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> 428f408ab6SDaniel Jurgens #include <linux/security.h> 438f408ab6SDaniel Jurgens #include <linux/notifier.h> 44324e227eSJason Gunthorpe #include <linux/hashtable.h> 45b2cbae2cSRoland Dreier #include <rdma/rdma_netlink.h> 4603db3a2dSMatan Barak #include <rdma/ib_addr.h> 4703db3a2dSMatan Barak #include <rdma/ib_cache.h> 48413d3347SMark Zhang #include <rdma/rdma_counter.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 969cd58817SJason Gunthorpe static u32 highest_client_id; 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 101621e55ffSJason Gunthorpe static void ib_client_put(struct ib_client *client) 102621e55ffSJason Gunthorpe { 103621e55ffSJason Gunthorpe if (refcount_dec_and_test(&client->uses)) 104621e55ffSJason Gunthorpe complete(&client->uses_zero); 105621e55ffSJason Gunthorpe } 106621e55ffSJason Gunthorpe 1071da177e4SLinus Torvalds /* 1080df91bb6SJason Gunthorpe * If client_data is registered then the corresponding client must also still 1090df91bb6SJason Gunthorpe * be registered. 1100df91bb6SJason Gunthorpe */ 1110df91bb6SJason Gunthorpe #define CLIENT_DATA_REGISTERED XA_MARK_1 1124e0f7b90SParav Pandit 1131d2fedd8SParav Pandit unsigned int rdma_dev_net_id; 1144e0f7b90SParav Pandit 1154e0f7b90SParav Pandit /* 1164e0f7b90SParav Pandit * A list of net namespaces is maintained in an xarray. This is necessary 1174e0f7b90SParav Pandit * because we can't get the locking right using the existing net ns list. We 1184e0f7b90SParav Pandit * would require a init_net callback after the list is updated. 1194e0f7b90SParav Pandit */ 1204e0f7b90SParav Pandit static DEFINE_XARRAY_FLAGS(rdma_nets, XA_FLAGS_ALLOC); 1214e0f7b90SParav Pandit /* 1224e0f7b90SParav Pandit * rwsem to protect accessing the rdma_nets xarray entries. 1234e0f7b90SParav Pandit */ 1244e0f7b90SParav Pandit static DECLARE_RWSEM(rdma_nets_rwsem); 1254e0f7b90SParav Pandit 126cb7e0e13SParav Pandit bool ib_devices_shared_netns = true; 127a56bc45bSParav Pandit module_param_named(netns_mode, ib_devices_shared_netns, bool, 0444); 128a56bc45bSParav Pandit MODULE_PARM_DESC(netns_mode, 129a56bc45bSParav Pandit "Share device among net namespaces; default=1 (shared)"); 13041c61401SParav Pandit /** 131d6537c1aSrd.dunlab@gmail.com * rdma_dev_access_netns() - Return whether an rdma device can be accessed 13241c61401SParav Pandit * from a specified net namespace or not. 133d6537c1aSrd.dunlab@gmail.com * @dev: Pointer to rdma device which needs to be checked 13441c61401SParav Pandit * @net: Pointer to net namesapce for which access to be checked 13541c61401SParav Pandit * 136d6537c1aSrd.dunlab@gmail.com * When the rdma device is in shared mode, it ignores the net namespace. 137d6537c1aSrd.dunlab@gmail.com * When the rdma device is exclusive to a net namespace, rdma device net 138d6537c1aSrd.dunlab@gmail.com * namespace is checked against the specified one. 13941c61401SParav Pandit */ 14041c61401SParav Pandit bool rdma_dev_access_netns(const struct ib_device *dev, const struct net *net) 14141c61401SParav Pandit { 14241c61401SParav Pandit return (ib_devices_shared_netns || 14341c61401SParav Pandit net_eq(read_pnet(&dev->coredev.rdma_net), net)); 14441c61401SParav Pandit } 14541c61401SParav Pandit EXPORT_SYMBOL(rdma_dev_access_netns); 14641c61401SParav Pandit 1470df91bb6SJason Gunthorpe /* 1480df91bb6SJason Gunthorpe * xarray has this behavior where it won't iterate over NULL values stored in 1490df91bb6SJason Gunthorpe * allocated arrays. So we need our own iterator to see all values stored in 1500df91bb6SJason Gunthorpe * the array. This does the same thing as xa_for_each except that it also 1510df91bb6SJason Gunthorpe * returns NULL valued entries if the array is allocating. Simplified to only 1520df91bb6SJason Gunthorpe * work on simple xarrays. 1530df91bb6SJason Gunthorpe */ 1540df91bb6SJason Gunthorpe static void *xan_find_marked(struct xarray *xa, unsigned long *indexp, 1550df91bb6SJason Gunthorpe xa_mark_t filter) 1560df91bb6SJason Gunthorpe { 1570df91bb6SJason Gunthorpe XA_STATE(xas, xa, *indexp); 1580df91bb6SJason Gunthorpe void *entry; 1590df91bb6SJason Gunthorpe 1600df91bb6SJason Gunthorpe rcu_read_lock(); 1610df91bb6SJason Gunthorpe do { 1620df91bb6SJason Gunthorpe entry = xas_find_marked(&xas, ULONG_MAX, filter); 1630df91bb6SJason Gunthorpe if (xa_is_zero(entry)) 1640df91bb6SJason Gunthorpe break; 1650df91bb6SJason Gunthorpe } while (xas_retry(&xas, entry)); 1660df91bb6SJason Gunthorpe rcu_read_unlock(); 1670df91bb6SJason Gunthorpe 1680df91bb6SJason Gunthorpe if (entry) { 1690df91bb6SJason Gunthorpe *indexp = xas.xa_index; 1700df91bb6SJason Gunthorpe if (xa_is_zero(entry)) 1710df91bb6SJason Gunthorpe return NULL; 1720df91bb6SJason Gunthorpe return entry; 1730df91bb6SJason Gunthorpe } 1740df91bb6SJason Gunthorpe return XA_ERROR(-ENOENT); 1750df91bb6SJason Gunthorpe } 1760df91bb6SJason Gunthorpe #define xan_for_each_marked(xa, index, entry, filter) \ 1770df91bb6SJason Gunthorpe for (index = 0, entry = xan_find_marked(xa, &(index), filter); \ 1780df91bb6SJason Gunthorpe !xa_is_err(entry); \ 1790df91bb6SJason Gunthorpe (index)++, entry = xan_find_marked(xa, &(index), filter)) 1800df91bb6SJason Gunthorpe 181324e227eSJason Gunthorpe /* RCU hash table mapping netdevice pointers to struct ib_port_data */ 182324e227eSJason Gunthorpe static DEFINE_SPINLOCK(ndev_hash_lock); 183324e227eSJason Gunthorpe static DECLARE_HASHTABLE(ndev_hash, 5); 184324e227eSJason Gunthorpe 185c2261dd7SJason Gunthorpe static void free_netdevs(struct ib_device *ib_dev); 186d0899892SJason Gunthorpe static void ib_unregister_work(struct work_struct *work); 187d0899892SJason Gunthorpe static void __ib_unregister_device(struct ib_device *device); 1888f408ab6SDaniel Jurgens static int ib_security_change(struct notifier_block *nb, unsigned long event, 1898f408ab6SDaniel Jurgens void *lsm_data); 1908f408ab6SDaniel Jurgens static void ib_policy_change_task(struct work_struct *work); 1918f408ab6SDaniel Jurgens static DECLARE_WORK(ib_policy_change_work, ib_policy_change_task); 1928f408ab6SDaniel Jurgens 193923abb9dSGal Pressman static void __ibdev_printk(const char *level, const struct ib_device *ibdev, 194923abb9dSGal Pressman struct va_format *vaf) 195923abb9dSGal Pressman { 196923abb9dSGal Pressman if (ibdev && ibdev->dev.parent) 197923abb9dSGal Pressman dev_printk_emit(level[1] - '0', 198923abb9dSGal Pressman ibdev->dev.parent, 199923abb9dSGal Pressman "%s %s %s: %pV", 200923abb9dSGal Pressman dev_driver_string(ibdev->dev.parent), 201923abb9dSGal Pressman dev_name(ibdev->dev.parent), 202923abb9dSGal Pressman dev_name(&ibdev->dev), 203923abb9dSGal Pressman vaf); 204923abb9dSGal Pressman else if (ibdev) 205923abb9dSGal Pressman printk("%s%s: %pV", 206923abb9dSGal Pressman level, dev_name(&ibdev->dev), vaf); 207923abb9dSGal Pressman else 208923abb9dSGal Pressman printk("%s(NULL ib_device): %pV", level, vaf); 209923abb9dSGal Pressman } 210923abb9dSGal Pressman 211923abb9dSGal Pressman void ibdev_printk(const char *level, const struct ib_device *ibdev, 212923abb9dSGal Pressman const char *format, ...) 213923abb9dSGal Pressman { 214923abb9dSGal Pressman struct va_format vaf; 215923abb9dSGal Pressman va_list args; 216923abb9dSGal Pressman 217923abb9dSGal Pressman va_start(args, format); 218923abb9dSGal Pressman 219923abb9dSGal Pressman vaf.fmt = format; 220923abb9dSGal Pressman vaf.va = &args; 221923abb9dSGal Pressman 222923abb9dSGal Pressman __ibdev_printk(level, ibdev, &vaf); 223923abb9dSGal Pressman 224923abb9dSGal Pressman va_end(args); 225923abb9dSGal Pressman } 226923abb9dSGal Pressman EXPORT_SYMBOL(ibdev_printk); 227923abb9dSGal Pressman 228923abb9dSGal Pressman #define define_ibdev_printk_level(func, level) \ 229923abb9dSGal Pressman void func(const struct ib_device *ibdev, const char *fmt, ...) \ 230923abb9dSGal Pressman { \ 231923abb9dSGal Pressman struct va_format vaf; \ 232923abb9dSGal Pressman va_list args; \ 233923abb9dSGal Pressman \ 234923abb9dSGal Pressman va_start(args, fmt); \ 235923abb9dSGal Pressman \ 236923abb9dSGal Pressman vaf.fmt = fmt; \ 237923abb9dSGal Pressman vaf.va = &args; \ 238923abb9dSGal Pressman \ 239923abb9dSGal Pressman __ibdev_printk(level, ibdev, &vaf); \ 240923abb9dSGal Pressman \ 241923abb9dSGal Pressman va_end(args); \ 242923abb9dSGal Pressman } \ 243923abb9dSGal Pressman EXPORT_SYMBOL(func); 244923abb9dSGal Pressman 245923abb9dSGal Pressman define_ibdev_printk_level(ibdev_emerg, KERN_EMERG); 246923abb9dSGal Pressman define_ibdev_printk_level(ibdev_alert, KERN_ALERT); 247923abb9dSGal Pressman define_ibdev_printk_level(ibdev_crit, KERN_CRIT); 248923abb9dSGal Pressman define_ibdev_printk_level(ibdev_err, KERN_ERR); 249923abb9dSGal Pressman define_ibdev_printk_level(ibdev_warn, KERN_WARNING); 250923abb9dSGal Pressman define_ibdev_printk_level(ibdev_notice, KERN_NOTICE); 251923abb9dSGal Pressman define_ibdev_printk_level(ibdev_info, KERN_INFO); 252923abb9dSGal Pressman 2538f408ab6SDaniel Jurgens static struct notifier_block ibdev_lsm_nb = { 2548f408ab6SDaniel Jurgens .notifier_call = ib_security_change, 2558f408ab6SDaniel Jurgens }; 2561da177e4SLinus Torvalds 257decbc7a6SParav Pandit static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, 258decbc7a6SParav Pandit struct net *net); 259decbc7a6SParav Pandit 260324e227eSJason Gunthorpe /* Pointer to the RCU head at the start of the ib_port_data array */ 261324e227eSJason Gunthorpe struct ib_port_data_rcu { 262324e227eSJason Gunthorpe struct rcu_head rcu_head; 263324e227eSJason Gunthorpe struct ib_port_data pdata[]; 264324e227eSJason Gunthorpe }; 265324e227eSJason Gunthorpe 266deee3c7eSKamal Heib static void ib_device_check_mandatory(struct ib_device *device) 2671da177e4SLinus Torvalds { 2683023a1e9SKamal Heib #define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device_ops, x), #x } 2691da177e4SLinus Torvalds static const struct { 2701da177e4SLinus Torvalds size_t offset; 2711da177e4SLinus Torvalds char *name; 2721da177e4SLinus Torvalds } mandatory_table[] = { 2731da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_device), 2741da177e4SLinus Torvalds IB_MANDATORY_FUNC(query_port), 2751da177e4SLinus Torvalds IB_MANDATORY_FUNC(alloc_pd), 2761da177e4SLinus Torvalds IB_MANDATORY_FUNC(dealloc_pd), 2771da177e4SLinus Torvalds IB_MANDATORY_FUNC(create_qp), 2781da177e4SLinus Torvalds IB_MANDATORY_FUNC(modify_qp), 2791da177e4SLinus Torvalds IB_MANDATORY_FUNC(destroy_qp), 2801da177e4SLinus Torvalds IB_MANDATORY_FUNC(post_send), 2811da177e4SLinus Torvalds IB_MANDATORY_FUNC(post_recv), 2821da177e4SLinus Torvalds IB_MANDATORY_FUNC(create_cq), 2831da177e4SLinus Torvalds IB_MANDATORY_FUNC(destroy_cq), 2841da177e4SLinus Torvalds IB_MANDATORY_FUNC(poll_cq), 2851da177e4SLinus Torvalds IB_MANDATORY_FUNC(req_notify_cq), 2861da177e4SLinus Torvalds IB_MANDATORY_FUNC(get_dma_mr), 2877738613eSIra Weiny IB_MANDATORY_FUNC(dereg_mr), 2887738613eSIra Weiny IB_MANDATORY_FUNC(get_port_immutable) 2891da177e4SLinus Torvalds }; 2901da177e4SLinus Torvalds int i; 2911da177e4SLinus Torvalds 2926780c4faSGal Pressman device->kverbs_provider = true; 2939a6b090cSAhmed S. Darwish for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) { 2943023a1e9SKamal Heib if (!*(void **) ((void *) &device->ops + 2953023a1e9SKamal Heib mandatory_table[i].offset)) { 2966780c4faSGal Pressman device->kverbs_provider = false; 2976780c4faSGal Pressman break; 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds } 3001da177e4SLinus Torvalds } 3011da177e4SLinus Torvalds 302f8978bd9SLeon Romanovsky /* 30301b67117SParav Pandit * Caller must perform ib_device_put() to return the device reference count 30401b67117SParav Pandit * when ib_device_get_by_index() returns valid device pointer. 305f8978bd9SLeon Romanovsky */ 30637eeab55SParav Pandit struct ib_device *ib_device_get_by_index(const struct net *net, u32 index) 307f8978bd9SLeon Romanovsky { 308f8978bd9SLeon Romanovsky struct ib_device *device; 309f8978bd9SLeon Romanovsky 310921eab11SJason Gunthorpe down_read(&devices_rwsem); 3110df91bb6SJason Gunthorpe device = xa_load(&devices, index); 31201b67117SParav Pandit if (device) { 31337eeab55SParav Pandit if (!rdma_dev_access_netns(device, net)) { 31437eeab55SParav Pandit device = NULL; 31537eeab55SParav Pandit goto out; 31637eeab55SParav Pandit } 31737eeab55SParav Pandit 318d79af724SJason Gunthorpe if (!ib_device_try_get(device)) 31901b67117SParav Pandit device = NULL; 32001b67117SParav Pandit } 32137eeab55SParav Pandit out: 322921eab11SJason Gunthorpe up_read(&devices_rwsem); 323f8978bd9SLeon Romanovsky return device; 324f8978bd9SLeon Romanovsky } 325f8978bd9SLeon Romanovsky 326d79af724SJason Gunthorpe /** 327d79af724SJason Gunthorpe * ib_device_put - Release IB device reference 328d79af724SJason Gunthorpe * @device: device whose reference to be released 329d79af724SJason Gunthorpe * 330d79af724SJason Gunthorpe * ib_device_put() releases reference to the IB device to allow it to be 331d79af724SJason Gunthorpe * unregistered and eventually free. 332d79af724SJason Gunthorpe */ 33301b67117SParav Pandit void ib_device_put(struct ib_device *device) 33401b67117SParav Pandit { 33501b67117SParav Pandit if (refcount_dec_and_test(&device->refcount)) 33601b67117SParav Pandit complete(&device->unreg_completion); 33701b67117SParav Pandit } 338d79af724SJason Gunthorpe EXPORT_SYMBOL(ib_device_put); 33901b67117SParav Pandit 3401da177e4SLinus Torvalds static struct ib_device *__ib_device_get_by_name(const char *name) 3411da177e4SLinus Torvalds { 3421da177e4SLinus Torvalds struct ib_device *device; 3430df91bb6SJason Gunthorpe unsigned long index; 3441da177e4SLinus Torvalds 3450df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) 346896de009SJason Gunthorpe if (!strcmp(name, dev_name(&device->dev))) 3471da177e4SLinus Torvalds return device; 3481da177e4SLinus Torvalds 3491da177e4SLinus Torvalds return NULL; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 3526cc2c8e5SJason Gunthorpe /** 3536cc2c8e5SJason Gunthorpe * ib_device_get_by_name - Find an IB device by name 3546cc2c8e5SJason Gunthorpe * @name: The name to look for 3556cc2c8e5SJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 3566cc2c8e5SJason Gunthorpe * 3576cc2c8e5SJason Gunthorpe * Find and hold an ib_device by its name. The caller must call 3586cc2c8e5SJason Gunthorpe * ib_device_put() on the returned pointer. 3596cc2c8e5SJason Gunthorpe */ 3606cc2c8e5SJason Gunthorpe struct ib_device *ib_device_get_by_name(const char *name, 3616cc2c8e5SJason Gunthorpe enum rdma_driver_id driver_id) 3626cc2c8e5SJason Gunthorpe { 3636cc2c8e5SJason Gunthorpe struct ib_device *device; 3646cc2c8e5SJason Gunthorpe 3656cc2c8e5SJason Gunthorpe down_read(&devices_rwsem); 3666cc2c8e5SJason Gunthorpe device = __ib_device_get_by_name(name); 3676cc2c8e5SJason Gunthorpe if (device && driver_id != RDMA_DRIVER_UNKNOWN && 368b9560a41SJason Gunthorpe device->ops.driver_id != driver_id) 3696cc2c8e5SJason Gunthorpe device = NULL; 3706cc2c8e5SJason Gunthorpe 3716cc2c8e5SJason Gunthorpe if (device) { 3726cc2c8e5SJason Gunthorpe if (!ib_device_try_get(device)) 3736cc2c8e5SJason Gunthorpe device = NULL; 3746cc2c8e5SJason Gunthorpe } 3756cc2c8e5SJason Gunthorpe up_read(&devices_rwsem); 3766cc2c8e5SJason Gunthorpe return device; 3776cc2c8e5SJason Gunthorpe } 3786cc2c8e5SJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_name); 3796cc2c8e5SJason Gunthorpe 3804e0f7b90SParav Pandit static int rename_compat_devs(struct ib_device *device) 3814e0f7b90SParav Pandit { 3824e0f7b90SParav Pandit struct ib_core_device *cdev; 3834e0f7b90SParav Pandit unsigned long index; 3844e0f7b90SParav Pandit int ret = 0; 3854e0f7b90SParav Pandit 3864e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 3874e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) { 3884e0f7b90SParav Pandit ret = device_rename(&cdev->dev, dev_name(&device->dev)); 3894e0f7b90SParav Pandit if (ret) { 3904e0f7b90SParav Pandit dev_warn(&cdev->dev, 3914e0f7b90SParav Pandit "Fail to rename compatdev to new name %s\n", 3924e0f7b90SParav Pandit dev_name(&device->dev)); 3934e0f7b90SParav Pandit break; 3944e0f7b90SParav Pandit } 3954e0f7b90SParav Pandit } 3964e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 3974e0f7b90SParav Pandit return ret; 3984e0f7b90SParav Pandit } 3994e0f7b90SParav Pandit 400d21943ddSLeon Romanovsky int ib_device_rename(struct ib_device *ibdev, const char *name) 401d21943ddSLeon Romanovsky { 402dc1435c0SLeon Romanovsky unsigned long index; 403dc1435c0SLeon Romanovsky void *client_data; 404e3593b56SJason Gunthorpe int ret; 405d21943ddSLeon Romanovsky 406921eab11SJason Gunthorpe down_write(&devices_rwsem); 407e3593b56SJason Gunthorpe if (!strcmp(name, dev_name(&ibdev->dev))) { 408dc1435c0SLeon Romanovsky up_write(&devices_rwsem); 409dc1435c0SLeon Romanovsky return 0; 410e3593b56SJason Gunthorpe } 411e3593b56SJason Gunthorpe 412344684e6SJason Gunthorpe if (__ib_device_get_by_name(name)) { 413dc1435c0SLeon Romanovsky up_write(&devices_rwsem); 414dc1435c0SLeon Romanovsky return -EEXIST; 415d21943ddSLeon Romanovsky } 416d21943ddSLeon Romanovsky 417d21943ddSLeon Romanovsky ret = device_rename(&ibdev->dev, name); 418dc1435c0SLeon Romanovsky if (ret) { 419921eab11SJason Gunthorpe up_write(&devices_rwsem); 420d21943ddSLeon Romanovsky return ret; 421d21943ddSLeon Romanovsky } 422d21943ddSLeon Romanovsky 423dc1435c0SLeon Romanovsky strlcpy(ibdev->name, name, IB_DEVICE_NAME_MAX); 424dc1435c0SLeon Romanovsky ret = rename_compat_devs(ibdev); 425dc1435c0SLeon Romanovsky 426dc1435c0SLeon Romanovsky downgrade_write(&devices_rwsem); 427dc1435c0SLeon Romanovsky down_read(&ibdev->client_data_rwsem); 428dc1435c0SLeon Romanovsky xan_for_each_marked(&ibdev->client_data, index, client_data, 429dc1435c0SLeon Romanovsky CLIENT_DATA_REGISTERED) { 430dc1435c0SLeon Romanovsky struct ib_client *client = xa_load(&clients, index); 431dc1435c0SLeon Romanovsky 432dc1435c0SLeon Romanovsky if (!client || !client->rename) 433dc1435c0SLeon Romanovsky continue; 