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/netdevice.h>
42 #include <linux/security.h>
43 #include <linux/notifier.h>
44 #include <rdma/rdma_netlink.h>
45 #include <rdma/ib_addr.h>
46 #include <rdma/ib_cache.h>
48 #include "core_priv.h"
50 MODULE_AUTHOR("Roland Dreier");
51 MODULE_DESCRIPTION("core kernel InfiniBand API");
52 MODULE_LICENSE("Dual BSD/GPL");
54 struct ib_client_data {
55 struct list_head list;
56 struct ib_client *client;
58 /* The device or client is going down. Do not call client or device
59 * callbacks other than remove(). */
63 struct workqueue_struct *ib_comp_wq;
64 struct workqueue_struct *ib_wq;
65 EXPORT_SYMBOL_GPL(ib_wq);
67 /* The device_list and client_list contain devices and clients after their
68 * registration has completed, and the devices and clients are removed
69 * during unregistration. */
70 static LIST_HEAD(device_list);
71 static LIST_HEAD(client_list);
74 * device_mutex and lists_rwsem protect access to both device_list and
75 * client_list. device_mutex protects writer access by device and client
76 * registration / de-registration. lists_rwsem protects reader access to
77 * these lists. Iterators of these lists must lock it for read, while updates
78 * to the lists must be done with a write lock. A special case is when the
79 * device_mutex is locked. In this case locking the lists for read access is
80 * not necessary as the device_mutex implies it.
82 * lists_rwsem also protects access to the client data list.
84 static DEFINE_MUTEX(device_mutex);
85 static DECLARE_RWSEM(lists_rwsem);
87 static int ib_security_change(struct notifier_block *nb, unsigned long event,
89 static void ib_policy_change_task(struct work_struct *work);
90 static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task);
92 static struct notifier_block ibdev_lsm_nb = {
93 .notifier_call = ib_security_change,
96 static int ib_device_check_mandatory(struct ib_device *device)
98 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
102 } mandatory_table[] = {
103 IB_MANDATORY_FUNC(query_device),
104 IB_MANDATORY_FUNC(query_port),
105 IB_MANDATORY_FUNC(query_pkey),
106 IB_MANDATORY_FUNC(query_gid),
107 IB_MANDATORY_FUNC(alloc_pd),
108 IB_MANDATORY_FUNC(dealloc_pd),
109 IB_MANDATORY_FUNC(create_ah),
110 IB_MANDATORY_FUNC(destroy_ah),
111 IB_MANDATORY_FUNC(create_qp),
112 IB_MANDATORY_FUNC(modify_qp),
113 IB_MANDATORY_FUNC(destroy_qp),
114 IB_MANDATORY_FUNC(post_send),
115 IB_MANDATORY_FUNC(post_recv),
116 IB_MANDATORY_FUNC(create_cq),
117 IB_MANDATORY_FUNC(destroy_cq),
118 IB_MANDATORY_FUNC(poll_cq),
119 IB_MANDATORY_FUNC(req_notify_cq),
120 IB_MANDATORY_FUNC(get_dma_mr),
121 IB_MANDATORY_FUNC(dereg_mr),
122 IB_MANDATORY_FUNC(get_port_immutable)
126 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
127 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
128 pr_warn("Device %s is missing mandatory function %s\n",
129 device->name, mandatory_table[i].name);
137 struct ib_device *__ib_device_get_by_index(u32 index)
139 struct ib_device *device;
141 list_for_each_entry(device, &device_list, core_list)
142 if (device->index == index)
148 static struct ib_device *__ib_device_get_by_name(const char *name)
150 struct ib_device *device;
152 list_for_each_entry(device, &device_list, core_list)
153 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
159 static int alloc_name(char *name)
161 unsigned long *inuse;
162 char buf[IB_DEVICE_NAME_MAX];
163 struct ib_device *device;
166 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
170 list_for_each_entry(device, &device_list, core_list) {
171 if (!sscanf(device->name, name, &i))
173 if (i < 0 || i >= PAGE_SIZE * 8)
175 snprintf(buf, sizeof buf, name, i);
176 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
180 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
181 free_page((unsigned long) inuse);
182 snprintf(buf, sizeof buf, name, i);
184 if (__ib_device_get_by_name(buf))
187 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
191 static void ib_device_release(struct device *device)
193 struct ib_device *dev = container_of(device, struct ib_device, dev);
195 WARN_ON(dev->reg_state == IB_DEV_REGISTERED);
196 if (dev->reg_state == IB_DEV_UNREGISTERED) {
198 * In IB_DEV_UNINITIALIZED state, cache or port table
199 * is not even created. Free cache and port table only when
200 * device reaches UNREGISTERED state.
