11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * Copyright (c) 2004 Topspin Communications. All rights reserved. 32a1d9b7fSRoland Dreier * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * This software is available to you under a choice of one of two 61da177e4SLinus Torvalds * licenses. You may choose to be licensed under the terms of the GNU 71da177e4SLinus Torvalds * General Public License (GPL) Version 2, available from the file 81da177e4SLinus Torvalds * COPYING in the main directory of this source tree, or the 91da177e4SLinus Torvalds * OpenIB.org BSD license below: 101da177e4SLinus Torvalds * 111da177e4SLinus Torvalds * Redistribution and use in source and binary forms, with or 121da177e4SLinus Torvalds * without modification, are permitted provided that the following 131da177e4SLinus Torvalds * conditions are met: 141da177e4SLinus Torvalds * 151da177e4SLinus Torvalds * - Redistributions of source code must retain the above 161da177e4SLinus Torvalds * copyright notice, this list of conditions and the following 171da177e4SLinus Torvalds * disclaimer. 181da177e4SLinus Torvalds * 191da177e4SLinus Torvalds * - Redistributions in binary form must reproduce the above 201da177e4SLinus Torvalds * copyright notice, this list of conditions and the following 211da177e4SLinus Torvalds * disclaimer in the documentation and/or other materials 221da177e4SLinus Torvalds * provided with the distribution. 231da177e4SLinus Torvalds * 241da177e4SLinus Torvalds * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 251da177e4SLinus Torvalds * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 261da177e4SLinus Torvalds * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 271da177e4SLinus Torvalds * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 281da177e4SLinus Torvalds * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 291da177e4SLinus Torvalds * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 301da177e4SLinus Torvalds * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 311da177e4SLinus Torvalds * SOFTWARE. 321da177e4SLinus Torvalds */ 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds #include <linux/module.h> 351da177e4SLinus Torvalds #include <linux/string.h> 361da177e4SLinus Torvalds #include <linux/errno.h> 379a6b090cSAhmed S. Darwish #include <linux/kernel.h> 381da177e4SLinus Torvalds #include <linux/slab.h> 391da177e4SLinus Torvalds #include <linux/init.h> 409268f72dSYotam Kenneth #include <linux/netdevice.h> 414e0f7b90SParav Pandit #include <net/net_namespace.h> 424e0f7b90SParav Pandit #include <net/netns/generic.h> 438f408ab6SDaniel Jurgens #include <linux/security.h> 448f408ab6SDaniel Jurgens #include <linux/notifier.h> 45324e227eSJason Gunthorpe #include <linux/hashtable.h> 46b2cbae2cSRoland Dreier #include <rdma/rdma_netlink.h> 4703db3a2dSMatan Barak #include <rdma/ib_addr.h> 4803db3a2dSMatan Barak #include <rdma/ib_cache.h> 491da177e4SLinus Torvalds 501da177e4SLinus Torvalds #include "core_priv.h" 5141eda65cSLeon Romanovsky #include "restrack.h" 521da177e4SLinus Torvalds 531da177e4SLinus Torvalds MODULE_AUTHOR("Roland Dreier"); 541da177e4SLinus Torvalds MODULE_DESCRIPTION("core kernel InfiniBand API"); 551da177e4SLinus Torvalds MODULE_LICENSE("Dual BSD/GPL"); 561da177e4SLinus Torvalds 5714d3a3b2SChristoph Hellwig struct workqueue_struct *ib_comp_wq; 58f794809aSJack Morgenstein struct workqueue_struct *ib_comp_unbound_wq; 59f0626710STejun Heo struct workqueue_struct *ib_wq; 60f0626710STejun Heo EXPORT_SYMBOL_GPL(ib_wq); 61f0626710STejun Heo 620df91bb6SJason Gunthorpe /* 63921eab11SJason Gunthorpe * Each of the three rwsem locks (devices, clients, client_data) protects the 64921eab11SJason Gunthorpe * xarray of the same name. Specifically it allows the caller to assert that 65921eab11SJason Gunthorpe * the MARK will/will not be changing under the lock, and for devices and 66921eab11SJason Gunthorpe * clients, that the value in the xarray is still a valid pointer. Change of 67921eab11SJason Gunthorpe * the MARK is linked to the object state, so holding the lock and testing the 68921eab11SJason Gunthorpe * MARK also asserts that the contained object is in a certain state. 69921eab11SJason Gunthorpe * 70921eab11SJason Gunthorpe * This is used to build a two stage register/unregister flow where objects 71921eab11SJason Gunthorpe * can continue to be in the xarray even though they are still in progress to 72921eab11SJason Gunthorpe * register/unregister. 73921eab11SJason Gunthorpe * 74921eab11SJason Gunthorpe * The xarray itself provides additional locking, and restartable iteration, 75921eab11SJason Gunthorpe * which is also relied on. 76921eab11SJason Gunthorpe * 77921eab11SJason Gunthorpe * Locks should not be nested, with the exception of client_data, which is 78921eab11SJason Gunthorpe * allowed to nest under the read side of the other two locks. 79921eab11SJason Gunthorpe * 80921eab11SJason Gunthorpe * The devices_rwsem also protects the device name list, any change or 81921eab11SJason Gunthorpe * assignment of device name must also hold the write side to guarantee unique 82921eab11SJason Gunthorpe * names. 83921eab11SJason Gunthorpe */ 84921eab11SJason Gunthorpe 85921eab11SJason Gunthorpe /* 860df91bb6SJason Gunthorpe * devices contains devices that have had their names assigned. The 870df91bb6SJason Gunthorpe * devices may not be registered. Users that care about the registration 880df91bb6SJason Gunthorpe * status need to call ib_device_try_get() on the device to ensure it is 890df91bb6SJason Gunthorpe * registered, and keep it registered, for the required duration. 900df91bb6SJason Gunthorpe * 910df91bb6SJason Gunthorpe */ 920df91bb6SJason Gunthorpe static DEFINE_XARRAY_FLAGS(devices, XA_FLAGS_ALLOC); 93921eab11SJason Gunthorpe static DECLARE_RWSEM(devices_rwsem); 940df91bb6SJason Gunthorpe #define DEVICE_REGISTERED XA_MARK_1 950df91bb6SJason Gunthorpe 961da177e4SLinus Torvalds static LIST_HEAD(client_list); 97e59178d8SJason Gunthorpe #define CLIENT_REGISTERED XA_MARK_1 98e59178d8SJason Gunthorpe static DEFINE_XARRAY_FLAGS(clients, XA_FLAGS_ALLOC); 99921eab11SJason Gunthorpe static DECLARE_RWSEM(clients_rwsem); 1001da177e4SLinus Torvalds 1011da177e4SLinus Torvalds /* 1020df91bb6SJason Gunthorpe * If client_data is registered then the corresponding client must also still 1030df91bb6SJason Gunthorpe * be registered. 1040df91bb6SJason Gunthorpe */ 1050df91bb6SJason Gunthorpe #define CLIENT_DATA_REGISTERED XA_MARK_1 1064e0f7b90SParav Pandit 1074e0f7b90SParav Pandit /** 1084e0f7b90SParav Pandit * struct rdma_dev_net - rdma net namespace metadata for a net 1094e0f7b90SParav Pandit * @net: Pointer to owner net namespace 1104e0f7b90SParav Pandit * @id: xarray id to identify the net namespace. 1114e0f7b90SParav Pandit */ 1124e0f7b90SParav Pandit struct rdma_dev_net { 1134e0f7b90SParav Pandit possible_net_t net; 1144e0f7b90SParav Pandit u32 id; 1154e0f7b90SParav Pandit }; 1164e0f7b90SParav Pandit 1174e0f7b90SParav Pandit static unsigned int rdma_dev_net_id; 1184e0f7b90SParav Pandit 1194e0f7b90SParav Pandit /* 1204e0f7b90SParav Pandit * A list of net namespaces is maintained in an xarray. This is necessary 1214e0f7b90SParav Pandit * because we can't get the locking right using the existing net ns list. We 1224e0f7b90SParav Pandit * would require a init_net callback after the list is updated. 1234e0f7b90SParav Pandit */ 1244e0f7b90SParav Pandit static DEFINE_XARRAY_FLAGS(rdma_nets, XA_FLAGS_ALLOC); 1254e0f7b90SParav Pandit /* 1264e0f7b90SParav Pandit * rwsem to protect accessing the rdma_nets xarray entries. 1274e0f7b90SParav Pandit */ 1284e0f7b90SParav Pandit static DECLARE_RWSEM(rdma_nets_rwsem); 1294e0f7b90SParav Pandit 1300df91bb6SJason Gunthorpe /* 1310df91bb6SJason Gunthorpe * xarray has this behavior where it won't iterate over NULL values stored in 1320df91bb6SJason Gunthorpe * allocated arrays. So we need our own iterator to see all values stored in 1330df91bb6SJason Gunthorpe * the array. This does the same thing as xa_for_each except that it also 1340df91bb6SJason Gunthorpe * returns NULL valued entries if the array is allocating. Simplified to only 1350df91bb6SJason Gunthorpe * work on simple xarrays. 1360df91bb6SJason Gunthorpe */ 1370df91bb6SJason Gunthorpe static void *xan_find_marked(struct xarray *xa, unsigned long *indexp, 1380df91bb6SJason Gunthorpe xa_mark_t filter) 1390df91bb6SJason Gunthorpe { 1400df91bb6SJason Gunthorpe XA_STATE(xas, xa, *indexp); 1410df91bb6SJason Gunthorpe void *entry; 1420df91bb6SJason Gunthorpe 1430df91bb6SJason Gunthorpe rcu_read_lock(); 1440df91bb6SJason Gunthorpe do { 1450df91bb6SJason Gunthorpe entry = xas_find_marked(&xas, ULONG_MAX, filter); 1460df91bb6SJason Gunthorpe if (xa_is_zero(entry)) 1470df91bb6SJason Gunthorpe break; 1480df91bb6SJason Gunthorpe } while (xas_retry(&xas, entry)); 1490df91bb6SJason Gunthorpe rcu_read_unlock(); 1500df91bb6SJason Gunthorpe 1510df91bb6SJason Gunthorpe if (entry) { 1520df91bb6SJason Gunthorpe *indexp = xas.xa_index; 1530df91bb6SJason Gunthorpe if (xa_is_zero(entry)) 1540df91bb6SJason Gunthorpe return NULL; 1550df91bb6SJason Gunthorpe return entry; 1560df91bb6SJason Gunthorpe } 1570df91bb6SJason Gunthorpe return XA_ERROR(-ENOENT); 1580df91bb6SJason Gunthorpe } 1590df91bb6SJason Gunthorpe #define xan_for_each_marked(xa, index, entry, filter) \ 1600df91bb6SJason Gunthorpe for (index = 0, entry = xan_find_marked(xa, &(index), filter); \ 1610df91bb6SJason Gunthorpe !xa_is_err(entry); \ 1620df91bb6SJason Gunthorpe (index)++, entry = xan_find_marked(xa, &(index), filter)) 1630df91bb6SJason Gunthorpe 164324e227eSJason Gunthorpe /* RCU hash table mapping netdevice pointers to struct ib_port_data */ 165324e227eSJason Gunthorpe static DEFINE_SPINLOCK(ndev_hash_lock); 166324e227eSJason Gunthorpe static DECLARE_HASHTABLE(ndev_hash, 5); 167324e227eSJason Gunthorpe 168c2261dd7SJason Gunthorpe static void free_netdevs(struct ib_device *ib_dev); 169d0899892SJason Gunthorpe static void ib_unregister_work(struct work_struct *work); 170d0899892SJason Gunthorpe static void __ib_unregister_device(struct ib_device *device); 1718f408ab6SDaniel Jurgens static int ib_security_change(struct notifier_block *nb, unsigned long event, 1728f408ab6SDaniel Jurgens void *lsm_data); 1738f408ab6SDaniel Jurgens static void ib_policy_change_task(struct work_struct *work); 1748f408ab6SDaniel Jurgens static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task); 1758f408ab6SDaniel Jurgens 1768f408ab6SDaniel Jurgens static struct notifier_block ibdev_lsm_nb = { 1778f408ab6SDaniel Jurgens .notifier_call = ib_security_change, 1788f408ab6SDaniel Jurgens }; 1791da177e4SLinus Torvalds 180324e227eSJason Gunthorpe /* Pointer to the RCU head at the start of the ib_port_data array */ 181324e227eSJason Gunthorpe struct ib_port_data_rcu { 182324e227eSJason Gunthorpe struct rcu_head rcu_head; 183324e227eSJason Gunthorpe struct ib_port_data pdata[]; 184324e227eSJason Gunthorpe }; 185324e227eSJason Gunthorpe 1861da177e4SLinus Torvalds static int ib_device_check_mandatory(struct ib_device *device) 1871da177e4SLinus Torvalds { 1883023a1e9SKamal Heib #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device_ops, x), #x } 1891da177e4SLinus Torvalds static const struct { 1901da177e4SLinus Torvalds size_t offset; 1911da177e4SLinus Torvalds char *name; 1921da177e4SLinus Torvalds } mandatory_table[] = { 1931da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_device), 1941da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_port), 1951da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_pkey), 1961da177e4SLinus Torvalds IB_MANDATORY_FUNC(alloc_pd), 1971da177e4SLinus Torvalds IB_MANDATORY_FUNC(dealloc_pd), 1981da177e4SLinus Torvalds IB_MANDATORY_FUNC(create_qp), 1991da177e4SLinus Torvalds IB_MANDATORY_FUNC(modify_qp), 2001da177e4SLinus Torvalds IB_MANDATORY_FUNC(destroy_qp), 2011da177e4SLinus Torvalds IB_MANDATORY_FUNC(post_send), 2021da177e4SLinus Torvalds IB_MANDATORY_FUNC(post_recv), 2031da177e4SLinus Torvalds IB_MANDATORY_FUNC(create_cq), 2041da177e4SLinus Torvalds IB_MANDATORY_FUNC(destroy_cq), 2051da177e4SLinus Torvalds IB_MANDATORY_FUNC(poll_cq), 2061da177e4SLinus Torvalds IB_MANDATORY_FUNC(req_notify_cq), 2071da177e4SLinus Torvalds IB_MANDATORY_FUNC(get_dma_mr), 2087738613eSIra Weiny IB_MANDATORY_FUNC(dereg_mr), 2097738613eSIra Weiny IB_MANDATORY_FUNC(get_port_immutable) 2101da177e4SLinus Torvalds }; 2111da177e4SLinus Torvalds int i; 2121da177e4SLinus Torvalds 2136780c4faSGal Pressman device->kverbs_provider = true; 2149a6b090cSAhmed S. Darwish for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { 2153023a1e9SKamal Heib if (!*(void **) ((void *) &device->ops + 2163023a1e9SKamal Heib mandatory_table[i].offset)) { 2176780c4faSGal Pressman device->kverbs_provider = false; 2186780c4faSGal Pressman break; 2191da177e4SLinus Torvalds } 2201da177e4SLinus Torvalds } 2211da177e4SLinus Torvalds 2221da177e4SLinus Torvalds return 0; 2231da177e4SLinus Torvalds } 2241da177e4SLinus Torvalds 225f8978bd9SLeon Romanovsky /* 22601b67117SParav Pandit * Caller must perform ib_device_put() to return the device reference count 22701b67117SParav Pandit * when ib_device_get_by_index() returns valid device pointer. 228f8978bd9SLeon Romanovsky */ 229f8978bd9SLeon Romanovsky struct ib_device *ib_device_get_by_index(u32 index) 230f8978bd9SLeon Romanovsky { 231f8978bd9SLeon Romanovsky struct ib_device *device; 232f8978bd9SLeon Romanovsky 233921eab11SJason Gunthorpe down_read(&devices_rwsem); 2340df91bb6SJason Gunthorpe device = xa_load(&devices, index); 23501b67117SParav Pandit if (device) { 236d79af724SJason Gunthorpe if (!ib_device_try_get(device)) 23701b67117SParav Pandit device = NULL; 23801b67117SParav Pandit } 239921eab11SJason Gunthorpe up_read(&devices_rwsem); 240f8978bd9SLeon Romanovsky return device; 241f8978bd9SLeon Romanovsky } 242f8978bd9SLeon Romanovsky 243d79af724SJason Gunthorpe /** 244d79af724SJason Gunthorpe * ib_device_put - Release IB device reference 245d79af724SJason Gunthorpe * @device: device whose reference to be released 246d79af724SJason Gunthorpe * 247d79af724SJason Gunthorpe * ib_device_put() releases reference to the IB device to allow it to be 248d79af724SJason Gunthorpe * unregistered and eventually free. 249d79af724SJason Gunthorpe */ 25001b67117SParav Pandit void ib_device_put(struct ib_device *device) 25101b67117SParav Pandit { 25201b67117SParav Pandit if (refcount_dec_and_test(&device->refcount)) 25301b67117SParav Pandit complete(&device->unreg_completion); 25401b67117SParav Pandit } 255d79af724SJason Gunthorpe EXPORT_SYMBOL(ib_device_put); 25601b67117SParav Pandit 2571da177e4SLinus Torvalds static struct ib_device *__ib_device_get_by_name(const char *name) 2581da177e4SLinus Torvalds { 2591da177e4SLinus Torvalds struct ib_device *device; 2600df91bb6SJason Gunthorpe unsigned long index; 2611da177e4SLinus Torvalds 2620df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) 263896de009SJason Gunthorpe if (!strcmp(name, dev_name(&device->dev))) 2641da177e4SLinus Torvalds return device; 2651da177e4SLinus Torvalds 2661da177e4SLinus Torvalds return NULL; 2671da177e4SLinus Torvalds } 2681da177e4SLinus Torvalds 2696cc2c8e5SJason Gunthorpe /** 2706cc2c8e5SJason Gunthorpe * ib_device_get_by_name - Find an IB device by name 2716cc2c8e5SJason Gunthorpe * @name: The name to look for 2726cc2c8e5SJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 2736cc2c8e5SJason Gunthorpe * 2746cc2c8e5SJason Gunthorpe * Find and hold an ib_device by its name. The caller must call 2756cc2c8e5SJason Gunthorpe * ib_device_put() on the returned pointer. 