434dc1435c0SLeon Romanovsky 435dc1435c0SLeon Romanovsky client->rename(ibdev, client_data); 436dc1435c0SLeon Romanovsky } 437dc1435c0SLeon Romanovsky up_read(&ibdev->client_data_rwsem); 438dc1435c0SLeon Romanovsky up_read(&devices_rwsem); 439dc1435c0SLeon Romanovsky return 0; 440dc1435c0SLeon Romanovsky } 441dc1435c0SLeon Romanovsky 442f8fc8cd9SYamin Friedman int ib_device_set_dim(struct ib_device *ibdev, u8 use_dim) 443f8fc8cd9SYamin Friedman { 444f8fc8cd9SYamin Friedman if (use_dim > 1) 445f8fc8cd9SYamin Friedman return -EINVAL; 446f8fc8cd9SYamin Friedman ibdev->use_cq_dim = use_dim; 447f8fc8cd9SYamin Friedman 448f8fc8cd9SYamin Friedman return 0; 449f8fc8cd9SYamin Friedman } 450f8fc8cd9SYamin Friedman 451e349f858SJason Gunthorpe static int alloc_name(struct ib_device *ibdev, const char *name) 4521da177e4SLinus Torvalds { 4531da177e4SLinus Torvalds struct ib_device *device; 4540df91bb6SJason Gunthorpe unsigned long index; 4553b88afd3SJason Gunthorpe struct ida inuse; 4563b88afd3SJason Gunthorpe int rc; 4571da177e4SLinus Torvalds int i; 4581da177e4SLinus Torvalds 4599ffbe8acSNikolay Borisov lockdep_assert_held_write(&devices_rwsem); 4603b88afd3SJason Gunthorpe ida_init(&inuse); 4610df91bb6SJason Gunthorpe xa_for_each (&devices, index, device) { 462e349f858SJason Gunthorpe char buf[IB_DEVICE_NAME_MAX]; 463e349f858SJason Gunthorpe 464896de009SJason Gunthorpe if (sscanf(dev_name(&device->dev), name, &i) != 1) 4651da177e4SLinus Torvalds continue; 4663b88afd3SJason Gunthorpe if (i < 0 || i >= INT_MAX) 4671da177e4SLinus Torvalds continue; 4681da177e4SLinus Torvalds snprintf(buf, sizeof buf, name, i); 4693b88afd3SJason Gunthorpe if (strcmp(buf, dev_name(&device->dev)) != 0) 4703b88afd3SJason Gunthorpe continue; 4713b88afd3SJason Gunthorpe 4723b88afd3SJason Gunthorpe rc = ida_alloc_range(&inuse, i, i, GFP_KERNEL); 4733b88afd3SJason Gunthorpe if (rc < 0) 4743b88afd3SJason Gunthorpe goto out; 4751da177e4SLinus Torvalds } 4761da177e4SLinus Torvalds 4773b88afd3SJason Gunthorpe rc = ida_alloc(&inuse, GFP_KERNEL); 4783b88afd3SJason Gunthorpe if (rc < 0) 4793b88afd3SJason Gunthorpe goto out; 4801da177e4SLinus Torvalds 4813b88afd3SJason Gunthorpe rc = dev_set_name(&ibdev->dev, name, rc); 4823b88afd3SJason Gunthorpe out: 4833b88afd3SJason Gunthorpe ida_destroy(&inuse); 4843b88afd3SJason Gunthorpe return rc; 4851da177e4SLinus Torvalds } 4861da177e4SLinus Torvalds 48755aeed06SJason Gunthorpe static void ib_device_release(struct device *device) 48855aeed06SJason Gunthorpe { 48955aeed06SJason Gunthorpe struct ib_device *dev = container_of(device, struct ib_device, dev); 49055aeed06SJason Gunthorpe 491c2261dd7SJason Gunthorpe free_netdevs(dev); 492652432f3SJason Gunthorpe WARN_ON(refcount_read(&dev->refcount)); 49346bdf370SKamal Heib if (dev->port_data) { 49403db3a2dSMatan Barak ib_cache_release_one(dev); 495b34b269aSJason Gunthorpe ib_security_release_port_pkey_list(dev); 496413d3347SMark Zhang rdma_counter_release(dev); 497324e227eSJason Gunthorpe kfree_rcu(container_of(dev->port_data, struct ib_port_data_rcu, 498324e227eSJason Gunthorpe pdata[0]), 499324e227eSJason Gunthorpe rcu_head); 50046bdf370SKamal Heib } 501413d3347SMark Zhang 50256594ae1SParav Pandit mutex_destroy(&dev->unregistration_lock); 50356594ae1SParav Pandit mutex_destroy(&dev->compat_devs_mutex); 50456594ae1SParav Pandit 50546bdf370SKamal Heib xa_destroy(&dev->compat_devs); 50646bdf370SKamal Heib xa_destroy(&dev->client_data); 507324e227eSJason Gunthorpe kfree_rcu(dev, rcu_head); 50855aeed06SJason Gunthorpe } 50955aeed06SJason Gunthorpe 51055aeed06SJason Gunthorpe static int ib_device_uevent(struct device *device, 51155aeed06SJason Gunthorpe struct kobj_uevent_env *env) 51255aeed06SJason Gunthorpe { 513896de009SJason Gunthorpe if (add_uevent_var(env, "NAME=%s", dev_name(device))) 51455aeed06SJason Gunthorpe return -ENOMEM; 51555aeed06SJason Gunthorpe 51655aeed06SJason Gunthorpe /* 51755aeed06SJason Gunthorpe * It would be nice to pass the node GUID with the event... 51855aeed06SJason Gunthorpe */ 51955aeed06SJason Gunthorpe 52055aeed06SJason Gunthorpe return 0; 52155aeed06SJason Gunthorpe } 52255aeed06SJason Gunthorpe 52362dfa795SParav Pandit static const void *net_namespace(struct device *d) 52462dfa795SParav Pandit { 5254e0f7b90SParav Pandit struct ib_core_device *coredev = 5264e0f7b90SParav Pandit container_of(d, struct ib_core_device, dev); 5274e0f7b90SParav Pandit 5284e0f7b90SParav Pandit return read_pnet(&coredev->rdma_net); 52962dfa795SParav Pandit } 53062dfa795SParav Pandit 53155aeed06SJason Gunthorpe static struct class ib_class = { 53255aeed06SJason Gunthorpe .name = "infiniband", 53355aeed06SJason Gunthorpe .dev_release = ib_device_release, 53455aeed06SJason Gunthorpe .dev_uevent = ib_device_uevent, 53562dfa795SParav Pandit .ns_type = &net_ns_type_operations, 53662dfa795SParav Pandit .namespace = net_namespace, 53755aeed06SJason Gunthorpe }; 53855aeed06SJason Gunthorpe 539cebe556bSParav Pandit static void rdma_init_coredev(struct ib_core_device *coredev, 5404e0f7b90SParav Pandit struct ib_device *dev, struct net *net) 541cebe556bSParav Pandit { 542cebe556bSParav Pandit /* This BUILD_BUG_ON is intended to catch layout change 543cebe556bSParav Pandit * of union of ib_core_device and device. 544cebe556bSParav Pandit * dev must be the first element as ib_core and providers 545cebe556bSParav Pandit * driver uses it. Adding anything in ib_core_device before 546cebe556bSParav Pandit * device will break this assumption. 547cebe556bSParav Pandit */ 548cebe556bSParav Pandit BUILD_BUG_ON(offsetof(struct ib_device, coredev.dev) != 549cebe556bSParav Pandit offsetof(struct ib_device, dev)); 550cebe556bSParav Pandit 551cebe556bSParav Pandit coredev->dev.class = &ib_class; 552cebe556bSParav Pandit coredev->dev.groups = dev->groups; 553cebe556bSParav Pandit device_initialize(&coredev->dev); 554cebe556bSParav Pandit coredev->owner = dev; 555cebe556bSParav Pandit INIT_LIST_HEAD(&coredev->port_list); 5564e0f7b90SParav Pandit write_pnet(&coredev->rdma_net, net); 557cebe556bSParav Pandit } 558cebe556bSParav Pandit 5591da177e4SLinus Torvalds /** 560459cc69fSLeon Romanovsky * _ib_alloc_device - allocate an IB device struct 5611da177e4SLinus Torvalds * @size:size of structure to allocate 5621da177e4SLinus Torvalds * 5631da177e4SLinus Torvalds * Low-level drivers should use ib_alloc_device() to allocate &struct 5641da177e4SLinus Torvalds * ib_device. @size is the size of the structure to be allocated, 5651da177e4SLinus Torvalds * including any private data used by the low-level driver. 5661da177e4SLinus Torvalds * ib_dealloc_device() must be used to free structures allocated with 5671da177e4SLinus Torvalds * ib_alloc_device(). 5681da177e4SLinus Torvalds */ 569459cc69fSLeon Romanovsky struct ib_device *_ib_alloc_device(size_t size) 5701da177e4SLinus Torvalds { 57155aeed06SJason Gunthorpe struct ib_device *device; 5721da177e4SLinus Torvalds 57355aeed06SJason Gunthorpe if (WARN_ON(size < sizeof(struct ib_device))) 57455aeed06SJason Gunthorpe return NULL; 57555aeed06SJason Gunthorpe 57655aeed06SJason Gunthorpe device = kzalloc(size, GFP_KERNEL); 57755aeed06SJason Gunthorpe if (!device) 57855aeed06SJason Gunthorpe return NULL; 57955aeed06SJason Gunthorpe 58041eda65cSLeon Romanovsky if (rdma_restrack_init(device)) { 58141eda65cSLeon Romanovsky kfree(device); 58241eda65cSLeon Romanovsky return NULL; 58341eda65cSLeon Romanovsky } 58402d8883fSLeon Romanovsky 5855f8f5499SParav Pandit device->groups[0] = &ib_dev_attr_group; 5864e0f7b90SParav Pandit rdma_init_coredev(&device->coredev, device, &init_net); 58755aeed06SJason Gunthorpe 58855aeed06SJason Gunthorpe INIT_LIST_HEAD(&device->event_handler_list); 58940adf686SParav Pandit spin_lock_init(&device->qp_open_list_lock); 5906b57cea9SParav Pandit init_rwsem(&device->event_handler_rwsem); 591d0899892SJason Gunthorpe mutex_init(&device->unregistration_lock); 5920df91bb6SJason Gunthorpe /* 5930df91bb6SJason Gunthorpe * client_data needs to be alloc because we don't want our mark to be 5940df91bb6SJason Gunthorpe * destroyed if the user stores NULL in the client data. 5950df91bb6SJason Gunthorpe */ 5960df91bb6SJason Gunthorpe xa_init_flags(&device->client_data, XA_FLAGS_ALLOC); 597921eab11SJason Gunthorpe init_rwsem(&device->client_data_rwsem); 5984e0f7b90SParav Pandit xa_init_flags(&device->compat_devs, XA_FLAGS_ALLOC); 5994e0f7b90SParav Pandit mutex_init(&device->compat_devs_mutex); 60001b67117SParav Pandit init_completion(&device->unreg_completion); 601d0899892SJason Gunthorpe INIT_WORK(&device->unregistration_work, ib_unregister_work); 60255aeed06SJason Gunthorpe 60355aeed06SJason Gunthorpe return device; 6041da177e4SLinus Torvalds } 605459cc69fSLeon Romanovsky EXPORT_SYMBOL(_ib_alloc_device); 6061da177e4SLinus Torvalds 6071da177e4SLinus Torvalds /** 6081da177e4SLinus Torvalds * ib_dealloc_device - free an IB device struct 6091da177e4SLinus Torvalds * @device:structure to free 6101da177e4SLinus Torvalds * 6111da177e4SLinus Torvalds * Free a structure allocated with ib_alloc_device(). 6121da177e4SLinus Torvalds */ 6131da177e4SLinus Torvalds void ib_dealloc_device(struct ib_device *device) 6141da177e4SLinus Torvalds { 615d0899892SJason Gunthorpe if (device->ops.dealloc_driver) 616d0899892SJason Gunthorpe device->ops.dealloc_driver(device); 617d0899892SJason Gunthorpe 618d0899892SJason Gunthorpe /* 619d0899892SJason Gunthorpe * ib_unregister_driver() requires all devices to remain in the xarray 620d0899892SJason Gunthorpe * while their ops are callable. The last op we call is dealloc_driver 621d0899892SJason Gunthorpe * above. This is needed to create a fence on op callbacks prior to 622d0899892SJason Gunthorpe * allowing the driver module to unload. 623d0899892SJason Gunthorpe */ 624d0899892SJason Gunthorpe down_write(&devices_rwsem); 625d0899892SJason Gunthorpe if (xa_load(&devices, device->index) == device) 626d0899892SJason Gunthorpe xa_erase(&devices, device->index); 627d0899892SJason Gunthorpe up_write(&devices_rwsem); 628d0899892SJason Gunthorpe 629c2261dd7SJason Gunthorpe /* Expedite releasing netdev references */ 630c2261dd7SJason Gunthorpe free_netdevs(device); 631c2261dd7SJason Gunthorpe 6324e0f7b90SParav Pandit WARN_ON(!xa_empty(&device->compat_devs)); 6330df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&device->client_data)); 634652432f3SJason Gunthorpe WARN_ON(refcount_read(&device->refcount)); 6350ad699c0SLeon Romanovsky rdma_restrack_clean(device); 636e155755eSParav Pandit /* Balances with device_initialize */ 637924b8900SLeon Romanovsky put_device(&device->dev); 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds EXPORT_SYMBOL(ib_dealloc_device); 6401da177e4SLinus Torvalds 641921eab11SJason Gunthorpe /* 642921eab11SJason Gunthorpe * add_client_context() and remove_client_context() must be safe against 643921eab11SJason Gunthorpe * parallel calls on the same device - registration/unregistration of both the 644921eab11SJason Gunthorpe * device and client can be occurring in parallel. 645921eab11SJason Gunthorpe * 646921eab11SJason Gunthorpe * The routines need to be a fence, any caller must not return until the add 647921eab11SJason Gunthorpe * or remove is fully completed. 648921eab11SJason Gunthorpe */ 649921eab11SJason Gunthorpe static int add_client_context(struct ib_device *device, 650921eab11SJason Gunthorpe struct ib_client *client) 6511da177e4SLinus Torvalds { 652921eab11SJason Gunthorpe int ret = 0; 6531da177e4SLinus Torvalds 6546780c4faSGal Pressman if (!device->kverbs_provider && !client->no_kverbs_req) 655921eab11SJason Gunthorpe return 0; 6566780c4faSGal Pressman 657921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 658921eab11SJason Gunthorpe /* 659621e55ffSJason Gunthorpe * So long as the client is registered hold both the client and device 660621e55ffSJason Gunthorpe * unregistration locks. 661621e55ffSJason Gunthorpe */ 662621e55ffSJason Gunthorpe if (!refcount_inc_not_zero(&client->uses)) 663621e55ffSJason Gunthorpe goto out_unlock; 664621e55ffSJason Gunthorpe refcount_inc(&device->refcount); 665621e55ffSJason Gunthorpe 666621e55ffSJason Gunthorpe /* 667921eab11SJason Gunthorpe * Another caller to add_client_context got here first and has already 668921eab11SJason Gunthorpe * completely initialized context. 669921eab11SJason Gunthorpe */ 670921eab11SJason Gunthorpe if (xa_get_mark(&device->client_data, client->client_id, 671921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) 672921eab11SJason Gunthorpe goto out; 673921eab11SJason Gunthorpe 674921eab11SJason Gunthorpe ret = xa_err(xa_store(&device->client_data, client->client_id, NULL, 675921eab11SJason Gunthorpe GFP_KERNEL)); 676921eab11SJason Gunthorpe if (ret) 677921eab11SJason Gunthorpe goto out; 678921eab11SJason Gunthorpe downgrade_write(&device->client_data_rwsem); 67911a0ae4cSJason Gunthorpe if (client->add) { 68011a0ae4cSJason Gunthorpe if (client->add(device)) { 68111a0ae4cSJason Gunthorpe /* 68211a0ae4cSJason Gunthorpe * If a client fails to add then the error code is 68311a0ae4cSJason Gunthorpe * ignored, but we won't call any more ops on this 68411a0ae4cSJason Gunthorpe * client. 68511a0ae4cSJason Gunthorpe */ 68611a0ae4cSJason Gunthorpe xa_erase(&device->client_data, client->client_id); 68711a0ae4cSJason Gunthorpe up_read(&device->client_data_rwsem); 68811a0ae4cSJason Gunthorpe ib_device_put(device); 68911a0ae4cSJason Gunthorpe ib_client_put(client); 69011a0ae4cSJason Gunthorpe return 0; 69111a0ae4cSJason Gunthorpe } 69211a0ae4cSJason Gunthorpe } 693921eab11SJason Gunthorpe 694921eab11SJason Gunthorpe /* Readers shall not see a client until add has been completed */ 6950df91bb6SJason Gunthorpe xa_set_mark(&device->client_data, client->client_id, 6960df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED); 697921eab11SJason Gunthorpe up_read(&device->client_data_rwsem); 698921eab11SJason Gunthorpe return 0; 6991da177e4SLinus Torvalds 700921eab11SJason Gunthorpe out: 701621e55ffSJason Gunthorpe ib_device_put(device); 702621e55ffSJason Gunthorpe ib_client_put(client); 703621e55ffSJason Gunthorpe out_unlock: 704921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 705921eab11SJason Gunthorpe return ret; 706921eab11SJason Gunthorpe } 707921eab11SJason Gunthorpe 708921eab11SJason Gunthorpe static void remove_client_context(struct ib_device *device, 709921eab11SJason Gunthorpe unsigned int client_id) 710921eab11SJason Gunthorpe { 711921eab11SJason Gunthorpe struct ib_client *client; 712921eab11SJason Gunthorpe void *client_data; 713921eab11SJason Gunthorpe 714921eab11SJason Gunthorpe down_write(&device->client_data_rwsem); 715921eab11SJason Gunthorpe if (!xa_get_mark(&device->client_data, client_id, 716921eab11SJason Gunthorpe CLIENT_DATA_REGISTERED)) { 717921eab11SJason Gunthorpe up_write(&device->client_data_rwsem); 718921eab11SJason Gunthorpe return; 719921eab11SJason Gunthorpe } 720921eab11SJason Gunthorpe client_data = xa_load(&device->client_data, client_id); 721921eab11SJason Gunthorpe xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED); 722921eab11SJason Gunthorpe client = xa_load(&clients, client_id); 723621e55ffSJason Gunthorpe up_write(&device->client_data_rwsem); 724921eab11SJason Gunthorpe 725921eab11SJason Gunthorpe /* 726921eab11SJason Gunthorpe * Notice we cannot be holding any exclusive locks when calling the 727921eab11SJason Gunthorpe * remove callback as the remove callback can recurse back into any 728921eab11SJason Gunthorpe * public functions in this module and thus try for any locks those 729921eab11SJason Gunthorpe * functions take. 730921eab11SJason Gunthorpe * 731921eab11SJason Gunthorpe * For this reason clients and drivers should not call the 732921eab11SJason Gunthorpe * unregistration functions will holdling any locks. 733921eab11SJason Gunthorpe */ 734921eab11SJason Gunthorpe if (client->remove) 735921eab11SJason Gunthorpe client->remove(device, client_data); 736921eab11SJason Gunthorpe 737921eab11SJason Gunthorpe xa_erase(&device->client_data, client_id); 738621e55ffSJason Gunthorpe ib_device_put(device); 739621e55ffSJason Gunthorpe ib_client_put(client); 7401da177e4SLinus Torvalds } 7411da177e4SLinus Torvalds 742c2261dd7SJason Gunthorpe static int alloc_port_data(struct ib_device *device) 7435eb620c8SYosef Etigin { 744324e227eSJason Gunthorpe struct ib_port_data_rcu *pdata_rcu; 745ea1075edSJason Gunthorpe unsigned int port; 746c2261dd7SJason Gunthorpe 747c2261dd7SJason Gunthorpe if (device->port_data) 748c2261dd7SJason Gunthorpe return 0; 749c2261dd7SJason Gunthorpe 750c2261dd7SJason Gunthorpe /* This can only be called once the physical port range is defined */ 751c2261dd7SJason Gunthorpe if (WARN_ON(!device->phys_port_cnt)) 752c2261dd7SJason Gunthorpe return -EINVAL; 7535eb620c8SYosef Etigin 7548ceb1357SJason Gunthorpe /* 7558ceb1357SJason Gunthorpe * device->port_data is indexed directly by the port number to make 7567738613eSIra Weiny * access to this data as efficient as possible. 7577738613eSIra Weiny * 7588ceb1357SJason Gunthorpe * Therefore port_data is declared as a 1 based array with potential 7598ceb1357SJason Gunthorpe * empty slots at the beginning. 