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(alloc_pd),
107 IB_MANDATORY_FUNC(dealloc_pd),
108 IB_MANDATORY_FUNC(create_ah),
109 IB_MANDATORY_FUNC(destroy_ah),
110 IB_MANDATORY_FUNC(create_qp),
111 IB_MANDATORY_FUNC(modify_qp),
112 IB_MANDATORY_FUNC(destroy_qp),
113 IB_MANDATORY_FUNC(post_send),
114 IB_MANDATORY_FUNC(post_recv),
115 IB_MANDATORY_FUNC(create_cq),
116 IB_MANDATORY_FUNC(destroy_cq),
117 IB_MANDATORY_FUNC(poll_cq),
118 IB_MANDATORY_FUNC(req_notify_cq),
119 IB_MANDATORY_FUNC(get_dma_mr),
120 IB_MANDATORY_FUNC(dereg_mr),
121 IB_MANDATORY_FUNC(get_port_immutable)
125 for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
126 if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
127 pr_warn("Device %s is missing mandatory function %s\n",
128 device->name, mandatory_table[i].name);
136 static struct ib_device *__ib_device_get_by_index(u32 index)
138 struct ib_device *device;
140 list_for_each_entry(device, &device_list, core_list)
141 if (device->index == index)
148 * Caller is responsible to return refrerence count by calling put_device()
150 struct ib_device *ib_device_get_by_index(u32 index)
152 struct ib_device *device;
154 down_read(&lists_rwsem);
155 device = __ib_device_get_by_index(index);
157 get_device(&device->dev);
159 up_read(&lists_rwsem);
163 static struct ib_device *__ib_device_get_by_name(const char *name)
165 struct ib_device *device;
167 list_for_each_entry(device, &device_list, core_list)
168 if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
174 static int alloc_name(char *name)
176 unsigned long *inuse;
177 char buf[IB_DEVICE_NAME_MAX];
178 struct ib_device *device;
181 inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
185 list_for_each_entry(device, &device_list, core_list) {
186 if (!sscanf(device->name, name, &i))
188 if (i < 0 || i >= PAGE_SIZE * 8)
190 snprintf(buf, sizeof buf, name, i);
191 if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
195 i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
196 free_page((unsigned long) inuse);
197 snprintf(buf, sizeof buf, name, i);
199 if (__ib_device_get_by_name(buf))
202 strlcpy(name, buf, IB_DEVICE_NAME_MAX);
206 static void ib_device_release(struct device *device)
208 struct ib_device *dev = container_of(device, struct ib_device, dev);
210 WARN_ON(dev->reg_state == IB_DEV_REGISTERED);
211 if (dev->reg_state == IB_DEV_UNREGISTERED) {
213 * In IB_DEV_UNINITIALIZED state, cache or port table
214 * is not even created. Free cache and port table only when
215 * device reaches UNREGISTERED state.
217 ib_cache_release_one(dev);
218 kfree(dev->port_immutable);
223 static int ib_device_uevent(struct device *device,
224 struct kobj_uevent_env *env)
226 struct ib_device *dev = container_of(device, struct ib_device, dev);
228 if (add_uevent_var(env, "NAME=%s", dev->name))
232 * It would be nice to pass the node GUID with the event...
238 static struct class ib_class = {
239 .name = "infiniband",
240 .dev_release = ib_device_release,
241 .dev_uevent = ib_device_uevent,
245 * ib_alloc_device - allocate an IB device struct
246 * @size:size of structure to allocate
248 * Low-level drivers should use ib_alloc_device() to allocate &struct
249 * ib_device. @size is the size of the structure to be allocated,
250 * including any private data used by the low-level driver.