2766cc2c8e5SJason Gunthorpe */ 2776cc2c8e5SJason Gunthorpe struct ib_device *ib_device_get_by_name(const char *name, 2786cc2c8e5SJason Gunthorpe enum rdma_driver_id driver_id) 2796cc2c8e5SJason Gunthorpe { 2806cc2c8e5SJason Gunthorpe struct ib_device *device; 2816cc2c8e5SJason Gunthorpe 2826cc2c8e5SJason Gunthorpe down_read(&devices_rwsem); 2836cc2c8e5SJason Gunthorpe device = __ib_device_get_by_name(name); 2846cc2c8e5SJason Gunthorpe if (device && driver_id != RDMA_DRIVER_UNKNOWN && 2856cc2c8e5SJason Gunthorpe device->driver_id != driver_id) 2866cc2c8e5SJason Gunthorpe device = NULL; 2876cc2c8e5SJason Gunthorpe 2886cc2c8e5SJason Gunthorpe if (device) { 2896cc2c8e5SJason Gunthorpe if (!ib_device_try_get(device)) 2906cc2c8e5SJason Gunthorpe device = NULL; 2916cc2c8e5SJason Gunthorpe } 2926cc2c8e5SJason Gunthorpe up_read(&devices_rwsem); 2936cc2c8e5SJason Gunthorpe return device; 2946cc2c8e5SJason Gunthorpe } 2956cc2c8e5SJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_name); 2966cc2c8e5SJason Gunthorpe 2974e0f7b90SParav Pandit static int rename_compat_devs(struct ib_device *device) 2984e0f7b90SParav Pandit { 2994e0f7b90SParav Pandit struct ib_core_device *cdev; 3004e0f7b90SParav Pandit unsigned long index; 3014e0f7b90SParav Pandit int ret = 0; 3024e0f7b90SParav Pandit 3034e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 3044e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) { 3054e0f7b90SParav Pandit ret = device_rename(&cdev->dev, dev_name(&device->dev)); 3064e0f7b90SParav Pandit if (ret) { 3074e0f7b90SParav Pandit dev_warn(&cdev->dev, 3084e0f7b90SParav Pandit "Fail to rename compatdev to new name %s\n", 3094e0f7b90SParav Pandit dev_name(&device->dev)); 3104e0f7b90SParav Pandit break; 3114e0f7b90SParav Pandit } 3124e0f7b90SParav Pandit } 3134e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 3144e0f7b90SParav Pandit return ret; 3154e0f7b90SParav Pandit } 3164e0f7b90SParav Pandit 317d21943ddSLeon Romanovsky int ib_device_rename(struct ib_device *ibdev, const char *name) 318d21943ddSLeon Romanovsky { 319e3593b56SJason Gunthorpe int ret; 320d21943ddSLeon Romanovsky 321921eab11SJason Gunthorpe down_write(&devices_rwsem); 322e3593b56SJason Gunthorpe if (!strcmp(name, dev_name(&ibdev->dev))) { 323e3593b56SJason Gunthorpe ret = 0; 324e3593b56SJason Gunthorpe goto out; 325e3593b56SJason Gunthorpe } 326e3593b56SJason Gunthorpe 327344684e6SJason Gunthorpe if (__ib_device_get_by_name(name)) { 328d21943ddSLeon Romanovsky ret = -EEXIST; 329d21943ddSLeon Romanovsky goto out; 330d21943ddSLeon Romanovsky } 331d21943ddSLeon Romanovsky 332d21943ddSLeon Romanovsky ret = device_rename(&ibdev->dev, name); 333d21943ddSLeon Romanovsky if (ret) 334d21943ddSLeon Romanovsky goto out; 335d21943ddSLeon Romanovsky strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); 3364e0f7b90SParav Pandit ret = rename_compat_devs(ibdev); 337d21943ddSLeon Romanovsky out: 338921eab11SJason Gunthorpe up_write(&devices_rwsem); 339d21943ddSLeon Romanovsky return ret; 340d21943ddSLeon Romanovsky } 341d21943ddSLeon Romanovsky 342e349f858SJason Gunthorpe static int alloc_name(struct ib_device *ibdev, const char *name) 3431da177e4SLinus Torvalds { 3441da177e4SLinus Torvalds struct ib_device *device; 3450df91bb6SJason Gunthorpe unsigned long index; 3463b88afd3SJason Gunthorpe struct ida inuse; 3473b88afd3SJason Gunthorpe int rc; 3481da177e4SLinus Torvalds int i; 3491da177e4SLinus Torvalds 350921eab11SJason Gunthorpe lockdep_assert_held_exclusive(&devices_rwsem); 3513b88afd3SJason Gunthorpe ida_init(&inuse); 3520df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) { 353e349f858SJason Gunthorpe char buf[IB_DEVICE_NAME_MAX]; 354e349f858SJason Gunthorpe 355896de009SJason Gunthorpe if (sscanf(dev_name(&device->dev), name, &i) != 1) 3561da177e4SLinus Torvalds continue; 3573b88afd3SJason Gunthorpe if (i < 0 || i >= INT_MAX) 3581da177e4SLinus Torvalds continue; 3591da177e4SLinus Torvalds snprintf(buf, sizeof buf, name, i); 3603b88afd3SJason Gunthorpe if (strcmp(buf, dev_name(&device->dev)) != 0) 3613b88afd3SJason Gunthorpe continue; 3623b88afd3SJason Gunthorpe 3633b88afd3SJason Gunthorpe rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); 3643b88afd3SJason Gunthorpe if (rc < 0) 3653b88afd3SJason Gunthorpe goto out; 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds 3683b88afd3SJason Gunthorpe rc = ida_alloc(&inuse, GFP_KERNEL); 3693b88afd3SJason Gunthorpe if (rc < 0) 3703b88afd3SJason Gunthorpe goto out; 3711da177e4SLinus Torvalds 3723b88afd3SJason Gunthorpe rc = dev_set_name(&ibdev->dev, name, rc); 3733b88afd3SJason Gunthorpe out: 3743b88afd3SJason Gunthorpe ida_destroy(&inuse); 3753b88afd3SJason Gunthorpe return rc; 3761da177e4SLinus Torvalds } 3771da177e4SLinus Torvalds 37855aeed06SJason Gunthorpe static void ib_device_release(struct device *device) 37955aeed06SJason Gunthorpe { 38055aeed06SJason Gunthorpe struct ib_device *dev = container_of(device, struct ib_device, dev); 38155aeed06SJason Gunthorpe 382c2261dd7SJason Gunthorpe free_netdevs(dev); 383652432f3SJason Gunthorpe WARN_ON(refcount_read(&dev->refcount)); 38403db3a2dSMatan Barak ib_cache_release_one(dev); 385b34b269aSJason Gunthorpe ib_security_release_port_pkey_list(dev); 3864e0f7b90SParav Pandit xa_destroy(&dev->compat_devs); 3870df91bb6SJason Gunthorpe xa_destroy(&dev->client_data); 388324e227eSJason Gunthorpe if (dev->port_data) 389324e227eSJason Gunthorpe kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, 390324e227eSJason Gunthorpe pdata[0]), 391324e227eSJason Gunthorpe rcu_head); 392324e227eSJason Gunthorpe kfree_rcu(dev, rcu_head); 39355aeed06SJason Gunthorpe } 39455aeed06SJason Gunthorpe 39555aeed06SJason Gunthorpe static int ib_device_uevent(struct device *device, 39655aeed06SJason Gunthorpe struct kobj_uevent_env *env) 39755aeed06SJason Gunthorpe { 398896de009SJason Gunthorpe if (add_uevent_var(env, "NAME=%s", dev_name(device))) 39955aeed06SJason Gunthorpe return -ENOMEM; 40055aeed06SJason Gunthorpe 40155aeed06SJason Gunthorpe /* 40255aeed06SJason Gunthorpe * It would be nice to pass the node GUID with the event... 40355aeed06SJason Gunthorpe */ 40455aeed06SJason Gunthorpe 40555aeed06SJason Gunthorpe return 0; 40655aeed06SJason Gunthorpe } 40755aeed06SJason Gunthorpe 40862dfa795SParav Pandit static const void *net_namespace(struct device *d) 40962dfa795SParav Pandit { 4104e0f7b90SParav Pandit struct ib_core_device *coredev = 4114e0f7b90SParav Pandit container_of(d, struct ib_core_device, dev); 4124e0f7b90SParav Pandit 4134e0f7b90SParav Pandit return read_pnet(&coredev->rdma_net); 41462dfa795SParav Pandit } 41562dfa795SParav Pandit 41655aeed06SJason Gunthorpe static struct class ib_class = { 41755aeed06SJason Gunthorpe .name = "infiniband", 41855aeed06SJason Gunthorpe .dev_release = ib_device_release, 41955aeed06SJason Gunthorpe .dev_uevent = ib_device_uevent, 42062dfa795SParav Pandit .ns_type = &net_ns_type_operations, 42162dfa795SParav Pandit .namespace = net_namespace, 42255aeed06SJason Gunthorpe }; 42355aeed06SJason Gunthorpe 424cebe556bSParav Pandit static void rdma_init_coredev(struct ib_core_device *coredev, 4254e0f7b90SParav Pandit struct ib_device *dev, struct net *net) 426cebe556bSParav Pandit { 427cebe556bSParav Pandit /* This BUILD_BUG_ON is intended to catch layout change 428cebe556bSParav Pandit * of union of ib_core_device and device. 429cebe556bSParav Pandit * dev must be the first element as ib_core and providers 430cebe556bSParav Pandit * driver uses it. Adding anything in ib_core_device before 431cebe556bSParav Pandit * device will break this assumption. 432cebe556bSParav Pandit */ 433cebe556bSParav Pandit BUILD_BUG_ON(offsetof(struct ib_device, coredev.dev) != 434cebe556bSParav Pandit offsetof(struct ib_device, dev)); 435cebe556bSParav Pandit 436cebe556bSParav Pandit coredev->dev.class = &ib_class; 437cebe556bSParav Pandit coredev->dev.groups = dev->groups; 438cebe556bSParav Pandit device_initialize(&coredev->dev); 439cebe556bSParav Pandit coredev->owner = dev; 440cebe556bSParav Pandit INIT_LIST_HEAD(&coredev->port_list); 4414e0f7b90SParav Pandit write_pnet(&coredev->rdma_net, net); 442cebe556bSParav Pandit } 443cebe556bSParav Pandit 4441da177e4SLinus Torvalds /** 445459cc69fSLeon Romanovsky * _ib_alloc_device - allocate an IB device struct 4461da177e4SLinus Torvalds * @size:size of structure to allocate 4471da177e4SLinus Torvalds * 4481da177e4SLinus Torvalds * Low-level drivers should use ib_alloc_device() to allocate &struct 4491da177e4SLinus Torvalds * ib_device. @size is the size of the structure to be allocated, 4501da177e4SLinus Torvalds * including any private data used by the low-level driver. 4511da177e4SLinus Torvalds * ib_dealloc_device() must be used to free structures allocated with 4521da177e4SLinus Torvalds * ib_alloc_device(). 4531da177e4SLinus Torvalds */ 454459cc69fSLeon Romanovsky struct ib_device *_ib_alloc_device(size_t size) 4551da177e4SLinus Torvalds { 45655aeed06SJason Gunthorpe struct ib_device *device; 4571da177e4SLinus Torvalds 45855aeed06SJason Gunthorpe if (WARN_ON(size < sizeof(struct ib_device))) 45955aeed06SJason Gunthorpe return NULL; 46055aeed06SJason Gunthorpe 46155aeed06SJason Gunthorpe device = kzalloc(size, GFP_KERNEL); 46255aeed06SJason Gunthorpe if (!device) 46355aeed06SJason Gunthorpe return NULL; 46455aeed06SJason Gunthorpe 46541eda65cSLeon Romanovsky if (rdma_restrack_init(device)) { 46641eda65cSLeon Romanovsky kfree(device); 46741eda65cSLeon Romanovsky return NULL; 46841eda65cSLeon Romanovsky } 46902d8883fSLeon Romanovsky 4705f8f5499SParav Pandit device->groups[0] = &ib_dev_attr_group; 4714e0f7b90SParav Pandit rdma_init_coredev(&device->coredev, device, &init_net); 47255aeed06SJason Gunthorpe 47355aeed06SJason Gunthorpe INIT_LIST_HEAD(&device->event_handler_list); 47455aeed06SJason Gunthorpe spin_lock_init(&device->event_handler_lock); 475d0899892SJason Gunthorpe mutex_init(&device->unregistration_lock); 4760df91bb6SJason Gunthorpe /* 4770df91bb6SJason Gunthorpe * client_data needs to be alloc because we don't want our mark to be 4780df91bb6SJason Gunthorpe * destroyed if the user stores NULL in the client data. 4790df91bb6SJason Gunthorpe */ 4800df91bb6SJason Gunthorpe xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); 481921eab11SJason Gunthorpe init_rwsem(&device->client_data_rwsem); 4824e0f7b90SParav Pandit xa_init_flags(&device->compat_devs, XA_FLAGS_ALLOC); 4834e0f7b90SParav Pandit mutex_init(&device->compat_devs_mutex); 48401b67117SParav Pandit init_completion(&device->unreg_completion); 485d0899892SJason Gunthorpe INIT_WORK(&device->unregistration_work, ib_unregister_work); 48655aeed06SJason Gunthorpe 48755aeed06SJason Gunthorpe return device; 4881da177e4SLinus Torvalds } 489459cc69fSLeon Romanovsky EXPORT_SYMBOL(_ib_alloc_device); 4901da177e4SLinus Torvalds 4911da177e4SLinus Torvalds /** 4921da177e4SLinus Torvalds * ib_dealloc_device - free an IB device struct 4931da177e4SLinus Torvalds * @device:structure to free 4941da177e4SLinus Torvalds * 4951da177e4SLinus Torvalds * Free a structure allocated with ib_alloc_device(). 4961da177e4SLinus Torvalds */ 4971da177e4SLinus Torvalds void ib_dealloc_device(struct ib_device *device) 4981da177e4SLinus Torvalds { 499d0899892SJason Gunthorpe if (device->ops.dealloc_driver) 500d0899892SJason Gunthorpe device->ops.dealloc_driver(device); 501d0899892SJason Gunthorpe 502d0899892SJason Gunthorpe /* 503d0899892SJason Gunthorpe * ib_unregister_driver() requires all devices to remain in the xarray 504d0899892SJason Gunthorpe * while their ops are callable. The last op we call is dealloc_driver 505d0899892SJason Gunthorpe * above. This is needed to create a fence on op callbacks prior to 506d0899892SJason Gunthorpe * allowing the driver module to unload. 507d0899892SJason Gunthorpe */ 508d0899892SJason Gunthorpe down_write(&devices_rwsem); 509d0899892SJason Gunthorpe if (xa_load(&devices, device->index) == device) 510d0899892SJason Gunthorpe xa_erase(&devices, device->index); 511d0899892SJason Gunthorpe up_write(&devices_rwsem); 512d0899892SJason Gunthorpe 513c2261dd7SJason Gunthorpe /* Expedite releasing netdev references */ 514c2261dd7SJason Gunthorpe free_netdevs(device); 515c2261dd7SJason Gunthorpe 5164e0f7b90SParav Pandit WARN_ON(!xa_empty(&device->compat_devs)); 5170df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&device->client_data)); 518652432f3SJason Gunthorpe WARN_ON(refcount_read(&device->refcount)); 5190ad699c0SLeon Romanovsky rdma_restrack_clean(device); 520e155755eSParav Pandit /* Balances with device_initialize */ 521924b8900SLeon Romanovsky put_device(&device->dev); 5221da177e4SLinus Torvalds } 5231da177e4SLinus Torvalds EXPORT_SYMBOL(ib_dealloc_device); 5241da177e4SLinus Torvalds 525921eab11SJason Gunthorpe /* 526921eab11SJason Gunthorpe * add_client_context() and remove_client_context() must be safe against 527921eab11SJason Gunthorpe * parallel calls on the same device - registration/unregistration of both the 528921eab11SJason Gunthorpe * device and client can be occurring in parallel. 529921eab11SJason Gunthorpe * 530921eab11SJason Gunthorpe * The routines need to be a fence, any caller must not return until the add 531921eab11SJason Gunthorpe * or remove is fully completed. 532921eab11SJason Gunthorpe */ 533921eab11SJason Gunthorpe static int add_client_context(struct ib_device *device, 534921eab11SJason Gunthorpe struct ib_client *client) 5351da177e4SLinus Torvalds { 536921eab11SJason Gunthorpe int ret = 0; 5371da177e4SLinus Torvalds 5386780c4faSGal Pressman if (!device->kverbs_provider && !client->no_kverbs_req) 539921eab11SJason Gunthorpe return 0; 5406780c4faSGal Pressman 541921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 542921eab11SJason Gunthorpe /* 543921eab11SJason Gunthorpe * Another caller to add_client_context got here first and has already 544921eab11SJason Gunthorpe * completely initialized context. 