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 273deee3c7eSKamal Heib static void 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 310f8978bd9SLeon Romanovsky /* 31101b67117SParav Pandit * Caller must perform ib_device_put() to return the device reference count 31201b67117SParav Pandit * when ib_device_get_by_index() returns valid device pointer. 313f8978bd9SLeon Romanovsky */ 31437eeab55SParav Pandit struct ib_device *ib_device_get_by_index(const struct net *net, u32 index) 315f8978bd9SLeon Romanovsky { 316f8978bd9SLeon Romanovsky struct ib_device *device; 317f8978bd9SLeon Romanovsky 318921eab11SJason Gunthorpe down_read(&devices_rwsem); 3190df91bb6SJason Gunthorpe device = xa_load(&devices, index); 32001b67117SParav Pandit if (device) { 32137eeab55SParav Pandit if (!rdma_dev_access_netns(device, net)) { 32237eeab55SParav Pandit device = NULL; 32337eeab55SParav Pandit goto out; 32437eeab55SParav Pandit } 32537eeab55SParav Pandit 326d79af724SJason Gunthorpe if (!ib_device_try_get(device)) 32701b67117SParav Pandit device = NULL; 32801b67117SParav Pandit } 32937eeab55SParav Pandit out: 330921eab11SJason Gunthorpe up_read(&devices_rwsem); 331f8978bd9SLeon Romanovsky return device; 332f8978bd9SLeon Romanovsky } 333f8978bd9SLeon Romanovsky 334d79af724SJason Gunthorpe /** 335d79af724SJason Gunthorpe * ib_device_put - Release IB device reference 336d79af724SJason Gunthorpe * @device: device whose reference to be released 337d79af724SJason Gunthorpe * 338d79af724SJason Gunthorpe * ib_device_put() releases reference to the IB device to allow it to be 339d79af724SJason Gunthorpe * unregistered and eventually free. 340d79af724SJason Gunthorpe */ 34101b67117SParav Pandit void ib_device_put(struct ib_device *device) 34201b67117SParav Pandit { 34301b67117SParav Pandit if (refcount_dec_and_test(&device->refcount)) 34401b67117SParav Pandit complete(&device->unreg_completion); 34501b67117SParav Pandit } 346d79af724SJason Gunthorpe EXPORT_SYMBOL(ib_device_put); 34701b67117SParav Pandit 3481da177e4SLinus Torvalds static struct ib_device *__ib_device_get_by_name(const char *name) 3491da177e4SLinus Torvalds { 3501da177e4SLinus Torvalds struct ib_device *device; 3510df91bb6SJason Gunthorpe unsigned long index; 3521da177e4SLinus Torvalds 3530df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) 354896de009SJason Gunthorpe if (!strcmp(name, dev_name(&device->dev))) 3551da177e4SLinus Torvalds return device; 3561da177e4SLinus Torvalds 3571da177e4SLinus Torvalds return NULL; 3581da177e4SLinus Torvalds } 3591da177e4SLinus Torvalds 3606cc2c8e5SJason Gunthorpe /** 3616cc2c8e5SJason Gunthorpe * ib_device_get_by_name - Find an IB device by name 3626cc2c8e5SJason Gunthorpe * @name: The name to look for 3636cc2c8e5SJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 3646cc2c8e5SJason Gunthorpe * 3656cc2c8e5SJason Gunthorpe * Find and hold an ib_device by its name. The caller must call 3666cc2c8e5SJason Gunthorpe * ib_device_put() on the returned pointer. 3676cc2c8e5SJason Gunthorpe */ 3686cc2c8e5SJason Gunthorpe struct ib_device *ib_device_get_by_name(const char *name, 3696cc2c8e5SJason Gunthorpe enum rdma_driver_id driver_id) 3706cc2c8e5SJason Gunthorpe { 3716cc2c8e5SJason Gunthorpe struct ib_device *device; 3726cc2c8e5SJason Gunthorpe 3736cc2c8e5SJason Gunthorpe down_read(&devices_rwsem); 3746cc2c8e5SJason Gunthorpe device = __ib_device_get_by_name(name); 3756cc2c8e5SJason Gunthorpe if (device && driver_id != RDMA_DRIVER_UNKNOWN && 376b9560a41SJason Gunthorpe device->ops.driver_id != driver_id) 3776cc2c8e5SJason Gunthorpe device = NULL; 3786cc2c8e5SJason Gunthorpe 3796cc2c8e5SJason Gunthorpe if (device) { 3806cc2c8e5SJason Gunthorpe if (!ib_device_try_get(device)) 3816cc2c8e5SJason Gunthorpe device = NULL; 3826cc2c8e5SJason Gunthorpe } 3836cc2c8e5SJason Gunthorpe up_read(&devices_rwsem); 3846cc2c8e5SJason Gunthorpe return device; 3856cc2c8e5SJason Gunthorpe } 3866cc2c8e5SJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_name); 3876cc2c8e5SJason Gunthorpe 3884e0f7b90SParav Pandit static int rename_compat_devs(struct ib_device *device) 3894e0f7b90SParav Pandit { 3904e0f7b90SParav Pandit struct ib_core_device *cdev; 3914e0f7b90SParav Pandit unsigned long index; 3924e0f7b90SParav Pandit int ret = 0; 3934e0f7b90SParav Pandit 3944e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 3954e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) { 3964e0f7b90SParav Pandit ret = device_rename(&cdev->dev, dev_name(&device->dev)); 3974e0f7b90SParav Pandit if (ret) { 3984e0f7b90SParav Pandit dev_warn(&cdev->dev, 3994e0f7b90SParav Pandit "Fail to rename compatdev to new name %s\n", 4004e0f7b90SParav Pandit dev_name(&device->dev)); 4014e0f7b90SParav Pandit break; 4024e0f7b90SParav Pandit } 4034e0f7b90SParav Pandit } 4044e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 4054e0f7b90SParav Pandit return ret; 4064e0f7b90SParav Pandit } 4074e0f7b90SParav Pandit 408d21943ddSLeon Romanovsky int ib_device_rename(struct ib_device *ibdev, const char *name) 409d21943ddSLeon Romanovsky { 410e3593b56SJason Gunthorpe int ret; 411d21943ddSLeon Romanovsky 412921eab11SJason Gunthorpe down_write(&devices_rwsem); 413e3593b56SJason Gunthorpe if (!strcmp(name, dev_name(&ibdev->dev))) { 414e3593b56SJason Gunthorpe ret = 0; 415e3593b56SJason Gunthorpe goto out; 416e3593b56SJason Gunthorpe } 417e3593b56SJason Gunthorpe 418344684e6SJason Gunthorpe if (__ib_device_get_by_name(name)) { 419d21943ddSLeon Romanovsky ret = -EEXIST; 420d21943ddSLeon Romanovsky goto out; 421d21943ddSLeon Romanovsky } 422d21943ddSLeon Romanovsky 423d21943ddSLeon Romanovsky ret = device_rename(&ibdev->dev, name); 424d21943ddSLeon Romanovsky if (ret) 425d21943ddSLeon Romanovsky goto out; 426d21943ddSLeon Romanovsky strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); 4274e0f7b90SParav Pandit ret = rename_compat_devs(ibdev); 428d21943ddSLeon Romanovsky out: 429921eab11SJason Gunthorpe up_write(&devices_rwsem); 430d21943ddSLeon Romanovsky return ret; 431d21943ddSLeon Romanovsky } 432d21943ddSLeon Romanovsky 433e349f858SJason Gunthorpe static int alloc_name(struct ib_device *ibdev, const char *name) 4341da177e4SLinus Torvalds { 4351da177e4SLinus Torvalds struct ib_device *device; 4360df91bb6SJason Gunthorpe unsigned long index; 4373b88afd3SJason Gunthorpe struct ida inuse; 4383b88afd3SJason Gunthorpe int rc; 4391da177e4SLinus Torvalds int i; 4401da177e4SLinus Torvalds 441921eab11SJason Gunthorpe lockdep_assert_held_exclusive(&devices_rwsem); 4423b88afd3SJason Gunthorpe ida_init(&inuse); 4430df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) { 444e349f858SJason Gunthorpe char buf[IB_DEVICE_NAME_MAX]; 445e349f858SJason Gunthorpe 446896de009SJason Gunthorpe if (sscanf(dev_name(&device->dev), name, &i) != 1) 4471da177e4SLinus Torvalds continue; 4483b88afd3SJason Gunthorpe if (i < 0 || i >= INT_MAX) 4491da177e4SLinus Torvalds continue; 4501da177e4SLinus Torvalds snprintf(buf, sizeof buf, name, i); 4513b88afd3SJason Gunthorpe if (strcmp(buf, dev_name(&device->dev)) != 0) 4523b88afd3SJason Gunthorpe continue; 4533b88afd3SJason Gunthorpe 4543b88afd3SJason Gunthorpe rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); 4553b88afd3SJason Gunthorpe if (rc < 0) 4563b88afd3SJason Gunthorpe goto out; 4571da177e4SLinus Torvalds } 4581da177e4SLinus Torvalds 4593b88afd3SJason Gunthorpe rc = ida_alloc(&inuse, GFP_KERNEL); 4603b88afd3SJason Gunthorpe if (rc < 0) 4613b88afd3SJason Gunthorpe goto out; 4621da177e4SLinus Torvalds 4633b88afd3SJason Gunthorpe rc = dev_set_name(&ibdev->dev, name, rc); 4643b88afd3SJason Gunthorpe out: 4653b88afd3SJason Gunthorpe ida_destroy(&inuse); 4663b88afd3SJason Gunthorpe return rc; 4671da177e4SLinus Torvalds } 4681da177e4SLinus Torvalds 46955aeed06SJason Gunthorpe static void ib_device_release(struct device *device) 47055aeed06SJason Gunthorpe { 47155aeed06SJason Gunthorpe struct ib_device *dev = container_of(device, struct ib_device, dev); 47255aeed06SJason Gunthorpe 473c2261dd7SJason Gunthorpe free_netdevs(dev); 474652432f3SJason Gunthorpe WARN_ON(refcount_read(&dev->refcount)); 47503db3a2dSMatan Barak ib_cache_release_one(dev); 476b34b269aSJason Gunthorpe ib_security_release_port_pkey_list(dev); 4774e0f7b90SParav Pandit xa_destroy(&dev->compat_devs); 4780df91bb6SJason Gunthorpe xa_destroy(&dev->client_data); 479324e227eSJason Gunthorpe if (dev->port_data) 480324e227eSJason Gunthorpe kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, 481324e227eSJason Gunthorpe pdata[0]), 482324e227eSJason Gunthorpe rcu_head); 483324e227eSJason Gunthorpe kfree_rcu(dev, rcu_head); 48455aeed06SJason Gunthorpe } 48555aeed06SJason Gunthorpe 48655aeed06SJason Gunthorpe static int ib_device_uevent(struct device *device, 48755aeed06SJason Gunthorpe struct kobj_uevent_env *env) 48855aeed06SJason Gunthorpe { 489896de009SJason Gunthorpe if (add_uevent_var(env, "NAME=%s", dev_name(device))) 49055aeed06SJason Gunthorpe return -ENOMEM; 49155aeed06SJason Gunthorpe 49255aeed06SJason Gunthorpe /* 49355aeed06SJason Gunthorpe * It would be nice to pass the node GUID with the event... 49455aeed06SJason Gunthorpe */ 49555aeed06SJason Gunthorpe 49655aeed06SJason Gunthorpe return 0; 49755aeed06SJason Gunthorpe } 49855aeed06SJason Gunthorpe 49962dfa795SParav Pandit static const void *net_namespace(struct device *d) 50062dfa795SParav Pandit { 5014e0f7b90SParav Pandit struct ib_core_device *coredev = 5024e0f7b90SParav Pandit container_of(d, struct ib_core_device, dev); 5034e0f7b90SParav Pandit 5044e0f7b90SParav Pandit return read_pnet(&coredev->rdma_net); 50562dfa795SParav Pandit } 50662dfa795SParav Pandit 50755aeed06SJason Gunthorpe static struct class ib_class = { 50855aeed06SJason Gunthorpe .name = "infiniband", 50955aeed06SJason Gunthorpe .dev_release = ib_device_release, 51055aeed06SJason Gunthorpe .dev_uevent = ib_device_uevent, 51162dfa795SParav Pandit .ns_type = &net_ns_type_operations, 51262dfa795SParav Pandit .namespace = net_namespace, 51355aeed06SJason Gunthorpe }; 51455aeed06SJason Gunthorpe 515cebe556bSParav Pandit static void rdma_init_coredev(struct ib_core_device *coredev, 5164e0f7b90SParav Pandit struct ib_device *dev, struct net *net) 517cebe556bSParav Pandit { 518cebe556bSParav Pandit /* This BUILD_BUG_ON is intended to catch layout change 519cebe556bSParav Pandit * of union of ib_core_device and device. 520cebe556bSParav Pandit * dev must be the first element as ib_core and providers 521cebe556bSParav Pandit * driver uses it. Adding anything in ib_core_device before 522cebe556bSParav Pandit * device will break this assumption. 523cebe556bSParav Pandit */ 524cebe556bSParav Pandit BUILD_BUG_ON(offsetof(struct ib_device, coredev.dev) != 525cebe556bSParav Pandit offsetof(struct ib_device, dev)); 526cebe556bSParav Pandit 527cebe556bSParav Pandit coredev->dev.class = &ib_class; 528cebe556bSParav Pandit coredev->dev.groups = dev->groups; 529cebe556bSParav Pandit device_initialize(&coredev->dev); 530cebe556bSParav Pandit coredev->owner = dev; 531cebe556bSParav Pandit INIT_LIST_HEAD(&coredev->port_list); 5324e0f7b90SParav Pandit write_pnet(&coredev->rdma_net, net); 533cebe556bSParav Pandit } 534cebe556bSParav Pandit 5351da177e4SLinus Torvalds /** 536459cc69fSLeon Romanovsky * _ib_alloc_device - allocate an IB device struct 5371da177e4SLinus Torvalds * @size:size of structure to allocate 5381da177e4SLinus Torvalds * 5391da177e4SLinus Torvalds * Low-level drivers should use ib_alloc_device() to allocate &struct 5401da177e4SLinus Torvalds * ib_device. @size is the size of the structure to be allocated, 5411da177e4SLinus Torvalds * including any private data used by the low-level driver. 5421da177e4SLinus Torvalds * ib_dealloc_device() must be used to free structures allocated with 5431da177e4SLinus Torvalds * ib_alloc_device(). 5441da177e4SLinus Torvalds */ 545459cc69fSLeon Romanovsky struct ib_device *_ib_alloc_device(size_t size) 5461da177e4SLinus Torvalds { 54755aeed06SJason Gunthorpe struct ib_device *device; 5481da177e4SLinus Torvalds 54955aeed06SJason Gunthorpe if (WARN_ON(size < sizeof(struct ib_device))) 55055aeed06SJason Gunthorpe return NULL; 55155aeed06SJason Gunthorpe 55255aeed06SJason Gunthorpe device = kzalloc(size, GFP_KERNEL); 55355aeed06SJason Gunthorpe if (!device) 55455aeed06SJason Gunthorpe return NULL; 55555aeed06SJason Gunthorpe 55641eda65cSLeon Romanovsky if (rdma_restrack_init(device)) { 55741eda65cSLeon Romanovsky kfree(device); 55841eda65cSLeon Romanovsky return NULL; 55941eda65cSLeon Romanovsky } 56002d8883fSLeon Romanovsky 5615f8f5499SParav Pandit device->groups[0] = &ib_dev_attr_group; 5624e0f7b90SParav Pandit rdma_init_coredev(&device->coredev, device, &init_net); 56355aeed06SJason Gunthorpe 56455aeed06SJason Gunthorpe INIT_LIST_HEAD(&device->event_handler_list); 56555aeed06SJason Gunthorpe spin_lock_init(&device->event_handler_lock); 566d0899892SJason Gunthorpe mutex_init(&device->unregistration_lock); 5670df91bb6SJason Gunthorpe /* 5680df91bb6SJason Gunthorpe * client_data needs to be alloc because we don't want our mark to be 5690df91bb6SJason Gunthorpe * destroyed if the user stores NULL in the client data. 5700df91bb6SJason Gunthorpe */ 5710df91bb6SJason Gunthorpe xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); 572921eab11SJason Gunthorpe init_rwsem(&device->client_data_rwsem); 5734e0f7b90SParav Pandit xa_init_flags(&device->compat_devs, XA_FLAGS_ALLOC); 5744e0f7b90SParav Pandit mutex_init(&device->compat_devs_mutex); 57501b67117SParav Pandit init_completion(&device->unreg_completion); 576d0899892SJason Gunthorpe INIT_WORK(&device->unregistration_work, ib_unregister_work); 57755aeed06SJason Gunthorpe 57855aeed06SJason Gunthorpe return device; 5791da177e4SLinus Torvalds } 580459cc69fSLeon Romanovsky EXPORT_SYMBOL(_ib_alloc_device); 5811da177e4SLinus Torvalds 5821da177e4SLinus Torvalds /** 5831da177e4SLinus Torvalds * ib_dealloc_device - free an IB device struct 5841da177e4SLinus Torvalds * @device:structure to free 5851da177e4SLinus Torvalds * 5861da177e4SLinus Torvalds * Free a structure allocated with ib_alloc_device(). 5871da177e4SLinus Torvalds */ 5881da177e4SLinus Torvalds void ib_dealloc_device(struct ib_device *device) 5891da177e4SLinus Torvalds { 590d0899892SJason Gunthorpe if (device->ops.dealloc_driver) 591d0899892SJason Gunthorpe device->ops.dealloc_driver(device); 592d0899892SJason Gunthorpe 593d0899892SJason Gunthorpe /* 594d0899892SJason Gunthorpe * ib_unregister_driver() requires all devices to remain in the xarray 595d0899892SJason Gunthorpe * while their ops are callable. The last op we call is dealloc_driver 596d0899892SJason Gunthorpe * above. This is needed to create a fence on op callbacks prior to 597d0899892SJason Gunthorpe * allowing the driver module to unload. 598d0899892SJason Gunthorpe */ 599d0899892SJason Gunthorpe down_write(&devices_rwsem); 600d0899892SJason Gunthorpe if (xa_load(&devices, device->index) == device) 601d0899892SJason Gunthorpe xa_erase(&devices, device->index); 602d0899892SJason Gunthorpe up_write(&devices_rwsem); 603d0899892SJason Gunthorpe 604c2261dd7SJason Gunthorpe /* Expedite releasing netdev references */ 605c2261dd7SJason Gunthorpe free_netdevs(device); 606c2261dd7SJason Gunthorpe 6074e0f7b90SParav Pandit WARN_ON(!xa_empty(&device->compat_devs)); 6080df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&device->client_data)); 609652432f3SJason Gunthorpe WARN_ON(refcount_read(&device->refcount)); 6100ad699c0SLeon Romanovsky rdma_restrack_clean(device); 611e155755eSParav Pandit /* Balances with device_initialize */ 612924b8900SLeon Romanovsky put_device(&device->dev); 6131da177e4SLinus Torvalds } 6141da177e4SLinus Torvalds EXPORT_SYMBOL(ib_dealloc_device); 6151da177e4SLinus Torvalds 616921eab11SJason Gunthorpe /* 617921eab11SJason Gunthorpe * add_client_context() and remove_client_context() must be safe against 618921eab11SJason Gunthorpe * parallel calls on the same device - registration/unregistration of both the 619921eab11SJason Gunthorpe * device and client can be occurring in parallel. 