7607738613eSIra Weiny */ 761324e227eSJason Gunthorpe pdata_rcu = kzalloc(struct_size(pdata_rcu, pdata, 762324e227eSJason Gunthorpe rdma_end_port(device) + 1), 763324e227eSJason Gunthorpe GFP_KERNEL); 764324e227eSJason Gunthorpe if (!pdata_rcu) 76555aeed06SJason Gunthorpe return -ENOMEM; 766324e227eSJason Gunthorpe /* 767324e227eSJason Gunthorpe * The rcu_head is put in front of the port data array and the stored 768324e227eSJason Gunthorpe * pointer is adjusted since we never need to see that member until 769324e227eSJason Gunthorpe * kfree_rcu. 770324e227eSJason Gunthorpe */ 771324e227eSJason Gunthorpe device->port_data = pdata_rcu->pdata; 7725eb620c8SYosef Etigin 773ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 7748ceb1357SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 7758ceb1357SJason Gunthorpe 776324e227eSJason Gunthorpe pdata->ib_dev = device; 7778ceb1357SJason Gunthorpe spin_lock_init(&pdata->pkey_list_lock); 7788ceb1357SJason Gunthorpe INIT_LIST_HEAD(&pdata->pkey_list); 779c2261dd7SJason Gunthorpe spin_lock_init(&pdata->netdev_lock); 780324e227eSJason Gunthorpe INIT_HLIST_NODE(&pdata->ndev_hash_link); 781c2261dd7SJason Gunthorpe } 782c2261dd7SJason Gunthorpe return 0; 783c2261dd7SJason Gunthorpe } 784c2261dd7SJason Gunthorpe 785c2261dd7SJason Gunthorpe static int verify_immutable(const struct ib_device *dev, u8 port) 786c2261dd7SJason Gunthorpe { 787c2261dd7SJason Gunthorpe return WARN_ON(!rdma_cap_ib_mad(dev, port) && 788c2261dd7SJason Gunthorpe rdma_max_mad_size(dev, port) != 0); 789c2261dd7SJason Gunthorpe } 790c2261dd7SJason Gunthorpe 791c2261dd7SJason Gunthorpe static int setup_port_data(struct ib_device *device) 792c2261dd7SJason Gunthorpe { 793c2261dd7SJason Gunthorpe unsigned int port; 794c2261dd7SJason Gunthorpe int ret; 795c2261dd7SJason Gunthorpe 796c2261dd7SJason Gunthorpe ret = alloc_port_data(device); 797c2261dd7SJason Gunthorpe if (ret) 798c2261dd7SJason Gunthorpe return ret; 799c2261dd7SJason Gunthorpe 800c2261dd7SJason Gunthorpe rdma_for_each_port (device, port) { 801c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &device->port_data[port]; 8028ceb1357SJason Gunthorpe 8038ceb1357SJason Gunthorpe ret = device->ops.get_port_immutable(device, port, 8048ceb1357SJason Gunthorpe &pdata->immutable); 8055eb620c8SYosef Etigin if (ret) 8065eb620c8SYosef Etigin return ret; 80755aeed06SJason Gunthorpe 80855aeed06SJason Gunthorpe if (verify_immutable(device, port)) 80955aeed06SJason Gunthorpe return -EINVAL; 81055aeed06SJason Gunthorpe } 81155aeed06SJason Gunthorpe return 0; 8125eb620c8SYosef Etigin } 8135eb620c8SYosef Etigin 8149abb0d1bSLeon Romanovsky void ib_get_device_fw_str(struct ib_device *dev, char *str) 8155fa76c20SIra Weiny { 8163023a1e9SKamal Heib if (dev->ops.get_dev_fw_str) 8173023a1e9SKamal Heib dev->ops.get_dev_fw_str(dev, str); 8185fa76c20SIra Weiny else 8195fa76c20SIra Weiny str[0] = '\0'; 8205fa76c20SIra Weiny } 8215fa76c20SIra Weiny EXPORT_SYMBOL(ib_get_device_fw_str); 8225fa76c20SIra Weiny 8238f408ab6SDaniel Jurgens static void ib_policy_change_task(struct work_struct *work) 8248f408ab6SDaniel Jurgens { 8258f408ab6SDaniel Jurgens struct ib_device *dev; 8260df91bb6SJason Gunthorpe unsigned long index; 8278f408ab6SDaniel Jurgens 828921eab11SJason Gunthorpe down_read(&devices_rwsem); 8290df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 830ea1075edSJason Gunthorpe unsigned int i; 8318f408ab6SDaniel Jurgens 832ea1075edSJason Gunthorpe rdma_for_each_port (dev, i) { 8338f408ab6SDaniel Jurgens u64 sp; 8348f408ab6SDaniel Jurgens int ret = ib_get_cached_subnet_prefix(dev, 8358f408ab6SDaniel Jurgens i, 8368f408ab6SDaniel Jurgens &sp); 8378f408ab6SDaniel Jurgens 8388f408ab6SDaniel Jurgens WARN_ONCE(ret, 8398f408ab6SDaniel Jurgens "ib_get_cached_subnet_prefix err: %d, this should never happen here\n", 8408f408ab6SDaniel Jurgens ret); 841a750cfdeSDaniel Jurgens if (!ret) 8428f408ab6SDaniel Jurgens ib_security_cache_change(dev, i, sp); 8438f408ab6SDaniel Jurgens } 8448f408ab6SDaniel Jurgens } 845921eab11SJason Gunthorpe up_read(&devices_rwsem); 8468f408ab6SDaniel Jurgens } 8478f408ab6SDaniel Jurgens 8488f408ab6SDaniel Jurgens static int ib_security_change(struct notifier_block *nb, unsigned long event, 8498f408ab6SDaniel Jurgens void *lsm_data) 8508f408ab6SDaniel Jurgens { 8518f408ab6SDaniel Jurgens if (event != LSM_POLICY_CHANGE) 8528f408ab6SDaniel Jurgens return NOTIFY_DONE; 8538f408ab6SDaniel Jurgens 8548f408ab6SDaniel Jurgens schedule_work(&ib_policy_change_work); 855c66f6741SDaniel Jurgens ib_mad_agent_security_change(); 8568f408ab6SDaniel Jurgens 8578f408ab6SDaniel Jurgens return NOTIFY_OK; 8588f408ab6SDaniel Jurgens } 8598f408ab6SDaniel Jurgens 8604e0f7b90SParav Pandit static void compatdev_release(struct device *dev) 8614e0f7b90SParav Pandit { 8624e0f7b90SParav Pandit struct ib_core_device *cdev = 8634e0f7b90SParav Pandit container_of(dev, struct ib_core_device, dev); 8644e0f7b90SParav Pandit 8654e0f7b90SParav Pandit kfree(cdev); 8664e0f7b90SParav Pandit } 8674e0f7b90SParav Pandit 8684e0f7b90SParav Pandit static int add_one_compat_dev(struct ib_device *device, 8694e0f7b90SParav Pandit struct rdma_dev_net *rnet) 8704e0f7b90SParav Pandit { 8714e0f7b90SParav Pandit struct ib_core_device *cdev; 8724e0f7b90SParav Pandit int ret; 8734e0f7b90SParav Pandit 8742b34c558SParav Pandit lockdep_assert_held(&rdma_nets_rwsem); 875a56bc45bSParav Pandit if (!ib_devices_shared_netns) 876a56bc45bSParav Pandit return 0; 877a56bc45bSParav Pandit 8784e0f7b90SParav Pandit /* 8794e0f7b90SParav Pandit * Create and add compat device in all namespaces other than where it 8804e0f7b90SParav Pandit * is currently bound to. 8814e0f7b90SParav Pandit */ 8824e0f7b90SParav Pandit if (net_eq(read_pnet(&rnet->net), 8834e0f7b90SParav Pandit read_pnet(&device->coredev.rdma_net))) 8844e0f7b90SParav Pandit return 0; 8854e0f7b90SParav Pandit 8864e0f7b90SParav Pandit /* 8874e0f7b90SParav Pandit * The first of init_net() or ib_register_device() to take the 8884e0f7b90SParav Pandit * compat_devs_mutex wins and gets to add the device. Others will wait 8894e0f7b90SParav Pandit * for completion here. 8904e0f7b90SParav Pandit */ 8914e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 8924e0f7b90SParav Pandit cdev = xa_load(&device->compat_devs, rnet->id); 8934e0f7b90SParav Pandit if (cdev) { 8944e0f7b90SParav Pandit ret = 0; 8954e0f7b90SParav Pandit goto done; 8964e0f7b90SParav Pandit } 8974e0f7b90SParav Pandit ret = xa_reserve(&device->compat_devs, rnet->id, GFP_KERNEL); 8984e0f7b90SParav Pandit if (ret) 8994e0f7b90SParav Pandit goto done; 9004e0f7b90SParav Pandit 9014e0f7b90SParav Pandit cdev = kzalloc(sizeof(*cdev), GFP_KERNEL); 9024e0f7b90SParav Pandit if (!cdev) { 9034e0f7b90SParav Pandit ret = -ENOMEM; 9044e0f7b90SParav Pandit goto cdev_err; 9054e0f7b90SParav Pandit } 9064e0f7b90SParav Pandit 9074e0f7b90SParav Pandit cdev->dev.parent = device->dev.parent; 9084e0f7b90SParav Pandit rdma_init_coredev(cdev, device, read_pnet(&rnet->net)); 9094e0f7b90SParav Pandit cdev->dev.release = compatdev_release; 910f2f2b3bbSJason Gunthorpe ret = dev_set_name(&cdev->dev, "%s", dev_name(&device->dev)); 911f2f2b3bbSJason Gunthorpe if (ret) 912f2f2b3bbSJason Gunthorpe goto add_err; 9134e0f7b90SParav Pandit 9144e0f7b90SParav Pandit ret = device_add(&cdev->dev); 9154e0f7b90SParav Pandit if (ret) 9164e0f7b90SParav Pandit goto add_err; 917eb15c78bSParav Pandit ret = ib_setup_port_attrs(cdev); 9185417783eSParav Pandit if (ret) 9195417783eSParav Pandit goto port_err; 9204e0f7b90SParav Pandit 9214e0f7b90SParav Pandit ret = xa_err(xa_store(&device->compat_devs, rnet->id, 9224e0f7b90SParav Pandit cdev, GFP_KERNEL)); 9234e0f7b90SParav Pandit if (ret) 9244e0f7b90SParav Pandit goto insert_err; 9254e0f7b90SParav Pandit 9264e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 9274e0f7b90SParav Pandit return 0; 9284e0f7b90SParav Pandit 9294e0f7b90SParav Pandit insert_err: 9305417783eSParav Pandit ib_free_port_attrs(cdev); 9315417783eSParav Pandit port_err: 9324e0f7b90SParav Pandit device_del(&cdev->dev); 9334e0f7b90SParav Pandit add_err: 9344e0f7b90SParav Pandit put_device(&cdev->dev); 9354e0f7b90SParav Pandit cdev_err: 9364e0f7b90SParav Pandit xa_release(&device->compat_devs, rnet->id); 9374e0f7b90SParav Pandit done: 9384e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 9394e0f7b90SParav Pandit return ret; 9404e0f7b90SParav Pandit } 9414e0f7b90SParav Pandit 9424e0f7b90SParav Pandit static void remove_one_compat_dev(struct ib_device *device, u32 id) 9434e0f7b90SParav Pandit { 9444e0f7b90SParav Pandit struct ib_core_device *cdev; 9454e0f7b90SParav Pandit 9464e0f7b90SParav Pandit mutex_lock(&device->compat_devs_mutex); 9474e0f7b90SParav Pandit cdev = xa_erase(&device->compat_devs, id); 9484e0f7b90SParav Pandit mutex_unlock(&device->compat_devs_mutex); 9494e0f7b90SParav Pandit if (cdev) { 9505417783eSParav Pandit ib_free_port_attrs(cdev); 9514e0f7b90SParav Pandit device_del(&cdev->dev); 9524e0f7b90SParav Pandit put_device(&cdev->dev); 9534e0f7b90SParav Pandit } 9544e0f7b90SParav Pandit } 9554e0f7b90SParav Pandit 9564e0f7b90SParav Pandit static void remove_compat_devs(struct ib_device *device) 9574e0f7b90SParav Pandit { 9584e0f7b90SParav Pandit struct ib_core_device *cdev; 9594e0f7b90SParav Pandit unsigned long index; 9604e0f7b90SParav Pandit 9614e0f7b90SParav Pandit xa_for_each (&device->compat_devs, index, cdev) 9624e0f7b90SParav Pandit remove_one_compat_dev(device, index); 9634e0f7b90SParav Pandit } 9644e0f7b90SParav Pandit 9654e0f7b90SParav Pandit static int add_compat_devs(struct ib_device *device) 9664e0f7b90SParav Pandit { 9674e0f7b90SParav Pandit struct rdma_dev_net *rnet; 9684e0f7b90SParav Pandit unsigned long index; 9694e0f7b90SParav Pandit int ret = 0; 9704e0f7b90SParav Pandit 971decbc7a6SParav Pandit lockdep_assert_held(&devices_rwsem); 972decbc7a6SParav Pandit 9734e0f7b90SParav Pandit down_read(&rdma_nets_rwsem); 9744e0f7b90SParav Pandit xa_for_each (&rdma_nets, index, rnet) { 9754e0f7b90SParav Pandit ret = add_one_compat_dev(device, rnet); 9764e0f7b90SParav Pandit if (ret) 9774e0f7b90SParav Pandit break; 9784e0f7b90SParav Pandit } 9794e0f7b90SParav Pandit up_read(&rdma_nets_rwsem); 9804e0f7b90SParav Pandit return ret; 9814e0f7b90SParav Pandit } 9824e0f7b90SParav Pandit 9832b34c558SParav Pandit static void remove_all_compat_devs(void) 9842b34c558SParav Pandit { 9852b34c558SParav Pandit struct ib_compat_device *cdev; 9862b34c558SParav Pandit struct ib_device *dev; 9872b34c558SParav Pandit unsigned long index; 9882b34c558SParav Pandit 9892b34c558SParav Pandit down_read(&devices_rwsem); 9902b34c558SParav Pandit xa_for_each (&devices, index, dev) { 9912b34c558SParav Pandit unsigned long c_index = 0; 9922b34c558SParav Pandit 9932b34c558SParav Pandit /* Hold nets_rwsem so that any other thread modifying this 9942b34c558SParav Pandit * system param can sync with this thread. 9952b34c558SParav Pandit */ 9962b34c558SParav Pandit down_read(&rdma_nets_rwsem); 9972b34c558SParav Pandit xa_for_each (&dev->compat_devs, c_index, cdev) 9982b34c558SParav Pandit remove_one_compat_dev(dev, c_index); 9992b34c558SParav Pandit up_read(&rdma_nets_rwsem); 10002b34c558SParav Pandit } 10012b34c558SParav Pandit up_read(&devices_rwsem); 10022b34c558SParav Pandit } 10032b34c558SParav Pandit 10042b34c558SParav Pandit static int add_all_compat_devs(void) 10052b34c558SParav Pandit { 10062b34c558SParav Pandit struct rdma_dev_net *rnet; 10072b34c558SParav Pandit struct ib_device *dev; 10082b34c558SParav Pandit unsigned long index; 10092b34c558SParav Pandit int ret = 0; 10102b34c558SParav Pandit 10112b34c558SParav Pandit down_read(&devices_rwsem); 10122b34c558SParav Pandit xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 10132b34c558SParav Pandit unsigned long net_index = 0; 10142b34c558SParav Pandit 10152b34c558SParav Pandit /* Hold nets_rwsem so that any other thread modifying this 10162b34c558SParav Pandit * system param can sync with this thread. 10172b34c558SParav Pandit */ 10182b34c558SParav Pandit down_read(&rdma_nets_rwsem); 10192b34c558SParav Pandit xa_for_each (&rdma_nets, net_index, rnet) { 10202b34c558SParav Pandit ret = add_one_compat_dev(dev, rnet); 10212b34c558SParav Pandit if (ret) 10222b34c558SParav Pandit break; 10232b34c558SParav Pandit } 10242b34c558SParav Pandit up_read(&rdma_nets_rwsem); 10252b34c558SParav Pandit } 10262b34c558SParav Pandit up_read(&devices_rwsem); 10272b34c558SParav Pandit if (ret) 10282b34c558SParav Pandit remove_all_compat_devs(); 10292b34c558SParav Pandit return ret; 10302b34c558SParav Pandit } 10312b34c558SParav Pandit 10322b34c558SParav Pandit int rdma_compatdev_set(u8 enable) 10332b34c558SParav Pandit { 10342b34c558SParav Pandit struct rdma_dev_net *rnet; 10352b34c558SParav Pandit unsigned long index; 10362b34c558SParav Pandit int ret = 0; 10372b34c558SParav Pandit 10382b34c558SParav Pandit down_write(&rdma_nets_rwsem); 10392b34c558SParav Pandit if (ib_devices_shared_netns == enable) { 10402b34c558SParav Pandit up_write(&rdma_nets_rwsem); 10412b34c558SParav Pandit return 0; 10422b34c558SParav Pandit } 10432b34c558SParav Pandit 10442b34c558SParav Pandit /* enable/disable of compat devices is not supported 10452b34c558SParav Pandit * when more than default init_net exists. 10462b34c558SParav Pandit */ 10472b34c558SParav Pandit xa_for_each (&rdma_nets, index, rnet) { 10482b34c558SParav Pandit ret++; 10492b34c558SParav Pandit break; 10502b34c558SParav Pandit } 10512b34c558SParav Pandit if (!ret) 10522b34c558SParav Pandit ib_devices_shared_netns = enable; 10532b34c558SParav Pandit up_write(&rdma_nets_rwsem); 10542b34c558SParav Pandit if (ret) 10552b34c558SParav Pandit return -EBUSY; 10562b34c558SParav Pandit 10572b34c558SParav Pandit if (enable) 10582b34c558SParav Pandit ret = add_all_compat_devs(); 10592b34c558SParav Pandit else 10602b34c558SParav Pandit remove_all_compat_devs(); 10612b34c558SParav Pandit return ret; 10622b34c558SParav Pandit } 10632b34c558SParav Pandit 10644e0f7b90SParav Pandit static void rdma_dev_exit_net(struct net *net) 10654e0f7b90SParav Pandit { 10661d2fedd8SParav Pandit struct rdma_dev_net *rnet = rdma_net_to_dev_net(net); 10674e0f7b90SParav Pandit struct ib_device *dev; 10684e0f7b90SParav Pandit unsigned long index; 10694e0f7b90SParav Pandit int ret; 10704e0f7b90SParav Pandit 10714e0f7b90SParav Pandit down_write(&rdma_nets_rwsem); 10724e0f7b90SParav Pandit /* 10734e0f7b90SParav Pandit * Prevent the ID from being re-used and hide the id from xa_for_each. 10744e0f7b90SParav Pandit */ 10754e0f7b90SParav Pandit ret = xa_err(xa_store(&rdma_nets, rnet->id, NULL, GFP_KERNEL)); 10764e0f7b90SParav Pandit WARN_ON(ret); 10774e0f7b90SParav Pandit up_write(&rdma_nets_rwsem); 10784e0f7b90SParav Pandit 10794e0f7b90SParav Pandit down_read(&devices_rwsem); 10804e0f7b90SParav Pandit xa_for_each (&devices, index, dev) { 10814e0f7b90SParav Pandit get_device(&dev->dev); 10824e0f7b90SParav Pandit /* 10834e0f7b90SParav Pandit * Release the devices_rwsem so that pontentially blocking 10844e0f7b90SParav Pandit * device_del, doesn't hold the devices_rwsem for too long. 10854e0f7b90SParav Pandit */ 10864e0f7b90SParav Pandit up_read(&devices_rwsem); 10874e0f7b90SParav Pandit 10884e0f7b90SParav Pandit remove_one_compat_dev(dev, rnet->id); 10894e0f7b90SParav Pandit 1090decbc7a6SParav Pandit /* 1091decbc7a6SParav Pandit * If the real device is in the NS then move it back to init. 1092decbc7a6SParav Pandit */ 1093decbc7a6SParav Pandit rdma_dev_change_netns(dev, net, &init_net); 1094decbc7a6SParav Pandit 10954e0f7b90SParav Pandit put_device(&dev->dev); 10964e0f7b90SParav Pandit down_read(&devices_rwsem); 10974e0f7b90SParav Pandit } 10984e0f7b90SParav Pandit up_read(&devices_rwsem); 10994e0f7b90SParav Pandit 11001d2fedd8SParav Pandit rdma_nl_net_exit(rnet); 11014e0f7b90SParav Pandit xa_erase(&rdma_nets, rnet->id); 11024e0f7b90SParav Pandit } 11034e0f7b90SParav Pandit 11044e0f7b90SParav Pandit static __net_init int rdma_dev_init_net(struct net *net) 11054e0f7b90SParav Pandit { 11061d2fedd8SParav Pandit struct rdma_dev_net *rnet = rdma_net_to_dev_net(net); 11074e0f7b90SParav Pandit unsigned long index; 11084e0f7b90SParav Pandit struct ib_device *dev; 11094e0f7b90SParav Pandit int ret; 11104e0f7b90SParav Pandit 11111d2fedd8SParav Pandit write_pnet(&rnet->net, net); 11121d2fedd8SParav Pandit 11131d2fedd8SParav Pandit ret = rdma_nl_net_init(rnet); 11141d2fedd8SParav Pandit if (ret) 11151d2fedd8SParav Pandit return ret; 11161d2fedd8SParav Pandit 11174e0f7b90SParav Pandit /* No need to create any compat devices in default init_net. */ 11184e0f7b90SParav Pandit if (net_eq(net, &init_net)) 11194e0f7b90SParav Pandit return 0; 11204e0f7b90SParav Pandit 11214e0f7b90SParav Pandit ret = xa_alloc(&rdma_nets, &rnet->id, rnet, xa_limit_32b, GFP_KERNEL); 11221d2fedd8SParav Pandit if (ret) { 11231d2fedd8SParav Pandit rdma_nl_net_exit(rnet); 11244e0f7b90SParav Pandit return ret; 11251d2fedd8SParav Pandit } 11264e0f7b90SParav Pandit 11274e0f7b90SParav Pandit down_read(&devices_rwsem); 11284e0f7b90SParav Pandit xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 11292b34c558SParav Pandit /* Hold nets_rwsem so that netlink command cannot change 11302b34c558SParav Pandit * system configuration for device sharing mode. 11312b34c558SParav Pandit */ 11322b34c558SParav Pandit down_read(&rdma_nets_rwsem); 11334e0f7b90SParav Pandit ret = add_one_compat_dev(dev, rnet); 11342b34c558SParav Pandit up_read(&rdma_nets_rwsem); 11354e0f7b90SParav Pandit if (ret) 11364e0f7b90SParav Pandit break; 11374e0f7b90SParav Pandit } 11384e0f7b90SParav Pandit up_read(&devices_rwsem); 11394e0f7b90SParav Pandit 11404e0f7b90SParav Pandit if (ret) 11414e0f7b90SParav Pandit rdma_dev_exit_net(net); 11424e0f7b90SParav Pandit 11434e0f7b90SParav Pandit return ret; 11444e0f7b90SParav Pandit } 11454e0f7b90SParav Pandit 1146ecc82c53SLeon Romanovsky /* 1147d0899892SJason Gunthorpe * Assign the unique string device name and the unique device index. This is 1148d0899892SJason Gunthorpe * undone by ib_dealloc_device. 