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_comp_unbound_wq;
65 struct workqueue_struct *ib_wq;
66 EXPORT_SYMBOL_GPL(ib_wq);
68 /* The device_list and client_list contain devices and clients after their
69 * registration has completed, and the devices and clients are removed
70 * during unregistration. */
71 static LIST_HEAD(device_list);
72 static LIST_HEAD(client_list);
75 * device_mutex and lists_rwsem protect access to both device_list and
76 * client_list. device_mutex protects writer access by device and client
77 * registration / de-registration. lists_rwsem protects reader access to
78 * these lists. Iterators of these lists must lock it for read, while updates
79 * to the lists must be done with a write lock. A special case is when the
80 * device_mutex is locked. In this case locking the lists for read access is
81 * not necessary as the device_mutex implies it.
83 * lists_rwsem also protects access to the client data list.
85 static DEFINE_MUTEX(device_mutex);
86 static DECLARE_RWSEM(lists_rwsem);
88 static int ib_security_change(struct notifier_block *nb, unsigned long event,
90 static void ib_policy_change_task(struct work_struct *work);
91 static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task);
93 static struct notifier_block ibdev_lsm_nb = {
94 .notifier_call = ib_security_change,
97 static int ib_device_check_mandatory(struct ib_device *device)
99 #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
100 static const struct {
103 } mandatory_table[] = {
104 IB_MANDATORY_FUNC(query_device),
105 IB_MANDATORY_FUNC(query_port),
106 IB_MANDATORY_FUNC(query_pkey),
107 IB_MANDATORY_FUNC(query_gid),
108 IB_MANDATORY_FUNC(alloc_pd),
109 IB_MANDATORY_FUNC(dealloc_pd),
110 IB_MANDATORY_FUNC(create_ah),
111 IB_MANDATORY_FUNC(destroy_ah),
112 IB_MANDATORY_FUNC(create_qp),
113 IB_MANDATORY_FUNC(modify_qp),
114 IB_MANDATORY_FUNC(destroy_qp),
115 IB_MANDATORY_FUNC(post_send),
116 IB_MANDATORY_FUNC(post_recv),
117 IB_MANDATORY_FUNC(create_cq),
118 IB_MANDATORY_FUNC(destroy_cq),
119 IB_MANDATORY_FUNC(poll_cq),
120 IB_MANDATORY_FUNC(req_notify_cq),
121 IB_MANDATORY_FUNC(get_dma_mr),
122 IB_MANDATORY_FUNC(dereg_mr),
123 IB_MANDATORY_FUNC(get_port_immutable)
127 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
128 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
129 pr_warn("Device %s is missing mandatory function %s\n",
130 device->name, mandatory_table[i].name);
138 static struct ib_device *__ib_device_get_by_index(u32 index)
140 struct ib_device *device;
142 list_for_each_entry(device, &device_list, core_list)
143 if (device->index == index)
150 * Caller is responsible to return refrerence count by calling put_device()
152 struct ib_device *ib_device_get_by_index(u32 index)
154 struct ib_device *device;
156 down_read(&lists_rwsem);
157 device = __ib_device_get_by_index(index);
159 get_device(&device->dev);
161 up_read(&lists_rwsem);
165 static struct ib_device *__ib_device_get_by_name(const char *name)
167 struct ib_device *device;
169 list_for_each_entry(device, &device_list, core_list)
170 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
176 static int alloc_name(char *name)
178 unsigned long *inuse;
179 char buf[IB_DEVICE_NAME_MAX];
180 struct ib_device *device;
183 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
187 list_for_each_entry(device, &device_list, core_list) {
188 if (!sscanf(device->name, name, &i))
190 if (i < 0 || i >= PAGE_SIZE * 8)
192 snprintf(buf, sizeof buf, name, i);
193 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
197 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
198 free_page((unsigned long) inuse);
199 snprintf(buf, sizeof buf, name, i);
201 if (__ib_device_get_by_name(buf))
204 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
208 static void ib_device_release(struct device *device)
210 struct ib_device *dev = container_of(device, struct ib_device, dev);
212 WARN_ON(dev->reg_state == IB_DEV_REGISTERED);
213 if (dev->reg_state == IB_DEV_UNREGISTERED) {
215 * In IB_DEV_UNINITIALIZED state, cache or port table
216 * is not even created. Free cache and port table only when
217 * device reaches UNREGISTERED state.
