Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / infiniband / core / device.c
diff --git a/kernel/drivers/infiniband/core/device.c b/kernel/drivers/infiniband/core/device.c
new file mode 100644 (file)
index 0000000..18c1ece
--- /dev/null
@@ -0,0 +1,785 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/init.h>
+#include <linux/mutex.h>
+#include <rdma/rdma_netlink.h>
+
+#include "core_priv.h"
+
+MODULE_AUTHOR("Roland Dreier");
+MODULE_DESCRIPTION("core kernel InfiniBand API");
+MODULE_LICENSE("Dual BSD/GPL");
+
+struct ib_client_data {
+       struct list_head  list;
+       struct ib_client *client;
+       void *            data;
+};
+
+struct workqueue_struct *ib_wq;
+EXPORT_SYMBOL_GPL(ib_wq);
+
+static LIST_HEAD(device_list);
+static LIST_HEAD(client_list);
+
+/*
+ * device_mutex protects access to both device_list and client_list.
+ * There's no real point to using multiple locks or something fancier
+ * like an rwsem: we always access both lists, and we're always
+ * modifying one list or the other list.  In any case this is not a
+ * hot path so there's no point in trying to optimize.
+ */
+static DEFINE_MUTEX(device_mutex);
+
+static int ib_device_check_mandatory(struct ib_device *device)
+{
+#define IB_MANDATORY_FUNC(x) { offsetof(struct ib_device, x), #x }
+       static const struct {
+               size_t offset;
+               char  *name;
+       } mandatory_table[] = {
+               IB_MANDATORY_FUNC(query_device),
+               IB_MANDATORY_FUNC(query_port),
+               IB_MANDATORY_FUNC(query_pkey),
+               IB_MANDATORY_FUNC(query_gid),
+               IB_MANDATORY_FUNC(alloc_pd),
+               IB_MANDATORY_FUNC(dealloc_pd),
+               IB_MANDATORY_FUNC(create_ah),
+               IB_MANDATORY_FUNC(destroy_ah),
+               IB_MANDATORY_FUNC(create_qp),
+               IB_MANDATORY_FUNC(modify_qp),
+               IB_MANDATORY_FUNC(destroy_qp),
+               IB_MANDATORY_FUNC(post_send),
+               IB_MANDATORY_FUNC(post_recv),
+               IB_MANDATORY_FUNC(create_cq),
+               IB_MANDATORY_FUNC(destroy_cq),
+               IB_MANDATORY_FUNC(poll_cq),
+               IB_MANDATORY_FUNC(req_notify_cq),
+               IB_MANDATORY_FUNC(get_dma_mr),
+               IB_MANDATORY_FUNC(dereg_mr)
+       };
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(mandatory_table); ++i) {
+               if (!*(void **) ((void *) device + mandatory_table[i].offset)) {
+                       printk(KERN_WARNING "Device %s is missing mandatory function %s\n",
+                              device->name, mandatory_table[i].name);
+                       return -EINVAL;
+               }
+       }
+
+       return 0;
+}
+
+static struct ib_device *__ib_device_get_by_name(const char *name)
+{
+       struct ib_device *device;
+
+       list_for_each_entry(device, &device_list, core_list)
+               if (!strncmp(name, device->name, IB_DEVICE_NAME_MAX))
+                       return device;
+
+       return NULL;
+}
+
+
+static int alloc_name(char *name)
+{
+       unsigned long *inuse;
+       char buf[IB_DEVICE_NAME_MAX];
+       struct ib_device *device;
+       int i;
+
+       inuse = (unsigned long *) get_zeroed_page(GFP_KERNEL);
+       if (!inuse)
+               return -ENOMEM;
+
+       list_for_each_entry(device, &device_list, core_list) {
+               if (!sscanf(device->name, name, &i))
+                       continue;
+               if (i < 0 || i >= PAGE_SIZE * 8)
+                       continue;
+               snprintf(buf, sizeof buf, name, i);
+               if (!strncmp(buf, device->name, IB_DEVICE_NAME_MAX))
+                       set_bit(i, inuse);
+       }
+
+       i = find_first_zero_bit(inuse, PAGE_SIZE * 8);
+       free_page((unsigned long) inuse);
+       snprintf(buf, sizeof buf, name, i);
+
+       if (__ib_device_get_by_name(buf))
+               return -ENFILE;
+
+       strlcpy(name, buf, IB_DEVICE_NAME_MAX);
+       return 0;
+}
+
+static int start_port(struct ib_device *device)
+{
+       return (device->node_type == RDMA_NODE_IB_SWITCH) ? 0 : 1;
+}
+
+
+static int end_port(struct ib_device *device)
+{
+       return (device->node_type == RDMA_NODE_IB_SWITCH) ?