545921eab11SJason Gunthorpe */ 546921eab11SJason Gunthorpe if (xa_get_mark(&device->client_data, client->client_id, 547921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) 548921eab11SJason Gunthorpe goto out; 549921eab11SJason Gunthorpe 550921eab11SJason Gunthorpe ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, 551921eab11SJason Gunthorpe GFP_KERNEL)); 552921eab11SJason Gunthorpe if (ret) 553921eab11SJason Gunthorpe goto out; 554921eab11SJason Gunthorpe downgrade_write(&device->client_data_rwsem); 555921eab11SJason Gunthorpe if (client->add) 556921eab11SJason Gunthorpe client->add(device); 557921eab11SJason Gunthorpe 558921eab11SJason Gunthorpe /* Readers shall not see a client until add has been completed */ 5590df91bb6SJason Gunthorpe xa_set_mark(&device->client_data, client->client_id, 5600df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED); 561921eab11SJason Gunthorpe up_read(&device->client_data_rwsem); 562921eab11SJason Gunthorpe return 0; 5631da177e4SLinus Torvalds 564921eab11SJason Gunthorpe out: 565921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 566921eab11SJason Gunthorpe return ret; 567921eab11SJason Gunthorpe } 568921eab11SJason Gunthorpe 569921eab11SJason Gunthorpe static void remove_client_context(struct ib_device *device, 570921eab11SJason Gunthorpe unsigned int client_id) 571921eab11SJason Gunthorpe { 572921eab11SJason Gunthorpe struct ib_client *client; 573921eab11SJason Gunthorpe void *client_data; 574921eab11SJason Gunthorpe 575921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 576921eab11SJason Gunthorpe if (!xa_get_mark(&device->client_data, client_id, 577921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) { 578921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 579921eab11SJason Gunthorpe return; 580921eab11SJason Gunthorpe } 581921eab11SJason Gunthorpe client_data = xa_load(&device->client_data, client_id); 582921eab11SJason Gunthorpe xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); 583921eab11SJason Gunthorpe client = xa_load(&clients, client_id); 584921eab11SJason Gunthorpe downgrade_write(&device->client_data_rwsem); 585921eab11SJason Gunthorpe 586921eab11SJason Gunthorpe /* 587921eab11SJason Gunthorpe * Notice we cannot be holding any exclusive locks when calling the 588921eab11SJason Gunthorpe * remove callback as the remove callback can recurse back into any 589921eab11SJason Gunthorpe * public functions in this module and thus try for any locks those 590921eab11SJason Gunthorpe * functions take. 591921eab11SJason Gunthorpe * 592921eab11SJason Gunthorpe * For this reason clients and drivers should not call the 593921eab11SJason Gunthorpe * unregistration functions will holdling any locks. 594921eab11SJason Gunthorpe * 595921eab11SJason Gunthorpe * It tempting to drop the client_data_rwsem too, but this is required 596921eab11SJason Gunthorpe * to ensure that unregister_client does not return until all clients 597921eab11SJason Gunthorpe * are completely unregistered, which is required to avoid module 598921eab11SJason Gunthorpe * unloading races. 599921eab11SJason Gunthorpe */ 600921eab11SJason Gunthorpe if (client->remove) 601921eab11SJason Gunthorpe client->remove(device, client_data); 602921eab11SJason Gunthorpe 603921eab11SJason Gunthorpe xa_erase(&device->client_data, client_id); 604921eab11SJason Gunthorpe up_read(&device->client_data_rwsem); 6051da177e4SLinus Torvalds } 6061da177e4SLinus Torvalds 607c2261dd7SJason Gunthorpe static int alloc_port_data(struct ib_device *device) 6085eb620c8SYosef Etigin { 609324e227eSJason Gunthorpe struct ib_port_data_rcu *pdata_rcu; 610ea1075edSJason Gunthorpe unsigned int port; 611c2261dd7SJason Gunthorpe 612c2261dd7SJason Gunthorpe if (device->port_data) 613c2261dd7SJason Gunthorpe return 0; 614c2261dd7SJason Gunthorpe 615c2261dd7SJason Gunthorpe /* This can only be called once the physical port range is defined */ 616c2261dd7SJason Gunthorpe if (WARN_ON(!device->phys_port_cnt)) 617c2261dd7SJason Gunthorpe return -EINVAL; 6185eb620c8SYosef Etigin 6198ceb1357SJason Gunthorpe /* 6208ceb1357SJason Gunthorpe * device->port_data is indexed directly by the port number to make 6217738613eSIra Weiny * access to this data as efficient as possible. 6227738613eSIra Weiny * 6238ceb1357SJason Gunthorpe * Therefore port_data is declared as a 1 based array with potential 6248ceb1357SJason Gunthorpe * empty slots at the beginning. 6257738613eSIra Weiny */ 626324e227eSJason Gunthorpe pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, 627324e227eSJason Gunthorpe rdma_end_port(device) + 1), 628324e227eSJason Gunthorpe GFP_KERNEL); 629324e227eSJason Gunthorpe if (!pdata_rcu) 63055aeed06SJason Gunthorpe return -ENOMEM; 631324e227eSJason Gunthorpe /* 632324e227eSJason Gunthorpe * The rcu_head is put in front of the port data array and the stored 633324e227eSJason Gunthorpe * pointer is adjusted since we never need to see that member until 634324e227eSJason Gunthorpe * kfree_rcu. 635324e227eSJason Gunthorpe */ 636324e227eSJason Gunthorpe device->port_data = pdata_rcu->pdata; 6375eb620c8SYosef Etigin 638ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 6398ceb1357SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 6408ceb1357SJason Gunthorpe 641324e227eSJason Gunthorpe pdata->ib_dev = device; 6428ceb1357SJason Gunthorpe spin_lock_init(&pdata->pkey_list_lock); 6438ceb1357SJason Gunthorpe INIT_LIST_HEAD(&pdata->pkey_list); 644c2261dd7SJason Gunthorpe spin_lock_init(&pdata->netdev_lock); 645324e227eSJason Gunthorpe INIT_HLIST_NODE(&pdata->ndev_hash_link); 646c2261dd7SJason Gunthorpe } 647c2261dd7SJason Gunthorpe return 0; 648c2261dd7SJason Gunthorpe } 649c2261dd7SJason Gunthorpe 650c2261dd7SJason Gunthorpe static int verify_immutable(const struct ib_device *dev, u8 port) 651c2261dd7SJason Gunthorpe { 652c2261dd7SJason Gunthorpe return WARN_ON(!rdma_cap_ib_mad(dev, port) && 653c2261dd7SJason Gunthorpe rdma_max_mad_size(dev, port) != 0); 654c2261dd7SJason Gunthorpe } 655c2261dd7SJason Gunthorpe 656c2261dd7SJason Gunthorpe static int setup_port_data(struct ib_device *device) 657c2261dd7SJason Gunthorpe { 658c2261dd7SJason Gunthorpe unsigned int port; 659c2261dd7SJason Gunthorpe int ret; 660c2261dd7SJason Gunthorpe 661c2261dd7SJason Gunthorpe ret = alloc_port_data(device); 662c2261dd7SJason Gunthorpe if (ret) 663c2261dd7SJason Gunthorpe return ret; 664c2261dd7SJason Gunthorpe 665c2261dd7SJason Gunthorpe rdma_for_each_port (device, port) { 666c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 6678ceb1357SJason Gunthorpe 6688ceb1357SJason Gunthorpe ret = device->ops.get_port_immutable(device, port, 6698ceb1357SJason Gunthorpe &pdata->immutable); 6705eb620c8SYosef Etigin if (ret) 6715eb620c8SYosef Etigin return ret; 67255aeed06SJason Gunthorpe 67355aeed06SJason Gunthorpe if (verify_immutable(device, port)) 67455aeed06SJason Gunthorpe return -EINVAL; 67555aeed06SJason Gunthorpe } 67655aeed06SJason Gunthorpe return 0; 6775eb620c8SYosef Etigin } 6785eb620c8SYosef Etigin 6799abb0d1bSLeon Romanovsky void ib_get_device_fw_str(struct ib_device *dev, char *str) 6805fa76c20SIra Weiny { 6813023a1e9SKamal Heib if (dev->ops.get_dev_fw_str) 6823023a1e9SKamal Heib dev->ops.get_dev_fw_str(dev, str); 6835fa76c20SIra Weiny else 6845fa76c20SIra Weiny str[0] = '\0'; 6855fa76c20SIra Weiny } 6865fa76c20SIra Weiny EXPORT_SYMBOL(ib_get_device_fw_str); 6875fa76c20SIra Weiny 6888f408ab6SDaniel Jurgens static void ib_policy_change_task(struct work_struct *work) 6898f408ab6SDaniel Jurgens { 6908f408ab6SDaniel Jurgens struct ib_device *dev; 6910df91bb6SJason Gunthorpe unsigned long index; 6928f408ab6SDaniel Jurgens 693921eab11SJason Gunthorpe down_read(&devices_rwsem); 6940df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 695ea1075edSJason Gunthorpe unsigned int i; 6968f408ab6SDaniel Jurgens 697ea1075edSJason Gunthorpe rdma_for_each_port (dev, i) { 6988f408ab6SDaniel Jurgens u64 sp; 6998f408ab6SDaniel Jurgens int ret = ib_get_cached_subnet_prefix(dev, 7008f408ab6SDaniel Jurgens i, 7018f408ab6SDaniel Jurgens &sp); 7028f408ab6SDaniel Jurgens 7038f408ab6SDaniel Jurgens WARN_ONCE(ret, 7048f408ab6SDaniel Jurgens "ib_get_cached_subnet_prefix err: %d, this should never happen here\n", 7058f408ab6SDaniel Jurgens ret); 706a750cfdeSDaniel Jurgens if (!ret) 7078f408ab6SDaniel Jurgens ib_security_cache_change(dev, i, sp); 7088f408ab6SDaniel Jurgens } 7098f408ab6SDaniel Jurgens } 710921eab11SJason Gunthorpe up_read(&devices_rwsem); 7118f408ab6SDaniel Jurgens } 7128f408ab6SDaniel Jurgens 7138f408ab6SDaniel Jurgens static int ib_security_change(struct notifier_block *nb, unsigned long event, 7148f408ab6SDaniel Jurgens void *lsm_data) 7158f408ab6SDaniel Jurgens { 7168f408ab6SDaniel Jurgens if (event != LSM_POLICY_CHANGE) 7178f408ab6SDaniel Jurgens return NOTIFY_DONE; 7188f408ab6SDaniel Jurgens 7198f408ab6SDaniel Jurgens schedule_work(&ib_policy_change_work); 720c66f6741SDaniel Jurgens ib_mad_agent_security_change(); 7218f408ab6SDaniel Jurgens 7228f408ab6SDaniel Jurgens return NOTIFY_OK; 7238f408ab6SDaniel Jurgens } 7248f408ab6SDaniel Jurgens 7254e0f7b90SParav Pandit static void compatdev_release(struct device *dev) 7264e0f7b90SParav Pandit { 7274e0f7b90SParav Pandit struct ib_core_device *cdev = 7284e0f7b90SParav Pandit container_of(dev, struct ib_core_device, dev); 7294e0f7b90SParav Pandit 7304e0f7b90SParav Pandit kfree(cdev); 7314e0f7b90SParav Pandit } 7324e0f7b90SParav Pandit 7334e0f7b90SParav Pandit static int add_one_compat_dev(struct ib_device *device, 7344e0f7b90SParav Pandit struct rdma_dev_net *rnet) 7354e0f7b90SParav Pandit { 7364e0f7b90SParav Pandit struct ib_core_device *cdev; 7374e0f7b90SParav Pandit int ret; 7384e0f7b90SParav Pandit 7394e0f7b90SParav Pandit /* 7404e0f7b90SParav Pandit * Create and add compat device in all namespaces other than where it 7414e0f7b90SParav Pandit * is currently bound to. 7424e0f7b90SParav Pandit */ 7434e0f7b90SParav Pandit if (net_eq(read_pnet(&rnet->net), 7444e0f7b90SParav Pandit read_pnet(&device->coredev.rdma_net))) 7454e0f7b90SParav Pandit return 0; 7464e0f7b90SParav Pandit 7474e0f7b90SParav Pandit /* 7484e0f7b90SParav Pandit * The first of init_net() or ib_register_device() to take the 7494e0f7b90SParav Pandit * compat_devs_mutex wins and gets to add the device. Others will wait 7504e0f7b90SParav Pandit * for completion here. 7514e0f7b90SParav Pandit */ 7524e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 7534e0f7b90SParav Pandit cdev = xa_load(&device->compat_devs, rnet->id); 7544e0f7b90SParav Pandit if (cdev) { 7554e0f7b90SParav Pandit ret = 0; 7564e0f7b90SParav Pandit goto done; 7574e0f7b90SParav Pandit } 7584e0f7b90SParav Pandit ret = xa_reserve(&device->compat_devs, rnet->id, GFP_KERNEL); 7594e0f7b90SParav Pandit if (ret) 7604e0f7b90SParav Pandit goto done; 7614e0f7b90SParav Pandit 7624e0f7b90SParav Pandit cdev = kzalloc(sizeof(*cdev), GFP_KERNEL); 7634e0f7b90SParav Pandit if (!cdev) { 7644e0f7b90SParav Pandit ret = -ENOMEM; 7654e0f7b90SParav Pandit goto cdev_err; 7664e0f7b90SParav Pandit } 7674e0f7b90SParav Pandit 7684e0f7b90SParav Pandit cdev->dev.parent = device->dev.parent; 7694e0f7b90SParav Pandit rdma_init_coredev(cdev, device, read_pnet(&rnet->net)); 7704e0f7b90SParav Pandit cdev->dev.release = compatdev_release; 7714e0f7b90SParav Pandit dev_set_name(&cdev->dev, "%s", dev_name(&device->dev)); 7724e0f7b90SParav Pandit 7734e0f7b90SParav Pandit ret = device_add(&cdev->dev); 7744e0f7b90SParav Pandit if (ret) 7754e0f7b90SParav Pandit goto add_err; 7765417783eSParav Pandit ret = ib_setup_port_attrs(cdev, false); 7775417783eSParav Pandit if (ret) 7785417783eSParav Pandit goto port_err; 7794e0f7b90SParav Pandit 7804e0f7b90SParav Pandit ret = xa_err(xa_store(&device->compat_devs, rnet->id, 7814e0f7b90SParav Pandit cdev, GFP_KERNEL)); 7824e0f7b90SParav Pandit if (ret) 7834e0f7b90SParav Pandit goto insert_err; 7844e0f7b90SParav Pandit 7854e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 7864e0f7b90SParav Pandit return 0; 7874e0f7b90SParav Pandit 7884e0f7b90SParav Pandit insert_err: 7895417783eSParav Pandit ib_free_port_attrs(cdev); 7905417783eSParav Pandit port_err: 7914e0f7b90SParav Pandit device_del(&cdev->dev); 7924e0f7b90SParav Pandit add_err: 7934e0f7b90SParav Pandit put_device(&cdev->dev); 7944e0f7b90SParav Pandit cdev_err: 7954e0f7b90SParav Pandit xa_release(&device->compat_devs, rnet->id); 7964e0f7b90SParav Pandit done: 7974e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 7984e0f7b90SParav Pandit return ret; 7994e0f7b90SParav Pandit } 8004e0f7b90SParav Pandit 8014e0f7b90SParav Pandit static void remove_one_compat_dev(struct ib_device *device, u32 id) 8024e0f7b90SParav Pandit { 8034e0f7b90SParav Pandit struct ib_core_device *cdev; 8044e0f7b90SParav Pandit 8054e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 8064e0f7b90SParav Pandit cdev = xa_erase(&device->compat_devs, id); 8074e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 8084e0f7b90SParav Pandit if (cdev) { 8095417783eSParav Pandit ib_free_port_attrs(cdev); 8104e0f7b90SParav Pandit device_del(&cdev->dev); 8114e0f7b90SParav Pandit put_device(&cdev->dev); 8124e0f7b90SParav Pandit } 8134e0f7b90SParav Pandit } 8144e0f7b90SParav Pandit 8154e0f7b90SParav Pandit static void remove_compat_devs(struct ib_device *device) 8164e0f7b90SParav Pandit { 8174e0f7b90SParav Pandit struct ib_core_device *cdev; 8184e0f7b90SParav Pandit unsigned long index; 8194e0f7b90SParav Pandit 8204e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) 8214e0f7b90SParav Pandit remove_one_compat_dev(device, index); 8224e0f7b90SParav Pandit } 8234e0f7b90SParav Pandit 8244e0f7b90SParav Pandit static int add_compat_devs(struct ib_device *device) 8254e0f7b90SParav Pandit { 8264e0f7b90SParav Pandit struct rdma_dev_net *rnet; 8274e0f7b90SParav Pandit unsigned long index; 8284e0f7b90SParav Pandit int ret = 0; 8294e0f7b90SParav Pandit 8304e0f7b90SParav Pandit down_read(&rdma_nets_rwsem); 8314e0f7b90SParav Pandit xa_for_each (&rdma_nets, index, rnet) { 8324e0f7b90SParav Pandit ret = add_one_compat_dev(device, rnet); 8334e0f7b90SParav Pandit if (ret) 8344e0f7b90SParav Pandit break; 8354e0f7b90SParav Pandit } 8364e0f7b90SParav Pandit up_read(&rdma_nets_rwsem); 8374e0f7b90SParav Pandit return ret; 8384e0f7b90SParav Pandit } 8394e0f7b90SParav Pandit 8404e0f7b90SParav Pandit static void rdma_dev_exit_net(struct net *net) 8414e0f7b90SParav Pandit { 8424e0f7b90SParav Pandit struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); 8434e0f7b90SParav Pandit struct ib_device *dev; 8444e0f7b90SParav Pandit unsigned long index; 8454e0f7b90SParav Pandit int ret; 8464e0f7b90SParav Pandit 8474e0f7b90SParav Pandit down_write(&rdma_nets_rwsem); 8484e0f7b90SParav Pandit /* 8494e0f7b90SParav Pandit * Prevent the ID from being re-used and hide the id from xa_for_each. 8504e0f7b90SParav Pandit */ 8514e0f7b90SParav Pandit ret = xa_err(xa_store(&rdma_nets, rnet->id, NULL, GFP_KERNEL)); 8524e0f7b90SParav Pandit WARN_ON(ret); 8534e0f7b90SParav Pandit up_write(&rdma_nets_rwsem); 8544e0f7b90SParav