219 ib_cache_release_one(dev);
220 kfree(dev->port_immutable);
225 static int ib_device_uevent(struct device *device,
226 struct kobj_uevent_env *env)
228 struct ib_device *dev = container_of(device, struct ib_device, dev);
230 if (add_uevent_var(env, "NAME=%s", dev->name))
234 * It would be nice to pass the node GUID with the event...
240 static struct class ib_class = {
241 .name = "infiniband",
242 .dev_release = ib_device_release,
243 .dev_uevent = ib_device_uevent,
247 * ib_alloc_device - allocate an IB device struct
248 * @size:size of structure to allocate
250 * Low-level drivers should use ib_alloc_device() to allocate &struct
251 * ib_device. @size is the size of the structure to be allocated,
252 * including any private data used by the low-level driver.
253 * ib_dealloc_device() must be used to free structures allocated with
256 struct ib_device *ib_alloc_device(size_t size)
258 struct ib_device *device;
260 if (WARN_ON(size < sizeof(struct ib_device)))
263 device = kzalloc(size, GFP_KERNEL);
267 device->dev.class = &ib_class;
268 device_initialize(&device->dev);
270 dev_set_drvdata(&device->dev, device);
272 INIT_LIST_HEAD(&device->event_handler_list);
273 spin_lock_init(&device->event_handler_lock);
274 spin_lock_init(&device->client_data_lock);
275 INIT_LIST_HEAD(&device->client_data_list);
276 INIT_LIST_HEAD(&device->port_list);
280 EXPORT_SYMBOL(ib_alloc_device);
283 * ib_dealloc_device - free an IB device struct
284 * @device:structure to free
286 * Free a structure allocated with ib_alloc_device().
288 void ib_dealloc_device(struct ib_device *device)
290 WARN_ON(device->reg_state != IB_DEV_UNREGISTERED &&
291 device->reg_state != IB_DEV_UNINITIALIZED);
292 kobject_put(&device->dev.kobj);
294 EXPORT_SYMBOL(ib_dealloc_device);
296 static int add_client_context(struct ib_device *device, struct ib_client *client)
298 struct ib_client_data *context;
301 context = kmalloc(sizeof *context, GFP_KERNEL);
305 context->client = client;
306 context->data = NULL;
307 context->going_down = false;
309 down_write(&lists_rwsem);
310 spin_lock_irqsave(&device->client_data_lock, flags);
311 list_add(&context->list, &device->client_data_list);
312 spin_unlock_irqrestore(&device->client_data_lock, flags);
313 up_write(&lists_rwsem);
318 static int verify_immutable(const struct ib_device *dev, u8 port)
320 return WARN_ON(!rdma_cap_ib_mad(dev, port) &&
321 rdma_max_mad_size(dev, port) != 0);
324 static int read_port_immutable(struct ib_device *device)
327 u8 start_port = rdma_start_port(device);
328 u8 end_port = rdma_end_port(device);
332 * device->port_immutable is indexed directly by the port number to make
333 * access to this data as efficient as possible.
335 * Therefore port_immutable is declared as a 1 based array with
336 * potential empty slots at the beginning.
338 device->port_immutable = kzalloc(sizeof(*device->port_immutable)
341 if (!device->port_immutable)
344 for (port = start_port; port <= end_port; ++port) {
345 ret = device->get_port_immutable(device, port,
346 &device->port_immutable[port]);
350 if (verify_immutable(device, port))
356 void ib_get_device_fw_str(struct ib_device *dev, char *str)
358 if (dev->get_dev_fw_str)
359 dev->get_dev_fw_str(dev, str);
363 EXPORT_SYMBOL(ib_get_device_fw_str);
365 static int setup_port_pkey_list(struct ib_device *device)
370 * device->port_pkey_list is indexed directly by the port number,
371 * Therefore it is declared as a 1 based array with potential empty
372 * slots at the beginning.