620921eab11SJason Gunthorpe * 621921eab11SJason Gunthorpe * The routines need to be a fence, any caller must not return until the add 622921eab11SJason Gunthorpe * or remove is fully completed. 623921eab11SJason Gunthorpe */ 624921eab11SJason Gunthorpe static int add_client_context(struct ib_device *device, 625921eab11SJason Gunthorpe struct ib_client *client) 6261da177e4SLinus Torvalds { 627921eab11SJason Gunthorpe int ret = 0; 6281da177e4SLinus Torvalds 6296780c4faSGal Pressman if (!device->kverbs_provider && !client->no_kverbs_req) 630921eab11SJason Gunthorpe return 0; 6316780c4faSGal Pressman 632921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 633921eab11SJason Gunthorpe /* 634921eab11SJason Gunthorpe * Another caller to add_client_context got here first and has already 635921eab11SJason Gunthorpe * completely initialized context. 636921eab11SJason Gunthorpe */ 637921eab11SJason Gunthorpe if (xa_get_mark(&device->client_data, client->client_id, 638921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) 639921eab11SJason Gunthorpe goto out; 640921eab11SJason Gunthorpe 641921eab11SJason Gunthorpe ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, 642921eab11SJason Gunthorpe GFP_KERNEL)); 643921eab11SJason Gunthorpe if (ret) 644921eab11SJason Gunthorpe goto out; 645921eab11SJason Gunthorpe downgrade_write(&device->client_data_rwsem); 646921eab11SJason Gunthorpe if (client->add) 647921eab11SJason Gunthorpe client->add(device); 648921eab11SJason Gunthorpe 649921eab11SJason Gunthorpe /* Readers shall not see a client until add has been completed */ 6500df91bb6SJason Gunthorpe xa_set_mark(&device->client_data, client->client_id, 6510df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED); 652921eab11SJason Gunthorpe up_read(&device->client_data_rwsem); 653921eab11SJason Gunthorpe return 0; 6541da177e4SLinus Torvalds 655921eab11SJason Gunthorpe out: 656921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 657921eab11SJason Gunthorpe return ret; 658921eab11SJason Gunthorpe } 659921eab11SJason Gunthorpe 660921eab11SJason Gunthorpe static void remove_client_context(struct ib_device *device, 661921eab11SJason Gunthorpe unsigned int client_id) 662921eab11SJason Gunthorpe { 663921eab11SJason Gunthorpe struct ib_client *client; 664921eab11SJason Gunthorpe void *client_data; 665921eab11SJason Gunthorpe 666921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 667921eab11SJason Gunthorpe if (!xa_get_mark(&device->client_data, client_id, 668921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) { 669921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 670921eab11SJason Gunthorpe return; 671921eab11SJason Gunthorpe } 672921eab11SJason Gunthorpe client_data = xa_load(&device->client_data, client_id); 673921eab11SJason Gunthorpe xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); 674921eab11SJason Gunthorpe client = xa_load(&clients, client_id); 675921eab11SJason Gunthorpe downgrade_write(&device->client_data_rwsem); 676921eab11SJason Gunthorpe 677921eab11SJason Gunthorpe /* 678921eab11SJason Gunthorpe * Notice we cannot be holding any exclusive locks when calling the 679921eab11SJason Gunthorpe * remove callback as the remove callback can recurse back into any 680921eab11SJason Gunthorpe * public functions in this module and thus try for any locks those 681921eab11SJason Gunthorpe * functions take. 682921eab11SJason Gunthorpe * 683921eab11SJason Gunthorpe * For this reason clients and drivers should not call the 684921eab11SJason Gunthorpe * unregistration functions will holdling any locks. 685921eab11SJason Gunthorpe * 686921eab11SJason Gunthorpe * It tempting to drop the client_data_rwsem too, but this is required 687921eab11SJason Gunthorpe * to ensure that unregister_client does not return until all clients 688921eab11SJason Gunthorpe * are completely unregistered, which is required to avoid module 689921eab11SJason Gunthorpe * unloading races. 690921eab11SJason Gunthorpe */ 691921eab11SJason Gunthorpe if (client->remove) 692921eab11SJason Gunthorpe client->remove(device, client_data); 693921eab11SJason Gunthorpe 694921eab11SJason Gunthorpe xa_erase(&device->client_data, client_id); 695921eab11SJason Gunthorpe up_read(&device->client_data_rwsem); 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds 698c2261dd7SJason Gunthorpe static int alloc_port_data(struct ib_device *device) 6995eb620c8SYosef Etigin { 700324e227eSJason Gunthorpe struct ib_port_data_rcu *pdata_rcu; 701ea1075edSJason Gunthorpe unsigned int port; 702c2261dd7SJason Gunthorpe 703c2261dd7SJason Gunthorpe if (device->port_data) 704c2261dd7SJason Gunthorpe return 0; 705c2261dd7SJason Gunthorpe 706c2261dd7SJason Gunthorpe /* This can only be called once the physical port range is defined */ 707c2261dd7SJason Gunthorpe if (WARN_ON(!device->phys_port_cnt)) 708c2261dd7SJason Gunthorpe return -EINVAL; 7095eb620c8SYosef Etigin 7108ceb1357SJason Gunthorpe /* 7118ceb1357SJason Gunthorpe * device->port_data is indexed directly by the port number to make 7127738613eSIra Weiny * access to this data as efficient as possible. 7137738613eSIra Weiny * 7148ceb1357SJason Gunthorpe * Therefore port_data is declared as a 1 based array with potential 7158ceb1357SJason Gunthorpe * empty slots at the beginning. 7167738613eSIra Weiny */ 717324e227eSJason Gunthorpe pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, 718324e227eSJason Gunthorpe rdma_end_port(device) + 1), 719324e227eSJason Gunthorpe GFP_KERNEL); 720324e227eSJason Gunthorpe if (!pdata_rcu) 72155aeed06SJason Gunthorpe return -ENOMEM; 722324e227eSJason Gunthorpe /* 723324e227eSJason Gunthorpe * The rcu_head is put in front of the port data array and the stored 724324e227eSJason Gunthorpe * pointer is adjusted since we never need to see that member until 725324e227eSJason Gunthorpe * kfree_rcu. 726324e227eSJason Gunthorpe */ 727324e227eSJason Gunthorpe device->port_data = pdata_rcu->pdata; 7285eb620c8SYosef Etigin 729ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 7308ceb1357SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 7318ceb1357SJason Gunthorpe 732324e227eSJason Gunthorpe pdata->ib_dev = device; 7338ceb1357SJason Gunthorpe spin_lock_init(&pdata->pkey_list_lock); 7348ceb1357SJason Gunthorpe INIT_LIST_HEAD(&pdata->pkey_list); 735c2261dd7SJason Gunthorpe spin_lock_init(&pdata->netdev_lock); 736324e227eSJason Gunthorpe INIT_HLIST_NODE(&pdata->ndev_hash_link); 737c2261dd7SJason Gunthorpe } 738c2261dd7SJason Gunthorpe return 0; 739c2261dd7SJason Gunthorpe } 740c2261dd7SJason Gunthorpe 741c2261dd7SJason Gunthorpe static int verify_immutable(const struct ib_device *dev, u8 port) 742c2261dd7SJason Gunthorpe { 743c2261dd7SJason Gunthorpe return WARN_ON(!rdma_cap_ib_mad(dev, port) && 744c2261dd7SJason Gunthorpe rdma_max_mad_size(dev, port) != 0); 745c2261dd7SJason Gunthorpe } 746c2261dd7SJason Gunthorpe 747c2261dd7SJason Gunthorpe static int setup_port_data(struct ib_device *device) 748c2261dd7SJason Gunthorpe { 749c2261dd7SJason Gunthorpe unsigned int port; 750c2261dd7SJason Gunthorpe int ret; 751c2261dd7SJason Gunthorpe 752c2261dd7SJason Gunthorpe ret = alloc_port_data(device); 753c2261dd7SJason Gunthorpe if (ret) 754c2261dd7SJason Gunthorpe return ret; 755c2261dd7SJason Gunthorpe 756c2261dd7SJason Gunthorpe rdma_for_each_port (device, port) { 757c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 7588ceb1357SJason Gunthorpe 7598ceb1357SJason Gunthorpe ret = device->ops.get_port_immutable(device, port, 7608ceb1357SJason Gunthorpe &pdata->immutable); 7615eb620c8SYosef Etigin if (ret) 7625eb620c8SYosef Etigin return ret; 76355aeed06SJason Gunthorpe 76455aeed06SJason Gunthorpe if (verify_immutable(device, port)) 76555aeed06SJason Gunthorpe return -EINVAL; 76655aeed06SJason Gunthorpe } 76755aeed06SJason Gunthorpe return 0; 7685eb620c8SYosef Etigin } 7695eb620c8SYosef Etigin 7709abb0d1bSLeon Romanovsky void ib_get_device_fw_str(struct ib_device *dev, char *str) 7715fa76c20SIra Weiny { 7723023a1e9SKamal Heib if (dev->ops.get_dev_fw_str) 7733023a1e9SKamal Heib dev->ops.get_dev_fw_str(dev, str); 7745fa76c20SIra Weiny else 7755fa76c20SIra Weiny str[0] = '\0'; 7765fa76c20SIra Weiny } 7775fa76c20SIra Weiny EXPORT_SYMBOL(ib_get_device_fw_str); 7785fa76c20SIra Weiny 7798f408ab6SDaniel Jurgens static void ib_policy_change_task(struct work_struct *work) 7808f408ab6SDaniel Jurgens { 7818f408ab6SDaniel Jurgens struct ib_device *dev; 7820df91bb6SJason Gunthorpe unsigned long index; 7838f408ab6SDaniel Jurgens 784921eab11SJason Gunthorpe down_read(&devices_rwsem); 7850df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 786ea1075edSJason Gunthorpe unsigned int i; 7878f408ab6SDaniel Jurgens 788ea1075edSJason Gunthorpe rdma_for_each_port (dev, i) { 7898f408ab6SDaniel Jurgens u64 sp; 7908f408ab6SDaniel Jurgens int ret = ib_get_cached_subnet_prefix(dev, 7918f408ab6SDaniel Jurgens i, 7928f408ab6SDaniel Jurgens &sp); 7938f408ab6SDaniel Jurgens 7948f408ab6SDaniel Jurgens WARN_ONCE(ret, 7958f408ab6SDaniel Jurgens "ib_get_cached_subnet_prefix err: %d, this should never happen here\n", 7968f408ab6SDaniel Jurgens ret); 797a750cfdeSDaniel Jurgens if (!ret) 7988f408ab6SDaniel Jurgens ib_security_cache_change(dev, i, sp); 7998f408ab6SDaniel Jurgens } 8008f408ab6SDaniel Jurgens } 801921eab11SJason Gunthorpe up_read(&devices_rwsem); 8028f408ab6SDaniel Jurgens } 8038f408ab6SDaniel Jurgens 8048f408ab6SDaniel Jurgens static int ib_security_change(struct notifier_block *nb, unsigned long event, 8058f408ab6SDaniel Jurgens void *lsm_data) 8068f408ab6SDaniel Jurgens { 8078f408ab6SDaniel Jurgens if (event != LSM_POLICY_CHANGE) 8088f408ab6SDaniel Jurgens return NOTIFY_DONE; 8098f408ab6SDaniel Jurgens 8108f408ab6SDaniel Jurgens schedule_work(&ib_policy_change_work); 811c66f6741SDaniel Jurgens ib_mad_agent_security_change(); 8128f408ab6SDaniel Jurgens 8138f408ab6SDaniel Jurgens return NOTIFY_OK; 8148f408ab6SDaniel Jurgens } 8158f408ab6SDaniel Jurgens 8164e0f7b90SParav Pandit static void compatdev_release(struct device *dev) 8174e0f7b90SParav Pandit { 8184e0f7b90SParav Pandit struct ib_core_device *cdev = 8194e0f7b90SParav Pandit container_of(dev, struct ib_core_device, dev); 8204e0f7b90SParav Pandit 8214e0f7b90SParav Pandit kfree(cdev); 8224e0f7b90SParav Pandit } 8234e0f7b90SParav Pandit 8244e0f7b90SParav Pandit static int add_one_compat_dev(struct ib_device *device, 8254e0f7b90SParav Pandit struct rdma_dev_net *rnet) 8264e0f7b90SParav Pandit { 8274e0f7b90SParav Pandit struct ib_core_device *cdev; 8284e0f7b90SParav Pandit int ret; 8294e0f7b90SParav Pandit 8302b34c558SParav Pandit lockdep_assert_held(&rdma_nets_rwsem); 831a56bc45bSParav Pandit if (!ib_devices_shared_netns) 832a56bc45bSParav Pandit return 0; 833a56bc45bSParav Pandit 8344e0f7b90SParav Pandit /* 8354e0f7b90SParav Pandit * Create and add compat device in all namespaces other than where it 8364e0f7b90SParav Pandit * is currently bound to. 8374e0f7b90SParav Pandit */ 8384e0f7b90SParav Pandit if (net_eq(read_pnet(&rnet->net), 8394e0f7b90SParav Pandit read_pnet(&device->coredev.rdma_net))) 8404e0f7b90SParav Pandit return 0; 8414e0f7b90SParav Pandit 8424e0f7b90SParav Pandit /* 8434e0f7b90SParav Pandit * The first of init_net() or ib_register_device() to take the 8444e0f7b90SParav Pandit * compat_devs_mutex wins and gets to add the device. Others will wait 8454e0f7b90SParav Pandit * for completion here. 8464e0f7b90SParav Pandit */ 8474e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 8484e0f7b90SParav Pandit cdev = xa_load(&device->compat_devs, rnet->id); 8494e0f7b90SParav Pandit if (cdev) { 8504e0f7b90SParav Pandit ret = 0; 8514e0f7b90SParav Pandit goto done; 8524e0f7b90SParav Pandit } 8534e0f7b90SParav Pandit ret = xa_reserve(&device->compat_devs, rnet->id, GFP_KERNEL); 8544e0f7b90SParav Pandit if (ret) 8554e0f7b90SParav Pandit goto done; 8564e0f7b90SParav Pandit 8574e0f7b90SParav Pandit cdev = kzalloc(sizeof(*cdev), GFP_KERNEL); 8584e0f7b90SParav Pandit if (!cdev) { 8594e0f7b90SParav Pandit ret = -ENOMEM; 8604e0f7b90SParav Pandit goto cdev_err; 8614e0f7b90SParav Pandit } 8624e0f7b90SParav Pandit 8634e0f7b90SParav Pandit cdev->dev.parent = device->dev.parent; 8644e0f7b90SParav Pandit rdma_init_coredev(cdev, device, read_pnet(&rnet->net)); 8654e0f7b90SParav Pandit cdev->dev.release = compatdev_release; 8664e0f7b90SParav Pandit dev_set_name(&cdev->dev, "%s", dev_name(&device->dev)); 8674e0f7b90SParav Pandit 8684e0f7b90SParav Pandit ret = device_add(&cdev->dev); 8694e0f7b90SParav Pandit if (ret) 8704e0f7b90SParav Pandit goto add_err; 871eb15c78bSParav Pandit ret = ib_setup_port_attrs(cdev); 8725417783eSParav Pandit if (ret) 8735417783eSParav Pandit goto port_err; 8744e0f7b90SParav Pandit 8754e0f7b90SParav Pandit ret = xa_err(xa_store(&device->compat_devs, rnet->id, 8764e0f7b90SParav Pandit cdev, GFP_KERNEL)); 8774e0f7b90SParav Pandit if (ret) 8784e0f7b90SParav Pandit goto insert_err; 8794e0f7b90SParav Pandit 8804e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 8814e0f7b90SParav Pandit return 0; 8824e0f7b90SParav Pandit 8834e0f7b90SParav Pandit insert_err: 8845417783eSParav Pandit ib_free_port_attrs(cdev); 8855417783eSParav Pandit port_err: 8864e0f7b90SParav Pandit device_del(&cdev->dev); 8874e0f7b90SParav Pandit add_err: 8884e0f7b90SParav Pandit put_device(&cdev->dev); 8894e0f7b90SParav Pandit cdev_err: 8904e0f7b90SParav Pandit xa_release(&device->compat_devs, rnet->id); 8914e0f7b90SParav Pandit done: 8924e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 8934e0f7b90SParav Pandit return ret; 8944e0f7b90SParav Pandit } 8954e0f7b90SParav Pandit 8964e0f7b90SParav Pandit static void remove_one_compat_dev(struct ib_device *device, u32 id) 8974e0f7b90SParav Pandit { 8984e0f7b90SParav Pandit struct ib_core_device *cdev; 8994e0f7b90SParav Pandit 9004e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 9014e0f7b90SParav Pandit cdev = xa_erase(&device->compat_devs, id); 9024e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 9034e0f7b90SParav Pandit if (cdev) { 9045417783eSParav Pandit ib_free_port_attrs(cdev); 9054e0f7b90SParav Pandit device_del(&cdev->dev); 9064e0f7b90SParav Pandit put_device(&cdev->dev); 9074e0f7b90SParav Pandit } 9084e0f7b90SParav Pandit } 9094e0f7b90SParav Pandit 9104e0f7b90SParav Pandit static void remove_compat_devs(struct ib_device *device) 9114e0f7b90SParav Pandit { 9124e0f7b90SParav Pandit struct ib_core_device *cdev; 9134e0f7b90SParav Pandit unsigned long index; 9144e0f7b90SParav Pandit 9154e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) 9164e0f7b90SParav Pandit remove_one_compat_dev(device, index); 9174e0f7b90SParav Pandit } 9184e0f7b90SParav Pandit 9194e0f7b90SParav Pandit static int add_compat_devs(struct ib_device *device) 9204e0f7b90SParav Pandit { 9214e0f7b90SParav Pandit struct rdma_dev_net *rnet; 9224e0f7b90SParav Pandit unsigned long index; 9234e0f7b90SParav Pandit int ret = 0; 9244e0f7b90SParav Pandit 925decbc7a6SParav Pandit lockdep_assert_held(&devices_rwsem); 926decbc7a6SParav Pandit 9274e0f7b90SParav Pandit down_read(&rdma_nets_rwsem); 9284e0f7b90SParav Pandit xa_for_each (&rdma_nets, index, rnet) { 9294e0f7b90SParav Pandit ret = add_one_compat_dev(device, rnet); 9304e0f7b90SParav Pandit if (ret) 9314e0f7b90SParav Pandit break; 9324e0f7b90SParav Pandit } 9334e0f7b90SParav Pandit up_read(&rdma_nets_rwsem); 9344e0f7b90SParav Pandit return ret; 9354e0f7b90SParav Pandit } 9364e0f7b90SParav Pandit 9372b34c558SParav Pandit static void remove_all_compat_devs(void) 9382b34c558SParav Pandit { 9392b34c558SParav Pandit struct ib_compat_device *cdev; 9402b34c558SParav Pandit struct ib_device *dev; 9412b34c558SParav Pandit unsigned long index; 9422b34c558SParav Pandit 9432b34c558SParav Pandit down_read(&devices_rwsem); 9442b34c558SParav Pandit xa_for_each (&devices, index, dev) { 9452b34c558SParav Pandit unsigned long c_index = 0; 9462b34c558SParav Pandit 9472b34c558SParav Pandit /* Hold nets_rwsem so that any other thread modifying this 9482b34c558SParav Pandit * system param can sync with this thread. 