[openib-general] [RFC] [PATCH 2/7] ib_multicast 2.6.20: add ib_multicast module to track join requests from the same port
Sean Hefty
sean.hefty at intel.com
Tue Oct 10 15:58:31 PDT 2006
The IB SA tracks multicast join / leave requests on a per port basis.
In order to support multiple users of the same multicast group from
the same port, we need to perform local reference counting on each
of the nodes.
Add an ib_multicast module to perform reference counting of multicast
join / leave requests. Modify ib_ipoib to use the multicast module.
Signed-off-by: Sean Hefty <sean.hefty at intel.com>
---
diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile
index 163d991..76cc988 100644
--- a/drivers/infiniband/core/Makefile
+++ b/drivers/infiniband/core/Makefile
@@ -1,6 +1,7 @@
infiniband-$(CONFIG_INFINIBAND_ADDR_TRANS) := ib_addr.o rdma_cm.o
obj-$(CONFIG_INFINIBAND) += ib_core.o ib_mad.o ib_sa.o \
+ ib_multicast.o \
ib_cm.o iw_cm.o $(infiniband-y)
obj-$(CONFIG_INFINIBAND_USER_MAD) += ib_umad.o
obj-$(CONFIG_INFINIBAND_USER_ACCESS) += ib_uverbs.o ib_ucm.o
@@ -12,6 +13,8 @@ ib_mad-y := mad.o smi.o agent.o mad_rm
ib_sa-y := sa_query.o
+ib_multicast-y := multicast.o
+
ib_cm-y := cm.o
iw_cm-y := iwcm.o
diff --git a/drivers/infiniband/core/multicast.c b/drivers/infiniband/core/multicast.c
new file mode 100755
index 0000000..e7204b4
--- /dev/null
+++ b/drivers/infiniband/core/multicast.c
@@ -0,0 +1,795 @@
+/*
+ * Copyright (c) 2006 Intel Corporation. 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/completion.h>
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/bitops.h>
+#include <linux/random.h>
+
+#include <rdma/ib_multicast.h>
+#include <rdma/ib_cache.h>
+
+MODULE_AUTHOR("Sean Hefty");
+MODULE_DESCRIPTION("InfiniBand multicast membership handling");
+MODULE_LICENSE("Dual BSD/GPL");
+
+static int retry_timer = 5000; /* 5 sec */
+module_param(retry_timer, int, 0444);
+MODULE_PARM_DESC(retry_timer, "Time in ms between retried requests.");
+
+static int retries = 3;
+module_param(retries, int, 0444);
+MODULE_PARM_DESC(retries, "Number of times to retry a request.");
+
+static void mcast_add_one(struct ib_device *device);
+static void mcast_remove_one(struct ib_device *device);
+
+static struct ib_client mcast_client = {
+ .name = "ib_multicast",
+ .add = mcast_add_one,
+ .remove = mcast_remove_one
+};
+
+static struct ib_sa_client sa_client;
+static struct ib_event_handler event_handler;
+static struct workqueue_struct *mcast_wq;
+static union ib_gid mgid0;
+
+struct mcast_device;
+
+struct mcast_port {
+ struct mcast_device *dev;
+ spinlock_t lock;
+ struct rb_root table;
+ atomic_t refcount;
+ struct completion comp;
+ u8 port_num;
+};
+
+struct mcast_device {
+ struct ib_device *device;
+ int start_port;
+ int end_port;
+ struct mcast_port port[0];
+};
+
+enum mcast_state {
+ MCAST_IDLE,
+ MCAST_JOINING,
+ MCAST_MEMBER,
+ MCAST_BUSY,
+ MCAST_ERROR
+};
+
+struct mcast_member;
+
+struct mcast_group {
+ struct ib_sa_mcmember_rec rec;
+ struct rb_node node;
+ struct mcast_port *port;
+ spinlock_t lock;
+ struct work_struct work;
+ struct list_head pending_list;
+ struct list_head active_list;
+ struct mcast_member *last_join;
+ int members[3];
+ atomic_t refcount;
+ enum mcast_state state;
+ struct ib_sa_query *query;
+ int query_id;
+};
+
+struct mcast_member {
+ struct ib_multicast multicast;
+ struct mcast_group *group;
+ struct list_head list;
+ enum mcast_state state;
+ atomic_t refcount;
+ struct completion comp;
+};
+
+static void join_handler(int status, struct ib_sa_mcmember_rec *rec,
+ void *context);
+static void leave_handler(int status, struct ib_sa_mcmember_rec *rec,
+ void *context);
+
+static struct mcast_group *mcast_find(struct mcast_port *port,
+ union ib_gid *mgid)
+{
+ struct rb_node *node = port->table.rb_node;
+ struct mcast_group *group;
+ int ret;
+
+ while (node) {
+ group = rb_entry(node, struct mcast_group, node);
+ ret = memcmp(mgid->raw, group->rec.mgid.raw, sizeof *mgid);
+ if (!ret)
+ return group;
+
+ if (ret < 0)
+ node = node->rb_left;
+ else
+ node = node->rb_right;
+ }
+ return NULL;
+}
+
+static struct mcast_group *mcast_insert(struct mcast_port *port,
