cur_cm_id_priv = rb_entry(parent, struct cm_id_private,
service_node);
if ((cur_cm_id_priv->id.service_mask & service_id) ==
- (service_mask & cur_cm_id_priv->id.service_id))
- return cm_id_priv;
- if (service_id < cur_cm_id_priv->id.service_id)
+ (service_mask & cur_cm_id_priv->id.service_id) &&
+ (cm_id_priv->id.device == cur_cm_id_priv->id.device))
+ return cur_cm_id_priv;
+
+ if (cm_id_priv->id.device < cur_cm_id_priv->id.device)
+ link = &(*link)->rb_left;
+ else if (cm_id_priv->id.device > cur_cm_id_priv->id.device)
+ link = &(*link)->rb_right;
+ else if (service_id < cur_cm_id_priv->id.service_id)
link = &(*link)->rb_left;
else
link = &(*link)->rb_right;
return NULL;
}
-static struct cm_id_private * cm_find_listen(__be64 service_id)
+static struct cm_id_private * cm_find_listen(struct ib_device *device,
+ __be64 service_id)
{
struct rb_node *node = cm.listen_service_table.rb_node;
struct cm_id_private *cm_id_priv;
while (node) {
cm_id_priv = rb_entry(node, struct cm_id_private, service_node);
if ((cm_id_priv->id.service_mask & service_id) ==
- (cm_id_priv->id.service_mask & cm_id_priv->id.service_id))
+ cm_id_priv->id.service_id &&
+ (cm_id_priv->id.device == device))
return cm_id_priv;
- if (service_id < cm_id_priv->id.service_id)
+
+ if (device < cm_id_priv->id.device)
+ node = node->rb_left;
+ else if (device > cm_id_priv->id.device)
+ node = node->rb_right;
+ else if (service_id < cm_id_priv->id.service_id)
node = node->rb_left;
else
node = node->rb_right;
ib_send_cm_sidr_rep(&cm_id_priv->id, ¶m);
}
-struct ib_cm_id *ib_create_cm_id(ib_cm_handler cm_handler,
+struct ib_cm_id *ib_create_cm_id(struct ib_device *device,
+ ib_cm_handler cm_handler,
void *context)
{
struct cm_id_private *cm_id_priv;
memset(cm_id_priv, 0, sizeof *cm_id_priv);
cm_id_priv->id.state = IB_CM_IDLE;
+ cm_id_priv->id.device = device;
cm_id_priv->id.cm_handler = cm_handler;
cm_id_priv->id.context = context;
cm_id_priv->id.remote_cm_qpn = 1;
req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
param = &work->cm_event.param.req_rcvd;
param->listen_id = listen_id;
- param->device = cm_id_priv->av.port->mad_agent->device;
param->port = cm_id_priv->av.port->port_num;
param->primary_path = &work->path[0];
if (req_msg->alt_local_lid)
}
/* Find matching listen request. */
- listen_cm_id_priv = cm_find_listen(req_msg->service_id);
+ listen_cm_id_priv = cm_find_listen(cm_id_priv->id.device,
+ req_msg->service_id);
if (!listen_cm_id_priv) {
spin_unlock_irqrestore(&cm.lock, flags);
cm_issue_rej(work->port, work->mad_recv_wc,
req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
- cm_id = ib_create_cm_id(NULL, NULL);
+ cm_id = ib_create_cm_id(work->port->cm_dev->device, NULL, NULL);
if (IS_ERR(cm_id))
return PTR_ERR(cm_id);
param = &work->cm_event.param.sidr_req_rcvd;
param->pkey = __be16_to_cpu(sidr_req_msg->pkey);
param->listen_id = listen_id;
- param->device = work->port->mad_agent->device;
param->port = work->port->port_num;
work->cm_event.private_data = &sidr_req_msg->private_data;
}
struct ib_wc *wc;
unsigned long flags;
- cm_id = ib_create_cm_id(NULL, NULL);
+ cm_id = ib_create_cm_id(work->port->cm_dev->device, NULL, NULL);
if (IS_ERR(cm_id))
return PTR_ERR(cm_id);
cm_id_priv = container_of(cm_id, struct cm_id_private, id);
spin_unlock_irqrestore(&cm.lock, flags);
goto out; /* Duplicate message. */
}
- cur_cm_id_priv = cm_find_listen(sidr_req_msg->service_id);
+ cur_cm_id_priv = cm_find_listen(cm_id->device,
+ sidr_req_msg->service_id);
if (!cur_cm_id_priv) {
rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
spin_unlock_irqrestore(&cm.lock, flags);
flush_workqueue(cm.wq);
destroy_workqueue(cm.wq);
ib_unregister_client(&cm_client);
+ idr_destroy(&cm.local_id_table);
}
module_init(ib_cm_init);
{
struct ib_sa_path_query *query;
struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
- struct ib_sa_port *port = &sa_dev->port[port_num - sa_dev->start_port];