Pandit 8554e0f7b90SParav Pandit down_read(&devices_rwsem); 8564e0f7b90SParav Pandit xa_for_each (&devices, index, dev) { 8574e0f7b90SParav Pandit get_device(&dev->dev); 8584e0f7b90SParav Pandit /* 8594e0f7b90SParav Pandit * Release the devices_rwsem so that pontentially blocking 8604e0f7b90SParav Pandit * device_del, doesn't hold the devices_rwsem for too long. 8614e0f7b90SParav Pandit */ 8624e0f7b90SParav Pandit up_read(&devices_rwsem); 8634e0f7b90SParav Pandit 8644e0f7b90SParav Pandit remove_one_compat_dev(dev, rnet->id); 8654e0f7b90SParav Pandit 8664e0f7b90SParav Pandit put_device(&dev->dev); 8674e0f7b90SParav Pandit down_read(&devices_rwsem); 8684e0f7b90SParav Pandit } 8694e0f7b90SParav Pandit up_read(&devices_rwsem); 8704e0f7b90SParav Pandit 8714e0f7b90SParav Pandit xa_erase(&rdma_nets, rnet->id); 8724e0f7b90SParav Pandit } 8734e0f7b90SParav Pandit 8744e0f7b90SParav Pandit static __net_init int rdma_dev_init_net(struct net *net) 8754e0f7b90SParav Pandit { 8764e0f7b90SParav Pandit struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); 8774e0f7b90SParav Pandit unsigned long index; 8784e0f7b90SParav Pandit struct ib_device *dev; 8794e0f7b90SParav Pandit int ret; 8804e0f7b90SParav Pandit 8814e0f7b90SParav Pandit /* No need to create any compat devices in default init_net. */ 8824e0f7b90SParav Pandit if (net_eq(net, &init_net)) 8834e0f7b90SParav Pandit return 0; 8844e0f7b90SParav Pandit 8854e0f7b90SParav Pandit write_pnet(&rnet->net, net); 8864e0f7b90SParav Pandit 8874e0f7b90SParav Pandit ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b, GFP_KERNEL); 8884e0f7b90SParav Pandit if (ret) 8894e0f7b90SParav Pandit return ret; 8904e0f7b90SParav Pandit 8914e0f7b90SParav Pandit down_read(&devices_rwsem); 8924e0f7b90SParav Pandit xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 8934e0f7b90SParav Pandit ret = add_one_compat_dev(dev, rnet); 8944e0f7b90SParav Pandit if (ret) 8954e0f7b90SParav Pandit break; 8964e0f7b90SParav Pandit } 8974e0f7b90SParav Pandit up_read(&devices_rwsem); 8984e0f7b90SParav Pandit 8994e0f7b90SParav Pandit if (ret) 9004e0f7b90SParav Pandit rdma_dev_exit_net(net); 9014e0f7b90SParav Pandit 9024e0f7b90SParav Pandit return ret; 9034e0f7b90SParav Pandit } 9044e0f7b90SParav Pandit 905ecc82c53SLeon Romanovsky /* 906d0899892SJason Gunthorpe * Assign the unique string device name and the unique device index. This is 907d0899892SJason Gunthorpe * undone by ib_dealloc_device. 908ecc82c53SLeon Romanovsky */ 9090df91bb6SJason Gunthorpe static int assign_name(struct ib_device *device, const char *name) 9100df91bb6SJason Gunthorpe { 9110df91bb6SJason Gunthorpe static u32 last_id; 9120df91bb6SJason Gunthorpe int ret; 913ecc82c53SLeon Romanovsky 914921eab11SJason Gunthorpe down_write(&devices_rwsem); 9150df91bb6SJason Gunthorpe /* Assign a unique name to the device */ 9160df91bb6SJason Gunthorpe if (strchr(name, '%')) 9170df91bb6SJason Gunthorpe ret = alloc_name(device, name); 9180df91bb6SJason Gunthorpe else 9190df91bb6SJason Gunthorpe ret = dev_set_name(&device->dev, name); 9200df91bb6SJason Gunthorpe if (ret) 9210df91bb6SJason Gunthorpe goto out; 922ecc82c53SLeon Romanovsky 9230df91bb6SJason Gunthorpe if (__ib_device_get_by_name(dev_name(&device->dev))) { 9240df91bb6SJason Gunthorpe ret = -ENFILE; 9250df91bb6SJason Gunthorpe goto out; 926ecc82c53SLeon Romanovsky } 9270df91bb6SJason Gunthorpe strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); 9280df91bb6SJason Gunthorpe 929ea295481SLinus Torvalds ret = xa_alloc_cyclic(&devices, &device->index, device, xa_limit_31b, 930ea295481SLinus Torvalds &last_id, GFP_KERNEL); 931ea295481SLinus Torvalds if (ret > 0) 9320df91bb6SJason Gunthorpe ret = 0; 933921eab11SJason Gunthorpe 9340df91bb6SJason Gunthorpe out: 935921eab11SJason Gunthorpe up_write(&devices_rwsem); 9360df91bb6SJason Gunthorpe return ret; 9370df91bb6SJason Gunthorpe } 9380df91bb6SJason Gunthorpe 939548cb4fbSParav Pandit static void setup_dma_device(struct ib_device *device) 9401da177e4SLinus Torvalds { 94199db9494SBart Van Assche struct device *parent = device->dev.parent; 9421da177e4SLinus Torvalds 9430957c29fSBart Van Assche WARN_ON_ONCE(device->dma_device); 9440957c29fSBart Van Assche if (device->dev.dma_ops) { 9450957c29fSBart Van Assche /* 9460957c29fSBart Van Assche * The caller provided custom DMA operations. Copy the 9470957c29fSBart Van Assche * DMA-related fields that are used by e.g. dma_alloc_coherent() 9480957c29fSBart Van Assche * into device->dev. 9490957c29fSBart Van Assche */ 9500957c29fSBart Van Assche device->dma_device = &device->dev; 95102ee9da3SBart Van Assche if (!device->dev.dma_mask) { 95202ee9da3SBart Van Assche if (parent) 95399db9494SBart Van Assche device->dev.dma_mask = parent->dma_mask; 95402ee9da3SBart Van Assche else 95502ee9da3SBart Van Assche WARN_ON_ONCE(true); 95602ee9da3SBart Van Assche } 95702ee9da3SBart Van Assche if (!device->dev.coherent_dma_mask) { 95802ee9da3SBart Van Assche if (parent) 9590957c29fSBart Van Assche device->dev.coherent_dma_mask = 9600957c29fSBart Van Assche parent->coherent_dma_mask; 96102ee9da3SBart Van Assche else 96202ee9da3SBart Van Assche WARN_ON_ONCE(true); 96302ee9da3SBart Van Assche } 9640957c29fSBart Van Assche } else { 9650957c29fSBart Van Assche /* 9660957c29fSBart Van Assche * The caller did not provide custom DMA operations. Use the 9670957c29fSBart Van Assche * DMA mapping operations of the parent device. 9680957c29fSBart Van Assche */ 96902ee9da3SBart Van Assche WARN_ON_ONCE(!parent); 9700957c29fSBart Van Assche device->dma_device = parent; 9710957c29fSBart Van Assche } 972548cb4fbSParav Pandit } 973548cb4fbSParav Pandit 974921eab11SJason Gunthorpe /* 975921eab11SJason Gunthorpe * setup_device() allocates memory and sets up data that requires calling the 976921eab11SJason Gunthorpe * device ops, this is the only reason these actions are not done during 977921eab11SJason Gunthorpe * ib_alloc_device. It is undone by ib_dealloc_device(). 978921eab11SJason Gunthorpe */ 979548cb4fbSParav Pandit static int setup_device(struct ib_device *device) 980548cb4fbSParav Pandit { 981548cb4fbSParav Pandit struct ib_udata uhw = {.outlen = 0, .inlen = 0}; 982548cb4fbSParav Pandit int ret; 983548cb4fbSParav Pandit 984921eab11SJason Gunthorpe setup_dma_device(device); 985921eab11SJason Gunthorpe 986548cb4fbSParav Pandit ret = ib_device_check_mandatory(device); 987548cb4fbSParav Pandit if (ret) 988548cb4fbSParav Pandit return ret; 989548cb4fbSParav Pandit 9908ceb1357SJason Gunthorpe ret = setup_port_data(device); 991548cb4fbSParav Pandit if (ret) { 9928ceb1357SJason Gunthorpe dev_warn(&device->dev, "Couldn't create per-port data\n"); 993548cb4fbSParav Pandit return ret; 994548cb4fbSParav Pandit } 995548cb4fbSParav Pandit 996548cb4fbSParav Pandit memset(&device->attrs, 0, sizeof(device->attrs)); 9973023a1e9SKamal Heib ret = device->ops.query_device(device, &device->attrs, &uhw); 998548cb4fbSParav Pandit if (ret) { 999548cb4fbSParav Pandit dev_warn(&device->dev, 1000548cb4fbSParav Pandit "Couldn't query the device attributes\n"); 1001d45f89d5SJason Gunthorpe return ret; 1002548cb4fbSParav Pandit } 1003548cb4fbSParav Pandit 1004548cb4fbSParav Pandit return 0; 1005548cb4fbSParav Pandit } 1006548cb4fbSParav Pandit 1007921eab11SJason Gunthorpe static void disable_device(struct ib_device *device) 1008921eab11SJason Gunthorpe { 1009921eab11SJason Gunthorpe struct ib_client *client; 1010921eab11SJason Gunthorpe 1011921eab11SJason Gunthorpe WARN_ON(!refcount_read(&device->refcount)); 1012921eab11SJason Gunthorpe 1013921eab11SJason Gunthorpe down_write(&devices_rwsem); 1014921eab11SJason Gunthorpe xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); 1015921eab11SJason Gunthorpe up_write(&devices_rwsem); 1016921eab11SJason Gunthorpe 1017921eab11SJason Gunthorpe down_read(&clients_rwsem); 1018921eab11SJason Gunthorpe list_for_each_entry_reverse(client, &client_list, list) 1019921eab11SJason Gunthorpe remove_client_context(device, client->client_id); 1020921eab11SJason Gunthorpe up_read(&clients_rwsem); 1021921eab11SJason Gunthorpe 1022921eab11SJason Gunthorpe /* Pairs with refcount_set in enable_device */ 1023921eab11SJason Gunthorpe ib_device_put(device); 1024921eab11SJason Gunthorpe wait_for_completion(&device->unreg_completion); 1025c2261dd7SJason Gunthorpe 10264e0f7b90SParav Pandit /* 10274e0f7b90SParav Pandit * compat devices must be removed after device refcount drops to zero. 10284e0f7b90SParav Pandit * Otherwise init_net() may add more compatdevs after removing compat 10294e0f7b90SParav Pandit * devices and before device is disabled. 10304e0f7b90SParav Pandit */ 10314e0f7b90SParav Pandit remove_compat_devs(device); 10324e0f7b90SParav Pandit 1033c2261dd7SJason Gunthorpe /* Expedite removing unregistered pointers from the hash table */ 1034c2261dd7SJason Gunthorpe free_netdevs(device); 1035921eab11SJason Gunthorpe } 1036921eab11SJason Gunthorpe 1037921eab11SJason Gunthorpe /* 1038921eab11SJason Gunthorpe * An enabled device is visible to all clients and to all the public facing 1039d0899892SJason Gunthorpe * APIs that return a device pointer. This always returns with a new get, even 1040d0899892SJason Gunthorpe * if it fails. 1041921eab11SJason Gunthorpe */ 1042d0899892SJason Gunthorpe static int enable_device_and_get(struct ib_device *device) 1043921eab11SJason Gunthorpe { 1044921eab11SJason Gunthorpe struct ib_client *client; 1045921eab11SJason Gunthorpe unsigned long index; 1046d0899892SJason Gunthorpe int ret = 0; 1047921eab11SJason Gunthorpe 1048d0899892SJason Gunthorpe /* 1049d0899892SJason Gunthorpe * One ref belongs to the xa and the other belongs to this 1050d0899892SJason Gunthorpe * thread. This is needed to guard against parallel unregistration. 1051d0899892SJason Gunthorpe */ 1052d0899892SJason Gunthorpe refcount_set(&device->refcount, 2); 1053921eab11SJason Gunthorpe down_write(&devices_rwsem); 1054921eab11SJason Gunthorpe xa_set_mark(&devices, device->index, DEVICE_REGISTERED); 1055d0899892SJason Gunthorpe 1056d0899892SJason Gunthorpe /* 1057d0899892SJason Gunthorpe * By using downgrade_write() we ensure that no other thread can clear 1058d0899892SJason Gunthorpe * DEVICE_REGISTERED while we are completing the client setup. 1059d0899892SJason Gunthorpe */ 1060d0899892SJason Gunthorpe downgrade_write(&devices_rwsem); 1061921eab11SJason Gunthorpe 1062ca22354bSJason Gunthorpe if (device->ops.enable_driver) { 1063ca22354bSJason Gunthorpe ret = device->ops.enable_driver(device); 1064ca22354bSJason Gunthorpe if (ret) 1065ca22354bSJason Gunthorpe goto out; 1066ca22354bSJason Gunthorpe } 1067ca22354bSJason Gunthorpe 1068921eab11SJason Gunthorpe down_read(&clients_rwsem); 1069921eab11SJason Gunthorpe xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { 1070921eab11SJason Gunthorpe ret = add_client_context(device, client); 1071d0899892SJason Gunthorpe if (ret) 1072d0899892SJason Gunthorpe break; 1073d0899892SJason Gunthorpe } 1074921eab11SJason Gunthorpe up_read(&clients_rwsem); 10754e0f7b90SParav Pandit if (!ret) 10764e0f7b90SParav Pandit ret = add_compat_devs(device); 1077ca22354bSJason Gunthorpe out: 1078d0899892SJason Gunthorpe up_read(&devices_rwsem); 1079921eab11SJason Gunthorpe return ret; 1080921eab11SJason Gunthorpe } 1081921eab11SJason Gunthorpe 1082548cb4fbSParav Pandit /** 1083548cb4fbSParav Pandit * ib_register_device - Register an IB device with IB core 1084548cb4fbSParav Pandit * @device:Device to register 1085548cb4fbSParav Pandit * 1086548cb4fbSParav Pandit * Low-level drivers use ib_register_device() to register their 1087548cb4fbSParav Pandit * devices with the IB core. All registered clients will receive a 1088548cb4fbSParav Pandit * callback for each device that is added. @device must be allocated 1089548cb4fbSParav Pandit * with ib_alloc_device(). 1090d0899892SJason Gunthorpe * 1091d0899892SJason Gunthorpe * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() 1092d0899892SJason Gunthorpe * asynchronously then the device pointer may become freed as soon as this 1093d0899892SJason Gunthorpe * function returns. 1094548cb4fbSParav Pandit */ 1095ea4baf7fSParav Pandit int ib_register_device(struct ib_device *device, const char *name) 1096548cb4fbSParav Pandit { 1097548cb4fbSParav Pandit int ret; 10981da177e4SLinus Torvalds 10990df91bb6SJason Gunthorpe ret = assign_name(device, name); 1100e349f858SJason Gunthorpe if (ret) 1101921eab11SJason Gunthorpe return ret; 11021da177e4SLinus Torvalds 1103548cb4fbSParav Pandit ret = setup_device(device); 1104548cb4fbSParav Pandit if (ret) 1105d0899892SJason Gunthorpe return ret; 110603db3a2dSMatan Barak 1107d45f89d5SJason Gunthorpe ret = ib_cache_setup_one(device); 1108d45f89d5SJason Gunthorpe if (ret) { 1109d45f89d5SJason Gunthorpe dev_warn(&device->dev, 1110d45f89d5SJason Gunthorpe "Couldn't set up InfiniBand P_Key/GID cache\n"); 1111d0899892SJason Gunthorpe return ret; 1112d45f89d5SJason Gunthorpe } 1113d45f89d5SJason Gunthorpe 11147527a7b1SParav Pandit ib_device_register_rdmacg(device); 11153e153a93SIra Weiny 11165f8f5499SParav Pandit ret = device_add(&device->dev); 11175f8f5499SParav Pandit if (ret) 11185f8f5499SParav Pandit goto cg_cleanup; 11195f8f5499SParav Pandit 1120ea4baf7fSParav Pandit ret = ib_device_register_sysfs(device); 11211da177e4SLinus Torvalds if (ret) { 112243c7c851SJason Gunthorpe dev_warn(&device->dev, 112343c7c851SJason Gunthorpe "Couldn't register device with driver model\n"); 11245f8f5499SParav Pandit goto dev_cleanup; 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds 1127d0899892SJason Gunthorpe ret = enable_device_and_get(device); 1128d0899892SJason Gunthorpe if (ret) { 1129d0899892SJason Gunthorpe void (*dealloc_fn)(struct ib_device *); 1130d0899892SJason Gunthorpe 1131d0899892SJason Gunthorpe /* 1132d0899892SJason Gunthorpe * If we hit this error flow then we don't want to 1133d0899892SJason Gunthorpe * automatically dealloc the device since the caller is 1134d0899892SJason Gunthorpe * expected to call ib_dealloc_device() after 1135d0899892SJason Gunthorpe * ib_register_device() fails. This is tricky due to the 1136d0899892SJason Gunthorpe * possibility for a parallel unregistration along with this 1137d0899892SJason Gunthorpe * error flow. Since we have a refcount here we know any 1138d0899892SJason Gunthorpe * parallel flow is stopped in disable_device and will see the 1139d0899892SJason Gunthorpe * NULL pointers, causing the responsibility to 1140d0899892SJason Gunthorpe * ib_dealloc_device() to revert back to this thread. 