374 device->port_pkey_list = kcalloc(rdma_end_port(device) + 1,
375 sizeof(*device->port_pkey_list),
378 if (!device->port_pkey_list)
381 for (i = 0; i < (rdma_end_port(device) + 1); i++) {
382 spin_lock_init(&device->port_pkey_list[i].list_lock);
383 INIT_LIST_HEAD(&device->port_pkey_list[i].pkey_list);
389 static void ib_policy_change_task(struct work_struct *work)
391 struct ib_device *dev;
393 down_read(&lists_rwsem);
394 list_for_each_entry(dev, &device_list, core_list) {
397 for (i = rdma_start_port(dev); i <= rdma_end_port(dev); i++) {
399 int ret = ib_get_cached_subnet_prefix(dev,
404 "ib_get_cached_subnet_prefix err: %d, this should never happen here\n",
407 ib_security_cache_change(dev, i, sp);
410 up_read(&lists_rwsem);
413 static int ib_security_change(struct notifier_block *nb, unsigned long event,
416 if (event != LSM_POLICY_CHANGE)
419 schedule_work(&ib_policy_change_work);
425 * __dev_new_index - allocate an device index
427 * Returns a suitable unique value for a new device interface
428 * number. It assumes that there are less than 2^32-1 ib devices
429 * will be present in the system.
431 static u32 __dev_new_index(void)
434 * The device index to allow stable naming.
435 * Similar to struct net -> ifindex.
443 if (!__ib_device_get_by_index(index))
449 * ib_register_device - Register an IB device with IB core
450 * @device:Device to register
452 * Low-level drivers use ib_register_device() to register their
453 * devices with the IB core. All registered clients will receive a
454 * callback for each device that is added. @device must be allocated
455 * with ib_alloc_device().
457 int ib_register_device(struct ib_device *device,
458 int (*port_callback)(struct ib_device *,
459 u8, struct kobject *))
462 struct ib_client *client;
463 struct ib_udata uhw = {.outlen = 0, .inlen = 0};
464 struct device *parent = device->dev.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) {
476 device->dev.dma_mask = parent->dma_mask;
480 if (!device->dev.coherent_dma_mask) {
482 device->dev.coherent_dma_mask =
483 parent->coherent_dma_mask;
489 * The caller did not provide custom DMA operations. Use the
490 * DMA mapping operations of the parent device.
492 WARN_ON_ONCE(!parent);
493 device->dma_device = parent;
496 mutex_lock(&device_mutex);
498 if (strchr(device->name, '%')) {
499 ret = alloc_name(device->name);
504 if (ib_device_check_mandatory(device)) {
509 ret = read_port_immutable(device);
511 pr_warn("Couldn't create per port immutable data %s\n",
516 ret = setup_port_pkey_list(device);
518 pr_warn("Couldn't create per port_pkey_list\n");
522 ret = ib_cache_setup_one(device);
524 pr_warn("Couldn't set up InfiniBand P_Key/GID cache\n");
528 ret = ib_device_register_rdmacg(device);
530 pr_warn("Couldn't register device with rdma cgroup\n");
534 memset(&device->attrs, 0, sizeof(device->attrs));
535 ret = device->query_device(device, &device->attrs, &uhw);
537 pr_warn("Couldn't query the device attributes\n");
541 ret = ib_device_register_sysfs(device, port_callback);
543 pr_warn("Couldn't register device %s with driver model\n",
548 device->reg_state = IB_DEV_REGISTERED;
550 list_for_each_entry(client, &client_list, list)
551 if (!add_client_context(device, client) && client->add)
554 device->index = __dev_new_index();
555 down_write(&lists_rwsem);
556 list_add_tail(&device->core_list, &device_list);
557 up_write(&lists_rwsem);
558 mutex_unlock(&device_mutex);
562 ib_device_unregister_rdmacg(device);
564 ib_cache_cleanup_one(device);
565 ib_cache_release_one(device);
567 kfree(device->port_immutable);
569 mutex_unlock(&device_mutex);
572 EXPORT_SYMBOL(ib_register_device);
575 * ib_unregister_device - Unregister an IB device
576 * @device:Device to unregister
578 * Unregister an IB device. All clients will receive a remove callback.
