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 static 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)
149 * Caller is responsible to return refrerence count by calling put_device()
151 struct ib_device *ib_device_get_by_index(u32 index)
153 struct ib_device *device;
155 down_read(&lists_rwsem);
156 device = __ib_device_get_by_index(index);
158 get_device(&device->dev);
160 up_read(&lists_rwsem);
164 static struct ib_device *__ib_device_get_by_name(const char *name)
166 struct ib_device *device;
168 list_for_each_entry(device, &device_list, core_list)
169 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
175 static int alloc_name(char *name)
177 unsigned long *inuse;
178 char buf[IB_DEVICE_NAME_MAX];
179 struct ib_device *device;
182 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
186 list_for_each_entry(device, &device_list, core_list) {
187 if (!sscanf(device->name, name, &i))
189 if (i < 0 || i >= PAGE_SIZE * 8)
191 snprintf(buf, sizeof buf, name, i);
192 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
196 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
197 free_page((unsigned long) inuse);
198 snprintf(buf, sizeof buf, name, i);
200 if (__ib_device_get_by_name(buf))
203 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
207 static void ib_device_release(struct device *device)
209 struct ib_device *dev = container_of(device, struct ib_device, dev);
211 WARN_ON(dev->reg_state == IB_DEV_REGISTERED);
212 if (dev->reg_state == IB_DEV_UNREGISTERED) {
214 * In IB_DEV_UNINITIALIZED state, cache or port table
215 * is not even created. Free cache and port table only when
216 * device reaches UNREGISTERED state.
218 ib_cache_release_one(dev);
219 kfree(dev->port_immutable);
224 static int ib_device_uevent(struct device *device,
225 struct kobj_uevent_env *env)
227 struct ib_device *dev = container_of(device, struct ib_device, dev);
229 if (add_uevent_var(env, "NAME=%s", dev->name))
233 * It would be nice to pass the node GUID with the event...
239 static struct class ib_class = {
240 .name = "infiniband",
241 .dev_release = ib_device_release,
242 .dev_uevent = ib_device_uevent,
246 * ib_alloc_device - allocate an IB device struct
247 * @size:size of structure to allocate
249 * Low-level drivers should use ib_alloc_device() to allocate &struct
250 * ib_device. @size is the size of the structure to be allocated,
251 * including any private data used by the low-level driver.
252 * ib_dealloc_device() must be used to free structures allocated with
255 struct ib_device *ib_alloc_device(size_t size)
257 struct ib_device *device;
259 if (WARN_ON(size < sizeof(struct ib_device)))
262 device = kzalloc(size, GFP_KERNEL);
266 device->dev.class = &ib_class;
267 device_initialize(&device->dev);
269 dev_set_drvdata(&device->dev, device);
271 INIT_LIST_HEAD(&device->event_handler_list);
272 spin_lock_init(&device->event_handler_lock);
273 spin_lock_init(&device->client_data_lock);
274 INIT_LIST_HEAD(&device->client_data_list);
275 INIT_LIST_HEAD(&device->port_list);
279 EXPORT_SYMBOL(ib_alloc_device);
282 * ib_dealloc_device - free an IB device struct
283 * @device:structure to free
285 * Free a structure allocated with ib_alloc_device().
287 void ib_dealloc_device(struct ib_device *device)
289 WARN_ON(device->reg_state != IB_DEV_UNREGISTERED &&
290 device->reg_state != IB_DEV_UNINITIALIZED);
291 kobject_put(&device->dev.kobj);
293 EXPORT_SYMBOL(ib_dealloc_device);
295 static int add_client_context(struct ib_device *device, struct ib_client *client)
297 struct ib_client_data *context;
300 context = kmalloc(sizeof *context, GFP_KERNEL);
304 context->client = client;
305 context->data = NULL;
306 context->going_down = false;
308 down_write(&lists_rwsem);
309 spin_lock_irqsave(&device->client_data_lock, flags);
310 list_add(&context->list, &device->client_data_list);
311 spin_unlock_irqrestore(&device->client_data_lock, flags);
312 up_write(&lists_rwsem);
317 static int verify_immutable(const struct ib_device *dev, u8 port)
319 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
320 rdma_max_mad_size(dev, port) != 0);
323 static int read_port_immutable(struct ib_device *device)
326 u8 start_port = rdma_start_port(device);
327 u8 end_port = rdma_end_port(device);
331 * device->port_immutable is indexed directly by the port number to make
332 * access to this data as efficient as possible.