1141d0899892SJason Gunthorpe */ 1142d0899892SJason Gunthorpe dealloc_fn = device->ops.dealloc_driver; 1143d0899892SJason Gunthorpe device->ops.dealloc_driver = NULL; 1144d0899892SJason Gunthorpe ib_device_put(device); 1145d0899892SJason Gunthorpe __ib_unregister_device(device); 1146d0899892SJason Gunthorpe device->ops.dealloc_driver = dealloc_fn; 1147d0899892SJason Gunthorpe return ret; 1148d0899892SJason Gunthorpe } 1149d0899892SJason Gunthorpe ib_device_put(device); 11501da177e4SLinus Torvalds 11514be3a4faSParav Pandit return 0; 11524be3a4faSParav Pandit 11535f8f5499SParav Pandit dev_cleanup: 11545f8f5499SParav Pandit device_del(&device->dev); 11552fb4f4eaSParav Pandit cg_cleanup: 11562fb4f4eaSParav Pandit ib_device_unregister_rdmacg(device); 1157d45f89d5SJason Gunthorpe ib_cache_cleanup_one(device); 11581da177e4SLinus Torvalds return ret; 11591da177e4SLinus Torvalds } 11601da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_device); 11611da177e4SLinus Torvalds 1162d0899892SJason Gunthorpe /* Callers must hold a get on the device. */ 1163d0899892SJason Gunthorpe static void __ib_unregister_device(struct ib_device *ib_dev) 1164d0899892SJason Gunthorpe { 1165d0899892SJason Gunthorpe /* 1166d0899892SJason Gunthorpe * We have a registration lock so that all the calls to unregister are 1167d0899892SJason Gunthorpe * fully fenced, once any unregister returns the device is truely 1168d0899892SJason Gunthorpe * unregistered even if multiple callers are unregistering it at the 1169d0899892SJason Gunthorpe * same time. This also interacts with the registration flow and 1170d0899892SJason Gunthorpe * provides sane semantics if register and unregister are racing. 1171d0899892SJason Gunthorpe */ 1172d0899892SJason Gunthorpe mutex_lock(&ib_dev->unregistration_lock); 1173d0899892SJason Gunthorpe if (!refcount_read(&ib_dev->refcount)) 1174d0899892SJason Gunthorpe goto out; 1175d0899892SJason Gunthorpe 1176d0899892SJason Gunthorpe disable_device(ib_dev); 1177d0899892SJason Gunthorpe ib_device_unregister_sysfs(ib_dev); 1178d0899892SJason Gunthorpe device_del(&ib_dev->dev); 1179d0899892SJason Gunthorpe ib_device_unregister_rdmacg(ib_dev); 1180d0899892SJason Gunthorpe ib_cache_cleanup_one(ib_dev); 1181d0899892SJason Gunthorpe 1182d0899892SJason Gunthorpe /* 1183d0899892SJason Gunthorpe * Drivers using the new flow may not call ib_dealloc_device except 1184d0899892SJason Gunthorpe * in error unwind prior to registration success. 1185d0899892SJason Gunthorpe */ 1186d0899892SJason Gunthorpe if (ib_dev->ops.dealloc_driver) { 1187d0899892SJason Gunthorpe WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); 1188d0899892SJason Gunthorpe ib_dealloc_device(ib_dev); 1189d0899892SJason Gunthorpe } 1190d0899892SJason Gunthorpe out: 1191d0899892SJason Gunthorpe mutex_unlock(&ib_dev->unregistration_lock); 1192d0899892SJason Gunthorpe } 1193d0899892SJason Gunthorpe 11941da177e4SLinus Torvalds /** 11951da177e4SLinus Torvalds * ib_unregister_device - Unregister an IB device 1196d0899892SJason Gunthorpe * @device: The device to unregister 11971da177e4SLinus Torvalds * 11981da177e4SLinus Torvalds * Unregister an IB device. All clients will receive a remove callback. 1199d0899892SJason Gunthorpe * 1200d0899892SJason Gunthorpe * Callers should call this routine only once, and protect against races with 1201d0899892SJason Gunthorpe * registration. Typically it should only be called as part of a remove 1202d0899892SJason Gunthorpe * callback in an implementation of driver core's struct device_driver and 1203d0899892SJason Gunthorpe * related. 1204d0899892SJason Gunthorpe * 1205d0899892SJason Gunthorpe * If ops.dealloc_driver is used then ib_dev will be freed upon return from 1206d0899892SJason Gunthorpe * this function. 12071da177e4SLinus Torvalds */ 1208d0899892SJason Gunthorpe void ib_unregister_device(struct ib_device *ib_dev) 12091da177e4SLinus Torvalds { 1210d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1211d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1212d0899892SJason Gunthorpe put_device(&ib_dev->dev); 12131da177e4SLinus Torvalds } 12141da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_device); 12151da177e4SLinus Torvalds 1216d0899892SJason Gunthorpe /** 1217d0899892SJason Gunthorpe * ib_unregister_device_and_put - Unregister a device while holding a 'get' 1218d0899892SJason Gunthorpe * device: The device to unregister 1219d0899892SJason Gunthorpe * 1220d0899892SJason Gunthorpe * This is the same as ib_unregister_device(), except it includes an internal 1221d0899892SJason Gunthorpe * ib_device_put() that should match a 'get' obtained by the caller. 1222d0899892SJason Gunthorpe * 1223d0899892SJason Gunthorpe * It is safe to call this routine concurrently from multiple threads while 1224d0899892SJason Gunthorpe * holding the 'get'. When the function returns the device is fully 1225d0899892SJason Gunthorpe * unregistered. 1226d0899892SJason Gunthorpe * 1227d0899892SJason Gunthorpe * Drivers using this flow MUST use the driver_unregister callback to clean up 1228d0899892SJason Gunthorpe * their resources associated with the device and dealloc it. 1229d0899892SJason Gunthorpe */ 1230d0899892SJason Gunthorpe void ib_unregister_device_and_put(struct ib_device *ib_dev) 1231d0899892SJason Gunthorpe { 1232d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1233d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1234d0899892SJason Gunthorpe ib_device_put(ib_dev); 1235d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1236d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1237d0899892SJason Gunthorpe } 1238d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_and_put); 1239d0899892SJason Gunthorpe 1240d0899892SJason Gunthorpe /** 1241d0899892SJason Gunthorpe * ib_unregister_driver - Unregister all IB devices for a driver 1242d0899892SJason Gunthorpe * @driver_id: The driver to unregister 1243d0899892SJason Gunthorpe * 1244d0899892SJason Gunthorpe * This implements a fence for device unregistration. It only returns once all 1245d0899892SJason Gunthorpe * devices associated with the driver_id have fully completed their 1246d0899892SJason Gunthorpe * unregistration and returned from ib_unregister_device*(). 1247d0899892SJason Gunthorpe * 1248d0899892SJason Gunthorpe * If device's are not yet unregistered it goes ahead and starts unregistering 1249d0899892SJason Gunthorpe * them. 1250d0899892SJason Gunthorpe * 1251d0899892SJason Gunthorpe * This does not block creation of new devices with the given driver_id, that 1252d0899892SJason Gunthorpe * is the responsibility of the caller. 1253d0899892SJason Gunthorpe */ 1254d0899892SJason Gunthorpe void ib_unregister_driver(enum rdma_driver_id driver_id) 1255d0899892SJason Gunthorpe { 1256d0899892SJason Gunthorpe struct ib_device *ib_dev; 1257d0899892SJason Gunthorpe unsigned long index; 1258d0899892SJason Gunthorpe 1259d0899892SJason Gunthorpe down_read(&devices_rwsem); 1260d0899892SJason Gunthorpe xa_for_each (&devices, index, ib_dev) { 1261d0899892SJason Gunthorpe if (ib_dev->driver_id != driver_id) 1262d0899892SJason Gunthorpe continue; 1263d0899892SJason Gunthorpe 1264d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1265d0899892SJason Gunthorpe up_read(&devices_rwsem); 1266d0899892SJason Gunthorpe 1267d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1268d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1269d0899892SJason Gunthorpe 1270d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1271d0899892SJason Gunthorpe down_read(&devices_rwsem); 1272d0899892SJason Gunthorpe } 1273d0899892SJason Gunthorpe up_read(&devices_rwsem); 1274d0899892SJason Gunthorpe } 1275d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_driver); 1276d0899892SJason Gunthorpe 1277d0899892SJason Gunthorpe static void ib_unregister_work(struct work_struct *work) 1278d0899892SJason Gunthorpe { 1279d0899892SJason Gunthorpe struct ib_device *ib_dev = 1280d0899892SJason Gunthorpe container_of(work, struct ib_device, unregistration_work); 1281d0899892SJason Gunthorpe 1282d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1283d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1284d0899892SJason Gunthorpe } 1285d0899892SJason Gunthorpe 1286d0899892SJason Gunthorpe /** 1287d0899892SJason Gunthorpe * ib_unregister_device_queued - Unregister a device using a work queue 1288d0899892SJason Gunthorpe * device: The device to unregister 1289d0899892SJason Gunthorpe * 1290d0899892SJason Gunthorpe * This schedules an asynchronous unregistration using a WQ for the device. A 1291d0899892SJason Gunthorpe * driver should use this to avoid holding locks while doing unregistration, 1292d0899892SJason Gunthorpe * such as holding the RTNL lock. 1293d0899892SJason Gunthorpe * 1294d0899892SJason Gunthorpe * Drivers using this API must use ib_unregister_driver before module unload 1295d0899892SJason Gunthorpe * to ensure that all scheduled unregistrations have completed. 1296d0899892SJason Gunthorpe */ 1297d0899892SJason Gunthorpe void ib_unregister_device_queued(struct ib_device *ib_dev) 1298d0899892SJason Gunthorpe { 1299d0899892SJason Gunthorpe WARN_ON(!refcount_read(&ib_dev->refcount)); 1300d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1301d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1302d0899892SJason Gunthorpe if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) 1303d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1304d0899892SJason Gunthorpe } 1305d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_queued); 1306d0899892SJason Gunthorpe 13074e0f7b90SParav Pandit static struct pernet_operations rdma_dev_net_ops = { 13084e0f7b90SParav Pandit .init = rdma_dev_init_net, 13094e0f7b90SParav Pandit .exit = rdma_dev_exit_net, 13104e0f7b90SParav Pandit .id = &rdma_dev_net_id, 13114e0f7b90SParav Pandit .size = sizeof(struct rdma_dev_net), 13124e0f7b90SParav Pandit }; 13134e0f7b90SParav Pandit 1314e59178d8SJason Gunthorpe static int assign_client_id(struct ib_client *client) 1315e59178d8SJason Gunthorpe { 1316e59178d8SJason Gunthorpe int ret; 1317e59178d8SJason Gunthorpe 1318921eab11SJason Gunthorpe down_write(&clients_rwsem); 1319e59178d8SJason Gunthorpe /* 1320e59178d8SJason Gunthorpe * The add/remove callbacks must be called in FIFO/LIFO order. To 1321e59178d8SJason Gunthorpe * achieve this we assign client_ids so they are sorted in 1322e59178d8SJason Gunthorpe * registration order, and retain a linked list we can reverse iterate 1323e59178d8SJason Gunthorpe * to get the LIFO order. The extra linked list can go away if xarray 1324e59178d8SJason Gunthorpe * learns to reverse iterate. 1325e59178d8SJason Gunthorpe */ 1326ea295481SLinus Torvalds if (list_empty(&client_list)) { 1327e59178d8SJason Gunthorpe client->client_id = 0; 1328ea295481SLinus Torvalds } else { 1329ea295481SLinus Torvalds struct ib_client *last; 1330ea295481SLinus Torvalds 1331ea295481SLinus Torvalds last = list_last_entry(&client_list, struct ib_client, list); 1332ea295481SLinus Torvalds client->client_id = last->client_id + 1; 1333ea295481SLinus Torvalds } 1334ea295481SLinus Torvalds ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); 1335e59178d8SJason Gunthorpe if (ret) 1336e59178d8SJason Gunthorpe goto out; 1337e59178d8SJason Gunthorpe 1338921eab11SJason Gunthorpe xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); 1339921eab11SJason Gunthorpe list_add_tail(&client->list, &client_list); 1340921eab11SJason Gunthorpe 1341e59178d8SJason Gunthorpe out: 1342921eab11SJason Gunthorpe up_write(&clients_rwsem); 1343e59178d8SJason Gunthorpe return ret; 1344e59178d8SJason Gunthorpe } 1345e59178d8SJason Gunthorpe 13461da177e4SLinus Torvalds /** 13471da177e4SLinus Torvalds * ib_register_client - Register an IB client 13481da177e4SLinus Torvalds * @client:Client to register 13491da177e4SLinus Torvalds * 13501da177e4SLinus Torvalds * Upper level users of the IB drivers can use ib_register_client() to 13511da177e4SLinus Torvalds * register callbacks for IB device addition and removal. When an IB 13521da177e4SLinus Torvalds * device is added, each registered client's add method will be called 13531da177e4SLinus Torvalds * (in the order the clients were registered), and when a device is 13541da177e4SLinus Torvalds * removed, each client's remove method will be called (in the reverse 13551da177e4SLinus Torvalds * order that clients were registered). In addition, when 13561da177e4SLinus Torvalds * ib_register_client() is called, the client will receive an add 13571da177e4SLinus Torvalds * callback for all devices already registered. 13581da177e4SLinus Torvalds */ 13591da177e4SLinus Torvalds int ib_register_client(struct ib_client *client) 13601da177e4SLinus Torvalds { 13611da177e4SLinus Torvalds struct ib_device *device; 13620df91bb6SJason Gunthorpe unsigned long index; 1363e59178d8SJason Gunthorpe int ret; 13641da177e4SLinus Torvalds 1365e59178d8SJason Gunthorpe ret = assign_client_id(client); 1366921eab11SJason Gunthorpe if (ret) 1367921eab11SJason Gunthorpe return ret; 1368921eab11SJason Gunthorpe 1369921eab11SJason Gunthorpe down_read(&devices_rwsem); 1370921eab11SJason Gunthorpe xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { 1371921eab11SJason Gunthorpe ret = add_client_context(device, client); 1372e59178d8SJason Gunthorpe if (ret) { 1373921eab11SJason Gunthorpe up_read(&devices_rwsem); 1374921eab11SJason Gunthorpe ib_unregister_client(client); 1375e59178d8SJason Gunthorpe return ret; 1376e59178d8SJason Gunthorpe } 1377921eab11SJason Gunthorpe } 1378921eab11SJason Gunthorpe up_read(&devices_rwsem); 13791da177e4SLinus Torvalds return 0; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_client); 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds /** 13841da177e4SLinus Torvalds * ib_unregister_client - Unregister an IB client 13851da177e4SLinus Torvalds * @client:Client to unregister 13861da177e4SLinus Torvalds * 13871da177e4SLinus Torvalds * Upper level users use ib_unregister_client() to remove their client 13881da177e4SLinus Torvalds * registration. When ib_unregister_client() is called, the client 13891da177e4SLinus Torvalds * will receive a remove callback for each IB device still registered. 1390921eab11SJason Gunthorpe * 1391921eab11SJason Gunthorpe * This is a full fence, once it returns no client callbacks will be called, 1392921eab11SJason Gunthorpe * or are running in another thread. 13931da177e4SLinus Torvalds */ 13941da177e4SLinus Torvalds void ib_unregister_client(struct ib_client *client) 13951da177e4SLinus Torvalds { 13961da177e4SLinus Torvalds struct ib_device *device; 13970df91bb6SJason Gunthorpe unsigned long index; 13981da177e4SLinus Torvalds 1399921eab11SJason Gunthorpe down_write(&clients_rwsem); 1400e59178d8SJason Gunthorpe xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); 1401921eab11SJason Gunthorpe up_write(&clients_rwsem); 1402921eab11SJason Gunthorpe /* 1403921eab11SJason Gunthorpe * Every device still known must be serialized to make sure we are 1404921eab11SJason Gunthorpe * done with the client callbacks before we return. 