- struct ib_mad_agent *agent = port->agent;
+ struct ib_sa_port *port;
+ struct ib_mad_agent *agent;
int ret;
+ if (!sa_dev)
+ return -ENODEV;
+
+ port = &sa_dev->port[port_num - sa_dev->start_port];
+ agent = port->agent;
+
query = kmalloc(sizeof *query, gfp_mask);
if (!query)
return -ENOMEM;
{
struct ib_sa_service_query *query;
struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
- struct ib_sa_port *port = &sa_dev->port[port_num - sa_dev->start_port];
- struct ib_mad_agent *agent = port->agent;
+ struct ib_sa_port *port;
+ struct ib_mad_agent *agent;
int ret;
+ if (!sa_dev)
+ return -ENODEV;
+
+ port = &sa_dev->port[port_num - sa_dev->start_port];
+ agent = port->agent;
+
if (method != IB_MGMT_METHOD_GET &&
method != IB_MGMT_METHOD_SET &&
method != IB_SA_METHOD_DELETE)
{
struct ib_sa_mcmember_query *query;
struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
- struct ib_sa_port *port = &sa_dev->port[port_num - sa_dev->start_port];
- struct ib_mad_agent *agent = port->agent;
+ struct ib_sa_port *port;
+ struct ib_mad_agent *agent;
int ret;
+ if (!sa_dev)
+ return -ENODEV;
+
+ port = &sa_dev->port[port_num - sa_dev->start_port];
+ agent = port->agent;
+
query = kmalloc(sizeof *query, gfp_mask);
if (!query)
return -ENOMEM;
static void __exit ib_sa_cleanup(void)
{
ib_unregister_client(&sa_client);
+ idr_destroy(&query_idr);
}
module_init(ib_sa_init);
#include <linux/file.h>
#include <linux/mount.h>
#include <linux/cdev.h>
+#include <linux/idr.h>
#include <asm/uaccess.h>
-#include "ucm.h"
+#include <rdma/ib_cm.h>
+#include <rdma/ib_user_cm.h>
MODULE_AUTHOR("Libor Michalek");
MODULE_DESCRIPTION("InfiniBand userspace Connection Manager access");
MODULE_LICENSE("Dual BSD/GPL");
-static int ucm_debug_level;
+struct ib_ucm_device {
+ int devnum;
+ struct cdev dev;
+ struct class_device class_dev;
+ struct ib_device *ib_dev;
+};
+
+struct ib_ucm_file {
+ struct semaphore mutex;
+ struct file *filp;
+ struct ib_ucm_device *device;
+
+ struct list_head ctxs;
+ struct list_head events;
+ wait_queue_head_t poll_wait;
+};
+
+struct ib_ucm_context {
+ int id;
+ wait_queue_head_t wait;
+ atomic_t ref;
+ int events_reported;
+
+ struct ib_ucm_file *file;
+ struct ib_cm_id *cm_id;
+ __u64 uid;
+
+ struct list_head events; /* list of pending events. */
+ struct list_head file_list; /* member in file ctx list */
+};
+
+struct ib_ucm_event {
+ struct ib_ucm_context *ctx;
+ struct list_head file_list; /* member in file event list */
+ struct list_head ctx_list; /* member in ctx event list */
-module_param_named(debug_level, ucm_debug_level, int, 0644);
-MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
+ struct ib_cm_id *cm_id;
+ struct ib_ucm_event_resp resp;
+ void *data;
+ void *info;
+ int data_len;
+ int info_len;
+};
enum {
IB_UCM_MAJOR = 231,
- IB_UCM_MINOR = 255
+ IB_UCM_BASE_MINOR = 224,
+ IB_UCM_MAX_DEVICES = 32
};
-#define IB_UCM_DEV MKDEV(IB_UCM_MAJOR, IB_UCM_MINOR)
+#define IB_UCM_BASE_DEV MKDEV(IB_UCM_MAJOR, IB_UCM_BASE_MINOR)
-#define PFX "UCM: "
+static void ib_ucm_add_one(struct ib_device *device);
+static void ib_ucm_remove_one(struct ib_device *device);
-#define ucm_dbg(format, arg...) \
- do { \
- if (ucm_debug_level > 0) \
- printk(KERN_DEBUG PFX format, ## arg); \
- } while (0)
+static struct ib_client ucm_client = {
+ .name = "ucm",
+ .add = ib_ucm_add_one,
+ .remove = ib_ucm_remove_one
+};
-static struct semaphore ctx_id_mutex;
-static struct idr ctx_id_table;
+static DECLARE_MUTEX(ctx_id_mutex);
+static DEFINE_IDR(ctx_id_table);
+static DECLARE_BITMAP(dev_map, IB_UCM_MAX_DEVICES);
static struct ib_ucm_context *ib_ucm_ctx_get(struct ib_ucm_file *file, int id)
{
goto error;
list_add_tail(&ctx->file_list, &file->ctxs);
- ucm_dbg("Allocated CM ID <%d>\n", ctx->id);
return ctx;
error:
kfree(ctx);
return NULL;
}
-/*
- * Event portion of the API, handle CM events
- * and allow event polling.
