IB/sa: Track multicast join/leave requests
authorSean Hefty <sean.hefty@intel.com>
Fri, 16 Feb 2007 01:00:17 +0000 (17:00 -0800)
committerRoland Dreier <rolandd@cisco.com>
Fri, 16 Feb 2007 22:20:02 +0000 (14:20 -0800)
The IB SA tracks multicast join/leave requests on a per port basis and
does not do any reference counting: if two users of the same port join
the same group, and one leaves that group, then the SA will remove the
port from the group even though there is one user who wants to stay a
member left.  Therefore, in order to support multiple users of the
same multicast group from the same port, we need to perform reference
counting locally.

To do this, add an multicast submodule to ib_sa to perform reference
counting of multicast join/leave operations.  Modify ib_ipoib (the
only in-kernel user of multicast) to use the new interface.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
drivers/infiniband/core/Makefile
drivers/infiniband/core/multicast.c [new file with mode: 0644]
drivers/infiniband/core/sa.h [new file with mode: 0644]
drivers/infiniband/core/sa_query.c
drivers/infiniband/ulp/ipoib/ipoib_multicast.c
include/rdma/ib_addr.h
include/rdma/ib_sa.h

index 50fb1cd447b76d80de09d6f565e737cd37101af5..189e5d4b9b17ed271027e2c015b252f929fbd32d 100644 (file)
@@ -12,7 +12,7 @@ ib_core-y :=                  packer.o ud_header.o verbs.o sysfs.o \
 
 ib_mad-y :=                    mad.o smi.o agent.o mad_rmpp.o
 
-ib_sa-y :=                     sa_query.o
+ib_sa-y :=                     sa_query.o multicast.o
 