9492b34c558SParav Pandit */ 9502b34c558SParav Pandit down_read(&rdma_nets_rwsem); 9512b34c558SParav Pandit xa_for_each (&dev->compat_devs, c_index, cdev) 9522b34c558SParav Pandit remove_one_compat_dev(dev, c_index); 9532b34c558SParav Pandit up_read(&rdma_nets_rwsem); 9542b34c558SParav Pandit } 9552b34c558SParav Pandit up_read(&devices_rwsem); 9562b34c558SParav Pandit } 9572b34c558SParav Pandit 9582b34c558SParav Pandit static int add_all_compat_devs(void) 9592b34c558SParav Pandit { 9602b34c558SParav Pandit struct rdma_dev_net *rnet; 9612b34c558SParav Pandit struct ib_device *dev; 9622b34c558SParav Pandit unsigned long index; 9632b34c558SParav Pandit int ret = 0; 9642b34c558SParav Pandit 9652b34c558SParav Pandit down_read(&devices_rwsem); 9662b34c558SParav Pandit xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 9672b34c558SParav Pandit unsigned long net_index = 0; 9682b34c558SParav Pandit 9692b34c558SParav Pandit /* Hold nets_rwsem so that any other thread modifying this 9702b34c558SParav Pandit * system param can sync with this thread. 9712b34c558SParav Pandit */ 9722b34c558SParav Pandit down_read(&rdma_nets_rwsem); 9732b34c558SParav Pandit xa_for_each (&rdma_nets, net_index, rnet) { 9742b34c558SParav Pandit ret = add_one_compat_dev(dev, rnet); 9752b34c558SParav Pandit if (ret) 9762b34c558SParav Pandit break; 9772b34c558SParav Pandit } 9782b34c558SParav Pandit up_read(&rdma_nets_rwsem); 9792b34c558SParav Pandit } 9802b34c558SParav Pandit up_read(&devices_rwsem); 9812b34c558SParav Pandit if (ret) 9822b34c558SParav Pandit remove_all_compat_devs(); 9832b34c558SParav Pandit return ret; 9842b34c558SParav Pandit } 9852b34c558SParav Pandit 9862b34c558SParav Pandit int rdma_compatdev_set(u8 enable) 9872b34c558SParav Pandit { 9882b34c558SParav Pandit struct rdma_dev_net *rnet; 9892b34c558SParav Pandit unsigned long index; 9902b34c558SParav Pandit int ret = 0; 9912b34c558SParav Pandit 9922b34c558SParav Pandit down_write(&rdma_nets_rwsem); 9932b34c558SParav Pandit if (ib_devices_shared_netns == enable) { 9942b34c558SParav Pandit up_write(&rdma_nets_rwsem); 9952b34c558SParav Pandit return 0; 9962b34c558SParav Pandit } 9972b34c558SParav Pandit 9982b34c558SParav Pandit /* enable/disable of compat devices is not supported 9992b34c558SParav Pandit * when more than default init_net exists. 10002b34c558SParav Pandit */ 10012b34c558SParav Pandit xa_for_each (&rdma_nets, index, rnet) { 10022b34c558SParav Pandit ret++; 10032b34c558SParav Pandit break; 10042b34c558SParav Pandit } 10052b34c558SParav Pandit if (!ret) 10062b34c558SParav Pandit ib_devices_shared_netns = enable; 10072b34c558SParav Pandit up_write(&rdma_nets_rwsem); 10082b34c558SParav Pandit if (ret) 10092b34c558SParav Pandit return -EBUSY; 10102b34c558SParav Pandit 10112b34c558SParav Pandit if (enable) 10122b34c558SParav Pandit ret = add_all_compat_devs(); 10132b34c558SParav Pandit else 10142b34c558SParav Pandit remove_all_compat_devs(); 10152b34c558SParav Pandit return ret; 10162b34c558SParav Pandit } 10172b34c558SParav Pandit 10184e0f7b90SParav Pandit static void rdma_dev_exit_net(struct net *net) 10194e0f7b90SParav Pandit { 10204e0f7b90SParav Pandit struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); 10214e0f7b90SParav Pandit struct ib_device *dev; 10224e0f7b90SParav Pandit unsigned long index; 10234e0f7b90SParav Pandit int ret; 10244e0f7b90SParav Pandit 10254e0f7b90SParav Pandit down_write(&rdma_nets_rwsem); 10264e0f7b90SParav Pandit /* 10274e0f7b90SParav Pandit * Prevent the ID from being re-used and hide the id from xa_for_each. 10284e0f7b90SParav Pandit */ 10294e0f7b90SParav Pandit ret = xa_err(xa_store(&rdma_nets, rnet->id, NULL, GFP_KERNEL)); 10304e0f7b90SParav Pandit WARN_ON(ret); 10314e0f7b90SParav Pandit up_write(&rdma_nets_rwsem); 10324e0f7b90SParav Pandit 10334e0f7b90SParav Pandit down_read(&devices_rwsem); 10344e0f7b90SParav Pandit xa_for_each (&devices, index, dev) { 10354e0f7b90SParav Pandit get_device(&dev->dev); 10364e0f7b90SParav Pandit /* 10374e0f7b90SParav Pandit * Release the devices_rwsem so that pontentially blocking 10384e0f7b90SParav Pandit * device_del, doesn't hold the devices_rwsem for too long. 10394e0f7b90SParav Pandit */ 10404e0f7b90SParav Pandit up_read(&devices_rwsem); 10414e0f7b90SParav Pandit 10424e0f7b90SParav Pandit remove_one_compat_dev(dev, rnet->id); 10434e0f7b90SParav Pandit 1044decbc7a6SParav Pandit /* 1045decbc7a6SParav Pandit * If the real device is in the NS then move it back to init. 1046decbc7a6SParav Pandit */ 1047decbc7a6SParav Pandit rdma_dev_change_netns(dev, net, &init_net); 1048decbc7a6SParav Pandit 10494e0f7b90SParav Pandit put_device(&dev->dev); 10504e0f7b90SParav Pandit down_read(&devices_rwsem); 10514e0f7b90SParav Pandit } 10524e0f7b90SParav Pandit up_read(&devices_rwsem); 10534e0f7b90SParav Pandit 10544e0f7b90SParav Pandit xa_erase(&rdma_nets, rnet->id); 10554e0f7b90SParav Pandit } 10564e0f7b90SParav Pandit 10574e0f7b90SParav Pandit static __net_init int rdma_dev_init_net(struct net *net) 10584e0f7b90SParav Pandit { 10594e0f7b90SParav Pandit struct rdma_dev_net *rnet = net_generic(net, rdma_dev_net_id); 10604e0f7b90SParav Pandit unsigned long index; 10614e0f7b90SParav Pandit struct ib_device *dev; 10624e0f7b90SParav Pandit int ret; 10634e0f7b90SParav Pandit 10644e0f7b90SParav Pandit /* No need to create any compat devices in default init_net. */ 10654e0f7b90SParav Pandit if (net_eq(net, &init_net)) 10664e0f7b90SParav Pandit return 0; 10674e0f7b90SParav Pandit 10684e0f7b90SParav Pandit write_pnet(&rnet->net, net); 10694e0f7b90SParav Pandit 10704e0f7b90SParav Pandit ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b, GFP_KERNEL); 10714e0f7b90SParav Pandit if (ret) 10724e0f7b90SParav Pandit return ret; 10734e0f7b90SParav Pandit 10744e0f7b90SParav Pandit down_read(&devices_rwsem); 10754e0f7b90SParav Pandit xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 10762b34c558SParav Pandit /* Hold nets_rwsem so that netlink command cannot change 10772b34c558SParav Pandit * system configuration for device sharing mode. 10782b34c558SParav Pandit */ 10792b34c558SParav Pandit down_read(&rdma_nets_rwsem); 10804e0f7b90SParav Pandit ret = add_one_compat_dev(dev, rnet); 10812b34c558SParav Pandit up_read(&rdma_nets_rwsem); 10824e0f7b90SParav Pandit if (ret) 10834e0f7b90SParav Pandit break; 10844e0f7b90SParav Pandit } 10854e0f7b90SParav Pandit up_read(&devices_rwsem); 10864e0f7b90SParav Pandit 10874e0f7b90SParav Pandit if (ret) 10884e0f7b90SParav Pandit rdma_dev_exit_net(net); 10894e0f7b90SParav Pandit 10904e0f7b90SParav Pandit return ret; 10914e0f7b90SParav Pandit } 10924e0f7b90SParav Pandit 1093ecc82c53SLeon Romanovsky /* 1094d0899892SJason Gunthorpe * Assign the unique string device name and the unique device index. This is 1095d0899892SJason Gunthorpe * undone by ib_dealloc_device. 1096ecc82c53SLeon Romanovsky */ 10970df91bb6SJason Gunthorpe static int assign_name(struct ib_device *device, const char *name) 10980df91bb6SJason Gunthorpe { 10990df91bb6SJason Gunthorpe static u32 last_id; 11000df91bb6SJason Gunthorpe int ret; 1101ecc82c53SLeon Romanovsky 1102921eab11SJason Gunthorpe down_write(&devices_rwsem); 11030df91bb6SJason Gunthorpe /* Assign a unique name to the device */ 11040df91bb6SJason Gunthorpe if (strchr(name, '%')) 11050df91bb6SJason Gunthorpe ret = alloc_name(device, name); 11060df91bb6SJason Gunthorpe else 11070df91bb6SJason Gunthorpe ret = dev_set_name(&device->dev, name); 11080df91bb6SJason Gunthorpe if (ret) 11090df91bb6SJason Gunthorpe goto out; 1110ecc82c53SLeon Romanovsky 11110df91bb6SJason Gunthorpe if (__ib_device_get_by_name(dev_name(&device->dev))) { 11120df91bb6SJason Gunthorpe ret = -ENFILE; 11130df91bb6SJason Gunthorpe goto out; 1114ecc82c53SLeon Romanovsky } 11150df91bb6SJason Gunthorpe strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); 11160df91bb6SJason Gunthorpe 1117ea295481SLinus Torvalds ret = xa_alloc_cyclic(&devices, &device->index, device, xa_limit_31b, 1118ea295481SLinus Torvalds &last_id, GFP_KERNEL); 1119ea295481SLinus Torvalds if (ret > 0) 11200df91bb6SJason Gunthorpe ret = 0; 1121921eab11SJason Gunthorpe 11220df91bb6SJason Gunthorpe out: 1123921eab11SJason Gunthorpe up_write(&devices_rwsem); 11240df91bb6SJason Gunthorpe return ret; 11250df91bb6SJason Gunthorpe } 11260df91bb6SJason Gunthorpe 1127548cb4fbSParav Pandit static void setup_dma_device(struct ib_device *device) 11281da177e4SLinus Torvalds { 112999db9494SBart Van Assche struct device *parent = device->dev.parent; 11301da177e4SLinus Torvalds 11310957c29fSBart Van Assche WARN_ON_ONCE(device->dma_device); 11320957c29fSBart Van Assche if (device->dev.dma_ops) { 11330957c29fSBart Van Assche /* 11340957c29fSBart Van Assche * The caller provided custom DMA operations. Copy the 11350957c29fSBart Van Assche * DMA-related fields that are used by e.g. dma_alloc_coherent() 11360957c29fSBart Van Assche * into device->dev. 11370957c29fSBart Van Assche */ 11380957c29fSBart Van Assche device->dma_device = &device->dev; 113902ee9da3SBart Van Assche if (!device->dev.dma_mask) { 114002ee9da3SBart Van Assche if (parent) 114199db9494SBart Van Assche device->dev.dma_mask = parent->dma_mask; 114202ee9da3SBart Van Assche else 114302ee9da3SBart Van Assche WARN_ON_ONCE(true); 114402ee9da3SBart Van Assche } 114502ee9da3SBart Van Assche if (!device->dev.coherent_dma_mask) { 114602ee9da3SBart Van Assche if (parent) 11470957c29fSBart Van Assche device->dev.coherent_dma_mask = 11480957c29fSBart Van Assche parent->coherent_dma_mask; 114902ee9da3SBart Van Assche else 115002ee9da3SBart Van Assche WARN_ON_ONCE(true); 115102ee9da3SBart Van Assche } 11520957c29fSBart Van Assche } else { 11530957c29fSBart Van Assche /* 11540957c29fSBart Van Assche * The caller did not provide custom DMA operations. Use the 11550957c29fSBart Van Assche * DMA mapping operations of the parent device. 11560957c29fSBart Van Assche */ 115702ee9da3SBart Van Assche WARN_ON_ONCE(!parent); 11580957c29fSBart Van Assche device->dma_device = parent; 11590957c29fSBart Van Assche } 1160d10bcf94SShiraz Saleem /* Setup default max segment size for all IB devices */ 1161d10bcf94SShiraz Saleem dma_set_max_seg_size(device->dma_device, SZ_2G); 1162d10bcf94SShiraz Saleem 1163548cb4fbSParav Pandit } 1164548cb4fbSParav Pandit 1165921eab11SJason Gunthorpe /* 1166921eab11SJason Gunthorpe * setup_device() allocates memory and sets up data that requires calling the 1167921eab11SJason Gunthorpe * device ops, this is the only reason these actions are not done during 1168921eab11SJason Gunthorpe * ib_alloc_device. It is undone by ib_dealloc_device(). 1169921eab11SJason Gunthorpe */ 1170548cb4fbSParav Pandit static int setup_device(struct ib_device *device) 1171548cb4fbSParav Pandit { 1172548cb4fbSParav Pandit struct ib_udata uhw = {.outlen = 0, .inlen = 0}; 1173548cb4fbSParav Pandit int ret; 1174548cb4fbSParav Pandit 1175921eab11SJason Gunthorpe setup_dma_device(device); 1176deee3c7eSKamal Heib ib_device_check_mandatory(device); 1177548cb4fbSParav Pandit 11788ceb1357SJason Gunthorpe ret = setup_port_data(device); 1179548cb4fbSParav Pandit if (ret) { 11808ceb1357SJason Gunthorpe dev_warn(&device->dev, "Couldn't create per-port data\n"); 1181548cb4fbSParav Pandit return ret; 1182548cb4fbSParav Pandit } 1183548cb4fbSParav Pandit 1184548cb4fbSParav Pandit memset(&device->attrs, 0, sizeof(device->attrs)); 11853023a1e9SKamal Heib ret = device->ops.query_device(device, &device->attrs, &uhw); 1186548cb4fbSParav Pandit if (ret) { 1187548cb4fbSParav Pandit dev_warn(&device->dev, 1188548cb4fbSParav Pandit "Couldn't query the device attributes\n"); 1189d45f89d5SJason Gunthorpe return ret; 1190548cb4fbSParav Pandit } 1191548cb4fbSParav Pandit 1192548cb4fbSParav Pandit return 0; 1193548cb4fbSParav Pandit } 1194548cb4fbSParav Pandit 1195921eab11SJason Gunthorpe static void disable_device(struct ib_device *device) 1196921eab11SJason Gunthorpe { 1197921eab11SJason Gunthorpe struct ib_client *client; 1198921eab11SJason Gunthorpe 1199921eab11SJason Gunthorpe WARN_ON(!refcount_read(&device->refcount)); 1200921eab11SJason Gunthorpe 1201921eab11SJason Gunthorpe down_write(&devices_rwsem); 1202921eab11SJason Gunthorpe xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); 1203921eab11SJason Gunthorpe up_write(&devices_rwsem); 1204921eab11SJason Gunthorpe 1205921eab11SJason Gunthorpe down_read(&clients_rwsem); 1206921eab11SJason Gunthorpe list_for_each_entry_reverse(client, &client_list, list) 1207921eab11SJason Gunthorpe remove_client_context(device, client->client_id); 1208921eab11SJason Gunthorpe up_read(&clients_rwsem); 1209921eab11SJason Gunthorpe 1210921eab11SJason Gunthorpe /* Pairs with refcount_set in enable_device */ 1211921eab11SJason Gunthorpe ib_device_put(device); 1212921eab11SJason Gunthorpe wait_for_completion(&device->unreg_completion); 1213c2261dd7SJason Gunthorpe 12144e0f7b90SParav Pandit /* 12154e0f7b90SParav Pandit * compat devices must be removed after device refcount drops to zero. 12164e0f7b90SParav Pandit * Otherwise init_net() may add more compatdevs after removing compat 12174e0f7b90SParav Pandit * devices and before device is disabled. 12184e0f7b90SParav Pandit */ 12194e0f7b90SParav Pandit remove_compat_devs(device); 1220921eab11SJason Gunthorpe } 1221921eab11SJason Gunthorpe 1222921eab11SJason Gunthorpe /* 1223921eab11SJason Gunthorpe * An enabled device is visible to all clients and to all the public facing 1224d0899892SJason Gunthorpe * APIs that return a device pointer. This always returns with a new get, even 1225d0899892SJason Gunthorpe * if it fails. 1226921eab11SJason Gunthorpe */ 1227d0899892SJason Gunthorpe static int enable_device_and_get(struct ib_device *device) 1228921eab11SJason Gunthorpe { 1229921eab11SJason Gunthorpe struct ib_client *client; 1230921eab11SJason Gunthorpe unsigned long index; 1231d0899892SJason Gunthorpe int ret = 0; 1232921eab11SJason Gunthorpe 1233d0899892SJason Gunthorpe /* 1234d0899892SJason Gunthorpe * One ref belongs to the xa and the other belongs to this 1235d0899892SJason Gunthorpe * thread. This is needed to guard against parallel unregistration. 1236d0899892SJason Gunthorpe */ 1237d0899892SJason Gunthorpe refcount_set(&device->refcount, 2); 1238921eab11SJason Gunthorpe down_write(&devices_rwsem); 1239921eab11SJason Gunthorpe xa_set_mark(&devices, device->index, DEVICE_REGISTERED); 1240d0899892SJason Gunthorpe 1241d0899892SJason Gunthorpe /* 1242d0899892SJason Gunthorpe * By using downgrade_write() we ensure that no other thread can clear 1243d0899892SJason Gunthorpe * DEVICE_REGISTERED while we are completing the client setup. 