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 130cb7e0e13SParav Pandit bool ib_devices_shared_netns = true; 131a56bc45bSParav Pandit module_param_named(netns_mode, ib_devices_shared_netns, bool, 0444); 132a56bc45bSParav Pandit MODULE_PARM_DESC(netns_mode, 133a56bc45bSParav Pandit "Share device among net namespaces; default=1 (shared)"); 13441c61401SParav Pandit /** 13541c61401SParav Pandit * rdma_dev_access_netns() - Return whether a rdma device can be accessed 13641c61401SParav Pandit * from a specified net namespace or not. 13741c61401SParav Pandit * @device: Pointer to rdma device which needs to be checked 13841c61401SParav Pandit * @net: Pointer to net namesapce for which access to be checked 13941c61401SParav Pandit * 14041c61401SParav Pandit * rdma_dev_access_netns() - Return whether a rdma device can be accessed 14141c61401SParav Pandit * from a specified net namespace or not. When 14241c61401SParav Pandit * rdma device is in shared mode, it ignores the 14341c61401SParav Pandit * net namespace. When rdma device is exclusive 14441c61401SParav Pandit * to a net namespace, rdma device net namespace is 14541c61401SParav Pandit * checked against the specified one. 14641c61401SParav Pandit */ 14741c61401SParav Pandit bool rdma_dev_access_netns(const struct ib_device *dev, const struct net *net) 14841c61401SParav Pandit { 14941c61401SParav Pandit return (ib_devices_shared_netns || 15041c61401SParav Pandit net_eq(read_pnet(&dev->coredev.rdma_net), net)); 15141c61401SParav Pandit } 15241c61401SParav Pandit EXPORT_SYMBOL(rdma_dev_access_netns); 15341c61401SParav Pandit 1540df91bb6SJason Gunthorpe /* 1550df91bb6SJason Gunthorpe * xarray has this behavior where it won't iterate over NULL values stored in 1560df91bb6SJason Gunthorpe * allocated arrays. So we need our own iterator to see all values stored in 1570df91bb6SJason Gunthorpe * the array. This does the same thing as xa_for_each except that it also 1580df91bb6SJason Gunthorpe * returns NULL valued entries if the array is allocating. Simplified to only 1590df91bb6SJason Gunthorpe * work on simple xarrays. 1600df91bb6SJason Gunthorpe */ 1610df91bb6SJason Gunthorpe static void *xan_find_marked(struct xarray *xa, unsigned long *indexp, 1620df91bb6SJason Gunthorpe xa_mark_t filter) 1630df91bb6SJason Gunthorpe { 1640df91bb6SJason Gunthorpe XA_STATE(xas, xa, *indexp); 1650df91bb6SJason Gunthorpe void *entry; 1660df91bb6SJason Gunthorpe 1670df91bb6SJason Gunthorpe rcu_read_lock(); 1680df91bb6SJason Gunthorpe do { 1690df91bb6SJason Gunthorpe entry = xas_find_marked(&xas, ULONG_MAX, filter); 1700df91bb6SJason Gunthorpe if (xa_is_zero(entry)) 1710df91bb6SJason Gunthorpe break; 1720df91bb6SJason Gunthorpe } while (xas_retry(&xas, entry)); 1730df91bb6SJason Gunthorpe rcu_read_unlock(); 1740df91bb6SJason Gunthorpe 1750df91bb6SJason Gunthorpe if (entry) { 1760df91bb6SJason Gunthorpe *indexp = xas.xa_index; 1770df91bb6SJason Gunthorpe if (xa_is_zero(entry)) 1780df91bb6SJason Gunthorpe return NULL; 1790df91bb6SJason Gunthorpe return entry; 1800df91bb6SJason Gunthorpe } 1810df91bb6SJason Gunthorpe return XA_ERROR(-ENOENT); 1820df91bb6SJason Gunthorpe } 1830df91bb6SJason Gunthorpe #define xan_for_each_marked(xa, index, entry, filter) \ 1840df91bb6SJason Gunthorpe for (index = 0, entry = xan_find_marked(xa, &(index), filter); \ 1850df91bb6SJason Gunthorpe !xa_is_err(entry); \ 1860df91bb6SJason Gunthorpe (index)++, entry = xan_find_marked(xa, &(index), filter)) 1870df91bb6SJason Gunthorpe 188324e227eSJason Gunthorpe /* RCU hash table mapping netdevice pointers to struct ib_port_data */ 189324e227eSJason Gunthorpe static DEFINE_SPINLOCK(ndev_hash_lock); 190324e227eSJason Gunthorpe static DECLARE_HASHTABLE(ndev_hash, 5); 191324e227eSJason Gunthorpe 192c2261dd7SJason Gunthorpe static void free_netdevs(struct ib_device *ib_dev); 193d0899892SJason Gunthorpe static void ib_unregister_work(struct work_struct *work); 194d0899892SJason Gunthorpe static void __ib_unregister_device(struct ib_device *device); 1958f408ab6SDaniel Jurgens static int ib_security_change(struct notifier_block *nb, unsigned long event, 1968f408ab6SDaniel Jurgens void *lsm_data); 1978f408ab6SDaniel Jurgens static void ib_policy_change_task(struct work_struct *work); 1988f408ab6SDaniel Jurgens static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task); 1998f408ab6SDaniel Jurgens 200923abb9dSGal Pressman static void __ibdev_printk(const char *level, const struct ib_device *ibdev, 201923abb9dSGal Pressman struct va_format *vaf) 202923abb9dSGal Pressman { 203923abb9dSGal Pressman if (ibdev && ibdev->dev.parent) 204923abb9dSGal Pressman dev_printk_emit(level[1] - '0', 205923abb9dSGal Pressman ibdev->dev.parent, 206923abb9dSGal Pressman "%s %s %s: %pV", 207923abb9dSGal Pressman dev_driver_string(ibdev->dev.parent), 208923abb9dSGal Pressman dev_name(ibdev->dev.parent), 209923abb9dSGal Pressman dev_name(&ibdev->dev), 210923abb9dSGal Pressman vaf); 211923abb9dSGal Pressman else if (ibdev) 212923abb9dSGal Pressman printk("%s%s: %pV", 213923abb9dSGal Pressman level, dev_name(&ibdev->dev), vaf); 214923abb9dSGal Pressman else 215923abb9dSGal Pressman printk("%s(NULL ib_device): %pV", level, vaf); 216923abb9dSGal Pressman } 217923abb9dSGal Pressman 218923abb9dSGal Pressman void ibdev_printk(const char *level, const struct ib_device *ibdev, 219923abb9dSGal Pressman const char *format, ...) 220923abb9dSGal Pressman { 221923abb9dSGal Pressman struct va_format vaf; 222923abb9dSGal Pressman va_list args; 223923abb9dSGal Pressman 224923abb9dSGal Pressman va_start(args, format); 225923abb9dSGal Pressman 226923abb9dSGal Pressman vaf.fmt = format; 227923abb9dSGal Pressman vaf.va = &args; 228923abb9dSGal Pressman 229923abb9dSGal Pressman __ibdev_printk(level, ibdev, &vaf); 230923abb9dSGal Pressman 231923abb9dSGal Pressman va_end(args); 232923abb9dSGal Pressman } 233923abb9dSGal Pressman EXPORT_SYMBOL(ibdev_printk); 234923abb9dSGal Pressman 235923abb9dSGal Pressman #define define_ibdev_printk_level(func, level) \ 236923abb9dSGal Pressman void func(const struct ib_device *ibdev, const char *fmt, ...) \ 237923abb9dSGal Pressman { \ 238923abb9dSGal Pressman struct va_format vaf; \ 239923abb9dSGal Pressman va_list args; \ 240923abb9dSGal Pressman \ 241923abb9dSGal Pressman va_start(args, fmt); \ 242923abb9dSGal Pressman \ 243923abb9dSGal Pressman vaf.fmt = fmt; \ 244923abb9dSGal Pressman vaf.va = &args; \ 245923abb9dSGal Pressman \ 246923abb9dSGal Pressman __ibdev_printk(level, ibdev, &vaf); \ 247923abb9dSGal Pressman \ 248923abb9dSGal Pressman va_end(args); \ 249923abb9dSGal Pressman } \ 250923abb9dSGal Pressman EXPORT_SYMBOL(func); 251923abb9dSGal Pressman 252923abb9dSGal Pressman define_ibdev_printk_level(ibdev_emerg, KERN_EMERG); 253923abb9dSGal Pressman define_ibdev_printk_level(ibdev_alert, KERN_ALERT); 254923abb9dSGal Pressman define_ibdev_printk_level(ibdev_crit, KERN_CRIT); 255923abb9dSGal Pressman define_ibdev_printk_level(ibdev_err, KERN_ERR); 256923abb9dSGal Pressman define_ibdev_printk_level(ibdev_warn, KERN_WARNING); 257923abb9dSGal Pressman define_ibdev_printk_level(ibdev_notice, KERN_NOTICE); 258923abb9dSGal Pressman define_ibdev_printk_level(ibdev_info, KERN_INFO); 259923abb9dSGal Pressman 2608f408ab6SDaniel Jurgens static struct notifier_block ibdev_lsm_nb = { 2618f408ab6SDaniel Jurgens .notifier_call = ib_security_change, 2628f408ab6SDaniel Jurgens }; 2631da177e4SLinus Torvalds 264decbc7a6SParav Pandit static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, 265decbc7a6SParav Pandit struct net *net); 266decbc7a6SParav Pandit 267324e227eSJason Gunthorpe /* Pointer to the RCU head at the start of the ib_port_data array */ 268324e227eSJason Gunthorpe struct ib_port_data_rcu { 269324e227eSJason Gunthorpe struct rcu_head rcu_head; 270324e227eSJason Gunthorpe struct ib_port_data pdata[]; 271324e227eSJason Gunthorpe }; 272324e227eSJason Gunthorpe 2731da177e4SLinus Torvalds static int ib_device_check_mandatory(struct ib_device *device) 2741da177e4SLinus Torvalds { 2753023a1e9SKamal Heib #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device_ops, x), #x } 2761da177e4SLinus Torvalds static const struct { 2771da177e4SLinus Torvalds size_t offset; 2781da177e4SLinus Torvalds char *name; 2791da177e4SLinus Torvalds } mandatory_table[] = { 2801da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_device), 2811da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_port), 2821da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_pkey), 2831da177e4SLinus Torvalds IB_MANDATORY_FUNC(alloc_pd), 2841da177e4SLinus Torvalds IB_MANDATORY_FUNC(dealloc_pd), 2851da177e4SLinus Torvalds IB_MANDATORY_FUNC(create_qp), 2861da177e4SLinus Torvalds IB_MANDATORY_FUNC(modify_qp), 2871da177e4SLinus Torvalds IB_MANDATORY_FUNC(destroy_qp), 2881da177e4SLinus Torvalds IB_MANDATORY_FUNC(post_send), 2891da177e4SLinus Torvalds IB_MANDATORY_FUNC(post_recv), 2901da177e4SLinus Torvalds IB_MANDATORY_FUNC(create_cq), 2911da177e4SLinus Torvalds IB_MANDATORY_FUNC(destroy_cq), 2921da177e4SLinus Torvalds IB_MANDATORY_FUNC(poll_cq), 2931da177e4SLinus Torvalds IB_MANDATORY_FUNC(req_notify_cq), 2941da177e4SLinus Torvalds IB_MANDATORY_FUNC(get_dma_mr), 2957738613eSIra Weiny IB_MANDATORY_FUNC(dereg_mr), 2967738613eSIra Weiny IB_MANDATORY_FUNC(get_port_immutable) 2971da177e4SLinus Torvalds }; 2981da177e4SLinus Torvalds int i; 2991da177e4SLinus Torvalds 3006780c4faSGal Pressman device->kverbs_provider = true; 3019a6b090cSAhmed S. Darwish for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { 3023023a1e9SKamal Heib if (!*(void **) ((void *) &device->ops + 3033023a1e9SKamal Heib mandatory_table[i].offset)) { 3046780c4faSGal Pressman device->kverbs_provider = false; 3056780c4faSGal Pressman break; 3061da177e4SLinus Torvalds } 3071da177e4SLinus Torvalds } 3081da177e4SLinus Torvalds 3091da177e4SLinus Torvalds return 0; 3101da177e4SLinus Torvalds } 3111da177e4SLinus Torvalds 312f8978bd9SLeon Romanovsky /* 31301b67117SParav Pandit * Caller must perform ib_device_put() to return the device reference count 31401b67117SParav Pandit * when ib_device_get_by_index() returns valid device pointer. 315f8978bd9SLeon Romanovsky */ 31637eeab55SParav Pandit struct ib_device *ib_device_get_by_index(const struct net *net, u32 index) 317f8978bd9SLeon Romanovsky { 318f8978bd9SLeon Romanovsky struct ib_device *device; 319f8978bd9SLeon Romanovsky 320921eab11SJason Gunthorpe down_read(&devices_rwsem); 3210df91bb6SJason Gunthorpe device = xa_load(&devices, index); 32201b67117SParav Pandit if (device) { 32337eeab55SParav Pandit if (!rdma_dev_access_netns(device, net)) { 32437eeab55SParav Pandit device = NULL; 32537eeab55SParav Pandit goto out; 32637eeab55SParav Pandit } 32737eeab55SParav Pandit 328d79af724SJason Gunthorpe if (!ib_device_try_get(device)) 32901b67117SParav Pandit device = NULL; 33001b67117SParav Pandit } 33137eeab55SParav Pandit out: 332921eab11SJason Gunthorpe up_read(&devices_rwsem); 333f8978bd9SLeon Romanovsky return device; 334f8978bd9SLeon Romanovsky } 335f8978bd9SLeon Romanovsky 336d79af724SJason Gunthorpe /** 337d79af724SJason Gunthorpe * ib_device_put - Release IB device reference 338d79af724SJason Gunthorpe * @device: device whose reference to be released 339d79af724SJason Gunthorpe * 340d79af724SJason Gunthorpe * ib_device_put() releases reference to the IB device to allow it to be 341d79af724SJason Gunthorpe * unregistered and eventually free. 342d79af724SJason Gunthorpe */ 34301b67117SParav Pandit void ib_device_put(struct ib_device *device) 34401b67117SParav Pandit { 34501b67117SParav Pandit if (refcount_dec_and_test(&device->refcount)) 34601b67117SParav Pandit complete(&device->unreg_completion); 34701b67117SParav Pandit } 348d79af724SJason Gunthorpe EXPORT_SYMBOL(ib_device_put); 34901b67117SParav Pandit 3501da177e4SLinus Torvalds static struct ib_device *__ib_device_get_by_name(const char *name) 3511da177e4SLinus Torvalds { 3521da177e4SLinus Torvalds struct ib_device *device; 3530df91bb6SJason Gunthorpe unsigned long index; 3541da177e4SLinus Torvalds 3550df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) 356896de009SJason Gunthorpe if (!strcmp(name, dev_name(&device->dev))) 3571da177e4SLinus Torvalds return device; 3581da177e4SLinus Torvalds 3591da177e4SLinus Torvalds return NULL; 3601da177e4SLinus Torvalds } 3611da177e4SLinus Torvalds 3626cc2c8e5SJason Gunthorpe /** 3636cc2c8e5SJason Gunthorpe * ib_device_get_by_name - Find an IB device by name 3646cc2c8e5SJason Gunthorpe * @name: The name to look for 3656cc2c8e5SJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 3666cc2c8e5SJason Gunthorpe * 3676cc2c8e5SJason Gunthorpe * Find and hold an ib_device by its name. The caller must call 3686cc2c8e5SJason Gunthorpe * ib_device_put() on the returned pointer. 3696cc2c8e5SJason Gunthorpe */ 3706cc2c8e5SJason Gunthorpe struct ib_device *ib_device_get_by_name(const char *name, 3716cc2c8e5SJason Gunthorpe enum rdma_driver_id driver_id) 3726cc2c8e5SJason Gunthorpe { 3736cc2c8e5SJason Gunthorpe struct ib_device *device; 3746cc2c8e5SJason Gunthorpe 3756cc2c8e5SJason Gunthorpe down_read(&devices_rwsem); 3766cc2c8e5SJason Gunthorpe device = __ib_device_get_by_name(name); 3776cc2c8e5SJason Gunthorpe if (device && driver_id != RDMA_DRIVER_UNKNOWN && 3786cc2c8e5SJason Gunthorpe device->driver_id != driver_id) 3796cc2c8e5SJason Gunthorpe device = NULL; 3806cc2c8e5SJason Gunthorpe 3816cc2c8e5SJason Gunthorpe if (device) { 3826cc2c8e5SJason Gunthorpe if (!ib_device_try_get(device)) 3836cc2c8e5SJason Gunthorpe device = NULL; 3846cc2c8e5SJason Gunthorpe } 3856cc2c8e5SJason Gunthorpe up_read(&devices_rwsem); 3866cc2c8e5SJason Gunthorpe return device; 3876cc2c8e5SJason Gunthorpe } 3886cc2c8e5SJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_name); 3896cc2c8e5SJason Gunthorpe 3904e0f7b90SParav Pandit static int rename_compat_devs(struct ib_device *device) 3914e0f7b90SParav Pandit { 3924e0f7b90SParav Pandit struct ib_core_device *cdev; 3934e0f7b90SParav Pandit unsigned long index; 3944e0f7b90SParav Pandit int ret = 0; 3954e0f7b90SParav Pandit 3964e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 3974e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) { 3984e0f7b90SParav Pandit ret = device_rename(&cdev->dev, dev_name(&device->dev)); 3994e0f7b90SParav Pandit if (ret) { 4004e0f7b90SParav Pandit dev_warn(&cdev->dev, 4014e0f7b90SParav Pandit "Fail to rename compatdev to new name %s\n", 4024e0f7b90SParav Pandit dev_name(&device->dev)); 4034e0f7b90SParav Pandit break; 4044e0f7b90SParav Pandit } 4054e0f7b90SParav Pandit } 4064e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 4074e0f7b90SParav Pandit return ret; 4084e0f7b90SParav Pandit } 4094e0f7b90SParav Pandit 410d21943ddSLeon Romanovsky int ib_device_rename(struct ib_device *ibdev, const char *name) 411d21943ddSLeon Romanovsky { 412dc1435c0SLeon Romanovsky unsigned long index; 413dc1435c0SLeon Romanovsky void *client_data; 414e3593b56SJason Gunthorpe int ret; 415d21943ddSLeon Romanovsky 416921eab11SJason Gunthorpe down_write(&devices_rwsem); 417e3593b56SJason Gunthorpe if (!strcmp(name, dev_name(&ibdev->dev))) { 418dc1435c0SLeon Romanovsky up_write(&devices_rwsem); 419dc1435c0SLeon Romanovsky return 0; 420e3593b56SJason Gunthorpe } 421e3593b56SJason Gunthorpe 422344684e6SJason Gunthorpe if (__ib_device_get_by_name(name)) { 423dc1435c0SLeon Romanovsky up_write(&devices_rwsem); 424dc1435c0SLeon Romanovsky return -EEXIST; 425d21943ddSLeon Romanovsky } 426d21943ddSLeon Romanovsky 427d21943ddSLeon Romanovsky ret = device_rename(&ibdev->dev, name); 428dc1435c0SLeon Romanovsky if (ret) { 429921eab11SJason Gunthorpe up_write(&devices_rwsem); 430d21943ddSLeon Romanovsky return ret; 431d21943ddSLeon Romanovsky } 432d21943ddSLeon Romanovsky 433dc1435c0SLeon Romanovsky strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); 434dc1435c0SLeon Romanovsky ret = rename_compat_devs(ibdev); 435dc1435c0SLeon Romanovsky 436dc1435c0SLeon Romanovsky downgrade_write(&devices_rwsem); 437dc1435c0SLeon Romanovsky down_read(&ibdev->client_data_rwsem); 438dc1435c0SLeon Romanovsky