1149ecc82c53SLeon Romanovsky */ 11500df91bb6SJason Gunthorpe static int assign_name(struct ib_device *device, const char *name) 11510df91bb6SJason Gunthorpe { 11520df91bb6SJason Gunthorpe static u32 last_id; 11530df91bb6SJason Gunthorpe int ret; 1154ecc82c53SLeon Romanovsky 1155921eab11SJason Gunthorpe down_write(&devices_rwsem); 11560df91bb6SJason Gunthorpe /* Assign a unique name to the device */ 11570df91bb6SJason Gunthorpe if (strchr(name, '%')) 11580df91bb6SJason Gunthorpe ret = alloc_name(device, name); 11590df91bb6SJason Gunthorpe else 11600df91bb6SJason Gunthorpe ret = dev_set_name(&device->dev, name); 11610df91bb6SJason Gunthorpe if (ret) 11620df91bb6SJason Gunthorpe goto out; 1163ecc82c53SLeon Romanovsky 11640df91bb6SJason Gunthorpe if (__ib_device_get_by_name(dev_name(&device->dev))) { 11650df91bb6SJason Gunthorpe ret = -ENFILE; 11660df91bb6SJason Gunthorpe goto out; 1167ecc82c53SLeon Romanovsky } 11680df91bb6SJason Gunthorpe strlcpy(device->name, dev_name(&device->dev), IB_DEVICE_NAME_MAX); 11690df91bb6SJason Gunthorpe 1170ea295481SLinus Torvalds ret = xa_alloc_cyclic(&devices, &device->index, device, xa_limit_31b, 1171ea295481SLinus Torvalds &last_id, GFP_KERNEL); 1172ea295481SLinus Torvalds if (ret > 0) 11730df91bb6SJason Gunthorpe ret = 0; 1174921eab11SJason Gunthorpe 11750df91bb6SJason Gunthorpe out: 1176921eab11SJason Gunthorpe up_write(&devices_rwsem); 11770df91bb6SJason Gunthorpe return ret; 11780df91bb6SJason Gunthorpe } 11790df91bb6SJason Gunthorpe 1180548cb4fbSParav Pandit static void setup_dma_device(struct ib_device *device) 11811da177e4SLinus Torvalds { 118299db9494SBart Van Assche struct device *parent = device->dev.parent; 11831da177e4SLinus Torvalds 11840957c29fSBart Van Assche WARN_ON_ONCE(device->dma_device); 11852f9237d4SChristoph Hellwig 11862f9237d4SChristoph Hellwig #ifdef CONFIG_DMA_OPS 11870957c29fSBart Van Assche if (device->dev.dma_ops) { 11880957c29fSBart Van Assche /* 11890957c29fSBart Van Assche * The caller provided custom DMA operations. Copy the 11900957c29fSBart Van Assche * DMA-related fields that are used by e.g. dma_alloc_coherent() 11910957c29fSBart Van Assche * into device->dev. 11920957c29fSBart Van Assche */ 11930957c29fSBart Van Assche device->dma_device = &device->dev; 119402ee9da3SBart Van Assche if (!device->dev.dma_mask) { 119502ee9da3SBart Van Assche if (parent) 119699db9494SBart Van Assche device->dev.dma_mask = parent->dma_mask; 119702ee9da3SBart Van Assche else 119802ee9da3SBart Van Assche WARN_ON_ONCE(true); 119902ee9da3SBart Van Assche } 120002ee9da3SBart Van Assche if (!device->dev.coherent_dma_mask) { 120102ee9da3SBart Van Assche if (parent) 12020957c29fSBart Van Assche device->dev.coherent_dma_mask = 12030957c29fSBart Van Assche parent->coherent_dma_mask; 120402ee9da3SBart Van Assche else 120502ee9da3SBart Van Assche WARN_ON_ONCE(true); 120602ee9da3SBart Van Assche } 12072f9237d4SChristoph Hellwig } else 12082f9237d4SChristoph Hellwig #endif /* CONFIG_DMA_OPS */ 12092f9237d4SChristoph Hellwig { 12100957c29fSBart Van Assche /* 12110957c29fSBart Van Assche * The caller did not provide custom DMA operations. Use the 12120957c29fSBart Van Assche * DMA mapping operations of the parent device. 12130957c29fSBart Van Assche */ 121402ee9da3SBart Van Assche WARN_ON_ONCE(!parent); 12150957c29fSBart Van Assche device->dma_device = parent; 12160957c29fSBart Van Assche } 1217d10bcf94SShiraz Saleem 1218c9121262SBart Van Assche if (!device->dev.dma_parms) { 1219c9121262SBart Van Assche if (parent) { 1220c9121262SBart Van Assche /* 1221c9121262SBart Van Assche * The caller did not provide DMA parameters, so 1222c9121262SBart Van Assche * 'parent' probably represents a PCI device. The PCI 1223c9121262SBart Van Assche * core sets the maximum segment size to 64 1224c9121262SBart Van Assche * KB. Increase this parameter to 2 GB. 1225c9121262SBart Van Assche */ 1226c9121262SBart Van Assche device->dev.dma_parms = parent->dma_parms; 1227c9121262SBart Van Assche dma_set_max_seg_size(device->dma_device, SZ_2G); 1228c9121262SBart Van Assche } else { 1229c9121262SBart Van Assche WARN_ON_ONCE(true); 1230c9121262SBart Van Assche } 1231c9121262SBart Van Assche } 1232548cb4fbSParav Pandit } 1233548cb4fbSParav Pandit 1234921eab11SJason Gunthorpe /* 1235921eab11SJason Gunthorpe * setup_device() allocates memory and sets up data that requires calling the 1236921eab11SJason Gunthorpe * device ops, this is the only reason these actions are not done during 1237921eab11SJason Gunthorpe * ib_alloc_device. It is undone by ib_dealloc_device(). 1238921eab11SJason Gunthorpe */ 1239548cb4fbSParav Pandit static int setup_device(struct ib_device *device) 1240548cb4fbSParav Pandit { 1241548cb4fbSParav Pandit struct ib_udata uhw = {.outlen = 0, .inlen = 0}; 1242548cb4fbSParav Pandit int ret; 1243548cb4fbSParav Pandit 1244921eab11SJason Gunthorpe setup_dma_device(device); 1245deee3c7eSKamal Heib ib_device_check_mandatory(device); 1246548cb4fbSParav Pandit 12478ceb1357SJason Gunthorpe ret = setup_port_data(device); 1248548cb4fbSParav Pandit if (ret) { 12498ceb1357SJason Gunthorpe dev_warn(&device->dev, "Couldn't create per-port data\n"); 1250548cb4fbSParav Pandit return ret; 1251548cb4fbSParav Pandit } 1252548cb4fbSParav Pandit 1253548cb4fbSParav Pandit memset(&device->attrs, 0, sizeof(device->attrs)); 12543023a1e9SKamal Heib ret = device->ops.query_device(device, &device->attrs, &uhw); 1255548cb4fbSParav Pandit if (ret) { 1256548cb4fbSParav Pandit dev_warn(&device->dev, 1257548cb4fbSParav Pandit "Couldn't query the device attributes\n"); 1258d45f89d5SJason Gunthorpe return ret; 1259548cb4fbSParav Pandit } 1260548cb4fbSParav Pandit 1261548cb4fbSParav Pandit return 0; 1262548cb4fbSParav Pandit } 1263548cb4fbSParav Pandit 1264921eab11SJason Gunthorpe static void disable_device(struct ib_device *device) 1265921eab11SJason Gunthorpe { 12669cd58817SJason Gunthorpe u32 cid; 1267921eab11SJason Gunthorpe 1268921eab11SJason Gunthorpe WARN_ON(!refcount_read(&device->refcount)); 1269921eab11SJason Gunthorpe 1270921eab11SJason Gunthorpe down_write(&devices_rwsem); 1271921eab11SJason Gunthorpe xa_clear_mark(&devices, device->index, DEVICE_REGISTERED); 1272921eab11SJason Gunthorpe up_write(&devices_rwsem); 1273921eab11SJason Gunthorpe 12749cd58817SJason Gunthorpe /* 12759cd58817SJason Gunthorpe * Remove clients in LIFO order, see assign_client_id. This could be 12769cd58817SJason Gunthorpe * more efficient if xarray learns to reverse iterate. Since no new 12779cd58817SJason Gunthorpe * clients can be added to this ib_device past this point we only need 12789cd58817SJason Gunthorpe * the maximum possible client_id value here. 12799cd58817SJason Gunthorpe */ 1280921eab11SJason Gunthorpe down_read(&clients_rwsem); 12819cd58817SJason Gunthorpe cid = highest_client_id; 1282921eab11SJason Gunthorpe up_read(&clients_rwsem); 12839cd58817SJason Gunthorpe while (cid) { 12849cd58817SJason Gunthorpe cid--; 12859cd58817SJason Gunthorpe remove_client_context(device, cid); 12869cd58817SJason Gunthorpe } 1287921eab11SJason Gunthorpe 1288921eab11SJason Gunthorpe /* Pairs with refcount_set in enable_device */ 1289921eab11SJason Gunthorpe ib_device_put(device); 1290921eab11SJason Gunthorpe wait_for_completion(&device->unreg_completion); 1291c2261dd7SJason Gunthorpe 12924e0f7b90SParav Pandit /* 12934e0f7b90SParav Pandit * compat devices must be removed after device refcount drops to zero. 12944e0f7b90SParav Pandit * Otherwise init_net() may add more compatdevs after removing compat 12954e0f7b90SParav Pandit * devices and before device is disabled. 12964e0f7b90SParav Pandit */ 12974e0f7b90SParav Pandit remove_compat_devs(device); 1298921eab11SJason Gunthorpe } 1299921eab11SJason Gunthorpe 1300921eab11SJason Gunthorpe /* 1301921eab11SJason Gunthorpe * An enabled device is visible to all clients and to all the public facing 1302d0899892SJason Gunthorpe * APIs that return a device pointer. This always returns with a new get, even 1303d0899892SJason Gunthorpe * if it fails. 1304921eab11SJason Gunthorpe */ 1305d0899892SJason Gunthorpe static int enable_device_and_get(struct ib_device *device) 1306921eab11SJason Gunthorpe { 1307921eab11SJason Gunthorpe struct ib_client *client; 1308921eab11SJason Gunthorpe unsigned long index; 1309d0899892SJason Gunthorpe int ret = 0; 1310921eab11SJason Gunthorpe 1311d0899892SJason Gunthorpe /* 1312d0899892SJason Gunthorpe * One ref belongs to the xa and the other belongs to this 1313d0899892SJason Gunthorpe * thread. This is needed to guard against parallel unregistration. 1314d0899892SJason Gunthorpe */ 1315d0899892SJason Gunthorpe refcount_set(&device->refcount, 2); 1316921eab11SJason Gunthorpe down_write(&devices_rwsem); 1317921eab11SJason Gunthorpe xa_set_mark(&devices, device->index, DEVICE_REGISTERED); 1318d0899892SJason Gunthorpe 1319d0899892SJason Gunthorpe /* 1320d0899892SJason Gunthorpe * By using downgrade_write() we ensure that no other thread can clear 1321d0899892SJason Gunthorpe * DEVICE_REGISTERED while we are completing the client setup. 1322d0899892SJason Gunthorpe */ 1323d0899892SJason Gunthorpe downgrade_write(&devices_rwsem); 1324921eab11SJason Gunthorpe 1325ca22354bSJason Gunthorpe if (device->ops.enable_driver) { 1326ca22354bSJason Gunthorpe ret = device->ops.enable_driver(device); 1327ca22354bSJason Gunthorpe if (ret) 1328ca22354bSJason Gunthorpe goto out; 1329ca22354bSJason Gunthorpe } 1330ca22354bSJason Gunthorpe 1331921eab11SJason Gunthorpe down_read(&clients_rwsem); 1332921eab11SJason Gunthorpe xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { 1333921eab11SJason Gunthorpe ret = add_client_context(device, client); 1334d0899892SJason Gunthorpe if (ret) 1335d0899892SJason Gunthorpe break; 1336d0899892SJason Gunthorpe } 1337921eab11SJason Gunthorpe up_read(&clients_rwsem); 13384e0f7b90SParav Pandit if (!ret) 13394e0f7b90SParav Pandit ret = add_compat_devs(device); 1340ca22354bSJason Gunthorpe out: 1341d0899892SJason Gunthorpe up_read(&devices_rwsem); 1342921eab11SJason Gunthorpe return ret; 1343921eab11SJason Gunthorpe } 1344921eab11SJason Gunthorpe 13450cb42c02SJason Gunthorpe static void prevent_dealloc_device(struct ib_device *ib_dev) 13460cb42c02SJason Gunthorpe { 13470cb42c02SJason Gunthorpe } 13480cb42c02SJason Gunthorpe 1349548cb4fbSParav Pandit /** 1350548cb4fbSParav Pandit * ib_register_device - Register an IB device with IB core 1351548cb4fbSParav Pandit * @device: Device to register 1352d6537c1aSrd.dunlab@gmail.com * @name: unique string device name. This may include a '%' which will 1353d6537c1aSrd.dunlab@gmail.com * cause a unique index to be added to the passed device name. 1354548cb4fbSParav Pandit * 1355548cb4fbSParav Pandit * Low-level drivers use ib_register_device() to register their 1356548cb4fbSParav Pandit * devices with the IB core. All registered clients will receive a 1357548cb4fbSParav Pandit * callback for each device that is added. @device must be allocated 1358548cb4fbSParav Pandit * with ib_alloc_device(). 1359d0899892SJason Gunthorpe * 1360d0899892SJason Gunthorpe * If the driver uses ops.dealloc_driver and calls any ib_unregister_device() 1361d0899892SJason Gunthorpe * asynchronously then the device pointer may become freed as soon as this 1362d0899892SJason Gunthorpe * function returns. 1363548cb4fbSParav Pandit */ 1364ea4baf7fSParav Pandit int ib_register_device(struct ib_device *device, const char *name) 1365548cb4fbSParav Pandit { 1366548cb4fbSParav Pandit int ret; 13671da177e4SLinus Torvalds 13680df91bb6SJason Gunthorpe ret = assign_name(device, name); 1369e349f858SJason Gunthorpe if (ret) 1370921eab11SJason Gunthorpe return ret; 13711da177e4SLinus Torvalds 1372548cb4fbSParav Pandit ret = setup_device(device); 1373548cb4fbSParav Pandit if (ret) 1374d0899892SJason Gunthorpe return ret; 137503db3a2dSMatan Barak 1376d45f89d5SJason Gunthorpe ret = ib_cache_setup_one(device); 1377d45f89d5SJason Gunthorpe if (ret) { 1378d45f89d5SJason Gunthorpe dev_warn(&device->dev, 1379d45f89d5SJason Gunthorpe "Couldn't set up InfiniBand P_Key/GID cache\n"); 1380d0899892SJason Gunthorpe return ret; 1381d45f89d5SJason Gunthorpe } 1382d45f89d5SJason Gunthorpe 13837527a7b1SParav Pandit ib_device_register_rdmacg(device); 13843e153a93SIra Weiny 1385413d3347SMark Zhang rdma_counter_init(device); 1386413d3347SMark Zhang 1387e7a5b4aaSLeon Romanovsky /* 1388e7a5b4aaSLeon Romanovsky * Ensure that ADD uevent is not fired because it 1389e7a5b4aaSLeon Romanovsky * is too early amd device is not initialized yet. 1390e7a5b4aaSLeon Romanovsky */ 1391e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, true); 13925f8f5499SParav Pandit ret = device_add(&device->dev); 13935f8f5499SParav Pandit if (ret) 13945f8f5499SParav Pandit goto cg_cleanup; 13955f8f5499SParav Pandit 1396ea4baf7fSParav Pandit ret = ib_device_register_sysfs(device); 13971da177e4SLinus Torvalds if (ret) { 139843c7c851SJason Gunthorpe dev_warn(&device->dev, 139943c7c851SJason Gunthorpe "Couldn't register device with driver model\n"); 14005f8f5499SParav Pandit goto dev_cleanup; 14011da177e4SLinus Torvalds } 14021da177e4SLinus Torvalds 1403c7ff819aSYamin Friedman ib_cq_pool_init(device); 1404d0899892SJason Gunthorpe ret = enable_device_and_get(device); 1405e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, false); 1406e7a5b4aaSLeon Romanovsky /* Mark for userspace that device is ready */ 1407e7a5b4aaSLeon Romanovsky kobject_uevent(&device->dev.kobj, KOBJ_ADD); 1408d0899892SJason Gunthorpe if (ret) { 1409d0899892SJason Gunthorpe void (*dealloc_fn)(struct ib_device *); 1410d0899892SJason Gunthorpe 1411d0899892SJason Gunthorpe /* 1412d0899892SJason Gunthorpe * If we hit this error flow then we don't want to 1413d0899892SJason Gunthorpe * automatically dealloc the device since the caller is 1414d0899892SJason Gunthorpe * expected to call ib_dealloc_device() after 1415d0899892SJason Gunthorpe * ib_register_device() fails. This is tricky due to the 1416d0899892SJason Gunthorpe * possibility for a parallel unregistration along with this 1417d0899892SJason Gunthorpe * error flow. Since we have a refcount here we know any 1418d0899892SJason Gunthorpe * parallel flow is stopped in disable_device and will see the 14190cb42c02SJason Gunthorpe * special dealloc_driver pointer, causing the responsibility to 1420d0899892SJason Gunthorpe * ib_dealloc_device() to revert back to this thread. 1421d0899892SJason Gunthorpe */ 1422d0899892SJason Gunthorpe dealloc_fn = device->ops.dealloc_driver; 14230cb42c02SJason Gunthorpe device->ops.dealloc_driver = prevent_dealloc_device; 1424d0899892SJason Gunthorpe ib_device_put(device); 1425d0899892SJason Gunthorpe __ib_unregister_device(device); 1426d0899892SJason Gunthorpe device->ops.dealloc_driver = dealloc_fn; 1427d0899892SJason Gunthorpe return ret; 1428d0899892SJason Gunthorpe } 1429d0899892SJason Gunthorpe ib_device_put(device); 14301da177e4SLinus Torvalds 14314be3a4faSParav Pandit return 0; 14324be3a4faSParav Pandit 14335f8f5499SParav Pandit dev_cleanup: 14345f8f5499SParav Pandit device_del(&device->dev); 14352fb4f4eaSParav Pandit cg_cleanup: 1436e7a5b4aaSLeon Romanovsky dev_set_uevent_suppress(&device->dev, false); 14372fb4f4eaSParav Pandit ib_device_unregister_rdmacg(device); 1438d45f89d5SJason Gunthorpe ib_cache_cleanup_one(device); 14391da177e4SLinus Torvalds return ret; 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_device); 14421da177e4SLinus Torvalds 1443d0899892SJason Gunthorpe /* Callers must hold a get on the device. */ 1444d0899892SJason Gunthorpe static void __ib_unregister_device(struct ib_device *ib_dev) 1445d0899892SJason Gunthorpe { 1446d0899892SJason Gunthorpe /* 1447d0899892SJason Gunthorpe * We have a registration lock so that all the calls to unregister are 1448d0899892SJason Gunthorpe * fully fenced, once any unregister returns the device is truely 1449d0899892SJason Gunthorpe * unregistered even if multiple callers are unregistering it at the 1450d0899892SJason Gunthorpe * same time. This also interacts with the registration flow and 1451d0899892SJason Gunthorpe * provides sane semantics if register and unregister are racing. 1452d0899892SJason Gunthorpe */ 1453d0899892SJason Gunthorpe mutex_lock(&ib_dev->unregistration_lock); 1454d0899892SJason Gunthorpe if (!refcount_read(&ib_dev->refcount)) 1455d0899892SJason Gunthorpe goto out; 1456d0899892SJason Gunthorpe 1457d0899892SJason Gunthorpe disable_device(ib_dev); 1458c7ff819aSYamin Friedman ib_cq_pool_destroy(ib_dev); 14593042492bSParav Pandit 14603042492bSParav Pandit /* Expedite removing unregistered pointers from the hash table */ 14613042492bSParav Pandit free_netdevs(ib_dev); 14623042492bSParav Pandit 1463d0899892SJason Gunthorpe ib_device_unregister_sysfs(ib_dev); 1464d0899892SJason Gunthorpe device_del(&ib_dev->dev); 1465d0899892SJason Gunthorpe ib_device_unregister_rdmacg(ib_dev); 1466d0899892SJason Gunthorpe ib_cache_cleanup_one(ib_dev); 1467d0899892SJason Gunthorpe 1468d0899892SJason Gunthorpe /* 1469d0899892SJason Gunthorpe * Drivers using the new flow may not call ib_dealloc_device except 1470d0899892SJason Gunthorpe * in error unwind prior to registration success. 1471d0899892SJason Gunthorpe */ 14720cb42c02SJason Gunthorpe if (ib_dev->ops.dealloc_driver && 14730cb42c02SJason Gunthorpe ib_dev->ops.dealloc_driver != prevent_dealloc_device) { 1474d0899892SJason Gunthorpe WARN_ON(kref_read(&ib_dev->dev.kobj.kref) <= 1); 1475d0899892SJason Gunthorpe ib_dealloc_device(ib_dev); 1476d0899892SJason Gunthorpe } 1477d0899892SJason Gunthorpe out: 1478d0899892SJason Gunthorpe mutex_unlock(&ib_dev->unregistration_lock); 1479d0899892SJason Gunthorpe } 1480d0899892SJason Gunthorpe 14811da177e4SLinus Torvalds /** 14821da177e4SLinus Torvalds * ib_unregister_device - Unregister an IB device 1483d6537c1aSrd.dunlab@gmail.com * @ib_dev: The device to unregister 14841da177e4SLinus Torvalds * 14851da177e4SLinus Torvalds * Unregister an IB device. All clients will receive a remove callback. 1486d0899892SJason Gunthorpe * 1487d0899892SJason Gunthorpe * Callers should call this routine only once, and protect against races with 1488d0899892SJason Gunthorpe * registration. Typically it should only be called as part of a remove 1489d0899892SJason Gunthorpe * callback in an implementation of driver core's struct device_driver and 1490d0899892SJason Gunthorpe * related. 1491d0899892SJason Gunthorpe * 1492d0899892SJason Gunthorpe * If ops.dealloc_driver is used then ib_dev will be freed upon return from 1493d0899892SJason Gunthorpe * this function. 