- */
+
static void ib_ucm_event_path_get(struct ib_ucm_path_rec *upath,
struct ib_sa_path_rec *kpath)
{
ureq->retry_count = kreq->retry_count;
ureq->rnr_retry_count = kreq->rnr_retry_count;
ureq->srq = kreq->srq;
+ ureq->port = kreq->port;
ib_ucm_event_path_get(&ureq->primary_path, kreq->primary_path);
ib_ucm_event_path_get(&ureq->alternate_path, kreq->alternate_path);
case IB_CM_SIDR_REQ_RECEIVED:
uvt->resp.u.sidr_req_resp.pkey =
evt->param.sidr_req_rcvd.pkey;
+ uvt->resp.u.sidr_req_resp.port =
+ evt->param.sidr_req_rcvd.port;
uvt->data_len = IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE;
break;
case IB_CM_SIDR_REP_RECEIVED:
if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
return -EFAULT;
- /*
- * wait
- */
+
down(&file->mutex);
while (list_empty(&file->events)) {
return result;
}
-
static ssize_t ib_ucm_create_id(struct ib_ucm_file *file,
const char __user *inbuf,
int in_len, int out_len)
return -ENOMEM;
ctx->uid = cmd.uid;
- ctx->cm_id = ib_create_cm_id(ib_ucm_event_handler, ctx);
+ ctx->cm_id = ib_create_cm_id(file->device->ib_dev,
+ ib_ucm_event_handler, ctx);
if (IS_ERR(ctx->cm_id)) {
result = PTR_ERR(ctx->cm_id);
- goto err;
+ goto err1;
}
resp.id = ctx->id;
if (copy_to_user((void __user *)(unsigned long)cmd.response,
&resp, sizeof(resp))) {
result = -EFAULT;
- goto err;
+ goto err2;
}
-
return 0;
-err:
+err2:
+ ib_destroy_cm_id(ctx->cm_id);
+err1:
down(&ctx_id_mutex);
idr_remove(&ctx_id_table, ctx->id);
up(&ctx_id_mutex);
-
- if (!IS_ERR(ctx->cm_id))
- ib_destroy_cm_id(ctx->cm_id);
-
kfree(ctx);
return result;
}
if (copy_from_user(&hdr, buf, sizeof(hdr)))
return -EFAULT;
- ucm_dbg("Write. cmd <%d> in <%d> out <%d> len <%Zu>\n",
- hdr.cmd, hdr.in, hdr.out, len);
-
if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucm_cmd_table))
return -EINVAL;
filp->private_data = file;
file->filp = filp;
-
- ucm_dbg("Created struct\n");
+ file->device = container_of(inode->i_cdev, struct ib_ucm_device, dev);
return 0;
}
return 0;
}
-static struct file_operations ib_ucm_fops = {
+static void ib_ucm_release_class_dev(struct class_device *class_dev)
+{
+ struct ib_ucm_device *dev;
+
+ dev = container_of(class_dev, struct ib_ucm_device, class_dev);
+ cdev_del(&dev->dev);
+ clear_bit(dev->devnum, dev_map);
+ kfree(dev);
+}
+
+static struct file_operations ucm_fops = {
.owner = THIS_MODULE,
.open = ib_ucm_open,
.release = ib_ucm_close,
.poll = ib_ucm_poll,
};
+static struct class ucm_class = {
+ .name = "infiniband_cm",
+ .release = ib_ucm_release_class_dev
+};
-static struct class *ib_ucm_class;
-static struct cdev ib_ucm_cdev;
+static ssize_t show_dev(struct class_device *class_dev, char *buf)
+{
+ struct ib_ucm_device *dev;
+
+ dev = container_of(class_dev, struct ib_ucm_device, class_dev);
+ return print_dev_t(buf, dev->dev.dev);
+}
+static CLASS_DEVICE_ATTR(dev, S_IRUGO, show_dev, NULL);
-static int __init ib_ucm_init(void)
+static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
{
- int result;
+ struct ib_ucm_device *dev;
+
+ dev = container_of(class_dev, struct ib_ucm_device, class_dev);
+ return sprintf(buf, "%s\n", dev->ib_dev->name);
+}
+static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
- result = register_chrdev_region(IB_UCM_DEV, 1, "infiniband_cm");
- if (result) {
- ucm_dbg("Error <%d> registering dev\n", result);
- goto err_chr;
- }
+static void ib_ucm_add_one(struct ib_device *device)
+{
+ struct ib_ucm_device *ucm_dev;
+
+ if (!device->alloc_ucontext)
+ return;
+
+ ucm_dev = kmalloc(sizeof *ucm_dev, GFP_KERNEL);