334 * Therefore port_immutable is declared as a 1 based array with
335 * potential empty slots at the beginning.
337 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
340 if (!device->port_immutable)
343 for (port = start_port; port <= end_port; ++port) {
344 ret = device->get_port_immutable(device, port,
345 &device->port_immutable[port]);
349 if (verify_immutable(device, port))
355 void ib_get_device_fw_str(struct ib_device *dev, char *str)
357 if (dev->get_dev_fw_str)
358 dev->get_dev_fw_str(dev, str);
362 EXPORT_SYMBOL(ib_get_device_fw_str);
364 static int setup_port_pkey_list(struct ib_device *device)
369 * device->port_pkey_list is indexed directly by the port number,
370 * Therefore it is declared as a 1 based array with potential empty
371 * slots at the beginning.
373 device->port_pkey_list = kcalloc(rdma_end_port(device) + 1,
374 sizeof(*device->port_pkey_list),
377 if (!device->port_pkey_list)
380 for (i = 0; i < (rdma_end_port(device) + 1); i++) {
381 spin_lock_init(&device->port_pkey_list[i].list_lock);
382 INIT_LIST_HEAD(&device->port_pkey_list[i].pkey_list);
388 static void ib_policy_change_task(struct work_struct *work)
390 struct ib_device *dev;
392 down_read(&lists_rwsem);
393 list_for_each_entry(dev, &device_list, core_list) {
396 for (i = rdma_start_port(dev); i <= rdma_end_port(dev); i++) {
398 int ret = ib_get_cached_subnet_prefix(dev,
403 "ib_get_cached_subnet_prefix err: %d, this should never happen here\n",
406 ib_security_cache_change(dev, i, sp);
409 up_read(&lists_rwsem);
412 static int ib_security_change(struct notifier_block *nb, unsigned long event,
415 if (event != LSM_POLICY_CHANGE)
418 schedule_work(&ib_policy_change_work);
424 * __dev_new_index - allocate an device index
426 * Returns a suitable unique value for a new device interface
427 * number. It assumes that there are less than 2^32-1 ib devices
428 * will be present in the system.
430 static u32 __dev_new_index(void)
433 * The device index to allow stable naming.
434 * Similar to struct net -> ifindex.
442 if (!__ib_device_get_by_index(index))
448 * ib_register_device - Register an IB device with IB core
449 * @device:Device to register
451 * Low-level drivers use ib_register_device() to register their
452 * devices with the IB core. All registered clients will receive a
453 * callback for each device that is added. @device must be allocated
454 * with ib_alloc_device().
456 int ib_register_device(struct ib_device *device,
457 int (*port_callback)(struct ib_device *,
458 u8, struct kobject *))
461 struct ib_client *client;
462 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
463 struct device *parent = device->dev.parent;
465 WARN_ON_ONCE(!parent);
466 WARN_ON_ONCE(device->dma_device);
467 if (device->dev.dma_ops) {
469 * The caller provided custom DMA operations. Copy the
470 * DMA-related fields that are used by e.g. dma_alloc_coherent()
473 device->dma_device = &device->dev;
474 if (!device->dev.dma_mask)
475 device->dev.dma_mask = parent->dma_mask;
476 if (!device->dev.coherent_dma_mask)
477 device->dev.coherent_dma_mask =
478 parent->coherent_dma_mask;
481 * The caller did not provide custom DMA operations. Use the
482 * DMA mapping operations of the parent device.
484 device->dma_device = parent;
487 mutex_lock(&device_mutex);
489 if (strchr(device->name, '%')) {
490 ret = alloc_name(device->name);
495 if (ib_device_check_mandatory(device)) {
500 ret = read_port_immutable(device);
502 pr_warn("Couldn't create per port immutable data %s\n",
507 ret = setup_port_pkey_list(device);
509 pr_warn("Couldn't create per port_pkey_list\n");
513 ret = ib_cache_setup_one(device);
515 pr_warn("Couldn't set up InfiniBand P_Key/GID cache\n");
519 ret = ib_device_register_rdmacg(device);
521 pr_warn("Couldn't register device with rdma cgroup\n");
525 memset(&device->attrs, 0, sizeof(device->attrs));
526 ret = device->query_device(device, &device->attrs, &uhw);
528 pr_warn("Couldn't query the device attributes\n");
532 ret = ib_device_register_sysfs(device, port_callback);
534 pr_warn("Couldn't register device %s with driver model\n",
539 device->reg_state = IB_DEV_REGISTERED;
541 list_for_each_entry(client, &client_list, list)
542 if (!add_client_context(device, client) && client->add)
545 device->index = __dev_new_index();
546 down_write(&lists_rwsem);
547 list_add_tail(&device->core_list, &device_list);
548 up_write(&lists_rwsem);
549 mutex_unlock(&device_mutex);
553 ib_cache_cleanup_one(device);
554 ib_cache_release_one(device);
556 kfree(device->port_immutable);
558 mutex_unlock(&device_mutex);
561 EXPORT_SYMBOL(ib_register_device);
564 * ib_unregister_device - Unregister an IB device
565 * @device:Device to unregister
567 * Unregister an IB device. All clients will receive a remove callback.