+ struct mcast_group *group,
+ int allow_duplicates)
+{
+ struct rb_node **link = &port->table.rb_node;
+ struct rb_node *parent = NULL;
+ struct mcast_group *cur_group;
+ int ret;
+
+ while (*link) {
+ parent = *link;
+ cur_group = rb_entry(parent, struct mcast_group, node);
+
+ ret = memcmp(group->rec.mgid.raw, cur_group->rec.mgid.raw,
+ sizeof group->rec.mgid);
+ if (ret < 0)
+ link = &(*link)->rb_left;
+ else if (ret > 0)
+ link = &(*link)->rb_right;
+ else if (allow_duplicates)
+ link = &(*link)->rb_left;
+ else
+ return cur_group;
+ }
+ rb_link_node(&group->node, parent, link);
+ rb_insert_color(&group->node, &port->table);
+ return NULL;
+}
+
+static void deref_port(struct mcast_port *port)
+{
+ if (atomic_dec_and_test(&port->refcount))
+ complete(&port->comp);
+}
+
+static void release_group(struct mcast_group *group)
+{
+ struct mcast_port *port = group->port;
+ unsigned long flags;
+
+ spin_lock_irqsave(&port->lock, flags);
+ if (atomic_dec_and_test(&group->refcount)) {
+ rb_erase(&group->node, &port->table);
+ spin_unlock_irqrestore(&port->lock, flags);
+ kfree(group);
+ deref_port(port);
+ } else
+ spin_unlock_irqrestore(&port->lock, flags);
+}
+
+static void deref_member(struct mcast_member *member)
+{
+ if (atomic_dec_and_test(&member->refcount))
+ complete(&member->comp);
+}
+
+static void queue_join(struct mcast_member *member)
+{
+ struct mcast_group *group = member->group;
+ unsigned long flags;
+
+ spin_lock_irqsave(&group->lock, flags);
+ list_add(&member->list, &group->pending_list);
+ if (group->state == MCAST_IDLE) {
+ group->state = MCAST_BUSY;
+ atomic_inc(&group->refcount);
+ queue_work(mcast_wq, &group->work);
+ }
+ spin_unlock_irqrestore(&group->lock, flags);
+}
+
+/*
+ * A multicast group has three types of members: full member, non member, and
+ * send only member. We need to keep track of the number of members of each
+ * type based on their join state. Adjust the number of members the belong to
+ * the specified join states.
+ */
+static void adjust_membership(struct mcast_group *group, u8 join_state, int inc)
+{
+ int i;
+
+ for (i = 0; i < 3; i++, join_state >>= 1)
+ if (join_state & 0x1)
+ group->members[i] += inc;
+}
+
+/*
+ * If a multicast group has zero members left for a particular join state, but
+ * the group is still a member with the SA, we need to leave that join state.
+ * Determine which join states we still belong to, but that do not have any
+ * active members.
+ */
+static u8 get_leave_state(struct mcast_group *group)
+{
+ u8 leave_state = 0;
+ int i;
+
+ for (i = 0; i < 3; i++)
+ if (!group->members[i])
+ leave_state |= (0x1 << i);
+
+ return leave_state & group->rec.join_state;
+}
+
+static int cmp_rec(struct ib_sa_mcmember_rec *src,
+ struct ib_sa_mcmember_rec *dst, ib_sa_comp_mask comp_mask)
+{
+ /* MGID must already match */
+
+ if (comp_mask & IB_SA_MCMEMBER_REC_PORT_GID &&
+ memcmp(&src->port_gid, &dst->port_gid, sizeof src->port_gid))
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_QKEY && src->qkey != dst->qkey)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_MLID && src->mlid != dst->mlid)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_MTU_SELECTOR &&
+ src->mtu_selector != dst->mtu_selector)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_MTU && src->mtu != dst->mtu)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_TRAFFIC_CLASS &&
+ src->traffic_class != dst->traffic_class)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_PKEY && src->pkey != dst->pkey)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_RATE_SELECTOR &&
+ src->rate_selector != dst->rate_selector)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_RATE && src->rate != dst->rate)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME_SELECTOR &&
+ src->packet_life_time_selector != dst->packet_life_time_selector)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME &&
+ src->packet_life_time != dst->packet_life_time)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_SL && src->sl != dst->sl)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_FLOW_LABEL &&
+ src->flow_label != dst->flow_label)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_HOP_LIMIT &&