+ if (!ucm_dev)
+ return;
- cdev_init(&ib_ucm_cdev, &ib_ucm_fops);
+ memset(ucm_dev, 0, sizeof *ucm_dev);
+ ucm_dev->ib_dev = device;
+
+ ucm_dev->devnum = find_first_zero_bit(dev_map, IB_UCM_MAX_DEVICES);
+ if (ucm_dev->devnum >= IB_UCM_MAX_DEVICES)
+ goto err;
+
+ set_bit(ucm_dev->devnum, dev_map);
+
+ cdev_init(&ucm_dev->dev, &ucm_fops);
+ ucm_dev->dev.owner = THIS_MODULE;
+ kobject_set_name(&ucm_dev->dev.kobj, "ucm%d", ucm_dev->devnum);
+ if (cdev_add(&ucm_dev->dev, IB_UCM_BASE_DEV + ucm_dev->devnum, 1))
+ goto err;
- result = cdev_add(&ib_ucm_cdev, IB_UCM_DEV, 1);
- if (result) {
- ucm_dbg("Error <%d> adding cdev\n", result);
+ ucm_dev->class_dev.class = &ucm_class;
+ ucm_dev->class_dev.dev = device->dma_device;
+ snprintf(ucm_dev->class_dev.class_id, BUS_ID_SIZE, "ucm%d",
+ ucm_dev->devnum);
+ if (class_device_register(&ucm_dev->class_dev))
goto err_cdev;
- }
- ib_ucm_class = class_create(THIS_MODULE, "infiniband_cm");
- if (IS_ERR(ib_ucm_class)) {
- result = PTR_ERR(ib_ucm_class);
- ucm_dbg("Error <%d> creating class\n", result);
+ if (class_device_create_file(&ucm_dev->class_dev,
+ &class_device_attr_dev))
+ goto err_class;
+ if (class_device_create_file(&ucm_dev->class_dev,
+ &class_device_attr_ibdev))
goto err_class;
+
+ ib_set_client_data(device, &ucm_client, ucm_dev);
+ return;
+
+err_class:
+ class_device_unregister(&ucm_dev->class_dev);
+err_cdev:
+ cdev_del(&ucm_dev->dev);
+ clear_bit(ucm_dev->devnum, dev_map);
+err:
+ kfree(ucm_dev);
+ return;
+}
+
+static void ib_ucm_remove_one(struct ib_device *device)
+{
+ struct ib_ucm_device *ucm_dev = ib_get_client_data(device, &ucm_client);
+
+ if (!ucm_dev)
+ return;
+
+ class_device_unregister(&ucm_dev->class_dev);
+}
+
+static ssize_t show_abi_version(struct class *class, char *buf)
+{
+ return sprintf(buf, "%d\n", IB_USER_CM_ABI_VERSION);
+}
+static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
+
+static int __init ib_ucm_init(void)
+{
+ int ret;
+
+ ret = register_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES,
+ "infiniband_cm");
+ if (ret) {
+ printk(KERN_ERR "ucm: couldn't register device number\n");
+ goto err;
}
- class_device_create(ib_ucm_class, IB_UCM_DEV, NULL, "ucm");
+ ret = class_register(&ucm_class);
+ if (ret) {
+ printk(KERN_ERR "ucm: couldn't create class infiniband_cm\n");
+ goto err_chrdev;
+ }
- idr_init(&ctx_id_table);
- init_MUTEX(&ctx_id_mutex);
+ ret = class_create_file(&ucm_class, &class_attr_abi_version);
+ if (ret) {
+ printk(KERN_ERR "ucm: couldn't create abi_version attribute\n");
+ goto err_class;
+ }
+ ret = ib_register_client(&ucm_client);
+ if (ret) {
+ printk(KERN_ERR "ucm: couldn't register client\n");
+ goto err_class;
+ }
return 0;
+
err_class:
- cdev_del(&ib_ucm_cdev);
-err_cdev:
- unregister_chrdev_region(IB_UCM_DEV, 1);
-err_chr:
- return result;
+ class_unregister(&ucm_class);
+err_chrdev:
+ unregister_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES);
+err:
+ return ret;
}
static void __exit ib_ucm_cleanup(void)
{
- class_device_destroy(ib_ucm_class, IB_UCM_DEV);
- class_destroy(ib_ucm_class);
- cdev_del(&ib_ucm_cdev);
- unregister_chrdev_region(IB_UCM_DEV, 1);
+ ib_unregister_client(&ucm_client);
+ class_unregister(&ucm_class);
+ unregister_chrdev_region(IB_UCM_BASE_DEV, IB_UCM_MAX_DEVICES);
+ idr_destroy(&ctx_id_table);
}
module_init(ib_ucm_init);
* Copyright (c) 2005 Cisco Systems. All rights reserved.
* Copyright (c) 2005 Mellanox Technologies. All rights reserved.
* Copyright (c) 2005 Voltaire, Inc. All rights reserved.