569 void ib_unregister_device(struct ib_device *device)
571 struct ib_client_data *context, *tmp;
574 mutex_lock(&device_mutex);
576 down_write(&lists_rwsem);
577 list_del(&device->core_list);
578 spin_lock_irqsave(&device->client_data_lock, flags);
579 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
580 context->going_down = true;
581 spin_unlock_irqrestore(&device->client_data_lock, flags);
582 downgrade_write(&lists_rwsem);
584 list_for_each_entry_safe(context, tmp, &device->client_data_list,
586 if (context->client->remove)
587 context->client->remove(device, context->data);
589 up_read(&lists_rwsem);
591 ib_device_unregister_rdmacg(device);
592 ib_device_unregister_sysfs(device);
594 mutex_unlock(&device_mutex);
596 ib_cache_cleanup_one(device);
598 ib_security_destroy_port_pkey_list(device);
599 kfree(device->port_pkey_list);
601 down_write(&lists_rwsem);
602 spin_lock_irqsave(&device->client_data_lock, flags);
603 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
605 spin_unlock_irqrestore(&device->client_data_lock, flags);
606 up_write(&lists_rwsem);
608 device->reg_state = IB_DEV_UNREGISTERED;
610 EXPORT_SYMBOL(ib_unregister_device);
613 * ib_register_client - Register an IB client
614 * @client:Client to register
616 * Upper level users of the IB drivers can use ib_register_client() to
617 * register callbacks for IB device addition and removal. When an IB
618 * device is added, each registered client's add method will be called
619 * (in the order the clients were registered), and when a device is
620 * removed, each client's remove method will be called (in the reverse
621 * order that clients were registered). In addition, when
622 * ib_register_client() is called, the client will receive an add
623 * callback for all devices already registered.
625 int ib_register_client(struct ib_client *client)
627 struct ib_device *device;
629 mutex_lock(&device_mutex);
631 list_for_each_entry(device, &device_list, core_list)
632 if (!add_client_context(device, client) && client->add)
635 down_write(&lists_rwsem);
636 list_add_tail(&client->list, &client_list);
637 up_write(&lists_rwsem);
639 mutex_unlock(&device_mutex);
643 EXPORT_SYMBOL(ib_register_client);
646 * ib_unregister_client - Unregister an IB client
647 * @client:Client to unregister
649 * Upper level users use ib_unregister_client() to remove their client
650 * registration. When ib_unregister_client() is called, the client
651 * will receive a remove callback for each IB device still registered.
653 void ib_unregister_client(struct ib_client *client)
655 struct ib_client_data *context, *tmp;
656 struct ib_device *device;
659 mutex_lock(&device_mutex);
661 down_write(&lists_rwsem);
662 list_del(&client->list);
663 up_write(&lists_rwsem);
665 list_for_each_entry(device, &device_list, core_list) {
666 struct ib_client_data *found_context = NULL;
668 down_write(&lists_rwsem);
669 spin_lock_irqsave(&device->client_data_lock, flags);
670 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
671 if (context->client == client) {
672 context->going_down = true;
673 found_context = context;
676 spin_unlock_irqrestore(&device->client_data_lock, flags);
677 up_write(&lists_rwsem);
680 client->remove(device, found_context ?
681 found_context->data : NULL);
683 if (!found_context) {
684 pr_warn("No client context found for %s/%s\n",
685 device->name, client->name);
689 down_write(&lists_rwsem);
690 spin_lock_irqsave(&device->client_data_lock, flags);
691 list_del(&found_context->list);
692 kfree(found_context);
693 spin_unlock_irqrestore(&device->client_data_lock, flags);
694 up_write(&lists_rwsem);
697 mutex_unlock(&device_mutex);
699 EXPORT_SYMBOL(ib_unregister_client);
702 * ib_get_client_data - Get IB client context
703 * @device:Device to get context for
704 * @client:Client to get context for
706 * ib_get_client_data() returns client context set with
707 * ib_set_client_data().