+ src->hop_limit != dst->hop_limit)
+ return -EINVAL;
+ if (comp_mask & IB_SA_MCMEMBER_REC_SCOPE && src->scope != dst->scope)
+ return -EINVAL;
+
+ /* join_state checked separately, proxy_join ignored */
+
+ return 0;
+}
+
+static int send_join(struct mcast_group *group, struct mcast_member *member)
+{
+ struct mcast_port *port = group->port;
+ int ret;
+
+ group->last_join = member;
+ ret = ib_sa_mcmember_rec_set(&sa_client, port->dev->device,
+ port->port_num, &member->multicast.rec,
+ member->multicast.comp_mask,
+ retry_timer, retries, GFP_KERNEL,
+ join_handler, group, &group->query);
+ if (ret >= 0) {
+ group->query_id = ret;
+ ret = 0;
+ }
+ return ret;
+}
+
+static int send_leave(struct mcast_group *group, u8 leave_state)
+{
+ struct mcast_port *port = group->port;
+ struct ib_sa_mcmember_rec rec;
+ int ret;
+
+ rec = group->rec;
+ rec.join_state = leave_state;
+
+ ret = ib_sa_mcmember_rec_delete(&sa_client, port->dev->device,
+ port->port_num, &rec,
+ IB_SA_MCMEMBER_REC_MGID |
+ IB_SA_MCMEMBER_REC_PORT_GID |
+ IB_SA_MCMEMBER_REC_JOIN_STATE,
+ retry_timer, retries, GFP_KERNEL,
+ leave_handler, group, &group->query);
+ if (ret >= 0) {
+ group->query_id = ret;
+ ret = 0;
+ }
+ return ret;
+}
+
+static void join_group(struct mcast_group *group, struct mcast_member *member,
+ u8 join_state)
+{
+ member->state = MCAST_MEMBER;
+ adjust_membership(group, join_state, 1);
+ group->rec.join_state |= join_state;
+ member->multicast.rec = group->rec;
+ member->multicast.rec.join_state = join_state;
+ list_del(&member->list);
+ list_add(&member->list, &group->active_list);
+}
+
+static int fail_join(struct mcast_group *group, struct mcast_member *member,
+ int status)
+{
+ spin_lock_irq(&group->lock);
+ list_del_init(&member->list);
+ spin_unlock_irq(&group->lock);
+ return member->multicast.callback(status, &member->multicast);
+}
+
+static void process_group_error(struct mcast_group *group)
+{
+ struct mcast_member *member;
+ int ret;
+
+ spin_lock_irq(&group->lock);
+ while (!list_empty(&group->active_list)) {
+ member = list_entry(group->active_list.next,
+ struct mcast_member, list);
+ atomic_inc(&member->refcount);
+ list_del_init(&member->list);
+ adjust_membership(group, member->multicast.rec.join_state, -1);
+ member->state = MCAST_ERROR;
+ spin_unlock_irq(&group->lock);
+
+ ret = member->multicast.callback(-ENETRESET,
+ &member->multicast);
+ deref_member(member);
+ if (ret)
+ ib_free_multicast(&member->multicast);
+ spin_lock_irq(&group->lock);
+ }
+
+ group->rec.join_state = 0;
+ group->state = MCAST_BUSY;
+ spin_unlock_irq(&group->lock);
+}
+
+static void mcast_work_handler(void *data)
+{
+ struct mcast_group *group = data;
+ struct mcast_member *member;
+ struct ib_multicast *multicast;
+ int status, ret;
+ u8 join_state;
+
+retest:
+ spin_lock_irq(&group->lock);
+ if (group->state == MCAST_ERROR) {
+ spin_unlock_irq(&group->lock);
+ process_group_error(group);
+ goto retest;
+ }
+
+ while (!list_empty(&group->pending_list)) {
+ member = list_entry(group->pending_list.next,
+ struct mcast_member, list);
+ multicast = &member->multicast;
+ join_state = multicast->rec.join_state;
+ atomic_inc(&member->refcount);
+
+ if (join_state == (group->rec.join_state & join_state)) {
+ status = cmp_rec(&group->rec, &multicast->rec,
+ multicast->comp_mask);
+ if (!status)
+ join_group(group, member, join_state);
+ else
+ list_del_init(&member->list);
+ spin_unlock_irq(&group->lock);
+ ret = multicast->callback(status, multicast);
+ } else {
+ spin_unlock_irq(&group->lock);
+ status = send_join(group, member);
+ if (!status) {
+ deref_member(member);
+ return;
+ }
+ ret = fail_join(group, member, status);
+ }
+
+ deref_member(member);
+ if (ret)
+ ib_free_multicast(&member->multicast);
+ spin_lock_irq(&group->lock);
+ }
+
+ join_state = get_leave_state(group);
+ if (join_state) {
+ group->rec.join_state &= ~join_state;
+ spin_unlock_irq(&group->lock);
+ if (send_leave(group, join_state))
+ goto retest;
+ } else {
+ group->state = MCAST_IDLE;
+ spin_unlock_irq(&group->lock);
+ release_group(group);
+ }
+}
+
+/*
+ * Fail a join request if it is still active - at the head of the pending queue.