+ * Copyright (c) 2005 PathScale, 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
static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
const char __user *buf, int in_len,
int out_len) = {
- [IB_USER_VERBS_CMD_QUERY_PARAMS] = ib_uverbs_query_params,
- [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
- [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
- [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
- [IB_USER_VERBS_CMD_QUERY_GID] = ib_uverbs_query_gid,
- [IB_USER_VERBS_CMD_QUERY_PKEY] = ib_uverbs_query_pkey,
- [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
- [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
- [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
- [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
- [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
- [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
- [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
- [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
- [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
- [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
- [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
- [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq,
- [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
- [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
+ [IB_USER_VERBS_CMD_GET_CONTEXT] = ib_uverbs_get_context,
+ [IB_USER_VERBS_CMD_QUERY_DEVICE] = ib_uverbs_query_device,
+ [IB_USER_VERBS_CMD_QUERY_PORT] = ib_uverbs_query_port,
+ [IB_USER_VERBS_CMD_ALLOC_PD] = ib_uverbs_alloc_pd,
+ [IB_USER_VERBS_CMD_DEALLOC_PD] = ib_uverbs_dealloc_pd,
+ [IB_USER_VERBS_CMD_REG_MR] = ib_uverbs_reg_mr,
+ [IB_USER_VERBS_CMD_DEREG_MR] = ib_uverbs_dereg_mr,
+ [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
+ [IB_USER_VERBS_CMD_CREATE_CQ] = ib_uverbs_create_cq,
+ [IB_USER_VERBS_CMD_POLL_CQ] = ib_uverbs_poll_cq,
+ [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ] = ib_uverbs_req_notify_cq,
+ [IB_USER_VERBS_CMD_DESTROY_CQ] = ib_uverbs_destroy_cq,
+ [IB_USER_VERBS_CMD_CREATE_QP] = ib_uverbs_create_qp,
+ [IB_USER_VERBS_CMD_MODIFY_QP] = ib_uverbs_modify_qp,
+ [IB_USER_VERBS_CMD_DESTROY_QP] = ib_uverbs_destroy_qp,
+ [IB_USER_VERBS_CMD_POST_SEND] = ib_uverbs_post_send,
+ [IB_USER_VERBS_CMD_POST_RECV] = ib_uverbs_post_recv,
+ [IB_USER_VERBS_CMD_POST_SRQ_RECV] = ib_uverbs_post_srq_recv,
+ [IB_USER_VERBS_CMD_CREATE_AH] = ib_uverbs_create_ah,
+ [IB_USER_VERBS_CMD_DESTROY_AH] = ib_uverbs_destroy_ah,
+ [IB_USER_VERBS_CMD_ATTACH_MCAST] = ib_uverbs_attach_mcast,
+ [IB_USER_VERBS_CMD_DETACH_MCAST] = ib_uverbs_detach_mcast,
+ [IB_USER_VERBS_CMD_CREATE_SRQ] = ib_uverbs_create_srq,
+ [IB_USER_VERBS_CMD_MODIFY_SRQ] = ib_uverbs_modify_srq,
+ [IB_USER_VERBS_CMD_DESTROY_SRQ] = ib_uverbs_destroy_srq,
};
static struct vfsmount *uverbs_event_mnt;
down(&ib_uverbs_idr_mutex);
- /* XXX Free AHs */
+ list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
+ struct ib_ah *ah = idr_find(&ib_uverbs_ah_idr, uobj->id);
+ idr_remove(&ib_uverbs_ah_idr, uobj->id);
+ ib_destroy_ah(ah);
+ list_del(&uobj->list);
+ kfree(uobj);
+ }
list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
struct ib_qp *qp = idr_find(&ib_uverbs_qp_idr, uobj->id);
spin_lock_irq(&file->lock);
- while (list_empty(&file->event_list) && file->fd >= 0) {
+ while (list_empty(&file->event_list)) {
spin_unlock_irq(&file->lock);
if (filp->f_flags & O_NONBLOCK)
return -EAGAIN;
if (wait_event_interruptible(file->poll_wait,
- !list_empty(&file->event_list) ||
- file->fd < 0))
+ !list_empty(&file->event_list)))
return -ERESTARTSYS;
spin_lock_irq(&file->lock);
}
- if (file->fd < 0) {
- spin_unlock_irq(&file->lock);
- return -ENODEV;
- }
-
event = list_entry(file->event_list.next, struct ib_uverbs_event, list);
if (file->is_async)
poll_wait(filp, &file->poll_wait, wait);
spin_lock_irq(&file->lock);
- if (file->fd < 0)
- pollflags = POLLERR;
- else if (!list_empty(&file->event_list))
+ if (!list_empty(&file->event_list))
pollflags = POLLIN | POLLRDNORM;
spin_unlock_irq(&file->lock);
return pollflags;
}
-static void ib_uverbs_event_release(struct ib_uverbs_event_file *file)
+void ib_uverbs_release_event_file(struct kref *ref)
{
- struct ib_uverbs_event *entry, *tmp;
+ struct ib_uverbs_event_file *file =
+ container_of(ref, struct ib_uverbs_event_file, ref);
- spin_lock_irq(&file->lock);
- if (file->fd != -1) {
- file->fd = -1;
- list_for_each_entry_safe(entry, tmp, &file->event_list, list)
- kfree(entry);
- }
- spin_unlock_irq(&file->lock);
+ kfree(file);
}
static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
{
struct ib_uverbs_event_file *file = filp->private_data;
+ struct ib_uverbs_event *entry, *tmp;
+
+ spin_lock_irq(&file->lock);
+ file->file = NULL;
+ list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
+ if (entry->counter)
+ list_del(&entry->obj_list);
+ kfree(entry);
+ }
+ spin_unlock_irq(&file->lock);
- ib_uverbs_event_release(file);
ib_uverbs_event_fasync(-1, filp, 0);
- kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
+
+ if (file->is_async) {
+ ib_unregister_event_handler(&file->uverbs_file->event_handler);
+ kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
+ }
+ kref_put(&file->ref, ib_uverbs_release_event_file);
return 0;
}
static struct file_operations uverbs_event_fops = {
- /*
- * No .owner field since we artificially create event files,
- * so there is no increment to the module reference count in
- * the open path. All event files come from a uverbs command
- * file, which already takes a module reference, so this is OK.