xan_for_each_marked(&ibdev->client_data, index, client_data, 439dc1435c0SLeon Romanovsky CLIENT_DATA_REGISTERED) { 440dc1435c0SLeon Romanovsky struct ib_client *client = xa_load(&clients, index); 441dc1435c0SLeon Romanovsky 442dc1435c0SLeon Romanovsky if (!client || !client->rename) 443dc1435c0SLeon Romanovsky continue; 444dc1435c0SLeon Romanovsky 445dc1435c0SLeon Romanovsky client->rename(ibdev, client_data); 446dc1435c0SLeon Romanovsky } 447dc1435c0SLeon Romanovsky up_read(&ibdev->client_data_rwsem); 448dc1435c0SLeon Romanovsky up_read(&devices_rwsem); 449dc1435c0SLeon Romanovsky return 0; 450dc1435c0SLeon Romanovsky } 451dc1435c0SLeon Romanovsky 452e349f858SJason Gunthorpe static int alloc_name(struct ib_device *ibdev, const char *name) 4531da177e4SLinus Torvalds { 4541da177e4SLinus Torvalds struct ib_device *device; 4550df91bb6SJason Gunthorpe unsigned long index; 4563b88afd3SJason Gunthorpe struct ida inuse; 4573b88afd3SJason Gunthorpe int rc; 4581da177e4SLinus Torvalds int i; 4591da177e4SLinus Torvalds 460921eab11SJason Gunthorpe lockdep_assert_held_exclusive(&devices_rwsem); 4613b88afd3SJason Gunthorpe ida_init(&inuse); 4620df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) { 463e349f858SJason Gunthorpe char buf[IB_DEVICE_NAME_MAX]; 464e349f858SJason Gunthorpe 465896de009SJason Gunthorpe if (sscanf(dev_name(&device->dev), name, &i) != 1) 4661da177e4SLinus Torvalds continue; 4673b88afd3SJason Gunthorpe if (i < 0 || i >= INT_MAX) 4681da177e4SLinus Torvalds continue; 4691da177e4SLinus Torvalds snprintf(buf, sizeof buf, name, i); 4703b88afd3SJason Gunthorpe if (strcmp(buf, dev_name(&device->dev)) != 0) 4713b88afd3SJason Gunthorpe continue; 4723b88afd3SJason Gunthorpe 4733b88afd3SJason Gunthorpe rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); 4743b88afd3SJason Gunthorpe if (rc < 0) 4753b88afd3SJason Gunthorpe goto out; 4761da177e4SLinus Torvalds } 4771da177e4SLinus Torvalds 4783b88afd3SJason Gunthorpe rc = ida_alloc(&inuse, GFP_KERNEL); 4793b88afd3SJason Gunthorpe if (rc < 0) 4803b88afd3SJason Gunthorpe goto out; 4811da177e4SLinus Torvalds 4823b88afd3SJason Gunthorpe rc = dev_set_name(&ibdev->dev, name, rc); 4833b88afd3SJason Gunthorpe out: 4843b88afd3SJason Gunthorpe ida_destroy(&inuse); 4853b88afd3SJason Gunthorpe return rc; 4861da177e4SLinus Torvalds } 4871da177e4SLinus Torvalds 48855aeed06SJason Gunthorpe static void ib_device_release(struct device *device) 48955aeed06SJason Gunthorpe { 49055aeed06SJason Gunthorpe struct ib_device *dev = container_of(device, struct ib_device, dev); 49155aeed06SJason Gunthorpe 492c2261dd7SJason Gunthorpe free_netdevs(dev); 493652432f3SJason Gunthorpe WARN_ON(refcount_read(&dev->refcount)); 49446bdf370SKamal Heib if (dev->port_data) { 49503db3a2dSMatan Barak ib_cache_release_one(dev); 496b34b269aSJason Gunthorpe ib_security_release_port_pkey_list(dev); 497324e227eSJason Gunthorpe kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, 498324e227eSJason Gunthorpe pdata[0]), 499324e227eSJason Gunthorpe rcu_head); 50046bdf370SKamal Heib } 50146bdf370SKamal Heib xa_destroy(&dev->compat_devs); 50246bdf370SKamal Heib xa_destroy(&dev->client_data); 503324e227eSJason Gunthorpe kfree_rcu(dev, rcu_head); 50455aeed06SJason Gunthorpe } 50555aeed06SJason Gunthorpe 50655aeed06SJason Gunthorpe static int ib_device_uevent(struct device *device, 50755aeed06SJason Gunthorpe struct kobj_uevent_env *env) 50855aeed06SJason Gunthorpe { 509896de009SJason Gunthorpe if (add_uevent_var(env, "NAME=%s", dev_name(device))) 51055aeed06SJason Gunthorpe return -ENOMEM; 51155aeed06SJason Gunthorpe 51255aeed06SJason Gunthorpe /* 51355aeed06SJason Gunthorpe * It would be nice to pass the node GUID with the event... 51455aeed06SJason Gunthorpe */ 51555aeed06SJason Gunthorpe 51655aeed06SJason Gunthorpe return 0; 51755aeed06SJason Gunthorpe } 51855aeed06SJason Gunthorpe 51962dfa795SParav Pandit static const void *net_namespace(struct device *d) 52062dfa795SParav Pandit { 5214e0f7b90SParav Pandit struct ib_core_device *coredev = 5224e0f7b90SParav Pandit container_of(d, struct ib_core_device, dev); 5234e0f7b90SParav Pandit 5244e0f7b90SParav Pandit return read_pnet(&coredev->rdma_net); 52562dfa795SParav Pandit } 52662dfa795SParav Pandit 52755aeed06SJason Gunthorpe static struct class ib_class = { 52855aeed06SJason Gunthorpe .name = "infiniband", 52955aeed06SJason Gunthorpe .dev_release = ib_device_release, 53055aeed06SJason Gunthorpe .dev_uevent = ib_device_uevent, 53162dfa795SParav Pandit .ns_type = &net_ns_type_operations, 53262dfa795SParav Pandit .namespace = net_namespace, 53355aeed06SJason Gunthorpe }; 53455aeed06SJason Gunthorpe 535cebe556bSParav Pandit static void rdma_init_coredev(struct ib_core_device *coredev, 5364e0f7b90SParav Pandit struct ib_device *dev, struct net *net) 537cebe556bSParav Pandit { 538cebe556bSParav Pandit /* This BUILD_BUG_ON is intended to catch layout change 539cebe556bSParav Pandit * of union of ib_core_device and device. 540cebe556bSParav Pandit * dev must be the first element as ib_core and providers 541cebe556bSParav Pandit * driver uses it. Adding anything in ib_core_device before 542cebe556bSParav Pandit * device will break this assumption. 543cebe556bSParav Pandit */ 544cebe556bSParav Pandit BUILD_BUG_ON(offsetof(struct ib_device, coredev.dev) != 545cebe556bSParav Pandit offsetof(struct ib_device, dev)); 546cebe556bSParav Pandit 547cebe556bSParav Pandit coredev->dev.class = &ib_class; 548cebe556bSParav Pandit coredev->dev.groups = dev->groups; 549cebe556bSParav Pandit device_initialize(&coredev->dev); 550cebe556bSParav Pandit coredev->owner = dev; 551cebe556bSParav Pandit INIT_LIST_HEAD(&coredev->port_list); 5524e0f7b90SParav Pandit write_pnet(&coredev->rdma_net, net); 553cebe556bSParav Pandit } 554cebe556bSParav Pandit 5551da177e4SLinus Torvalds /** 556459cc69fSLeon Romanovsky * _ib_alloc_device - allocate an IB device struct 5571da177e4SLinus Torvalds * @size:size of structure to allocate 5581da177e4SLinus Torvalds * 5591da177e4SLinus Torvalds * Low-level drivers should use ib_alloc_device() to allocate &struct 5601da177e4SLinus Torvalds * ib_device. @size is the size of the structure to be allocated, 5611da177e4SLinus Torvalds * including any private data used by the low-level driver. 5621da177e4SLinus Torvalds * ib_dealloc_device() must be used to free structures allocated with 5631da177e4SLinus Torvalds * ib_alloc_device(). 5641da177e4SLinus Torvalds */ 565459cc69fSLeon Romanovsky struct ib_device *_ib_alloc_device(size_t size) 5661da177e4SLinus Torvalds { 56755aeed06SJason Gunthorpe struct ib_device *device; 5681da177e4SLinus Torvalds 56955aeed06SJason Gunthorpe if (WARN_ON(size < sizeof(struct ib_device))) 57055aeed06SJason Gunthorpe return NULL; 57155aeed06SJason Gunthorpe 57255aeed06SJason Gunthorpe device = kzalloc(size, GFP_KERNEL); 57355aeed06SJason Gunthorpe if (!device) 57455aeed06SJason Gunthorpe return NULL; 57555aeed06SJason Gunthorpe 57641eda65cSLeon Romanovsky if (rdma_restrack_init(device)) { 57741eda65cSLeon Romanovsky kfree(device); 57841eda65cSLeon Romanovsky return NULL; 57941eda65cSLeon Romanovsky } 58002d8883fSLeon Romanovsky 5815f8f5499SParav Pandit device->groups[0] = &ib_dev_attr_group; 5824e0f7b90SParav Pandit rdma_init_coredev(&device->coredev, device, &init_net); 58355aeed06SJason Gunthorpe 58455aeed06SJason Gunthorpe INIT_LIST_HEAD(&device->event_handler_list); 58555aeed06SJason Gunthorpe spin_lock_init(&device->event_handler_lock); 586d0899892SJason Gunthorpe mutex_init(&device->unregistration_lock); 5870df91bb6SJason Gunthorpe /* 5880df91bb6SJason Gunthorpe * client_data needs to be alloc because we don't want our mark to be 5890df91bb6SJason Gunthorpe * destroyed if the user stores NULL in the client data. 5900df91bb6SJason Gunthorpe */ 5910df91bb6SJason Gunthorpe xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); 592921eab11SJason Gunthorpe init_rwsem(&device->client_data_rwsem); 5934e0f7b90SParav Pandit xa_init_flags(&device->compat_devs, XA_FLAGS_ALLOC); 5944e0f7b90SParav Pandit mutex_init(&device->compat_devs_mutex); 59501b67117SParav Pandit init_completion(&device->unreg_completion); 596d0899892SJason Gunthorpe INIT_WORK(&device->unregistration_work, ib_unregister_work); 59755aeed06SJason Gunthorpe 59855aeed06SJason Gunthorpe return device; 5991da177e4SLinus Torvalds } 600459cc69fSLeon Romanovsky EXPORT_SYMBOL(_ib_alloc_device); 6011da177e4SLinus Torvalds 6021da177e4SLinus Torvalds /** 6031da177e4SLinus Torvalds * ib_dealloc_device - free an IB device struct 6041da177e4SLinus Torvalds * @device:structure to free 6051da177e4SLinus Torvalds * 6061da177e4SLinus Torvalds * Free a structure allocated with ib_alloc_device(). 6071da177e4SLinus Torvalds */ 6081da177e4SLinus Torvalds void ib_dealloc_device(struct ib_device *device) 6091da177e4SLinus Torvalds { 610d0899892SJason Gunthorpe if (device->ops.dealloc_driver) 611d0899892SJason Gunthorpe device->ops.dealloc_driver(device); 612d0899892SJason Gunthorpe 613d0899892SJason Gunthorpe /* 614d0899892SJason Gunthorpe * ib_unregister_driver() requires all devices to remain in the xarray 615d0899892SJason Gunthorpe * while their ops are callable. The last op we call is dealloc_driver 616d0899892SJason Gunthorpe * above. This is needed to create a fence on op callbacks prior to 617d0899892SJason Gunthorpe * allowing the driver module to unload. 618d0899892SJason Gunthorpe */ 619d0899892SJason Gunthorpe down_write(&devices_rwsem); 620d0899892SJason Gunthorpe if (xa_load(&devices, device->index) == device) 621d0899892SJason Gunthorpe xa_erase(&devices, device->index); 622d0899892SJason Gunthorpe up_write(&devices_rwsem); 623d0899892SJason Gunthorpe 624c2261dd7SJason Gunthorpe /* Expedite releasing netdev references */ 625c2261dd7SJason Gunthorpe free_netdevs(device); 626c2261dd7SJason Gunthorpe 6274e0f7b90SParav Pandit WARN_ON(!xa_empty(&device->compat_devs)); 6280df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&device->client_data)); 629652432f3SJason Gunthorpe WARN_ON(refcount_read(&device->refcount)); 6300ad699c0SLeon Romanovsky rdma_restrack_clean(device); 631e155755eSParav Pandit /* Balances with device_initialize */ 632924b8900SLeon Romanovsky put_device(&device->dev); 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds EXPORT_SYMBOL(ib_dealloc_device); 6351da177e4SLinus Torvalds 636921eab11SJason Gunthorpe /* 637921eab11SJason Gunthorpe * add_client_context() and remove_client_context() must be safe against 638921eab11SJason Gunthorpe * parallel calls on the same device - registration/unregistration of both the 639921eab11SJason Gunthorpe * device and client can be occurring in parallel. 640921eab11SJason Gunthorpe * 641921eab11SJason Gunthorpe * The routines need to be a fence, any caller must not return until the add 642921eab11SJason Gunthorpe * or remove is fully completed. 643921eab11SJason Gunthorpe */ 644921eab11SJason Gunthorpe static int add_client_context(struct ib_device *device, 645921eab11SJason Gunthorpe struct ib_client *client) 6461da177e4SLinus Torvalds { 647921eab11SJason Gunthorpe int ret = 0; 6481da177e4SLinus Torvalds 6496780c4faSGal Pressman if (!device->kverbs_provider && !client->no_kverbs_req) 650921eab11SJason Gunthorpe return 0; 6516780c4faSGal Pressman 652921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 653921eab11SJason Gunthorpe /* 654921eab11SJason Gunthorpe * Another caller to add_client_context got here first and has already 655921eab11SJason Gunthorpe * completely initialized context. 656921eab11SJason Gunthorpe */ 657921eab11SJason Gunthorpe if (xa_get_mark(&device->client_data, client->client_id, 658921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) 659921eab11SJason Gunthorpe goto out; 660921eab11SJason Gunthorpe 661921eab11SJason Gunthorpe ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, 662921eab11SJason Gunthorpe GFP_KERNEL)); 663921eab11SJason Gunthorpe if (ret) 664921eab11SJason Gunthorpe goto out; 665921eab11SJason Gunthorpe downgrade_write(&device->client_data_rwsem); 666921eab11SJason Gunthorpe if (client->add) 667921eab11SJason Gunthorpe client->add(device); 668921eab11SJason Gunthorpe 669921eab11SJason Gunthorpe /* Readers shall not see a client until add has been completed */ 6700df91bb6SJason Gunthorpe xa_set_mark(&device->client_data, client->client_id, 6710df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED); 672921eab11SJason Gunthorpe up_read(&device->client_data_rwsem); 673921eab11SJason Gunthorpe return 0; 6741da177e4SLinus Torvalds 675921eab11SJason Gunthorpe out: 676921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 677921eab11SJason Gunthorpe return ret; 678921eab11SJason Gunthorpe } 679921eab11SJason Gunthorpe 680921eab11SJason Gunthorpe static void remove_client_context(struct ib_device *device, 681921eab11SJason Gunthorpe unsigned int client_id) 682921eab11SJason Gunthorpe { 683921eab11SJason Gunthorpe struct ib_client *client; 684921eab11SJason Gunthorpe void *client_data; 685921eab11SJason Gunthorpe 686921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 687921eab11SJason Gunthorpe if (!xa_get_mark(&device->client_data, client_id, 688921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) { 689921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 690921eab11SJason Gunthorpe return; 691921eab11SJason Gunthorpe } 692921eab11SJason Gunthorpe client_data = xa_load(&device->client_data, client_id); 693921eab11SJason Gunthorpe xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); 694921eab11SJason Gunthorpe client = xa_load(&clients, client_id); 695921eab11SJason Gunthorpe downgrade_write(&device->client_data_rwsem); 696921eab11SJason Gunthorpe 697921eab11SJason Gunthorpe /* 698921eab11SJason Gunthorpe * Notice we cannot be holding any exclusive locks when calling the 699921eab11SJason Gunthorpe * remove callback as the remove callback can recurse back into any 700921eab11SJason Gunthorpe * public functions in this module and thus try for any locks those 701921eab11SJason Gunthorpe * functions take. 702921eab11SJason Gunthorpe * 703921eab11SJason Gunthorpe * For this reason clients and drivers should not call the 704921eab11SJason Gunthorpe * unregistration functions will holdling any locks. 705921eab11SJason Gunthorpe * 706921eab11SJason Gunthorpe * It tempting to drop the client_data_rwsem too, but this is required 707921eab11SJason Gunthorpe * to ensure that unregister_client does not return until all clients 708921eab11SJason Gunthorpe * are completely unregistered, which is required to avoid module 709921eab11SJason Gunthorpe * unloading races. 710921eab11SJason Gunthorpe */ 711921eab11SJason Gunthorpe if (client->remove) 712921eab11SJason Gunthorpe client->remove(device, client_data); 713921eab11SJason Gunthorpe 714921eab11SJason Gunthorpe xa_erase(&device->client_data, client_id); 715921eab11SJason Gunthorpe up_read(&device->client_data_rwsem); 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 718c2261dd7SJason Gunthorpe static int alloc_port_data(struct ib_device *device) 7195eb620c8SYosef Etigin { 720324e227eSJason Gunthorpe struct ib_port_data_rcu *pdata_rcu; 721ea1075edSJason Gunthorpe unsigned int port; 722c2261dd7SJason Gunthorpe 723c2261dd7SJason Gunthorpe if (device->port_data) 724c2261dd7SJason Gunthorpe return 0; 725c2261dd7SJason Gunthorpe 726c2261dd7SJason Gunthorpe /* This can only be called once the physical port range is defined */ 727c2261dd7SJason Gunthorpe if (WARN_ON(!device->phys_port_cnt)) 728c2261dd7SJason Gunthorpe return -EINVAL; 7295eb620c8SYosef Etigin 7308ceb1357SJason Gunthorpe /* 7318ceb1357SJason Gunthorpe * device->port_data is indexed directly by the port number to make 7327738613eSIra Weiny * access to this data as efficient as possible. 7337738613eSIra Weiny * 7348ceb1357SJason Gunthorpe * Therefore port_data is declared as a 1 based array with potential 7358ceb1357SJason Gunthorpe * empty slots at the beginning. 