580 void ib_unregister_device(struct ib_device *device)
582 struct ib_client_data *context, *tmp;
585 mutex_lock(&device_mutex);
587 down_write(&lists_rwsem);
588 list_del(&device->core_list);
589 spin_lock_irqsave(&device->client_data_lock, flags);
590 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
591 context->going_down = true;
592 spin_unlock_irqrestore(&device->client_data_lock, flags);
593 downgrade_write(&lists_rwsem);
595 list_for_each_entry_safe(context, tmp, &device->client_data_list,
597 if (context->client->remove)
598 context->client->remove(device, context->data);
600 up_read(&lists_rwsem);
602 ib_device_unregister_sysfs(device);
603 ib_device_unregister_rdmacg(device);
605 mutex_unlock(&device_mutex);
607 ib_cache_cleanup_one(device);
609 ib_security_destroy_port_pkey_list(device);
610 kfree(device->port_pkey_list);
612 down_write(&lists_rwsem);
613 spin_lock_irqsave(&device->client_data_lock, flags);
614 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
616 spin_unlock_irqrestore(&device->client_data_lock, flags);
617 up_write(&lists_rwsem);
619 device->reg_state = IB_DEV_UNREGISTERED;
621 EXPORT_SYMBOL(ib_unregister_device);
624 * ib_register_client - Register an IB client
625 * @client:Client to register
627 * Upper level users of the IB drivers can use ib_register_client() to
628 * register callbacks for IB device addition and removal. When an IB
629 * device is added, each registered client's add method will be called
630 * (in the order the clients were registered), and when a device is
631 * removed, each client's remove method will be called (in the reverse
632 * order that clients were registered). In addition, when
633 * ib_register_client() is called, the client will receive an add
634 * callback for all devices already registered.
636 int ib_register_client(struct ib_client *client)
638 struct ib_device *device;
640 mutex_lock(&device_mutex);
642 list_for_each_entry(device, &device_list, core_list)
643 if (!add_client_context(device, client) && client->add)
646 down_write(&lists_rwsem);
647 list_add_tail(&client->list, &client_list);
648 up_write(&lists_rwsem);
650 mutex_unlock(&device_mutex);
654 EXPORT_SYMBOL(ib_register_client);
657 * ib_unregister_client - Unregister an IB client
658 * @client:Client to unregister
660 * Upper level users use ib_unregister_client() to remove their client
661 * registration. When ib_unregister_client() is called, the client
662 * will receive a remove callback for each IB device still registered.
664 void ib_unregister_client(struct ib_client *client)
666 struct ib_client_data *context, *tmp;
667 struct ib_device *device;
670 mutex_lock(&device_mutex);
672 down_write(&lists_rwsem);
673 list_del(&client->list);
674 up_write(&lists_rwsem);
676 list_for_each_entry(device, &device_list, core_list) {
677 struct ib_client_data *found_context = NULL;
679 down_write(&lists_rwsem);
680 spin_lock_irqsave(&device->client_data_lock, flags);
681 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
682 if (context->client == client) {
683 context->going_down = true;
684 found_context = context;
687 spin_unlock_irqrestore(&device->client_data_lock, flags);
688 up_write(&lists_rwsem);
691 client->remove(device, found_context ?
692 found_context->data : NULL);
694 if (!found_context) {
695 pr_warn("No client context found for %s/%s\n",
696 device->name, client->name);
700 down_write(&lists_rwsem);
701 spin_lock_irqsave(&device->client_data_lock, flags);
702 list_del(&found_context->list);
703 kfree(found_context);
704 spin_unlock_irqrestore(&device->client_data_lock, flags);
705 up_write(&lists_rwsem);
708 mutex_unlock(&device_mutex);
710 EXPORT_SYMBOL(ib_unregister_client);
713 * ib_get_client_data - Get IB client context
714 * @device:Device to get context for
715 * @client:Client to get context for
717 * ib_get_client_data() returns client context set with
718 * ib_set_client_data().
720 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
722 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) {
732 spin_unlock_irqrestore(&device->client_data_lock, flags);
736 EXPORT_SYMBOL(ib_get_client_data);
739 * ib_set_client_data - Set IB client context
740 * @device:Device to set context for
741 * @client:Client to set context for
742 * @data:Context to set
744 * ib_set_client_data() sets client context that can be retrieved with
745 * ib_get_client_data().