1405921eab11SJason Gunthorpe */ 1406921eab11SJason Gunthorpe down_read(&devices_rwsem); 1407921eab11SJason Gunthorpe xa_for_each (&devices, index, device) 1408921eab11SJason Gunthorpe remove_client_context(device, client->client_id); 1409921eab11SJason Gunthorpe up_read(&devices_rwsem); 14105aa44bb9SHaggai Eran 1411921eab11SJason Gunthorpe down_write(&clients_rwsem); 1412e59178d8SJason Gunthorpe list_del(&client->list); 1413e59178d8SJason Gunthorpe xa_erase(&clients, client->client_id); 1414921eab11SJason Gunthorpe up_write(&clients_rwsem); 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_client); 14171da177e4SLinus Torvalds 14181da177e4SLinus Torvalds /** 14199cd330d3SKrishna Kumar * ib_set_client_data - Set IB client context 14201da177e4SLinus Torvalds * @device:Device to set context for 14211da177e4SLinus Torvalds * @client:Client to set context for 14221da177e4SLinus Torvalds * @data:Context to set 14231da177e4SLinus Torvalds * 14240df91bb6SJason Gunthorpe * ib_set_client_data() sets client context data that can be retrieved with 14250df91bb6SJason Gunthorpe * ib_get_client_data(). This can only be called while the client is 14260df91bb6SJason Gunthorpe * registered to the device, once the ib_client remove() callback returns this 14270df91bb6SJason Gunthorpe * cannot be called. 14281da177e4SLinus Torvalds */ 14291da177e4SLinus Torvalds void ib_set_client_data(struct ib_device *device, struct ib_client *client, 14301da177e4SLinus Torvalds void *data) 14311da177e4SLinus Torvalds { 14320df91bb6SJason Gunthorpe void *rc; 14331da177e4SLinus Torvalds 14340df91bb6SJason Gunthorpe if (WARN_ON(IS_ERR(data))) 14350df91bb6SJason Gunthorpe data = NULL; 14361da177e4SLinus Torvalds 14370df91bb6SJason Gunthorpe rc = xa_store(&device->client_data, client->client_id, data, 14380df91bb6SJason Gunthorpe GFP_KERNEL); 14390df91bb6SJason Gunthorpe WARN_ON(xa_is_err(rc)); 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds EXPORT_SYMBOL(ib_set_client_data); 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds /** 14441da177e4SLinus Torvalds * ib_register_event_handler - Register an IB event handler 14451da177e4SLinus Torvalds * @event_handler:Handler to register 14461da177e4SLinus Torvalds * 14471da177e4SLinus Torvalds * ib_register_event_handler() registers an event handler that will be 14481da177e4SLinus Torvalds * called back when asynchronous IB events occur (as defined in 14491da177e4SLinus Torvalds * chapter 11 of the InfiniBand Architecture Specification). This 14501da177e4SLinus Torvalds * callback may occur in interrupt context. 14511da177e4SLinus Torvalds */ 1452dcc9881eSLeon Romanovsky void ib_register_event_handler(struct ib_event_handler *event_handler) 14531da177e4SLinus Torvalds { 14541da177e4SLinus Torvalds unsigned long flags; 14551da177e4SLinus Torvalds 14561da177e4SLinus Torvalds spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 14571da177e4SLinus Torvalds list_add_tail(&event_handler->list, 14581da177e4SLinus Torvalds &event_handler->device->event_handler_list); 14591da177e4SLinus Torvalds spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 14601da177e4SLinus Torvalds } 14611da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_event_handler); 14621da177e4SLinus Torvalds 14631da177e4SLinus Torvalds /** 14641da177e4SLinus Torvalds * ib_unregister_event_handler - Unregister an event handler 14651da177e4SLinus Torvalds * @event_handler:Handler to unregister 14661da177e4SLinus Torvalds * 14671da177e4SLinus Torvalds * Unregister an event handler registered with 14681da177e4SLinus Torvalds * ib_register_event_handler(). 14691da177e4SLinus Torvalds */ 1470dcc9881eSLeon Romanovsky void ib_unregister_event_handler(struct ib_event_handler *event_handler) 14711da177e4SLinus Torvalds { 14721da177e4SLinus Torvalds unsigned long flags; 14731da177e4SLinus Torvalds 14741da177e4SLinus Torvalds spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 14751da177e4SLinus Torvalds list_del(&event_handler->list); 14761da177e4SLinus Torvalds spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 14771da177e4SLinus Torvalds } 14781da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_event_handler); 14791da177e4SLinus Torvalds 14801da177e4SLinus Torvalds /** 14811da177e4SLinus Torvalds * ib_dispatch_event - Dispatch an asynchronous event 14821da177e4SLinus Torvalds * @event:Event to dispatch 14831da177e4SLinus Torvalds * 14841da177e4SLinus Torvalds * Low-level drivers must call ib_dispatch_event() to dispatch the 14851da177e4SLinus Torvalds * event to all registered event handlers when an asynchronous event 14861da177e4SLinus Torvalds * occurs. 14871da177e4SLinus Torvalds */ 14881da177e4SLinus Torvalds void ib_dispatch_event(struct ib_event *event) 14891da177e4SLinus Torvalds { 14901da177e4SLinus Torvalds unsigned long flags; 14911da177e4SLinus Torvalds struct ib_event_handler *handler; 14921da177e4SLinus Torvalds 14931da177e4SLinus Torvalds spin_lock_irqsave(&event->device->event_handler_lock, flags); 14941da177e4SLinus Torvalds 14951da177e4SLinus Torvalds list_for_each_entry(handler, &event->device->event_handler_list, list) 14961da177e4SLinus Torvalds handler->handler(handler, event); 14971da177e4SLinus Torvalds 14981da177e4SLinus Torvalds spin_unlock_irqrestore(&event->device->event_handler_lock, flags); 14991da177e4SLinus Torvalds } 15001da177e4SLinus Torvalds EXPORT_SYMBOL(ib_dispatch_event); 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds /** 15031da177e4SLinus Torvalds * ib_query_port - Query IB port attributes 15041da177e4SLinus Torvalds * @device:Device to query 15051da177e4SLinus Torvalds * @port_num:Port number to query 15061da177e4SLinus Torvalds * @port_attr:Port attributes 15071da177e4SLinus Torvalds * 15081da177e4SLinus Torvalds * ib_query_port() returns the attributes of a port through the 15091da177e4SLinus Torvalds * @port_attr pointer. 15101da177e4SLinus Torvalds */ 15111da177e4SLinus Torvalds int ib_query_port(struct ib_device *device, 15121da177e4SLinus Torvalds u8 port_num, 15131da177e4SLinus Torvalds struct ib_port_attr *port_attr) 15141da177e4SLinus Torvalds { 1515fad61ad4SEli Cohen union ib_gid gid; 1516fad61ad4SEli Cohen int err; 1517fad61ad4SEli Cohen 151824dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 1519116c0074SRoland Dreier return -EINVAL; 1520116c0074SRoland Dreier 1521fad61ad4SEli Cohen memset(port_attr, 0, sizeof(*port_attr)); 15223023a1e9SKamal Heib err = device->ops.query_port(device, port_num, port_attr); 1523fad61ad4SEli Cohen if (err || port_attr->subnet_prefix) 1524fad61ad4SEli Cohen return err; 1525fad61ad4SEli Cohen 1526d7012467SEli Cohen if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND) 1527d7012467SEli Cohen return 0; 1528d7012467SEli Cohen 15293023a1e9SKamal Heib err = device->ops.query_gid(device, port_num, 0, &gid); 1530fad61ad4SEli Cohen if (err) 1531fad61ad4SEli Cohen return err; 1532fad61ad4SEli Cohen 1533fad61ad4SEli Cohen port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); 1534fad61ad4SEli Cohen return 0; 15351da177e4SLinus Torvalds } 15361da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_port); 15371da177e4SLinus Torvalds 1538324e227eSJason Gunthorpe static void add_ndev_hash(struct ib_port_data *pdata) 1539324e227eSJason Gunthorpe { 1540324e227eSJason Gunthorpe unsigned long flags; 1541324e227eSJason Gunthorpe 1542324e227eSJason Gunthorpe might_sleep(); 1543324e227eSJason Gunthorpe 1544324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 1545324e227eSJason Gunthorpe if (hash_hashed(&pdata->ndev_hash_link)) { 1546324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 1547324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 1548324e227eSJason Gunthorpe /* 1549324e227eSJason Gunthorpe * We cannot do hash_add_rcu after a hash_del_rcu until the 1550324e227eSJason Gunthorpe * grace period 1551324e227eSJason Gunthorpe */ 1552324e227eSJason Gunthorpe synchronize_rcu(); 1553324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 1554324e227eSJason Gunthorpe } 1555324e227eSJason Gunthorpe if (pdata->netdev) 1556324e227eSJason Gunthorpe hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, 1557324e227eSJason Gunthorpe (uintptr_t)pdata->netdev); 1558324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 1559324e227eSJason Gunthorpe } 1560324e227eSJason Gunthorpe 15611da177e4SLinus Torvalds /** 1562c2261dd7SJason Gunthorpe * ib_device_set_netdev - Associate the ib_dev with an underlying net_device 1563c2261dd7SJason Gunthorpe * @ib_dev: Device to modify 1564c2261dd7SJason Gunthorpe * @ndev: net_device to affiliate, may be NULL 1565c2261dd7SJason Gunthorpe * @port: IB port the net_device is connected to 1566c2261dd7SJason Gunthorpe * 1567c2261dd7SJason Gunthorpe * Drivers should use this to link the ib_device to a netdev so the netdev 1568c2261dd7SJason Gunthorpe * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be 1569c2261dd7SJason Gunthorpe * affiliated with any port. 1570c2261dd7SJason Gunthorpe * 1571c2261dd7SJason Gunthorpe * The caller must ensure that the given ndev is not unregistered or 1572c2261dd7SJason Gunthorpe * unregistering, and that either the ib_device is unregistered or 1573c2261dd7SJason Gunthorpe * ib_device_set_netdev() is called with NULL when the ndev sends a 1574c2261dd7SJason Gunthorpe * NETDEV_UNREGISTER event. 1575c2261dd7SJason Gunthorpe */ 1576c2261dd7SJason Gunthorpe int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 1577c2261dd7SJason Gunthorpe unsigned int port) 1578c2261dd7SJason Gunthorpe { 1579c2261dd7SJason Gunthorpe struct net_device *old_ndev; 1580c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 1581c2261dd7SJason Gunthorpe unsigned long flags; 1582c2261dd7SJason Gunthorpe int ret; 1583c2261dd7SJason Gunthorpe 1584c2261dd7SJason Gunthorpe /* 1585c2261dd7SJason Gunthorpe * Drivers wish to call this before ib_register_driver, so we have to 1586c2261dd7SJason Gunthorpe * setup the port data early. 1587c2261dd7SJason Gunthorpe */ 1588c2261dd7SJason Gunthorpe ret = alloc_port_data(ib_dev); 1589c2261dd7SJason Gunthorpe if (ret) 1590c2261dd7SJason Gunthorpe return ret; 1591c2261dd7SJason Gunthorpe 1592c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 1593c2261dd7SJason Gunthorpe return -EINVAL; 1594c2261dd7SJason Gunthorpe 1595c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 1596c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 1597324e227eSJason Gunthorpe old_ndev = rcu_dereference_protected( 1598324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 1599324e227eSJason Gunthorpe if (old_ndev == ndev) { 1600c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 1601c2261dd7SJason Gunthorpe return 0; 1602c2261dd7SJason Gunthorpe } 1603c2261dd7SJason Gunthorpe 1604c2261dd7SJason Gunthorpe if (ndev) 1605c2261dd7SJason Gunthorpe dev_hold(ndev); 1606324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, ndev); 1607c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 1608c2261dd7SJason Gunthorpe 1609324e227eSJason Gunthorpe add_ndev_hash(pdata); 1610c2261dd7SJason Gunthorpe if (old_ndev) 1611c2261dd7SJason Gunthorpe dev_put(old_ndev); 1612c2261dd7SJason Gunthorpe 1613c2261dd7SJason Gunthorpe return 0; 1614c2261dd7SJason Gunthorpe } 1615c2261dd7SJason Gunthorpe EXPORT_SYMBOL(ib_device_set_netdev); 1616c2261dd7SJason Gunthorpe 1617c2261dd7SJason Gunthorpe static void free_netdevs(struct ib_device *ib_dev) 1618c2261dd7SJason Gunthorpe { 1619c2261dd7SJason Gunthorpe unsigned long flags; 1620c2261dd7SJason Gunthorpe unsigned int port; 1621c2261dd7SJason Gunthorpe 1622c2261dd7SJason Gunthorpe rdma_for_each_port (ib_dev, port) { 1623c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &ib_dev->port_data[port]; 1624324e227eSJason Gunthorpe struct net_device *ndev; 1625c2261dd7SJason Gunthorpe 1626c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 1627324e227eSJason Gunthorpe ndev = rcu_dereference_protected( 1628324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 1629324e227eSJason Gunthorpe if (ndev) { 1630324e227eSJason Gunthorpe spin_lock(&ndev_hash_lock); 1631324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 1632324e227eSJason Gunthorpe spin_unlock(&ndev_hash_lock); 1633324e227eSJason Gunthorpe 1634324e227eSJason Gunthorpe /* 1635324e227eSJason Gunthorpe * If this is the last dev_put there is still a 1636324e227eSJason Gunthorpe * synchronize_rcu before the netdev is kfreed, so we 1637324e227eSJason Gunthorpe * can continue to rely on unlocked pointer 1638324e227eSJason Gunthorpe * comparisons after the put 1639324e227eSJason Gunthorpe */ 1640324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, NULL); 1641324e227eSJason Gunthorpe dev_put(ndev); 1642c2261dd7SJason Gunthorpe } 1643c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 1644c2261dd7SJason Gunthorpe } 1645c2261dd7SJason Gunthorpe } 1646c2261dd7SJason Gunthorpe 1647c2261dd7SJason Gunthorpe struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, 1648c2261dd7SJason Gunthorpe unsigned int port) 1649c2261dd7SJason Gunthorpe { 1650c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 1651c2261dd7SJason Gunthorpe struct net_device *res; 1652c2261dd7SJason Gunthorpe 1653c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 1654c2261dd7SJason Gunthorpe return NULL; 1655c2261dd7SJason Gunthorpe 1656c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 1657c2261dd7SJason Gunthorpe 1658c2261dd7SJason Gunthorpe /* 1659c2261dd7SJason Gunthorpe * New drivers should use ib_device_set_netdev() not the legacy 1660c2261dd7SJason Gunthorpe * get_netdev(). 1661c2261dd7SJason Gunthorpe */ 1662c2261dd7SJason Gunthorpe if (ib_dev->ops.get_netdev) 1663c2261dd7SJason Gunthorpe res = ib_dev->ops.get_netdev(ib_dev, port); 1664c2261dd7SJason Gunthorpe else { 1665c2261dd7SJason Gunthorpe spin_lock(&pdata->netdev_lock); 1666324e227eSJason Gunthorpe res = rcu_dereference_protected( 1667324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 1668c2261dd7SJason Gunthorpe if (res) 1669c2261dd7SJason Gunthorpe dev_hold(res); 1670c2261dd7SJason Gunthorpe spin_unlock(&pdata->netdev_lock); 1671c2261dd7SJason Gunthorpe } 1672c2261dd7SJason Gunthorpe 1673c2261dd7SJason Gunthorpe /* 1674c2261dd7SJason Gunthorpe * If we are starting to unregister expedite things by preventing 1675c2261dd7SJason Gunthorpe * propagation of an unregistering netdev. 1676c2261dd7SJason Gunthorpe */ 1677c2261dd7SJason Gunthorpe if (res && res->reg_state != NETREG_REGISTERED) { 1678c2261dd7SJason Gunthorpe dev_put(res); 1679c2261dd7SJason Gunthorpe return NULL; 1680c2261dd7SJason Gunthorpe } 1681c2261dd7SJason Gunthorpe 1682c2261dd7SJason Gunthorpe return res; 1683c2261dd7SJason Gunthorpe } 1684c2261dd7SJason Gunthorpe 1685c2261dd7SJason Gunthorpe /** 1686324e227eSJason Gunthorpe * ib_device_get_by_netdev - Find an IB device associated with a netdev 1687324e227eSJason Gunthorpe * @ndev: netdev to locate 1688324e227eSJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 1689324e227eSJason Gunthorpe * 1690324e227eSJason Gunthorpe * Find and hold an ib_device that is associated with a netdev via 1691324e227eSJason Gunthorpe * ib_device_set_netdev(). The caller must call ib_device_put() on the 1692324e227eSJason Gunthorpe * returned pointer. 