7367738613eSIra Weiny */ 737324e227eSJason Gunthorpe pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, 738324e227eSJason Gunthorpe rdma_end_port(device) + 1), 739324e227eSJason Gunthorpe GFP_KERNEL); 740324e227eSJason Gunthorpe if (!pdata_rcu) 74155aeed06SJason Gunthorpe return -ENOMEM; 742324e227eSJason Gunthorpe /* 743324e227eSJason Gunthorpe * The rcu_head is put in front of the port data array and the stored 744324e227eSJason Gunthorpe * pointer is adjusted since we never need to see that member until 745324e227eSJason Gunthorpe * kfree_rcu. 746324e227eSJason Gunthorpe */ 747324e227eSJason Gunthorpe device->port_data = pdata_rcu->pdata; 7485eb620c8SYosef Etigin 749ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 7508ceb1357SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 7518ceb1357SJason Gunthorpe 752324e227eSJason Gunthorpe pdata->ib_dev = device; 7538ceb1357SJason Gunthorpe spin_lock_init(&pdata->pkey_list_lock); 7548ceb1357SJason Gunthorpe INIT_LIST_HEAD(&pdata->pkey_list); 755c2261dd7SJason Gunthorpe spin_lock_init(&pdata->netdev_lock); 756324e227eSJason Gunthorpe INIT_HLIST_NODE(&pdata->ndev_hash_link); 757c2261dd7SJason Gunthorpe } 758c2261dd7SJason Gunthorpe return 0; 759c2261dd7SJason Gunthorpe } 760c2261dd7SJason Gunthorpe 761c2261dd7SJason Gunthorpe static int verify_immutable(const struct ib_device *dev, u8 port) 762c2261dd7SJason Gunthorpe { 763c2261dd7SJason Gunthorpe return WARN_ON(!rdma_cap_ib_mad(dev, port) && 764c2261dd7SJason Gunthorpe rdma_max_mad_size(dev, port) != 0); 765c2261dd7SJason Gunthorpe } 766c2261dd7SJason Gunthorpe 767c2261dd7SJason Gunthorpe static int setup_port_data(struct ib_device *device) 768c2261dd7SJason Gunthorpe { 769c2261dd7SJason Gunthorpe unsigned int port; 770c2261dd7SJason Gunthorpe int ret; 771c2261dd7SJason Gunthorpe 772c2261dd7SJason Gunthorpe ret = alloc_port_data(device); 773c2261dd7SJason Gunthorpe if (ret) 774c2261dd7SJason Gunthorpe return ret; 775c2261dd7SJason Gunthorpe 776c2261dd7SJason Gunthorpe rdma_for_each_port (device, port) { 777c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 7788ceb1357SJason Gunthorpe 7798ceb1357SJason Gunthorpe ret = device->ops.get_port_immutable(device, port, 7808ceb1357SJason Gunthorpe &pdata->immutable); 7815eb620c8SYosef Etigin if (ret) 7825eb620c8SYosef Etigin return ret; 78355aeed06SJason Gunthorpe 78455aeed06SJason Gunthorpe if (verify_immutable(device, port)) 78555aeed06SJason Gunthorpe return -EINVAL; 78655aeed06SJason Gunthorpe } 78755aeed06SJason Gunthorpe return 0; 7885eb620c8SYosef Etigin } 7895eb620c8SYosef Etigin 7909abb0d1bSLeon Romanovsky void ib_get_device_fw_str(struct ib_device *dev, char *str) 7915fa76c20SIra Weiny { 7923023a1e9SKamal Heib if (dev->ops.get_dev_fw_str) 7933023a1e9SKamal Heib dev->ops.get_dev_fw_str(dev, str); 7945fa76c20SIra Weiny else 7955fa76c20SIra Weiny str[0] = '\0'; 7965fa76c20SIra Weiny } 7975fa76c20SIra Weiny EXPORT_SYMBOL(ib_get_device_fw_str); 7985fa76c20SIra Weiny 7998f408ab6SDaniel Jurgens static void ib_policy_change_task(struct work_struct *work) 8008f408ab6SDaniel Jurgens { 8018f408ab6SDaniel Jurgens struct ib_device *dev; 8020df91bb6SJason Gunthorpe unsigned long index; 8038f408ab6SDaniel Jurgens 804921eab11SJason Gunthorpe down_read(&devices_rwsem); 8050df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 806ea1075edSJason Gunthorpe unsigned int i; 8078f408ab6SDaniel Jurgens 808ea1075edSJason Gunthorpe rdma_for_each_port (dev, i) { 8098f408ab6SDaniel Jurgens u64 sp; 8108f408ab6SDaniel Jurgens int ret = ib_get_cached_subnet_prefix(dev, 8118f408ab6SDaniel Jurgens i, 8128f408ab6SDaniel Jurgens &sp); 8138f408ab6SDaniel Jurgens 8148f408ab6SDaniel Jurgens WARN_ONCE(ret, 8158f408ab6SDaniel Jurgens "ib_get_cached_subnet_prefix err: %d, this should never happen here\n", 8168f408ab6SDaniel Jurgens ret); 817a750cfdeSDaniel Jurgens if (!ret) 8188f408ab6SDaniel Jurgens ib_security_cache_change(dev, i, sp); 8198f408ab6SDaniel Jurgens } 8208f408ab6SDaniel Jurgens } 821921eab11SJason Gunthorpe up_read(&devices_rwsem); 8228f408ab6SDaniel Jurgens } 8238f408ab6SDaniel Jurgens 8248f408ab6SDaniel Jurgens static int ib_security_change(struct notifier_block *nb, unsigned long event, 8258f408ab6SDaniel Jurgens void *lsm_data) 8268f408ab6SDaniel Jurgens { 8278f408ab6SDaniel Jurgens if (event != LSM_POLICY_CHANGE) 8288f408ab6SDaniel Jurgens return NOTIFY_DONE; 8298f408ab6SDaniel Jurgens 8308f408ab6SDaniel Jurgens schedule_work(&ib_policy_change_work); 831c66f6741SDaniel Jurgens ib_mad_agent_security_change(); 8328f408ab6SDaniel Jurgens 8338f408ab6SDaniel Jurgens return NOTIFY_OK; 8348f408ab6SDaniel Jurgens } 8358f408ab6SDaniel Jurgens 8364e0f7b90SParav Pandit static void compatdev_release(struct device *dev) 8374e0f7b90SParav Pandit { 8384e0f7b90SParav Pandit struct ib_core_device *cdev = 8394e0f7b90SParav Pandit container_of(dev, struct ib_core_device, dev); 8404e0f7b90SParav Pandit 8414e0f7b90SParav Pandit kfree(cdev); 8424e0f7b90SParav Pandit } 8434e0f7b90SParav Pandit 8444e0f7b90SParav Pandit static int add_one_compat_dev(struct ib_device *device, 8454e0f7b90SParav Pandit struct rdma_dev_net *rnet) 8464e0f7b90SParav Pandit { 8474e0f7b90SParav Pandit struct ib_core_device *cdev; 8484e0f7b90SParav Pandit int ret; 8494e0f7b90SParav Pandit 8502b34c558SParav Pandit lockdep_assert_held(&rdma_nets_rwsem); 851a56bc45bSParav Pandit if (!ib_devices_shared_netns) 852a56bc45bSParav Pandit return 0; 853a56bc45bSParav Pandit 8544e0f7b90SParav Pandit /* 8554e0f7b90SParav Pandit * Create and add compat device in all namespaces other than where it 8564e0f7b90SParav Pandit * is currently bound to. 8574e0f7b90SParav Pandit */ 8584e0f7b90SParav Pandit if (net_eq(read_pnet(&rnet->net), 8594e0f7b90SParav Pandit read_pnet(&device->coredev.rdma_net))) 8604e0f7b90SParav Pandit return 0; 8614e0f7b90SParav Pandit 8624e0f7b90SParav Pandit /* 8634e0f7b90SParav Pandit * The first of init_net() or ib_register_device() to take the 8644e0f7b90SParav Pandit * compat_devs_mutex wins and gets to add the device. Others will wait 8654e0f7b90SParav Pandit * for completion here. 8664e0f7b90SParav Pandit */ 8674e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 8684e0f7b90SParav Pandit cdev = xa_load(&device->compat_devs, rnet->id); 8694e0f7b90SParav Pandit if (cdev) { 8704e0f7b90SParav Pandit ret = 0; 8714e0f7b90SParav Pandit goto done; 8724e0f7b90SParav Pandit } 8734e0f7b90SParav Pandit ret = xa_reserve(&device->compat_devs, rnet->id, GFP_KERNEL); 8744e0f7b90SParav Pandit if (ret) 8754e0f7b90SParav Pandit goto done; 8764e0f7b90SParav Pandit 8774e0f7b90SParav Pandit cdev = kzalloc(sizeof(*cdev), GFP_KERNEL); 8784e0f7b90SParav Pandit if (!cdev) { 8794e0f7b90SParav Pandit ret = -ENOMEM; 8804e0f7b90SParav Pandit goto cdev_err; 8814e0f7b90SParav Pandit } 8824e0f7b90SParav Pandit 8834e0f7b90SParav Pandit cdev->dev.parent = device->dev.parent; 8844e0f7b90SParav Pandit rdma_init_coredev(cdev, device, read_pnet(&rnet->net)); 8854e0f7b90SParav Pandit cdev->dev.release = compatdev_release; 8864e0f7b90SParav Pandit dev_set_name(&cdev->dev, "%s", dev_name(&device->dev)); 8874e0f7b90SParav Pandit 8884e0f7b90SParav Pandit ret = device_add(&cdev->dev); 8894e0f7b90SParav Pandit if (ret) 8904e0f7b90SParav Pandit goto add_err; 891eb15c78bSParav Pandit ret = ib_setup_port_attrs(cdev); 8925417783eSParav Pandit if (ret) 8935417783eSParav Pandit goto port_err; 8944e0f7b90SParav Pandit 8954e0f7b90SParav Pandit ret = xa_err(xa_store(&device->compat_devs, rnet->id, 8964e0f7b90SParav Pandit cdev, GFP_KERNEL)); 8974e0f7b90SParav Pandit if (ret) 8984e0f7b90SParav Pandit goto insert_err; 8994e0f7b90SParav Pandit 9004e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 9014e0f7b90SParav Pandit return 0; 9024e0f7b90SParav Pandit 9034e0f7b90SParav Pandit insert_err: 9045417783eSParav Pandit ib_free_port_attrs(cdev); 9055417783eSParav Pandit port_err: 9064e0f7b90SParav Pandit device_del(&cdev->dev); 9074e0f7b90SParav Pandit add_err: 9084e0f7b90SParav Pandit put_device(&cdev->dev); 9094e0f7b90SParav Pandit cdev_err: 9104e0f7b90SParav Pandit xa_release(&device->compat_devs, rnet->id); 9114e0f7b90SParav Pandit done: 9124e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 9134e0f7b90SParav Pandit return ret; 9144e0f7b90SParav Pandit } 9154e0f7b90SParav Pandit 9164e0f7b90SParav Pandit static void remove_one_compat_dev(struct ib_device *device, u32 id) 9174e0f7b90SParav Pandit { 9184e0f7b90SParav Pandit struct ib_core_device *cdev; 9194e0f7b90SParav Pandit 9204e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 9214e0f7b90SParav Pandit cdev = xa_erase(&device->compat_devs, id); 9224e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 9234e0f7b90SParav Pandit if (cdev) { 9245417783eSParav Pandit ib_free_port_attrs(cdev); 9254e0f7b90SParav Pandit device_del(&cdev->dev); 9264e0f7b90SParav Pandit put_device(&cdev->dev); 9274e0f7b90SParav Pandit } 9284e0f7b90SParav Pandit } 9294e0f7b90SParav Pandit 9304e0f7b90SParav Pandit static void remove_compat_devs(struct ib_device *device) 9314e0f7b90SParav Pandit { 9324e0f7b90SParav Pandit struct ib_core_device *cdev; 9334e0f7b90SParav Pandit unsigned long index; 9344e0f7b90SParav Pandit 9354e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) 9364e0f7b90SParav Pandit remove_one_compat_dev(device, index); 9374e0f7b90SParav Pandit } 9384e0f7b90SParav Pandit 9394e0f7b90SParav Pandit static int add_compat_devs(struct ib_device *device) 9404e0f7b90SParav Pandit { 9414e0f7b90SParav Pandit struct rdma_dev_net *rnet; 9424e0f7b90SParav Pandit unsigned long index; 9434e0f7b90SParav Pandit int ret = 0; 9444e0f7b90SParav Pandit 945decbc7a6SParav Pandit lockdep_assert_held(&devices_rwsem); 946decbc7a6SParav Pandit 9474e0f7b90SParav Pandit down_read(&rdma_nets_rwsem); 9484e0f7b90SParav Pandit xa_for_each (&rdma_nets, index, rnet) { 9494e0f7b90SParav Pandit ret = add_one_compat_dev(device, rnet); 9504e0f7b90SParav Pandit if (ret) 9514e0f7b90SParav Pandit break; 9524e0f7b90SParav Pandit } 9534e0f7b90SParav Pandit up_read(&rdma_nets_rwsem); 9544e0f7b90SParav Pandit return ret; 9554e0f7b90SParav Pandit } 9564e0f7b90SParav Pandit 9572b34c558SParav Pandit static void remove_all_compat_devs(void) 9582b34c558SParav Pandit { 9592b34c558SParav Pandit struct ib_compat_device *cdev; 9602b34c558SParav Pandit struct ib_device *dev; 9612b34c558SParav Pandit unsigned long index; 9622b34c558SParav Pandit 9632b34c558SParav Pandit down_read(&devices_rwsem); 9642b34c558SParav Pandit xa_for_each (&devices, index, dev) { 9652b34c558SParav Pandit unsigned long c_index = 0; 9662b34c558SParav Pandit 9672b34c558SParav Pandit /* Hold nets_rwsem so that any other thread modifying this 9682b34c558SParav Pandit * system param can sync with this thread. 9692b34c558SParav Pandit */ 9702b34c558SParav Pandit down_read(&rdma_nets_rwsem); 9712b34c558SParav Pandit xa_for_each (&dev->compat_devs, c_index, cdev) 9722b34c558SParav Pandit remove_one_compat_dev(dev, c_index); 9732b34c558SParav Pandit up_read(&rdma_nets_rwsem); 9742b34c558SParav Pandit } 9752b34c558SParav Pandit up_read(&devices_rwsem); 9762b34c558SParav Pandit } 9772b34c558SParav Pandit 9782b34c558SParav Pandit static int add_all_compat_devs(void) 9792b34c558SParav Pandit { 9802b34c558SParav Pandit struct rdma_dev_net *rnet; 9812b34c558SParav Pandit struct ib_device *dev; 9822b34c558SParav Pandit unsigned long index; 9832b34c558SParav Pandit int ret = 0; 9842b34c558SParav Pandit 9852b34c558SParav Pandit down_read(&devices_rwsem); 9862b34c558SParav Pandit xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 9872b34c558SParav Pandit unsigned long net_index = 0; 9882b34c558SParav Pandit 9892b34c558SParav Pandit /* Hold nets_rwsem so that any other thread modifying this 9902b34c558SParav Pandit * system param can sync with this thread. 9912b34c558SParav Pandit */ 9922b34c558SParav Pandit down_read(&rdma_nets_rwsem); 9932b34c558SParav Pandit xa_for_each (&rdma_nets, net_index, rnet) { 9942b34c558SParav Pandit ret = add_one_compat_dev(dev, rnet); 9952b34c558SParav Pandit if (ret) 9962b34c558SParav Pandit break; 9972b34c558SParav Pandit } 9982b34c558SParav Pandit up_read(&rdma_nets_rwsem); 9992b34c558SParav Pandit } 10002b34c558SParav Pandit up_read(&devices_rwsem); 10012b34c558SParav Pandit if (ret) 10022b34c558SParav Pandit remove_all_compat_devs(); 10032b34c558SParav Pandit return ret; 10042b34c558SParav Pandit } 10052b34c558SParav Pandit 10062b34c558SParav Pandit int rdma_compatdev_set(u8 enable) 10072b34c558SParav Pandit { 10082b34c558SParav Pandit struct rdma_dev_net *rnet; 10092b34c558SParav Pandit unsigned long index; 10102b34c558SParav Pandit int ret = 0; 10112b34c558SParav Pandit 10122b34c558SParav Pandit down_write(&rdma_nets_rwsem); 10132b34c558SParav Pandit if (ib_devices_shared_netns == enable) { 10142b34c558SParav Pandit up_write(&rdma_nets_rwsem); 10152b34c558SParav Pandit return 0; 10162b34c558SParav Pandit } 10172b34c558SParav Pandit 10182b34c558SParav Pandit /* enable/disable of compat devices is not supported 10192b34c558SParav Pandit * when more than default init_net exists. 10202b34c558SParav Pandit */ 10212b34c558SParav Pandit xa_for_each (&rdma_nets, index, rnet) { 10222b34c558SParav Pandit ret++; 10232b34c558SParav Pandit break; 10242b34c558SParav Pandit } 10252b34c558SParav Pandit if (!ret) 10262b34c558SParav Pandit ib_devices_shared_netns = enable; 10272b34c558SParav Pandit up_write(&rdma_nets_rwsem); 10282b34c558SParav Pandit if (ret) 10292b34c558SParav Pandit return -EBUSY; 10302b34c558SParav Pandit 10312b34c558SParav Pandit if (enable) 10322b34c558SParav Pandit ret = add_all_compat_devs(); 10332b34c558SParav Pandit else 10342b34c558SParav Pandit remove_all_compat_devs(); 10352b34c558SParav Pandit return ret; 10362b34c558SParav Pandit } 10372b34c558SParav Pandit 10384e0f7b90SParav Pandit static void rdma_dev_exit_net(struct net *net) 10394e0f7b90SParav Pandit { 10404e0f7b90SParav Pandit struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); 10414e0f7b90SParav Pandit struct ib_device *dev; 10424e0f7b90SParav Pandit unsigned long index; 10434e0f7b90SParav Pandit int ret; 10444e0f7b90SParav Pandit 10454e0f7b90SParav Pandit down_write(&rdma_nets_rwsem); 10464e0f7b90SParav Pandit /* 10474e0f7b90SParav Pandit * Prevent the ID from being re-used and hide the id from xa_for_each. 10484e0f7b90SParav Pandit */ 10494e0f7b90SParav Pandit ret = xa_err(xa_store(&rdma_nets, rnet->id, NULL, GFP_KERNEL)); 10504e0f7b90SParav Pandit WARN_ON(ret); 10514e0f7b90SParav Pandit up_write(&rdma_nets_rwsem); 10524e0f7b90SParav Pandit 10534e0f7b90SParav Pandit down_read(&devices_rwsem); 10544e0f7b90SParav Pandit xa_for_each (&devices, index, dev) { 10554e0f7b90SParav Pandit get_device(&dev->dev); 10564e0f7b90SParav Pandit /* 10574e0f7b90SParav Pandit * Release the devices_rwsem so that pontentially blocking 10584e0f7b90SParav Pandit * device_del, doesn't hold the devices_rwsem for too long. 10594e0f7b90SParav Pandit */ 10604e0f7b90SParav Pandit up_read(&devices_rwsem); 10614e0f7b90SParav Pandit 10624e0f7b90SParav Pandit remove_one_compat_dev(dev, rnet->id); 10634e0f7b90SParav Pandit 1064decbc7a6SParav Pandit /* 1065decbc7a6SParav Pandit * If the real device is in the NS then move it back to init. 1066decbc7a6SParav Pandit */ 1067decbc7a6SParav Pandit rdma_dev_change_netns(dev, net, &init_net); 1068decbc7a6SParav Pandit 10694e0f7b90SParav Pandit put_device(&dev->dev); 10704e0f7b90SParav Pandit down_read(&devices_rwsem); 10714e0f7b90SParav Pandit } 10724e0f7b90SParav Pandit up_read(&devices_rwsem); 10734e0f7b90SParav Pandit 10744e0f7b90SParav Pandit xa_erase(&rdma_nets, rnet->id); 10754e0f7b90SParav Pandit } 10764e0f7b90SParav Pandit 10774e0f7b90SParav Pandit static __net_init int rdma_dev_init_net(struct net *net) 10784e0f7b90SParav Pandit { 10794e0f7b90SParav Pandit struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); 10804e0f7b90SParav Pandit unsigned long index; 10814e0f7b90SParav Pandit struct ib_device *dev; 10824e0f7b90SParav Pandit int ret; 10834e0f7b90SParav Pandit 10844e0f7b90SParav Pandit /* No need to create any compat devices in default init_net. */ 10854e0f7b90SParav Pandit if (net_eq(net, &init_net)) 10864e0f7b90SParav Pandit return 0; 10874e0f7b90SParav Pandit 10884e0f7b90SParav Pandit write_pnet(&rnet->net, net); 10894e0f7b90SParav Pandit 10904e0f7b90SParav Pandit ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b, GFP_KERNEL); 10914e0f7b90SParav Pandit if (ret) 10924e0f7b90SParav Pandit return ret; 10934e0f7b90SParav Pandit 10944e0f7b90SParav Pandit down_read(&devices_rwsem); 10954e0f7b90SParav Pandit xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 10962b34c558SParav Pandit /* Hold nets_rwsem so that netlink command cannot change 10972b34c558SParav Pandit * system configuration for device sharing mode. 10982b34c558SParav Pandit */ 10992b34c558SParav Pandit down_read(&rdma_nets_rwsem); 11004e0f7b90SParav Pandit ret = add_one_compat_dev(dev, rnet); 11012b34c558SParav Pandit up_read(&rdma_nets_rwsem); 11024e0f7b90SParav Pandit if (ret) 11034e0f7b90SParav Pandit break; 11044e0f7b90SParav Pandit } 11054e0f7b90SParav Pandit up_read(&devices_rwsem); 11064e0f7b90SParav Pandit 11074e0f7b90SParav Pandit if (ret) 11084e0f7b90SParav Pandit rdma_dev_exit_net(net); 11094e0f7b90SParav Pandit 11104e0f7b90SParav Pandit return ret; 11114e0f7b90SParav Pandit } 11124e0f7b90SParav Pandit 1113ecc82c53SLeon Romanovsky /* 1114d0899892SJason Gunthorpe * Assign the unique string device name and the unique device index. This is 1115d0899892SJason Gunthorpe * undone by ib_dealloc_device. 