1244d0899892SJason Gunthorpe */ 1245d0899892SJason Gunthorpe downgrade_write(&devices_rwsem); 1246921eab11SJason Gunthorpe 1247ca22354bSJason Gunthorpe if (device->ops.enable_driver) { 1248ca22354bSJason Gunthorpe ret = device->ops.enable_driver(device); 1249ca22354bSJason Gunthorpe if (ret) 1250ca22354bSJason Gunthorpe goto out; 1251ca22354bSJason Gunthorpe } 1252ca22354bSJason Gunthorpe 1253921eab11SJason Gunthorpe down_read(&clients_rwsem); 1254921eab11SJason Gunthorpe xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { 1255921eab11SJason Gunthorpe ret = add_client_context(device, client); 1256d0899892SJason Gunthorpe if (ret) 1257d0899892SJason Gunthorpe break; 1258d0899892SJason Gunthorpe } 1259921eab11SJason Gunthorpe up_read(&clients_rwsem); 12604e0f7b90SParav Pandit if (!ret) 12614e0f7b90SParav Pandit ret = add_compat_devs(device); 1262ca22354bSJason Gunthorpe out: 1263d0899892SJason Gunthorpe up_read(&devices_rwsem); 1264921eab11SJason Gunthorpe return ret; 1265921eab11SJason Gunthorpe } 1266921eab11SJason Gunthorpe 1267548cb4fbSParav Pandit /** 1268548cb4fbSParav Pandit * ib_register_device - Register an IB device with IB core 1269548cb4fbSParav Pandit * @device:Device to register 1270548cb4fbSParav Pandit * 1271548cb4fbSParav Pandit * Low-level drivers use ib_register_device() to register their 1272548cb4fbSParav Pandit * devices with the IB core. All registered clients will receive a 1273548cb4fbSParav Pandit * callback for each device that is added. @device must be allocated 1274548cb4fbSParav Pandit * with ib_alloc_device(). 1275d0899892SJason Gunthorpe * 1276d0899892SJason Gunthorpe * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() 1277d0899892SJason Gunthorpe * asynchronously then the device pointer may become freed as soon as this 1278d0899892SJason Gunthorpe * function returns. 1279548cb4fbSParav Pandit */ 1280ea4baf7fSParav Pandit int ib_register_device(struct ib_device *device, const char *name) 1281548cb4fbSParav Pandit { 1282548cb4fbSParav Pandit int ret; 12831da177e4SLinus Torvalds 12840df91bb6SJason Gunthorpe ret = assign_name(device, name); 1285e349f858SJason Gunthorpe if (ret) 1286921eab11SJason Gunthorpe return ret; 12871da177e4SLinus Torvalds 1288548cb4fbSParav Pandit ret = setup_device(device); 1289548cb4fbSParav Pandit if (ret) 1290d0899892SJason Gunthorpe return ret; 129103db3a2dSMatan Barak 1292d45f89d5SJason Gunthorpe ret = ib_cache_setup_one(device); 1293d45f89d5SJason Gunthorpe if (ret) { 1294d45f89d5SJason Gunthorpe dev_warn(&device->dev, 1295d45f89d5SJason Gunthorpe "Couldn't set up InfiniBand P_Key/GID cache\n"); 1296d0899892SJason Gunthorpe return ret; 1297d45f89d5SJason Gunthorpe } 1298d45f89d5SJason Gunthorpe 12997527a7b1SParav Pandit ib_device_register_rdmacg(device); 13003e153a93SIra Weiny 1301e7a5b4aaSLeon Romanovsky /* 1302e7a5b4aaSLeon Romanovsky * Ensure that ADD uevent is not fired because it 1303e7a5b4aaSLeon Romanovsky * is too early amd device is not initialized yet. 1304e7a5b4aaSLeon Romanovsky */ 1305e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, true); 13065f8f5499SParav Pandit ret = device_add(&device->dev); 13075f8f5499SParav Pandit if (ret) 13085f8f5499SParav Pandit goto cg_cleanup; 13095f8f5499SParav Pandit 1310ea4baf7fSParav Pandit ret = ib_device_register_sysfs(device); 13111da177e4SLinus Torvalds if (ret) { 131243c7c851SJason Gunthorpe dev_warn(&device->dev, 131343c7c851SJason Gunthorpe "Couldn't register device with driver model\n"); 13145f8f5499SParav Pandit goto dev_cleanup; 13151da177e4SLinus Torvalds } 13161da177e4SLinus Torvalds 1317d0899892SJason Gunthorpe ret = enable_device_and_get(device); 1318e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, false); 1319e7a5b4aaSLeon Romanovsky /* Mark for userspace that device is ready */ 1320e7a5b4aaSLeon Romanovsky kobject_uevent(&device->dev.kobj, KOBJ_ADD); 1321d0899892SJason Gunthorpe if (ret) { 1322d0899892SJason Gunthorpe void (*dealloc_fn)(struct ib_device *); 1323d0899892SJason Gunthorpe 1324d0899892SJason Gunthorpe /* 1325d0899892SJason Gunthorpe * If we hit this error flow then we don't want to 1326d0899892SJason Gunthorpe * automatically dealloc the device since the caller is 1327d0899892SJason Gunthorpe * expected to call ib_dealloc_device() after 1328d0899892SJason Gunthorpe * ib_register_device() fails. This is tricky due to the 1329d0899892SJason Gunthorpe * possibility for a parallel unregistration along with this 1330d0899892SJason Gunthorpe * error flow. Since we have a refcount here we know any 1331d0899892SJason Gunthorpe * parallel flow is stopped in disable_device and will see the 1332d0899892SJason Gunthorpe * NULL pointers, causing the responsibility to 1333d0899892SJason Gunthorpe * ib_dealloc_device() to revert back to this thread. 1334d0899892SJason Gunthorpe */ 1335d0899892SJason Gunthorpe dealloc_fn = device->ops.dealloc_driver; 1336d0899892SJason Gunthorpe device->ops.dealloc_driver = NULL; 1337d0899892SJason Gunthorpe ib_device_put(device); 1338d0899892SJason Gunthorpe __ib_unregister_device(device); 1339d0899892SJason Gunthorpe device->ops.dealloc_driver = dealloc_fn; 1340d0899892SJason Gunthorpe return ret; 1341d0899892SJason Gunthorpe } 1342d0899892SJason Gunthorpe ib_device_put(device); 13431da177e4SLinus Torvalds 13444be3a4faSParav Pandit return 0; 13454be3a4faSParav Pandit 13465f8f5499SParav Pandit dev_cleanup: 13475f8f5499SParav Pandit device_del(&device->dev); 13482fb4f4eaSParav Pandit cg_cleanup: 1349e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, false); 13502fb4f4eaSParav Pandit ib_device_unregister_rdmacg(device); 1351d45f89d5SJason Gunthorpe ib_cache_cleanup_one(device); 13521da177e4SLinus Torvalds return ret; 13531da177e4SLinus Torvalds } 13541da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_device); 13551da177e4SLinus Torvalds 1356d0899892SJason Gunthorpe /* Callers must hold a get on the device. */ 1357d0899892SJason Gunthorpe static void __ib_unregister_device(struct ib_device *ib_dev) 1358d0899892SJason Gunthorpe { 1359d0899892SJason Gunthorpe /* 1360d0899892SJason Gunthorpe * We have a registration lock so that all the calls to unregister are 1361d0899892SJason Gunthorpe * fully fenced, once any unregister returns the device is truely 1362d0899892SJason Gunthorpe * unregistered even if multiple callers are unregistering it at the 1363d0899892SJason Gunthorpe * same time. This also interacts with the registration flow and 1364d0899892SJason Gunthorpe * provides sane semantics if register and unregister are racing. 1365d0899892SJason Gunthorpe */ 1366d0899892SJason Gunthorpe mutex_lock(&ib_dev->unregistration_lock); 1367d0899892SJason Gunthorpe if (!refcount_read(&ib_dev->refcount)) 1368d0899892SJason Gunthorpe goto out; 1369d0899892SJason Gunthorpe 1370d0899892SJason Gunthorpe disable_device(ib_dev); 13713042492bSParav Pandit 13723042492bSParav Pandit /* Expedite removing unregistered pointers from the hash table */ 13733042492bSParav Pandit free_netdevs(ib_dev); 13743042492bSParav Pandit 1375d0899892SJason Gunthorpe ib_device_unregister_sysfs(ib_dev); 1376d0899892SJason Gunthorpe device_del(&ib_dev->dev); 1377d0899892SJason Gunthorpe ib_device_unregister_rdmacg(ib_dev); 1378d0899892SJason Gunthorpe ib_cache_cleanup_one(ib_dev); 1379d0899892SJason Gunthorpe 1380d0899892SJason Gunthorpe /* 1381d0899892SJason Gunthorpe * Drivers using the new flow may not call ib_dealloc_device except 1382d0899892SJason Gunthorpe * in error unwind prior to registration success. 1383d0899892SJason Gunthorpe */ 1384d0899892SJason Gunthorpe if (ib_dev->ops.dealloc_driver) { 1385d0899892SJason Gunthorpe WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); 1386d0899892SJason Gunthorpe ib_dealloc_device(ib_dev); 1387d0899892SJason Gunthorpe } 1388d0899892SJason Gunthorpe out: 1389d0899892SJason Gunthorpe mutex_unlock(&ib_dev->unregistration_lock); 1390d0899892SJason Gunthorpe } 1391d0899892SJason Gunthorpe 13921da177e4SLinus Torvalds /** 13931da177e4SLinus Torvalds * ib_unregister_device - Unregister an IB device 1394d0899892SJason Gunthorpe * @device: The device to unregister 13951da177e4SLinus Torvalds * 13961da177e4SLinus Torvalds * Unregister an IB device. All clients will receive a remove callback. 1397d0899892SJason Gunthorpe * 1398d0899892SJason Gunthorpe * Callers should call this routine only once, and protect against races with 1399d0899892SJason Gunthorpe * registration. Typically it should only be called as part of a remove 1400d0899892SJason Gunthorpe * callback in an implementation of driver core's struct device_driver and 1401d0899892SJason Gunthorpe * related. 1402d0899892SJason Gunthorpe * 1403d0899892SJason Gunthorpe * If ops.dealloc_driver is used then ib_dev will be freed upon return from 1404d0899892SJason Gunthorpe * this function. 14051da177e4SLinus Torvalds */ 1406d0899892SJason Gunthorpe void ib_unregister_device(struct ib_device *ib_dev) 14071da177e4SLinus Torvalds { 1408d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1409d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1410d0899892SJason Gunthorpe put_device(&ib_dev->dev); 14111da177e4SLinus Torvalds } 14121da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_device); 14131da177e4SLinus Torvalds 1414d0899892SJason Gunthorpe /** 1415d0899892SJason Gunthorpe * ib_unregister_device_and_put - Unregister a device while holding a 'get' 1416d0899892SJason Gunthorpe * device: The device to unregister 1417d0899892SJason Gunthorpe * 1418d0899892SJason Gunthorpe * This is the same as ib_unregister_device(), except it includes an internal 1419d0899892SJason Gunthorpe * ib_device_put() that should match a 'get' obtained by the caller. 1420d0899892SJason Gunthorpe * 1421d0899892SJason Gunthorpe * It is safe to call this routine concurrently from multiple threads while 1422d0899892SJason Gunthorpe * holding the 'get'. When the function returns the device is fully 1423d0899892SJason Gunthorpe * unregistered. 1424d0899892SJason Gunthorpe * 1425d0899892SJason Gunthorpe * Drivers using this flow MUST use the driver_unregister callback to clean up 1426d0899892SJason Gunthorpe * their resources associated with the device and dealloc it. 1427d0899892SJason Gunthorpe */ 1428d0899892SJason Gunthorpe void ib_unregister_device_and_put(struct ib_device *ib_dev) 1429d0899892SJason Gunthorpe { 1430d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1431d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1432d0899892SJason Gunthorpe ib_device_put(ib_dev); 1433d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1434d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1435d0899892SJason Gunthorpe } 1436d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_and_put); 1437d0899892SJason Gunthorpe 1438d0899892SJason Gunthorpe /** 1439d0899892SJason Gunthorpe * ib_unregister_driver - Unregister all IB devices for a driver 1440d0899892SJason Gunthorpe * @driver_id: The driver to unregister 1441d0899892SJason Gunthorpe * 1442d0899892SJason Gunthorpe * This implements a fence for device unregistration. It only returns once all 1443d0899892SJason Gunthorpe * devices associated with the driver_id have fully completed their 1444d0899892SJason Gunthorpe * unregistration and returned from ib_unregister_device*(). 1445d0899892SJason Gunthorpe * 1446d0899892SJason Gunthorpe * If device's are not yet unregistered it goes ahead and starts unregistering 1447d0899892SJason Gunthorpe * them. 1448d0899892SJason Gunthorpe * 1449d0899892SJason Gunthorpe * This does not block creation of new devices with the given driver_id, that 1450d0899892SJason Gunthorpe * is the responsibility of the caller. 1451d0899892SJason Gunthorpe */ 1452d0899892SJason Gunthorpe void ib_unregister_driver(enum rdma_driver_id driver_id) 1453d0899892SJason Gunthorpe { 1454d0899892SJason Gunthorpe struct ib_device *ib_dev; 1455d0899892SJason Gunthorpe unsigned long index; 1456d0899892SJason Gunthorpe 1457d0899892SJason Gunthorpe down_read(&devices_rwsem); 1458d0899892SJason Gunthorpe xa_for_each (&devices, index, ib_dev) { 1459b9560a41SJason Gunthorpe if (ib_dev->ops.driver_id != driver_id) 1460d0899892SJason Gunthorpe continue; 1461d0899892SJason Gunthorpe 1462d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1463d0899892SJason Gunthorpe up_read(&devices_rwsem); 1464d0899892SJason Gunthorpe 1465d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1466d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1467d0899892SJason Gunthorpe 1468d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1469d0899892SJason Gunthorpe down_read(&devices_rwsem); 1470d0899892SJason Gunthorpe } 1471d0899892SJason Gunthorpe up_read(&devices_rwsem); 1472d0899892SJason Gunthorpe } 1473d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_driver); 1474d0899892SJason Gunthorpe 1475d0899892SJason Gunthorpe static void ib_unregister_work(struct work_struct *work) 1476d0899892SJason Gunthorpe { 1477d0899892SJason Gunthorpe struct ib_device *ib_dev = 1478d0899892SJason Gunthorpe container_of(work, struct ib_device, unregistration_work); 1479d0899892SJason Gunthorpe 1480d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1481d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1482d0899892SJason Gunthorpe } 1483d0899892SJason Gunthorpe 1484d0899892SJason Gunthorpe /** 1485d0899892SJason Gunthorpe * ib_unregister_device_queued - Unregister a device using a work queue 1486d0899892SJason Gunthorpe * device: The device to unregister 1487d0899892SJason Gunthorpe * 1488d0899892SJason Gunthorpe * This schedules an asynchronous unregistration using a WQ for the device. A 1489d0899892SJason Gunthorpe * driver should use this to avoid holding locks while doing unregistration, 1490d0899892SJason Gunthorpe * such as holding the RTNL lock. 1491d0899892SJason Gunthorpe * 1492d0899892SJason Gunthorpe * Drivers using this API must use ib_unregister_driver before module unload 1493d0899892SJason Gunthorpe * to ensure that all scheduled unregistrations have completed. 1494d0899892SJason Gunthorpe */ 1495d0899892SJason Gunthorpe void ib_unregister_device_queued(struct ib_device *ib_dev) 1496d0899892SJason Gunthorpe { 1497d0899892SJason Gunthorpe WARN_ON(!refcount_read(&ib_dev->refcount)); 1498d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1499d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1500d0899892SJason Gunthorpe if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) 1501d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1502d0899892SJason Gunthorpe } 1503d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_queued); 1504d0899892SJason Gunthorpe 1505decbc7a6SParav Pandit /* 1506decbc7a6SParav Pandit * The caller must pass in a device that has the kref held and the refcount 1507decbc7a6SParav Pandit * released. If the device is in cur_net and still registered then it is moved 1508decbc7a6SParav Pandit * into net. 1509decbc7a6SParav Pandit */ 1510decbc7a6SParav Pandit static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, 1511decbc7a6SParav Pandit struct net *net) 1512decbc7a6SParav Pandit { 1513decbc7a6SParav Pandit int ret2 = -EINVAL; 1514decbc7a6SParav Pandit int ret; 1515decbc7a6SParav Pandit 1516decbc7a6SParav Pandit mutex_lock(&device->unregistration_lock); 1517decbc7a6SParav Pandit 1518decbc7a6SParav Pandit /* 15192e5b8a01SParav Pandit * If a device not under ib_device_get() or if the unregistration_lock 15202e5b8a01SParav Pandit * is not held, the namespace can be changed, or it can be unregistered. 