709 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
711 struct ib_client_data *context;
715 spin_lock_irqsave(&device->client_data_lock, flags);
716 list_for_each_entry(context, &device->client_data_list, list)
717 if (context->client == client) {
721 spin_unlock_irqrestore(&device->client_data_lock, flags);
725 EXPORT_SYMBOL(ib_get_client_data);
728 * ib_set_client_data - Set IB client context
729 * @device:Device to set context for
730 * @client:Client to set context for
731 * @data:Context to set
733 * ib_set_client_data() sets client context that can be retrieved with
734 * ib_get_client_data().
736 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
739 struct ib_client_data *context;
742 spin_lock_irqsave(&device->client_data_lock, flags);
743 list_for_each_entry(context, &device->client_data_list, list)
744 if (context->client == client) {
745 context->data = data;
749 pr_warn("No client context found for %s/%s\n",
750 device->name, client->name);
753 spin_unlock_irqrestore(&device->client_data_lock, flags);
755 EXPORT_SYMBOL(ib_set_client_data);
758 * ib_register_event_handler - Register an IB event handler
759 * @event_handler:Handler to register
761 * ib_register_event_handler() registers an event handler that will be
762 * called back when asynchronous IB events occur (as defined in
763 * chapter 11 of the InfiniBand Architecture Specification). This
764 * callback may occur in interrupt context.
766 void ib_register_event_handler(struct ib_event_handler *event_handler)
770 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
771 list_add_tail(&event_handler->list,
772 &event_handler->device->event_handler_list);
773 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
775 EXPORT_SYMBOL(ib_register_event_handler);
778 * ib_unregister_event_handler - Unregister an event handler
779 * @event_handler:Handler to unregister
781 * Unregister an event handler registered with
782 * ib_register_event_handler().
784 void ib_unregister_event_handler(struct ib_event_handler *event_handler)
788 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
789 list_del(&event_handler->list);
790 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
792 EXPORT_SYMBOL(ib_unregister_event_handler);
795 * ib_dispatch_event - Dispatch an asynchronous event
796 * @event:Event to dispatch
798 * Low-level drivers must call ib_dispatch_event() to dispatch the
799 * event to all registered event handlers when an asynchronous event
802 void ib_dispatch_event(struct ib_event *event)
805 struct ib_event_handler *handler;
807 spin_lock_irqsave(&event->device->event_handler_lock, flags);
809 list_for_each_entry(handler, &event->device->event_handler_list, list)
810 handler->handler(handler, event);
812 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
814 EXPORT_SYMBOL(ib_dispatch_event);
817 * ib_query_port - Query IB port attributes
818 * @device:Device to query
819 * @port_num:Port number to query
820 * @port_attr:Port attributes
822 * ib_query_port() returns the attributes of a port through the
823 * @port_attr pointer.
825 int ib_query_port(struct ib_device *device,
827 struct ib_port_attr *port_attr)
832 if (!rdma_is_port_valid(device, port_num))
835 memset(port_attr, 0, sizeof(*port_attr));
836 err = device->query_port(device, port_num, port_attr);
837 if (err || port_attr->subnet_prefix)
840 if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND)
843 err = ib_query_gid(device, port_num, 0, &gid, NULL);
847 port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix);
850 EXPORT_SYMBOL(ib_query_port);
853 * ib_query_gid - Get GID table entry
854 * @device:Device to query
855 * @port_num:Port number to query
856 * @index:GID table index to query
858 * @attr: Returned GID attributes related to this GID index (only in RoCE).
861 * ib_query_gid() fetches the specified GID table entry.
863 int ib_query_gid(struct ib_device *device,
864 u8 port_num, int index, union ib_gid *gid,
865 struct ib_gid_attr *attr)
867 if (rdma_cap_roce_gid_table(device, port_num))
868 return ib_get_cached_gid(device, port_num, index, gid, attr);
873 return device->query_gid(device, port_num, index, gid);
875 EXPORT_SYMBOL(ib_query_gid);
878 * ib_enum_roce_netdev - enumerate all RoCE ports
879 * @ib_dev : IB device we want to query
880 * @filter: Should we call the callback?