- */
+ .owner = THIS_MODULE,
.read = ib_uverbs_event_read,
.poll = ib_uverbs_event_poll,
.release = ib_uverbs_event_close,
void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
{
- struct ib_uverbs_file *file = cq_context;
- struct ib_ucq_object *uobj;
- struct ib_uverbs_event *entry;
- unsigned long flags;
+ struct ib_uverbs_event_file *file = cq_context;
+ struct ib_ucq_object *uobj;
+ struct ib_uverbs_event *entry;
+ unsigned long flags;
+
+ if (!file)
+ return;
+
+ spin_lock_irqsave(&file->lock, flags);
+ if (!file->file) {
+ spin_unlock_irqrestore(&file->lock, flags);
+ return;
+ }
entry = kmalloc(sizeof *entry, GFP_ATOMIC);
- if (!entry)
+ if (!entry) {
+ spin_unlock_irqrestore(&file->lock, flags);
return;
+ }
uobj = container_of(cq->uobject, struct ib_ucq_object, uobject);
entry->desc.comp.cq_handle = cq->uobject->user_handle;
entry->counter = &uobj->comp_events_reported;
- spin_lock_irqsave(&file->comp_file[0].lock, flags);
- list_add_tail(&entry->list, &file->comp_file[0].event_list);
+ list_add_tail(&entry->list, &file->event_list);
list_add_tail(&entry->obj_list, &uobj->comp_list);
- spin_unlock_irqrestore(&file->comp_file[0].lock, flags);
+ spin_unlock_irqrestore(&file->lock, flags);
- wake_up_interruptible(&file->comp_file[0].poll_wait);
- kill_fasync(&file->comp_file[0].async_queue, SIGIO, POLL_IN);
+ wake_up_interruptible(&file->poll_wait);
+ kill_fasync(&file->async_queue, SIGIO, POLL_IN);
}
static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
struct ib_uverbs_event *entry;
unsigned long flags;
+ spin_lock_irqsave(&file->async_file->lock, flags);
+ if (!file->async_file->file) {
+ spin_unlock_irqrestore(&file->async_file->lock, flags);
+ return;
+ }
+
entry = kmalloc(sizeof *entry, GFP_ATOMIC);
- if (!entry)
+ if (!entry) {
+ spin_unlock_irqrestore(&file->async_file->lock, flags);
return;
+ }
entry->desc.async.element = element;
entry->desc.async.event_type = event;
entry->counter = counter;
- spin_lock_irqsave(&file->async_file.lock, flags);
- list_add_tail(&entry->list, &file->async_file.event_list);
+ list_add_tail(&entry->list, &file->async_file->event_list);
if (obj_list)
list_add_tail(&entry->obj_list, obj_list);
- spin_unlock_irqrestore(&file->async_file.lock, flags);
+ spin_unlock_irqrestore(&file->async_file->lock, flags);
- wake_up_interruptible(&file->async_file.poll_wait);
- kill_fasync(&file->async_file.async_queue, SIGIO, POLL_IN);
+ wake_up_interruptible(&file->async_file->poll_wait);
+ kill_fasync(&file->async_file->async_queue, SIGIO, POLL_IN);
}
void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
{
+ struct ib_uverbs_event_file *ev_file = context_ptr;
struct ib_ucq_object *uobj;
uobj = container_of(event->element.cq->uobject,
struct ib_ucq_object, uobject);
- ib_uverbs_async_handler(context_ptr, uobj->uobject.user_handle,
+ ib_uverbs_async_handler(ev_file->uverbs_file, uobj->uobject.user_handle,
event->event, &uobj->async_list,
&uobj->async_events_reported);
&uobj->events_reported);
}
-static void ib_uverbs_event_handler(struct ib_event_handler *handler,
- struct ib_event *event)
+void ib_uverbs_event_handler(struct ib_event_handler *handler,
+ struct ib_event *event)
{
struct ib_uverbs_file *file =
container_of(handler, struct ib_uverbs_file, event_handler);
NULL, NULL);
}
-static int ib_uverbs_event_init(struct ib_uverbs_event_file *file,
- struct ib_uverbs_file *uverbs_file)
+struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
+ int is_async, int *fd)
{
+ struct ib_uverbs_event_file *ev_file;
struct file *filp;
+ int ret;
- spin_lock_init(&file->lock);
- INIT_LIST_HEAD(&file->event_list);
- init_waitqueue_head(&file->poll_wait);
- file->uverbs_file = uverbs_file;
- file->async_queue = NULL;
-
- file->fd = get_unused_fd();
- if (file->fd < 0)
- return file->fd;
+ ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
+ if (!ev_file)
+ return ERR_PTR(-ENOMEM);
+
+ kref_init(&ev_file->ref);
+ spin_lock_init(&ev_file->lock);
+ INIT_LIST_HEAD(&ev_file->event_list);
+ init_waitqueue_head(&ev_file->poll_wait);
+ ev_file->uverbs_file = uverbs_file;
+ ev_file->async_queue = NULL;
+ ev_file->is_async = is_async;
+
+ *fd = get_unused_fd();
+ if (*fd < 0) {
+ ret = *fd;
+ goto err;
+ }
filp = get_empty_filp();
if (!filp) {
- put_unused_fd(file->fd);
- return -ENFILE;
+ ret = -ENFILE;
+ goto err_fd;
}
- filp->f_op = &uverbs_event_fops;
+ ev_file->file = filp;
+
+ /*
+ * fops_get() can't fail here, because we're coming from a
+ * system call on a uverbs file, which will already have a
+ * module reference.