14941da177e4SLinus Torvalds */ 1495d0899892SJason Gunthorpe void ib_unregister_device(struct ib_device *ib_dev) 14961da177e4SLinus Torvalds { 1497d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1498d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1499d0899892SJason Gunthorpe put_device(&ib_dev->dev); 15001da177e4SLinus Torvalds } 15011da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_device); 15021da177e4SLinus Torvalds 1503d0899892SJason Gunthorpe /** 1504d0899892SJason Gunthorpe * ib_unregister_device_and_put - Unregister a device while holding a 'get' 1505d6537c1aSrd.dunlab@gmail.com * @ib_dev: The device to unregister 1506d0899892SJason Gunthorpe * 1507d0899892SJason Gunthorpe * This is the same as ib_unregister_device(), except it includes an internal 1508d0899892SJason Gunthorpe * ib_device_put() that should match a 'get' obtained by the caller. 1509d0899892SJason Gunthorpe * 1510d0899892SJason Gunthorpe * It is safe to call this routine concurrently from multiple threads while 1511d0899892SJason Gunthorpe * holding the 'get'. When the function returns the device is fully 1512d0899892SJason Gunthorpe * unregistered. 1513d0899892SJason Gunthorpe * 1514d0899892SJason Gunthorpe * Drivers using this flow MUST use the driver_unregister callback to clean up 1515d0899892SJason Gunthorpe * their resources associated with the device and dealloc it. 1516d0899892SJason Gunthorpe */ 1517d0899892SJason Gunthorpe void ib_unregister_device_and_put(struct ib_device *ib_dev) 1518d0899892SJason Gunthorpe { 1519d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1520d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1521d0899892SJason Gunthorpe ib_device_put(ib_dev); 1522d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1523d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1524d0899892SJason Gunthorpe } 1525d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_and_put); 1526d0899892SJason Gunthorpe 1527d0899892SJason Gunthorpe /** 1528d0899892SJason Gunthorpe * ib_unregister_driver - Unregister all IB devices for a driver 1529d0899892SJason Gunthorpe * @driver_id: The driver to unregister 1530d0899892SJason Gunthorpe * 1531d0899892SJason Gunthorpe * This implements a fence for device unregistration. It only returns once all 1532d0899892SJason Gunthorpe * devices associated with the driver_id have fully completed their 1533d0899892SJason Gunthorpe * unregistration and returned from ib_unregister_device*(). 1534d0899892SJason Gunthorpe * 1535d0899892SJason Gunthorpe * If device's are not yet unregistered it goes ahead and starts unregistering 1536d0899892SJason Gunthorpe * them. 1537d0899892SJason Gunthorpe * 1538d0899892SJason Gunthorpe * This does not block creation of new devices with the given driver_id, that 1539d0899892SJason Gunthorpe * is the responsibility of the caller. 1540d0899892SJason Gunthorpe */ 1541d0899892SJason Gunthorpe void ib_unregister_driver(enum rdma_driver_id driver_id) 1542d0899892SJason Gunthorpe { 1543d0899892SJason Gunthorpe struct ib_device *ib_dev; 1544d0899892SJason Gunthorpe unsigned long index; 1545d0899892SJason Gunthorpe 1546d0899892SJason Gunthorpe down_read(&devices_rwsem); 1547d0899892SJason Gunthorpe xa_for_each (&devices, index, ib_dev) { 1548b9560a41SJason Gunthorpe if (ib_dev->ops.driver_id != driver_id) 1549d0899892SJason Gunthorpe continue; 1550d0899892SJason Gunthorpe 1551d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1552d0899892SJason Gunthorpe up_read(&devices_rwsem); 1553d0899892SJason Gunthorpe 1554d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1555d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1556d0899892SJason Gunthorpe 1557d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1558d0899892SJason Gunthorpe down_read(&devices_rwsem); 1559d0899892SJason Gunthorpe } 1560d0899892SJason Gunthorpe up_read(&devices_rwsem); 1561d0899892SJason Gunthorpe } 1562d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_driver); 1563d0899892SJason Gunthorpe 1564d0899892SJason Gunthorpe static void ib_unregister_work(struct work_struct *work) 1565d0899892SJason Gunthorpe { 1566d0899892SJason Gunthorpe struct ib_device *ib_dev = 1567d0899892SJason Gunthorpe container_of(work, struct ib_device, unregistration_work); 1568d0899892SJason Gunthorpe 1569d0899892SJason Gunthorpe __ib_unregister_device(ib_dev); 1570d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1571d0899892SJason Gunthorpe } 1572d0899892SJason Gunthorpe 1573d0899892SJason Gunthorpe /** 1574d0899892SJason Gunthorpe * ib_unregister_device_queued - Unregister a device using a work queue 1575d6537c1aSrd.dunlab@gmail.com * @ib_dev: The device to unregister 1576d0899892SJason Gunthorpe * 1577d0899892SJason Gunthorpe * This schedules an asynchronous unregistration using a WQ for the device. A 1578d0899892SJason Gunthorpe * driver should use this to avoid holding locks while doing unregistration, 1579d0899892SJason Gunthorpe * such as holding the RTNL lock. 1580d0899892SJason Gunthorpe * 1581d0899892SJason Gunthorpe * Drivers using this API must use ib_unregister_driver before module unload 1582d0899892SJason Gunthorpe * to ensure that all scheduled unregistrations have completed. 1583d0899892SJason Gunthorpe */ 1584d0899892SJason Gunthorpe void ib_unregister_device_queued(struct ib_device *ib_dev) 1585d0899892SJason Gunthorpe { 1586d0899892SJason Gunthorpe WARN_ON(!refcount_read(&ib_dev->refcount)); 1587d0899892SJason Gunthorpe WARN_ON(!ib_dev->ops.dealloc_driver); 1588d0899892SJason Gunthorpe get_device(&ib_dev->dev); 1589d0899892SJason Gunthorpe if (!queue_work(system_unbound_wq, &ib_dev->unregistration_work)) 1590d0899892SJason Gunthorpe put_device(&ib_dev->dev); 1591d0899892SJason Gunthorpe } 1592d0899892SJason Gunthorpe EXPORT_SYMBOL(ib_unregister_device_queued); 1593d0899892SJason Gunthorpe 1594decbc7a6SParav Pandit /* 1595decbc7a6SParav Pandit * The caller must pass in a device that has the kref held and the refcount 1596decbc7a6SParav Pandit * released. If the device is in cur_net and still registered then it is moved 1597decbc7a6SParav Pandit * into net. 1598decbc7a6SParav Pandit */ 1599decbc7a6SParav Pandit static int rdma_dev_change_netns(struct ib_device *device, struct net *cur_net, 1600decbc7a6SParav Pandit struct net *net) 1601decbc7a6SParav Pandit { 1602decbc7a6SParav Pandit int ret2 = -EINVAL; 1603decbc7a6SParav Pandit int ret; 1604decbc7a6SParav Pandit 1605decbc7a6SParav Pandit mutex_lock(&device->unregistration_lock); 1606decbc7a6SParav Pandit 1607decbc7a6SParav Pandit /* 16082e5b8a01SParav Pandit * If a device not under ib_device_get() or if the unregistration_lock 16092e5b8a01SParav Pandit * is not held, the namespace can be changed, or it can be unregistered. 16102e5b8a01SParav Pandit * Check again under the lock. 1611decbc7a6SParav Pandit */ 1612decbc7a6SParav Pandit if (refcount_read(&device->refcount) == 0 || 1613decbc7a6SParav Pandit !net_eq(cur_net, read_pnet(&device->coredev.rdma_net))) { 1614decbc7a6SParav Pandit ret = -ENODEV; 1615decbc7a6SParav Pandit goto out; 1616decbc7a6SParav Pandit } 1617decbc7a6SParav Pandit 1618decbc7a6SParav Pandit kobject_uevent(&device->dev.kobj, KOBJ_REMOVE); 1619decbc7a6SParav Pandit disable_device(device); 1620decbc7a6SParav Pandit 1621decbc7a6SParav Pandit /* 1622decbc7a6SParav Pandit * At this point no one can be using the device, so it is safe to 1623decbc7a6SParav Pandit * change the namespace. 1624decbc7a6SParav Pandit */ 1625decbc7a6SParav Pandit write_pnet(&device->coredev.rdma_net, net); 1626decbc7a6SParav Pandit 16272e5b8a01SParav Pandit down_read(&devices_rwsem); 1628decbc7a6SParav Pandit /* 1629decbc7a6SParav Pandit * Currently rdma devices are system wide unique. So the device name 1630decbc7a6SParav Pandit * is guaranteed free in the new namespace. Publish the new namespace 1631decbc7a6SParav Pandit * at the sysfs level. 1632decbc7a6SParav Pandit */ 1633decbc7a6SParav Pandit ret = device_rename(&device->dev, dev_name(&device->dev)); 1634decbc7a6SParav Pandit up_read(&devices_rwsem); 1635decbc7a6SParav Pandit if (ret) { 1636decbc7a6SParav Pandit dev_warn(&device->dev, 1637decbc7a6SParav Pandit "%s: Couldn't rename device after namespace change\n", 1638decbc7a6SParav Pandit __func__); 1639decbc7a6SParav Pandit /* Try and put things back and re-enable the device */ 1640decbc7a6SParav Pandit write_pnet(&device->coredev.rdma_net, cur_net); 1641decbc7a6SParav Pandit } 1642decbc7a6SParav Pandit 1643decbc7a6SParav Pandit ret2 = enable_device_and_get(device); 16442e5b8a01SParav Pandit if (ret2) { 1645decbc7a6SParav Pandit /* 1646decbc7a6SParav Pandit * This shouldn't really happen, but if it does, let the user 1647decbc7a6SParav Pandit * retry at later point. So don't disable the device. 1648decbc7a6SParav Pandit */ 1649decbc7a6SParav Pandit dev_warn(&device->dev, 1650decbc7a6SParav Pandit "%s: Couldn't re-enable device after namespace change\n", 1651decbc7a6SParav Pandit __func__); 16522e5b8a01SParav Pandit } 1653decbc7a6SParav Pandit kobject_uevent(&device->dev.kobj, KOBJ_ADD); 16542e5b8a01SParav Pandit 1655decbc7a6SParav Pandit ib_device_put(device); 1656decbc7a6SParav Pandit out: 1657decbc7a6SParav Pandit mutex_unlock(&device->unregistration_lock); 1658decbc7a6SParav Pandit if (ret) 1659decbc7a6SParav Pandit return ret; 1660decbc7a6SParav Pandit return ret2; 1661decbc7a6SParav Pandit } 1662decbc7a6SParav Pandit 16632e5b8a01SParav Pandit int ib_device_set_netns_put(struct sk_buff *skb, 16642e5b8a01SParav Pandit struct ib_device *dev, u32 ns_fd) 16652e5b8a01SParav Pandit { 16662e5b8a01SParav Pandit struct net *net; 16672e5b8a01SParav Pandit int ret; 16682e5b8a01SParav Pandit 16692e5b8a01SParav Pandit net = get_net_ns_by_fd(ns_fd); 16702e5b8a01SParav Pandit if (IS_ERR(net)) { 16712e5b8a01SParav Pandit ret = PTR_ERR(net); 16722e5b8a01SParav Pandit goto net_err; 16732e5b8a01SParav Pandit } 16742e5b8a01SParav Pandit 16752e5b8a01SParav Pandit if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN)) { 16762e5b8a01SParav Pandit ret = -EPERM; 16772e5b8a01SParav Pandit goto ns_err; 16782e5b8a01SParav Pandit } 16792e5b8a01SParav Pandit 16802e5b8a01SParav Pandit /* 16812e5b8a01SParav Pandit * Currently supported only for those providers which support 16822e5b8a01SParav Pandit * disassociation and don't do port specific sysfs init. Once a 16832e5b8a01SParav Pandit * port_cleanup infrastructure is implemented, this limitation will be 16842e5b8a01SParav Pandit * removed. 16852e5b8a01SParav Pandit */ 16862e5b8a01SParav Pandit if (!dev->ops.disassociate_ucontext || dev->ops.init_port || 16872e5b8a01SParav Pandit ib_devices_shared_netns) { 16882e5b8a01SParav Pandit ret = -EOPNOTSUPP; 16892e5b8a01SParav Pandit goto ns_err; 16902e5b8a01SParav Pandit } 16912e5b8a01SParav Pandit 16922e5b8a01SParav Pandit get_device(&dev->dev); 16932e5b8a01SParav Pandit ib_device_put(dev); 16942e5b8a01SParav Pandit ret = rdma_dev_change_netns(dev, current->nsproxy->net_ns, net); 16952e5b8a01SParav Pandit put_device(&dev->dev); 16962e5b8a01SParav Pandit 16972e5b8a01SParav Pandit put_net(net); 16982e5b8a01SParav Pandit return ret; 16992e5b8a01SParav Pandit 17002e5b8a01SParav Pandit ns_err: 17012e5b8a01SParav Pandit put_net(net); 17022e5b8a01SParav Pandit net_err: 17032e5b8a01SParav Pandit ib_device_put(dev); 17042e5b8a01SParav Pandit return ret; 17052e5b8a01SParav Pandit } 17062e5b8a01SParav Pandit 17074e0f7b90SParav Pandit static struct pernet_operations rdma_dev_net_ops = { 17084e0f7b90SParav Pandit .init = rdma_dev_init_net, 17094e0f7b90SParav Pandit .exit = rdma_dev_exit_net, 17104e0f7b90SParav Pandit .id = &rdma_dev_net_id, 17114e0f7b90SParav Pandit .size = sizeof(struct rdma_dev_net), 17124e0f7b90SParav Pandit }; 17134e0f7b90SParav Pandit 1714e59178d8SJason Gunthorpe static int assign_client_id(struct ib_client *client) 1715e59178d8SJason Gunthorpe { 1716e59178d8SJason Gunthorpe int ret; 1717e59178d8SJason Gunthorpe 1718921eab11SJason Gunthorpe down_write(&clients_rwsem); 1719e59178d8SJason Gunthorpe /* 1720e59178d8SJason Gunthorpe * The add/remove callbacks must be called in FIFO/LIFO order. To 1721e59178d8SJason Gunthorpe * achieve this we assign client_ids so they are sorted in 17229cd58817SJason Gunthorpe * registration order. 1723e59178d8SJason Gunthorpe */ 17249cd58817SJason Gunthorpe client->client_id = highest_client_id; 1725ea295481SLinus Torvalds ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL); 1726e59178d8SJason Gunthorpe if (ret) 1727e59178d8SJason Gunthorpe goto out; 1728e59178d8SJason Gunthorpe 17299cd58817SJason Gunthorpe highest_client_id++; 1730921eab11SJason Gunthorpe xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED); 1731921eab11SJason Gunthorpe 1732e59178d8SJason Gunthorpe out: 1733921eab11SJason Gunthorpe up_write(&clients_rwsem); 1734e59178d8SJason Gunthorpe return ret; 1735e59178d8SJason Gunthorpe } 1736e59178d8SJason Gunthorpe 17379cd58817SJason Gunthorpe static void remove_client_id(struct ib_client *client) 17389cd58817SJason Gunthorpe { 17399cd58817SJason Gunthorpe down_write(&clients_rwsem); 17409cd58817SJason Gunthorpe xa_erase(&clients, client->client_id); 17419cd58817SJason Gunthorpe for (; highest_client_id; highest_client_id--) 17429cd58817SJason Gunthorpe if (xa_load(&clients, highest_client_id - 1)) 17439cd58817SJason Gunthorpe break; 17449cd58817SJason Gunthorpe up_write(&clients_rwsem); 17459cd58817SJason Gunthorpe } 17469cd58817SJason Gunthorpe 17471da177e4SLinus Torvalds /** 17481da177e4SLinus Torvalds * ib_register_client - Register an IB client 17491da177e4SLinus Torvalds * @client:Client to register 17501da177e4SLinus Torvalds * 17511da177e4SLinus Torvalds * Upper level users of the IB drivers can use ib_register_client() to 17521da177e4SLinus Torvalds * register callbacks for IB device addition and removal. When an IB 17531da177e4SLinus Torvalds * device is added, each registered client's add method will be called 17541da177e4SLinus Torvalds * (in the order the clients were registered), and when a device is 17551da177e4SLinus Torvalds * removed, each client's remove method will be called (in the reverse 17561da177e4SLinus Torvalds * order that clients were registered). In addition, when 17571da177e4SLinus Torvalds * ib_register_client() is called, the client will receive an add 17581da177e4SLinus Torvalds * callback for all devices already registered. 17591da177e4SLinus Torvalds */ 17601da177e4SLinus Torvalds int ib_register_client(struct ib_client *client) 17611da177e4SLinus Torvalds { 17621da177e4SLinus Torvalds struct ib_device *device; 17630df91bb6SJason Gunthorpe unsigned long index; 1764e59178d8SJason Gunthorpe int ret; 17651da177e4SLinus Torvalds 1766621e55ffSJason Gunthorpe refcount_set(&client->uses, 1); 1767621e55ffSJason Gunthorpe init_completion(&client->uses_zero); 1768e59178d8SJason Gunthorpe ret = assign_client_id(client); 1769921eab11SJason Gunthorpe if (ret) 1770921eab11SJason Gunthorpe return ret; 1771921eab11SJason Gunthorpe 1772921eab11SJason Gunthorpe down_read(&devices_rwsem); 1773921eab11SJason Gunthorpe xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) { 1774921eab11SJason Gunthorpe ret = add_client_context(device, client); 1775e59178d8SJason Gunthorpe if (ret) { 1776921eab11SJason Gunthorpe up_read(&devices_rwsem); 1777921eab11SJason Gunthorpe ib_unregister_client(client); 1778e59178d8SJason Gunthorpe return ret; 1779e59178d8SJason Gunthorpe } 1780921eab11SJason Gunthorpe } 1781921eab11SJason Gunthorpe up_read(&devices_rwsem); 17821da177e4SLinus Torvalds return 0; 17831da177e4SLinus Torvalds } 17841da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_client); 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds /** 17871da177e4SLinus Torvalds * ib_unregister_client - Unregister an IB client 17881da177e4SLinus Torvalds * @client:Client to unregister 17891da177e4SLinus Torvalds * 17901da177e4SLinus Torvalds * Upper level users use ib_unregister_client() to remove their client 17911da177e4SLinus Torvalds * registration. When ib_unregister_client() is called, the client 17921da177e4SLinus Torvalds * will receive a remove callback for each IB device still registered. 1793921eab11SJason Gunthorpe * 1794921eab11SJason Gunthorpe * This is a full fence, once it returns no client callbacks will be called, 1795921eab11SJason Gunthorpe * or are running in another thread. 17961da177e4SLinus Torvalds */ 17971da177e4SLinus Torvalds void ib_unregister_client(struct ib_client *client) 17981da177e4SLinus Torvalds { 17991da177e4SLinus Torvalds struct ib_device *device; 18000df91bb6SJason Gunthorpe unsigned long index; 18011da177e4SLinus Torvalds 1802921eab11SJason Gunthorpe down_write(&clients_rwsem); 1803621e55ffSJason Gunthorpe ib_client_put(client); 1804e59178d8SJason Gunthorpe xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED); 1805921eab11SJason Gunthorpe up_write(&clients_rwsem); 18065aa44bb9SHaggai Eran 1807621e55ffSJason Gunthorpe /* We do not want to have locks while calling client->remove() */ 1808621e55ffSJason Gunthorpe rcu_read_lock(); 1809621e55ffSJason Gunthorpe xa_for_each (&devices, index, device) { 1810621e55ffSJason Gunthorpe if (!ib_device_try_get(device)) 1811621e55ffSJason Gunthorpe continue; 1812621e55ffSJason Gunthorpe rcu_read_unlock(); 1813621e55ffSJason Gunthorpe 18141da177e4SLinus Torvalds remove_client_context(device, client->client_id); 1815621e55ffSJason Gunthorpe 1816621e55ffSJason Gunthorpe ib_device_put(device); 1817621e55ffSJason Gunthorpe rcu_read_lock(); 1818621e55ffSJason Gunthorpe } 1819621e55ffSJason Gunthorpe rcu_read_unlock(); 1820621e55ffSJason Gunthorpe 1821621e55ffSJason Gunthorpe /* 1822621e55ffSJason Gunthorpe * remove_client_context() is not a fence, it can return even though a 1823621e55ffSJason Gunthorpe * removal is ongoing. Wait until all removals are completed. 