15212e5b8a01SParav Pandit * Check again under the lock. 1522decbc7a6SParav Pandit */ 1523decbc7a6SParav Pandit if (refcount_read(&device->refcount) == 0 || 1524decbc7a6SParav Pandit !net_eq(cur_net, read_pnet(&device->coredev.rdma_net))) { 1525decbc7a6SParav Pandit ret = -ENODEV; 1526decbc7a6SParav Pandit goto out; 1527decbc7a6SParav Pandit } 1528decbc7a6SParav Pandit 1529decbc7a6SParav Pandit kobject_uevent(&device->dev.kobj, KOBJ_REMOVE); 1530decbc7a6SParav Pandit disable_device(device); 1531decbc7a6SParav Pandit 1532decbc7a6SParav Pandit /* 1533decbc7a6SParav Pandit * At this point no one can be using the device, so it is safe to 1534decbc7a6SParav Pandit * change the namespace. 1535decbc7a6SParav Pandit */ 1536decbc7a6SParav Pandit write_pnet(&device->coredev.rdma_net, net); 1537decbc7a6SParav Pandit 15382e5b8a01SParav Pandit down_read(&devices_rwsem); 1539decbc7a6SParav Pandit /* 1540decbc7a6SParav Pandit * Currently rdma devices are system wide unique. So the device name 1541decbc7a6SParav Pandit * is guaranteed free in the new namespace. Publish the new namespace 1542decbc7a6SParav Pandit * at the sysfs level. 1543decbc7a6SParav Pandit */ 1544decbc7a6SParav Pandit ret = device_rename(&device->dev, dev_name(&device->dev)); 1545decbc7a6SParav Pandit up_read(&devices_rwsem); 1546decbc7a6SParav Pandit if (ret) { 1547decbc7a6SParav Pandit dev_warn(&device->dev, 1548decbc7a6SParav Pandit "%s: Couldn't rename device after namespace change\n", 1549decbc7a6SParav Pandit __func__); 1550decbc7a6SParav Pandit /* Try and put things back and re-enable the device */ 1551decbc7a6SParav Pandit write_pnet(&device->coredev.rdma_net, cur_net); 1552decbc7a6SParav Pandit } 1553decbc7a6SParav Pandit 1554decbc7a6SParav Pandit ret2 = enable_device_and_get(device); 15552e5b8a01SParav Pandit if (ret2) { 1556decbc7a6SParav Pandit /* 1557decbc7a6SParav Pandit * This shouldn't really happen, but if it does, let the user 1558decbc7a6SParav Pandit * retry at later point. So don't disable the device. 1559decbc7a6SParav Pandit */ 1560decbc7a6SParav Pandit dev_warn(&device->dev, 1561decbc7a6SParav Pandit "%s: Couldn't re-enable device after namespace change\n", 1562decbc7a6SParav Pandit __func__); 15632e5b8a01SParav Pandit } 1564decbc7a6SParav Pandit kobject_uevent(&device->dev.kobj, KOBJ_ADD); 15652e5b8a01SParav Pandit 1566decbc7a6SParav Pandit ib_device_put(device); 1567decbc7a6SParav Pandit out: 1568decbc7a6SParav Pandit mutex_unlock(&device->unregistration_lock); 1569decbc7a6SParav Pandit if (ret) 1570decbc7a6SParav Pandit return ret; 1571decbc7a6SParav Pandit return ret2; 1572decbc7a6SParav Pandit } 1573decbc7a6SParav Pandit 15742e5b8a01SParav Pandit int ib_device_set_netns_put(struct sk_buff *skb, 15752e5b8a01SParav Pandit struct ib_device *dev, u32 ns_fd) 15762e5b8a01SParav Pandit { 15772e5b8a01SParav Pandit struct net *net; 15782e5b8a01SParav Pandit int ret; 15792e5b8a01SParav Pandit 15802e5b8a01SParav Pandit net = get_net_ns_by_fd(ns_fd); 15812e5b8a01SParav Pandit if (IS_ERR(net)) { 15822e5b8a01SParav Pandit ret = PTR_ERR(net); 15832e5b8a01SParav Pandit goto net_err; 15842e5b8a01SParav Pandit } 15852e5b8a01SParav Pandit 15862e5b8a01SParav Pandit if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) { 15872e5b8a01SParav Pandit ret = -EPERM; 15882e5b8a01SParav Pandit goto ns_err; 15892e5b8a01SParav Pandit } 15902e5b8a01SParav Pandit 15912e5b8a01SParav Pandit /* 15922e5b8a01SParav Pandit * Currently supported only for those providers which support 15932e5b8a01SParav Pandit * disassociation and don't do port specific sysfs init. Once a 15942e5b8a01SParav Pandit * port_cleanup infrastructure is implemented, this limitation will be 15952e5b8a01SParav Pandit * removed. 15962e5b8a01SParav Pandit */ 15972e5b8a01SParav Pandit if (!dev->ops.disassociate_ucontext || dev->ops.init_port || 15982e5b8a01SParav Pandit ib_devices_shared_netns) { 15992e5b8a01SParav Pandit ret = -EOPNOTSUPP; 16002e5b8a01SParav Pandit goto ns_err; 16012e5b8a01SParav Pandit } 16022e5b8a01SParav Pandit 16032e5b8a01SParav Pandit get_device(&dev->dev); 16042e5b8a01SParav Pandit ib_device_put(dev); 16052e5b8a01SParav Pandit ret = rdma_dev_change_netns(dev, current->nsproxy->net_ns, net); 16062e5b8a01SParav Pandit put_device(&dev->dev); 16072e5b8a01SParav Pandit 16082e5b8a01SParav Pandit put_net(net); 16092e5b8a01SParav Pandit return ret; 16102e5b8a01SParav Pandit 16112e5b8a01SParav Pandit ns_err: 16122e5b8a01SParav Pandit put_net(net); 16132e5b8a01SParav Pandit net_err: 16142e5b8a01SParav Pandit ib_device_put(dev); 16152e5b8a01SParav Pandit return ret; 16162e5b8a01SParav Pandit } 16172e5b8a01SParav Pandit 16184e0f7b90SParav Pandit static struct pernet_operations rdma_dev_net_ops = { 16194e0f7b90SParav Pandit .init = rdma_dev_init_net, 16204e0f7b90SParav Pandit .exit = rdma_dev_exit_net, 16214e0f7b90SParav Pandit .id = &rdma_dev_net_id, 16224e0f7b90SParav Pandit .size = sizeof(struct rdma_dev_net), 16234e0f7b90SParav Pandit }; 16244e0f7b90SParav Pandit 1625e59178d8SJason Gunthorpe static int assign_client_id(struct ib_client *client) 1626e59178d8SJason Gunthorpe { 1627e59178d8SJason Gunthorpe int ret; 1628e59178d8SJason Gunthorpe 1629921eab11SJason Gunthorpe down_write(&clients_rwsem); 1630e59178d8SJason Gunthorpe /* 1631e59178d8SJason Gunthorpe * The add/remove callbacks must be called in FIFO/LIFO order. To 1632e59178d8SJason Gunthorpe * achieve this we assign client_ids so they are sorted in 1633e59178d8SJason Gunthorpe * registration order, and retain a linked list we can reverse iterate 1634e59178d8SJason Gunthorpe * to get the LIFO order. The extra linked list can go away if xarray 1635e59178d8SJason Gunthorpe * learns to reverse iterate. 1636e59178d8SJason Gunthorpe */ 1637ea295481SLinus Torvalds if (list_empty(&client_list)) { 1638e59178d8SJason Gunthorpe client->client_id = 0; 1639ea295481SLinus Torvalds } else { 1640ea295481SLinus Torvalds struct ib_client *last; 1641ea295481SLinus Torvalds 1642ea295481SLinus Torvalds last = list_last_entry(&client_list, struct ib_client, list); 1643ea295481SLinus Torvalds client->client_id = last->client_id + 1; 1644ea295481SLinus Torvalds } 1645ea295481SLinus Torvalds ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); 1646e59178d8SJason Gunthorpe if (ret) 1647e59178d8SJason Gunthorpe goto out; 1648e59178d8SJason Gunthorpe 1649921eab11SJason Gunthorpe xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); 1650921eab11SJason Gunthorpe list_add_tail(&client->list, &client_list); 1651921eab11SJason Gunthorpe 1652e59178d8SJason Gunthorpe out: 1653921eab11SJason Gunthorpe up_write(&clients_rwsem); 1654e59178d8SJason Gunthorpe return ret; 1655e59178d8SJason Gunthorpe } 1656e59178d8SJason Gunthorpe 16571da177e4SLinus Torvalds /** 16581da177e4SLinus Torvalds * ib_register_client - Register an IB client 16591da177e4SLinus Torvalds * @client:Client to register 16601da177e4SLinus Torvalds * 16611da177e4SLinus Torvalds * Upper level users of the IB drivers can use ib_register_client() to 16621da177e4SLinus Torvalds * register callbacks for IB device addition and removal. When an IB 16631da177e4SLinus Torvalds * device is added, each registered client's add method will be called 16641da177e4SLinus Torvalds * (in the order the clients were registered), and when a device is 16651da177e4SLinus Torvalds * removed, each client's remove method will be called (in the reverse 16661da177e4SLinus Torvalds * order that clients were registered). In addition, when 16671da177e4SLinus Torvalds * ib_register_client() is called, the client will receive an add 16681da177e4SLinus Torvalds * callback for all devices already registered. 16691da177e4SLinus Torvalds */ 16701da177e4SLinus Torvalds int ib_register_client(struct ib_client *client) 16711da177e4SLinus Torvalds { 16721da177e4SLinus Torvalds struct ib_device *device; 16730df91bb6SJason Gunthorpe unsigned long index; 1674e59178d8SJason Gunthorpe int ret; 16751da177e4SLinus Torvalds 1676e59178d8SJason Gunthorpe ret = assign_client_id(client); 1677921eab11SJason Gunthorpe if (ret) 1678921eab11SJason Gunthorpe return ret; 1679921eab11SJason Gunthorpe 1680921eab11SJason Gunthorpe down_read(&devices_rwsem); 1681921eab11SJason Gunthorpe xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { 1682921eab11SJason Gunthorpe ret = add_client_context(device, client); 1683e59178d8SJason Gunthorpe if (ret) { 1684921eab11SJason Gunthorpe up_read(&devices_rwsem); 1685921eab11SJason Gunthorpe ib_unregister_client(client); 1686e59178d8SJason Gunthorpe return ret; 1687e59178d8SJason Gunthorpe } 1688921eab11SJason Gunthorpe } 1689921eab11SJason Gunthorpe up_read(&devices_rwsem); 16901da177e4SLinus Torvalds return 0; 16911da177e4SLinus Torvalds } 16921da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_client); 16931da177e4SLinus Torvalds 16941da177e4SLinus Torvalds /** 16951da177e4SLinus Torvalds * ib_unregister_client - Unregister an IB client 16961da177e4SLinus Torvalds * @client:Client to unregister 16971da177e4SLinus Torvalds * 16981da177e4SLinus Torvalds * Upper level users use ib_unregister_client() to remove their client 16991da177e4SLinus Torvalds * registration. When ib_unregister_client() is called, the client 17001da177e4SLinus Torvalds * will receive a remove callback for each IB device still registered. 1701921eab11SJason Gunthorpe * 1702921eab11SJason Gunthorpe * This is a full fence, once it returns no client callbacks will be called, 1703921eab11SJason Gunthorpe * or are running in another thread. 17041da177e4SLinus Torvalds */ 17051da177e4SLinus Torvalds void ib_unregister_client(struct ib_client *client) 17061da177e4SLinus Torvalds { 17071da177e4SLinus Torvalds struct ib_device *device; 17080df91bb6SJason Gunthorpe unsigned long index; 17091da177e4SLinus Torvalds 1710921eab11SJason Gunthorpe down_write(&clients_rwsem); 1711e59178d8SJason Gunthorpe xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); 1712921eab11SJason Gunthorpe up_write(&clients_rwsem); 1713921eab11SJason Gunthorpe /* 1714921eab11SJason Gunthorpe * Every device still known must be serialized to make sure we are 1715921eab11SJason Gunthorpe * done with the client callbacks before we return. 1716921eab11SJason Gunthorpe */ 1717921eab11SJason Gunthorpe down_read(&devices_rwsem); 1718921eab11SJason Gunthorpe xa_for_each (&devices, index, device) 1719921eab11SJason Gunthorpe remove_client_context(device, client->client_id); 1720921eab11SJason Gunthorpe up_read(&devices_rwsem); 17215aa44bb9SHaggai Eran 1722921eab11SJason Gunthorpe down_write(&clients_rwsem); 1723e59178d8SJason Gunthorpe list_del(&client->list); 1724e59178d8SJason Gunthorpe xa_erase(&clients, client->client_id); 1725921eab11SJason Gunthorpe up_write(&clients_rwsem); 17261da177e4SLinus Torvalds } 17271da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_client); 17281da177e4SLinus Torvalds 17290e2d00ebSJason Gunthorpe static int __ib_get_global_client_nl_info(const char *client_name, 17300e2d00ebSJason Gunthorpe struct ib_client_nl_info *res) 17310e2d00ebSJason Gunthorpe { 17320e2d00ebSJason Gunthorpe struct ib_client *client; 17330e2d00ebSJason Gunthorpe unsigned long index; 17340e2d00ebSJason Gunthorpe int ret = -ENOENT; 17350e2d00ebSJason Gunthorpe 17360e2d00ebSJason Gunthorpe down_read(&clients_rwsem); 17370e2d00ebSJason Gunthorpe xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { 17380e2d00ebSJason Gunthorpe if (strcmp(client->name, client_name) != 0) 17390e2d00ebSJason Gunthorpe continue; 17400e2d00ebSJason Gunthorpe if (!client->get_global_nl_info) { 17410e2d00ebSJason Gunthorpe ret = -EOPNOTSUPP; 17420e2d00ebSJason Gunthorpe break; 17430e2d00ebSJason Gunthorpe } 17440e2d00ebSJason Gunthorpe ret = client->get_global_nl_info(res); 17450e2d00ebSJason Gunthorpe if (WARN_ON(ret == -ENOENT)) 17460e2d00ebSJason Gunthorpe ret = -EINVAL; 17470e2d00ebSJason Gunthorpe if (!ret && res->cdev) 17480e2d00ebSJason Gunthorpe get_device(res->cdev); 17490e2d00ebSJason Gunthorpe break; 17500e2d00ebSJason Gunthorpe } 17510e2d00ebSJason Gunthorpe up_read(&clients_rwsem); 17520e2d00ebSJason Gunthorpe return ret; 17530e2d00ebSJason Gunthorpe } 17540e2d00ebSJason Gunthorpe 17550e2d00ebSJason Gunthorpe static int __ib_get_client_nl_info(struct ib_device *ibdev, 17560e2d00ebSJason Gunthorpe const char *client_name, 17570e2d00ebSJason Gunthorpe struct ib_client_nl_info *res) 17580e2d00ebSJason Gunthorpe { 17590e2d00ebSJason Gunthorpe unsigned long index; 17600e2d00ebSJason Gunthorpe void *client_data; 17610e2d00ebSJason Gunthorpe int ret = -ENOENT; 17620e2d00ebSJason Gunthorpe 17630e2d00ebSJason Gunthorpe down_read(&ibdev->client_data_rwsem); 17640e2d00ebSJason Gunthorpe xan_for_each_marked (&ibdev->client_data, index, client_data, 17650e2d00ebSJason Gunthorpe CLIENT_DATA_REGISTERED) { 17660e2d00ebSJason Gunthorpe struct ib_client *client = xa_load(&clients, index); 17670e2d00ebSJason Gunthorpe 17680e2d00ebSJason Gunthorpe if (!client || strcmp(client->name, client_name) != 0) 17690e2d00ebSJason Gunthorpe continue; 17700e2d00ebSJason Gunthorpe if (!client->get_nl_info) { 17710e2d00ebSJason Gunthorpe ret = -EOPNOTSUPP; 17720e2d00ebSJason Gunthorpe break; 17730e2d00ebSJason Gunthorpe } 17740e2d00ebSJason Gunthorpe ret = client->get_nl_info(ibdev, client_data, res); 17750e2d00ebSJason Gunthorpe if (WARN_ON(ret == -ENOENT)) 17760e2d00ebSJason Gunthorpe ret = -EINVAL; 17770e2d00ebSJason Gunthorpe 17780e2d00ebSJason Gunthorpe /* 17790e2d00ebSJason Gunthorpe * The cdev is guaranteed valid as long as we are inside the 17800e2d00ebSJason Gunthorpe * client_data_rwsem as remove_one can't be called. Keep it 17810e2d00ebSJason Gunthorpe * valid for the caller. 17820e2d00ebSJason Gunthorpe */ 17830e2d00ebSJason Gunthorpe if (!ret && res->cdev) 17840e2d00ebSJason Gunthorpe get_device(res->cdev); 17850e2d00ebSJason Gunthorpe break; 17860e2d00ebSJason Gunthorpe } 17870e2d00ebSJason Gunthorpe up_read(&ibdev->client_data_rwsem); 17880e2d00ebSJason Gunthorpe 17890e2d00ebSJason Gunthorpe return ret; 17900e2d00ebSJason Gunthorpe } 17910e2d00ebSJason Gunthorpe 17920e2d00ebSJason Gunthorpe /** 17930e2d00ebSJason Gunthorpe * ib_get_client_nl_info - Fetch the nl_info from a client 17940e2d00ebSJason Gunthorpe * @device - IB device 17950e2d00ebSJason Gunthorpe * @client_name - Name of the client 17960e2d00ebSJason Gunthorpe * @res - Result of the query 17970e2d00ebSJason Gunthorpe */ 17980e2d00ebSJason Gunthorpe int ib_get_client_nl_info(struct ib_device *ibdev, const char *client_name, 17990e2d00ebSJason Gunthorpe struct ib_client_nl_info *res) 18000e2d00ebSJason Gunthorpe { 18010e2d00ebSJason Gunthorpe int ret; 18020e2d00ebSJason Gunthorpe 18030e2d00ebSJason Gunthorpe if (ibdev) 18040e2d00ebSJason Gunthorpe ret = __ib_get_client_nl_info(ibdev, client_name, res); 18050e2d00ebSJason Gunthorpe else 18060e2d00ebSJason Gunthorpe ret = __ib_get_global_client_nl_info(client_name, res); 18070e2d00ebSJason Gunthorpe #ifdef CONFIG_MODULES 18080e2d00ebSJason Gunthorpe if (ret == -ENOENT) { 18090e2d00ebSJason Gunthorpe request_module("rdma-client-%s", client_name); 18100e2d00ebSJason Gunthorpe if (ibdev) 18110e2d00ebSJason Gunthorpe ret = __ib_get_client_nl_info(ibdev, client_name, res); 18120e2d00ebSJason Gunthorpe else 18130e2d00ebSJason Gunthorpe ret = __ib_get_global_client_nl_info(client_name, res); 18140e2d00ebSJason Gunthorpe } 18150e2d00ebSJason Gunthorpe #endif 18160e2d00ebSJason Gunthorpe if (ret) { 18170e2d00ebSJason Gunthorpe if (ret == -ENOENT) 18180e2d00ebSJason Gunthorpe return -EOPNOTSUPP; 18190e2d00ebSJason Gunthorpe return ret; 18200e2d00ebSJason Gunthorpe } 18210e2d00ebSJason Gunthorpe 18220e2d00ebSJason Gunthorpe if (WARN_ON(!res->cdev)) 18230e2d00ebSJason Gunthorpe return -EINVAL; 18240e2d00ebSJason Gunthorpe return 0; 18250e2d00ebSJason Gunthorpe } 18260e2d00ebSJason Gunthorpe 18271da177e4SLinus Torvalds /** 18289cd330d3SKrishna Kumar * ib_set_client_data - Set IB client context 18291da177e4SLinus Torvalds * @device:Device to set context for 18301da177e4SLinus Torvalds * @client:Client to set context for 18311da177e4SLinus Torvalds * @data:Context to set 18321da177e4SLinus Torvalds * 18330df91bb6SJason Gunthorpe * ib_set_client_data() sets client context data that can be retrieved with 18340df91bb6SJason Gunthorpe * ib_get_client_data(). This can only be called while the client is 18350df91bb6SJason Gunthorpe * registered to the device, once the ib_client remove() callback returns this 18360df91bb6SJason Gunthorpe * cannot be called. 