+ */
+static void process_join_error(struct mcast_group *group, int status)
+{
+ struct mcast_member *member;
+ int ret;
+
+ spin_lock_irq(&group->lock);
+ member = list_entry(group->pending_list.next,
+ struct mcast_member, list);
+ if (group->last_join == member) {
+ atomic_inc(&member->refcount);
+ list_del_init(&member->list);
+ spin_unlock_irq(&group->lock);
+ ret = member->multicast.callback(status, &member->multicast);
+ deref_member(member);
+ if (ret)
+ ib_free_multicast(&member->multicast);
+ } else
+ spin_unlock_irq(&group->lock);
+}
+
+static void join_handler(int status, struct ib_sa_mcmember_rec *rec,
+ void *context)
+{
+ struct mcast_group *group = context;
+
+ if (status)
+ process_join_error(group, status);
+ else {
+ spin_lock_irq(&group->port->lock);
+ group->rec = *rec;
+ if (!memcmp(&mgid0, &group->rec.mgid, sizeof mgid0)) {
+ rb_erase(&group->node, &group->port->table);
+ mcast_insert(group->port, group, 1);
+ }
+ spin_unlock_irq(&group->port->lock);
+ }
+ mcast_work_handler(group);
+}
+
+static void leave_handler(int status, struct ib_sa_mcmember_rec *rec,
+ void *context)
+{
+ mcast_work_handler(context);
+}
+
+static struct mcast_group *acquire_group(struct mcast_port *port,
+ union ib_gid *mgid, gfp_t gfp_mask)
+{
+ struct mcast_group *group, *cur_group;
+ unsigned long flags;
+ int is_mgid0;
+
+ is_mgid0 = !memcmp(&mgid0, mgid, sizeof mgid0);
+ if (!is_mgid0) {
+ spin_lock_irqsave(&port->lock, flags);
+ group = mcast_find(port, mgid);
+ if (group)
+ goto found;
+ spin_unlock_irqrestore(&port->lock, flags);
+ }
+
+ group = kzalloc(sizeof *group, gfp_mask);
+ if (!group)
+ return NULL;
+
+ group->port = port;
+ group->rec.mgid = *mgid;
+ INIT_LIST_HEAD(&group->pending_list);
+ INIT_LIST_HEAD(&group->active_list);
+ INIT_WORK(&group->work, mcast_work_handler, group);
+ spin_lock_init(&group->lock);
+
+ spin_lock_irqsave(&port->lock, flags);
+ cur_group = mcast_insert(port, group, is_mgid0);
+ if (cur_group) {
+ kfree(group);
+ group = cur_group;
+ } else
+ atomic_inc(&port->refcount);
+found:
+ atomic_inc(&group->refcount);
+ spin_unlock_irqrestore(&port->lock, flags);
+ return group;
+}
+
+/*
+ * We serialize all join requests to a single group to make our lives much
+ * easier. Otherwise, two users could try to join the same group
+ * simultaneously, with different configurations, one could leave while the
+ * join is in progress, etc., which makes locking around error recovery
+ * difficult.
+ */
+struct ib_multicast *ib_join_multicast(struct ib_device *device, u8 port_num,
+ struct ib_sa_mcmember_rec *rec,
+ ib_sa_comp_mask comp_mask, gfp_t gfp_mask,
+ int (*callback)(int status,
+ struct ib_multicast
+ *multicast),
+ void *context)
+{
+ struct mcast_device *dev;
+ struct mcast_member *member;
+ struct ib_multicast *multicast;
+ int ret;
+
+ dev = ib_get_client_data(device, &mcast_client);
+ if (!dev)
+ return ERR_PTR(-ENODEV);
+
+ member = kzalloc(sizeof *member, gfp_mask);
+ if (!member)
+ return ERR_PTR(-ENOMEM);
+
+ member->multicast.rec = *rec;
+ member->multicast.comp_mask = comp_mask;
+ member->multicast.callback = callback;
+ member->multicast.context = context;
+ init_completion(&member->comp);
+ atomic_set(&member->refcount, 1);
+ member->state = MCAST_JOINING;
+
+ member->group = acquire_group(&dev->port[port_num - dev->start_port],
+ &rec->mgid, gfp_mask);
+ if (!member->group) {
+ ret = -ENOMEM;
+ goto err;
+ }
+
+ /*
+ * The user will get the multicast structure in their callback. They
+ * could then free the multicast structure before we can return from
+ * this routine. So we save the pointer to return before queuing
+ * any callback.