881 * @filter_cookie: Cookie passed to filter
882 * @cb: Callback to call for each found RoCE ports
883 * @cookie: Cookie passed back to the callback
885 * Enumerates all of the physical RoCE ports of ib_dev
886 * which are related to netdevice and calls callback() on each
887 * device for which filter() function returns non zero.
889 void ib_enum_roce_netdev(struct ib_device *ib_dev,
890 roce_netdev_filter filter,
892 roce_netdev_callback cb,
897 for (port = rdma_start_port(ib_dev); port <= rdma_end_port(ib_dev);
899 if (rdma_protocol_roce(ib_dev, port)) {
900 struct net_device *idev = NULL;
902 if (ib_dev->get_netdev)
903 idev = ib_dev->get_netdev(ib_dev, port);
906 idev->reg_state >= NETREG_UNREGISTERED) {
911 if (filter(ib_dev, port, idev, filter_cookie))
912 cb(ib_dev, port, idev, cookie);
920 * ib_enum_all_roce_netdevs - enumerate all RoCE devices
921 * @filter: Should we call the callback?
922 * @filter_cookie: Cookie passed to filter
923 * @cb: Callback to call for each found RoCE ports
924 * @cookie: Cookie passed back to the callback
926 * Enumerates all RoCE devices' physical ports which are related
927 * to netdevices and calls callback() on each device for which
928 * filter() function returns non zero.
930 void ib_enum_all_roce_netdevs(roce_netdev_filter filter,
932 roce_netdev_callback cb,
935 struct ib_device *dev;
937 down_read(&lists_rwsem);
938 list_for_each_entry(dev, &device_list, core_list)
939 ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie);
940 up_read(&lists_rwsem);
944 * ib_enum_all_devs - enumerate all ib_devices
945 * @cb: Callback to call for each found ib_device
947 * Enumerates all ib_devices and calls callback() on each device.
949 int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb,
950 struct netlink_callback *cb)
952 struct ib_device *dev;
953 unsigned int idx = 0;
956 down_read(&lists_rwsem);
957 list_for_each_entry(dev, &device_list, core_list) {
958 ret = nldev_cb(dev, skb, cb, idx);
964 up_read(&lists_rwsem);
969 * ib_query_pkey - Get P_Key table entry
970 * @device:Device to query
971 * @port_num:Port number to query
972 * @index:P_Key table index to query
973 * @pkey:Returned P_Key
975 * ib_query_pkey() fetches the specified P_Key table entry.
977 int ib_query_pkey(struct ib_device *device,
978 u8 port_num, u16 index, u16 *pkey)
980 return device->query_pkey(device, port_num, index, pkey);
982 EXPORT_SYMBOL(ib_query_pkey);
985 * ib_modify_device - Change IB device attributes
986 * @device:Device to modify
987 * @device_modify_mask:Mask of attributes to change
988 * @device_modify:New attribute values
990 * ib_modify_device() changes a device's attributes as specified by
991 * the @device_modify_mask and @device_modify structure.
993 int ib_modify_device(struct ib_device *device,
994 int device_modify_mask,
995 struct ib_device_modify *device_modify)
997 if (!device->modify_device)
1000 return device->modify_device(device, device_modify_mask,
1003 EXPORT_SYMBOL(ib_modify_device);
1006 * ib_modify_port - Modifies the attributes for the specified port.
1007 * @device: The device to modify.
1008 * @port_num: The number of the port to modify.
1009 * @port_modify_mask: Mask used to specify which attributes of the port
1011 * @port_modify: New attribute values for the port.
1013 * ib_modify_port() changes a port's attributes as specified by the
1014 * @port_modify_mask and @port_modify structure.
1016 int ib_modify_port(struct ib_device *device,
1017 u8 port_num, int port_modify_mask,
1018 struct ib_port_modify *port_modify)
1022 if (!rdma_is_port_valid(device, port_num))
1025 if (device->modify_port)
1026 rc = device->modify_port(device, port_num, port_modify_mask,
1029 rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS;
1032 EXPORT_SYMBOL(ib_modify_port);
1035 * ib_find_gid - Returns the port number and GID table index where
1036 * a specified GID value occurs.
1037 * @device: The device to query.
1038 * @gid: The GID value to search for.
1039 * @gid_type: Type of GID.
1040 * @ndev: The ndev related to the GID to search for.
1041 * @port_num: The port number of the device where the GID value was found.
1042 * @index: The index into the GID table where the GID was found. This
1043 * parameter may be NULL.