747 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
750 struct ib_client_data *context;
753 spin_lock_irqsave(&device->client_data_lock, flags);
754 list_for_each_entry(context, &device->client_data_list, list)
755 if (context->client == client) {
756 context->data = data;
760 pr_warn("No client context found for %s/%s\n",
761 device->name, client->name);
764 spin_unlock_irqrestore(&device->client_data_lock, flags);
766 EXPORT_SYMBOL(ib_set_client_data);
769 * ib_register_event_handler - Register an IB event handler
770 * @event_handler:Handler to register
772 * ib_register_event_handler() registers an event handler that will be
773 * called back when asynchronous IB events occur (as defined in
774 * chapter 11 of the InfiniBand Architecture Specification). This
775 * callback may occur in interrupt context.
777 void ib_register_event_handler(struct ib_event_handler *event_handler)
781 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
782 list_add_tail(&event_handler->list,
783 &event_handler->device->event_handler_list);
784 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
786 EXPORT_SYMBOL(ib_register_event_handler);
789 * ib_unregister_event_handler - Unregister an event handler
790 * @event_handler:Handler to unregister
792 * Unregister an event handler registered with
793 * ib_register_event_handler().
795 void ib_unregister_event_handler(struct ib_event_handler *event_handler)
799 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
800 list_del(&event_handler->list);
801 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
803 EXPORT_SYMBOL(ib_unregister_event_handler);
806 * ib_dispatch_event - Dispatch an asynchronous event
807 * @event:Event to dispatch
809 * Low-level drivers must call ib_dispatch_event() to dispatch the
810 * event to all registered event handlers when an asynchronous event
813 void ib_dispatch_event(struct ib_event *event)
816 struct ib_event_handler *handler;
818 spin_lock_irqsave(&event->device->event_handler_lock, flags);
820 list_for_each_entry(handler, &event->device->event_handler_list, list)
821 handler->handler(handler, event);
823 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
825 EXPORT_SYMBOL(ib_dispatch_event);
828 * ib_query_port - Query IB port attributes
829 * @device:Device to query
830 * @port_num:Port number to query
831 * @port_attr:Port attributes
833 * ib_query_port() returns the attributes of a port through the
834 * @port_attr pointer.
836 int ib_query_port(struct ib_device *device,
838 struct ib_port_attr *port_attr)
843 if (!rdma_is_port_valid(device, port_num))
846 memset(port_attr, 0, sizeof(*port_attr));
847 err = device->query_port(device, port_num, port_attr);
848 if (err || port_attr->subnet_prefix)
851 if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND)
854 err = ib_query_gid(device, port_num, 0, &gid, NULL);
858 port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix);
861 EXPORT_SYMBOL(ib_query_port);
864 * ib_query_gid - Get GID table entry
865 * @device:Device to query
866 * @port_num:Port number to query
867 * @index:GID table index to query
869 * @attr: Returned GID attributes related to this GID index (only in RoCE).
872 * ib_query_gid() fetches the specified GID table entry.
874 int ib_query_gid(struct ib_device *device,
875 u8 port_num, int index, union ib_gid *gid,
876 struct ib_gid_attr *attr)
878 if (rdma_cap_roce_gid_table(device, port_num))
879 return ib_get_cached_gid(device, port_num, index, gid, attr);
884 return device->query_gid(device, port_num, index, gid);
886 EXPORT_SYMBOL(ib_query_gid);
889 * ib_enum_roce_netdev - enumerate all RoCE ports
890 * @ib_dev : IB device we want to query
891 * @filter: Should we call the callback?
892 * @filter_cookie: Cookie passed to filter
893 * @cb: Callback to call for each found RoCE ports
894 * @cookie: Cookie passed back to the callback
896 * Enumerates all of the physical RoCE ports of ib_dev
897 * which are related to netdevice and calls callback() on each
898 * device for which filter() function returns non zero.
900 void ib_enum_roce_netdev(struct ib_device *ib_dev,
901 roce_netdev_filter filter,
903 roce_netdev_callback cb,
908 for (port = rdma_start_port(ib_dev); port <= rdma_end_port(ib_dev);
910 if (rdma_protocol_roce(ib_dev, port)) {
911 struct net_device *idev = NULL;
913 if (ib_dev->get_netdev)
914 idev = ib_dev->get_netdev(ib_dev, port);
917 idev->reg_state >= NETREG_UNREGISTERED) {
922 if (filter(ib_dev, port, idev, filter_cookie))
923 cb(ib_dev, port, idev, cookie);
931 * ib_enum_all_roce_netdevs - enumerate all RoCE devices
932 * @filter: Should we call the callback?