251 * ib_dealloc_device() must be used to free structures allocated with
254 struct ib_device *ib_alloc_device(size_t size)
256 struct ib_device *device;
258 if (WARN_ON(size < sizeof(struct ib_device)))
261 device = kzalloc(size, GFP_KERNEL);
265 rdma_restrack_init(&device->res);
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 put_device(&device->dev);
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 rdma_restrack_clean(&device->res);
604 ib_device_unregister_rdmacg(device);
605 ib_device_unregister_sysfs(device);
607 mutex_unlock(&device_mutex);
609 ib_cache_cleanup_one(device);
611 ib_security_destroy_port_pkey_list(device);
612 kfree(device->port_pkey_list);
614 down_write(&lists_rwsem);
615 spin_lock_irqsave(&device->client_data_lock, flags);
616 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
618 spin_unlock_irqrestore(&device->client_data_lock, flags);
619 up_write(&lists_rwsem);
621 device->reg_state = IB_DEV_UNREGISTERED;
623 EXPORT_SYMBOL(ib_unregister_device);
626 * ib_register_client - Register an IB client
627 * @client:Client to register
629 * Upper level users of the IB drivers can use ib_register_client() to
630 * register callbacks for IB device addition and removal. When an IB
631 * device is added, each registered client's add method will be called
632 * (in the order the clients were registered), and when a device is
633 * removed, each client's remove method will be called (in the reverse
634 * order that clients were registered). In addition, when
635 * ib_register_client() is called, the client will receive an add
636 * callback for all devices already registered.
638 int ib_register_client(struct ib_client *client)
640 struct ib_device *device;
642 mutex_lock(&device_mutex);
644 list_for_each_entry(device, &device_list, core_list)
645 if (!add_client_context(device, client) && client->add)
648 down_write(&lists_rwsem);
649 list_add_tail(&client->list, &client_list);
650 up_write(&lists_rwsem);
652 mutex_unlock(&device_mutex);
656 EXPORT_SYMBOL(ib_register_client);
659 * ib_unregister_client - Unregister an IB client
660 * @client:Client to unregister
662 * Upper level users use ib_unregister_client() to remove their client
663 * registration. When ib_unregister_client() is called, the client
664 * will receive a remove callback for each IB device still registered.
666 void ib_unregister_client(struct ib_client *client)
668 struct ib_client_data *context, *tmp;
669 struct ib_device *device;
672 mutex_lock(&device_mutex);
674 down_write(&lists_rwsem);
675 list_del(&client->list);
676 up_write(&lists_rwsem);
678 list_for_each_entry(device, &device_list, core_list) {
679 struct ib_client_data *found_context = NULL;
681 down_write(&lists_rwsem);
682 spin_lock_irqsave(&device->client_data_lock, flags);
683 list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
684 if (context->client == client) {
685 context->going_down = true;
686 found_context = context;
689 spin_unlock_irqrestore(&device->client_data_lock, flags);
690 up_write(&lists_rwsem);
693 client->remove(device, found_context ?
694 found_context->data : NULL);
696 if (!found_context) {
697 pr_warn("No client context found for %s/%s\n",
698 device->name, client->name);
702 down_write(&lists_rwsem);
703 spin_lock_irqsave(&device->client_data_lock, flags);
704 list_del(&found_context->list);
705 kfree(found_context);
706 spin_unlock_irqrestore(&device->client_data_lock, flags);
707 up_write(&lists_rwsem);
710 mutex_unlock(&device_mutex);
712 EXPORT_SYMBOL(ib_unregister_client);
715 * ib_get_client_data - Get IB client context
716 * @device:Device to get context for
717 * @client:Client to get context for
719 * ib_get_client_data() returns client context set with
720 * ib_set_client_data().
722 void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
724 struct ib_client_data *context;
728 spin_lock_irqsave(&device->client_data_lock, flags);
729 list_for_each_entry(context, &device->client_data_list, list)
730 if (context->client == client) {
734 spin_unlock_irqrestore(&device->client_data_lock, flags);
738 EXPORT_SYMBOL(ib_get_client_data);
741 * ib_set_client_data - Set IB client context
742 * @device:Device to set context for
743 * @client:Client to set context for
744 * @data:Context to set
746 * ib_set_client_data() sets client context that can be retrieved with
747 * ib_get_client_data().