1116ecc82c53SLeon Romanovsky */ 11170df91bb6SJason Gunthorpe static int assign_name(struct ib_device *device, const char *name) 11180df91bb6SJason Gunthorpe { 11190df91bb6SJason Gunthorpe static u32 last_id; 11200df91bb6SJason Gunthorpe int ret; 1121ecc82c53SLeon Romanovsky 1122921eab11SJason Gunthorpe down_write(&devices_rwsem); 11230df91bb6SJason Gunthorpe /* Assign a unique name to the device */ 11240df91bb6SJason Gunthorpe if (strchr(name, '%')) 11250df91bb6SJason Gunthorpe ret = alloc_name(device, name); 11260df91bb6SJason Gunthorpe else 11270df91bb6SJason Gunthorpe ret = dev_set_name(&device->dev, name); 11280df91bb6SJason Gunthorpe if (ret) 11290df91bb6SJason Gunthorpe goto out; 1130ecc82c53SLeon Romanovsky 11310df91bb6SJason Gunthorpe if (__ib_device_get_by_name(dev_name(&device->dev))) { 11320df91bb6SJason Gunthorpe ret = -ENFILE; 11330df91bb6SJason Gunthorpe goto out; 1134ecc82c53SLeon Romanovsky } 11350df91bb6SJason Gunthorpe strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); 11360df91bb6SJason Gunthorpe 1137ea295481SLinus Torvalds ret = xa_alloc_cyclic(&devices, &device->index, device, xa_limit_31b, 1138ea295481SLinus Torvalds &last_id, GFP_KERNEL); 1139ea295481SLinus Torvalds if (ret > 0) 11400df91bb6SJason Gunthorpe ret = 0; 1141921eab11SJason Gunthorpe 11420df91bb6SJason Gunthorpe out: 1143921eab11SJason Gunthorpe up_write(&devices_rwsem); 11440df91bb6SJason Gunthorpe return ret; 11450df91bb6SJason Gunthorpe } 11460df91bb6SJason Gunthorpe 1147548cb4fbSParav Pandit static void setup_dma_device(struct ib_device *device) 11481da177e4SLinus Torvalds { 114999db9494SBart Van Assche struct device *parent = device->dev.parent; 11501da177e4SLinus Torvalds 11510957c29fSBart Van Assche WARN_ON_ONCE(device->dma_device); 11520957c29fSBart Van Assche if (device->dev.dma_ops) { 11530957c29fSBart Van Assche /* 11540957c29fSBart Van Assche * The caller provided custom DMA operations. Copy the 11550957c29fSBart Van Assche * DMA-related fields that are used by e.g. dma_alloc_coherent() 11560957c29fSBart Van Assche * into device->dev. 11570957c29fSBart Van Assche */ 11580957c29fSBart Van Assche device->dma_device = &device->dev; 115902ee9da3SBart Van Assche if (!device->dev.dma_mask) { 116002ee9da3SBart Van Assche if (parent) 116199db9494SBart Van Assche device->dev.dma_mask = parent->dma_mask; 116202ee9da3SBart Van Assche else 116302ee9da3SBart Van Assche WARN_ON_ONCE(true); 116402ee9da3SBart Van Assche } 116502ee9da3SBart Van Assche if (!device->dev.coherent_dma_mask) { 116602ee9da3SBart Van Assche if (parent) 11670957c29fSBart Van Assche device->dev.coherent_dma_mask = 11680957c29fSBart Van Assche parent->coherent_dma_mask; 116902ee9da3SBart Van Assche else 117002ee9da3SBart Van Assche WARN_ON_ONCE(true); 117102ee9da3SBart Van Assche } 11720957c29fSBart Van Assche } else { 11730957c29fSBart Van Assche /* 11740957c29fSBart Van Assche * The caller did not provide custom DMA operations. Use the 11750957c29fSBart Van Assche * DMA mapping operations of the parent device. 11760957c29fSBart Van Assche */ 117702ee9da3SBart Van Assche WARN_ON_ONCE(!parent); 11780957c29fSBart Van Assche device->dma_device = parent; 11790957c29fSBart Van Assche } 1180d10bcf94SShiraz Saleem /* Setup default max segment size for all IB devices */ 1181d10bcf94SShiraz Saleem dma_set_max_seg_size(device->dma_device, SZ_2G); 1182d10bcf94SShiraz Saleem 1183548cb4fbSParav Pandit } 1184548cb4fbSParav Pandit 1185921eab11SJason Gunthorpe /* 1186921eab11SJason Gunthorpe * setup_device() allocates memory and sets up data that requires calling the 1187921eab11SJason Gunthorpe * device ops, this is the only reason these actions are not done during 1188921eab11SJason Gunthorpe * ib_alloc_device. It is undone by ib_dealloc_device(). 1189921eab11SJason Gunthorpe */ 1190548cb4fbSParav Pandit static int setup_device(struct ib_device *device) 1191548cb4fbSParav Pandit { 1192548cb4fbSParav Pandit struct ib_udata uhw = {.outlen = 0, .inlen = 0}; 1193548cb4fbSParav Pandit int ret; 1194548cb4fbSParav Pandit 1195921eab11SJason Gunthorpe setup_dma_device(device); 1196921eab11SJason Gunthorpe 1197548cb4fbSParav Pandit ret = ib_device_check_mandatory(device); 1198548cb4fbSParav Pandit if (ret) 1199548cb4fbSParav Pandit return ret; 1200548cb4fbSParav Pandit 12018ceb1357SJason Gunthorpe ret = setup_port_data(device); 1202548cb4fbSParav Pandit if (ret) { 12038ceb1357SJason Gunthorpe dev_warn(&device->dev, "Couldn't create per-port data\n"); 1204548cb4fbSParav Pandit return ret; 1205548cb4fbSParav Pandit } 1206548cb4fbSParav Pandit 1207548cb4fbSParav Pandit memset(&device->attrs, 0, sizeof(device->attrs)); 12083023a1e9SKamal Heib ret = device->ops.query_device(device, &device->attrs, &uhw); 1209548cb4fbSParav Pandit if (ret) { 1210548cb4fbSParav Pandit dev_warn(&device->dev, 1211548cb4fbSParav Pandit "Couldn't query the device attributes\n"); 1212d45f89d5SJason Gunthorpe return ret; 1213548cb4fbSParav Pandit } 1214548cb4fbSParav Pandit 1215548cb4fbSParav Pandit return 0; 1216548cb4fbSParav Pandit } 1217548cb4fbSParav Pandit 1218921eab11SJason Gunthorpe static void disable_device(struct ib_device *device) 1219921eab11SJason Gunthorpe { 1220921eab11SJason Gunthorpe struct ib_client *client; 1221921eab11SJason Gunthorpe 1222921eab11SJason Gunthorpe WARN_ON(!refcount_read(&device->refcount)); 1223921eab11SJason Gunthorpe 1224921eab11SJason Gunthorpe down_write(&devices_rwsem); 1225921eab11SJason Gunthorpe xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); 1226921eab11SJason Gunthorpe up_write(&devices_rwsem); 1227921eab11SJason Gunthorpe 1228921eab11SJason Gunthorpe down_read(&clients_rwsem); 1229921eab11SJason Gunthorpe list_for_each_entry_reverse(client, &client_list, list) 1230921eab11SJason Gunthorpe remove_client_context(device, client->client_id); 1231921eab11SJason Gunthorpe up_read(&clients_rwsem); 1232921eab11SJason Gunthorpe 1233921eab11SJason Gunthorpe /* Pairs with refcount_set in enable_device */ 1234921eab11SJason Gunthorpe ib_device_put(device); 1235921eab11SJason Gunthorpe wait_for_completion(&device->unreg_completion); 1236c2261dd7SJason Gunthorpe 12374e0f7b90SParav Pandit /* 12384e0f7b90SParav Pandit * compat devices must be removed after device refcount drops to zero. 12394e0f7b90SParav Pandit * Otherwise init_net() may add more compatdevs after removing compat 12404e0f7b90SParav Pandit * devices and before device is disabled. 12414e0f7b90SParav Pandit */ 12424e0f7b90SParav Pandit remove_compat_devs(device); 1243921eab11SJason Gunthorpe } 1244921eab11SJason Gunthorpe 1245921eab11SJason Gunthorpe /* 1246921eab11SJason Gunthorpe * An enabled device is visible to all clients and to all the public facing 1247d0899892SJason Gunthorpe * APIs that return a device pointer. This always returns with a new get, even 1248d0899892SJason Gunthorpe * if it fails. 1249921eab11SJason Gunthorpe */ 1250d0899892SJason Gunthorpe static int enable_device_and_get(struct ib_device *device) 1251921eab11SJason Gunthorpe { 1252921eab11SJason Gunthorpe struct ib_client *client; 1253921eab11SJason Gunthorpe unsigned long index; 1254d0899892SJason Gunthorpe int ret = 0; 1255921eab11SJason Gunthorpe 1256d0899892SJason Gunthorpe /* 1257d0899892SJason Gunthorpe * One ref belongs to the xa and the other belongs to this 1258d0899892SJason Gunthorpe * thread. This is needed to guard against parallel unregistration. 1259d0899892SJason Gunthorpe */ 1260d0899892SJason Gunthorpe refcount_set(&device->refcount, 2); 1261921eab11SJason Gunthorpe down_write(&devices_rwsem); 1262921eab11SJason Gunthorpe xa_set_mark(&devices, device->index, DEVICE_REGISTERED); 1263d0899892SJason Gunthorpe 1264d0899892SJason Gunthorpe /* 1265d0899892SJason Gunthorpe * By using downgrade_write() we ensure that no other thread can clear 1266d0899892SJason Gunthorpe * DEVICE_REGISTERED while we are completing the client setup. 1267d0899892SJason Gunthorpe */ 1268d0899892SJason Gunthorpe downgrade_write(&devices_rwsem); 1269921eab11SJason Gunthorpe 1270ca22354bSJason Gunthorpe if (device->ops.enable_driver) { 1271ca22354bSJason Gunthorpe ret = device->ops.enable_driver(device); 1272ca22354bSJason Gunthorpe if (ret) 1273ca22354bSJason Gunthorpe goto out; 1274ca22354bSJason Gunthorpe } 1275ca22354bSJason Gunthorpe 1276921eab11SJason Gunthorpe down_read(&clients_rwsem); 1277921eab11SJason Gunthorpe xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { 1278921eab11SJason Gunthorpe ret = add_client_context(device, client); 1279d0899892SJason Gunthorpe if (ret) 1280d0899892SJason Gunthorpe break; 1281d0899892SJason Gunthorpe } 1282921eab11SJason Gunthorpe up_read(&clients_rwsem); 12834e0f7b90SParav Pandit if (!ret) 12844e0f7b90SParav Pandit ret = add_compat_devs(device); 1285ca22354bSJason Gunthorpe out: 1286d0899892SJason Gunthorpe up_read(&devices_rwsem); 1287921eab11SJason Gunthorpe return ret; 1288921eab11SJason Gunthorpe } 1289921eab11SJason Gunthorpe 1290548cb4fbSParav Pandit /** 1291548cb4fbSParav Pandit * ib_register_device - Register an IB device with IB core 1292548cb4fbSParav Pandit * @device:Device to register 1293548cb4fbSParav Pandit * 1294548cb4fbSParav Pandit * Low-level drivers use ib_register_device() to register their 1295548cb4fbSParav Pandit * devices with the IB core. All registered clients will receive a 1296548cb4fbSParav Pandit * callback for each device that is added. @device must be allocated 1297548cb4fbSParav Pandit * with ib_alloc_device(). 1298d0899892SJason Gunthorpe * 1299d0899892SJason Gunthorpe * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() 1300d0899892SJason Gunthorpe * asynchronously then the device pointer may become freed as soon as this 1301d0899892SJason Gunthorpe * function returns. 1302548cb4fbSParav Pandit */ 1303ea4baf7fSParav Pandit int ib_register_device(struct ib_device *device, const char *name) 1304548cb4fbSParav Pandit { 1305548cb4fbSParav Pandit int ret; 13061da177e4SLinus Torvalds 13070df91bb6SJason Gunthorpe ret = assign_name(device, name); 1308e349f858SJason Gunthorpe if (ret) 1309921eab11SJason Gunthorpe return ret; 13101da177e4SLinus Torvalds 1311548cb4fbSParav Pandit ret = setup_device(device); 1312548cb4fbSParav Pandit if (ret) 1313d0899892SJason Gunthorpe return ret; 131403db3a2dSMatan Barak 1315d45f89d5SJason Gunthorpe ret = ib_cache_setup_one(device); 1316d45f89d5SJason Gunthorpe if (ret) { 1317d45f89d5SJason Gunthorpe dev_warn(&device->dev, 1318d45f89d5SJason Gunthorpe "Couldn't set up InfiniBand P_Key/GID cache\n"); 1319d0899892SJason Gunthorpe return ret; 1320d45f89d5SJason Gunthorpe } 1321d45f89d5SJason Gunthorpe 13227527a7b1SParav Pandit ib_device_register_rdmacg(device); 13233e153a93SIra Weiny 1324e7a5b4aaSLeon Romanovsky /* 1325e7a5b4aaSLeon Romanovsky * Ensure that ADD uevent is not fired because it 1326e7a5b4aaSLeon Romanovsky * is too early amd device is not initialized yet. 1327e7a5b4aaSLeon Romanovsky */ 1328e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, true); 13295f8f5499SParav Pandit ret = device_add(&device->dev); 13305f8f5499SParav Pandit if (ret) 13315f8f5499SParav Pandit goto cg_cleanup; 13325f8f5499SParav Pandit 1333ea4baf7fSParav Pandit ret = ib_device_register_sysfs(device); 13341da177e4SLinus Torvalds if (ret) { 133543c7c851SJason Gunthorpe dev_warn(&device->dev, 133643c7c851SJason Gunthorpe "Couldn't register device with driver model\n"); 13375f8f5499SParav Pandit goto dev_cleanup; 13381da177e4SLinus Torvalds } 13391da177e4SLinus Torvalds 1340d0899892SJason Gunthorpe ret = enable_device_and_get(device); 1341e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, false); 1342e7a5b4aaSLeon Romanovsky /* Mark for userspace that device is ready */ 1343e7a5b4aaSLeon Romanovsky kobject_uevent(&device->dev.kobj, KOBJ_ADD); 1344d0899892SJason Gunthorpe if (ret) { 1345d0899892SJason Gunthorpe void (*dealloc_fn)(struct ib_device *); 1346d0899892SJason Gunthorpe 1347d0899892SJason Gunthorpe /* 1348d0899892SJason Gunthorpe * If we hit this error flow then we don't want to 1349d0899892SJason Gunthorpe * automatically dealloc the device since the caller is 1350d0899892SJason Gunthorpe * expected to call ib_dealloc_device() after 1351d0899892SJason Gunthorpe * ib_register_device() fails. This is tricky due to the 1352d0899892SJason Gunthorpe * possibility for a parallel unregistration along with this 1353d0899892SJason Gunthorpe * error flow. Since we have a refcount here we know any 1354d0899892SJason Gunthorpe * parallel flow is stopped in disable_device and will see the 1355d0899892SJason Gunthorpe * NULL pointers, causing the responsibility to 1356d0899892SJason Gunthorpe * ib_dealloc_device() to revert back to this thread. 1357d0899892SJason Gunthorpe */ 1358d0899892SJason Gunthorpe dealloc_fn = device->ops.dealloc_driver; 1359d0899892SJason Gunthorpe device->ops.dealloc_driver = NULL; 1360d0899892SJason Gunthorpe ib_device_put(device); 1361d0899892SJason Gunthorpe __ib_unregister_device(device); 1362d0899892SJason Gunthorpe device->ops.dealloc_driver = dealloc_fn; 1363d0899892SJason Gunthorpe return ret; 1364d0899892SJason Gunthorpe } 1365d0899892SJason Gunthorpe ib_device_put(device); 13661da177e4SLinus Torvalds 13674be3a4faSParav Pandit return 0; 13684be3a4faSParav Pandit 13695f8f5499SParav Pandit dev_cleanup: 13705f8f5499SParav Pandit device_del(&device->dev); 13712fb4f4eaSParav Pandit cg_cleanup: 1372e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, false); 13732fb4f4eaSParav Pandit ib_device_unregister_rdmacg(device); 1374d45f89d5SJason Gunthorpe ib_cache_cleanup_one(device); 13751da177e4SLinus Torvalds return ret; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_device); 13781da177e4SLinus Torvalds 1379d0899892SJason Gunthorpe /* Callers must hold a get on the device. */ 1380d0899892SJason Gunthorpe static void __ib_unregister_device(struct ib_device *ib_dev) 1381d0899892SJason Gunthorpe { 1382d0899892SJason Gunthorpe /* 1383d0899892SJason Gunthorpe * We have a registration lock so that all the calls to unregister are 1384d0899892SJason Gunthorpe * fully fenced, once any unregister returns the device is truely 1385d0899892SJason Gunthorpe * unregistered even if multiple callers are unregistering it at the 1386d0899892SJason Gunthorpe * same time. This also interacts with the registration flow and 1387d0899892SJason Gunthorpe * provides sane semantics if register and unregister are racing. 