18371da177e4SLinus Torvalds */ 18381da177e4SLinus Torvalds void ib_set_client_data(struct ib_device *device, struct ib_client *client, 18391da177e4SLinus Torvalds void *data) 18401da177e4SLinus Torvalds { 18410df91bb6SJason Gunthorpe void *rc; 18421da177e4SLinus Torvalds 18430df91bb6SJason Gunthorpe if (WARN_ON(IS_ERR(data))) 18440df91bb6SJason Gunthorpe data = NULL; 18451da177e4SLinus Torvalds 18460df91bb6SJason Gunthorpe rc = xa_store(&device->client_data, client->client_id, data, 18470df91bb6SJason Gunthorpe GFP_KERNEL); 18480df91bb6SJason Gunthorpe WARN_ON(xa_is_err(rc)); 18491da177e4SLinus Torvalds } 18501da177e4SLinus Torvalds EXPORT_SYMBOL(ib_set_client_data); 18511da177e4SLinus Torvalds 18521da177e4SLinus Torvalds /** 18531da177e4SLinus Torvalds * ib_register_event_handler - Register an IB event handler 18541da177e4SLinus Torvalds * @event_handler:Handler to register 18551da177e4SLinus Torvalds * 18561da177e4SLinus Torvalds * ib_register_event_handler() registers an event handler that will be 18571da177e4SLinus Torvalds * called back when asynchronous IB events occur (as defined in 18581da177e4SLinus Torvalds * chapter 11 of the InfiniBand Architecture Specification). This 18591da177e4SLinus Torvalds * callback may occur in interrupt context. 18601da177e4SLinus Torvalds */ 1861dcc9881eSLeon Romanovsky void ib_register_event_handler(struct ib_event_handler *event_handler) 18621da177e4SLinus Torvalds { 18631da177e4SLinus Torvalds unsigned long flags; 18641da177e4SLinus Torvalds 18651da177e4SLinus Torvalds spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 18661da177e4SLinus Torvalds list_add_tail(&event_handler->list, 18671da177e4SLinus Torvalds &event_handler->device->event_handler_list); 18681da177e4SLinus Torvalds spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 18691da177e4SLinus Torvalds } 18701da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_event_handler); 18711da177e4SLinus Torvalds 18721da177e4SLinus Torvalds /** 18731da177e4SLinus Torvalds * ib_unregister_event_handler - Unregister an event handler 18741da177e4SLinus Torvalds * @event_handler:Handler to unregister 18751da177e4SLinus Torvalds * 18761da177e4SLinus Torvalds * Unregister an event handler registered with 18771da177e4SLinus Torvalds * ib_register_event_handler(). 18781da177e4SLinus Torvalds */ 1879dcc9881eSLeon Romanovsky void ib_unregister_event_handler(struct ib_event_handler *event_handler) 18801da177e4SLinus Torvalds { 18811da177e4SLinus Torvalds unsigned long flags; 18821da177e4SLinus Torvalds 18831da177e4SLinus Torvalds spin_lock_irqsave(&event_handler->device->event_handler_lock, flags); 18841da177e4SLinus Torvalds list_del(&event_handler->list); 18851da177e4SLinus Torvalds spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags); 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_event_handler); 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds /** 18901da177e4SLinus Torvalds * ib_dispatch_event - Dispatch an asynchronous event 18911da177e4SLinus Torvalds * @event:Event to dispatch 18921da177e4SLinus Torvalds * 18931da177e4SLinus Torvalds * Low-level drivers must call ib_dispatch_event() to dispatch the 18941da177e4SLinus Torvalds * event to all registered event handlers when an asynchronous event 18951da177e4SLinus Torvalds * occurs. 18961da177e4SLinus Torvalds */ 18971da177e4SLinus Torvalds void ib_dispatch_event(struct ib_event *event) 18981da177e4SLinus Torvalds { 18991da177e4SLinus Torvalds unsigned long flags; 19001da177e4SLinus Torvalds struct ib_event_handler *handler; 19011da177e4SLinus Torvalds 19021da177e4SLinus Torvalds spin_lock_irqsave(&event->device->event_handler_lock, flags); 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds list_for_each_entry(handler, &event->device->event_handler_list, list) 19051da177e4SLinus Torvalds handler->handler(handler, event); 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds spin_unlock_irqrestore(&event->device->event_handler_lock, flags); 19081da177e4SLinus Torvalds } 19091da177e4SLinus Torvalds EXPORT_SYMBOL(ib_dispatch_event); 19101da177e4SLinus Torvalds 19111da177e4SLinus Torvalds /** 19121da177e4SLinus Torvalds * ib_query_port - Query IB port attributes 19131da177e4SLinus Torvalds * @device:Device to query 19141da177e4SLinus Torvalds * @port_num:Port number to query 19151da177e4SLinus Torvalds * @port_attr:Port attributes 19161da177e4SLinus Torvalds * 19171da177e4SLinus Torvalds * ib_query_port() returns the attributes of a port through the 19181da177e4SLinus Torvalds * @port_attr pointer. 19191da177e4SLinus Torvalds */ 19201da177e4SLinus Torvalds int ib_query_port(struct ib_device *device, 19211da177e4SLinus Torvalds u8 port_num, 19221da177e4SLinus Torvalds struct ib_port_attr *port_attr) 19231da177e4SLinus Torvalds { 1924fad61ad4SEli Cohen union ib_gid gid; 1925fad61ad4SEli Cohen int err; 1926fad61ad4SEli Cohen 192724dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 1928116c0074SRoland Dreier return -EINVAL; 1929116c0074SRoland Dreier 1930fad61ad4SEli Cohen memset(port_attr, 0, sizeof(*port_attr)); 19313023a1e9SKamal Heib err = device->ops.query_port(device, port_num, port_attr); 1932fad61ad4SEli Cohen if (err || port_attr->subnet_prefix) 1933fad61ad4SEli Cohen return err; 1934fad61ad4SEli Cohen 1935d7012467SEli Cohen if (rdma_port_get_link_layer(device, port_num) != IB_LINK_LAYER_INFINIBAND) 1936d7012467SEli Cohen return 0; 1937d7012467SEli Cohen 19383023a1e9SKamal Heib err = device->ops.query_gid(device, port_num, 0, &gid); 1939fad61ad4SEli Cohen if (err) 1940fad61ad4SEli Cohen return err; 1941fad61ad4SEli Cohen 1942fad61ad4SEli Cohen port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); 1943fad61ad4SEli Cohen return 0; 19441da177e4SLinus Torvalds } 19451da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_port); 19461da177e4SLinus Torvalds 1947324e227eSJason Gunthorpe static void add_ndev_hash(struct ib_port_data *pdata) 1948324e227eSJason Gunthorpe { 1949324e227eSJason Gunthorpe unsigned long flags; 1950324e227eSJason Gunthorpe 1951324e227eSJason Gunthorpe might_sleep(); 1952324e227eSJason Gunthorpe 1953324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 1954324e227eSJason Gunthorpe if (hash_hashed(&pdata->ndev_hash_link)) { 1955324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 1956324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 1957324e227eSJason Gunthorpe /* 1958324e227eSJason Gunthorpe * We cannot do hash_add_rcu after a hash_del_rcu until the 1959324e227eSJason Gunthorpe * grace period 1960324e227eSJason Gunthorpe */ 1961324e227eSJason Gunthorpe synchronize_rcu(); 1962324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 1963324e227eSJason Gunthorpe } 1964324e227eSJason Gunthorpe if (pdata->netdev) 1965324e227eSJason Gunthorpe hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, 1966324e227eSJason Gunthorpe (uintptr_t)pdata->netdev); 1967324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 1968324e227eSJason Gunthorpe } 1969324e227eSJason Gunthorpe 19701da177e4SLinus Torvalds /** 1971c2261dd7SJason Gunthorpe * ib_device_set_netdev - Associate the ib_dev with an underlying net_device 1972c2261dd7SJason Gunthorpe * @ib_dev: Device to modify 1973c2261dd7SJason Gunthorpe * @ndev: net_device to affiliate, may be NULL 1974c2261dd7SJason Gunthorpe * @port: IB port the net_device is connected to 1975c2261dd7SJason Gunthorpe * 1976c2261dd7SJason Gunthorpe * Drivers should use this to link the ib_device to a netdev so the netdev 1977c2261dd7SJason Gunthorpe * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be 1978c2261dd7SJason Gunthorpe * affiliated with any port. 1979c2261dd7SJason Gunthorpe * 1980c2261dd7SJason Gunthorpe * The caller must ensure that the given ndev is not unregistered or 1981c2261dd7SJason Gunthorpe * unregistering, and that either the ib_device is unregistered or 1982c2261dd7SJason Gunthorpe * ib_device_set_netdev() is called with NULL when the ndev sends a 1983c2261dd7SJason Gunthorpe * NETDEV_UNREGISTER event. 1984c2261dd7SJason Gunthorpe */ 1985c2261dd7SJason Gunthorpe int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 1986c2261dd7SJason Gunthorpe unsigned int port) 1987c2261dd7SJason Gunthorpe { 1988c2261dd7SJason Gunthorpe struct net_device *old_ndev; 1989c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 1990c2261dd7SJason Gunthorpe unsigned long flags; 1991c2261dd7SJason Gunthorpe int ret; 1992c2261dd7SJason Gunthorpe 1993c2261dd7SJason Gunthorpe /* 1994c2261dd7SJason Gunthorpe * Drivers wish to call this before ib_register_driver, so we have to 1995c2261dd7SJason Gunthorpe * setup the port data early. 1996c2261dd7SJason Gunthorpe */ 1997c2261dd7SJason Gunthorpe ret = alloc_port_data(ib_dev); 1998c2261dd7SJason Gunthorpe if (ret) 1999c2261dd7SJason Gunthorpe return ret; 2000c2261dd7SJason Gunthorpe 2001c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 2002c2261dd7SJason Gunthorpe return -EINVAL; 2003c2261dd7SJason Gunthorpe 2004c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 2005c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 2006324e227eSJason Gunthorpe old_ndev = rcu_dereference_protected( 2007324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 2008324e227eSJason Gunthorpe if (old_ndev == ndev) { 2009c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 2010c2261dd7SJason Gunthorpe return 0; 2011c2261dd7SJason Gunthorpe } 2012c2261dd7SJason Gunthorpe 2013c2261dd7SJason Gunthorpe if (ndev) 2014c2261dd7SJason Gunthorpe dev_hold(ndev); 2015324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, ndev); 2016c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 2017c2261dd7SJason Gunthorpe 2018324e227eSJason Gunthorpe add_ndev_hash(pdata); 2019c2261dd7SJason Gunthorpe if (old_ndev) 2020c2261dd7SJason Gunthorpe dev_put(old_ndev); 2021c2261dd7SJason Gunthorpe 2022c2261dd7SJason Gunthorpe return 0; 2023c2261dd7SJason Gunthorpe } 2024c2261dd7SJason Gunthorpe EXPORT_SYMBOL(ib_device_set_netdev); 2025c2261dd7SJason Gunthorpe 2026c2261dd7SJason Gunthorpe static void free_netdevs(struct ib_device *ib_dev) 2027c2261dd7SJason Gunthorpe { 2028c2261dd7SJason Gunthorpe unsigned long flags; 2029c2261dd7SJason Gunthorpe unsigned int port; 2030c2261dd7SJason Gunthorpe 2031c2261dd7SJason Gunthorpe rdma_for_each_port (ib_dev, port) { 2032c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &ib_dev->port_data[port]; 2033324e227eSJason Gunthorpe struct net_device *ndev; 2034c2261dd7SJason Gunthorpe 2035c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 2036324e227eSJason Gunthorpe ndev = rcu_dereference_protected( 2037324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 2038324e227eSJason Gunthorpe if (ndev) { 2039324e227eSJason Gunthorpe spin_lock(&ndev_hash_lock); 2040324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 2041324e227eSJason Gunthorpe spin_unlock(&ndev_hash_lock); 2042324e227eSJason Gunthorpe 2043324e227eSJason Gunthorpe /* 2044324e227eSJason Gunthorpe * If this is the last dev_put there is still a 2045324e227eSJason Gunthorpe * synchronize_rcu before the netdev is kfreed, so we 2046324e227eSJason Gunthorpe * can continue to rely on unlocked pointer 2047324e227eSJason Gunthorpe * comparisons after the put 2048324e227eSJason Gunthorpe */ 2049324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, NULL); 2050324e227eSJason Gunthorpe dev_put(ndev); 2051c2261dd7SJason Gunthorpe } 2052c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 2053c2261dd7SJason Gunthorpe } 2054c2261dd7SJason Gunthorpe } 2055c2261dd7SJason Gunthorpe 2056c2261dd7SJason Gunthorpe struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, 2057c2261dd7SJason Gunthorpe unsigned int port) 2058c2261dd7SJason Gunthorpe { 2059c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 2060c2261dd7SJason Gunthorpe struct net_device *res; 2061c2261dd7SJason Gunthorpe 2062c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 2063c2261dd7SJason Gunthorpe return NULL; 2064c2261dd7SJason Gunthorpe 2065c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 2066c2261dd7SJason Gunthorpe 2067c2261dd7SJason Gunthorpe /* 2068c2261dd7SJason Gunthorpe * New drivers should use ib_device_set_netdev() not the legacy 2069c2261dd7SJason Gunthorpe * get_netdev(). 2070c2261dd7SJason Gunthorpe */ 2071c2261dd7SJason Gunthorpe if (ib_dev->ops.get_netdev) 2072c2261dd7SJason Gunthorpe res = ib_dev->ops.get_netdev(ib_dev, port); 2073c2261dd7SJason Gunthorpe else { 2074c2261dd7SJason Gunthorpe spin_lock(&pdata->netdev_lock); 2075324e227eSJason Gunthorpe res = rcu_dereference_protected( 2076324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 2077c2261dd7SJason Gunthorpe if (res) 2078c2261dd7SJason Gunthorpe dev_hold(res); 2079c2261dd7SJason Gunthorpe spin_unlock(&pdata->netdev_lock); 2080c2261dd7SJason Gunthorpe } 2081c2261dd7SJason Gunthorpe 2082c2261dd7SJason Gunthorpe /* 2083c2261dd7SJason Gunthorpe * If we are starting to unregister expedite things by preventing 2084c2261dd7SJason Gunthorpe * propagation of an unregistering netdev. 2085c2261dd7SJason Gunthorpe */ 2086c2261dd7SJason Gunthorpe if (res && res->reg_state != NETREG_REGISTERED) { 2087c2261dd7SJason Gunthorpe dev_put(res); 2088c2261dd7SJason Gunthorpe return NULL; 2089c2261dd7SJason Gunthorpe } 2090c2261dd7SJason Gunthorpe 2091c2261dd7SJason Gunthorpe return res; 2092c2261dd7SJason Gunthorpe } 2093c2261dd7SJason Gunthorpe 2094c2261dd7SJason Gunthorpe /** 2095324e227eSJason Gunthorpe * ib_device_get_by_netdev - Find an IB device associated with a netdev 2096324e227eSJason Gunthorpe * @ndev: netdev to locate 2097324e227eSJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 2098324e227eSJason Gunthorpe * 2099324e227eSJason Gunthorpe * Find and hold an ib_device that is associated with a netdev via 2100324e227eSJason Gunthorpe * ib_device_set_netdev(). The caller must call ib_device_put() on the 2101324e227eSJason Gunthorpe * returned pointer. 