+ */
+ multicast = &member->multicast;
+ queue_join(member);
+ return multicast;
+
+err:
+ kfree(member);
+ return ERR_PTR(ret);
+}
+EXPORT_SYMBOL(ib_join_multicast);
+
+void ib_free_multicast(struct ib_multicast *multicast)
+{
+ struct mcast_member *member;
+ struct mcast_group *group;
+
+ member = container_of(multicast, struct mcast_member, multicast);
+ group = member->group;
+
+ spin_lock_irq(&group->lock);
+ if (member->state == MCAST_MEMBER)
+ adjust_membership(group, multicast->rec.join_state, -1);
+
+ list_del_init(&member->list);
+
+ if (group->state == MCAST_IDLE) {
+ group->state = MCAST_BUSY;
+ spin_unlock_irq(&group->lock);
+ /* Continue to hold reference on group until callback */
+ queue_work(mcast_wq, &group->work);
+ } else {
+ spin_unlock_irq(&group->lock);
+ release_group(group);
+ }
+
+ deref_member(member);
+ wait_for_completion(&member->comp);
+ kfree(member);
+}
+EXPORT_SYMBOL(ib_free_multicast);
+
+int ib_get_mcmember_rec(struct ib_device *device, u8 port_num,
+ union ib_gid *mgid, struct ib_sa_mcmember_rec *rec)
+{
+ struct mcast_device *dev;
+ struct mcast_port *port;
+ struct mcast_group *group;
+ unsigned long flags;
+ int ret = 0;
+
+ dev = ib_get_client_data(device, &mcast_client);
+ if (!dev)
+ return -ENODEV;
+
+ port = &dev->port[port_num - dev->start_port];
+ if (mgid && memcmp(mgid, &mgid0, sizeof mgid0)) {
+ spin_lock_irqsave(&port->lock, flags);
+ group = mcast_find(port, mgid);
+ if (group)
+ *rec = group->rec;
+ else
+ ret = -EADDRNOTAVAIL;
+ spin_unlock_irqrestore(&port->lock, flags);
+ } else {
+ memset(rec, 0, sizeof *rec);
+ ib_get_cached_gid(device, port_num, 0, &rec->port_gid);
+ rec->pkey = 0xFFFF;
+ get_random_bytes(&rec->qkey, sizeof rec->qkey);
+ rec->join_state = 1;
+ }
+
+ return ret;
+}
+EXPORT_SYMBOL(ib_get_mcmember_rec);
+
+static void mcast_groups_lost(struct mcast_port *port)
+{
+ struct mcast_group *group;
+ struct rb_node *node;
+ unsigned long flags;
+
+ spin_lock_irqsave(&port->lock, flags);
+ for (node = rb_first(&port->table); node; node = rb_next(node)) {
+ group = rb_entry(node, struct mcast_group, node);
+ spin_lock(&group->lock);
+ if (group->state == MCAST_IDLE) {
+ atomic_inc(&group->refcount);
+ queue_work(mcast_wq, &group->work);
+ }
+ group->state = MCAST_ERROR;
+ spin_unlock(&group->lock);
+ }
+ spin_unlock_irqrestore(&port->lock, flags);
+}
+
+static void mcast_event_handler(struct ib_event_handler *handler,
+ struct ib_event *event)
+{
+ struct mcast_device *dev;
+
+ dev = ib_get_client_data(event->device, &mcast_client);
+ if (!dev)
+ return;
+
+ switch (event->event) {
+ case IB_EVENT_PORT_ERR:
+ case IB_EVENT_LID_CHANGE:
+ case IB_EVENT_SM_CHANGE:
+ case IB_EVENT_CLIENT_REREGISTER:
+ mcast_groups_lost(&dev->port[event->element.port_num -
+ dev->start_port]);
+ break;
+ default:
+ break;
+ }
+}
+
+static void mcast_add_one(struct ib_device *device)
+{
+ struct mcast_device *dev;
+ struct mcast_port *port;
+ int i;
+
+ if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
+ return;
+
+ dev = kmalloc(sizeof *dev + device->phys_port_cnt * sizeof *port,
+ GFP_KERNEL);
+ if (!dev)
+ return;
+
+ if (device->node_type == RDMA_NODE_IB_SWITCH)
+ dev->start_port = dev->end_port = 0;
+ else {
+ dev->start_port = 1;
+ dev->end_port = device->phys_port_cnt;
+ }
+
+ for (i = 0; i <= dev->end_port - dev->start_port; i++) {
+ port = &dev->port[i];
+ port->dev = dev;
+ port->port_num = dev->start_port + i;
+ spin_lock_init(&port->lock);
+ port->table = RB_ROOT;
+ init_completion(&port->comp);
+ atomic_set(&port->refcount, 1);
+ }
+
+ dev->device = device;
+ ib_set_client_data(device, &mcast_client, dev);
+
+ INIT_IB_EVENT_HANDLER(&event_handler, device, mcast_event_handler);
+ ib_register_event_handler(&event_handler);
+}
+
+static void mcast_remove_one(struct ib_device *device)
+{
+ struct mcast_device *dev;
+ struct mcast_port *port;
+ int i;
+
+ dev = ib_get_client_data(device, &mcast_client);
+ if (!dev)
+ return;
+
+ ib_unregister_event_handler(&event_handler);
+ flush_workqueue(mcast_wq);
+
+ for (i = 0; i < dev->end_port - dev->start_port; i++) {
+ port = &dev->port[i];
+ deref_port(port);
+ wait_for_completion(&port->comp);
+ }
+
+ kfree(dev);
+}
+
+static int __init mcast_init(void)
+{
+ int ret;
+
+ mcast_wq = create_singlethread_workqueue("ib_mcast_wq");
+ if (!mcast_wq)
+ return -ENOMEM;
+
+ ib_sa_register_client(&sa_client);
+
+ ret = ib_register_client(&mcast_client);
+ if (ret)
+ goto err;
+ return 0;
+
+err:
+ ib_sa_unregister_client(&sa_client);
+ destroy_workqueue(mcast_wq);