202 ib_cache_release_one(dev);
203 kfree(dev->port_immutable);
208 static int ib_device_uevent(struct device *device,
209 struct kobj_uevent_env *env)
211 struct ib_device *dev = container_of(device, struct ib_device, dev);
213 if (add_uevent_var(env, "NAME=%s", dev->name))
217 * It would be nice to pass the node GUID with the event...
223 static struct class ib_class = {
224 .name = "infiniband",
225 .dev_release = ib_device_release,
226 .dev_uevent = ib_device_uevent,
230 * ib_alloc_device - allocate an IB device struct
231 * @size:size of structure to allocate
233 * Low-level drivers should use ib_alloc_device() to allocate &struct
234 * ib_device. @size is the size of the structure to be allocated,
235 * including any private data used by the low-level driver.
236 * ib_dealloc_device() must be used to free structures allocated with
239 struct ib_device *ib_alloc_device(size_t size)
241 struct ib_device *device;
243 if (WARN_ON(size < sizeof(struct ib_device)))
246 device = kzalloc(size, GFP_KERNEL);
250 device->dev.class = &ib_class;
251 device_initialize(&device->dev);
253 dev_set_drvdata(&device->dev, device);
255 INIT_LIST_HEAD(&device->event_handler_list);
256 spin_lock_init(&device->event_handler_lock);
257 spin_lock_init(&device->client_data_lock);
258 INIT_LIST_HEAD(&device->client_data_list);
259 INIT_LIST_HEAD(&device->port_list);
263 EXPORT_SYMBOL(ib_alloc_device);
266 * ib_dealloc_device - free an IB device struct
267 * @device:structure to free
269 * Free a structure allocated with ib_alloc_device().
271 void ib_dealloc_device(struct ib_device *device)
273 WARN_ON(device->reg_state != IB_DEV_UNREGISTERED &&
274 device->reg_state != IB_DEV_UNINITIALIZED);
275 kobject_put(&device->dev.kobj);
277 EXPORT_SYMBOL(ib_dealloc_device);
279 static int add_client_context(struct ib_device *device, struct ib_client *client)
281 struct ib_client_data *context;
284 context = kmalloc(sizeof *context, GFP_KERNEL);
288 context->client = client;
289 context->data = NULL;
290 context->going_down = false;
292 down_write(&lists_rwsem);
293 spin_lock_irqsave(&device->client_data_lock, flags);
294 list_add(&context->list, &device->client_data_list);
295 spin_unlock_irqrestore(&device->client_data_lock, flags);
296 up_write(&lists_rwsem);
301 static int verify_immutable(const struct ib_device *dev, u8 port)
303 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
304 rdma_max_mad_size(dev, port) != 0);
307 static int read_port_immutable(struct ib_device *device)
310 u8 start_port = rdma_start_port(device);
311 u8 end_port = rdma_end_port(device);
315 * device->port_immutable is indexed directly by the port number to make
316 * access to this data as efficient as possible.
318 * Therefore port_immutable is declared as a 1 based array with
319 * potential empty slots at the beginning.
321 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
324 if (!device->port_immutable)
327 for (port = start_port; port <= end_port; ++port) {
328 ret = device->get_port_immutable(device, port,
329 &device->port_immutable[port]);
333 if (verify_immutable(device, port))
339 void ib_get_device_fw_str(struct ib_device *dev, char *str)
341 if (dev->get_dev_fw_str)
342 dev->get_dev_fw_str(dev, str);
346 EXPORT_SYMBOL(ib_get_device_fw_str);
348 static int setup_port_pkey_list(struct ib_device *device)
353 * device->port_pkey_list is indexed directly by the port number,
354 * Therefore it is declared as a 1 based array with potential empty
355 * slots at the beginning.