+               0 : device->phys_port_cnt;
+}
+
+/**
+ * ib_alloc_device - allocate an IB device struct
+ * @size:size of structure to allocate
+ *
+ * Low-level drivers should use ib_alloc_device() to allocate &struct
+ * ib_device.  @size is the size of the structure to be allocated,
+ * including any private data used by the low-level driver.
+ * ib_dealloc_device() must be used to free structures allocated with
+ * ib_alloc_device().
+ */
+struct ib_device *ib_alloc_device(size_t size)
+{
+       BUG_ON(size < sizeof (struct ib_device));
+
+       return kzalloc(size, GFP_KERNEL);
+}
+EXPORT_SYMBOL(ib_alloc_device);
+
+/**
+ * ib_dealloc_device - free an IB device struct
+ * @device:structure to free
+ *
+ * Free a structure allocated with ib_alloc_device().
+ */
+void ib_dealloc_device(struct ib_device *device)
+{
+       if (device->reg_state == IB_DEV_UNINITIALIZED) {
+               kfree(device);
+               return;
+       }
+
+       BUG_ON(device->reg_state != IB_DEV_UNREGISTERED);
+
+       kobject_put(&device->dev.kobj);
+}
+EXPORT_SYMBOL(ib_dealloc_device);
+
+static int add_client_context(struct ib_device *device, struct ib_client *client)
+{
+       struct ib_client_data *context;
+       unsigned long flags;
+
+       context = kmalloc(sizeof *context, GFP_KERNEL);
+       if (!context) {
+               printk(KERN_WARNING "Couldn't allocate client context for %s/%s\n",
+                      device->name, client->name);
+               return -ENOMEM;
+       }
+
+       context->client = client;
+       context->data   = NULL;
+
+       spin_lock_irqsave(&device->client_data_lock, flags);
+       list_add(&context->list, &device->client_data_list);
+       spin_unlock_irqrestore(&device->client_data_lock, flags);
+
+       return 0;
+}
+
+static int read_port_table_lengths(struct ib_device *device)
+{
+       struct ib_port_attr *tprops = NULL;
+       int num_ports, ret = -ENOMEM;
+       u8 port_index;
+
+       tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
+       if (!tprops)
+               goto out;
+
+       num_ports = end_port(device) - start_port(device) + 1;
+
+       device->pkey_tbl_len = kmalloc(sizeof *device->pkey_tbl_len * num_ports,
+                                      GFP_KERNEL);
+       device->gid_tbl_len = kmalloc(sizeof *device->gid_tbl_len * num_ports,
+                                     GFP_KERNEL);
+       if (!device->pkey_tbl_len || !device->gid_tbl_len)
+               goto err;
+
+       for (port_index = 0; port_index < num_ports; ++port_index) {
+               ret = ib_query_port(device, port_index + start_port(device),
+                                       tprops);
+               if (ret)
+                       goto err;
+               device->pkey_tbl_len[port_index] = tprops->pkey_tbl_len;
+               device->gid_tbl_len[port_index]  = tprops->gid_tbl_len;
+       }
+
+       ret = 0;
+       goto out;
+
+err:
+       kfree(device->gid_tbl_len);
+       kfree(device->pkey_tbl_len);
+out:
+       kfree(tprops);
+       return ret;
+}
+
+/**
+ * ib_register_device - Register an IB device with IB core
+ * @device:Device to register
+ *
+ * Low-level drivers use ib_register_device() to register their
+ * devices with the IB core.  All registered clients will receive a
+ * callback for each device that is added. @device must be allocated
+ * with ib_alloc_device().