+ */
+ filp->f_op = fops_get(&uverbs_event_fops);
filp->f_vfsmnt = mntget(uverbs_event_mnt);
filp->f_dentry = dget(uverbs_event_mnt->mnt_root);
filp->f_mapping = filp->f_dentry->d_inode->i_mapping;
filp->f_flags = O_RDONLY;
filp->f_mode = FMODE_READ;
- filp->private_data = file;
+ filp->private_data = ev_file;
- fd_install(file->fd, filp);
+ return filp;
- return 0;
+err_fd:
+ put_unused_fd(*fd);
+
+err:
+ kfree(ev_file);
+ return ERR_PTR(ret);
+}
+
+/*
+ * Look up a completion event file by FD. If lookup is successful,
+ * takes a ref to the event file struct that it returns; if
+ * unsuccessful, returns NULL.
+ */
+struct ib_uverbs_event_file *ib_uverbs_lookup_comp_file(int fd)
+{
+ struct ib_uverbs_event_file *ev_file = NULL;
+ struct file *filp;
+
+ filp = fget(fd);
+ if (!filp)
+ return NULL;
+
+ if (filp->f_op != &uverbs_event_fops)
+ goto out;
+
+ ev_file = filp->private_data;
+ if (ev_file->is_async) {
+ ev_file = NULL;
+ goto out;
+ }
+
+ kref_get(&ev_file->ref);
+
+out:
+ fput(filp);
+ return ev_file;
}
static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
if (hdr.command < 0 ||
hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
- !uverbs_cmd_table[hdr.command])
+ !uverbs_cmd_table[hdr.command] ||
+ !(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
return -EINVAL;
- if (!file->ucontext &&
- hdr.command != IB_USER_VERBS_CMD_QUERY_PARAMS &&
+ if (!file->ucontext &&
hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
return -EINVAL;
struct ib_uverbs_device *dev =
container_of(inode->i_cdev, struct ib_uverbs_device, dev);
struct ib_uverbs_file *file;
- int i = 0;
- int ret;
if (!try_module_get(dev->ib_dev->owner))
return -ENODEV;
- file = kmalloc(sizeof *file +
- (dev->num_comp - 1) * sizeof (struct ib_uverbs_event_file),
- GFP_KERNEL);
+ file = kmalloc(sizeof *file, GFP_KERNEL);
if (!file) {
- ret = -ENOMEM;
- goto err;
+ module_put(dev->ib_dev->owner);
+ return -ENOMEM;
}
- file->device = dev;
+ file->device = dev;
+ file->ucontext = NULL;
kref_init(&file->ref);
init_MUTEX(&file->mutex);
- file->ucontext = NULL;
-
- kref_get(&file->ref);
- ret = ib_uverbs_event_init(&file->async_file, file);
- if (ret)
- goto err_kref;
-
- file->async_file.is_async = 1;
-
- for (i = 0; i < dev->num_comp; ++i) {
- kref_get(&file->ref);
- ret = ib_uverbs_event_init(&file->comp_file[i], file);
- if (ret)
- goto err_async;
- file->comp_file[i].is_async = 0;
- }
-
-
filp->private_data = file;
- INIT_IB_EVENT_HANDLER(&file->event_handler, dev->ib_dev,
- ib_uverbs_event_handler);
- if (ib_register_event_handler(&file->event_handler))
- goto err_async;
-
return 0;
-
-err_async:
- while (i--)
- ib_uverbs_event_release(&file->comp_file[i]);
-
- ib_uverbs_event_release(&file->async_file);
-
-err_kref:
- /*
- * One extra kref_put() because we took a reference before the
- * event file creation that failed and got us here.