357 device->port_pkey_list = kcalloc(rdma_end_port(device) + 1,
358 sizeof(*device->port_pkey_list),
361 if (!device->port_pkey_list)
364 for (i = 0; i < (rdma_end_port(device) + 1); i++) {
365 spin_lock_init(&device->port_pkey_list[i].list_lock);
366 INIT_LIST_HEAD(&device->port_pkey_list[i].pkey_list);
372 static void ib_policy_change_task(struct work_struct *work)
374 struct ib_device *dev;
376 down_read(&lists_rwsem);
377 list_for_each_entry(dev, &device_list, core_list) {
380 for (i = rdma_start_port(dev); i <= rdma_end_port(dev); i++) {
382 int ret = ib_get_cached_subnet_prefix(dev,
387 "ib_get_cached_subnet_prefix err: %d, this should never happen here\n",
390 ib_security_cache_change(dev, i, sp);
393 up_read(&lists_rwsem);
396 static int ib_security_change(struct notifier_block *nb, unsigned long event,
399 if (event != LSM_POLICY_CHANGE)
402 schedule_work(&ib_policy_change_work);
408 * __dev_new_index - allocate an device index
410 * Returns a suitable unique value for a new device interface
411 * number. It assumes that there are less than 2^32-1 ib devices
412 * will be present in the system.
414 static u32 __dev_new_index(void)
417 * The device index to allow stable naming.
418 * Similar to struct net -> ifindex.
426 if (!__ib_device_get_by_index(index))
432 * ib_register_device - Register an IB device with IB core
433 * @device:Device to register
435 * Low-level drivers use ib_register_device() to register their
436 * devices with the IB core. All registered clients will receive a
437 * callback for each device that is added. @device must be allocated
438 * with ib_alloc_device().
440 int ib_register_device(struct ib_device *device,
441 int (*port_callback)(struct ib_device *,
442 u8, struct kobject *))
445 struct ib_client *client;
446 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
447 struct device *parent = device->dev.parent;
449 WARN_ON_ONCE(!parent);
450 WARN_ON_ONCE(device->dma_device);
451 if (device->dev.dma_ops) {
453 * The caller provided custom DMA operations. Copy the
454 * DMA-related fields that are used by e.g. dma_alloc_coherent()
457 device->dma_device = &device->dev;
458 if (!device->dev.dma_mask)
459 device->dev.dma_mask = parent->dma_mask;
460 if (!device->dev.coherent_dma_mask)
461 device->dev.coherent_dma_mask =
462 parent->coherent_dma_mask;
465 * The caller did not provide custom DMA operations. Use the
466 * DMA mapping operations of the parent device.
468 device->dma_device = parent;
471 mutex_lock(&device_mutex);
473 if (strchr(device->name, '%')) {
474 ret = alloc_name(device->name);
479 if (ib_device_check_mandatory(device)) {
484 ret = read_port_immutable(device);
486 pr_warn("Couldn't create per port immutable data %s\n",
491 ret = setup_port_pkey_list(device);
493 pr_warn("Couldn't create per port_pkey_list\n");
497 ret = ib_cache_setup_one(device);
499 pr_warn("Couldn't set up InfiniBand P_Key/GID cache\n");
503 ret = ib_device_register_rdmacg(device);
505 pr_warn("Couldn't register device with rdma cgroup\n");
509 memset(&device->attrs, 0, sizeof(device->attrs));
510 ret = device->query_device(device, &device->attrs, &uhw);
512 pr_warn("Couldn't query the device attributes\n");
516 ret = ib_device_register_sysfs(device, port_callback);
518 pr_warn("Couldn't register device %s with driver model\n",
523 device->reg_state = IB_DEV_REGISTERED;
525 list_for_each_entry(client, &client_list, list)
526 if (!add_client_context(device, client) && client->add)
529 device->index = __dev_new_index();
530 down_write(&lists_rwsem);
531 list_add_tail(&device->core_list, &device_list);
532 up_write(&lists_rwsem);
533 mutex_unlock(&device_mutex);
537 ib_cache_cleanup_one(device);
538 ib_cache_release_one(device);
540 kfree(device->port_immutable);
542 mutex_unlock(&device_mutex);
545 EXPORT_SYMBOL(ib_register_device);
548 * ib_unregister_device - Unregister an IB device
549 * @device:Device to unregister
551 * Unregister an IB device. All clients will receive a remove callback.