749 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
752 struct ib_client_data *context;
755 spin_lock_irqsave(&device->client_data_lock, flags);
756 list_for_each_entry(context, &device->client_data_list, list)
757 if (context->client == client) {
758 context->data = data;
762 pr_warn("No client context found for %s/%s\n",
763 device->name, client->name);
766 spin_unlock_irqrestore(&device->client_data_lock, flags);
768 EXPORT_SYMBOL(ib_set_client_data);
771 * ib_register_event_handler - Register an IB event handler
772 * @event_handler:Handler to register
774 * ib_register_event_handler() registers an event handler that will be
775 * called back when asynchronous IB events occur (as defined in
776 * chapter 11 of the InfiniBand Architecture Specification). This
777 * callback may occur in interrupt context.
779 void ib_register_event_handler(struct ib_event_handler *event_handler)
783 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
784 list_add_tail(&event_handler->list,
785 &event_handler->device->event_handler_list);
786 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
788 EXPORT_SYMBOL(ib_register_event_handler);
791 * ib_unregister_event_handler - Unregister an event handler
792 * @event_handler:Handler to unregister
794 * Unregister an event handler registered with
795 * ib_register_event_handler().
797 void ib_unregister_event_handler(struct ib_event_handler *event_handler)
801 spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
802 list_del(&event_handler->list);
803 spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
805 EXPORT_SYMBOL(ib_unregister_event_handler);
808 * ib_dispatch_event - Dispatch an asynchronous event
809 * @event:Event to dispatch
811 * Low-level drivers must call ib_dispatch_event() to dispatch the
812 * event to all registered event handlers when an asynchronous event
815 void ib_dispatch_event(struct ib_event *event)
818 struct ib_event_handler *handler;
820 spin_lock_irqsave(&event->device->event_handler_lock, flags);
822 list_for_each_entry(handler, &event->device->event_handler_list, list)
823 handler->handler(handler, event);
825 spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
827 EXPORT_SYMBOL(ib_dispatch_event);
830 * ib_query_port - Query IB port attributes
831 * @device:Device to query
832 * @port_num:Port number to query
833 * @port_attr:Port attributes
835 * ib_query_port() returns the attributes of a port through the
836 * @port_attr pointer.
838 int ib_query_port(struct ib_device *device,
840 struct ib_port_attr *port_attr)
845 if (!rdma_is_port_valid(device, port_num))
848 memset(port_attr, 0, sizeof(*port_attr));
849 err = device->query_port(device, port_num, port_attr);
850 if (err || port_attr->subnet_prefix)
853 if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND)
856 err = device->query_gid(device, port_num, 0, &gid);
860 port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix);
863 EXPORT_SYMBOL(ib_query_port);
866 * ib_query_gid - Get GID table entry
867 * @device:Device to query
868 * @port_num:Port number to query
869 * @index:GID table index to query
871 * @attr: Returned GID attributes related to this GID index (only in RoCE).
874 * ib_query_gid() fetches the specified GID table entry from the cache.
876 int ib_query_gid(struct ib_device *device,
877 u8 port_num, int index, union ib_gid *gid,
878 struct ib_gid_attr *attr)
880 return ib_get_cached_gid(device, port_num, index, gid, attr);
882 EXPORT_SYMBOL(ib_query_gid);
885 * ib_enum_roce_netdev - enumerate all RoCE ports
886 * @ib_dev : IB device we want to query
887 * @filter: Should we call the callback?
888 * @filter_cookie: Cookie passed to filter
889 * @cb: Callback to call for each found RoCE ports
890 * @cookie: Cookie passed back to the callback
892 * Enumerates all of the physical RoCE ports of ib_dev
893 * which are related to netdevice and calls callback() on each
894 * device for which filter() function returns non zero.
896 void ib_enum_roce_netdev(struct ib_device *ib_dev,
897 roce_netdev_filter filter,
899 roce_netdev_callback cb,
904 for (port = rdma_start_port(ib_dev); port <= rdma_end_port(ib_dev);
906 if (rdma_protocol_roce(ib_dev, port)) {
907 struct net_device *idev = NULL;
909 if (ib_dev->get_netdev)
910 idev = ib_dev->get_netdev(ib_dev, port);
913 idev->reg_state >= NETREG_UNREGISTERED) {
918 if (filter(ib_dev, port, idev, filter_cookie))
919 cb(ib_dev, port, idev, cookie);
927 * ib_enum_all_roce_netdevs - enumerate all RoCE devices
928 * @filter: Should we call the callback?