+ return ret;
+}
+
+static void __exit mcast_cleanup(void)
+{
+ ib_unregister_client(&mcast_client);
+ ib_sa_unregister_client(&sa_client);
+ destroy_workqueue(mcast_wq);
+}
+
+module_init(mcast_init);
+module_exit(mcast_cleanup);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
index 3faa182..b993cb1 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
@@ -45,6 +45,8 @@ #include <linux/completion.h>
#include <net/dst.h>
+#include <rdma/ib_multicast.h>
+
#include "ipoib.h"
#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
@@ -60,14 +62,11 @@ static DEFINE_MUTEX(mcast_mutex);
/* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
struct ipoib_mcast {
struct ib_sa_mcmember_rec mcmember;
+ struct ib_multicast *mc;
struct ipoib_ah *ah;
struct rb_node rb_node;
struct list_head list;
- struct completion done;
-
- int query_id;
- struct ib_sa_query *query;
unsigned long created;
unsigned long backoff;
@@ -299,18 +298,22 @@ static int ipoib_mcast_join_finish(struc
return 0;
}
-static void
+static int
ipoib_mcast_sendonly_join_complete(int status,
- struct ib_sa_mcmember_rec *mcmember,
- void *mcast_ptr)
+ struct ib_multicast *multicast)
{
- struct ipoib_mcast *mcast = mcast_ptr;
+ struct ipoib_mcast *mcast = multicast->context;
struct net_device *dev = mcast->dev;
struct ipoib_dev_priv *priv = netdev_priv(dev);
+ /* We trap for port events ourselves. */
+ if (status == -ENETRESET)
+ return 0;
+
if (!status)
- ipoib_mcast_join_finish(mcast, mcmember);
- else {
+ status = ipoib_mcast_join_finish(mcast, &multicast->rec);
+
+ if (status) {
if (mcast->logcount++ < 20)
ipoib_dbg_mcast(netdev_priv(dev), "multicast join failed for "
IPOIB_GID_FMT ", status %d\n",
@@ -325,11 +328,10 @@ ipoib_mcast_sendonly_join_complete(int s
spin_unlock_irq(&priv->tx_lock);
/* Clear the busy flag so we try again */
- clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
- mcast->query = NULL;
+ status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
+ &mcast->flags);
}
-
- complete(&mcast->done);
+ return status;
}
static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
@@ -359,35 +361,32 @@ #endif
rec.port_gid = priv->local_gid;
rec.pkey = cpu_to_be16(priv->pkey);
- init_completion(&mcast->done);
-
- ret = ib_sa_mcmember_rec_set(&ipoib_sa_client, priv->ca, priv->port, &rec,
- IB_SA_MCMEMBER_REC_MGID |
- IB_SA_MCMEMBER_REC_PORT_GID |
- IB_SA_MCMEMBER_REC_PKEY |
- IB_SA_MCMEMBER_REC_JOIN_STATE,
- 1000, GFP_ATOMIC,
- ipoib_mcast_sendonly_join_complete,
- mcast, &mcast->query);
- if (ret < 0) {
- ipoib_warn(priv, "ib_sa_mcmember_rec_set failed (ret = %d)\n",
+ mcast->mc = ib_join_multicast(priv->ca, priv->port, &rec,
+ IB_SA_MCMEMBER_REC_MGID |
+ IB_SA_MCMEMBER_REC_PORT_GID |
+ IB_SA_MCMEMBER_REC_PKEY |
+ IB_SA_MCMEMBER_REC_JOIN_STATE,
+ GFP_ATOMIC,
+ ipoib_mcast_sendonly_join_complete,
+ mcast);
+ if (IS_ERR(mcast->mc)) {
+ ret = PTR_ERR(mcast->mc);
+ clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
+ ipoib_warn(priv, "ib_join_multicast failed (ret = %d)\n",
ret);
} else {
ipoib_dbg_mcast(priv, "no multicast record for " IPOIB_GID_FMT
", starting join\n",
IPOIB_GID_ARG(mcast->mcmember.mgid));
-
- mcast->query_id = ret;
}
return ret;
}
-static void ipoib_mcast_join_complete(int status,
- struct ib_sa_mcmember_rec *mcmember,
- void *mcast_ptr)
+static int ipoib_mcast_join_complete(int status,
+ struct ib_multicast *multicast)
{
- struct ipoib_mcast *mcast = mcast_ptr;
+ struct ipoib_mcast *mcast = multicast->context;
struct net_device *dev = mcast->dev;
struct ipoib_dev_priv *priv = netdev_priv(dev);
@@ -395,23 +394,24 @@ static void ipoib_mcast_join_complete(in
" (status %d)\n",
IPOIB_GID_ARG(mcast->mcmember.mgid), status);
- if (!status && !ipoib_mcast_join_finish(mcast, mcmember)) {
+ /* We trap for port events ourselves. */
+ if (status == -ENETRESET)
+ return 0;
+
+ if (!status)
+ status = ipoib_mcast_join_finish(mcast, &multicast->rec);
+
+ if (!status) {
mcast->backoff = 1;
mutex_lock(&mcast_mutex);
if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
queue_work(ipoib_workqueue, &priv->mcast_task);
mutex_unlock(&mcast_mutex);
- complete(&mcast->done);
- return;
- }
-
- if (status == -EINTR) {
- complete(&mcast->done);
- return;
+ return 0;
}
- if (status && mcast->logcount++ < 20) {
- if (status == -ETIMEDOUT || status == -EINTR) {
+ if (mcast->logcount++ < 20) {
+ if (status == -ETIMEDOUT) {
ipoib_dbg_mcast(priv, "multicast join failed for " IPOIB_GID_FMT
", status %d\n",
IPOIB_GID_ARG(mcast->mcmember.mgid),
@@ -428,23 +428,18 @@ static void ipoib_mcast_join_complete(in