553 void ib_unregister_device(struct ib_device *device)
555 struct ib_client_data *context, *tmp;
558 mutex_lock(&device_mutex);
560 down_write(&lists_rwsem);
561 list_del(&device->core_list);
562 spin_lock_irqsave(&device->client_data_lock, flags);
563 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
564 context->going_down = true;
565 spin_unlock_irqrestore(&device->client_data_lock, flags);
566 downgrade_write(&lists_rwsem);
568 list_for_each_entry_safe(context, tmp, &device->client_data_list,
570 if (context->client->remove)
571 context->client->remove(device, context->data);
573 up_read(&lists_rwsem);
575 ib_device_unregister_rdmacg(device);
576 ib_device_unregister_sysfs(device);
578 mutex_unlock(&device_mutex);
580 ib_cache_cleanup_one(device);
582 ib_security_destroy_port_pkey_list(device);
583 kfree(device->port_pkey_list);
585 down_write(&lists_rwsem);
586 spin_lock_irqsave(&device->client_data_lock, flags);
587 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
589 spin_unlock_irqrestore(&device->client_data_lock, flags);
590 up_write(&lists_rwsem);
592 device->reg_state = IB_DEV_UNREGISTERED;
594 EXPORT_SYMBOL(ib_unregister_device);
597 * ib_register_client - Register an IB client
598 * @client:Client to register
600 * Upper level users of the IB drivers can use ib_register_client() to
601 * register callbacks for IB device addition and removal. When an IB
602 * device is added, each registered client's add method will be called
603 * (in the order the clients were registered), and when a device is
604 * removed, each client's remove method will be called (in the reverse
605 * order that clients were registered). In addition, when
606 * ib_register_client() is called, the client will receive an add
607 * callback for all devices already registered.
609 int ib_register_client(struct ib_client *client)
611 struct ib_device *device;
613 mutex_lock(&device_mutex);
615 list_for_each_entry(device, &device_list, core_list)
616 if (!add_client_context(device, client) && client->add)
619 down_write(&lists_rwsem);
620 list_add_tail(&client->list, &client_list);
621 up_write(&lists_rwsem);
623 mutex_unlock(&device_mutex);
627 EXPORT_SYMBOL(ib_register_client);
630 * ib_unregister_client - Unregister an IB client
631 * @client:Client to unregister
633 * Upper level users use ib_unregister_client() to remove their client
634 * registration. When ib_unregister_client() is called, the client
635 * will receive a remove callback for each IB device still registered.
637 void ib_unregister_client(struct ib_client *client)
639 struct ib_client_data *context, *tmp;
640 struct ib_device *device;
643 mutex_lock(&device_mutex);
645 down_write(&lists_rwsem);
646 list_del(&client->list);
647 up_write(&lists_rwsem);
649 list_for_each_entry(device, &device_list, core_list) {
650 struct ib_client_data *found_context = NULL;
652 down_write(&lists_rwsem);
653 spin_lock_irqsave(&device->client_data_lock, flags);
654 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
655 if (context->client == client) {
656 context->going_down = true;
657 found_context = context;
660 spin_unlock_irqrestore(&device->client_data_lock, flags);
661 up_write(&lists_rwsem);
664 client->remove(device, found_context ?
665 found_context->data : NULL);
667 if (!found_context) {
668 pr_warn("No client context found for %s/%s\n",
669 device->name, client->name);
673 down_write(&lists_rwsem);
674 spin_lock_irqsave(&device->client_data_lock, flags);
675 list_del(&found_context->list);
676 kfree(found_context);
677 spin_unlock_irqrestore(&device->client_data_lock, flags);
678 up_write(&lists_rwsem);
681 mutex_unlock(&device_mutex);
683 EXPORT_SYMBOL(ib_unregister_client);
686 * ib_get_client_data - Get IB client context
687 * @device:Device to get context for
688 * @client:Client to get context for
690 * ib_get_client_data() returns client context set with
691 * ib_set_client_data().
693 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
695 struct ib_client_data *context;
699 spin_lock_irqsave(&device->client_data_lock, flags);
700 list_for_each_entry(context, &device->client_data_list, list)
701 if (context->client == client) {
705 spin_unlock_irqrestore(&device->client_data_lock, flags);
709 EXPORT_SYMBOL(ib_get_client_data);
712 * ib_set_client_data - Set IB client context
713 * @device:Device to set context for
714 * @client:Client to set context for
715 * @data:Context to set
717 * ib_set_client_data() sets client context that can be retrieved with
718 * ib_get_client_data().