1824621e55ffSJason Gunthorpe */ 1825621e55ffSJason Gunthorpe wait_for_completion(&client->uses_zero); 18269cd58817SJason Gunthorpe remove_client_id(client); 18271da177e4SLinus Torvalds } 18281da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_client); 18291da177e4SLinus Torvalds 18300e2d00ebSJason Gunthorpe static int __ib_get_global_client_nl_info(const char *client_name, 18310e2d00ebSJason Gunthorpe struct ib_client_nl_info *res) 18320e2d00ebSJason Gunthorpe { 18330e2d00ebSJason Gunthorpe struct ib_client *client; 18340e2d00ebSJason Gunthorpe unsigned long index; 18350e2d00ebSJason Gunthorpe int ret = -ENOENT; 18360e2d00ebSJason Gunthorpe 18370e2d00ebSJason Gunthorpe down_read(&clients_rwsem); 18380e2d00ebSJason Gunthorpe xa_for_each_marked (&clients, index, client, CLIENT_REGISTERED) { 18390e2d00ebSJason Gunthorpe if (strcmp(client->name, client_name) != 0) 18400e2d00ebSJason Gunthorpe continue; 18410e2d00ebSJason Gunthorpe if (!client->get_global_nl_info) { 18420e2d00ebSJason Gunthorpe ret = -EOPNOTSUPP; 18430e2d00ebSJason Gunthorpe break; 18440e2d00ebSJason Gunthorpe } 18450e2d00ebSJason Gunthorpe ret = client->get_global_nl_info(res); 18460e2d00ebSJason Gunthorpe if (WARN_ON(ret == -ENOENT)) 18470e2d00ebSJason Gunthorpe ret = -EINVAL; 18480e2d00ebSJason Gunthorpe if (!ret && res->cdev) 18490e2d00ebSJason Gunthorpe get_device(res->cdev); 18500e2d00ebSJason Gunthorpe break; 18510e2d00ebSJason Gunthorpe } 18520e2d00ebSJason Gunthorpe up_read(&clients_rwsem); 18530e2d00ebSJason Gunthorpe return ret; 18540e2d00ebSJason Gunthorpe } 18550e2d00ebSJason Gunthorpe 18560e2d00ebSJason Gunthorpe static int __ib_get_client_nl_info(struct ib_device *ibdev, 18570e2d00ebSJason Gunthorpe const char *client_name, 18580e2d00ebSJason Gunthorpe struct ib_client_nl_info *res) 18590e2d00ebSJason Gunthorpe { 18600e2d00ebSJason Gunthorpe unsigned long index; 18610e2d00ebSJason Gunthorpe void *client_data; 18620e2d00ebSJason Gunthorpe int ret = -ENOENT; 18630e2d00ebSJason Gunthorpe 18640e2d00ebSJason Gunthorpe down_read(&ibdev->client_data_rwsem); 18650e2d00ebSJason Gunthorpe xan_for_each_marked (&ibdev->client_data, index, client_data, 18660e2d00ebSJason Gunthorpe CLIENT_DATA_REGISTERED) { 18670e2d00ebSJason Gunthorpe struct ib_client *client = xa_load(&clients, index); 18680e2d00ebSJason Gunthorpe 18690e2d00ebSJason Gunthorpe if (!client || strcmp(client->name, client_name) != 0) 18700e2d00ebSJason Gunthorpe continue; 18710e2d00ebSJason Gunthorpe if (!client->get_nl_info) { 18720e2d00ebSJason Gunthorpe ret = -EOPNOTSUPP; 18730e2d00ebSJason Gunthorpe break; 18740e2d00ebSJason Gunthorpe } 18750e2d00ebSJason Gunthorpe ret = client->get_nl_info(ibdev, client_data, res); 18760e2d00ebSJason Gunthorpe if (WARN_ON(ret == -ENOENT)) 18770e2d00ebSJason Gunthorpe ret = -EINVAL; 18780e2d00ebSJason Gunthorpe 18790e2d00ebSJason Gunthorpe /* 18800e2d00ebSJason Gunthorpe * The cdev is guaranteed valid as long as we are inside the 18810e2d00ebSJason Gunthorpe * client_data_rwsem as remove_one can't be called. Keep it 18820e2d00ebSJason Gunthorpe * valid for the caller. 18830e2d00ebSJason Gunthorpe */ 18840e2d00ebSJason Gunthorpe if (!ret && res->cdev) 18850e2d00ebSJason Gunthorpe get_device(res->cdev); 18860e2d00ebSJason Gunthorpe break; 18870e2d00ebSJason Gunthorpe } 18880e2d00ebSJason Gunthorpe up_read(&ibdev->client_data_rwsem); 18890e2d00ebSJason Gunthorpe 18900e2d00ebSJason Gunthorpe return ret; 18910e2d00ebSJason Gunthorpe } 18920e2d00ebSJason Gunthorpe 18930e2d00ebSJason Gunthorpe /** 18940e2d00ebSJason Gunthorpe * ib_get_client_nl_info - Fetch the nl_info from a client 18950e2d00ebSJason Gunthorpe * @device - IB device 18960e2d00ebSJason Gunthorpe * @client_name - Name of the client 18970e2d00ebSJason Gunthorpe * @res - Result of the query 18980e2d00ebSJason Gunthorpe */ 18990e2d00ebSJason Gunthorpe int ib_get_client_nl_info(struct ib_device *ibdev, const char *client_name, 19000e2d00ebSJason Gunthorpe struct ib_client_nl_info *res) 19010e2d00ebSJason Gunthorpe { 19020e2d00ebSJason Gunthorpe int ret; 19030e2d00ebSJason Gunthorpe 19040e2d00ebSJason Gunthorpe if (ibdev) 19050e2d00ebSJason Gunthorpe ret = __ib_get_client_nl_info(ibdev, client_name, res); 19060e2d00ebSJason Gunthorpe else 19070e2d00ebSJason Gunthorpe ret = __ib_get_global_client_nl_info(client_name, res); 19080e2d00ebSJason Gunthorpe #ifdef CONFIG_MODULES 19090e2d00ebSJason Gunthorpe if (ret == -ENOENT) { 19100e2d00ebSJason Gunthorpe request_module("rdma-client-%s", client_name); 19110e2d00ebSJason Gunthorpe if (ibdev) 19120e2d00ebSJason Gunthorpe ret = __ib_get_client_nl_info(ibdev, client_name, res); 19130e2d00ebSJason Gunthorpe else 19140e2d00ebSJason Gunthorpe ret = __ib_get_global_client_nl_info(client_name, res); 19150e2d00ebSJason Gunthorpe } 19160e2d00ebSJason Gunthorpe #endif 19170e2d00ebSJason Gunthorpe if (ret) { 19180e2d00ebSJason Gunthorpe if (ret == -ENOENT) 19190e2d00ebSJason Gunthorpe return -EOPNOTSUPP; 19200e2d00ebSJason Gunthorpe return ret; 19210e2d00ebSJason Gunthorpe } 19220e2d00ebSJason Gunthorpe 19230e2d00ebSJason Gunthorpe if (WARN_ON(!res->cdev)) 19240e2d00ebSJason Gunthorpe return -EINVAL; 19250e2d00ebSJason Gunthorpe return 0; 19260e2d00ebSJason Gunthorpe } 19270e2d00ebSJason Gunthorpe 19281da177e4SLinus Torvalds /** 19299cd330d3SKrishna Kumar * ib_set_client_data - Set IB client context 19301da177e4SLinus Torvalds * @device:Device to set context for 19311da177e4SLinus Torvalds * @client:Client to set context for 19321da177e4SLinus Torvalds * @data:Context to set 19331da177e4SLinus Torvalds * 19340df91bb6SJason Gunthorpe * ib_set_client_data() sets client context data that can be retrieved with 19350df91bb6SJason Gunthorpe * ib_get_client_data(). This can only be called while the client is 19360df91bb6SJason Gunthorpe * registered to the device, once the ib_client remove() callback returns this 19370df91bb6SJason Gunthorpe * cannot be called. 19381da177e4SLinus Torvalds */ 19391da177e4SLinus Torvalds void ib_set_client_data(struct ib_device *device, struct ib_client *client, 19401da177e4SLinus Torvalds void *data) 19411da177e4SLinus Torvalds { 19420df91bb6SJason Gunthorpe void *rc; 19431da177e4SLinus Torvalds 19440df91bb6SJason Gunthorpe if (WARN_ON(IS_ERR(data))) 19450df91bb6SJason Gunthorpe data = NULL; 19461da177e4SLinus Torvalds 19470df91bb6SJason Gunthorpe rc = xa_store(&device->client_data, client->client_id, data, 19480df91bb6SJason Gunthorpe GFP_KERNEL); 19490df91bb6SJason Gunthorpe WARN_ON(xa_is_err(rc)); 19501da177e4SLinus Torvalds } 19511da177e4SLinus Torvalds EXPORT_SYMBOL(ib_set_client_data); 19521da177e4SLinus Torvalds 19531da177e4SLinus Torvalds /** 19541da177e4SLinus Torvalds * ib_register_event_handler - Register an IB event handler 19551da177e4SLinus Torvalds * @event_handler:Handler to register 19561da177e4SLinus Torvalds * 19571da177e4SLinus Torvalds * ib_register_event_handler() registers an event handler that will be 19581da177e4SLinus Torvalds * called back when asynchronous IB events occur (as defined in 19591da177e4SLinus Torvalds * chapter 11 of the InfiniBand Architecture Specification). This 19606b57cea9SParav Pandit * callback occurs in workqueue context. 19611da177e4SLinus Torvalds */ 1962dcc9881eSLeon Romanovsky void ib_register_event_handler(struct ib_event_handler *event_handler) 19631da177e4SLinus Torvalds { 19646b57cea9SParav Pandit down_write(&event_handler->device->event_handler_rwsem); 19651da177e4SLinus Torvalds list_add_tail(&event_handler->list, 19661da177e4SLinus Torvalds &event_handler->device->event_handler_list); 19676b57cea9SParav Pandit up_write(&event_handler->device->event_handler_rwsem); 19681da177e4SLinus Torvalds } 19691da177e4SLinus Torvalds EXPORT_SYMBOL(ib_register_event_handler); 19701da177e4SLinus Torvalds 19711da177e4SLinus Torvalds /** 19721da177e4SLinus Torvalds * ib_unregister_event_handler - Unregister an event handler 19731da177e4SLinus Torvalds * @event_handler:Handler to unregister 19741da177e4SLinus Torvalds * 19751da177e4SLinus Torvalds * Unregister an event handler registered with 19761da177e4SLinus Torvalds * ib_register_event_handler(). 19771da177e4SLinus Torvalds */ 1978dcc9881eSLeon Romanovsky void ib_unregister_event_handler(struct ib_event_handler *event_handler) 19791da177e4SLinus Torvalds { 19806b57cea9SParav Pandit down_write(&event_handler->device->event_handler_rwsem); 19811da177e4SLinus Torvalds list_del(&event_handler->list); 19826b57cea9SParav Pandit up_write(&event_handler->device->event_handler_rwsem); 19831da177e4SLinus Torvalds } 19841da177e4SLinus Torvalds EXPORT_SYMBOL(ib_unregister_event_handler); 19851da177e4SLinus Torvalds 19866b57cea9SParav Pandit void ib_dispatch_event_clients(struct ib_event *event) 19871da177e4SLinus Torvalds { 19881da177e4SLinus Torvalds struct ib_event_handler *handler; 19891da177e4SLinus Torvalds 19906b57cea9SParav Pandit down_read(&event->device->event_handler_rwsem); 19911da177e4SLinus Torvalds 19921da177e4SLinus Torvalds list_for_each_entry(handler, &event->device->event_handler_list, list) 19931da177e4SLinus Torvalds handler->handler(handler, event); 19941da177e4SLinus Torvalds 19956b57cea9SParav Pandit up_read(&event->device->event_handler_rwsem); 19961da177e4SLinus Torvalds } 19971da177e4SLinus Torvalds 19984929116bSKamal Heib static int iw_query_port(struct ib_device *device, 19994929116bSKamal Heib u8 port_num, 20004929116bSKamal Heib struct ib_port_attr *port_attr) 20014929116bSKamal Heib { 20024929116bSKamal Heib struct in_device *inetdev; 20034929116bSKamal Heib struct net_device *netdev; 20044929116bSKamal Heib 20054929116bSKamal Heib memset(port_attr, 0, sizeof(*port_attr)); 20064929116bSKamal Heib 20074929116bSKamal Heib netdev = ib_device_get_netdev(device, port_num); 20084929116bSKamal Heib if (!netdev) 20094929116bSKamal Heib return -ENODEV; 20104929116bSKamal Heib 20114929116bSKamal Heib port_attr->max_mtu = IB_MTU_4096; 20124929116bSKamal Heib port_attr->active_mtu = ib_mtu_int_to_enum(netdev->mtu); 20134929116bSKamal Heib 20144929116bSKamal Heib if (!netif_carrier_ok(netdev)) { 20154929116bSKamal Heib port_attr->state = IB_PORT_DOWN; 20164929116bSKamal Heib port_attr->phys_state = IB_PORT_PHYS_STATE_DISABLED; 20174929116bSKamal Heib } else { 2018390d3fdcSMichal Kalderon rcu_read_lock(); 2019390d3fdcSMichal Kalderon inetdev = __in_dev_get_rcu(netdev); 20204929116bSKamal Heib 20214929116bSKamal Heib if (inetdev && inetdev->ifa_list) { 20224929116bSKamal Heib port_attr->state = IB_PORT_ACTIVE; 20234929116bSKamal Heib port_attr->phys_state = IB_PORT_PHYS_STATE_LINK_UP; 20244929116bSKamal Heib } else { 20254929116bSKamal Heib port_attr->state = IB_PORT_INIT; 20264929116bSKamal Heib port_attr->phys_state = 20274929116bSKamal Heib IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING; 20284929116bSKamal Heib } 2029390d3fdcSMichal Kalderon 2030390d3fdcSMichal Kalderon rcu_read_unlock(); 20314929116bSKamal Heib } 20324929116bSKamal Heib 2033390d3fdcSMichal Kalderon dev_put(netdev); 20341e123d96SGuoqing Jiang return device->ops.query_port(device, port_num, port_attr); 20354929116bSKamal Heib } 20364929116bSKamal Heib 20374929116bSKamal Heib static int __ib_query_port(struct ib_device *device, 20384929116bSKamal Heib u8 port_num, 20394929116bSKamal Heib struct ib_port_attr *port_attr) 20404929116bSKamal Heib { 20414929116bSKamal Heib union ib_gid gid = {}; 20424929116bSKamal Heib int err; 20434929116bSKamal Heib 20444929116bSKamal Heib memset(port_attr, 0, sizeof(*port_attr)); 20454929116bSKamal Heib 20464929116bSKamal Heib err = device->ops.query_port(device, port_num, port_attr); 20474929116bSKamal Heib if (err || port_attr->subnet_prefix) 20484929116bSKamal Heib return err; 20494929116bSKamal Heib 20504929116bSKamal Heib if (rdma_port_get_link_layer(device, port_num) != 20514929116bSKamal Heib IB_LINK_LAYER_INFINIBAND) 20524929116bSKamal Heib return 0; 20534929116bSKamal Heib 20544929116bSKamal Heib err = device->ops.query_gid(device, port_num, 0, &gid); 20554929116bSKamal Heib if (err) 20564929116bSKamal Heib return err; 20574929116bSKamal Heib 20584929116bSKamal Heib port_attr->subnet_prefix = be64_to_cpu(gid.global.subnet_prefix); 20594929116bSKamal Heib return 0; 20604929116bSKamal Heib } 20614929116bSKamal Heib 20621da177e4SLinus Torvalds /** 20631da177e4SLinus Torvalds * ib_query_port - Query IB port attributes 20641da177e4SLinus Torvalds * @device:Device to query 20651da177e4SLinus Torvalds * @port_num:Port number to query 20661da177e4SLinus Torvalds * @port_attr:Port attributes 20671da177e4SLinus Torvalds * 20681da177e4SLinus Torvalds * ib_query_port() returns the attributes of a port through the 20691da177e4SLinus Torvalds * @port_attr pointer. 20701da177e4SLinus Torvalds */ 20711da177e4SLinus Torvalds int ib_query_port(struct ib_device *device, 20721da177e4SLinus Torvalds u8 port_num, 20731da177e4SLinus Torvalds struct ib_port_attr *port_attr) 20741da177e4SLinus Torvalds { 207524dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 2076116c0074SRoland Dreier return -EINVAL; 2077116c0074SRoland Dreier 20784929116bSKamal Heib if (rdma_protocol_iwarp(device, port_num)) 20794929116bSKamal Heib return iw_query_port(device, port_num, port_attr); 20804929116bSKamal Heib else 20814929116bSKamal Heib return __ib_query_port(device, port_num, port_attr); 20821da177e4SLinus Torvalds } 20831da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_port); 20841da177e4SLinus Torvalds 2085324e227eSJason Gunthorpe static void add_ndev_hash(struct ib_port_data *pdata) 2086324e227eSJason Gunthorpe { 2087324e227eSJason Gunthorpe unsigned long flags; 2088324e227eSJason Gunthorpe 2089324e227eSJason Gunthorpe might_sleep(); 2090324e227eSJason Gunthorpe 2091324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 2092324e227eSJason Gunthorpe if (hash_hashed(&pdata->ndev_hash_link)) { 2093324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 2094324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 2095324e227eSJason Gunthorpe /* 2096324e227eSJason Gunthorpe * We cannot do hash_add_rcu after a hash_del_rcu until the 2097324e227eSJason Gunthorpe * grace period 2098324e227eSJason Gunthorpe */ 2099324e227eSJason Gunthorpe synchronize_rcu(); 2100324e227eSJason Gunthorpe spin_lock_irqsave(&ndev_hash_lock, flags); 2101324e227eSJason Gunthorpe } 2102324e227eSJason Gunthorpe if (pdata->netdev) 2103324e227eSJason Gunthorpe hash_add_rcu(ndev_hash, &pdata->ndev_hash_link, 2104324e227eSJason Gunthorpe (uintptr_t)pdata->netdev); 2105324e227eSJason Gunthorpe spin_unlock_irqrestore(&ndev_hash_lock, flags); 2106324e227eSJason Gunthorpe } 2107324e227eSJason Gunthorpe 21081da177e4SLinus Torvalds /** 2109c2261dd7SJason Gunthorpe * ib_device_set_netdev - Associate the ib_dev with an underlying net_device 2110c2261dd7SJason Gunthorpe * @ib_dev: Device to modify 2111c2261dd7SJason Gunthorpe * @ndev: net_device to affiliate, may be NULL 2112c2261dd7SJason Gunthorpe * @port: IB port the net_device is connected to 2113c2261dd7SJason Gunthorpe * 2114c2261dd7SJason Gunthorpe * Drivers should use this to link the ib_device to a netdev so the netdev 2115c2261dd7SJason Gunthorpe * shows up in interfaces like ib_enum_roce_netdev. Only one netdev may be 2116c2261dd7SJason Gunthorpe * affiliated with any port. 2117c2261dd7SJason Gunthorpe * 2118c2261dd7SJason Gunthorpe * The caller must ensure that the given ndev is not unregistered or 2119c2261dd7SJason Gunthorpe * unregistering, and that either the ib_device is unregistered or 2120c2261dd7SJason Gunthorpe * ib_device_set_netdev() is called with NULL when the ndev sends a 2121c2261dd7SJason Gunthorpe * NETDEV_UNREGISTER event. 2122c2261dd7SJason Gunthorpe */ 2123c2261dd7SJason Gunthorpe int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 2124c2261dd7SJason Gunthorpe unsigned int port) 2125c2261dd7SJason Gunthorpe { 2126c2261dd7SJason Gunthorpe struct net_device *old_ndev; 2127c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 2128c2261dd7SJason Gunthorpe unsigned long flags; 2129c2261dd7SJason Gunthorpe int ret; 2130c2261dd7SJason Gunthorpe 2131c2261dd7SJason Gunthorpe /* 2132c2261dd7SJason Gunthorpe * Drivers wish to call this before ib_register_driver, so we have to 2133c2261dd7SJason Gunthorpe * setup the port data early. 2134c2261dd7SJason Gunthorpe */ 2135c2261dd7SJason Gunthorpe ret = alloc_port_data(ib_dev); 2136c2261dd7SJason Gunthorpe if (ret) 2137c2261dd7SJason Gunthorpe return ret; 2138c2261dd7SJason Gunthorpe 2139c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 2140c2261dd7SJason Gunthorpe return -EINVAL; 2141c2261dd7SJason Gunthorpe 2142c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 2143c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 2144324e227eSJason Gunthorpe old_ndev = rcu_dereference_protected( 2145324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 2146324e227eSJason Gunthorpe if (old_ndev == ndev) { 2147c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 2148c2261dd7SJason Gunthorpe return 0; 2149c2261dd7SJason Gunthorpe } 2150c2261dd7SJason Gunthorpe 2151c2261dd7SJason Gunthorpe if (ndev) 2152c2261dd7SJason Gunthorpe dev_hold(ndev); 2153324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, ndev); 2154c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 2155c2261dd7SJason Gunthorpe 2156324e227eSJason Gunthorpe add_ndev_hash(pdata); 2157c2261dd7SJason Gunthorpe if (old_ndev) 2158c2261dd7SJason Gunthorpe dev_put(old_ndev); 2159c2261dd7SJason Gunthorpe 2160c2261dd7SJason Gunthorpe return 0; 2161c2261dd7SJason Gunthorpe } 2162c2261dd7SJason Gunthorpe EXPORT_SYMBOL(ib_device_set_netdev); 2163c2261dd7SJason Gunthorpe 2164c2261dd7SJason Gunthorpe static void free_netdevs(struct ib_device *ib_dev) 2165c2261dd7SJason Gunthorpe { 2166c2261dd7SJason Gunthorpe unsigned long flags; 2167c2261dd7SJason Gunthorpe unsigned int port; 2168c2261dd7SJason Gunthorpe 216946bdf370SKamal Heib if (!ib_dev->port_data) 217046bdf370SKamal Heib return; 217146bdf370SKamal Heib 2172c2261dd7SJason Gunthorpe rdma_for_each_port (ib_dev, port) { 2173c2261dd7SJason Gunthorpe struct ib_port_data *pdata = &ib_dev->port_data[port]; 2174324e227eSJason Gunthorpe struct net_device *ndev; 2175c2261dd7SJason Gunthorpe 2176c2261dd7SJason Gunthorpe spin_lock_irqsave(&pdata->netdev_lock, flags); 2177324e227eSJason Gunthorpe ndev = rcu_dereference_protected( 2178324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 2179324e227eSJason Gunthorpe if (ndev) { 2180324e227eSJason Gunthorpe spin_lock(&ndev_hash_lock); 2181324e227eSJason Gunthorpe hash_del_rcu(&pdata->ndev_hash_link); 2182324e227eSJason Gunthorpe spin_unlock(&ndev_hash_lock); 2183324e227eSJason Gunthorpe 2184324e227eSJason Gunthorpe /* 2185324e227eSJason Gunthorpe * If this is the last dev_put there is still a 2186324e227eSJason Gunthorpe * synchronize_rcu before the netdev is kfreed, so we 2187324e227eSJason Gunthorpe * can continue to rely on unlocked pointer 2188324e227eSJason Gunthorpe * comparisons after the put 2189324e227eSJason Gunthorpe */ 2190324e227eSJason Gunthorpe rcu_assign_pointer(pdata->netdev, NULL); 2191324e227eSJason Gunthorpe dev_put(ndev); 2192c2261dd7SJason Gunthorpe } 2193c2261dd7SJason Gunthorpe spin_unlock_irqrestore(&pdata->netdev_lock, flags); 2194c2261dd7SJason Gunthorpe } 2195c2261dd7SJason Gunthorpe } 2196c2261dd7SJason Gunthorpe 2197c2261dd7SJason Gunthorpe struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, 2198c2261dd7SJason Gunthorpe unsigned int port) 2199c2261dd7SJason Gunthorpe { 2200c2261dd7SJason Gunthorpe struct ib_port_data *pdata; 2201c2261dd7SJason Gunthorpe struct net_device *res; 2202c2261dd7SJason Gunthorpe 2203c2261dd7SJason Gunthorpe if (!rdma_is_port_valid(ib_dev, port)) 2204c2261dd7SJason Gunthorpe return NULL; 2205c2261dd7SJason Gunthorpe 2206c2261dd7SJason Gunthorpe pdata = &ib_dev->port_data[port]; 2207c2261dd7SJason Gunthorpe 2208c2261dd7SJason Gunthorpe /* 2209c2261dd7SJason Gunthorpe * New drivers should use ib_device_set_netdev() not the legacy 2210c2261dd7SJason Gunthorpe * get_netdev(). 