933 * @filter_cookie: Cookie passed to filter
934 * @cb: Callback to call for each found RoCE ports
935 * @cookie: Cookie passed back to the callback
937 * Enumerates all RoCE devices' physical ports which are related
938 * to netdevices and calls callback() on each device for which
939 * filter() function returns non zero.
941 void ib_enum_all_roce_netdevs(roce_netdev_filter filter,
943 roce_netdev_callback cb,
946 struct ib_device *dev;
948 down_read(&lists_rwsem);
949 list_for_each_entry(dev, &device_list, core_list)
950 ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie);
951 up_read(&lists_rwsem);
955 * ib_enum_all_devs - enumerate all ib_devices
956 * @cb: Callback to call for each found ib_device
958 * Enumerates all ib_devices and calls callback() on each device.
960 int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb,
961 struct netlink_callback *cb)
963 struct ib_device *dev;
964 unsigned int idx = 0;
967 down_read(&lists_rwsem);
968 list_for_each_entry(dev, &device_list, core_list) {
969 ret = nldev_cb(dev, skb, cb, idx);
975 up_read(&lists_rwsem);
980 * ib_query_pkey - Get P_Key table entry
981 * @device:Device to query
982 * @port_num:Port number to query
983 * @index:P_Key table index to query
984 * @pkey:Returned P_Key
986 * ib_query_pkey() fetches the specified P_Key table entry.
988 int ib_query_pkey(struct ib_device *device,
989 u8 port_num, u16 index, u16 *pkey)
991 return device->query_pkey(device, port_num, index, pkey);
993 EXPORT_SYMBOL(ib_query_pkey);
996 * ib_modify_device - Change IB device attributes
997 * @device:Device to modify
998 * @device_modify_mask:Mask of attributes to change
999 * @device_modify:New attribute values
1001 * ib_modify_device() changes a device's attributes as specified by
1002 * the @device_modify_mask and @device_modify structure.
1004 int ib_modify_device(struct ib_device *device,
1005 int device_modify_mask,
1006 struct ib_device_modify *device_modify)
1008 if (!device->modify_device)
1011 return device->modify_device(device, device_modify_mask,
1014 EXPORT_SYMBOL(ib_modify_device);
1017 * ib_modify_port - Modifies the attributes for the specified port.
1018 * @device: The device to modify.
1019 * @port_num: The number of the port to modify.
1020 * @port_modify_mask: Mask used to specify which attributes of the port
1022 * @port_modify: New attribute values for the port.
1024 * ib_modify_port() changes a port's attributes as specified by the
1025 * @port_modify_mask and @port_modify structure.
1027 int ib_modify_port(struct ib_device *device,
1028 u8 port_num, int port_modify_mask,
1029 struct ib_port_modify *port_modify)
1033 if (!rdma_is_port_valid(device, port_num))
1036 if (device->modify_port)
1037 rc = device->modify_port(device, port_num, port_modify_mask,
1040 rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS;
1043 EXPORT_SYMBOL(ib_modify_port);
1046 * ib_find_gid - Returns the port number and GID table index where
1047 * a specified GID value occurs.
1048 * @device: The device to query.
1049 * @gid: The GID value to search for.
1050 * @gid_type: Type of GID.
1051 * @ndev: The ndev related to the GID to search for.
1052 * @port_num: The port number of the device where the GID value was found.
1053 * @index: The index into the GID table where the GID was found. This
1054 * parameter may be NULL.
1056 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1057 enum ib_gid_type gid_type, struct net_device *ndev,
1058 u8 *port_num, u16 *index)
1060 union ib_gid tmp_gid;
1063 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
1064 if (rdma_cap_roce_gid_table(device, port)) {
1065 if (!ib_find_cached_gid_by_port(device, gid, gid_type, port,
1072 if (gid_type != IB_GID_TYPE_IB)
1075 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
1076 ret = ib_query_gid(device, port, i, &tmp_gid, NULL);
1079 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
1090 EXPORT_SYMBOL(ib_find_gid);
1093 * ib_find_pkey - Returns the PKey table index where a specified
1094 * PKey value occurs.