+ */
+int ib_register_device(struct ib_device *device,
+                      int (*port_callback)(struct ib_device *,
+                                           u8, struct kobject *))
+{
+       int ret;
+
+       mutex_lock(&device_mutex);
+
+       if (strchr(device->name, '%')) {
+               ret = alloc_name(device->name);
+               if (ret)
+                       goto out;
+       }
+
+       if (ib_device_check_mandatory(device)) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       INIT_LIST_HEAD(&device->event_handler_list);
+       INIT_LIST_HEAD(&device->client_data_list);
+       spin_lock_init(&device->event_handler_lock);
+       spin_lock_init(&device->client_data_lock);
+
+       ret = read_port_table_lengths(device);
+       if (ret) {
+               printk(KERN_WARNING "Couldn't create table lengths cache for device %s\n",
+                      device->name);
+               goto out;
+       }
+
+       ret = ib_device_register_sysfs(device, port_callback);
+       if (ret) {
+               printk(KERN_WARNING "Couldn't register device %s with driver model\n",
+                      device->name);
+               kfree(device->gid_tbl_len);
+               kfree(device->pkey_tbl_len);
+               goto out;
+       }
+
+       list_add_tail(&device->core_list, &device_list);
+
+       device->reg_state = IB_DEV_REGISTERED;
+
+       {
+               struct ib_client *client;
+
+               list_for_each_entry(client, &client_list, list)
+                       if (client->add && !add_client_context(device, client))
+                               client->add(device);
+       }
+
+ out:
+       mutex_unlock(&device_mutex);
+       return ret;
+}
+EXPORT_SYMBOL(ib_register_device);
+
+/**
+ * ib_unregister_device - Unregister an IB device
+ * @device:Device to unregister
+ *
+ * Unregister an IB device.  All clients will receive a remove callback.
+ */
+void ib_unregister_device(struct ib_device *device)
+{
+       struct ib_client *client;
+       struct ib_client_data *context, *tmp;
+       unsigned long flags;
+
+       mutex_lock(&device_mutex);
+
+       list_for_each_entry_reverse(client, &client_list, list)
+               if (client->remove)
+                       client->remove(device);
+
+       list_del(&device->core_list);
+
+       kfree(device->gid_tbl_len);
+       kfree(device->pkey_tbl_len);
+
+       mutex_unlock(&device_mutex);
+
+       ib_device_unregister_sysfs(device);
+
+       spin_lock_irqsave(&device->client_data_lock, flags);
+       list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
+               kfree(context);
+       spin_unlock_irqrestore(&device->client_data_lock, flags);
+
+       device->reg_state = IB_DEV_UNREGISTERED;
+}
+EXPORT_SYMBOL(ib_unregister_device);
+
+/**
+ * ib_register_client - Register an IB client
+ * @client:Client to register
+ *
+ * Upper level users of the IB drivers can use ib_register_client() to
+ * register callbacks for IB device addition and removal.  When an IB
+ * device is added, each registered client's add method will be called
+ * (in the order the clients were registered), and when a device is
+ * removed, each client's remove method will be called (in the reverse
+ * order that clients were registered).  In addition, when
+ * ib_register_client() is called, the client will receive an add
+ * callback for all devices already registered.
+ */
+int ib_register_client(struct ib_client *client)
+{
+       struct ib_device *device;
+
+       mutex_lock(&device_mutex);
+
+       list_add_tail(&client->list, &client_list);
+       list_for_each_entry(device, &device_list, core_list)
+               if (client->add && !add_client_context(device, client))
+                       client->add(device);
+
+       mutex_unlock(&device_mutex);
+
+       return 0;
+}
+EXPORT_SYMBOL(ib_register_client);
+
+/**
+ * ib_unregister_client - Unregister an IB client
+ * @client:Client to unregister
+ *
+ * Upper level users use ib_unregister_client() to remove their client
+ * registration.  When ib_unregister_client() is called, the client
+ * will receive a remove callback for each IB device still registered.