720 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
723 struct ib_client_data *context;
726 spin_lock_irqsave(&device->client_data_lock, flags);
727 list_for_each_entry(context, &device->client_data_list, list)
728 if (context->client == client) {
729 context->data = data;
733 pr_warn("No client context found for %s/%s\n",
734 device->name, client->name);
737 spin_unlock_irqrestore(&device->client_data_lock, flags);
739 EXPORT_SYMBOL(ib_set_client_data);
742 * ib_register_event_handler - Register an IB event handler
743 * @event_handler:Handler to register
745 * ib_register_event_handler() registers an event handler that will be
746 * called back when asynchronous IB events occur (as defined in
747 * chapter 11 of the InfiniBand Architecture Specification). This
748 * callback may occur in interrupt context.
750 void ib_register_event_handler(struct ib_event_handler *event_handler)
754 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
755 list_add_tail(&event_handler->list,
756 &event_handler->device->event_handler_list);
757 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
759 EXPORT_SYMBOL(ib_register_event_handler);
762 * ib_unregister_event_handler - Unregister an event handler
763 * @event_handler:Handler to unregister
765 * Unregister an event handler registered with
766 * ib_register_event_handler().
768 void ib_unregister_event_handler(struct ib_event_handler *event_handler)
772 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
773 list_del(&event_handler->list);
774 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
776 EXPORT_SYMBOL(ib_unregister_event_handler);
779 * ib_dispatch_event - Dispatch an asynchronous event
780 * @event:Event to dispatch
782 * Low-level drivers must call ib_dispatch_event() to dispatch the
783 * event to all registered event handlers when an asynchronous event
786 void ib_dispatch_event(struct ib_event *event)
789 struct ib_event_handler *handler;
791 spin_lock_irqsave(&event->device->event_handler_lock, flags);
793 list_for_each_entry(handler, &event->device->event_handler_list, list)
794 handler->handler(handler, event);
796 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
798 EXPORT_SYMBOL(ib_dispatch_event);
801 * ib_query_port - Query IB port attributes
802 * @device:Device to query
803 * @port_num:Port number to query
804 * @port_attr:Port attributes
806 * ib_query_port() returns the attributes of a port through the
807 * @port_attr pointer.
809 int ib_query_port(struct ib_device *device,
811 struct ib_port_attr *port_attr)
816 if (!rdma_is_port_valid(device, port_num))
819 memset(port_attr, 0, sizeof(*port_attr));
820 err = device->query_port(device, port_num, port_attr);
821 if (err || port_attr->subnet_prefix)
824 if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND)
827 err = ib_query_gid(device, port_num, 0, &gid, NULL);
831 port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix);
834 EXPORT_SYMBOL(ib_query_port);
837 * ib_query_gid - Get GID table entry
838 * @device:Device to query
839 * @port_num:Port number to query
840 * @index:GID table index to query
842 * @attr: Returned GID attributes related to this GID index (only in RoCE).
845 * ib_query_gid() fetches the specified GID table entry.
847 int ib_query_gid(struct ib_device *device,
848 u8 port_num, int index, union ib_gid *gid,
849 struct ib_gid_attr *attr)
851 if (rdma_cap_roce_gid_table(device, port_num))
852 return ib_get_cached_gid(device, port_num, index, gid, attr);
857 return device->query_gid(device, port_num, index, gid);
859 EXPORT_SYMBOL(ib_query_gid);
862 * ib_enum_roce_netdev - enumerate all RoCE ports
863 * @ib_dev : IB device we want to query
864 * @filter: Should we call the callback?
865 * @filter_cookie: Cookie passed to filter
866 * @cb: Callback to call for each found RoCE ports
867 * @cookie: Cookie passed back to the callback
869 * Enumerates all of the physical RoCE ports of ib_dev
870 * which are related to netdevice and calls callback() on each
871 * device for which filter() function returns non zero.