2102324e227eSJason Gunthorpe */ 2103324e227eSJason Gunthorpe struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 2104324e227eSJason Gunthorpe enum rdma_driver_id driver_id) 2105324e227eSJason Gunthorpe { 2106324e227eSJason Gunthorpe struct ib_device *res = NULL; 2107324e227eSJason Gunthorpe struct ib_port_data *cur; 2108324e227eSJason Gunthorpe 2109324e227eSJason Gunthorpe rcu_read_lock(); 2110324e227eSJason Gunthorpe hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, 2111324e227eSJason Gunthorpe (uintptr_t)ndev) { 2112324e227eSJason Gunthorpe if (rcu_access_pointer(cur->netdev) == ndev && 2113324e227eSJason Gunthorpe (driver_id == RDMA_DRIVER_UNKNOWN || 2114b9560a41SJason Gunthorpe cur->ib_dev->ops.driver_id == driver_id) && 2115324e227eSJason Gunthorpe ib_device_try_get(cur->ib_dev)) { 2116324e227eSJason Gunthorpe res = cur->ib_dev; 2117324e227eSJason Gunthorpe break; 2118324e227eSJason Gunthorpe } 2119324e227eSJason Gunthorpe } 2120324e227eSJason Gunthorpe rcu_read_unlock(); 2121324e227eSJason Gunthorpe 2122324e227eSJason Gunthorpe return res; 2123324e227eSJason Gunthorpe } 2124324e227eSJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_netdev); 2125324e227eSJason Gunthorpe 2126324e227eSJason Gunthorpe /** 212703db3a2dSMatan Barak * ib_enum_roce_netdev - enumerate all RoCE ports 212803db3a2dSMatan Barak * @ib_dev : IB device we want to query 212903db3a2dSMatan Barak * @filter: Should we call the callback? 213003db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 213103db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 213203db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 213303db3a2dSMatan Barak * 213403db3a2dSMatan Barak * Enumerates all of the physical RoCE ports of ib_dev 213503db3a2dSMatan Barak * which are related to netdevice and calls callback() on each 213603db3a2dSMatan Barak * device for which filter() function returns non zero. 213703db3a2dSMatan Barak */ 213803db3a2dSMatan Barak void ib_enum_roce_netdev(struct ib_device *ib_dev, 213903db3a2dSMatan Barak roce_netdev_filter filter, 214003db3a2dSMatan Barak void *filter_cookie, 214103db3a2dSMatan Barak roce_netdev_callback cb, 214203db3a2dSMatan Barak void *cookie) 214303db3a2dSMatan Barak { 2144ea1075edSJason Gunthorpe unsigned int port; 214503db3a2dSMatan Barak 2146ea1075edSJason Gunthorpe rdma_for_each_port (ib_dev, port) 214703db3a2dSMatan Barak if (rdma_protocol_roce(ib_dev, port)) { 2148c2261dd7SJason Gunthorpe struct net_device *idev = 2149c2261dd7SJason Gunthorpe ib_device_get_netdev(ib_dev, port); 215003db3a2dSMatan Barak 215103db3a2dSMatan Barak if (filter(ib_dev, port, idev, filter_cookie)) 215203db3a2dSMatan Barak cb(ib_dev, port, idev, cookie); 215303db3a2dSMatan Barak 215403db3a2dSMatan Barak if (idev) 215503db3a2dSMatan Barak dev_put(idev); 215603db3a2dSMatan Barak } 215703db3a2dSMatan Barak } 215803db3a2dSMatan Barak 215903db3a2dSMatan Barak /** 216003db3a2dSMatan Barak * ib_enum_all_roce_netdevs - enumerate all RoCE devices 216103db3a2dSMatan Barak * @filter: Should we call the callback? 216203db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 216303db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 216403db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 216503db3a2dSMatan Barak * 216603db3a2dSMatan Barak * Enumerates all RoCE devices' physical ports which are related 216703db3a2dSMatan Barak * to netdevices and calls callback() on each device for which 216803db3a2dSMatan Barak * filter() function returns non zero. 216903db3a2dSMatan Barak */ 217003db3a2dSMatan Barak void ib_enum_all_roce_netdevs(roce_netdev_filter filter, 217103db3a2dSMatan Barak void *filter_cookie, 217203db3a2dSMatan Barak roce_netdev_callback cb, 217303db3a2dSMatan Barak void *cookie) 217403db3a2dSMatan Barak { 217503db3a2dSMatan Barak struct ib_device *dev; 21760df91bb6SJason Gunthorpe unsigned long index; 217703db3a2dSMatan Barak 2178921eab11SJason Gunthorpe down_read(&devices_rwsem); 21790df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) 218003db3a2dSMatan Barak ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); 2181921eab11SJason Gunthorpe up_read(&devices_rwsem); 218203db3a2dSMatan Barak } 218303db3a2dSMatan Barak 218403db3a2dSMatan Barak /** 21858030c835SLeon Romanovsky * ib_enum_all_devs - enumerate all ib_devices 21868030c835SLeon Romanovsky * @cb: Callback to call for each found ib_device 21878030c835SLeon Romanovsky * 21888030c835SLeon Romanovsky * Enumerates all ib_devices and calls callback() on each device. 21898030c835SLeon Romanovsky */ 21908030c835SLeon Romanovsky int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, 21918030c835SLeon Romanovsky struct netlink_callback *cb) 21928030c835SLeon Romanovsky { 21930df91bb6SJason Gunthorpe unsigned long index; 21948030c835SLeon Romanovsky struct ib_device *dev; 21958030c835SLeon Romanovsky unsigned int idx = 0; 21968030c835SLeon Romanovsky int ret = 0; 21978030c835SLeon Romanovsky 2198921eab11SJason Gunthorpe down_read(&devices_rwsem); 21990df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 220037eeab55SParav Pandit if (!rdma_dev_access_netns(dev, sock_net(skb->sk))) 220137eeab55SParav Pandit continue; 220237eeab55SParav Pandit 22038030c835SLeon Romanovsky ret = nldev_cb(dev, skb, cb, idx); 22048030c835SLeon Romanovsky if (ret) 22058030c835SLeon Romanovsky break; 22068030c835SLeon Romanovsky idx++; 22078030c835SLeon Romanovsky } 2208921eab11SJason Gunthorpe up_read(&devices_rwsem); 22098030c835SLeon Romanovsky return ret; 22108030c835SLeon Romanovsky } 22118030c835SLeon Romanovsky 22128030c835SLeon Romanovsky /** 22131da177e4SLinus Torvalds * ib_query_pkey - Get P_Key table entry 22141da177e4SLinus Torvalds * @device:Device to query 22151da177e4SLinus Torvalds * @port_num:Port number to query 22161da177e4SLinus Torvalds * @index:P_Key table index to query 22171da177e4SLinus Torvalds * @pkey:Returned P_Key 22181da177e4SLinus Torvalds * 22191da177e4SLinus Torvalds * ib_query_pkey() fetches the specified P_Key table entry. 22201da177e4SLinus Torvalds */ 22211da177e4SLinus Torvalds int ib_query_pkey(struct ib_device *device, 22221da177e4SLinus Torvalds u8 port_num, u16 index, u16 *pkey) 22231da177e4SLinus Torvalds { 22249af3f5cfSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 22259af3f5cfSYuval Shaia return -EINVAL; 22269af3f5cfSYuval Shaia 22273023a1e9SKamal Heib return device->ops.query_pkey(device, port_num, index, pkey); 22281da177e4SLinus Torvalds } 22291da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_pkey); 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds /** 22321da177e4SLinus Torvalds * ib_modify_device - Change IB device attributes 22331da177e4SLinus Torvalds * @device:Device to modify 22341da177e4SLinus Torvalds * @device_modify_mask:Mask of attributes to change 22351da177e4SLinus Torvalds * @device_modify:New attribute values 22361da177e4SLinus Torvalds * 22371da177e4SLinus Torvalds * ib_modify_device() changes a device's attributes as specified by 22381da177e4SLinus Torvalds * the @device_modify_mask and @device_modify structure. 22391da177e4SLinus Torvalds */ 22401da177e4SLinus Torvalds int ib_modify_device(struct ib_device *device, 22411da177e4SLinus Torvalds int device_modify_mask, 22421da177e4SLinus Torvalds struct ib_device_modify *device_modify) 22431da177e4SLinus Torvalds { 22443023a1e9SKamal Heib if (!device->ops.modify_device) 224510e1b54bSBart Van Assche return -ENOSYS; 224610e1b54bSBart Van Assche 22473023a1e9SKamal Heib return device->ops.modify_device(device, device_modify_mask, 22481da177e4SLinus Torvalds device_modify); 22491da177e4SLinus Torvalds } 22501da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_device); 22511da177e4SLinus Torvalds 22521da177e4SLinus Torvalds /** 22531da177e4SLinus Torvalds * ib_modify_port - Modifies the attributes for the specified port. 22541da177e4SLinus Torvalds * @device: The device to modify. 22551da177e4SLinus Torvalds * @port_num: The number of the port to modify. 22561da177e4SLinus Torvalds * @port_modify_mask: Mask used to specify which attributes of the port 22571da177e4SLinus Torvalds * to change. 22581da177e4SLinus Torvalds * @port_modify: New attribute values for the port. 22591da177e4SLinus Torvalds * 22601da177e4SLinus Torvalds * ib_modify_port() changes a port's attributes as specified by the 22611da177e4SLinus Torvalds * @port_modify_mask and @port_modify structure. 22621da177e4SLinus Torvalds */ 22631da177e4SLinus Torvalds int ib_modify_port(struct ib_device *device, 22641da177e4SLinus Torvalds u8 port_num, int port_modify_mask, 22651da177e4SLinus Torvalds struct ib_port_modify *port_modify) 22661da177e4SLinus Torvalds { 226761e0962dSSelvin Xavier int rc; 226810e1b54bSBart Van Assche 226924dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 2270116c0074SRoland Dreier return -EINVAL; 2271116c0074SRoland Dreier 22723023a1e9SKamal Heib if (device->ops.modify_port) 22733023a1e9SKamal Heib rc = device->ops.modify_port(device, port_num, 22743023a1e9SKamal Heib port_modify_mask, 22751da177e4SLinus Torvalds port_modify); 227661e0962dSSelvin Xavier else 227761e0962dSSelvin Xavier rc = rdma_protocol_roce(device, port_num) ? 0 : -ENOSYS; 227861e0962dSSelvin Xavier return rc; 22791da177e4SLinus Torvalds } 22801da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_port); 22811da177e4SLinus Torvalds 22825eb620c8SYosef Etigin /** 22835eb620c8SYosef Etigin * ib_find_gid - Returns the port number and GID table index where 2284dbb12562SParav Pandit * a specified GID value occurs. Its searches only for IB link layer. 22855eb620c8SYosef Etigin * @device: The device to query. 22865eb620c8SYosef Etigin * @gid: The GID value to search for. 22875eb620c8SYosef Etigin * @port_num: The port number of the device where the GID value was found. 22885eb620c8SYosef Etigin * @index: The index into the GID table where the GID was found. This 22895eb620c8SYosef Etigin * parameter may be NULL. 22905eb620c8SYosef Etigin */ 22915eb620c8SYosef Etigin int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2292b26c4a11SParav Pandit u8 *port_num, u16 *index) 22935eb620c8SYosef Etigin { 22945eb620c8SYosef Etigin union ib_gid tmp_gid; 2295ea1075edSJason Gunthorpe unsigned int port; 2296ea1075edSJason Gunthorpe int ret, i; 22975eb620c8SYosef Etigin 2298ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 229922d24f75SParav Pandit if (!rdma_protocol_ib(device, port)) 2300b39ffa1dSMatan Barak continue; 2301b39ffa1dSMatan Barak 23028ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; 23038ceb1357SJason Gunthorpe ++i) { 23041dfce294SParav Pandit ret = rdma_query_gid(device, port, i, &tmp_gid); 23055eb620c8SYosef Etigin if (ret) 23065eb620c8SYosef Etigin return ret; 23075eb620c8SYosef Etigin if (!memcmp(&tmp_gid, gid, sizeof *gid)) { 23085eb620c8SYosef Etigin *port_num = port; 23095eb620c8SYosef Etigin if (index) 23105eb620c8SYosef Etigin *index = i; 23115eb620c8SYosef Etigin return 0; 23125eb620c8SYosef Etigin } 23135eb620c8SYosef Etigin } 23145eb620c8SYosef Etigin } 23155eb620c8SYosef Etigin 23165eb620c8SYosef Etigin return -ENOENT; 23175eb620c8SYosef Etigin } 23185eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_gid); 23195eb620c8SYosef Etigin 23205eb620c8SYosef Etigin /** 23215eb620c8SYosef Etigin * ib_find_pkey - Returns the PKey table index where a specified 23225eb620c8SYosef Etigin * PKey value occurs. 23235eb620c8SYosef Etigin * @device: The device to query. 23245eb620c8SYosef Etigin * @port_num: The port number of the device to search for the PKey. 23255eb620c8SYosef Etigin * @pkey: The PKey value to search for. 23265eb620c8SYosef Etigin * @index: The index into the PKey table where the PKey was found. 23275eb620c8SYosef Etigin */ 23285eb620c8SYosef Etigin int ib_find_pkey(struct ib_device *device, 23295eb620c8SYosef Etigin u8 port_num, u16 pkey, u16 *index) 23305eb620c8SYosef Etigin { 23315eb620c8SYosef Etigin int ret, i; 23325eb620c8SYosef Etigin u16 tmp_pkey; 2333ff7166c4SJack Morgenstein int partial_ix = -1; 23345eb620c8SYosef Etigin 23358ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; 23368ceb1357SJason Gunthorpe ++i) { 23375eb620c8SYosef Etigin ret = ib_query_pkey(device, port_num, i, &tmp_pkey); 23385eb620c8SYosef Etigin if (ret) 23395eb620c8SYosef Etigin return ret; 234036026eccSMoni Shoua if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { 2341ff7166c4SJack Morgenstein /* if there is full-member pkey take it.*/ 2342ff7166c4SJack Morgenstein if (tmp_pkey & 0x8000) { 23435eb620c8SYosef Etigin *index = i; 23445eb620c8SYosef Etigin return 0; 23455eb620c8SYosef Etigin } 2346ff7166c4SJack Morgenstein if (partial_ix < 0) 2347ff7166c4SJack Morgenstein partial_ix = i; 2348ff7166c4SJack Morgenstein } 23495eb620c8SYosef Etigin } 23505eb620c8SYosef Etigin 2351ff7166c4SJack Morgenstein /*no full-member, if exists take the limited*/ 2352ff7166c4SJack Morgenstein if (partial_ix >= 0) { 2353ff7166c4SJack Morgenstein *index = partial_ix; 2354ff7166c4SJack Morgenstein return 0; 2355ff7166c4SJack Morgenstein } 23565eb620c8SYosef Etigin return -ENOENT; 23575eb620c8SYosef Etigin } 23585eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_pkey); 23595eb620c8SYosef Etigin 23609268f72dSYotam Kenneth /** 23619268f72dSYotam Kenneth * ib_get_net_dev_by_params() - Return the appropriate net_dev 23629268f72dSYotam Kenneth * for a received CM request 23639268f72dSYotam Kenneth * @dev: An RDMA device on which the request has been received. 23649268f72dSYotam Kenneth * @port: Port number on the RDMA device. 23659268f72dSYotam Kenneth * @pkey: The Pkey the request came on. 23669268f72dSYotam Kenneth * @gid: A GID that the net_dev uses to communicate. 23679268f72dSYotam Kenneth * @addr: Contains the IP address that the request specified as its 23689268f72dSYotam Kenneth * destination. 2369921eab11SJason Gunthorpe * 23709268f72dSYotam Kenneth */ 23719268f72dSYotam Kenneth struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, 23729268f72dSYotam Kenneth u8 port, 23739268f72dSYotam Kenneth u16 pkey, 23749268f72dSYotam Kenneth const union ib_gid *gid, 23759268f72dSYotam Kenneth const struct sockaddr *addr) 23769268f72dSYotam Kenneth { 23779268f72dSYotam Kenneth struct net_device *net_dev = NULL; 23780df91bb6SJason Gunthorpe unsigned long index; 23790df91bb6SJason Gunthorpe void *client_data; 23809268f72dSYotam Kenneth 23819268f72dSYotam Kenneth if (!rdma_protocol_ib(dev, port)) 23829268f72dSYotam Kenneth return NULL; 23839268f72dSYotam Kenneth 2384921eab11SJason Gunthorpe /* 2385921eab11SJason Gunthorpe * Holding the read side guarantees that the client will not become 2386921eab11SJason Gunthorpe * unregistered while we are calling get_net_dev_by_params() 2387921eab11SJason Gunthorpe */ 2388921eab11SJason Gunthorpe down_read(&dev->client_data_rwsem); 23890df91bb6SJason Gunthorpe xan_for_each_marked (&dev->client_data, index, client_data, 23900df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED) { 23910df91bb6SJason Gunthorpe struct ib_client *client = xa_load(&clients, index); 23929268f72dSYotam Kenneth 23930df91bb6SJason Gunthorpe if (!client || !client->get_net_dev_by_params) 23949268f72dSYotam Kenneth continue; 23959268f72dSYotam Kenneth 23960df91bb6SJason Gunthorpe net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, 23970df91bb6SJason Gunthorpe addr, client_data); 23989268f72dSYotam Kenneth if (net_dev) 23999268f72dSYotam Kenneth break; 24009268f72dSYotam Kenneth } 2401921eab11SJason Gunthorpe up_read(&dev->client_data_rwsem); 24029268f72dSYotam Kenneth 24039268f72dSYotam Kenneth return net_dev; 24049268f72dSYotam Kenneth } 24059268f72dSYotam Kenneth EXPORT_SYMBOL(ib_get_net_dev_by_params); 24069268f72dSYotam Kenneth 2407521ed0d9SKamal Heib void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) 2408521ed0d9SKamal Heib { 24093023a1e9SKamal Heib struct ib_device_ops *dev_ops = &dev->ops; 2410521ed0d9SKamal Heib #define SET_DEVICE_OP(ptr, name) \ 2411521ed0d9SKamal Heib do { \ 2412521ed0d9SKamal Heib if (ops->name) \ 2413521ed0d9SKamal Heib if (!