if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
- mutex_lock(&mcast_mutex);
+ /* Clear the busy flag so we try again */
+ status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
+ mutex_lock(&mcast_mutex);
spin_lock_irq(&priv->lock);
- mcast->query = NULL;
-
- if (test_bit(IPOIB_MCAST_RUN, &priv->flags)) {
- if (status == -ETIMEDOUT)
- queue_work(ipoib_workqueue, &priv->mcast_task);
- else
- queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
- mcast->backoff * HZ);
- } else
- complete(&mcast->done);
+ if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
+ queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
+ mcast->backoff * HZ);
spin_unlock_irq(&priv->lock);
mutex_unlock(&mcast_mutex);
- return;
+ return status;
}
static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
@@ -493,15 +488,14 @@ static void ipoib_mcast_join(struct net_
rec.hop_limit = priv->broadcast->mcmember.hop_limit;
}
- init_completion(&mcast->done);
-
- ret = ib_sa_mcmember_rec_set(&ipoib_sa_client, priv->ca, priv->port,
- &rec, comp_mask, mcast->backoff * 1000,
- GFP_ATOMIC, ipoib_mcast_join_complete,
- mcast, &mcast->query);
-
- if (ret < 0) {
- ipoib_warn(priv, "ib_sa_mcmember_rec_set failed, status %d\n", ret);
+ set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
+ mcast->mc = ib_join_multicast(priv->ca, priv->port, &rec, comp_mask,
+ GFP_KERNEL, ipoib_mcast_join_complete,
+ mcast);
+ if (IS_ERR(mcast->mc)) {
+ clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
+ ret = PTR_ERR(mcast->mc);
+ ipoib_warn(priv, "ib_join_multicast failed, status %d\n", ret);
mcast->backoff *= 2;
if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
@@ -513,8 +507,7 @@ static void ipoib_mcast_join(struct net_
&priv->mcast_task,
mcast->backoff * HZ);
mutex_unlock(&mcast_mutex);
- } else
- mcast->query_id = ret;
+ }
}
void ipoib_mcast_join_task(void *dev_ptr)
@@ -538,7 +531,7 @@ void ipoib_mcast_join_task(void *dev_ptr
priv->local_rate = attr.active_speed *
ib_width_enum_to_int(attr.active_width);
} else
- ipoib_warn(priv, "ib_query_port failed\n");
+ ipoib_warn(priv, "ib_query_port failed\n");
}
if (!priv->broadcast) {
@@ -565,7 +558,8 @@ void ipoib_mcast_join_task(void *dev_ptr
}
if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
- ipoib_mcast_join(dev, priv->broadcast, 0);
+ if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
+ ipoib_mcast_join(dev, priv->broadcast, 0);
return;
}
@@ -620,26 +614,9 @@ int ipoib_mcast_start_thread(struct net_
return 0;
}
-static void wait_for_mcast_join(struct ipoib_dev_priv *priv,
- struct ipoib_mcast *mcast)
-{
- spin_lock_irq(&priv->lock);
- if (mcast && mcast->query) {
- ib_sa_cancel_query(mcast->query_id, mcast->query);
- mcast->query = NULL;
- spin_unlock_irq(&priv->lock);
- ipoib_dbg_mcast(priv, "waiting for MGID " IPOIB_GID_FMT "\n",
- IPOIB_GID_ARG(mcast->mcmember.mgid));
- wait_for_completion(&mcast->done);
- }
- else
- spin_unlock_irq(&priv->lock);
-}
-
int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
- struct ipoib_mcast *mcast;
ipoib_dbg_mcast(priv, "stopping multicast thread\n");
@@ -655,52 +632,27 @@ int ipoib_mcast_stop_thread(struct net_d
if (flush)
flush_workqueue(ipoib_workqueue);
- wait_for_mcast_join(priv, priv->broadcast);
-
- list_for_each_entry(mcast, &priv->multicast_list, list)
- wait_for_mcast_join(priv, mcast);
-
return 0;
}
static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
{
struct ipoib_dev_priv *priv = netdev_priv(dev);
- struct ib_sa_mcmember_rec rec = {
- .join_state = 1
- };
int ret = 0;
- if (!test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags))
- return 0;
-
- ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
- IPOIB_GID_ARG(mcast->mcmember.mgid));
-
- rec.mgid = mcast->mcmember.mgid;
- rec.port_gid = priv->local_gid;
- rec.pkey = cpu_to_be16(priv->pkey);
+ if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
+ ipoib_dbg_mcast(priv, "leaving MGID " IPOIB_GID_FMT "\n",
+ IPOIB_GID_ARG(mcast->mcmember.mgid));
- /* Remove ourselves from the multicast group */
- ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
- &mcast->mcmember.mgid);
- if (ret)
- ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
+ /* Remove ourselves from the multicast group */
+ ret = ipoib_mcast_detach(dev, be16_to_cpu(mcast->mcmember.mlid),
+ &mcast->mcmember.mgid);
+ if (ret)
+ ipoib_warn(priv, "ipoib_mcast_detach failed (result = %d)\n", ret);
+ }
- /*
- * Just make one shot at leaving and don't wait for a reply;
- * if we fail, too bad.