873 void ib_enum_roce_netdev(struct ib_device *ib_dev,
874 roce_netdev_filter filter,
876 roce_netdev_callback cb,
881 for (port = rdma_start_port(ib_dev); port <= rdma_end_port(ib_dev);
883 if (rdma_protocol_roce(ib_dev, port)) {
884 struct net_device *idev = NULL;
886 if (ib_dev->get_netdev)
887 idev = ib_dev->get_netdev(ib_dev, port);
890 idev->reg_state >= NETREG_UNREGISTERED) {
895 if (filter(ib_dev, port, idev, filter_cookie))
896 cb(ib_dev, port, idev, cookie);
904 * ib_enum_all_roce_netdevs - enumerate all RoCE devices
905 * @filter: Should we call the callback?
906 * @filter_cookie: Cookie passed to filter
907 * @cb: Callback to call for each found RoCE ports
908 * @cookie: Cookie passed back to the callback
910 * Enumerates all RoCE devices' physical ports which are related
911 * to netdevices and calls callback() on each device for which
912 * filter() function returns non zero.
914 void ib_enum_all_roce_netdevs(roce_netdev_filter filter,
916 roce_netdev_callback cb,
919 struct ib_device *dev;
921 down_read(&lists_rwsem);
922 list_for_each_entry(dev, &device_list, core_list)
923 ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie);
924 up_read(&lists_rwsem);
928 * ib_enum_all_devs - enumerate all ib_devices
929 * @cb: Callback to call for each found ib_device
931 * Enumerates all ib_devices and calls callback() on each device.
933 int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb,
934 struct netlink_callback *cb)
936 struct ib_device *dev;
937 unsigned int idx = 0;
940 down_read(&lists_rwsem);
941 list_for_each_entry(dev, &device_list, core_list) {
942 ret = nldev_cb(dev, skb, cb, idx);
948 up_read(&lists_rwsem);
953 * ib_query_pkey - Get P_Key table entry
954 * @device:Device to query
955 * @port_num:Port number to query
956 * @index:P_Key table index to query
957 * @pkey:Returned P_Key
959 * ib_query_pkey() fetches the specified P_Key table entry.
961 int ib_query_pkey(struct ib_device *device,
962 u8 port_num, u16 index, u16 *pkey)
964 return device->query_pkey(device, port_num, index, pkey);
966 EXPORT_SYMBOL(ib_query_pkey);
969 * ib_modify_device - Change IB device attributes
970 * @device:Device to modify
971 * @device_modify_mask:Mask of attributes to change
972 * @device_modify:New attribute values
974 * ib_modify_device() changes a device's attributes as specified by
975 * the @device_modify_mask and @device_modify structure.
977 int ib_modify_device(struct ib_device *device,
978 int device_modify_mask,
979 struct ib_device_modify *device_modify)
981 if (!device->modify_device)
984 return device->modify_device(device, device_modify_mask,
987 EXPORT_SYMBOL(ib_modify_device);
990 * ib_modify_port - Modifies the attributes for the specified port.
991 * @device: The device to modify.
992 * @port_num: The number of the port to modify.
993 * @port_modify_mask: Mask used to specify which attributes of the port
995 * @port_modify: New attribute values for the port.
997 * ib_modify_port() changes a port's attributes as specified by the
998 * @port_modify_mask and @port_modify structure.
1000 int ib_modify_port(struct ib_device *device,
1001 u8 port_num, int port_modify_mask,
1002 struct ib_port_modify *port_modify)
1006 if (!rdma_is_port_valid(device, port_num))
1009 if (device->modify_port)
1010 rc = device->modify_port(device, port_num, port_modify_mask,
1013 rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS;
1016 EXPORT_SYMBOL(ib_modify_port);
1019 * ib_find_gid - Returns the port number and GID table index where
1020 * a specified GID value occurs.
1021 * @device: The device to query.
1022 * @gid: The GID value to search for.
1023 * @gid_type: Type of GID.
1024 * @ndev: The ndev related to the GID to search for.
1025 * @port_num: The port number of the device where the GID value was found.
1026 * @index: The index into the GID table where the GID was found. This
1027 * parameter may be NULL.