2211c2261dd7SJason Gunthorpe */ 2212c2261dd7SJason Gunthorpe if (ib_dev->ops.get_netdev) 2213c2261dd7SJason Gunthorpe res = ib_dev->ops.get_netdev(ib_dev, port); 2214c2261dd7SJason Gunthorpe else { 2215c2261dd7SJason Gunthorpe spin_lock(&pdata->netdev_lock); 2216324e227eSJason Gunthorpe res = rcu_dereference_protected( 2217324e227eSJason Gunthorpe pdata->netdev, lockdep_is_held(&pdata->netdev_lock)); 2218c2261dd7SJason Gunthorpe if (res) 2219c2261dd7SJason Gunthorpe dev_hold(res); 2220c2261dd7SJason Gunthorpe spin_unlock(&pdata->netdev_lock); 2221c2261dd7SJason Gunthorpe } 2222c2261dd7SJason Gunthorpe 2223c2261dd7SJason Gunthorpe /* 2224c2261dd7SJason Gunthorpe * If we are starting to unregister expedite things by preventing 2225c2261dd7SJason Gunthorpe * propagation of an unregistering netdev. 2226c2261dd7SJason Gunthorpe */ 2227c2261dd7SJason Gunthorpe if (res && res->reg_state != NETREG_REGISTERED) { 2228c2261dd7SJason Gunthorpe dev_put(res); 2229c2261dd7SJason Gunthorpe return NULL; 2230c2261dd7SJason Gunthorpe } 2231c2261dd7SJason Gunthorpe 2232c2261dd7SJason Gunthorpe return res; 2233c2261dd7SJason Gunthorpe } 2234c2261dd7SJason Gunthorpe 2235c2261dd7SJason Gunthorpe /** 2236324e227eSJason Gunthorpe * ib_device_get_by_netdev - Find an IB device associated with a netdev 2237324e227eSJason Gunthorpe * @ndev: netdev to locate 2238324e227eSJason Gunthorpe * @driver_id: The driver ID that must match (RDMA_DRIVER_UNKNOWN matches all) 2239324e227eSJason Gunthorpe * 2240324e227eSJason Gunthorpe * Find and hold an ib_device that is associated with a netdev via 2241324e227eSJason Gunthorpe * ib_device_set_netdev(). The caller must call ib_device_put() on the 2242324e227eSJason Gunthorpe * returned pointer. 2243324e227eSJason Gunthorpe */ 2244324e227eSJason Gunthorpe struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 2245324e227eSJason Gunthorpe enum rdma_driver_id driver_id) 2246324e227eSJason Gunthorpe { 2247324e227eSJason Gunthorpe struct ib_device *res = NULL; 2248324e227eSJason Gunthorpe struct ib_port_data *cur; 2249324e227eSJason Gunthorpe 2250324e227eSJason Gunthorpe rcu_read_lock(); 2251324e227eSJason Gunthorpe hash_for_each_possible_rcu (ndev_hash, cur, ndev_hash_link, 2252324e227eSJason Gunthorpe (uintptr_t)ndev) { 2253324e227eSJason Gunthorpe if (rcu_access_pointer(cur->netdev) == ndev && 2254324e227eSJason Gunthorpe (driver_id == RDMA_DRIVER_UNKNOWN || 2255b9560a41SJason Gunthorpe cur->ib_dev->ops.driver_id == driver_id) && 2256324e227eSJason Gunthorpe ib_device_try_get(cur->ib_dev)) { 2257324e227eSJason Gunthorpe res = cur->ib_dev; 2258324e227eSJason Gunthorpe break; 2259324e227eSJason Gunthorpe } 2260324e227eSJason Gunthorpe } 2261324e227eSJason Gunthorpe rcu_read_unlock(); 2262324e227eSJason Gunthorpe 2263324e227eSJason Gunthorpe return res; 2264324e227eSJason Gunthorpe } 2265324e227eSJason Gunthorpe EXPORT_SYMBOL(ib_device_get_by_netdev); 2266324e227eSJason Gunthorpe 2267324e227eSJason Gunthorpe /** 226803db3a2dSMatan Barak * ib_enum_roce_netdev - enumerate all RoCE ports 226903db3a2dSMatan Barak * @ib_dev : IB device we want to query 227003db3a2dSMatan Barak * @filter: Should we call the callback? 227103db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 227203db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 227303db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 227403db3a2dSMatan Barak * 227503db3a2dSMatan Barak * Enumerates all of the physical RoCE ports of ib_dev 227603db3a2dSMatan Barak * which are related to netdevice and calls callback() on each 227703db3a2dSMatan Barak * device for which filter() function returns non zero. 227803db3a2dSMatan Barak */ 227903db3a2dSMatan Barak void ib_enum_roce_netdev(struct ib_device *ib_dev, 228003db3a2dSMatan Barak roce_netdev_filter filter, 228103db3a2dSMatan Barak void *filter_cookie, 228203db3a2dSMatan Barak roce_netdev_callback cb, 228303db3a2dSMatan Barak void *cookie) 228403db3a2dSMatan Barak { 2285ea1075edSJason Gunthorpe unsigned int port; 228603db3a2dSMatan Barak 2287ea1075edSJason Gunthorpe rdma_for_each_port (ib_dev, port) 228803db3a2dSMatan Barak if (rdma_protocol_roce(ib_dev, port)) { 2289c2261dd7SJason Gunthorpe struct net_device *idev = 2290c2261dd7SJason Gunthorpe ib_device_get_netdev(ib_dev, port); 229103db3a2dSMatan Barak 229203db3a2dSMatan Barak if (filter(ib_dev, port, idev, filter_cookie)) 229303db3a2dSMatan Barak cb(ib_dev, port, idev, cookie); 229403db3a2dSMatan Barak 229503db3a2dSMatan Barak if (idev) 229603db3a2dSMatan Barak dev_put(idev); 229703db3a2dSMatan Barak } 229803db3a2dSMatan Barak } 229903db3a2dSMatan Barak 230003db3a2dSMatan Barak /** 230103db3a2dSMatan Barak * ib_enum_all_roce_netdevs - enumerate all RoCE devices 230203db3a2dSMatan Barak * @filter: Should we call the callback? 230303db3a2dSMatan Barak * @filter_cookie: Cookie passed to filter 230403db3a2dSMatan Barak * @cb: Callback to call for each found RoCE ports 230503db3a2dSMatan Barak * @cookie: Cookie passed back to the callback 230603db3a2dSMatan Barak * 230703db3a2dSMatan Barak * Enumerates all RoCE devices' physical ports which are related 230803db3a2dSMatan Barak * to netdevices and calls callback() on each device for which 230903db3a2dSMatan Barak * filter() function returns non zero. 231003db3a2dSMatan Barak */ 231103db3a2dSMatan Barak void ib_enum_all_roce_netdevs(roce_netdev_filter filter, 231203db3a2dSMatan Barak void *filter_cookie, 231303db3a2dSMatan Barak roce_netdev_callback cb, 231403db3a2dSMatan Barak void *cookie) 231503db3a2dSMatan Barak { 231603db3a2dSMatan Barak struct ib_device *dev; 23170df91bb6SJason Gunthorpe unsigned long index; 231803db3a2dSMatan Barak 2319921eab11SJason Gunthorpe down_read(&devices_rwsem); 23200df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) 232103db3a2dSMatan Barak ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie); 2322921eab11SJason Gunthorpe up_read(&devices_rwsem); 232303db3a2dSMatan Barak } 232403db3a2dSMatan Barak 232503db3a2dSMatan Barak /** 23268030c835SLeon Romanovsky * ib_enum_all_devs - enumerate all ib_devices 23278030c835SLeon Romanovsky * @cb: Callback to call for each found ib_device 23288030c835SLeon Romanovsky * 23298030c835SLeon Romanovsky * Enumerates all ib_devices and calls callback() on each device. 23308030c835SLeon Romanovsky */ 23318030c835SLeon Romanovsky int ib_enum_all_devs(nldev_callback nldev_cb, struct sk_buff *skb, 23328030c835SLeon Romanovsky struct netlink_callback *cb) 23338030c835SLeon Romanovsky { 23340df91bb6SJason Gunthorpe unsigned long index; 23358030c835SLeon Romanovsky struct ib_device *dev; 23368030c835SLeon Romanovsky unsigned int idx = 0; 23378030c835SLeon Romanovsky int ret = 0; 23388030c835SLeon Romanovsky 2339921eab11SJason Gunthorpe down_read(&devices_rwsem); 23400df91bb6SJason Gunthorpe xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED) { 234137eeab55SParav Pandit if (!rdma_dev_access_netns(dev, sock_net(skb->sk))) 234237eeab55SParav Pandit continue; 234337eeab55SParav Pandit 23448030c835SLeon Romanovsky ret = nldev_cb(dev, skb, cb, idx); 23458030c835SLeon Romanovsky if (ret) 23468030c835SLeon Romanovsky break; 23478030c835SLeon Romanovsky idx++; 23488030c835SLeon Romanovsky } 2349921eab11SJason Gunthorpe up_read(&devices_rwsem); 23508030c835SLeon Romanovsky return ret; 23518030c835SLeon Romanovsky } 23528030c835SLeon Romanovsky 23538030c835SLeon Romanovsky /** 23541da177e4SLinus Torvalds * ib_query_pkey - Get P_Key table entry 23551da177e4SLinus Torvalds * @device:Device to query 23561da177e4SLinus Torvalds * @port_num:Port number to query 23571da177e4SLinus Torvalds * @index:P_Key table index to query 23581da177e4SLinus Torvalds * @pkey:Returned P_Key 23591da177e4SLinus Torvalds * 23601da177e4SLinus Torvalds * ib_query_pkey() fetches the specified P_Key table entry. 23611da177e4SLinus Torvalds */ 23621da177e4SLinus Torvalds int ib_query_pkey(struct ib_device *device, 23631da177e4SLinus Torvalds u8 port_num, u16 index, u16 *pkey) 23641da177e4SLinus Torvalds { 23659af3f5cfSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 23669af3f5cfSYuval Shaia return -EINVAL; 23679af3f5cfSYuval Shaia 2368ab75a6cbSKamal Heib if (!device->ops.query_pkey) 2369ab75a6cbSKamal Heib return -EOPNOTSUPP; 2370ab75a6cbSKamal Heib 23713023a1e9SKamal Heib return device->ops.query_pkey(device, port_num, index, pkey); 23721da177e4SLinus Torvalds } 23731da177e4SLinus Torvalds EXPORT_SYMBOL(ib_query_pkey); 23741da177e4SLinus Torvalds 23751da177e4SLinus Torvalds /** 23761da177e4SLinus Torvalds * ib_modify_device - Change IB device attributes 23771da177e4SLinus Torvalds * @device:Device to modify 23781da177e4SLinus Torvalds * @device_modify_mask:Mask of attributes to change 23791da177e4SLinus Torvalds * @device_modify:New attribute values 23801da177e4SLinus Torvalds * 23811da177e4SLinus Torvalds * ib_modify_device() changes a device's attributes as specified by 23821da177e4SLinus Torvalds * the @device_modify_mask and @device_modify structure. 23831da177e4SLinus Torvalds */ 23841da177e4SLinus Torvalds int ib_modify_device(struct ib_device *device, 23851da177e4SLinus Torvalds int device_modify_mask, 23861da177e4SLinus Torvalds struct ib_device_modify *device_modify) 23871da177e4SLinus Torvalds { 23883023a1e9SKamal Heib if (!device->ops.modify_device) 2389d0f3ef36SKamal Heib return -EOPNOTSUPP; 239010e1b54bSBart Van Assche 23913023a1e9SKamal Heib return device->ops.modify_device(device, device_modify_mask, 23921da177e4SLinus Torvalds device_modify); 23931da177e4SLinus Torvalds } 23941da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_device); 23951da177e4SLinus Torvalds 23961da177e4SLinus Torvalds /** 23971da177e4SLinus Torvalds * ib_modify_port - Modifies the attributes for the specified port. 23981da177e4SLinus Torvalds * @device: The device to modify. 23991da177e4SLinus Torvalds * @port_num: The number of the port to modify. 24001da177e4SLinus Torvalds * @port_modify_mask: Mask used to specify which attributes of the port 24011da177e4SLinus Torvalds * to change. 24021da177e4SLinus Torvalds * @port_modify: New attribute values for the port. 24031da177e4SLinus Torvalds * 24041da177e4SLinus Torvalds * ib_modify_port() changes a port's attributes as specified by the 24051da177e4SLinus Torvalds * @port_modify_mask and @port_modify structure. 24061da177e4SLinus Torvalds */ 24071da177e4SLinus Torvalds int ib_modify_port(struct ib_device *device, 24081da177e4SLinus Torvalds u8 port_num, int port_modify_mask, 24091da177e4SLinus Torvalds struct ib_port_modify *port_modify) 24101da177e4SLinus Torvalds { 241161e0962dSSelvin Xavier int rc; 241210e1b54bSBart Van Assche 241324dc831bSYuval Shaia if (!rdma_is_port_valid(device, port_num)) 2414116c0074SRoland Dreier return -EINVAL; 2415116c0074SRoland Dreier 24163023a1e9SKamal Heib if (device->ops.modify_port) 24173023a1e9SKamal Heib rc = device->ops.modify_port(device, port_num, 24183023a1e9SKamal Heib port_modify_mask, 24191da177e4SLinus Torvalds port_modify); 242055bfe905SKamal Heib else if (rdma_protocol_roce(device, port_num) && 242155bfe905SKamal Heib ((port_modify->set_port_cap_mask & ~IB_PORT_CM_SUP) == 0 || 242255bfe905SKamal Heib (port_modify->clr_port_cap_mask & ~IB_PORT_CM_SUP) == 0)) 242355bfe905SKamal Heib rc = 0; 242461e0962dSSelvin Xavier else 242555bfe905SKamal Heib rc = -EOPNOTSUPP; 242661e0962dSSelvin Xavier return rc; 24271da177e4SLinus Torvalds } 24281da177e4SLinus Torvalds EXPORT_SYMBOL(ib_modify_port); 24291da177e4SLinus Torvalds 24305eb620c8SYosef Etigin /** 24315eb620c8SYosef Etigin * ib_find_gid - Returns the port number and GID table index where 2432dbb12562SParav Pandit * a specified GID value occurs. Its searches only for IB link layer. 24335eb620c8SYosef Etigin * @device: The device to query. 24345eb620c8SYosef Etigin * @gid: The GID value to search for. 24355eb620c8SYosef Etigin * @port_num: The port number of the device where the GID value was found. 24365eb620c8SYosef Etigin * @index: The index into the GID table where the GID was found. This 24375eb620c8SYosef Etigin * parameter may be NULL. 24385eb620c8SYosef Etigin */ 24395eb620c8SYosef Etigin int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2440b26c4a11SParav Pandit u8 *port_num, u16 *index) 24415eb620c8SYosef Etigin { 24425eb620c8SYosef Etigin union ib_gid tmp_gid; 2443ea1075edSJason Gunthorpe unsigned int port; 2444ea1075edSJason Gunthorpe int ret, i; 24455eb620c8SYosef Etigin 2446ea1075edSJason Gunthorpe rdma_for_each_port (device, port) { 244722d24f75SParav Pandit if (!rdma_protocol_ib(device, port)) 2448b39ffa1dSMatan Barak continue; 2449b39ffa1dSMatan Barak 24508ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port].immutable.gid_tbl_len; 24518ceb1357SJason Gunthorpe ++i) { 24521dfce294SParav Pandit ret = rdma_query_gid(device, port, i, &tmp_gid); 24535eb620c8SYosef Etigin if (ret) 24545eb620c8SYosef Etigin return ret; 24555eb620c8SYosef Etigin if (!memcmp(&tmp_gid, gid, sizeof *gid)) { 24565eb620c8SYosef Etigin *port_num = port; 24575eb620c8SYosef Etigin if (index) 24585eb620c8SYosef Etigin *index = i; 24595eb620c8SYosef Etigin return 0; 24605eb620c8SYosef Etigin } 24615eb620c8SYosef Etigin } 24625eb620c8SYosef Etigin } 24635eb620c8SYosef Etigin 24645eb620c8SYosef Etigin return -ENOENT; 24655eb620c8SYosef Etigin } 24665eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_gid); 24675eb620c8SYosef Etigin 24685eb620c8SYosef Etigin /** 24695eb620c8SYosef Etigin * ib_find_pkey - Returns the PKey table index where a specified 24705eb620c8SYosef Etigin * PKey value occurs. 24715eb620c8SYosef Etigin * @device: The device to query. 24725eb620c8SYosef Etigin * @port_num: The port number of the device to search for the PKey. 24735eb620c8SYosef Etigin * @pkey: The PKey value to search for. 24745eb620c8SYosef Etigin * @index: The index into the PKey table where the PKey was found. 24755eb620c8SYosef Etigin */ 24765eb620c8SYosef Etigin int ib_find_pkey(struct ib_device *device, 24775eb620c8SYosef Etigin u8 port_num, u16 pkey, u16 *index) 24785eb620c8SYosef Etigin { 24795eb620c8SYosef Etigin int ret, i; 24805eb620c8SYosef Etigin u16 tmp_pkey; 2481ff7166c4SJack Morgenstein int partial_ix = -1; 24825eb620c8SYosef Etigin 24838ceb1357SJason Gunthorpe for (i = 0; i < device->port_data[port_num].immutable.pkey_tbl_len; 24848ceb1357SJason Gunthorpe ++i) { 24855eb620c8SYosef Etigin ret = ib_query_pkey(device, port_num, i, &tmp_pkey); 24865eb620c8SYosef Etigin if (ret) 24875eb620c8SYosef Etigin return ret; 248836026eccSMoni Shoua if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) { 2489ff7166c4SJack Morgenstein /* if there is full-member pkey take it.*/ 2490ff7166c4SJack Morgenstein if (tmp_pkey & 0x8000) { 24915eb620c8SYosef Etigin *index = i; 24925eb620c8SYosef Etigin return 0; 24935eb620c8SYosef Etigin } 2494ff7166c4SJack Morgenstein if (partial_ix < 0) 2495ff7166c4SJack Morgenstein partial_ix = i; 2496ff7166c4SJack Morgenstein } 24975eb620c8SYosef Etigin } 24985eb620c8SYosef Etigin 2499ff7166c4SJack Morgenstein /*no full-member, if exists take the limited*/ 2500ff7166c4SJack Morgenstein if (partial_ix >= 0) { 2501ff7166c4SJack Morgenstein *index = partial_ix; 2502ff7166c4SJack Morgenstein return 0; 2503ff7166c4SJack Morgenstein } 25045eb620c8SYosef Etigin return -ENOENT; 25055eb620c8SYosef Etigin } 25065eb620c8SYosef Etigin EXPORT_SYMBOL(ib_find_pkey); 25075eb620c8SYosef Etigin 25089268f72dSYotam Kenneth /** 25099268f72dSYotam Kenneth * ib_get_net_dev_by_params() - Return the appropriate net_dev 25109268f72dSYotam Kenneth * for a received CM request 25119268f72dSYotam Kenneth * @dev: An RDMA device on which the request has been received. 25129268f72dSYotam Kenneth * @port: Port number on the RDMA device. 25139268f72dSYotam Kenneth * @pkey: The Pkey the request came on. 25149268f72dSYotam Kenneth * @gid: A GID that the net_dev uses to communicate. 25159268f72dSYotam Kenneth * @addr: Contains the IP address that the request specified as its 25169268f72dSYotam Kenneth * destination. 