929 * @filter_cookie: Cookie passed to filter
930 * @cb: Callback to call for each found RoCE ports
931 * @cookie: Cookie passed back to the callback
933 * Enumerates all RoCE devices' physical ports which are related
934 * to netdevices and calls callback() on each device for which
935 * filter() function returns non zero.
937 void ib_enum_all_roce_netdevs(roce_netdev_filter filter,
939 roce_netdev_callback cb,
942 struct ib_device *dev;
944 down_read(&lists_rwsem);
945 list_for_each_entry(dev, &device_list, core_list)
946 ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie);
947 up_read(&lists_rwsem);
951 * ib_enum_all_devs - enumerate all ib_devices
952 * @cb: Callback to call for each found ib_device
954 * Enumerates all ib_devices and calls callback() on each device.
956 int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb,
957 struct netlink_callback *cb)
959 struct ib_device *dev;
960 unsigned int idx = 0;
963 down_read(&lists_rwsem);
964 list_for_each_entry(dev, &device_list, core_list) {
965 ret = nldev_cb(dev, skb, cb, idx);
971 up_read(&lists_rwsem);
976 * ib_query_pkey - Get P_Key table entry
977 * @device:Device to query
978 * @port_num:Port number to query
979 * @index:P_Key table index to query
980 * @pkey:Returned P_Key
982 * ib_query_pkey() fetches the specified P_Key table entry.
984 int ib_query_pkey(struct ib_device *device,
985 u8 port_num, u16 index, u16 *pkey)
987 return device->query_pkey(device, port_num, index, pkey);
989 EXPORT_SYMBOL(ib_query_pkey);
992 * ib_modify_device - Change IB device attributes
993 * @device:Device to modify
994 * @device_modify_mask:Mask of attributes to change
995 * @device_modify:New attribute values
997 * ib_modify_device() changes a device's attributes as specified by
998 * the @device_modify_mask and @device_modify structure.
1000 int ib_modify_device(struct ib_device *device,
1001 int device_modify_mask,
1002 struct ib_device_modify *device_modify)
1004 if (!device->modify_device)
1007 return device->modify_device(device, device_modify_mask,
1010 EXPORT_SYMBOL(ib_modify_device);
1013 * ib_modify_port - Modifies the attributes for the specified port.
1014 * @device: The device to modify.
1015 * @port_num: The number of the port to modify.
1016 * @port_modify_mask: Mask used to specify which attributes of the port
1018 * @port_modify: New attribute values for the port.
1020 * ib_modify_port() changes a port's attributes as specified by the
1021 * @port_modify_mask and @port_modify structure.
1023 int ib_modify_port(struct ib_device *device,
1024 u8 port_num, int port_modify_mask,
1025 struct ib_port_modify *port_modify)
1029 if (!rdma_is_port_valid(device, port_num))
1032 if (device->modify_port)
1033 rc = device->modify_port(device, port_num, port_modify_mask,
1036 rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS;
1039 EXPORT_SYMBOL(ib_modify_port);
1042 * ib_find_gid - Returns the port number and GID table index where
1043 * a specified GID value occurs. Its searches only for IB link layer.
1044 * @device: The device to query.
1045 * @gid: The GID value to search for.
1046 * @port_num: The port number of the device where the GID value was found.
1047 * @index: The index into the GID table where the GID was found. This
1048 * parameter may be NULL.
1050 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1051 u8 *port_num, u16 *index)
1053 union ib_gid tmp_gid;
1056 for (port = rdma_start_port(device); port <= rdma_end_port(device); ++port) {
1057 if (!rdma_protocol_ib(device, port))
1060 for (i = 0; i < device->port_immutable[port].gid_tbl_len; ++i) {
1061 ret = ib_query_gid(device, port, i, &tmp_gid, NULL);
1064 if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
1075 EXPORT_SYMBOL(ib_find_gid);
1078 * ib_find_pkey - Returns the PKey table index where a specified
1079 * PKey value occurs.