1029 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1030 enum ib_gid_type gid_type, struct net_device *ndev,
1031 u8 *port_num, u16 *index)
1033 union ib_gid tmp_gid;
1036 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
1037 if (rdma_cap_roce_gid_table(device, port)) {
1038 if (!ib_find_cached_gid_by_port(device, gid, gid_type, port,
1045 if (gid_type != IB_GID_TYPE_IB)
1048 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
1049 ret = ib_query_gid(device, port, i, &tmp_gid, NULL);
1052 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
1063 EXPORT_SYMBOL(ib_find_gid);
1066 * ib_find_pkey - Returns the PKey table index where a specified
1067 * PKey value occurs.
1068 * @device: The device to query.
1069 * @port_num: The port number of the device to search for the PKey.
1070 * @pkey: The PKey value to search for.
1071 * @index: The index into the PKey table where the PKey was found.
1073 int ib_find_pkey(struct ib_device *device,
1074 u8 port_num, u16 pkey, u16 *index)
1078 int partial_ix = -1;
1080 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
1081 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
1084 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
1085 /* if there is full-member pkey take it.*/
1086 if (tmp_pkey & 0x8000) {
1095 /*no full-member, if exists take the limited*/
1096 if (partial_ix >= 0) {
1097 *index = partial_ix;
1102 EXPORT_SYMBOL(ib_find_pkey);
1105 * ib_get_net_dev_by_params() - Return the appropriate net_dev
1106 * for a received CM request
1107 * @dev: An RDMA device on which the request has been received.
1108 * @port: Port number on the RDMA device.
1109 * @pkey: The Pkey the request came on.
1110 * @gid: A GID that the net_dev uses to communicate.
1111 * @addr: Contains the IP address that the request specified as its
1114 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev,
1117 const union ib_gid *gid,
1118 const struct sockaddr *addr)
1120 struct net_device *net_dev = NULL;
1121 struct ib_client_data *context;
1123 if (!rdma_protocol_ib(dev, port))
1126 down_read(&lists_rwsem);
1128 list_for_each_entry(context, &dev->client_data_list, list) {
1129 struct ib_client *client = context->client;
1131 if (context->going_down)
1134 if (client->get_net_dev_by_params) {
1135 net_dev = client->get_net_dev_by_params(dev, port, pkey,
1143 up_read(&lists_rwsem);
1147 EXPORT_SYMBOL(ib_get_net_dev_by_params);
1149 static const struct rdma_nl_cbs ibnl_ls_cb_table[] = {
1150 [RDMA_NL_LS_OP_RESOLVE] = {
1151 .doit = ib_nl_handle_resolve_resp,
1152 .flags = RDMA_NL_ADMIN_PERM,
1154 [RDMA_NL_LS_OP_SET_TIMEOUT] = {
1155 .doit = ib_nl_handle_set_timeout,
1156 .flags = RDMA_NL_ADMIN_PERM,
1158 [RDMA_NL_LS_OP_IP_RESOLVE] = {
1159 .doit = ib_nl_handle_ip_res_resp,
1160 .flags = RDMA_NL_ADMIN_PERM,
1164 static int __init ib_core_init(void)
1168 ib_wq = alloc_workqueue("infiniband", 0, 0);
1172 ib_comp_wq = alloc_workqueue("ib-comp-wq",
1173 WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
1179 ret = class_register(&ib_class);
1181 pr_warn("Couldn't create InfiniBand device class\n");
1185 ret = rdma_nl_init();
1187 pr_warn("Couldn't init IB netlink interface: err %d\n", ret);
1193 pr_warn("Could't init IB address resolution\n");
1197 ret = ib_mad_init();
1199 pr_warn("Couldn't init IB MAD\n");
1205 pr_warn("Couldn't init SA\n");
1209 ret = register_lsm_notifier(&ibdev_lsm_nb);
1211 pr_warn("Couldn't register LSM notifier. ret %d\n", ret);
1216 rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table);
1230 class_unregister(&ib_class);
1232 destroy_workqueue(ib_comp_wq);
1234 destroy_workqueue(ib_wq);
1238 static void __exit ib_core_cleanup(void)
1242 rdma_nl_unregister(RDMA_NL_LS);
1243 unregister_lsm_notifier(&ibdev_lsm_nb);
1248 class_unregister(&ib_class);
1249 destroy_workqueue(ib_comp_wq);
1250 /* Make sure that any pending umem accounting work is done. */
1251 destroy_workqueue(ib_wq);
1254 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4);
1256 module_init(ib_core_init);
1257 module_exit(ib_core_cleanup);