- */
- kref_put(&file->ref, ib_uverbs_release_file);
- kref_put(&file->ref, ib_uverbs_release_file);
-
-err:
- module_put(dev->ib_dev->owner);
- return ret;
}
static int ib_uverbs_close(struct inode *inode, struct file *filp)
{
struct ib_uverbs_file *file = filp->private_data;
- int i;
- ib_unregister_event_handler(&file->event_handler);
- ib_uverbs_event_release(&file->async_file);
ib_dealloc_ucontext(file->ucontext);
- for (i = 0; i < file->device->num_comp; ++i)
- ib_uverbs_event_release(&file->comp_file[i]);
-
+ kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
kref_put(&file->ref, ib_uverbs_release_file);
return 0;
}
static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
+static ssize_t show_dev_abi_version(struct class_device *class_dev, char *buf)
+{
+ struct ib_uverbs_device *dev =
+ container_of(class_dev, struct ib_uverbs_device, class_dev);
+
+ return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
+}
+static CLASS_DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
+
static void ib_uverbs_release_class_dev(struct class_device *class_dev)
{
struct ib_uverbs_device *dev =
set_bit(uverbs_dev->devnum, dev_map);
spin_unlock(&map_lock);
- uverbs_dev->ib_dev = device;
- uverbs_dev->num_comp = 1;
+ uverbs_dev->ib_dev = device;
+ uverbs_dev->num_comp_vectors = 1;
if (device->mmap)
cdev_init(&uverbs_dev->dev, &uverbs_mmap_fops);
if (class_device_create_file(&uverbs_dev->class_dev, &class_device_attr_ibdev))
goto err_class;
+ if (class_device_create_file(&uverbs_dev->class_dev, &class_device_attr_abi_version))
+ goto err_class;
ib_set_client_data(device, &uverbs_client, uverbs_dev);
unregister_filesystem(&uverbs_event_fs);
class_unregister(&uverbs_class);
unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
+ idr_destroy(&ib_uverbs_pd_idr);
+ idr_destroy(&ib_uverbs_mr_idr);
+ idr_destroy(&ib_uverbs_mw_idr);
+ idr_destroy(&ib_uverbs_ah_idr);
+ idr_destroy(&ib_uverbs_cq_idr);
+ idr_destroy(&ib_uverbs_qp_idr);
+ idr_destroy(&ib_uverbs_srq_idr);
}
module_init(ib_uverbs_init);
MTHCA_EVENT_TYPE_PATH_MIG = 0x01,
MTHCA_EVENT_TYPE_COMM_EST = 0x02,
MTHCA_EVENT_TYPE_SQ_DRAINED = 0x03,
- MTHCA_EVENT_TYPE_SRQ_LAST_WQE = 0x13,
+ MTHCA_EVENT_TYPE_SRQ_QP_LAST_WQE = 0x13,
+ MTHCA_EVENT_TYPE_SRQ_LIMIT = 0x14,
MTHCA_EVENT_TYPE_CQ_ERROR = 0x04,
MTHCA_EVENT_TYPE_WQ_CATAS_ERROR = 0x05,
MTHCA_EVENT_TYPE_EEC_CATAS_ERROR = 0x06,
(1ULL << MTHCA_EVENT_TYPE_LOCAL_CATAS_ERROR) | \
(1ULL << MTHCA_EVENT_TYPE_PORT_CHANGE) | \
(1ULL << MTHCA_EVENT_TYPE_ECC_DETECT))
-#define MTHCA_SRQ_EVENT_MASK (1ULL << MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR) | \
- (1ULL << MTHCA_EVENT_TYPE_SRQ_LAST_WQE)
+#define MTHCA_SRQ_EVENT_MASK ((1ULL << MTHCA_EVENT_TYPE_SRQ_CATAS_ERROR) | \
+ (1ULL << MTHCA_EVENT_TYPE_SRQ_QP_LAST_WQE) | \
+ (1ULL << MTHCA_EVENT_TYPE_SRQ_LIMIT))
#define MTHCA_CMD_EVENT_MASK (1ULL << MTHCA_EVENT_TYPE_CMD)
#define MTHCA_EQ_DB_INC_CI (1 << 24)
struct {
__be32 qpn;
} __attribute__((packed)) qp;
+ struct {
+ __be32 srqn;
+ } __attribute__((packed)) srq;
struct {
__be32 cqn;
u32 reserved1;
IB_EVENT_SQ_DRAINED);
break;
+ case MTHCA_EVENT_TYPE_SRQ_QP_LAST_WQE:
+ mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
+ IB_EVENT_QP_LAST_WQE_REACHED);
+ break;
+
+ case MTHCA_EVENT_TYPE_SRQ_LIMIT:
+ mthca_srq_event(dev, be32_to_cpu(eqe->event.srq.srqn) & 0xffffff,
+ IB_EVENT_SRQ_LIMIT_REACHED);
+ break;
+
case MTHCA_EVENT_TYPE_WQ_CATAS_ERROR:
mthca_qp_event(dev, be32_to_cpu(eqe->event.qp.qpn) & 0xffffff,
IB_EVENT_QP_FATAL);
writel(dev->eq_table.clr_mask, dev->eq_table.clr_int);
ecr = readl(dev->eq_regs.tavor.ecr_base + 4);
- if (ecr) {
- writel(ecr, dev->eq_regs.tavor.ecr_base +
- MTHCA_ECR_CLR_BASE - MTHCA_ECR_BASE + 4);
+ if (!ecr)
+ return IRQ_NONE;
- for (i = 0; i < MTHCA_NUM_EQ; ++i)
- if (ecr & dev->eq_table.eq[i].eqn_mask &&
- mthca_eq_int(dev, &dev->eq_table.eq[i])) {
+ writel(ecr, dev->eq_regs.tavor.ecr_base +
+ MTHCA_ECR_CLR_BASE - MTHCA_ECR_BASE + 4);
+
+ for (i = 0; i < MTHCA_NUM_EQ; ++i)
+ if (ecr & dev->eq_table.eq[i].eqn_mask) {
+ if (mthca_eq_int(dev, &dev->eq_table.eq[i]))
tavor_set_eq_ci(dev, &dev->eq_table.eq[i],
dev->eq_table.eq[i].cons_index);
- tavor_eq_req_not(dev, dev->eq_table.eq[i].eqn);
- }
- }
+ tavor_eq_req_not(dev, dev->eq_table.eq[i].eqn);
+ }
- return IRQ_RETVAL(ecr);
+ return IRQ_HANDLED;
}
static irqreturn_t mthca_tavor_msi_x_interrupt(int irq, void *eq_ptr,