1388d0899892SJason Gunthorpe */ 1389d0899892SJason Gunthorpe mutex_lock(&ib_dev->unregistration_lock); 1390d0899892SJason Gunthorpe if (!refcount_read(&ib_dev->refcount)) 1391d0899892SJason Gunthorpe goto out; 1392d0899892SJason Gunthorpe 1393d0899892SJason Gunthorpe disable_device(ib_dev); 13943042492bSParav Pandit 13953042492bSParav Pandit /* Expedite removing unregistered pointers from the hash table */ 13963042492bSParav Pandit free_netdevs(ib_dev); 13973042492bSParav Pandit 1398d0899892SJason Gunthorpe ib_device_unregister_sysfs(ib_dev); 1399d0899892SJason Gunthorpe device_del(&ib_dev->dev); 1400d0899892SJason Gunthorpe ib_device_unregister_rdmacg(ib_dev); 1401d0899892SJason Gunthorpe ib_cache_cleanup_one(ib_dev); 1402d0899892SJason Gunthorpe 1403d0899892SJason Gunthorpe /* 1404d0899892SJason Gunthorpe * Drivers using the new flow may not call ib_dealloc_device except 1405d0899892SJason Gunthorpe * in error unwind prior to registration success. 1406d0899892SJason Gunthorpe */ 1407d0899892SJason Gunthorpe if (ib_dev->ops.dealloc_driver) { 1408d0899892SJason Gunthorpe WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); 1409d0899892SJason Gunthorpe ib_dealloc_device(ib_dev); 1410d0899892SJason Gunthorpe } 1411d0899892SJason Gunthorpe out: 1412d0899892SJason Gunthorpe mutex_unlock(&ib_dev->unregistration_lock); 1413d0899892SJason Gunthorpe } 1414d0899892SJason Gunthorpe 14151da177e4SLinus Torvalds /** 14161da177e4SLinus Torvalds * ib_unregister_device - Unregister an IB device 1417d0899892SJason Gunthorpe * @device: The device to unregister 14181da177e4SLinus Torvalds * 14191da177e4SLinus Torvalds * Unregister an IB device. All clients will receive a remove callback. 1420d0899892SJason Gunthorpe * 1421d0899892SJason Gunthorpe * Callers should call this routine only once, and protect against races with 1422d0899892SJason Gunthorpe * registration. Typically it should only be called as part of a remove 1423d0899892SJason Gunthorpe * callback in an implementation of driver core's struct device_driver and 1424d0899892SJason Gunthorpe * related. 1425d0899892SJason Gunthorpe * 1426d0899892SJason Gunthorpe * If ops.dealloc_driver is used then ib_dev will be freed upon return from 1427d0899892SJason Gunthorpe * this function. 14281da177e4SLinus Torvalds */ 1429d0899892SJason Gunthorpe void ib_unregister_device(struct ib_device *ib_dev) 14301da177e4SLinus Torvalds { 1431d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1432d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1433d0899892SJason Gunthorpe put_device(&ib_dev->dev); 14341da177e4SLinus Torvalds } 14351da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_device); 14361da177e4SLinus Torvalds 1437d0899892SJason Gunthorpe /** 1438d0899892SJason Gunthorpe * ib_unregister_device_and_put - Unregister a device while holding a 'get' 1439d0899892SJason Gunthorpe * device: The device to unregister 1440d0899892SJason Gunthorpe * 1441d0899892SJason Gunthorpe * This is the same as ib_unregister_device(), except it includes an internal 1442d0899892SJason Gunthorpe * ib_device_put() that should match a 'get' obtained by the caller. 1443d0899892SJason Gunthorpe * 1444d0899892SJason Gunthorpe * It is safe to call this routine concurrently from multiple threads while 1445d0899892SJason Gunthorpe * holding the 'get'. When the function returns the device is fully 1446d0899892SJason Gunthorpe * unregistered. 1447d0899892SJason Gunthorpe * 1448d0899892SJason Gunthorpe * Drivers using this flow MUST use the driver_unregister callback to clean up 1449d0899892SJason Gunthorpe * their resources associated with the device and dealloc it. 1450d0899892SJason Gunthorpe */ 1451d0899892SJason Gunthorpe void ib_unregister_device_and_put(struct ib_device *ib_dev) 1452d0899892SJason Gunthorpe { 1453d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1454d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1455d0899892SJason Gunthorpe ib_device_put(ib_dev); 1456d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1457d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1458d0899892SJason Gunthorpe } 1459d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_and_put); 1460d0899892SJason Gunthorpe 1461d0899892SJason Gunthorpe /** 1462d0899892SJason Gunthorpe * ib_unregister_driver - Unregister all IB devices for a driver 1463d0899892SJason Gunthorpe * @driver_id: The driver to unregister 1464d0899892SJason Gunthorpe * 1465d0899892SJason Gunthorpe * This implements a fence for device unregistration. It only returns once all 1466d0899892SJason Gunthorpe * devices associated with the driver_id have fully completed their 1467d0899892SJason Gunthorpe * unregistration and returned from ib_unregister_device*(). 1468d0899892SJason Gunthorpe * 1469d0899892SJason Gunthorpe * If device's are not yet unregistered it goes ahead and starts unregistering 1470d0899892SJason Gunthorpe * them. 1471d0899892SJason Gunthorpe * 1472d0899892SJason Gunthorpe * This does not block creation of new devices with the given driver_id, that 1473d0899892SJason Gunthorpe * is the responsibility of the caller. 1474d0899892SJason Gunthorpe */ 1475d0899892SJason Gunthorpe void ib_unregister_driver(enum rdma_driver_id driver_id) 1476d0899892SJason Gunthorpe { 1477d0899892SJason Gunthorpe struct ib_device *ib_dev; 1478d0899892SJason Gunthorpe unsigned long index; 1479d0899892SJason Gunthorpe 1480d0899892SJason Gunthorpe down_read(&devices_rwsem); 1481d0899892SJason Gunthorpe xa_for_each (&devices, index, ib_dev) { 1482d0899892SJason Gunthorpe if (ib_dev->driver_id != driver_id) 1483d0899892SJason Gunthorpe continue; 1484d0899892SJason Gunthorpe 1485d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1486d0899892SJason Gunthorpe up_read(&devices_rwsem); 1487d0899892SJason Gunthorpe 1488d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1489d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1490d0899892SJason Gunthorpe 1491d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1492d0899892SJason Gunthorpe down_read(&devices_rwsem); 1493d0899892SJason Gunthorpe } 1494d0899892SJason Gunthorpe up_read(&devices_rwsem); 1495d0899892SJason Gunthorpe } 1496d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_driver); 1497d0899892SJason Gunthorpe 1498d0899892SJason Gunthorpe static void ib_unregister_work(struct work_struct *work) 1499d0899892SJason Gunthorpe { 1500d0899892SJason Gunthorpe struct ib_device *ib_dev = 1501d0899892SJason Gunthorpe container_of(work, struct ib_device, unregistration_work); 1502d0899892SJason Gunthorpe 1503d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1504d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1505d0899892SJason Gunthorpe } 1506d0899892SJason Gunthorpe 1507d0899892SJason Gunthorpe /** 1508d0899892SJason Gunthorpe * ib_unregister_device_queued - Unregister a device using a work queue 1509d0899892SJason Gunthorpe * device: The device to unregister 1510d0899892SJason Gunthorpe * 1511d0899892SJason Gunthorpe * This schedules an asynchronous unregistration using a WQ for the device. A 1512d0899892SJason Gunthorpe * driver should use this to avoid holding locks while doing unregistration, 1513d0899892SJason Gunthorpe * such as holding the RTNL lock. 1514d0899892SJason Gunthorpe * 1515d0899892SJason Gunthorpe * Drivers using this API must use ib_unregister_driver before module unload 1516d0899892SJason Gunthorpe * to ensure that all scheduled unregistrations have completed. 1517d0899892SJason Gunthorpe */ 1518d0899892SJason Gunthorpe void ib_unregister_device_queued(struct ib_device *ib_dev) 1519d0899892SJason Gunthorpe { 1520d0899892SJason Gunthorpe WARN_ON(!refcount_read(&ib_dev->refcount)); 1521d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1522d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1523d0899892SJason Gunthorpe if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) 1524d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1525d0899892SJason Gunthorpe } 1526d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_queued); 1527d0899892SJason Gunthorpe 1528decbc7a6SParav Pandit /* 1529decbc7a6SParav Pandit * The caller must pass in a device that has the kref held and the refcount 1530decbc7a6SParav Pandit * released. If the device is in cur_net and still registered then it is moved 1531decbc7a6SParav Pandit * into net. 1532decbc7a6SParav Pandit */ 1533decbc7a6SParav Pandit static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, 1534decbc7a6SParav Pandit struct net *net) 1535decbc7a6SParav Pandit { 1536decbc7a6SParav Pandit int ret2 = -EINVAL; 1537decbc7a6SParav Pandit int ret; 1538decbc7a6SParav Pandit 1539decbc7a6SParav Pandit mutex_lock(&device->unregistration_lock); 1540decbc7a6SParav Pandit 1541decbc7a6SParav Pandit /* 15422e5b8a01SParav Pandit * If a device not under ib_device_get() or if the unregistration_lock 15432e5b8a01SParav Pandit * is not held, the namespace can be changed, or it can be unregistered. 15442e5b8a01SParav Pandit * Check again under the lock. 1545decbc7a6SParav Pandit */ 1546decbc7a6SParav Pandit if (refcount_read(&device->refcount) == 0 || 1547decbc7a6SParav Pandit !net_eq(cur_net, read_pnet(&device->coredev.rdma_net))) { 1548decbc7a6SParav Pandit ret = -ENODEV; 1549decbc7a6SParav Pandit goto out; 1550decbc7a6SParav Pandit } 1551decbc7a6SParav Pandit 1552decbc7a6SParav Pandit kobject_uevent(&device->dev.kobj, KOBJ_REMOVE); 1553decbc7a6SParav Pandit disable_device(device); 1554decbc7a6SParav Pandit 1555decbc7a6SParav Pandit /* 1556decbc7a6SParav Pandit * At this point no one can be using the device, so it is safe to 1557decbc7a6SParav Pandit * change the namespace. 1558decbc7a6SParav Pandit */ 1559decbc7a6SParav Pandit write_pnet(&device->coredev.rdma_net, net); 1560decbc7a6SParav Pandit 15612e5b8a01SParav Pandit down_read(&devices_rwsem); 1562decbc7a6SParav Pandit /* 1563decbc7a6SParav Pandit * Currently rdma devices are system wide unique. So the device name 1564decbc7a6SParav Pandit * is guaranteed free in the new namespace. Publish the new namespace 1565decbc7a6SParav Pandit * at the sysfs level. 1566decbc7a6SParav Pandit */ 1567decbc7a6SParav Pandit ret = device_rename(&device->dev, dev_name(&device->dev)); 1568decbc7a6SParav Pandit up_read(&devices_rwsem); 1569decbc7a6SParav Pandit if (ret) { 1570decbc7a6SParav Pandit dev_warn(&device->dev, 1571decbc7a6SParav Pandit "%s: Couldn't rename device after namespace change\n", 1572decbc7a6SParav Pandit __func__); 1573decbc7a6SParav Pandit /* Try and put things back and re-enable the device */ 1574decbc7a6SParav Pandit write_pnet(&device->coredev.rdma_net, cur_net); 1575decbc7a6SParav Pandit } 1576decbc7a6SParav Pandit 1577decbc7a6SParav Pandit ret2 = enable_device_and_get(device); 15782e5b8a01SParav Pandit if (ret2) { 1579decbc7a6SParav Pandit /* 1580decbc7a6SParav Pandit * This shouldn't really happen, but if it does, let the user 1581decbc7a6SParav Pandit * retry at later point. So don't disable the device. 1582decbc7a6SParav Pandit */ 1583decbc7a6SParav Pandit dev_warn(&device->dev, 1584decbc7a6SParav Pandit "%s: Couldn't re-enable device after namespace change\n", 1585decbc7a6SParav Pandit __func__); 15862e5b8a01SParav Pandit } 1587decbc7a6SParav Pandit kobject_uevent(&device->dev.kobj, KOBJ_ADD); 15882e5b8a01SParav Pandit 1589decbc7a6SParav Pandit ib_device_put(device); 1590decbc7a6SParav Pandit out: 1591decbc7a6SParav Pandit mutex_unlock(&device->unregistration_lock); 1592decbc7a6SParav Pandit if (ret) 1593decbc7a6SParav Pandit return ret; 1594decbc7a6SParav Pandit return ret2; 1595decbc7a6SParav Pandit } 1596decbc7a6SParav Pandit 15972e5b8a01SParav Pandit int ib_device_set_netns_put(struct sk_buff *skb, 15982e5b8a01SParav Pandit struct ib_device *dev, u32 ns_fd) 15992e5b8a01SParav Pandit { 16002e5b8a01SParav Pandit struct net *net; 16012e5b8a01SParav Pandit int ret; 16022e5b8a01SParav Pandit 16032e5b8a01SParav Pandit net = get_net_ns_by_fd(ns_fd); 16042e5b8a01SParav Pandit if (IS_ERR(net)) { 16052e5b8a01SParav Pandit ret = PTR_ERR(net); 16062e5b8a01SParav Pandit goto net_err; 16072e5b8a01SParav Pandit } 16082e5b8a01SParav Pandit 16092e5b8a01SParav Pandit if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) { 16102e5b8a01SParav Pandit ret = -EPERM; 16112e5b8a01SParav Pandit goto ns_err; 16122e5b8a01SParav Pandit } 16132e5b8a01SParav Pandit 16142e5b8a01SParav Pandit /* 16152e5b8a01SParav Pandit * Currently supported only for those providers which support 16162e5b8a01SParav Pandit * disassociation and don't do port specific sysfs init. Once a 16172e5b8a01SParav Pandit * port_cleanup infrastructure is implemented, this limitation will be 16182e5b8a01SParav Pandit * removed. 16192e5b8a01SParav Pandit */ 16202e5b8a01SParav Pandit if (!dev->ops.disassociate_ucontext || dev->ops.init_port || 16212e5b8a01SParav Pandit ib_devices_shared_netns) { 16222e5b8a01SParav Pandit ret = -EOPNOTSUPP; 16232e5b8a01SParav Pandit goto ns_err; 16242e5b8a01SParav Pandit } 16252e5b8a01SParav Pandit 16262e5b8a01SParav Pandit get_device(&dev->dev); 16272e5b8a01SParav Pandit ib_device_put(dev); 16282e5b8a01SParav Pandit ret = rdma_dev_change_netns(dev, current->nsproxy->net_ns, net); 16292e5b8a01SParav Pandit put_device(&dev->dev); 16302e5b8a01SParav Pandit 16312e5b8a01SParav Pandit put_net(net); 16322e5b8a01SParav Pandit return ret; 16332e5b8a01SParav Pandit 16342e5b8a01SParav Pandit ns_err: 16352e5b8a01SParav Pandit put_net(net); 16362e5b8a01SParav Pandit net_err: 16372e5b8a01SParav Pandit ib_device_put(dev); 16382e5b8a01SParav Pandit return ret; 16392e5b8a01SParav Pandit } 16402e5b8a01SParav Pandit 16414e0f7b90SParav Pandit static struct pernet_operations rdma_dev_net_ops = { 16424e0f7b90SParav Pandit .init = rdma_dev_init_net, 16434e0f7b90SParav Pandit .exit = rdma_dev_exit_net, 16444e0f7b90SParav Pandit .id = &rdma_dev_net_id, 16454e0f7b90SParav Pandit .size = sizeof(struct rdma_dev_net), 16464e0f7b90SParav Pandit }; 16474e0f7b90SParav Pandit 1648e59178d8SJason Gunthorpe static int assign_client_id(struct ib_client *client) 1649e59178d8SJason Gunthorpe { 1650e59178d8SJason Gunthorpe int ret; 1651e59178d8SJason Gunthorpe 1652921eab11SJason Gunthorpe down_write(&clients_rwsem); 1653e59178d8SJason Gunthorpe /* 1654e59178d8SJason Gunthorpe * The add/remove callbacks must be called in FIFO/LIFO order. To 1655e59178d8SJason Gunthorpe * achieve this we assign client_ids so they are sorted in 1656e59178d8SJason Gunthorpe * registration order, and retain a linked list we can reverse iterate 1657e59178d8SJason Gunthorpe * to get the LIFO order. The extra linked list can go away if xarray 1658e59178d8SJason Gunthorpe * learns to reverse iterate. 1659e59178d8SJason Gunthorpe */ 1660ea295481SLinus Torvalds if (list_empty(&client_list)) { 1661e59178d8SJason Gunthorpe client->client_id = 0; 1662ea295481SLinus Torvalds } else { 1663ea295481SLinus Torvalds struct ib_client *last; 1664ea295481SLinus Torvalds 1665ea295481SLinus Torvalds last = list_last_entry(&client_list, struct ib_client, list); 1666ea295481SLinus Torvalds client->client_id = last->client_id + 1; 1667ea295481SLinus Torvalds } 1668ea295481SLinus Torvalds ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); 1669e59178d8SJason Gunthorpe if (ret) 1670e59178d8SJason Gunthorpe goto out; 1671e59178d8SJason Gunthorpe 1672921eab11SJason Gunthorpe xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); 1673921eab11SJason Gunthorpe list_add_tail(&client->list, &client_list); 1674921eab11SJason Gunthorpe 1675e59178d8SJason Gunthorpe out: 1676921eab11SJason Gunthorpe up_write(&clients_rwsem); 1677e59178d8SJason Gunthorpe return ret; 1678e59178d8SJason Gunthorpe } 1679e59178d8SJason Gunthorpe 16801da177e4SLinus Torvalds /** 16811da177e4SLinus Torvalds * ib_register_client - Register an IB client 16821da177e4SLinus Torvalds * @client:Client to register 16831da177e4SLinus Torvalds * 16841da177e4SLinus Torvalds * Upper level users of the IB drivers can use ib_register_client() to 16851da177e4SLinus Torvalds * register callbacks for IB device addition and removal. When an IB 16861da177e4SLinus Torvalds * device is added, each registered client's add method will be called 16871da177e4SLinus Torvalds * (in the order the clients were registered), and when a device is 16881da177e4SLinus Torvalds * removed, each client's remove method will be called (in the reverse 16891da177e4SLinus Torvalds * order that clients were registered). In addition, when 16901da177e4SLinus Torvalds * ib_register_client() is called, the client will receive an add 16911da177e4SLinus Torvalds * callback for all devices already registered. 