- */
- ret = ib_sa_mcmember_rec_delete(&ipoib_sa_client, priv->ca, priv->port, &rec,
- IB_SA_MCMEMBER_REC_MGID |
- IB_SA_MCMEMBER_REC_PORT_GID |
- IB_SA_MCMEMBER_REC_PKEY |
- IB_SA_MCMEMBER_REC_JOIN_STATE,
- 0, GFP_ATOMIC, NULL,
- mcast, &mcast->query);
- if (ret < 0)
- ipoib_warn(priv, "ib_sa_mcmember_rec_delete failed "
- "for leave (result = %d)\n", ret);
+ if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
+ ib_free_multicast(mcast->mc);
return 0;
}
@@ -753,7 +705,7 @@ void ipoib_mcast_send(struct net_device
dev_kfree_skb_any(skb);
}
- if (mcast->query)
+ if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
ipoib_dbg_mcast(priv, "no address vector, "
"but multicast join already started\n");
else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
@@ -910,7 +862,6 @@ void ipoib_mcast_restart_task(void *dev_
/* We have to cancel outside of the spinlock */
list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
- wait_for_mcast_join(priv, mcast);
ipoib_mcast_leave(mcast->dev, mcast);
ipoib_mcast_free(mcast);
}
diff --git a/include/rdma/ib_multicast.h b/include/rdma/ib_multicast.h
new file mode 100755
index 0000000..423b754
--- /dev/null
+++ b/include/rdma/ib_multicast.h
@@ -0,0 +1,102 @@
+/*
+ * Copyright (c) 2006 Intel Corporation. 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.
+ */
+
+#ifndef IB_MULTICAST_H
+#define IB_MULTICAST_H
+
+#include <rdma/ib_sa.h>
+
+struct ib_multicast {
+ struct ib_sa_mcmember_rec rec;
+ ib_sa_comp_mask comp_mask;
+ int (*callback)(int status,
+ struct ib_multicast *multicast);
+ void *context;
+};
+
+/**
+ * ib_join_multicast - Initiates a join request to the specified multicast
+ * group.
+ * @device: Device associated with the multicast group.
+ * @port_num: Port on the specified device to associate with the multicast
+ * group.
+ * @rec: SA multicast member record specifying group attributes.
+ * @comp_mask: Component mask indicating which group attributes of %rec are
+ * valid.
+ * @gfp_mask: GFP mask for memory allocations.
+ * @callback: User callback invoked once the join operation completes.
+ * @context: User specified context stored with the ib_multicast structure.
+ *
+ * This call initiates a multicast join request with the SA for the specified
+ * multicast group. If the join operation is started successfully, it returns
+ * an ib_multicast structure that is used to track the multicast operation.
+ * Users must free this structure by calling ib_free_multicast, even if the
+ * join operation later fails. (The callback status is non-zero.)
+ */
+struct ib_multicast *ib_join_multicast(struct ib_device *device, u8 port_num,
+ struct ib_sa_mcmember_rec *rec,
+ ib_sa_comp_mask comp_mask, gfp_t gfp_mask,
+ int (*callback)(int status,
+ struct ib_multicast
+ *multicast),
+ void *context);
+
+/**
+ * ib_free_multicast - Frees the multicast tracking structure, and releases
+ * any reference on the multicast group.
+ * @multicast: Multicast tracking structure allocated by ib_join_multicast.
+ *
+ * This call blocks until the connection identifier is destroyed. It may
+ * not be called from within the multicast callback; however, returning a non-
+ * zero value from the callback will result in destroying the multicast
+ * tracking structure.
+ */
+void ib_free_multicast(struct ib_multicast *multicast);
+
+/**
+ * ib_get_mcmember_rec - Looks up a multicast member record by its MGID and
+ * returns it if found.
+ * @device: Device associated with the multicast group.
+ * @port_num: Port on the specified device to associate with the multicast
+ * group.
+ * @mgid: optional MGID of multicast group.
+ * @rec: Location to copy SA multicast member record.
+ *
+ * If an MGID is specified, returns an existing multicast member record if
+ * one is found for the local port. If no MGID is specified, or the specified
+ * MGID is 0, returns a multicast member record filled in with default values
+ * that may be used to create a new multicast group.
+ */
+int ib_get_mcmember_rec(struct ib_device *device, u8 port_num,
+ union ib_gid *mgid, struct ib_sa_mcmember_rec *rec);
+
+#endif /* IB_MULTICAST_H */
More information about the general
mailing list