+ */
+void ib_unregister_client(struct ib_client *client)
+{
+       struct ib_client_data *context, *tmp;
+       struct ib_device *device;
+       unsigned long flags;
+
+       mutex_lock(&device_mutex);
+
+       list_for_each_entry(device, &device_list, core_list) {
+               if (client->remove)
+                       client->remove(device);
+
+               spin_lock_irqsave(&device->client_data_lock, flags);
+               list_for_each_entry_safe(context, tmp, &device->client_data_list, list)
+                       if (context->client == client) {
+                               list_del(&context->list);
+                               kfree(context);
+                       }
+               spin_unlock_irqrestore(&device->client_data_lock, flags);
+       }
+       list_del(&client->list);
+
+       mutex_unlock(&device_mutex);
+}
+EXPORT_SYMBOL(ib_unregister_client);
+
+/**
+ * ib_get_client_data - Get IB client context
+ * @device:Device to get context for
+ * @client:Client to get context for
+ *
+ * ib_get_client_data() returns client context set with
+ * ib_set_client_data().
+ */
+void *ib_get_client_data(struct ib_device *device, struct ib_client *client)
+{
+       struct ib_client_data *context;
+       void *ret = NULL;
+       unsigned long flags;
+
+       spin_lock_irqsave(&device->client_data_lock, flags);
+       list_for_each_entry(context, &device->client_data_list, list)
+               if (context->client == client) {
+                       ret = context->data;
+                       break;
+               }
+       spin_unlock_irqrestore(&device->client_data_lock, flags);
+
+       return ret;
+}
+EXPORT_SYMBOL(ib_get_client_data);
+
+/**
+ * ib_set_client_data - Set IB client context
+ * @device:Device to set context for
+ * @client:Client to set context for
+ * @data:Context to set
+ *
+ * ib_set_client_data() sets client context that can be retrieved with
+ * ib_get_client_data().
+ */
+void ib_set_client_data(struct ib_device *device, struct ib_client *client,
+                       void *data)
+{
+       struct ib_client_data *context;
+       unsigned long flags;
+
+       spin_lock_irqsave(&device->client_data_lock, flags);
+       list_for_each_entry(context, &device->client_data_list, list)
+               if (context->client == client) {
+                       context->data = data;
+                       goto out;
+               }
+
+       printk(KERN_WARNING "No client context found for %s/%s\n",
+              device->name, client->name);
+
+out:
+       spin_unlock_irqrestore(&device->client_data_lock, flags);
+}
+EXPORT_SYMBOL(ib_set_client_data);
+
+/**
+ * ib_register_event_handler - Register an IB event handler
+ * @event_handler:Handler to register
+ *
+ * ib_register_event_handler() registers an event handler that will be
+ * called back when asynchronous IB events occur (as defined in
+ * chapter 11 of the InfiniBand Architecture Specification).  This
+ * callback may occur in interrupt context.
+ */
+int ib_register_event_handler  (struct ib_event_handler *event_handler)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
+       list_add_tail(&event_handler->list,
+                     &event_handler->device->event_handler_list);
+       spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
+
+       return 0;
+}
+EXPORT_SYMBOL(ib_register_event_handler);
+
+/**
+ * ib_unregister_event_handler - Unregister an event handler
+ * @event_handler:Handler to unregister
+ *
+ * Unregister an event handler registered with
+ * ib_register_event_handler().
+ */
+int ib_unregister_event_handler(struct ib_event_handler *event_handler)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&event_handler->device->event_handler_lock, flags);
+       list_del(&event_handler->list);
+       spin_unlock_irqrestore(&event_handler->device->event_handler_lock, flags);
+
+       return 0;
+}
+EXPORT_SYMBOL(ib_unregister_event_handler);
+
+/**
+ * ib_dispatch_event - Dispatch an asynchronous event
+ * @event:Event to dispatch
+ *
+ * Low-level drivers must call ib_dispatch_event() to dispatch the
+ * event to all registered event handlers when an asynchronous event
+ * occurs.