16921da177e4SLinus Torvalds */ 16931da177e4SLinus Torvalds int ib_register_client(struct ib_client *client) 16941da177e4SLinus Torvalds { 16951da177e4SLinus Torvalds struct ib_device *device; 16960df91bb6SJason Gunthorpe unsigned long index; 1697e59178d8SJason Gunthorpe int ret; 16981da177e4SLinus Torvalds 1699e59178d8SJason Gunthorpe ret = assign_client_id(client); 1700921eab11SJason Gunthorpe if (ret) 1701921eab11SJason Gunthorpe return ret; 1702921eab11SJason Gunthorpe 1703921eab11SJason Gunthorpe down_read(&devices_rwsem); 1704921eab11SJason Gunthorpe xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { 1705921eab11SJason Gunthorpe ret = add_client_context(device, client); 1706e59178d8SJason Gunthorpe if (ret) { 1707921eab11SJason Gunthorpe up_read(&devices_rwsem); 1708921eab11SJason Gunthorpe ib_unregister_client(client); 1709e59178d8SJason Gunthorpe return ret; 1710e59178d8SJason Gunthorpe } 1711921eab11SJason Gunthorpe } 1712921eab11SJason Gunthorpe up_read(&devices_rwsem); 17131da177e4SLinus Torvalds return 0; 17141da177e4SLinus Torvalds } 17151da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_client); 17161da177e4SLinus Torvalds 17171da177e4SLinus Torvalds /** 17181da177e4SLinus Torvalds * ib_unregister_client - Unregister an IB client 17191da177e4SLinus Torvalds * @client:Client to unregister 17201da177e4SLinus Torvalds * 17211da177e4SLinus Torvalds * Upper level users use ib_unregister_client() to remove their client 17221da177e4SLinus Torvalds * registration. When ib_unregister_client() is called, the client 17231da177e4SLinus Torvalds * will receive a remove callback for each IB device still registered. 1724921eab11SJason Gunthorpe * 1725921eab11SJason Gunthorpe * This is a full fence, once it returns no client callbacks will be called, 1726921eab11SJason Gunthorpe * or are running in another thread. 17271da177e4SLinus Torvalds */ 17281da177e4SLinus Torvalds void ib_unregister_client(struct ib_client *client) 17291da177e4SLinus Torvalds { 17301da177e4SLinus Torvalds struct ib_device *device; 17310df91bb6SJason Gunthorpe unsigned long index; 17321da177e4SLinus Torvalds 1733921eab11SJason Gunthorpe down_write(&clients_rwsem); 1734e59178d8SJason Gunthorpe xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); 1735921eab11SJason Gunthorpe up_write(&clients_rwsem); 1736921eab11SJason Gunthorpe /* 1737921eab11SJason Gunthorpe * Every device still known must be serialized to make sure we are 1738921eab11SJason Gunthorpe * done with the client callbacks before we return. 1739921eab11SJason Gunthorpe */ 1740921eab11SJason Gunthorpe down_read(&devices_rwsem); 1741921eab11SJason Gunthorpe xa_for_each (&devices, index, device) 1742921eab11SJason Gunthorpe remove_client_context(device, client->client_id); 1743921eab11SJason Gunthorpe up_read(&devices_rwsem); 17445aa44bb9SHaggai Eran 1745921eab11SJason Gunthorpe down_write(&clients_rwsem); 1746e59178d8SJason Gunthorpe list_del(&client->list); 1747e59178d8SJason Gunthorpe xa_erase(&clients, client->client_id); 1748921eab11SJason Gunthorpe up_write(&clients_rwsem); 17491da177e4SLinus Torvalds } 17501da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_client); 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds /** 17539cd330d3SKrishna Kumar * ib_set_client_data - Set IB client context 17541da177e4SLinus Torvalds * @device:Device to set context for 17551da177e4SLinus Torvalds * @client:Client to set context for 17561da177e4SLinus Torvalds * @data:Context to set 17571da177e4SLinus Torvalds * 17580df91bb6SJason Gunthorpe * ib_set_client_data() sets client context data that can be retrieved with 17590df91bb6SJason Gunthorpe * ib_get_client_data(). This can only be called while the client is 17600df91bb6SJason Gunthorpe * registered to the device, once the ib_client remove() callback returns this 17610df91bb6SJason Gunthorpe * cannot be called. 17621da177e4SLinus Torvalds */ 17631da177e4SLinus Torvalds void ib_set_client_data(struct ib_device *device, struct ib_client *client, 17641da177e4SLinus Torvalds void *data) 17651da177e4SLinus Torvalds { 17660df91bb6SJason Gunthorpe void *rc; 17671da177e4SLinus Torvalds 17680df91bb6SJason Gunthorpe if (WARN_ON(IS_ERR(data))) 17690df91bb6SJason Gunthorpe data = NULL; 17701da177e4SLinus Torvalds 17710df91bb6SJason Gunthorpe rc = xa_store(&device->client_data, client->client_id, data, 17720df91bb6SJason Gunthorpe GFP_KERNEL); 17730df91bb6SJason Gunthorpe WARN_ON(xa_is_err(rc)); 17741da177e4SLinus Torvalds } 17751da177e4SLinus Torvalds EXPORT_SYMBOL(ib_set_client_data); 17761da177e4SLinus Torvalds 17771da177e4SLinus Torvalds /** 17781da177e4SLinus Torvalds * ib_register_event_handler - Register an IB event handler 17791da177e4SLinus Torvalds * @event_handler:Handler to register 17801da177e4SLinus Torvalds * 17811da177e4SLinus Torvalds * ib_register_event_handler() registers an event handler that will be 17821da177e4SLinus Torvalds * called back when asynchronous IB events occur (as defined in 17831da177e4SLinus Torvalds * chapter 11 of the InfiniBand Architecture Specification). This 17841da177e4SLinus Torvalds * callback may occur in interrupt context. 17851da177e4SLinus Torvalds */ 1786dcc9881eSLeon Romanovsky void ib_register_event_handler(struct ib_event_handler *event_handler) 17871da177e4SLinus Torvalds { 17881da177e4SLinus Torvalds unsigned long flags; 17891da177e4SLinus Torvalds 17901da177e4SLinus Torvalds spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 17911da177e4SLinus Torvalds list_add_tail(&event_handler->list, 17921da177e4SLinus Torvalds &event_handler->device->event_handler_list); 17931da177e4SLinus Torvalds spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 17941da177e4SLinus Torvalds } 17951da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_event_handler); 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds /** 17981da177e4SLinus Torvalds * ib_unregister_event_handler - Unregister an event handler 17991da177e4SLinus Torvalds * @event_handler:Handler to unregister 18001da177e4SLinus Torvalds * 18011da177e4SLinus Torvalds * Unregister an event handler registered with 18021da177e4SLinus Torvalds * ib_register_event_handler(). 18031da177e4SLinus Torvalds */ 1804dcc9881eSLeon Romanovsky void ib_unregister_event_handler(struct ib_event_handler *event_handler) 18051da177e4SLinus Torvalds { 18061da177e4SLinus Torvalds unsigned long flags; 18071da177e4SLinus Torvalds 18081da177e4SLinus Torvalds spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 18091da177e4SLinus Torvalds list_del(&event_handler->list); 18101da177e4SLinus Torvalds spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_event_handler); 18131da177e4SLinus Torvalds 18141da177e4SLinus Torvalds /** 18151da177e4SLinus Torvalds * ib_dispatch_event - Dispatch an asynchronous event 18161da177e4SLinus Torvalds * @event:Event to dispatch 18171da177e4SLinus Torvalds * 18181da177e4SLinus Torvalds * Low-level drivers must call ib_dispatch_event() to dispatch the 18191da177e4SLinus Torvalds * event to all registered event handlers when an asynchronous event 18201da177e4SLinus Torvalds * occurs. 18211da177e4SLinus Torvalds */ 18221da177e4SLinus Torvalds void ib_dispatch_event(struct ib_event *event) 18231da177e4SLinus Torvalds { 18241da177e4SLinus Torvalds unsigned long flags; 18251da177e4SLinus Torvalds struct ib_event_handler *handler; 18261da177e4SLinus Torvalds 18271da177e4SLinus Torvalds spin_lock_irqsave(&event->device->event_handler_lock, flags); 18281da177e4SLinus Torvalds 18291da177e4SLinus Torvalds list_for_each_entry(handler, &event->device->event_handler_list, list) 18301da177e4SLinus Torvalds handler->handler(handler, event); 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds spin_unlock_irqrestore(&event->device->event_handler_lock, flags); 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds EXPORT_SYMBOL(ib_dispatch_event); 18351da177e4SLinus Torvalds 18361da177e4SLinus Torvalds /** 18371da177e4SLinus Torvalds * ib_query_port - Query IB port attributes 18381da177e4SLinus Torvalds * @device:Device to query 18391da177e4SLinus Torvalds * @port_num:Port number to query 18401da177e4SLinus Torvalds * @port_attr:Port attributes 18411da177e4SLinus Torvalds * 18421da177e4SLinus Torvalds * ib_query_port() returns the attributes of a port through the 18431da177e4SLinus Torvalds * @port_attr pointer. 18441da177e4SLinus Torvalds */ 18451da177e4SLinus Torvalds int ib_query_port(struct ib_device *device, 18461da177e4SLinus Torvalds u8 port_num, 18471da177e4SLinus Torvalds struct ib_port_attr *port_attr) 18481da177e4SLinus Torvalds { 1849fad61ad4SEli Cohen union ib_gid gid; 1850fad61ad4SEli Cohen int err; 1851fad61ad4SEli Cohen 185224dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 1853116c0074SRoland Dreier return -EINVAL; 1854116c0074SRoland Dreier 1855fad61ad4SEli Cohen memset(port_attr, 0, sizeof(*port_attr)); 18563023a1e9SKamal Heib err = device->ops.query_port(device, port_num, port_attr); 1857fad61ad4SEli Cohen if (err || port_attr->subnet_prefix) 1858fad61ad4SEli Cohen return err; 1859fad61ad4SEli Cohen 1860d7012467SEli Cohen if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND) 1861d7012467SEli Cohen return 0; 1862d7012467SEli Cohen 18633023a1e9SKamal Heib err = device->ops.query_gid(device, port_num, 0, &gid); 1864fad61ad4SEli Cohen if (err) 1865fad61ad4SEli Cohen return err; 1866fad61ad4SEli Cohen 1867fad61ad4SEli Cohen port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); 1868fad61ad4SEli Cohen return 0; 18691da177e4SLinus Torvalds } 18701da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_port); 18711da177e4SLinus Torvalds 1872324e227eSJason Gunthorpe static void add_ndev_hash(struct ib_port_data *pdata) 1873324e227eSJason Gunthorpe { 1874324e227eSJason Gunthorpe unsigned long flags; 1875324e227eSJason Gunthorpe 1876324e227eSJason Gunthorpe might_sleep(); 1877324e227eSJason Gunthorpe 1878324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 1879324e227eSJason Gunthorpe if (hash_hashed(&pdata->ndev_hash_link)) { 1880324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 1881324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 1882324e227eSJason Gunthorpe /* 1883324e227eSJason Gunthorpe * We cannot do hash_add_rcu after a hash_del_rcu until the 1884324e227eSJason Gunthorpe * grace period 1885324e227eSJason Gunthorpe */ 1886324e227eSJason Gunthorpe synchronize_rcu(); 1887324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 1888324e227eSJason Gunthorpe } 1889324e227eSJason Gunthorpe if (pdata->netdev) 1890324e227eSJason Gunthorpe hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, 1891324e227eSJason Gunthorpe (uintptr_t)pdata->netdev); 1892324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 1893324e227eSJason Gunthorpe } 1894324e227eSJason Gunthorpe 18951da177e4SLinus Torvalds /** 1896c2261dd7SJason Gunthorpe * ib_device_set_netdev - Associate the ib_dev with an underlying net_device 1897c2261dd7SJason Gunthorpe * @ib_dev: Device to modify 1898c2261dd7SJason Gunthorpe * @ndev: net_device to affiliate, may be NULL 1899c2261dd7SJason Gunthorpe * @port: IB port the net_device is connected to 1900c2261dd7SJason Gunthorpe * 1901c2261dd7SJason Gunthorpe * Drivers should use this to link the ib_device to a netdev so the netdev 1902c2261dd7SJason Gunthorpe * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be 1903c2261dd7SJason Gunthorpe * affiliated with any port. 1904c2261dd7SJason Gunthorpe * 1905c2261dd7SJason Gunthorpe * The caller must ensure that the given ndev is not unregistered or 1906c2261dd7SJason Gunthorpe * unregistering, and that either the ib_device is unregistered or 1907c2261dd7SJason Gunthorpe * ib_device_set_netdev() is called with NULL when the ndev sends a 1908c2261dd7SJason Gunthorpe * NETDEV_UNREGISTER event. 1909c2261dd7SJason Gunthorpe */ 1910c2261dd7SJason Gunthorpe int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 1911c2261dd7SJason Gunthorpe unsigned int port) 1912c2261dd7SJason Gunthorpe { 1913c2261dd7SJason Gunthorpe struct net_device *old_ndev; 1914c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 1915c2261dd7SJason Gunthorpe unsigned long flags; 1916c2261dd7SJason Gunthorpe int ret; 1917c2261dd7SJason Gunthorpe 1918c2261dd7SJason Gunthorpe /* 1919c2261dd7SJason Gunthorpe * Drivers wish to call this before ib_register_driver, so we have to 1920c2261dd7SJason Gunthorpe * setup the port data early. 1921c2261dd7SJason Gunthorpe */ 1922c2261dd7SJason Gunthorpe ret = alloc_port_data(ib_dev); 1923c2261dd7SJason Gunthorpe if (ret) 1924c2261dd7SJason Gunthorpe return ret; 1925c2261dd7SJason Gunthorpe 1926c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 1927c2261dd7SJason Gunthorpe return -EINVAL; 1928c2261dd7SJason Gunthorpe 1929c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 1930c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 1931324e227eSJason Gunthorpe old_ndev = rcu_dereference_protected( 1932324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 1933324e227eSJason Gunthorpe if (old_ndev == ndev) { 1934c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 1935c2261dd7SJason Gunthorpe return 0; 1936c2261dd7SJason Gunthorpe } 1937c2261dd7SJason Gunthorpe 1938c2261dd7SJason Gunthorpe if (ndev) 1939c2261dd7SJason Gunthorpe dev_hold(ndev); 1940324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, ndev); 1941c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 1942c2261dd7SJason Gunthorpe 1943324e227eSJason Gunthorpe add_ndev_hash(pdata); 1944c2261dd7SJason Gunthorpe if (old_ndev) 1945c2261dd7SJason Gunthorpe dev_put(old_ndev); 1946c2261dd7SJason Gunthorpe 1947c2261dd7SJason Gunthorpe return 0; 1948c2261dd7SJason Gunthorpe } 1949c2261dd7SJason Gunthorpe EXPORT_SYMBOL(ib_device_set_netdev); 1950c2261dd7SJason Gunthorpe 1951c2261dd7SJason Gunthorpe static void free_netdevs(struct ib_device *ib_dev) 1952c2261dd7SJason Gunthorpe { 1953c2261dd7SJason Gunthorpe unsigned long flags; 1954c2261dd7SJason Gunthorpe unsigned int port; 1955c2261dd7SJason Gunthorpe 195646bdf370SKamal Heib if (!ib_dev->port_data) 195746bdf370SKamal Heib return; 195846bdf370SKamal Heib 1959c2261dd7SJason Gunthorpe rdma_for_each_port (ib_dev, port) { 1960c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &ib_dev->port_data[port]; 1961324e227eSJason Gunthorpe struct net_device *ndev; 1962c2261dd7SJason Gunthorpe 1963c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 1964324e227eSJason Gunthorpe ndev = rcu_dereference_protected( 1965324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 1966324e227eSJason Gunthorpe if (ndev) { 1967324e227eSJason Gunthorpe spin_lock(&ndev_hash_lock); 1968324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 1969324e227eSJason Gunthorpe spin_unlock(&ndev_hash_lock); 1970324e227eSJason Gunthorpe 1971324e227eSJason Gunthorpe /* 1972324e227eSJason Gunthorpe * If this is the last dev_put there is still a 1973324e227eSJason Gunthorpe * synchronize_rcu before the netdev is kfreed, so we 1974324e227eSJason Gunthorpe * can continue to rely on unlocked pointer 1975324e227eSJason Gunthorpe * comparisons after the put 1976324e227eSJason Gunthorpe */ 1977324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, NULL); 1978324e227eSJason Gunthorpe dev_put(ndev); 1979c2261dd7SJason Gunthorpe } 1980c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 1981c2261dd7SJason Gunthorpe } 1982c2261dd7SJason Gunthorpe } 1983c2261dd7SJason Gunthorpe 1984c2261dd7SJason Gunthorpe struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, 1985c2261dd7SJason Gunthorpe unsigned int port) 1986c2261dd7SJason Gunthorpe { 1987c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 1988c2261dd7SJason Gunthorpe struct net_device *res; 1989c2261dd7SJason Gunthorpe 1990c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 1991c2261dd7SJason Gunthorpe return NULL; 1992c2261dd7SJason Gunthorpe 1993c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 1994c2261dd7SJason Gunthorpe 1995c2261dd7SJason Gunthorpe /* 1996c2261dd7SJason Gunthorpe * New drivers should use ib_device_set_netdev() not the legacy 1997c2261dd7SJason Gunthorpe * get_netdev(). 