1693324e227eSJason Gunthorpe */ 1694324e227eSJason Gunthorpe struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 1695324e227eSJason Gunthorpe enum rdma_driver_id driver_id) 1696324e227eSJason Gunthorpe { 1697324e227eSJason Gunthorpe struct ib_device *res = NULL; 1698324e227eSJason Gunthorpe struct ib_port_data *cur; 1699324e227eSJason Gunthorpe 1700324e227eSJason Gunthorpe rcu_read_lock(); 1701324e227eSJason Gunthorpe hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, 1702324e227eSJason Gunthorpe (uintptr_t)ndev) { 1703324e227eSJason Gunthorpe if (rcu_access_pointer(cur->netdev) == ndev && 1704324e227eSJason Gunthorpe (driver_id == RDMA_DRIVER_UNKNOWN || 1705324e227eSJason Gunthorpe cur->ib_dev->driver_id == driver_id) && 1706324e227eSJason Gunthorpe ib_device_try_get(cur->ib_dev)) { 1707324e227eSJason Gunthorpe res = cur->ib_dev; 1708324e227eSJason Gunthorpe break; 1709324e227eSJason Gunthorpe } 1710324e227eSJason Gunthorpe } 1711324e227eSJason Gunthorpe rcu_read_unlock(); 1712324e227eSJason Gunthorpe 1713324e227eSJason Gunthorpe return res; 1714324e227eSJason Gunthorpe } 1715324e227eSJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_netdev); 1716324e227eSJason Gunthorpe 1717324e227eSJason Gunthorpe /** 171803db3a2dSMatan Barak * ib_enum_roce_netdev - enumerate all RoCE ports 171903db3a2dSMatan Barak * @ib_dev : IB device we want to query 172003db3a2dSMatan Barak * @filter: Should we call the callback? 172103db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 172203db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 172303db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 172403db3a2dSMatan Barak * 172503db3a2dSMatan Barak * Enumerates all of the physical RoCE ports of ib_dev 172603db3a2dSMatan Barak * which are related to netdevice and calls callback() on each 172703db3a2dSMatan Barak * device for which filter() function returns non zero. 172803db3a2dSMatan Barak */ 172903db3a2dSMatan Barak void ib_enum_roce_netdev(struct ib_device *ib_dev, 173003db3a2dSMatan Barak roce_netdev_filter filter, 173103db3a2dSMatan Barak void *filter_cookie, 173203db3a2dSMatan Barak roce_netdev_callback cb, 173303db3a2dSMatan Barak void *cookie) 173403db3a2dSMatan Barak { 1735ea1075edSJason Gunthorpe unsigned int port; 173603db3a2dSMatan Barak 1737ea1075edSJason Gunthorpe rdma_for_each_port (ib_dev, port) 173803db3a2dSMatan Barak if (rdma_protocol_roce(ib_dev, port)) { 1739c2261dd7SJason Gunthorpe struct net_device *idev = 1740c2261dd7SJason Gunthorpe ib_device_get_netdev(ib_dev, port); 174103db3a2dSMatan Barak 174203db3a2dSMatan Barak if (filter(ib_dev, port, idev, filter_cookie)) 174303db3a2dSMatan Barak cb(ib_dev, port, idev, cookie); 174403db3a2dSMatan Barak 174503db3a2dSMatan Barak if (idev) 174603db3a2dSMatan Barak dev_put(idev); 174703db3a2dSMatan Barak } 174803db3a2dSMatan Barak } 174903db3a2dSMatan Barak 175003db3a2dSMatan Barak /** 175103db3a2dSMatan Barak * ib_enum_all_roce_netdevs - enumerate all RoCE devices 175203db3a2dSMatan Barak * @filter: Should we call the callback? 175303db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 175403db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 175503db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 175603db3a2dSMatan Barak * 175703db3a2dSMatan Barak * Enumerates all RoCE devices' physical ports which are related 175803db3a2dSMatan Barak * to netdevices and calls callback() on each device for which 175903db3a2dSMatan Barak * filter() function returns non zero. 176003db3a2dSMatan Barak */ 176103db3a2dSMatan Barak void ib_enum_all_roce_netdevs(roce_netdev_filter filter, 176203db3a2dSMatan Barak void *filter_cookie, 176303db3a2dSMatan Barak roce_netdev_callback cb, 176403db3a2dSMatan Barak void *cookie) 176503db3a2dSMatan Barak { 176603db3a2dSMatan Barak struct ib_device *dev; 17670df91bb6SJason Gunthorpe unsigned long index; 176803db3a2dSMatan Barak 1769921eab11SJason Gunthorpe down_read(&devices_rwsem); 17700df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) 177103db3a2dSMatan Barak ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); 1772921eab11SJason Gunthorpe up_read(&devices_rwsem); 177303db3a2dSMatan Barak } 177403db3a2dSMatan Barak 177503db3a2dSMatan Barak /** 17768030c835SLeon Romanovsky * ib_enum_all_devs - enumerate all ib_devices 17778030c835SLeon Romanovsky * @cb: Callback to call for each found ib_device 17788030c835SLeon Romanovsky * 17798030c835SLeon Romanovsky * Enumerates all ib_devices and calls callback() on each device. 17808030c835SLeon Romanovsky */ 17818030c835SLeon Romanovsky int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, 17828030c835SLeon Romanovsky struct netlink_callback *cb) 17838030c835SLeon Romanovsky { 17840df91bb6SJason Gunthorpe unsigned long index; 17858030c835SLeon Romanovsky struct ib_device *dev; 17868030c835SLeon Romanovsky unsigned int idx = 0; 17878030c835SLeon Romanovsky int ret = 0; 17888030c835SLeon Romanovsky 1789921eab11SJason Gunthorpe down_read(&devices_rwsem); 17900df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 17918030c835SLeon Romanovsky ret = nldev_cb(dev, skb, cb, idx); 17928030c835SLeon Romanovsky if (ret) 17938030c835SLeon Romanovsky break; 17948030c835SLeon Romanovsky idx++; 17958030c835SLeon Romanovsky } 1796921eab11SJason Gunthorpe up_read(&devices_rwsem); 17978030c835SLeon Romanovsky return ret; 17988030c835SLeon Romanovsky } 17998030c835SLeon Romanovsky 18008030c835SLeon Romanovsky /** 18011da177e4SLinus Torvalds * ib_query_pkey - Get P_Key table entry 18021da177e4SLinus Torvalds * @device:Device to query 18031da177e4SLinus Torvalds * @port_num:Port number to query 18041da177e4SLinus Torvalds * @index:P_Key table index to query 18051da177e4SLinus Torvalds * @pkey:Returned P_Key 18061da177e4SLinus Torvalds * 18071da177e4SLinus Torvalds * ib_query_pkey() fetches the specified P_Key table entry. 18081da177e4SLinus Torvalds */ 18091da177e4SLinus Torvalds int ib_query_pkey(struct ib_device *device, 18101da177e4SLinus Torvalds u8 port_num, u16 index, u16 *pkey) 18111da177e4SLinus Torvalds { 18129af3f5cfSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 18139af3f5cfSYuval Shaia return -EINVAL; 18149af3f5cfSYuval Shaia 18153023a1e9SKamal Heib return device->ops.query_pkey(device, port_num, index, pkey); 18161da177e4SLinus Torvalds } 18171da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_pkey); 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds /** 18201da177e4SLinus Torvalds * ib_modify_device - Change IB device attributes 18211da177e4SLinus Torvalds * @device:Device to modify 18221da177e4SLinus Torvalds * @device_modify_mask:Mask of attributes to change 18231da177e4SLinus Torvalds * @device_modify:New attribute values 18241da177e4SLinus Torvalds * 18251da177e4SLinus Torvalds * ib_modify_device() changes a device's attributes as specified by 18261da177e4SLinus Torvalds * the @device_modify_mask and @device_modify structure. 18271da177e4SLinus Torvalds */ 18281da177e4SLinus Torvalds int ib_modify_device(struct ib_device *device, 18291da177e4SLinus Torvalds int device_modify_mask, 18301da177e4SLinus Torvalds struct ib_device_modify *device_modify) 18311da177e4SLinus Torvalds { 18323023a1e9SKamal Heib if (!device->ops.modify_device) 183310e1b54bSBart Van Assche return -ENOSYS; 183410e1b54bSBart Van Assche 18353023a1e9SKamal Heib return device->ops.modify_device(device, device_modify_mask, 18361da177e4SLinus Torvalds device_modify); 18371da177e4SLinus Torvalds } 18381da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_device); 18391da177e4SLinus Torvalds 18401da177e4SLinus Torvalds /** 18411da177e4SLinus Torvalds * ib_modify_port - Modifies the attributes for the specified port. 18421da177e4SLinus Torvalds * @device: The device to modify. 18431da177e4SLinus Torvalds * @port_num: The number of the port to modify. 18441da177e4SLinus Torvalds * @port_modify_mask: Mask used to specify which attributes of the port 18451da177e4SLinus Torvalds * to change. 18461da177e4SLinus Torvalds * @port_modify: New attribute values for the port. 18471da177e4SLinus Torvalds * 18481da177e4SLinus Torvalds * ib_modify_port() changes a port's attributes as specified by the 18491da177e4SLinus Torvalds * @port_modify_mask and @port_modify structure. 18501da177e4SLinus Torvalds */ 18511da177e4SLinus Torvalds int ib_modify_port(struct ib_device *device, 18521da177e4SLinus Torvalds u8 port_num, int port_modify_mask, 18531da177e4SLinus Torvalds struct ib_port_modify *port_modify) 18541da177e4SLinus Torvalds { 185561e0962dSSelvin Xavier int rc; 185610e1b54bSBart Van Assche 185724dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 1858116c0074SRoland Dreier return -EINVAL; 1859116c0074SRoland Dreier 18603023a1e9SKamal Heib if (device->ops.modify_port) 18613023a1e9SKamal Heib rc = device->ops.modify_port(device, port_num, 18623023a1e9SKamal Heib port_modify_mask, 18631da177e4SLinus Torvalds port_modify); 186461e0962dSSelvin Xavier else 186561e0962dSSelvin Xavier rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS; 186661e0962dSSelvin Xavier return rc; 18671da177e4SLinus Torvalds } 18681da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_port); 18691da177e4SLinus Torvalds 18705eb620c8SYosef Etigin /** 18715eb620c8SYosef Etigin * ib_find_gid - Returns the port number and GID table index where 1872dbb12562SParav Pandit * a specified GID value occurs. Its searches only for IB link layer. 18735eb620c8SYosef Etigin * @device: The device to query. 18745eb620c8SYosef Etigin * @gid: The GID value to search for. 18755eb620c8SYosef Etigin * @port_num: The port number of the device where the GID value was found. 18765eb620c8SYosef Etigin * @index: The index into the GID table where the GID was found. This 18775eb620c8SYosef Etigin * parameter may be NULL. 18785eb620c8SYosef Etigin */ 18795eb620c8SYosef Etigin int ib_find_gid(struct ib_device *device, union ib_gid *gid, 1880b26c4a11SParav Pandit u8 *port_num, u16 *index) 18815eb620c8SYosef Etigin { 18825eb620c8SYosef Etigin union ib_gid tmp_gid; 1883ea1075edSJason Gunthorpe unsigned int port; 1884ea1075edSJason Gunthorpe int ret, i; 18855eb620c8SYosef Etigin 1886ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 188722d24f75SParav Pandit if (!rdma_protocol_ib(device, port)) 1888b39ffa1dSMatan Barak continue; 1889b39ffa1dSMatan Barak 18908ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; 18918ceb1357SJason Gunthorpe ++i) { 18921dfce294SParav Pandit ret = rdma_query_gid(device, port, i, &tmp_gid); 18935eb620c8SYosef Etigin if (ret) 18945eb620c8SYosef Etigin return ret; 18955eb620c8SYosef Etigin if (!memcmp(&tmp_gid, gid, sizeof *gid)) { 18965eb620c8SYosef Etigin *port_num = port; 18975eb620c8SYosef Etigin if (index) 18985eb620c8SYosef Etigin *index = i; 18995eb620c8SYosef Etigin return 0; 19005eb620c8SYosef Etigin } 19015eb620c8SYosef Etigin } 19025eb620c8SYosef Etigin } 19035eb620c8SYosef Etigin 19045eb620c8SYosef Etigin return -ENOENT; 19055eb620c8SYosef Etigin } 19065eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_gid); 19075eb620c8SYosef Etigin 19085eb620c8SYosef Etigin /** 19095eb620c8SYosef Etigin * ib_find_pkey - Returns the PKey table index where a specified 19105eb620c8SYosef Etigin * PKey value occurs. 19115eb620c8SYosef Etigin * @device: The device to query. 19125eb620c8SYosef Etigin * @port_num: The port number of the device to search for the PKey. 19135eb620c8SYosef Etigin * @pkey: The PKey value to search for. 19145eb620c8SYosef Etigin * @index: The index into the PKey table where the PKey was found. 19155eb620c8SYosef Etigin */ 19165eb620c8SYosef Etigin int ib_find_pkey(struct ib_device *device, 19175eb620c8SYosef Etigin u8 port_num, u16 pkey, u16 *index) 19185eb620c8SYosef Etigin { 19195eb620c8SYosef Etigin int ret, i; 19205eb620c8SYosef Etigin u16 tmp_pkey; 1921ff7166c4SJack Morgenstein int partial_ix = -1; 19225eb620c8SYosef Etigin 19238ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; 19248ceb1357SJason Gunthorpe ++i) { 19255eb620c8SYosef Etigin ret = ib_query_pkey(device, port_num, i, &tmp_pkey); 19265eb620c8SYosef Etigin if (ret) 19275eb620c8SYosef Etigin return ret; 192836026eccSMoni Shoua if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { 1929ff7166c4SJack Morgenstein /* if there is full-member pkey take it.*/ 1930ff7166c4SJack Morgenstein if (tmp_pkey & 0x8000) { 19315eb620c8SYosef Etigin *index = i; 19325eb620c8SYosef Etigin return 0; 19335eb620c8SYosef Etigin } 1934ff7166c4SJack Morgenstein if (partial_ix < 0) 1935ff7166c4SJack Morgenstein partial_ix = i; 1936ff7166c4SJack Morgenstein } 19375eb620c8SYosef Etigin } 19385eb620c8SYosef Etigin 1939ff7166c4SJack Morgenstein /*no full-member, if exists take the limited*/ 1940ff7166c4SJack Morgenstein if (partial_ix >= 0) { 1941ff7166c4SJack Morgenstein *index = partial_ix; 1942ff7166c4SJack Morgenstein return 0; 1943ff7166c4SJack Morgenstein } 19445eb620c8SYosef Etigin return -ENOENT; 19455eb620c8SYosef Etigin } 19465eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_pkey); 19475eb620c8SYosef Etigin 19489268f72dSYotam Kenneth /** 19499268f72dSYotam Kenneth * ib_get_net_dev_by_params() - Return the appropriate net_dev 19509268f72dSYotam Kenneth * for a received CM request 19519268f72dSYotam Kenneth * @dev: An RDMA device on which the request has been received. 19529268f72dSYotam Kenneth * @port: Port number on the RDMA device. 19539268f72dSYotam Kenneth * @pkey: The Pkey the request came on. 19549268f72dSYotam Kenneth * @gid: A GID that the net_dev uses to communicate. 19559268f72dSYotam Kenneth * @addr: Contains the IP address that the request specified as its 19569268f72dSYotam Kenneth * destination. 