2517921eab11SJason Gunthorpe * 25189268f72dSYotam Kenneth */ 25199268f72dSYotam Kenneth struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, 25209268f72dSYotam Kenneth u8 port, 25219268f72dSYotam Kenneth u16 pkey, 25229268f72dSYotam Kenneth const union ib_gid *gid, 25239268f72dSYotam Kenneth const struct sockaddr *addr) 25249268f72dSYotam Kenneth { 25259268f72dSYotam Kenneth struct net_device *net_dev = NULL; 25260df91bb6SJason Gunthorpe unsigned long index; 25270df91bb6SJason Gunthorpe void *client_data; 25289268f72dSYotam Kenneth 25299268f72dSYotam Kenneth if (!rdma_protocol_ib(dev, port)) 25309268f72dSYotam Kenneth return NULL; 25319268f72dSYotam Kenneth 2532921eab11SJason Gunthorpe /* 2533921eab11SJason Gunthorpe * Holding the read side guarantees that the client will not become 2534921eab11SJason Gunthorpe * unregistered while we are calling get_net_dev_by_params() 2535921eab11SJason Gunthorpe */ 2536921eab11SJason Gunthorpe down_read(&dev->client_data_rwsem); 25370df91bb6SJason Gunthorpe xan_for_each_marked (&dev->client_data, index, client_data, 25380df91bb6SJason Gunthorpe CLIENT_DATA_REGISTERED) { 25390df91bb6SJason Gunthorpe struct ib_client *client = xa_load(&clients, index); 25409268f72dSYotam Kenneth 25410df91bb6SJason Gunthorpe if (!client || !client->get_net_dev_by_params) 25429268f72dSYotam Kenneth continue; 25439268f72dSYotam Kenneth 25440df91bb6SJason Gunthorpe net_dev = client->get_net_dev_by_params(dev, port, pkey, gid, 25450df91bb6SJason Gunthorpe addr, client_data); 25469268f72dSYotam Kenneth if (net_dev) 25479268f72dSYotam Kenneth break; 25489268f72dSYotam Kenneth } 2549921eab11SJason Gunthorpe up_read(&dev->client_data_rwsem); 25509268f72dSYotam Kenneth 25519268f72dSYotam Kenneth return net_dev; 25529268f72dSYotam Kenneth } 25539268f72dSYotam Kenneth EXPORT_SYMBOL(ib_get_net_dev_by_params); 25549268f72dSYotam Kenneth 2555521ed0d9SKamal Heib void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) 2556521ed0d9SKamal Heib { 25573023a1e9SKamal Heib struct ib_device_ops *dev_ops = &dev->ops; 2558521ed0d9SKamal Heib #define SET_DEVICE_OP(ptr, name) \ 2559521ed0d9SKamal Heib do { \ 2560521ed0d9SKamal Heib if (ops->name) \ 2561521ed0d9SKamal Heib if (!((ptr)->name)) \ 2562521ed0d9SKamal Heib (ptr)->name = ops->name; \ 2563521ed0d9SKamal Heib } while (0) 2564521ed0d9SKamal Heib 256530471d4bSLeon Romanovsky #define SET_OBJ_SIZE(ptr, name) SET_DEVICE_OP(ptr, size_##name) 256630471d4bSLeon Romanovsky 2567b9560a41SJason Gunthorpe if (ops->driver_id != RDMA_DRIVER_UNKNOWN) { 2568b9560a41SJason Gunthorpe WARN_ON(dev_ops->driver_id != RDMA_DRIVER_UNKNOWN && 2569b9560a41SJason Gunthorpe dev_ops->driver_id != ops->driver_id); 2570b9560a41SJason Gunthorpe dev_ops->driver_id = ops->driver_id; 2571b9560a41SJason Gunthorpe } 25727a154142SJason Gunthorpe if (ops->owner) { 25737a154142SJason Gunthorpe WARN_ON(dev_ops->owner && dev_ops->owner != ops->owner); 25747a154142SJason Gunthorpe dev_ops->owner = ops->owner; 25757a154142SJason Gunthorpe } 257672c6ec18SJason Gunthorpe if (ops->uverbs_abi_ver) 257772c6ec18SJason Gunthorpe dev_ops->uverbs_abi_ver = ops->uverbs_abi_ver; 2578b9560a41SJason Gunthorpe 25798f71bb00SJason Gunthorpe dev_ops->uverbs_no_driver_id_binding |= 25808f71bb00SJason Gunthorpe ops->uverbs_no_driver_id_binding; 25818f71bb00SJason Gunthorpe 25823023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, add_gid); 25832f1927b0SMoni Shoua SET_DEVICE_OP(dev_ops, advise_mr); 25843023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_dm); 25853023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_hw_stats); 25863023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mr); 258726bc7eaeSIsrael Rukshin SET_DEVICE_OP(dev_ops, alloc_mr_integrity); 25883023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_mw); 25893023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_pd); 25903023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_rdma_netdev); 25913023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_ucontext); 25923023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, alloc_xrcd); 25933023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, attach_mcast); 25943023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, check_mr_status); 2595c4ffee7cSMark Zhang SET_DEVICE_OP(dev_ops, counter_alloc_stats); 259699fa331dSMark Zhang SET_DEVICE_OP(dev_ops, counter_bind_qp); 259799fa331dSMark Zhang SET_DEVICE_OP(dev_ops, counter_dealloc); 259899fa331dSMark Zhang SET_DEVICE_OP(dev_ops, counter_unbind_qp); 2599c4ffee7cSMark Zhang SET_DEVICE_OP(dev_ops, counter_update_stats); 26003023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_ah); 26013023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_counters); 26023023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_cq); 26033023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow); 26043023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_flow_action_esp); 26053023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_qp); 26063023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_rwq_ind_table); 26073023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_srq); 26083023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, create_wq); 26093023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_dm); 2610d0899892SJason Gunthorpe SET_DEVICE_OP(dev_ops, dealloc_driver); 26113023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_mw); 26123023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_pd); 26133023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_ucontext); 26143023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dealloc_xrcd); 26153023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, del_gid); 26163023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, dereg_mr); 26173023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_ah); 26183023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_counters); 26193023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_cq); 26203023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow); 26213023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_flow_action); 26223023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_qp); 26233023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_rwq_ind_table); 26243023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_srq); 26253023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, destroy_wq); 26263023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, detach_mcast); 26273023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, disassociate_ucontext); 26283023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_rq); 26293023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, drain_sq); 2630ca22354bSJason Gunthorpe SET_DEVICE_OP(dev_ops, enable_driver); 2631211cd945SMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_res_cm_id_entry); 26329e2a187aSMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_res_cq_entry); 263365959522SMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_res_cq_entry_raw); 2634f4434529SMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_res_mr_entry); 263565959522SMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_res_mr_entry_raw); 26365cc34116SMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_res_qp_entry); 263765959522SMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_res_qp_entry_raw); 2638f4434529SMaor Gottlieb SET_DEVICE_OP(dev_ops, fill_stat_mr_entry); 26393023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dev_fw_str); 26403023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_dma_mr); 26413023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_hw_stats); 26423023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_link_layer); 26433023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_netdev); 26443023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_port_immutable); 26453023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vector_affinity); 26463023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_config); 2647bfcb3c5dSDanit Goldberg SET_DEVICE_OP(dev_ops, get_vf_guid); 26483023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, get_vf_stats); 2649ea4baf7fSParav Pandit SET_DEVICE_OP(dev_ops, init_port); 2650dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_accept); 2651dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_add_ref); 2652dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_connect); 2653dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_create_listen); 2654dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_destroy_listen); 2655dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_get_qp); 2656dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_reject); 2657dd05cb82SKamal Heib SET_DEVICE_OP(dev_ops, iw_rem_ref); 26583023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, map_mr_sg); 26592cdfcdd8SMax Gurtovoy SET_DEVICE_OP(dev_ops, map_mr_sg_pi); 26603023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, mmap); 26613411f9f0SMichal Kalderon SET_DEVICE_OP(dev_ops, mmap_free); 26623023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_ah); 26633023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_cq); 26643023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_device); 26653023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_flow_action_esp); 26663023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_port); 26673023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_qp); 26683023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_srq); 26693023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, modify_wq); 26703023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, peek_cq); 26713023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, poll_cq); 26723023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_recv); 26733023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_send); 26743023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, post_srq_recv); 26753023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, process_mad); 26763023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_ah); 26773023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_device); 26783023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_gid); 26793023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_pkey); 26803023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_port); 26813023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_qp); 26823023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, query_srq); 26831c8fb1eaSYishai Hadas SET_DEVICE_OP(dev_ops, query_ucontext); 26843023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rdma_netdev_get_params); 26853023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, read_counters); 26863023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_dm_mr); 26873023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, reg_user_mr); 26883023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_ncomp_notif); 26893023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, req_notify_cq); 26903023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, rereg_user_mr); 26913023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, resize_cq); 26923023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_guid); 26933023a1e9SKamal Heib SET_DEVICE_OP(dev_ops, set_vf_link_state); 269421a428a0SLeon Romanovsky 2695d3456914SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_ah); 26963b023e1bSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_counters); 2697e39afe3dSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_cq); 269821a428a0SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_pd); 269968e326deSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_srq); 2700a2a074efSLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_ucontext); 270128ad5f65SLeon Romanovsky SET_OBJ_SIZE(dev_ops, ib_xrcd); 2702521ed0d9SKamal Heib } 2703521ed0d9SKamal Heib EXPORT_SYMBOL(ib_set_device_ops); 2704521ed0d9SKamal Heib 2705d0e312feSLeon Romanovsky static const struct rdma_nl_cbs ibnl_ls_cb_table[RDMA_NL_LS_NUM_OPS] = { 2706735c631aSMark Bloch [RDMA_NL_LS_OP_RESOLVE] = { 2707647c75acSLeon Romanovsky .doit = ib_nl_handle_resolve_resp, 2708e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2709e3a2b93dSLeon Romanovsky }, 2710735c631aSMark Bloch [RDMA_NL_LS_OP_SET_TIMEOUT] = { 2711647c75acSLeon Romanovsky .doit = ib_nl_handle_set_timeout, 2712e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2713e3a2b93dSLeon Romanovsky }, 2714ae43f828SMark Bloch [RDMA_NL_LS_OP_IP_RESOLVE] = { 2715647c75acSLeon Romanovsky .doit = ib_nl_handle_ip_res_resp, 2716e3a2b93dSLeon Romanovsky .flags = RDMA_NL_ADMIN_PERM, 2717e3a2b93dSLeon Romanovsky }, 2718735c631aSMark Bloch }; 2719735c631aSMark Bloch 27201da177e4SLinus Torvalds static int __init ib_core_init(void) 27211da177e4SLinus Torvalds { 27221da177e4SLinus Torvalds int ret; 27231da177e4SLinus Torvalds 2724f0626710STejun Heo ib_wq = alloc_workqueue("infiniband", 0, 0); 2725f0626710STejun Heo if (!ib_wq) 2726f0626710STejun Heo return -ENOMEM; 2727f0626710STejun Heo 272814d3a3b2SChristoph Hellwig ib_comp_wq = alloc_workqueue("ib-comp-wq", 2729b7363e67SSagi Grimberg WQ_HIGHPRI | WQ_MEM_RECLAIM | WQ_SYSFS, 0); 273014d3a3b2SChristoph Hellwig if (!ib_comp_wq) { 273114d3a3b2SChristoph Hellwig ret = -ENOMEM; 273214d3a3b2SChristoph Hellwig goto err; 273314d3a3b2SChristoph Hellwig } 273414d3a3b2SChristoph Hellwig 2735f794809aSJack Morgenstein ib_comp_unbound_wq = 2736f794809aSJack Morgenstein alloc_workqueue("ib-comp-unb-wq", 2737f794809aSJack Morgenstein WQ_UNBOUND | WQ_HIGHPRI | WQ_MEM_RECLAIM | 2738f794809aSJack Morgenstein WQ_SYSFS, WQ_UNBOUND_MAX_ACTIVE); 2739f794809aSJack Morgenstein if (!ib_comp_unbound_wq) { 2740f794809aSJack Morgenstein ret = -ENOMEM; 2741f794809aSJack Morgenstein goto err_comp; 2742f794809aSJack Morgenstein } 2743f794809aSJack Morgenstein 274455aeed06SJason Gunthorpe ret = class_register(&ib_class); 2745fd75c789SNir Muchtar if (ret) { 2746aba25a3eSParav Pandit pr_warn("Couldn't create InfiniBand device class\n"); 2747f794809aSJack Morgenstein goto err_comp_unbound; 2748fd75c789SNir Muchtar } 27491da177e4SLinus Torvalds 2750549af008SParav Pandit rdma_nl_init(); 2751549af008SParav Pandit 2752e3f20f02SLeon Romanovsky ret = addr_init(); 2753e3f20f02SLeon Romanovsky if (ret) { 27544469add9SColin Ian King pr_warn("Couldn't init IB address resolution\n"); 2755e3f20f02SLeon Romanovsky goto err_ibnl; 2756e3f20f02SLeon Romanovsky } 2757e3f20f02SLeon Romanovsky 27584c2cb422SMark Bloch ret = ib_mad_init(); 27594c2cb422SMark Bloch if (ret) { 27604c2cb422SMark Bloch pr_warn("Couldn't init IB MAD\n"); 27614c2cb422SMark Bloch goto err_addr; 27624c2cb422SMark Bloch } 27634c2cb422SMark Bloch 2764c2e49c92SMark Bloch ret = ib_sa_init(); 2765c2e49c92SMark Bloch if (ret) { 2766c2e49c92SMark Bloch pr_warn("Couldn't init SA\n"); 2767c2e49c92SMark Bloch goto err_mad; 2768c2e49c92SMark Bloch } 2769c2e49c92SMark Bloch 277042df744cSJanne Karhunen ret = register_blocking_lsm_notifier(&ibdev_lsm_nb); 27718f408ab6SDaniel Jurgens if (ret) { 27728f408ab6SDaniel Jurgens pr_warn("Couldn't register LSM notifier. ret %d\n", ret); 2773c9901724SLeon Romanovsky goto err_sa; 27748f408ab6SDaniel Jurgens } 27758f408ab6SDaniel Jurgens 27764e0f7b90SParav Pandit ret = register_pernet_device(&rdma_dev_net_ops); 27774e0f7b90SParav Pandit if (ret) { 27784e0f7b90SParav Pandit pr_warn("Couldn't init compat dev. ret %d\n", ret); 27794e0f7b90SParav Pandit goto err_compat; 27804e0f7b90SParav Pandit } 27814e0f7b90SParav Pandit 27826c80b41aSLeon Romanovsky nldev_init(); 2783c9901724SLeon Romanovsky rdma_nl_register(RDMA_NL_LS, ibnl_ls_cb_table); 27845ef8c0c1SJason Gunthorpe roce_gid_mgmt_init(); 2785b2cbae2cSRoland Dreier 2786fd75c789SNir Muchtar return 0; 2787fd75c789SNir Muchtar 27884e0f7b90SParav Pandit err_compat: 278942df744cSJanne Karhunen unregister_blocking_lsm_notifier(&ibdev_lsm_nb); 2790735c631aSMark Bloch err_sa: 2791735c631aSMark Bloch ib_sa_cleanup(); 2792c2e49c92SMark Bloch err_mad: 2793c2e49c92SMark Bloch ib_mad_cleanup(); 27944c2cb422SMark Bloch err_addr: 27954c2cb422SMark Bloch addr_cleanup(); 2796e3f20f02SLeon Romanovsky err_ibnl: 279755aeed06SJason Gunthorpe class_unregister(&ib_class); 2798f794809aSJack Morgenstein err_comp_unbound: 2799f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 280014d3a3b2SChristoph Hellwig err_comp: 280114d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2802fd75c789SNir Muchtar err: 2803fd75c789SNir Muchtar destroy_workqueue(ib_wq); 28041da177e4SLinus Torvalds return ret; 28051da177e4SLinus Torvalds } 28061da177e4SLinus Torvalds 28071da177e4SLinus Torvalds static void __exit ib_core_cleanup(void) 28081da177e4SLinus Torvalds { 28095ef8c0c1SJason Gunthorpe roce_gid_mgmt_cleanup(); 28106c80b41aSLeon Romanovsky nldev_exit(); 2811c9901724SLeon Romanovsky rdma_nl_unregister(RDMA_NL_LS); 28124e0f7b90SParav Pandit unregister_pernet_device(&rdma_dev_net_ops); 281342df744cSJanne Karhunen unregister_blocking_lsm_notifier(&ibdev_lsm_nb); 2814c2e49c92SMark Bloch ib_sa_cleanup(); 28154c2cb422SMark Bloch ib_mad_cleanup(); 2816e3f20f02SLeon Romanovsky addr_cleanup(); 2817c9901724SLeon Romanovsky rdma_nl_exit(); 281855aeed06SJason Gunthorpe class_unregister(&ib_class); 2819f794809aSJack Morgenstein destroy_workqueue(ib_comp_unbound_wq); 282014d3a3b2SChristoph Hellwig destroy_workqueue(ib_comp_wq); 2821f7c6a7b5SRoland Dreier /* Make sure that any pending umem accounting work is done. */ 2822f0626710STejun Heo destroy_workqueue(ib_wq); 2823d0899892SJason Gunthorpe flush_workqueue(system_unbound_wq); 2824e59178d8SJason Gunthorpe WARN_ON(!xa_empty(&clients)); 28250df91bb6SJason Gunthorpe WARN_ON(!xa_empty(&devices)); 28261da177e4SLinus Torvalds } 28271da177e4SLinus Torvalds 2828e3bf14bdSJason Gunthorpe MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_LS, 4); 2829e3bf14bdSJason Gunthorpe 283062dfa795SParav Pandit /* ib core relies on netdev stack to first register net_ns_type_operations 283162dfa795SParav Pandit * ns kobject type before ib_core initialization. 283262dfa795SParav Pandit */ 283362dfa795SParav Pandit fs_initcall(ib_core_init); 28341da177e4SLinus Torvalds module_exit(ib_core_cleanup); 2835