+ */
+void ib_dispatch_event(struct ib_event *event)
+{
+       unsigned long flags;
+       struct ib_event_handler *handler;
+
+       spin_lock_irqsave(&event->device->event_handler_lock, flags);
+
+       list_for_each_entry(handler, &event->device->event_handler_list, list)
+               handler->handler(handler, event);
+
+       spin_unlock_irqrestore(&event->device->event_handler_lock, flags);
+}
+EXPORT_SYMBOL(ib_dispatch_event);
+
+/**
+ * ib_query_device - Query IB device attributes
+ * @device:Device to query
+ * @device_attr:Device attributes
+ *
+ * ib_query_device() returns the attributes of a device through the
+ * @device_attr pointer.
+ */
+int ib_query_device(struct ib_device *device,
+                   struct ib_device_attr *device_attr)
+{
+       return device->query_device(device, device_attr);
+}
+EXPORT_SYMBOL(ib_query_device);
+
+/**
+ * ib_query_port - Query IB port attributes
+ * @device:Device to query
+ * @port_num:Port number to query
+ * @port_attr:Port attributes
+ *
+ * ib_query_port() returns the attributes of a port through the
+ * @port_attr pointer.
+ */
+int ib_query_port(struct ib_device *device,
+                 u8 port_num,
+                 struct ib_port_attr *port_attr)
+{
+       if (port_num < start_port(device) || port_num > end_port(device))
+               return -EINVAL;
+
+       return device->query_port(device, port_num, port_attr);
+}
+EXPORT_SYMBOL(ib_query_port);
+
+/**
+ * ib_query_gid - Get GID table entry
+ * @device:Device to query
+ * @port_num:Port number to query
+ * @index:GID table index to query
+ * @gid:Returned GID
+ *
+ * ib_query_gid() fetches the specified GID table entry.
+ */
+int ib_query_gid(struct ib_device *device,
+                u8 port_num, int index, union ib_gid *gid)
+{
+       return device->query_gid(device, port_num, index, gid);
+}
+EXPORT_SYMBOL(ib_query_gid);
+
+/**
+ * ib_query_pkey - Get P_Key table entry
+ * @device:Device to query
+ * @port_num:Port number to query
+ * @index:P_Key table index to query
+ * @pkey:Returned P_Key
+ *
+ * ib_query_pkey() fetches the specified P_Key table entry.
+ */
+int ib_query_pkey(struct ib_device *device,
+                 u8 port_num, u16 index, u16 *pkey)
+{
+       return device->query_pkey(device, port_num, index, pkey);
+}
+EXPORT_SYMBOL(ib_query_pkey);
+
+/**
+ * ib_modify_device - Change IB device attributes
+ * @device:Device to modify
+ * @device_modify_mask:Mask of attributes to change
+ * @device_modify:New attribute values
+ *
+ * ib_modify_device() changes a device's attributes as specified by
+ * the @device_modify_mask and @device_modify structure.
+ */
+int ib_modify_device(struct ib_device *device,
+                    int device_modify_mask,
+                    struct ib_device_modify *device_modify)
+{
+       if (!device->modify_device)
+               return -ENOSYS;
+
+       return device->modify_device(device, device_modify_mask,
+                                    device_modify);
+}
+EXPORT_SYMBOL(ib_modify_device);
+
+/**
+ * ib_modify_port - Modifies the attributes for the specified port.
+ * @device: The device to modify.
+ * @port_num: The number of the port to modify.
+ * @port_modify_mask: Mask used to specify which attributes of the port
+ *   to change.
+ * @port_modify: New attribute values for the port.
+ *
+ * ib_modify_port() changes a port's attributes as specified by the
+ * @port_modify_mask and @port_modify structure.