1998c2261dd7SJason Gunthorpe */ 1999c2261dd7SJason Gunthorpe if (ib_dev->ops.get_netdev) 2000c2261dd7SJason Gunthorpe res = ib_dev->ops.get_netdev(ib_dev, port); 2001c2261dd7SJason Gunthorpe else { 2002c2261dd7SJason Gunthorpe spin_lock(&pdata->netdev_lock); 2003324e227eSJason Gunthorpe res = rcu_dereference_protected( 2004324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 2005c2261dd7SJason Gunthorpe if (res) 2006c2261dd7SJason Gunthorpe dev_hold(res); 2007c2261dd7SJason Gunthorpe spin_unlock(&pdata->netdev_lock); 2008c2261dd7SJason Gunthorpe } 2009c2261dd7SJason Gunthorpe 2010c2261dd7SJason Gunthorpe /* 2011c2261dd7SJason Gunthorpe * If we are starting to unregister expedite things by preventing 2012c2261dd7SJason Gunthorpe * propagation of an unregistering netdev. 2013c2261dd7SJason Gunthorpe */ 2014c2261dd7SJason Gunthorpe if (res && res->reg_state != NETREG_REGISTERED) { 2015c2261dd7SJason Gunthorpe dev_put(res); 2016c2261dd7SJason Gunthorpe return NULL; 2017c2261dd7SJason Gunthorpe } 2018c2261dd7SJason Gunthorpe 2019c2261dd7SJason Gunthorpe return res; 2020c2261dd7SJason Gunthorpe } 2021c2261dd7SJason Gunthorpe 2022c2261dd7SJason Gunthorpe /** 2023324e227eSJason Gunthorpe * ib_device_get_by_netdev - Find an IB device associated with a netdev 2024324e227eSJason Gunthorpe * @ndev: netdev to locate 2025324e227eSJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 2026324e227eSJason Gunthorpe * 2027324e227eSJason Gunthorpe * Find and hold an ib_device that is associated with a netdev via 2028324e227eSJason Gunthorpe * ib_device_set_netdev(). The caller must call ib_device_put() on the 2029324e227eSJason Gunthorpe * returned pointer. 2030324e227eSJason Gunthorpe */ 2031324e227eSJason Gunthorpe struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 2032324e227eSJason Gunthorpe enum rdma_driver_id driver_id) 2033324e227eSJason Gunthorpe { 2034324e227eSJason Gunthorpe struct ib_device *res = NULL; 2035324e227eSJason Gunthorpe struct ib_port_data *cur; 2036324e227eSJason Gunthorpe 2037324e227eSJason Gunthorpe rcu_read_lock(); 2038324e227eSJason Gunthorpe hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, 2039324e227eSJason Gunthorpe (uintptr_t)ndev) { 2040324e227eSJason Gunthorpe if (rcu_access_pointer(cur->netdev) == ndev && 2041324e227eSJason Gunthorpe (driver_id == RDMA_DRIVER_UNKNOWN || 2042324e227eSJason Gunthorpe cur->ib_dev->driver_id == driver_id) && 2043324e227eSJason Gunthorpe ib_device_try_get(cur->ib_dev)) { 2044324e227eSJason Gunthorpe res = cur->ib_dev; 2045324e227eSJason Gunthorpe break; 2046324e227eSJason Gunthorpe } 2047324e227eSJason Gunthorpe } 2048324e227eSJason Gunthorpe rcu_read_unlock(); 2049324e227eSJason Gunthorpe 2050324e227eSJason Gunthorpe return res; 2051324e227eSJason Gunthorpe } 2052324e227eSJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_netdev); 2053324e227eSJason Gunthorpe 2054324e227eSJason Gunthorpe /** 205503db3a2dSMatan Barak * ib_enum_roce_netdev - enumerate all RoCE ports 205603db3a2dSMatan Barak * @ib_dev : IB device we want to query 205703db3a2dSMatan Barak * @filter: Should we call the callback? 205803db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 205903db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 206003db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 206103db3a2dSMatan Barak * 206203db3a2dSMatan Barak * Enumerates all of the physical RoCE ports of ib_dev 206303db3a2dSMatan Barak * which are related to netdevice and calls callback() on each 206403db3a2dSMatan Barak * device for which filter() function returns non zero. 206503db3a2dSMatan Barak */ 206603db3a2dSMatan Barak void ib_enum_roce_netdev(struct ib_device *ib_dev, 206703db3a2dSMatan Barak roce_netdev_filter filter, 206803db3a2dSMatan Barak void *filter_cookie, 206903db3a2dSMatan Barak roce_netdev_callback cb, 207003db3a2dSMatan Barak void *cookie) 207103db3a2dSMatan Barak { 2072ea1075edSJason Gunthorpe unsigned int port; 207303db3a2dSMatan Barak 2074ea1075edSJason Gunthorpe rdma_for_each_port (ib_dev, port) 207503db3a2dSMatan Barak if (rdma_protocol_roce(ib_dev, port)) { 2076c2261dd7SJason Gunthorpe struct net_device *idev = 2077c2261dd7SJason Gunthorpe ib_device_get_netdev(ib_dev, port); 207803db3a2dSMatan Barak 207903db3a2dSMatan Barak if (filter(ib_dev, port, idev, filter_cookie)) 208003db3a2dSMatan Barak cb(ib_dev, port, idev, cookie); 208103db3a2dSMatan Barak 208203db3a2dSMatan Barak if (idev) 208303db3a2dSMatan Barak dev_put(idev); 208403db3a2dSMatan Barak } 208503db3a2dSMatan Barak } 208603db3a2dSMatan Barak 208703db3a2dSMatan Barak /** 208803db3a2dSMatan Barak * ib_enum_all_roce_netdevs - enumerate all RoCE devices 208903db3a2dSMatan Barak * @filter: Should we call the callback? 209003db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 209103db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 209203db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 209303db3a2dSMatan Barak * 209403db3a2dSMatan Barak * Enumerates all RoCE devices' physical ports which are related 209503db3a2dSMatan Barak * to netdevices and calls callback() on each device for which 209603db3a2dSMatan Barak * filter() function returns non zero. 209703db3a2dSMatan Barak */ 209803db3a2dSMatan Barak void ib_enum_all_roce_netdevs(roce_netdev_filter filter, 209903db3a2dSMatan Barak void *filter_cookie, 210003db3a2dSMatan Barak roce_netdev_callback cb, 210103db3a2dSMatan Barak void *cookie) 210203db3a2dSMatan Barak { 210303db3a2dSMatan Barak struct ib_device *dev; 21040df91bb6SJason Gunthorpe unsigned long index; 210503db3a2dSMatan Barak 2106921eab11SJason Gunthorpe down_read(&devices_rwsem); 21070df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) 210803db3a2dSMatan Barak ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); 2109921eab11SJason Gunthorpe up_read(&devices_rwsem); 211003db3a2dSMatan Barak } 211103db3a2dSMatan Barak 211203db3a2dSMatan Barak /** 21138030c835SLeon Romanovsky * ib_enum_all_devs - enumerate all ib_devices 21148030c835SLeon Romanovsky * @cb: Callback to call for each found ib_device 21158030c835SLeon Romanovsky * 21168030c835SLeon Romanovsky * Enumerates all ib_devices and calls callback() on each device. 21178030c835SLeon Romanovsky */ 21188030c835SLeon Romanovsky int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, 21198030c835SLeon Romanovsky struct netlink_callback *cb) 21208030c835SLeon Romanovsky { 21210df91bb6SJason Gunthorpe unsigned long index; 21228030c835SLeon Romanovsky struct ib_device *dev; 21238030c835SLeon Romanovsky unsigned int idx = 0; 21248030c835SLeon Romanovsky int ret = 0; 21258030c835SLeon Romanovsky 2126921eab11SJason Gunthorpe down_read(&devices_rwsem); 21270df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 212837eeab55SParav Pandit if (!rdma_dev_access_netns(dev, sock_net(skb->sk))) 212937eeab55SParav Pandit continue; 213037eeab55SParav Pandit 21318030c835SLeon Romanovsky ret = nldev_cb(dev, skb, cb, idx); 21328030c835SLeon Romanovsky if (ret) 21338030c835SLeon Romanovsky break; 21348030c835SLeon Romanovsky idx++; 21358030c835SLeon Romanovsky } 2136921eab11SJason Gunthorpe up_read(&devices_rwsem); 21378030c835SLeon Romanovsky return ret; 21388030c835SLeon Romanovsky } 21398030c835SLeon Romanovsky 21408030c835SLeon Romanovsky /** 21411da177e4SLinus Torvalds * ib_query_pkey - Get P_Key table entry 21421da177e4SLinus Torvalds * @device:Device to query 21431da177e4SLinus Torvalds * @port_num:Port number to query 21441da177e4SLinus Torvalds * @index:P_Key table index to query 21451da177e4SLinus Torvalds * @pkey:Returned P_Key 21461da177e4SLinus Torvalds * 21471da177e4SLinus Torvalds * ib_query_pkey() fetches the specified P_Key table entry. 21481da177e4SLinus Torvalds */ 21491da177e4SLinus Torvalds int ib_query_pkey(struct ib_device *device, 21501da177e4SLinus Torvalds u8 port_num, u16 index, u16 *pkey) 21511da177e4SLinus Torvalds { 21529af3f5cfSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 21539af3f5cfSYuval Shaia return -EINVAL; 21549af3f5cfSYuval Shaia 21553023a1e9SKamal Heib return device->ops.query_pkey(device, port_num, index, pkey); 21561da177e4SLinus Torvalds } 21571da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_pkey); 21581da177e4SLinus Torvalds 21591da177e4SLinus Torvalds /** 21601da177e4SLinus Torvalds * ib_modify_device - Change IB device attributes 21611da177e4SLinus Torvalds * @device:Device to modify 21621da177e4SLinus Torvalds * @device_modify_mask:Mask of attributes to change 21631da177e4SLinus Torvalds * @device_modify:New attribute values 21641da177e4SLinus Torvalds * 21651da177e4SLinus Torvalds * ib_modify_device() changes a device's attributes as specified by 21661da177e4SLinus Torvalds * the @device_modify_mask and @device_modify structure. 21671da177e4SLinus Torvalds */ 21681da177e4SLinus Torvalds int ib_modify_device(struct ib_device *device, 21691da177e4SLinus Torvalds int device_modify_mask, 21701da177e4SLinus Torvalds struct ib_device_modify *device_modify) 21711da177e4SLinus Torvalds { 21723023a1e9SKamal Heib if (!device->ops.modify_device) 217310e1b54bSBart Van Assche return -ENOSYS; 217410e1b54bSBart Van Assche 21753023a1e9SKamal Heib return device->ops.modify_device(device, device_modify_mask, 21761da177e4SLinus Torvalds device_modify); 21771da177e4SLinus Torvalds } 21781da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_device); 21791da177e4SLinus Torvalds 21801da177e4SLinus Torvalds /** 21811da177e4SLinus Torvalds * ib_modify_port - Modifies the attributes for the specified port. 21821da177e4SLinus Torvalds * @device: The device to modify. 21831da177e4SLinus Torvalds * @port_num: The number of the port to modify. 21841da177e4SLinus Torvalds * @port_modify_mask: Mask used to specify which attributes of the port 21851da177e4SLinus Torvalds * to change. 21861da177e4SLinus Torvalds * @port_modify: New attribute values for the port. 21871da177e4SLinus Torvalds * 21881da177e4SLinus Torvalds * ib_modify_port() changes a port's attributes as specified by the 21891da177e4SLinus Torvalds * @port_modify_mask and @port_modify structure. 21901da177e4SLinus Torvalds */ 21911da177e4SLinus Torvalds int ib_modify_port(struct ib_device *device, 21921da177e4SLinus Torvalds u8 port_num, int port_modify_mask, 21931da177e4SLinus Torvalds struct ib_port_modify *port_modify) 21941da177e4SLinus Torvalds { 219561e0962dSSelvin Xavier int rc; 219610e1b54bSBart Van Assche 219724dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 2198116c0074SRoland Dreier return -EINVAL; 2199116c0074SRoland Dreier 22003023a1e9SKamal Heib if (device->ops.modify_port) 22013023a1e9SKamal Heib rc = device->ops.modify_port(device, port_num, 22023023a1e9SKamal Heib port_modify_mask, 22031da177e4SLinus Torvalds port_modify); 220461e0962dSSelvin Xavier else 220561e0962dSSelvin Xavier rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS; 220661e0962dSSelvin Xavier return rc; 22071da177e4SLinus Torvalds } 22081da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_port); 22091da177e4SLinus Torvalds 22105eb620c8SYosef Etigin /** 22115eb620c8SYosef Etigin * ib_find_gid - Returns the port number and GID table index where 2212dbb12562SParav Pandit * a specified GID value occurs. Its searches only for IB link layer. 22135eb620c8SYosef Etigin * @device: The device to query. 22145eb620c8SYosef Etigin * @gid: The GID value to search for. 22155eb620c8SYosef Etigin * @port_num: The port number of the device where the GID value was found. 22165eb620c8SYosef Etigin * @index: The index into the GID table where the GID was found. This 22175eb620c8SYosef Etigin * parameter may be NULL. 22185eb620c8SYosef Etigin */ 22195eb620c8SYosef Etigin int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2220b26c4a11SParav Pandit u8 *port_num, u16 *index) 22215eb620c8SYosef Etigin { 22225eb620c8SYosef Etigin union ib_gid tmp_gid; 2223ea1075edSJason Gunthorpe unsigned int port; 2224ea1075edSJason Gunthorpe int ret, i; 22255eb620c8SYosef Etigin 2226ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 222722d24f75SParav Pandit if (!rdma_protocol_ib(device, port)) 2228b39ffa1dSMatan Barak continue; 2229b39ffa1dSMatan Barak 22308ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; 22318ceb1357SJason Gunthorpe ++i) { 22321dfce294SParav Pandit ret = rdma_query_gid(device, port, i, &tmp_gid); 22335eb620c8SYosef Etigin if (ret) 22345eb620c8SYosef Etigin return ret; 22355eb620c8SYosef Etigin if (!memcmp(&tmp_gid, gid, sizeof *gid)) { 22365eb620c8SYosef Etigin *port_num = port; 22375eb620c8SYosef Etigin if (index) 22385eb620c8SYosef Etigin *index = i; 22395eb620c8SYosef Etigin return 0; 22405eb620c8SYosef Etigin } 22415eb620c8SYosef Etigin } 22425eb620c8SYosef Etigin } 22435eb620c8SYosef Etigin 22445eb620c8SYosef Etigin return -ENOENT; 22455eb620c8SYosef Etigin } 22465eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_gid); 22475eb620c8SYosef Etigin 22485eb620c8SYosef Etigin /** 22495eb620c8SYosef Etigin * ib_find_pkey - Returns the PKey table index where a specified 22505eb620c8SYosef Etigin * PKey value occurs. 22515eb620c8SYosef Etigin * @device: The device to query. 22525eb620c8SYosef Etigin * @port_num: The port number of the device to search for the PKey. 22535eb620c8SYosef Etigin * @pkey: The PKey value to search for. 22545eb620c8SYosef Etigin * @index: The index into the PKey table where the PKey was found. 22555eb620c8SYosef Etigin */ 22565eb620c8SYosef Etigin int ib_find_pkey(struct ib_device *device, 22575eb620c8SYosef Etigin u8 port_num, u16 pkey, u16 *index) 22585eb620c8SYosef Etigin { 22595eb620c8SYosef Etigin int ret, i; 22605eb620c8SYosef Etigin u16 tmp_pkey; 2261ff7166c4SJack Morgenstein int partial_ix = -1; 22625eb620c8SYosef Etigin 22638ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; 22648ceb1357SJason Gunthorpe ++i) { 22655eb620c8SYosef Etigin ret = ib_query_pkey(device, port_num, i, &tmp_pkey); 22665eb620c8SYosef Etigin if (ret) 22675eb620c8SYosef Etigin return ret; 226836026eccSMoni Shoua if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { 2269ff7166c4SJack Morgenstein /* if there is full-member pkey take it.*/ 2270ff7166c4SJack Morgenstein if (tmp_pkey & 0x8000) { 22715eb620c8SYosef Etigin *index = i; 22725eb620c8SYosef Etigin return 0; 22735eb620c8SYosef Etigin } 2274ff7166c4SJack Morgenstein if (partial_ix < 0) 2275ff7166c4SJack Morgenstein partial_ix = i; 2276ff7166c4SJack Morgenstein } 22775eb620c8SYosef Etigin } 22785eb620c8SYosef Etigin 2279ff7166c4SJack Morgenstein /*no full-member, if exists take the limited*/ 2280ff7166c4SJack Morgenstein if (partial_ix >= 0) { 2281ff7166c4SJack Morgenstein *index = partial_ix; 2282ff7166c4SJack Morgenstein return 0; 2283ff7166c4SJack Morgenstein } 22845eb620c8SYosef Etigin return -ENOENT; 22855eb620c8SYosef Etigin } 22865eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_pkey); 22875eb620c8SYosef Etigin 22889268f72dSYotam Kenneth /** 22899268f72dSYotam Kenneth * ib_get_net_dev_by_params() - Return the appropriate net_dev 22909268f72dSYotam Kenneth * for a received CM request 22919268f72dSYotam Kenneth * @dev: An RDMA device on which the request has been received. 22929268f72dSYotam Kenneth * @port: Port number on the RDMA device. 22939268f72dSYotam Kenneth * @pkey: The Pkey the request came on. 22949268f72dSYotam Kenneth * @gid: A GID that the net_dev uses to communicate. 22959268f72dSYotam Kenneth * @addr: Contains the IP address that the request specified as its 22969268f72dSYotam Kenneth * destination. 