1957921eab11SJason Gunthorpe * 19589268f72dSYotam Kenneth */ 19599268f72dSYotam Kenneth struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, 19609268f72dSYotam Kenneth u8 port, 19619268f72dSYotam Kenneth u16 pkey, 19629268f72dSYotam Kenneth const union ib_gid *gid, 19639268f72dSYotam Kenneth const struct sockaddr *addr) 19649268f72dSYotam Kenneth { 19659268f72dSYotam Kenneth struct net_device *net_dev = NULL; 19660df91bb6SJason Gunthorpe unsigned long index; 19670df91bb6SJason Gunthorpe void *client_data; 19689268f72dSYotam Kenneth 19699268f72dSYotam Kenneth if (!rdma_protocol_ib(dev, port)) 19709268f72dSYotam Kenneth return NULL; 19719268f72dSYotam Kenneth 1972921eab11SJason Gunthorpe /* 1973921eab11SJason Gunthorpe * Holding the read side guarantees that the client will not become 1974921eab11SJason Gunthorpe * unregistered while we are calling get_net_dev_by_params() 1975921eab11SJason Gunthorpe */ 1976921eab11SJason Gunthorpe down_read(&dev->client_data_rwsem); 19770df91bb6SJason Gunthorpe xan_for_each_marked (&dev->client_data, index, client_data, 19780df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED) { 19790df91bb6SJason Gunthorpe struct ib_client *client = xa_load(&clients, index); 19809268f72dSYotam Kenneth 19810df91bb6SJason Gunthorpe if (!client || !client->get_net_dev_by_params) 19829268f72dSYotam Kenneth continue; 19839268f72dSYotam Kenneth 19840df91bb6SJason Gunthorpe net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, 19850df91bb6SJason Gunthorpe addr, client_data); 19869268f72dSYotam Kenneth if (net_dev) 19879268f72dSYotam Kenneth break; 19889268f72dSYotam Kenneth } 1989921eab11SJason Gunthorpe up_read(&dev->client_data_rwsem); 19909268f72dSYotam Kenneth 19919268f72dSYotam Kenneth return net_dev; 19929268f72dSYotam Kenneth } 19939268f72dSYotam Kenneth EXPORT_SYMBOL(ib_get_net_dev_by_params); 19949268f72dSYotam Kenneth 1995521ed0d9SKamal Heib void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) 1996521ed0d9SKamal Heib { 19973023a1e9SKamal Heib struct ib_device_ops *dev_ops = &dev->ops; 1998521ed0d9SKamal Heib #define SET_DEVICE_OP(ptr, name) \ 1999521ed0d9SKamal Heib do { \ 2000521ed0d9SKamal Heib if (ops->name) \ 2001521ed0d9SKamal Heib if (!((ptr)->name)) \ 2002521ed0d9SKamal Heib (ptr)->name = ops->name; \ 2003521ed0d9SKamal Heib } while (0) 2004521ed0d9SKamal Heib 200530471d4bSLeon Romanovsky #define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) 200630471d4bSLeon Romanovsky 20073023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, add_gid); 20082f1927b0SMoni Shoua SET_DEVICE_OP(dev_ops, advise_mr); 20093023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_dm); 20103023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_fmr); 20113023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_hw_stats); 20123023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mr); 20133023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mw); 20143023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_pd); 20153023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_rdma_netdev); 20163023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_ucontext); 20173023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_xrcd); 20183023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, attach_mcast); 20193023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, check_mr_status); 20203023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_ah); 20213023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_counters); 20223023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_cq); 20233023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow); 20243023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow_action_esp); 20253023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_qp); 20263023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_rwq_ind_table); 20273023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_srq); 20283023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_wq); 20293023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_dm); 2030d0899892SJason Gunthorpe SET_DEVICE_OP(dev_ops, dealloc_driver); 20313023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_fmr); 20323023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_mw); 20333023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_pd); 20343023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_ucontext); 20353023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_xrcd); 20363023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, del_gid); 20373023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dereg_mr); 20383023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_ah); 20393023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_counters); 20403023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_cq); 20413023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow); 20423023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow_action); 20433023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_qp); 20443023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_rwq_ind_table); 20453023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_srq); 20463023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_wq); 20473023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, detach_mcast); 20483023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, disassociate_ucontext); 20493023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_rq); 20503023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_sq); 2051ca22354bSJason Gunthorpe SET_DEVICE_OP(dev_ops, enable_driver); 205202da3750SLeon Romanovsky SET_DEVICE_OP(dev_ops, fill_res_entry); 20533023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dev_fw_str); 20543023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dma_mr); 20553023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_hw_stats); 20563023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_link_layer); 20573023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_netdev); 20583023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_port_immutable); 20593023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vector_affinity); 20603023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_config); 20613023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_stats); 2062ea4baf7fSParav Pandit SET_DEVICE_OP(dev_ops, init_port); 20633023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, map_mr_sg); 20643023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, map_phys_fmr); 20653023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, mmap); 20663023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_ah); 20673023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_cq); 20683023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_device); 20693023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_flow_action_esp); 20703023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_port); 20713023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_qp); 20723023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_srq); 20733023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_wq); 20743023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, peek_cq); 20753023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, poll_cq); 20763023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_recv); 20773023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_send); 20783023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_srq_recv); 20793023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, process_mad); 20803023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_ah); 20813023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_device); 20823023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_gid); 20833023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_pkey); 20843023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_port); 20853023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_qp); 20863023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_srq); 20873023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rdma_netdev_get_params); 20883023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, read_counters); 20893023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_dm_mr); 20903023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_user_mr); 20913023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_ncomp_notif); 20923023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_notify_cq); 20933023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rereg_user_mr); 20943023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, resize_cq); 20953023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_guid); 20963023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_link_state); 20973023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, unmap_fmr); 209821a428a0SLeon Romanovsky 209921a428a0SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_pd); 2100a2a074efSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_ucontext); 2101521ed0d9SKamal Heib } 2102521ed0d9SKamal Heib EXPORT_SYMBOL(ib_set_device_ops); 2103521ed0d9SKamal Heib 2104d0e312feSLeon Romanovsky static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = { 2105735c631aSMark Bloch [RDMA_NL_LS_OP_RESOLVE] = { 2106647c75acSLeon Romanovsky .doit = ib_nl_handle_resolve_resp, 2107e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2108e3a2b93dSLeon Romanovsky }, 2109735c631aSMark Bloch [RDMA_NL_LS_OP_SET_TIMEOUT] = { 2110647c75acSLeon Romanovsky .doit = ib_nl_handle_set_timeout, 2111e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2112e3a2b93dSLeon Romanovsky }, 2113ae43f828SMark Bloch [RDMA_NL_LS_OP_IP_RESOLVE] = { 2114647c75acSLeon Romanovsky .doit = ib_nl_handle_ip_res_resp, 2115e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2116e3a2b93dSLeon Romanovsky }, 2117735c631aSMark Bloch }; 2118735c631aSMark Bloch 21191da177e4SLinus Torvalds static int __init ib_core_init(void) 21201da177e4SLinus Torvalds { 21211da177e4SLinus Torvalds int ret; 21221da177e4SLinus Torvalds 2123f0626710STejun Heo ib_wq = alloc_workqueue("infiniband", 0, 0); 2124f0626710STejun Heo if (!ib_wq) 2125f0626710STejun Heo return -ENOMEM; 2126f0626710STejun Heo 212714d3a3b2SChristoph Hellwig ib_comp_wq = alloc_workqueue("ib-comp-wq", 2128b7363e67SSagi Grimberg WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0); 212914d3a3b2SChristoph Hellwig if (!ib_comp_wq) { 213014d3a3b2SChristoph Hellwig ret = -ENOMEM; 213114d3a3b2SChristoph Hellwig goto err; 213214d3a3b2SChristoph Hellwig } 213314d3a3b2SChristoph Hellwig 2134f794809aSJack Morgenstein ib_comp_unbound_wq = 2135f794809aSJack Morgenstein alloc_workqueue("ib-comp-unb-wq", 2136f794809aSJack Morgenstein WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM | 2137f794809aSJack Morgenstein WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE); 2138f794809aSJack Morgenstein if (!ib_comp_unbound_wq) { 2139f794809aSJack Morgenstein ret = -ENOMEM; 2140f794809aSJack Morgenstein goto err_comp; 2141f794809aSJack Morgenstein } 2142f794809aSJack Morgenstein 214355aeed06SJason Gunthorpe ret = class_register(&ib_class); 2144fd75c789SNir Muchtar if (ret) { 2145aba25a3eSParav Pandit pr_warn("Couldn't create InfiniBand device class\n"); 2146f794809aSJack Morgenstein goto err_comp_unbound; 2147fd75c789SNir Muchtar } 21481da177e4SLinus Torvalds 2149c9901724SLeon Romanovsky ret = rdma_nl_init(); 21501da177e4SLinus Torvalds if (ret) { 2151c9901724SLeon Romanovsky pr_warn("Couldn't init IB netlink interface: err %d\n", ret); 2152fd75c789SNir Muchtar goto err_sysfs; 21531da177e4SLinus Torvalds } 21541da177e4SLinus Torvalds 2155e3f20f02SLeon Romanovsky ret = addr_init(); 2156e3f20f02SLeon Romanovsky if (ret) { 2157e3f20f02SLeon Romanovsky pr_warn("Could't init IB address resolution\n"); 2158e3f20f02SLeon Romanovsky goto err_ibnl; 2159e3f20f02SLeon Romanovsky } 2160e3f20f02SLeon Romanovsky 21614c2cb422SMark Bloch ret = ib_mad_init(); 21624c2cb422SMark Bloch if (ret) { 21634c2cb422SMark Bloch pr_warn("Couldn't init IB MAD\n"); 21644c2cb422SMark Bloch goto err_addr; 21654c2cb422SMark Bloch } 21664c2cb422SMark Bloch 2167c2e49c92SMark Bloch ret = ib_sa_init(); 2168c2e49c92SMark Bloch if (ret) { 2169c2e49c92SMark Bloch pr_warn("Couldn't init SA\n"); 2170c2e49c92SMark Bloch goto err_mad; 2171c2e49c92SMark Bloch } 2172c2e49c92SMark Bloch 21738f408ab6SDaniel Jurgens ret = register_lsm_notifier(&ibdev_lsm_nb); 21748f408ab6SDaniel Jurgens if (ret) { 21758f408ab6SDaniel Jurgens pr_warn("Couldn't register LSM notifier. ret %d\n", ret); 2176c9901724SLeon Romanovsky goto err_sa; 21778f408ab6SDaniel Jurgens } 21788f408ab6SDaniel Jurgens 21794e0f7b90SParav Pandit ret = register_pernet_device(&rdma_dev_net_ops); 21804e0f7b90SParav Pandit if (ret) { 21814e0f7b90SParav Pandit pr_warn("Couldn't init compat dev. ret %d\n", ret); 21824e0f7b90SParav Pandit goto err_compat; 21834e0f7b90SParav Pandit } 21844e0f7b90SParav Pandit 21856c80b41aSLeon Romanovsky nldev_init(); 2186c9901724SLeon Romanovsky rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table); 21875ef8c0c1SJason Gunthorpe roce_gid_mgmt_init(); 2188b2cbae2cSRoland Dreier 2189fd75c789SNir Muchtar return 0; 2190fd75c789SNir Muchtar 21914e0f7b90SParav Pandit err_compat: 21924e0f7b90SParav Pandit unregister_lsm_notifier(&ibdev_lsm_nb); 2193735c631aSMark Bloch err_sa: 2194735c631aSMark Bloch ib_sa_cleanup(); 2195c2e49c92SMark Bloch err_mad: 2196c2e49c92SMark Bloch ib_mad_cleanup(); 21974c2cb422SMark Bloch err_addr: 21984c2cb422SMark Bloch addr_cleanup(); 2199e3f20f02SLeon Romanovsky err_ibnl: 2200c9901724SLeon Romanovsky rdma_nl_exit(); 2201fd75c789SNir Muchtar err_sysfs: 220255aeed06SJason Gunthorpe class_unregister(&ib_class); 2203f794809aSJack Morgenstein err_comp_unbound: 2204f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 220514d3a3b2SChristoph Hellwig err_comp: 220614d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2207fd75c789SNir Muchtar err: 2208fd75c789SNir Muchtar destroy_workqueue(ib_wq); 22091da177e4SLinus Torvalds return ret; 22101da177e4SLinus Torvalds } 22111da177e4SLinus Torvalds 22121da177e4SLinus Torvalds static void __exit ib_core_cleanup(void) 22131da177e4SLinus Torvalds { 22145ef8c0c1SJason Gunthorpe roce_gid_mgmt_cleanup(); 22156c80b41aSLeon Romanovsky nldev_exit(); 2216c9901724SLeon Romanovsky rdma_nl_unregister(RDMA_NL_LS); 22174e0f7b90SParav Pandit unregister_pernet_device(&rdma_dev_net_ops); 2218c9901724SLeon Romanovsky unregister_lsm_notifier(&ibdev_lsm_nb); 2219c2e49c92SMark Bloch ib_sa_cleanup(); 22204c2cb422SMark Bloch ib_mad_cleanup(); 2221e3f20f02SLeon Romanovsky addr_cleanup(); 2222c9901724SLeon Romanovsky rdma_nl_exit(); 222355aeed06SJason Gunthorpe class_unregister(&ib_class); 2224f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 222514d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2226f7c6a7b5SRoland Dreier /* Make sure that any pending umem accounting work is done. */ 2227f0626710STejun Heo destroy_workqueue(ib_wq); 2228d0899892SJason Gunthorpe flush_workqueue(system_unbound_wq); 2229e59178d8SJason Gunthorpe WARN_ON(!xa_empty(&clients)); 22300df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&devices)); 22311da177e4SLinus Torvalds } 22321da177e4SLinus Torvalds 2233e3bf14bdSJason Gunthorpe MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); 2234e3bf14bdSJason Gunthorpe 223562dfa795SParav Pandit /* ib core relies on netdev stack to first register net_ns_type_operations 223662dfa795SParav Pandit * ns kobject type before ib_core initialization. 223762dfa795SParav Pandit */ 223862dfa795SParav Pandit fs_initcall(ib_core_init); 22391da177e4SLinus Torvalds module_exit(ib_core_cleanup); 2240