1045 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1046 enum ib_gid_type gid_type, struct net_device *ndev,
1047 u8 *port_num, u16 *index)
1049 union ib_gid tmp_gid;
1052 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
1053 if (rdma_cap_roce_gid_table(device, port)) {
1054 if (!ib_find_cached_gid_by_port(device, gid, gid_type, port,
1061 if (gid_type != IB_GID_TYPE_IB)
1064 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
1065 ret = ib_query_gid(device, port, i, &tmp_gid, NULL);
1068 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
1079 EXPORT_SYMBOL(ib_find_gid);
1082 * ib_find_pkey - Returns the PKey table index where a specified
1083 * PKey value occurs.
1084 * @device: The device to query.
1085 * @port_num: The port number of the device to search for the PKey.
1086 * @pkey: The PKey value to search for.
1087 * @index: The index into the PKey table where the PKey was found.
1089 int ib_find_pkey(struct ib_device *device,
1090 u8 port_num, u16 pkey, u16 *index)
1094 int partial_ix = -1;
1096 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
1097 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
1100 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
1101 /* if there is full-member pkey take it.*/
1102 if (tmp_pkey & 0x8000) {
1111 /*no full-member, if exists take the limited*/
1112 if (partial_ix >= 0) {
1113 *index = partial_ix;
1118 EXPORT_SYMBOL(ib_find_pkey);
1121 * ib_get_net_dev_by_params() - Return the appropriate net_dev
1122 * for a received CM request
1123 * @dev: An RDMA device on which the request has been received.
1124 * @port: Port number on the RDMA device.
1125 * @pkey: The Pkey the request came on.
1126 * @gid: A GID that the net_dev uses to communicate.
1127 * @addr: Contains the IP address that the request specified as its
1130 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev,
1133 const union ib_gid *gid,
1134 const struct sockaddr *addr)
1136 struct net_device *net_dev = NULL;
1137 struct ib_client_data *context;
1139 if (!rdma_protocol_ib(dev, port))
1142 down_read(&lists_rwsem);
1144 list_for_each_entry(context, &dev->client_data_list, list) {
1145 struct ib_client *client = context->client;
1147 if (context->going_down)
1150 if (client->get_net_dev_by_params) {
1151 net_dev = client->get_net_dev_by_params(dev, port, pkey,
1159 up_read(&lists_rwsem);
1163 EXPORT_SYMBOL(ib_get_net_dev_by_params);
1165 static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = {
1166 [RDMA_NL_LS_OP_RESOLVE] = {
1167 .doit = ib_nl_handle_resolve_resp,
1168 .flags = RDMA_NL_ADMIN_PERM,
1170 [RDMA_NL_LS_OP_SET_TIMEOUT] = {
1171 .doit = ib_nl_handle_set_timeout,
1172 .flags = RDMA_NL_ADMIN_PERM,
1174 [RDMA_NL_LS_OP_IP_RESOLVE] = {
1175 .doit = ib_nl_handle_ip_res_resp,
1176 .flags = RDMA_NL_ADMIN_PERM,
1180 static int __init ib_core_init(void)
1184 ib_wq = alloc_workqueue("infiniband", 0, 0);
1188 ib_comp_wq = alloc_workqueue("ib-comp-wq",
1189 WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
1195 ret = class_register(&ib_class);
1197 pr_warn("Couldn't create InfiniBand device class\n");
1201 ret = rdma_nl_init();
1203 pr_warn("Couldn't init IB netlink interface: err %d\n", ret);
1209 pr_warn("Could't init IB address resolution\n");
1213 ret = ib_mad_init();
1215 pr_warn("Couldn't init IB MAD\n");
1221 pr_warn("Couldn't init SA\n");
1225 ret = register_lsm_notifier(&ibdev_lsm_nb);
1227 pr_warn("Couldn't register LSM notifier. ret %d\n", ret);
1232 rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table);
1246 class_unregister(&ib_class);
1248 destroy_workqueue(ib_comp_wq);
1250 destroy_workqueue(ib_wq);
1254 static void __exit ib_core_cleanup(void)
1258 rdma_nl_unregister(RDMA_NL_LS);
1259 unregister_lsm_notifier(&ibdev_lsm_nb);
1264 class_unregister(&ib_class);
1265 destroy_workqueue(ib_comp_wq);
1266 /* Make sure that any pending umem accounting work is done. */
1267 destroy_workqueue(ib_wq);
1270 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4);
1272 subsys_initcall(ib_core_init);
1273 module_exit(ib_core_cleanup);