1080 * @device: The device to query.
1081 * @port_num: The port number of the device to search for the PKey.
1082 * @pkey: The PKey value to search for.
1083 * @index: The index into the PKey table where the PKey was found.
1085 int ib_find_pkey(struct ib_device *device,
1086 u8 port_num, u16 pkey, u16 *index)
1090 int partial_ix = -1;
1092 for (i = 0; i < device->port_immutable[port_num].pkey_tbl_len; ++i) {
1093 ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
1096 if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
1097 /* if there is full-member pkey take it.*/
1098 if (tmp_pkey & 0x8000) {
1107 /*no full-member, if exists take the limited*/
1108 if (partial_ix >= 0) {
1109 *index = partial_ix;
1114 EXPORT_SYMBOL(ib_find_pkey);
1117 * ib_get_net_dev_by_params() - Return the appropriate net_dev
1118 * for a received CM request
1119 * @dev: An RDMA device on which the request has been received.
1120 * @port: Port number on the RDMA device.
1121 * @pkey: The Pkey the request came on.
1122 * @gid: A GID that the net_dev uses to communicate.
1123 * @addr: Contains the IP address that the request specified as its
1126 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev,
1129 const union ib_gid *gid,
1130 const struct sockaddr *addr)
1132 struct net_device *net_dev = NULL;
1133 struct ib_client_data *context;
1135 if (!rdma_protocol_ib(dev, port))
1138 down_read(&lists_rwsem);
1140 list_for_each_entry(context, &dev->client_data_list, list) {
1141 struct ib_client *client = context->client;
1143 if (context->going_down)
1146 if (client->get_net_dev_by_params) {
1147 net_dev = client->get_net_dev_by_params(dev, port, pkey,
1155 up_read(&lists_rwsem);
1159 EXPORT_SYMBOL(ib_get_net_dev_by_params);
1161 static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = {
1162 [RDMA_NL_LS_OP_RESOLVE] = {
1163 .doit = ib_nl_handle_resolve_resp,
1164 .flags = RDMA_NL_ADMIN_PERM,
1166 [RDMA_NL_LS_OP_SET_TIMEOUT] = {
1167 .doit = ib_nl_handle_set_timeout,
1168 .flags = RDMA_NL_ADMIN_PERM,
1170 [RDMA_NL_LS_OP_IP_RESOLVE] = {
1171 .doit = ib_nl_handle_ip_res_resp,
1172 .flags = RDMA_NL_ADMIN_PERM,
1176 static int __init ib_core_init(void)
1180 ib_wq = alloc_workqueue("infiniband", 0, 0);
1184 ib_comp_wq = alloc_workqueue("ib-comp-wq",
1185 WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
1191 ret = class_register(&ib_class);
1193 pr_warn("Couldn't create InfiniBand device class\n");
1197 ret = rdma_nl_init();
1199 pr_warn("Couldn't init IB netlink interface: err %d\n", ret);
1205 pr_warn("Could't init IB address resolution\n");
1209 ret = ib_mad_init();
1211 pr_warn("Couldn't init IB MAD\n");
1217 pr_warn("Couldn't init SA\n");
1221 ret = register_lsm_notifier(&ibdev_lsm_nb);
1223 pr_warn("Couldn't register LSM notifier. ret %d\n", ret);
1228 rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table);
1242 class_unregister(&ib_class);
1244 destroy_workqueue(ib_comp_wq);
1246 destroy_workqueue(ib_wq);
1250 static void __exit ib_core_cleanup(void)
1254 rdma_nl_unregister(RDMA_NL_LS);
1255 unregister_lsm_notifier(&ibdev_lsm_nb);
1260 class_unregister(&ib_class);
1261 destroy_workqueue(ib_comp_wq);
1262 /* Make sure that any pending umem accounting work is done. */
1263 destroy_workqueue(ib_wq);
1266 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4);
1268 subsys_initcall(ib_core_init);
1269 module_exit(ib_core_cleanup);