1095 * @device: The device to query.
1096 * @port_num: The port number of the device to search for the PKey.
1097 * @pkey: The PKey value to search for.
1098 * @index: The index into the PKey table where the PKey was found.
1100 int ib_find_pkey(struct ib_device *device,
1101 u8 port_num, u16 pkey, u16 *index)
1105 int partial_ix = -1;
1107 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
1108 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
1111 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
1112 /* if there is full-member pkey take it.*/
1113 if (tmp_pkey & 0x8000) {
1122 /*no full-member, if exists take the limited*/
1123 if (partial_ix >= 0) {
1124 *index = partial_ix;
1129 EXPORT_SYMBOL(ib_find_pkey);
1132 * ib_get_net_dev_by_params() - Return the appropriate net_dev
1133 * for a received CM request
1134 * @dev: An RDMA device on which the request has been received.
1135 * @port: Port number on the RDMA device.
1136 * @pkey: The Pkey the request came on.
1137 * @gid: A GID that the net_dev uses to communicate.
1138 * @addr: Contains the IP address that the request specified as its
1141 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev,
1144 const union ib_gid *gid,
1145 const struct sockaddr *addr)
1147 struct net_device *net_dev = NULL;
1148 struct ib_client_data *context;
1150 if (!rdma_protocol_ib(dev, port))
1153 down_read(&lists_rwsem);
1155 list_for_each_entry(context, &dev->client_data_list, list) {
1156 struct ib_client *client = context->client;
1158 if (context->going_down)
1161 if (client->get_net_dev_by_params) {
1162 net_dev = client->get_net_dev_by_params(dev, port, pkey,
1170 up_read(&lists_rwsem);
1174 EXPORT_SYMBOL(ib_get_net_dev_by_params);
1176 static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = {
1177 [RDMA_NL_LS_OP_RESOLVE] = {
1178 .doit = ib_nl_handle_resolve_resp,
1179 .flags = RDMA_NL_ADMIN_PERM,
1181 [RDMA_NL_LS_OP_SET_TIMEOUT] = {
1182 .doit = ib_nl_handle_set_timeout,
1183 .flags = RDMA_NL_ADMIN_PERM,
1185 [RDMA_NL_LS_OP_IP_RESOLVE] = {
1186 .doit = ib_nl_handle_ip_res_resp,
1187 .flags = RDMA_NL_ADMIN_PERM,
1191 static int __init ib_core_init(void)
1195 ib_wq = alloc_workqueue("infiniband", 0, 0);
1199 ib_comp_wq = alloc_workqueue("ib-comp-wq",
1200 WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
1206 ib_comp_unbound_wq =
1207 alloc_workqueue("ib-comp-unb-wq",
1208 WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM |
1209 WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE);
1210 if (!ib_comp_unbound_wq) {
1215 ret = class_register(&ib_class);
1217 pr_warn("Couldn't create InfiniBand device class\n");
1218 goto err_comp_unbound;
1221 ret = rdma_nl_init();
1223 pr_warn("Couldn't init IB netlink interface: err %d\n", ret);
1229 pr_warn("Could't init IB address resolution\n");
1233 ret = ib_mad_init();
1235 pr_warn("Couldn't init IB MAD\n");
1241 pr_warn("Couldn't init SA\n");
1245 ret = register_lsm_notifier(&ibdev_lsm_nb);
1247 pr_warn("Couldn't register LSM notifier. ret %d\n", ret);
1252 rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table);
1266 class_unregister(&ib_class);
1268 destroy_workqueue(ib_comp_unbound_wq);
1270 destroy_workqueue(ib_comp_wq);
1272 destroy_workqueue(ib_wq);
1276 static void __exit ib_core_cleanup(void)
1280 rdma_nl_unregister(RDMA_NL_LS);
1281 unregister_lsm_notifier(&ibdev_lsm_nb);
1286 class_unregister(&ib_class);
1287 destroy_workqueue(ib_comp_unbound_wq);
1288 destroy_workqueue(ib_comp_wq);
1289 /* Make sure that any pending umem accounting work is done. */
1290 destroy_workqueue(ib_wq);
1293 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4);
1295 subsys_initcall(ib_core_init);
1296 module_exit(ib_core_cleanup);