2297921eab11SJason Gunthorpe * 22989268f72dSYotam Kenneth */ 22999268f72dSYotam Kenneth struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, 23009268f72dSYotam Kenneth u8 port, 23019268f72dSYotam Kenneth u16 pkey, 23029268f72dSYotam Kenneth const union ib_gid *gid, 23039268f72dSYotam Kenneth const struct sockaddr *addr) 23049268f72dSYotam Kenneth { 23059268f72dSYotam Kenneth struct net_device *net_dev = NULL; 23060df91bb6SJason Gunthorpe unsigned long index; 23070df91bb6SJason Gunthorpe void *client_data; 23089268f72dSYotam Kenneth 23099268f72dSYotam Kenneth if (!rdma_protocol_ib(dev, port)) 23109268f72dSYotam Kenneth return NULL; 23119268f72dSYotam Kenneth 2312921eab11SJason Gunthorpe /* 2313921eab11SJason Gunthorpe * Holding the read side guarantees that the client will not become 2314921eab11SJason Gunthorpe * unregistered while we are calling get_net_dev_by_params() 2315921eab11SJason Gunthorpe */ 2316921eab11SJason Gunthorpe down_read(&dev->client_data_rwsem); 23170df91bb6SJason Gunthorpe xan_for_each_marked (&dev->client_data, index, client_data, 23180df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED) { 23190df91bb6SJason Gunthorpe struct ib_client *client = xa_load(&clients, index); 23209268f72dSYotam Kenneth 23210df91bb6SJason Gunthorpe if (!client || !client->get_net_dev_by_params) 23229268f72dSYotam Kenneth continue; 23239268f72dSYotam Kenneth 23240df91bb6SJason Gunthorpe net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, 23250df91bb6SJason Gunthorpe addr, client_data); 23269268f72dSYotam Kenneth if (net_dev) 23279268f72dSYotam Kenneth break; 23289268f72dSYotam Kenneth } 2329921eab11SJason Gunthorpe up_read(&dev->client_data_rwsem); 23309268f72dSYotam Kenneth 23319268f72dSYotam Kenneth return net_dev; 23329268f72dSYotam Kenneth } 23339268f72dSYotam Kenneth EXPORT_SYMBOL(ib_get_net_dev_by_params); 23349268f72dSYotam Kenneth 2335521ed0d9SKamal Heib void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) 2336521ed0d9SKamal Heib { 23373023a1e9SKamal Heib struct ib_device_ops *dev_ops = &dev->ops; 2338521ed0d9SKamal Heib #define SET_DEVICE_OP(ptr, name) \ 2339521ed0d9SKamal Heib do { \ 2340521ed0d9SKamal Heib if (ops->name) \ 2341521ed0d9SKamal Heib if (!((ptr)->name)) \ 2342521ed0d9SKamal Heib (ptr)->name = ops->name; \ 2343521ed0d9SKamal Heib } while (0) 2344521ed0d9SKamal Heib 234530471d4bSLeon Romanovsky #define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) 234630471d4bSLeon Romanovsky 23473023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, add_gid); 23482f1927b0SMoni Shoua SET_DEVICE_OP(dev_ops, advise_mr); 23493023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_dm); 23503023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_fmr); 23513023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_hw_stats); 23523023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mr); 23533023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mw); 23543023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_pd); 23553023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_rdma_netdev); 23563023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_ucontext); 23573023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_xrcd); 23583023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, attach_mcast); 23593023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, check_mr_status); 23603023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_ah); 23613023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_counters); 23623023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_cq); 23633023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow); 23643023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow_action_esp); 23653023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_qp); 23663023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_rwq_ind_table); 23673023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_srq); 23683023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_wq); 23693023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_dm); 2370d0899892SJason Gunthorpe SET_DEVICE_OP(dev_ops, dealloc_driver); 23713023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_fmr); 23723023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_mw); 23733023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_pd); 23743023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_ucontext); 23753023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_xrcd); 23763023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, del_gid); 23773023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dereg_mr); 23783023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_ah); 23793023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_counters); 23803023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_cq); 23813023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow); 23823023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow_action); 23833023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_qp); 23843023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_rwq_ind_table); 23853023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_srq); 23863023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_wq); 23873023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, detach_mcast); 23883023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, disassociate_ucontext); 23893023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_rq); 23903023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_sq); 2391ca22354bSJason Gunthorpe SET_DEVICE_OP(dev_ops, enable_driver); 239202da3750SLeon Romanovsky SET_DEVICE_OP(dev_ops, fill_res_entry); 23933023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dev_fw_str); 23943023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dma_mr); 23953023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_hw_stats); 23963023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_link_layer); 23973023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_netdev); 23983023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_port_immutable); 23993023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vector_affinity); 24003023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_config); 24013023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_stats); 2402ea4baf7fSParav Pandit SET_DEVICE_OP(dev_ops, init_port); 2403dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_accept); 2404dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_add_ref); 2405dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_connect); 2406dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_create_listen); 2407dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_destroy_listen); 2408dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_get_qp); 2409dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_reject); 2410dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_rem_ref); 24113023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, map_mr_sg); 24123023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, map_phys_fmr); 24133023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, mmap); 24143023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_ah); 24153023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_cq); 24163023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_device); 24173023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_flow_action_esp); 24183023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_port); 24193023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_qp); 24203023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_srq); 24213023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_wq); 24223023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, peek_cq); 24233023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, poll_cq); 24243023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_recv); 24253023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_send); 24263023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_srq_recv); 24273023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, process_mad); 24283023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_ah); 24293023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_device); 24303023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_gid); 24313023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_pkey); 24323023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_port); 24333023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_qp); 24343023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_srq); 24353023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rdma_netdev_get_params); 24363023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, read_counters); 24373023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_dm_mr); 24383023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_user_mr); 24393023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_ncomp_notif); 24403023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_notify_cq); 24413023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rereg_user_mr); 24423023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, resize_cq); 24433023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_guid); 24443023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_link_state); 24453023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, unmap_fmr); 244621a428a0SLeon Romanovsky 2447d3456914SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_ah); 244821a428a0SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_pd); 244968e326deSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_srq); 2450a2a074efSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_ucontext); 2451521ed0d9SKamal Heib } 2452521ed0d9SKamal Heib EXPORT_SYMBOL(ib_set_device_ops); 2453521ed0d9SKamal Heib 2454d0e312feSLeon Romanovsky static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = { 2455735c631aSMark Bloch [RDMA_NL_LS_OP_RESOLVE] = { 2456647c75acSLeon Romanovsky .doit = ib_nl_handle_resolve_resp, 2457e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2458e3a2b93dSLeon Romanovsky }, 2459735c631aSMark Bloch [RDMA_NL_LS_OP_SET_TIMEOUT] = { 2460647c75acSLeon Romanovsky .doit = ib_nl_handle_set_timeout, 2461e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2462e3a2b93dSLeon Romanovsky }, 2463ae43f828SMark Bloch [RDMA_NL_LS_OP_IP_RESOLVE] = { 2464647c75acSLeon Romanovsky .doit = ib_nl_handle_ip_res_resp, 2465e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2466e3a2b93dSLeon Romanovsky }, 2467735c631aSMark Bloch }; 2468735c631aSMark Bloch 24691da177e4SLinus Torvalds static int __init ib_core_init(void) 24701da177e4SLinus Torvalds { 24711da177e4SLinus Torvalds int ret; 24721da177e4SLinus Torvalds 2473f0626710STejun Heo ib_wq = alloc_workqueue("infiniband", 0, 0); 2474f0626710STejun Heo if (!ib_wq) 2475f0626710STejun Heo return -ENOMEM; 2476f0626710STejun Heo 247714d3a3b2SChristoph Hellwig ib_comp_wq = alloc_workqueue("ib-comp-wq", 2478b7363e67SSagi Grimberg WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0); 247914d3a3b2SChristoph Hellwig if (!ib_comp_wq) { 248014d3a3b2SChristoph Hellwig ret = -ENOMEM; 248114d3a3b2SChristoph Hellwig goto err; 248214d3a3b2SChristoph Hellwig } 248314d3a3b2SChristoph Hellwig 2484f794809aSJack Morgenstein ib_comp_unbound_wq = 2485f794809aSJack Morgenstein alloc_workqueue("ib-comp-unb-wq", 2486f794809aSJack Morgenstein WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM | 2487f794809aSJack Morgenstein WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE); 2488f794809aSJack Morgenstein if (!ib_comp_unbound_wq) { 2489f794809aSJack Morgenstein ret = -ENOMEM; 2490f794809aSJack Morgenstein goto err_comp; 2491f794809aSJack Morgenstein } 2492f794809aSJack Morgenstein 249355aeed06SJason Gunthorpe ret = class_register(&ib_class); 2494fd75c789SNir Muchtar if (ret) { 2495aba25a3eSParav Pandit pr_warn("Couldn't create InfiniBand device class\n"); 2496f794809aSJack Morgenstein goto err_comp_unbound; 2497fd75c789SNir Muchtar } 24981da177e4SLinus Torvalds 2499c9901724SLeon Romanovsky ret = rdma_nl_init(); 25001da177e4SLinus Torvalds if (ret) { 2501c9901724SLeon Romanovsky pr_warn("Couldn't init IB netlink interface: err %d\n", ret); 2502fd75c789SNir Muchtar goto err_sysfs; 25031da177e4SLinus Torvalds } 25041da177e4SLinus Torvalds 2505e3f20f02SLeon Romanovsky ret = addr_init(); 2506e3f20f02SLeon Romanovsky if (ret) { 2507e3f20f02SLeon Romanovsky pr_warn("Could't init IB address resolution\n"); 2508e3f20f02SLeon Romanovsky goto err_ibnl; 2509e3f20f02SLeon Romanovsky } 2510e3f20f02SLeon Romanovsky 25114c2cb422SMark Bloch ret = ib_mad_init(); 25124c2cb422SMark Bloch if (ret) { 25134c2cb422SMark Bloch pr_warn("Couldn't init IB MAD\n"); 25144c2cb422SMark Bloch goto err_addr; 25154c2cb422SMark Bloch } 25164c2cb422SMark Bloch 2517c2e49c92SMark Bloch ret = ib_sa_init(); 2518c2e49c92SMark Bloch if (ret) { 2519c2e49c92SMark Bloch pr_warn("Couldn't init SA\n"); 2520c2e49c92SMark Bloch goto err_mad; 2521c2e49c92SMark Bloch } 2522c2e49c92SMark Bloch 25238f408ab6SDaniel Jurgens ret = register_lsm_notifier(&ibdev_lsm_nb); 25248f408ab6SDaniel Jurgens if (ret) { 25258f408ab6SDaniel Jurgens pr_warn("Couldn't register LSM notifier. ret %d\n", ret); 2526c9901724SLeon Romanovsky goto err_sa; 25278f408ab6SDaniel Jurgens } 25288f408ab6SDaniel Jurgens 25294e0f7b90SParav Pandit ret = register_pernet_device(&rdma_dev_net_ops); 25304e0f7b90SParav Pandit if (ret) { 25314e0f7b90SParav Pandit pr_warn("Couldn't init compat dev. ret %d\n", ret); 25324e0f7b90SParav Pandit goto err_compat; 25334e0f7b90SParav Pandit } 25344e0f7b90SParav Pandit 25356c80b41aSLeon Romanovsky nldev_init(); 2536c9901724SLeon Romanovsky rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table); 25375ef8c0c1SJason Gunthorpe roce_gid_mgmt_init(); 2538b2cbae2cSRoland Dreier 2539fd75c789SNir Muchtar return 0; 2540fd75c789SNir Muchtar 25414e0f7b90SParav Pandit err_compat: 25424e0f7b90SParav Pandit unregister_lsm_notifier(&ibdev_lsm_nb); 2543735c631aSMark Bloch err_sa: 2544735c631aSMark Bloch ib_sa_cleanup(); 2545c2e49c92SMark Bloch err_mad: 2546c2e49c92SMark Bloch ib_mad_cleanup(); 25474c2cb422SMark Bloch err_addr: 25484c2cb422SMark Bloch addr_cleanup(); 2549e3f20f02SLeon Romanovsky err_ibnl: 2550c9901724SLeon Romanovsky rdma_nl_exit(); 2551fd75c789SNir Muchtar err_sysfs: 255255aeed06SJason Gunthorpe class_unregister(&ib_class); 2553f794809aSJack Morgenstein err_comp_unbound: 2554f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 255514d3a3b2SChristoph Hellwig err_comp: 255614d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2557fd75c789SNir Muchtar err: 2558fd75c789SNir Muchtar destroy_workqueue(ib_wq); 25591da177e4SLinus Torvalds return ret; 25601da177e4SLinus Torvalds } 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds static void __exit ib_core_cleanup(void) 25631da177e4SLinus Torvalds { 25645ef8c0c1SJason Gunthorpe roce_gid_mgmt_cleanup(); 25656c80b41aSLeon Romanovsky nldev_exit(); 2566c9901724SLeon Romanovsky rdma_nl_unregister(RDMA_NL_LS); 25674e0f7b90SParav Pandit unregister_pernet_device(&rdma_dev_net_ops); 2568c9901724SLeon Romanovsky unregister_lsm_notifier(&ibdev_lsm_nb); 2569c2e49c92SMark Bloch ib_sa_cleanup(); 25704c2cb422SMark Bloch ib_mad_cleanup(); 2571e3f20f02SLeon Romanovsky addr_cleanup(); 2572c9901724SLeon Romanovsky rdma_nl_exit(); 257355aeed06SJason Gunthorpe class_unregister(&ib_class); 2574f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 257514d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2576f7c6a7b5SRoland Dreier /* Make sure that any pending umem accounting work is done. */ 2577f0626710STejun Heo destroy_workqueue(ib_wq); 2578d0899892SJason Gunthorpe flush_workqueue(system_unbound_wq); 2579e59178d8SJason Gunthorpe WARN_ON(!xa_empty(&clients)); 25800df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&devices)); 25811da177e4SLinus Torvalds } 25821da177e4SLinus Torvalds 2583e3bf14bdSJason Gunthorpe MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); 2584e3bf14bdSJason Gunthorpe 258562dfa795SParav Pandit /* ib core relies on netdev stack to first register net_ns_type_operations 258662dfa795SParav Pandit * ns kobject type before ib_core initialization. 258762dfa795SParav Pandit */ 258862dfa795SParav Pandit fs_initcall(ib_core_init); 25891da177e4SLinus Torvalds module_exit(ib_core_cleanup); 2590