((ptr)->name)) \ 2414521ed0d9SKamal Heib (ptr)->name = ops->name; \ 2415521ed0d9SKamal Heib } while (0) 2416521ed0d9SKamal Heib 241730471d4bSLeon Romanovsky #define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) 241830471d4bSLeon Romanovsky 2419b9560a41SJason Gunthorpe if (ops->driver_id != RDMA_DRIVER_UNKNOWN) { 2420b9560a41SJason Gunthorpe WARN_ON(dev_ops->driver_id != RDMA_DRIVER_UNKNOWN && 2421b9560a41SJason Gunthorpe dev_ops->driver_id != ops->driver_id); 2422b9560a41SJason Gunthorpe dev_ops->driver_id = ops->driver_id; 2423b9560a41SJason Gunthorpe } 24247a154142SJason Gunthorpe if (ops->owner) { 24257a154142SJason Gunthorpe WARN_ON(dev_ops->owner && dev_ops->owner != ops->owner); 24267a154142SJason Gunthorpe dev_ops->owner = ops->owner; 24277a154142SJason Gunthorpe } 242872c6ec18SJason Gunthorpe if (ops->uverbs_abi_ver) 242972c6ec18SJason Gunthorpe dev_ops->uverbs_abi_ver = ops->uverbs_abi_ver; 2430b9560a41SJason Gunthorpe 24318f71bb00SJason Gunthorpe dev_ops->uverbs_no_driver_id_binding |= 24328f71bb00SJason Gunthorpe ops->uverbs_no_driver_id_binding; 24338f71bb00SJason Gunthorpe 24343023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, add_gid); 24352f1927b0SMoni Shoua SET_DEVICE_OP(dev_ops, advise_mr); 24363023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_dm); 24373023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_fmr); 24383023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_hw_stats); 24393023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mr); 244026bc7eaeSIsrael Rukshin SET_DEVICE_OP(dev_ops, alloc_mr_integrity); 24413023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mw); 24423023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_pd); 24433023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_rdma_netdev); 24443023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_ucontext); 24453023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_xrcd); 24463023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, attach_mcast); 24473023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, check_mr_status); 24483023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_ah); 24493023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_counters); 24503023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_cq); 24513023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow); 24523023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow_action_esp); 24533023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_qp); 24543023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_rwq_ind_table); 24553023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_srq); 24563023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_wq); 24573023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_dm); 2458d0899892SJason Gunthorpe SET_DEVICE_OP(dev_ops, dealloc_driver); 24593023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_fmr); 24603023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_mw); 24613023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_pd); 24623023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_ucontext); 24633023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_xrcd); 24643023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, del_gid); 24653023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dereg_mr); 24663023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_ah); 24673023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_counters); 24683023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_cq); 24693023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow); 24703023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow_action); 24713023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_qp); 24723023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_rwq_ind_table); 24733023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_srq); 24743023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_wq); 24753023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, detach_mcast); 24763023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, disassociate_ucontext); 24773023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_rq); 24783023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_sq); 2479ca22354bSJason Gunthorpe SET_DEVICE_OP(dev_ops, enable_driver); 248002da3750SLeon Romanovsky SET_DEVICE_OP(dev_ops, fill_res_entry); 24813023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dev_fw_str); 24823023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dma_mr); 24833023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_hw_stats); 24843023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_link_layer); 24853023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_netdev); 24863023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_port_immutable); 24873023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vector_affinity); 24883023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_config); 24893023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_stats); 2490ea4baf7fSParav Pandit SET_DEVICE_OP(dev_ops, init_port); 2491dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_accept); 2492dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_add_ref); 2493dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_connect); 2494dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_create_listen); 2495dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_destroy_listen); 2496dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_get_qp); 2497dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_reject); 2498dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_rem_ref); 24993023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, map_mr_sg); 25003023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, map_phys_fmr); 25013023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, mmap); 25023023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_ah); 25033023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_cq); 25043023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_device); 25053023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_flow_action_esp); 25063023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_port); 25073023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_qp); 25083023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_srq); 25093023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_wq); 25103023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, peek_cq); 25113023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, poll_cq); 25123023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_recv); 25133023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_send); 25143023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_srq_recv); 25153023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, process_mad); 25163023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_ah); 25173023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_device); 25183023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_gid); 25193023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_pkey); 25203023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_port); 25213023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_qp); 25223023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_srq); 25233023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rdma_netdev_get_params); 25243023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, read_counters); 25253023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_dm_mr); 25263023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_user_mr); 25273023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_ncomp_notif); 25283023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_notify_cq); 25293023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rereg_user_mr); 25303023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, resize_cq); 25313023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_guid); 25323023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_link_state); 25333023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, unmap_fmr); 253421a428a0SLeon Romanovsky 2535d3456914SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_ah); 2536e39afe3dSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_cq); 253721a428a0SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_pd); 253868e326deSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_srq); 2539a2a074efSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_ucontext); 2540521ed0d9SKamal Heib } 2541521ed0d9SKamal Heib EXPORT_SYMBOL(ib_set_device_ops); 2542521ed0d9SKamal Heib 2543d0e312feSLeon Romanovsky static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = { 2544735c631aSMark Bloch [RDMA_NL_LS_OP_RESOLVE] = { 2545647c75acSLeon Romanovsky .doit = ib_nl_handle_resolve_resp, 2546e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2547e3a2b93dSLeon Romanovsky }, 2548735c631aSMark Bloch [RDMA_NL_LS_OP_SET_TIMEOUT] = { 2549647c75acSLeon Romanovsky .doit = ib_nl_handle_set_timeout, 2550e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2551e3a2b93dSLeon Romanovsky }, 2552ae43f828SMark Bloch [RDMA_NL_LS_OP_IP_RESOLVE] = { 2553647c75acSLeon Romanovsky .doit = ib_nl_handle_ip_res_resp, 2554e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2555e3a2b93dSLeon Romanovsky }, 2556735c631aSMark Bloch }; 2557735c631aSMark Bloch 25581da177e4SLinus Torvalds static int __init ib_core_init(void) 25591da177e4SLinus Torvalds { 25601da177e4SLinus Torvalds int ret; 25611da177e4SLinus Torvalds 2562f0626710STejun Heo ib_wq = alloc_workqueue("infiniband", 0, 0); 2563f0626710STejun Heo if (!ib_wq) 2564f0626710STejun Heo return -ENOMEM; 2565f0626710STejun Heo 256614d3a3b2SChristoph Hellwig ib_comp_wq = alloc_workqueue("ib-comp-wq", 2567b7363e67SSagi Grimberg WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0); 256814d3a3b2SChristoph Hellwig if (!ib_comp_wq) { 256914d3a3b2SChristoph Hellwig ret = -ENOMEM; 257014d3a3b2SChristoph Hellwig goto err; 257114d3a3b2SChristoph Hellwig } 257214d3a3b2SChristoph Hellwig 2573f794809aSJack Morgenstein ib_comp_unbound_wq = 2574f794809aSJack Morgenstein alloc_workqueue("ib-comp-unb-wq", 2575f794809aSJack Morgenstein WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM | 2576f794809aSJack Morgenstein WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE); 2577f794809aSJack Morgenstein if (!ib_comp_unbound_wq) { 2578f794809aSJack Morgenstein ret = -ENOMEM; 2579f794809aSJack Morgenstein goto err_comp; 2580f794809aSJack Morgenstein } 2581f794809aSJack Morgenstein 258255aeed06SJason Gunthorpe ret = class_register(&ib_class); 2583fd75c789SNir Muchtar if (ret) { 2584aba25a3eSParav Pandit pr_warn("Couldn't create InfiniBand device class\n"); 2585f794809aSJack Morgenstein goto err_comp_unbound; 2586fd75c789SNir Muchtar } 25871da177e4SLinus Torvalds 2588c9901724SLeon Romanovsky ret = rdma_nl_init(); 25891da177e4SLinus Torvalds if (ret) { 2590c9901724SLeon Romanovsky pr_warn("Couldn't init IB netlink interface: err %d\n", ret); 2591fd75c789SNir Muchtar goto err_sysfs; 25921da177e4SLinus Torvalds } 25931da177e4SLinus Torvalds 2594e3f20f02SLeon Romanovsky ret = addr_init(); 2595e3f20f02SLeon Romanovsky if (ret) { 2596e3f20f02SLeon Romanovsky pr_warn("Could't init IB address resolution\n"); 2597e3f20f02SLeon Romanovsky goto err_ibnl; 2598e3f20f02SLeon Romanovsky } 2599e3f20f02SLeon Romanovsky 26004c2cb422SMark Bloch ret = ib_mad_init(); 26014c2cb422SMark Bloch if (ret) { 26024c2cb422SMark Bloch pr_warn("Couldn't init IB MAD\n"); 26034c2cb422SMark Bloch goto err_addr; 26044c2cb422SMark Bloch } 26054c2cb422SMark Bloch 2606c2e49c92SMark Bloch ret = ib_sa_init(); 2607c2e49c92SMark Bloch if (ret) { 2608c2e49c92SMark Bloch pr_warn("Couldn't init SA\n"); 2609c2e49c92SMark Bloch goto err_mad; 2610c2e49c92SMark Bloch } 2611c2e49c92SMark Bloch 26128f408ab6SDaniel Jurgens ret = register_lsm_notifier(&ibdev_lsm_nb); 26138f408ab6SDaniel Jurgens if (ret) { 26148f408ab6SDaniel Jurgens pr_warn("Couldn't register LSM notifier. ret %d\n", ret); 2615c9901724SLeon Romanovsky goto err_sa; 26168f408ab6SDaniel Jurgens } 26178f408ab6SDaniel Jurgens 26184e0f7b90SParav Pandit ret = register_pernet_device(&rdma_dev_net_ops); 26194e0f7b90SParav Pandit if (ret) { 26204e0f7b90SParav Pandit pr_warn("Couldn't init compat dev. ret %d\n", ret); 26214e0f7b90SParav Pandit goto err_compat; 26224e0f7b90SParav Pandit } 26234e0f7b90SParav Pandit 26246c80b41aSLeon Romanovsky nldev_init(); 2625c9901724SLeon Romanovsky rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table); 26265ef8c0c1SJason Gunthorpe roce_gid_mgmt_init(); 2627b2cbae2cSRoland Dreier 2628fd75c789SNir Muchtar return 0; 2629fd75c789SNir Muchtar 26304e0f7b90SParav Pandit err_compat: 26314e0f7b90SParav Pandit unregister_lsm_notifier(&ibdev_lsm_nb); 2632735c631aSMark Bloch err_sa: 2633735c631aSMark Bloch ib_sa_cleanup(); 2634c2e49c92SMark Bloch err_mad: 2635c2e49c92SMark Bloch ib_mad_cleanup(); 26364c2cb422SMark Bloch err_addr: 26374c2cb422SMark Bloch addr_cleanup(); 2638e3f20f02SLeon Romanovsky err_ibnl: 2639c9901724SLeon Romanovsky rdma_nl_exit(); 2640fd75c789SNir Muchtar err_sysfs: 264155aeed06SJason Gunthorpe class_unregister(&ib_class); 2642f794809aSJack Morgenstein err_comp_unbound: 2643f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 264414d3a3b2SChristoph Hellwig err_comp: 264514d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2646fd75c789SNir Muchtar err: 2647fd75c789SNir Muchtar destroy_workqueue(ib_wq); 26481da177e4SLinus Torvalds return ret; 26491da177e4SLinus Torvalds } 26501da177e4SLinus Torvalds 26511da177e4SLinus Torvalds static void __exit ib_core_cleanup(void) 26521da177e4SLinus Torvalds { 26535ef8c0c1SJason Gunthorpe roce_gid_mgmt_cleanup(); 26546c80b41aSLeon Romanovsky nldev_exit(); 2655c9901724SLeon Romanovsky rdma_nl_unregister(RDMA_NL_LS); 26564e0f7b90SParav Pandit unregister_pernet_device(&rdma_dev_net_ops); 2657c9901724SLeon Romanovsky unregister_lsm_notifier(&ibdev_lsm_nb); 2658c2e49c92SMark Bloch ib_sa_cleanup(); 26594c2cb422SMark Bloch ib_mad_cleanup(); 2660e3f20f02SLeon Romanovsky addr_cleanup(); 2661c9901724SLeon Romanovsky rdma_nl_exit(); 266255aeed06SJason Gunthorpe class_unregister(&ib_class); 2663f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 266414d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2665f7c6a7b5SRoland Dreier /* Make sure that any pending umem accounting work is done. */ 2666f0626710STejun Heo destroy_workqueue(ib_wq); 2667d0899892SJason Gunthorpe flush_workqueue(system_unbound_wq); 2668e59178d8SJason Gunthorpe WARN_ON(!xa_empty(&clients)); 26690df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&devices)); 26701da177e4SLinus Torvalds } 26711da177e4SLinus Torvalds 2672e3bf14bdSJason Gunthorpe MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); 2673e3bf14bdSJason Gunthorpe 267462dfa795SParav Pandit /* ib core relies on netdev stack to first register net_ns_type_operations 267562dfa795SParav Pandit * ns kobject type before ib_core initialization. 267662dfa795SParav Pandit */ 267762dfa795SParav Pandit fs_initcall(ib_core_init); 26781da177e4SLinus Torvalds module_exit(ib_core_cleanup); 2679