+ */
+int ib_modify_port(struct ib_device *device,
+                  u8 port_num, int port_modify_mask,
+                  struct ib_port_modify *port_modify)
+{
+       if (!device->modify_port)
+               return -ENOSYS;
+
+       if (port_num < start_port(device) || port_num > end_port(device))
+               return -EINVAL;
+
+       return device->modify_port(device, port_num, port_modify_mask,
+                                  port_modify);
+}
+EXPORT_SYMBOL(ib_modify_port);
+
+/**
+ * ib_find_gid - Returns the port number and GID table index where
+ *   a specified GID value occurs.
+ * @device: The device to query.
+ * @gid: The GID value to search for.
+ * @port_num: The port number of the device where the GID value was found.
+ * @index: The index into the GID table where the GID was found.  This
+ *   parameter may be NULL.
+ */
+int ib_find_gid(struct ib_device *device, union ib_gid *gid,
+               u8 *port_num, u16 *index)
+{
+       union ib_gid tmp_gid;
+       int ret, port, i;
+
+       for (port = start_port(device); port <= end_port(device); ++port) {
+               for (i = 0; i < device->gid_tbl_len[port - start_port(device)]; ++i) {
+                       ret = ib_query_gid(device, port, i, &tmp_gid);
+                       if (ret)
+                               return ret;
+                       if (!memcmp(&tmp_gid, gid, sizeof *gid)) {
+                               *port_num = port;
+                               if (index)
+                                       *index = i;
+                               return 0;
+                       }
+               }
+       }
+
+       return -ENOENT;
+}
+EXPORT_SYMBOL(ib_find_gid);
+
+/**
+ * ib_find_pkey - Returns the PKey table index where a specified
+ *   PKey value occurs.
+ * @device: The device to query.
+ * @port_num: The port number of the device to search for the PKey.
+ * @pkey: The PKey value to search for.
+ * @index: The index into the PKey table where the PKey was found.
+ */
+int ib_find_pkey(struct ib_device *device,
+                u8 port_num, u16 pkey, u16 *index)
+{
+       int ret, i;
+       u16 tmp_pkey;
+       int partial_ix = -1;
+
+       for (i = 0; i < device->pkey_tbl_len[port_num - start_port(device)]; ++i) {
+               ret = ib_query_pkey(device, port_num, i, &tmp_pkey);
+               if (ret)
+                       return ret;
+               if ((pkey & 0x7fff) == (tmp_pkey & 0x7fff)) {
+                       /* if there is full-member pkey take it.*/
+                       if (tmp_pkey & 0x8000) {
+                               *index = i;
+                               return 0;
+                       }
+                       if (partial_ix < 0)
+                               partial_ix = i;
+               }
+       }
+
+       /*no full-member, if exists take the limited*/
+       if (partial_ix >= 0) {
+               *index = partial_ix;
+               return 0;
+       }
+       return -ENOENT;
+}
+EXPORT_SYMBOL(ib_find_pkey);
+
+static int __init ib_core_init(void)
+{
+       int ret;
+
+       ib_wq = alloc_workqueue("infiniband", 0, 0);
+       if (!ib_wq)
+               return -ENOMEM;
+
+       ret = ib_sysfs_setup();
+       if (ret) {
+               printk(KERN_WARNING "Couldn't create InfiniBand device class\n");
+               goto err;
+       }
+
+       ret = ibnl_init();
+       if (ret) {
+               printk(KERN_WARNING "Couldn't init IB netlink interface\n");
+               goto err_sysfs;
+       }
+
+       ret = ib_cache_setup();
+       if (ret) {
+               printk(KERN_WARNING "Couldn't set up InfiniBand P_Key/GID cache\n");
+               goto err_nl;
+       }
+
+       return 0;
+
+err_nl:
+       ibnl_cleanup();
+
+err_sysfs:
+       ib_sysfs_cleanup();
+
+err:
+       destroy_workqueue(ib_wq);
+       return ret;
+}
+
+static void __exit ib_core_cleanup(void)
+{
+       ib_cache_cleanup();
+       ibnl_cleanup();
+       ib_sysfs_cleanup();
+       /* Make sure that any pending umem accounting work is done. */
+       destroy_workqueue(ib_wq);
+}
+
+module_init(ib_core_init);
+module_exit(ib_core_cleanup);