Make sure that any pending umem accounting work is done. */
752 flush_scheduled_work();
755 module_init(ib_core_init);
756 module_exit(ib_core_cleanup);
758 #undef MODULE_VERSION
759 #include <sys/module.h>
761 ibcore_evhand(module_t mod, int event, void *arg)
766 static moduledata_t ibcore_mod = {
768 .evhand = ibcore_evhand,
771 MODULE_VERSION(ibcore, 1);
772 DECLARE_MODULE(ibcore, ibcore_mod, SI_SUB_SMP, SI_ORDER_ANY);