 ib_cm-y :=                     cm.o
 
diff --git a/drivers/infiniband/core/multicast.c b/drivers/infiniband/core/multicast.c
new file mode 100644 (file)
index 0000000..4a579b3
--- /dev/null
@@ -0,0 +1,837 @@
+/*
+ * 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_cache.h>
+#include "sa.h"
+
+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 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;
+       struct ib_event_handler event_handler;
+       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_sa_multicast  multicast;
+       struct ib_sa_client     *client;
+       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 check_selector(ib_sa_comp_mask comp_mask,
+                         ib_sa_comp_mask selector_mask,
+                         ib_sa_comp_mask value_mask,
+                         u8 selector, u8 src_value, u8 dst_value)
+{
+       int err;
+
+       if (!(comp_mask & selector_mask) || !(comp_mask & value_mask))
+               return 0;
+
+       switch (selector) {
+       case IB_SA_GT:
+               err = (src_value <= dst_value);
+               break;
+       case IB_SA_LT:
+               err = (src_value >= dst_value);
+               break;
+       case IB_SA_EQ:
+               err = (src_value != dst_value);
+               break;
+       default:
+               err = 0;
+               break;
+       }
+
+       return err;
+}
+
+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 (check_selector(comp_mask, IB_SA_MCMEMBER_REC_MTU_SELECTOR,
+                          IB_SA_MCMEMBER_REC_MTU, dst->mtu_selector,
+                          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 (check_selector(comp_mask, IB_SA_MCMEMBER_REC_RATE_SELECTOR,
+                          IB_SA_MCMEMBER_REC_RATE, dst->rate_selector,
+                          src->rate, dst->rate))
+               return -EINVAL;
+       if (check_selector(comp_mask,
+                          IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME_SELECTOR,
+                          IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME,
+                          dst->packet_life_time_selector,
+                          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_query(&sa_client, port->dev->device,
+                                      port->port_num, IB_MGMT_METHOD_SET,
+                                      &member->multicast.rec,
+                                      member->multicast.comp_mask,
+                                      3000, 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_query(&sa_client, port->dev->device,
+                                      port->port_num, IB_SA_METHOD_DELETE, &rec,
+                                      IB_SA_MCMEMBER_REC_MGID     |
+                                      IB_SA_MCMEMBER_REC_PORT_GID |
+                                      IB_SA_MCMEMBER_REC_JOIN_STATE,
+                                      3000, 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_move(&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_sa_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(struct work_struct *work)
+{
+       struct mcast_group *group;
+       struct mcast_member *member;
+       struct ib_sa_multicast *multicast;
+       int status, ret;
+       u8 join_state;
+
+       group = container_of(work, typeof(*group), work);
+retest:
+       spin_lock_irq(&group->lock);
+       while (!list_empty(&group->pending_list) ||
+              (group->state == MCAST_ERROR)) {
+
+               if (group->state == MCAST_ERROR) {
+                       spin_unlock_irq(&group->lock);
+                       process_group_error(group);
+                       goto retest;
+               }
+
+               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_sa_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_sa_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->work);
+}
+
+static void leave_handler(int status, struct ib_sa_mcmember_rec *rec,
+                         void *context)
+{
+       struct mcast_group *group = context;
+
+       mcast_work_handler(&group->work);
+}
+
+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);
+       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_sa_multicast *
+ib_sa_join_multicast(struct ib_sa_client *client,
+                    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_sa_multicast *multicast),
+                    void *context)
+{
+       struct mcast_device *dev;
+       struct mcast_member *member;
+       struct ib_sa_multicast *multicast;
+       int ret;
+
+       dev = ib_get_client_data(device, &mcast_client);
+       if (!dev)
+               return ERR_PTR(-ENODEV);
+
+       member = kmalloc(sizeof *member, gfp_mask);
+       if (!member)
+               return ERR_PTR(-ENOMEM);
+
+       ib_sa_client_get(client);
+       member->client = client;
+       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:
+       ib_sa_client_put(client);
+       kfree(member);
+       return ERR_PTR(ret);
+}
+EXPORT_SYMBOL(ib_sa_join_multicast);
+
+void ib_sa_free_multicast(struct ib_sa_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);
+       ib_sa_client_put(member->client);
+       kfree(member);
+}
+EXPORT_SYMBOL(ib_sa_free_multicast);
+
+int ib_sa_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];
+       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);
+
+       return ret;
+}
+EXPORT_SYMBOL(ib_sa_get_mcmember_rec);
+
+int ib_init_ah_from_mcmember(struct ib_device *device, u8 port_num,
+                            struct ib_sa_mcmember_rec *rec,
+                            struct ib_ah_attr *ah_attr)
+{
+       int ret;
+       u16 gid_index;
+       u8 p;
+
+       ret = ib_find_cached_gid(device, &rec->port_gid, &p, &gid_index);
+       if (ret)
+               return ret;
+
+       memset(ah_attr, 0, sizeof *ah_attr);
+       ah_attr->dlid = be16_to_cpu(rec->mlid);
+       ah_attr->sl = rec->sl;
+       ah_attr->port_num = port_num;
+       ah_attr->static_rate = rec->rate;
+
+       ah_attr->ah_flags = IB_AH_GRH;
+       ah_attr->grh.dgid = rec->mgid;
+
+       ah_attr->grh.sgid_index = (u8) gid_index;
+       ah_attr->grh.flow_label = be32_to_cpu(rec->flow_label);
+       ah_attr->grh.hop_limit = rec->hop_limit;
+       ah_attr->grh.traffic_class = rec->traffic_class;
+
+       return 0;
+}
+EXPORT_SYMBOL(ib_init_ah_from_mcmember);
+
+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 = container_of(handler, struct mcast_device, event_handler);
+
+       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(&dev->event_handler, device, mcast_event_handler);
+       ib_register_event_handler(&dev->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(&dev->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);
+}
+
+int mcast_init(void)
+{
+       int ret;
+
+       mcast_wq = create_singlethread_workqueue("ib_mcast");
+       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;
+}
+
+void mcast_cleanup(void)
+{
+       ib_unregister_client(&mcast_client);
+       ib_sa_unregister_client(&sa_client);
+       destroy_workqueue(mcast_wq);
+}
diff --git a/drivers/infiniband/core/sa.h b/drivers/infiniband/core/sa.h
new file mode 100644 (file)
index 0000000..24c93fd
--- /dev/null
@@ -0,0 +1,66 @@
+/*
+ * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
+ * 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 SA_H
+#define SA_H
+
+#include <rdma/ib_sa.h>
+
+static inline void ib_sa_client_get(struct ib_sa_client *client)
+{
+       atomic_inc(&client->users);
+}
+
+static inline void ib_sa_client_put(struct ib_sa_client *client)
+{
+       if (atomic_dec_and_test(&client->users))
+               complete(&client->comp);
+}
+
+int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
+                            struct ib_device *device, u8 port_num,
+                            u8 method,
+                            struct ib_sa_mcmember_rec *rec,
+                            ib_sa_comp_mask comp_mask,
+                            int timeout_ms, gfp_t gfp_mask,
+                            void (*callback)(int status,
+                                             struct ib_sa_mcmember_rec *resp,
+                                             void *context),
+                            void *context,
+                            struct ib_sa_query **sa_query);
+
+int mcast_init(void);
+void mcast_cleanup(void);
+
+#endif /* SA_H */
index e45afba75341ee86d80464406564c3492366edbd..d7d4a5309ba939325ebccf149c3f7be7bd27dd5f 100644 (file)
@@ -47,8 +47,8 @@
 #include <linux/workqueue.h>
 
 #include <rdma/ib_pack.h>
-#include <rdma/ib_sa.h>
 #include <rdma/ib_cache.h>
+#include "sa.h"
 
 MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
@@ -425,17 +425,6 @@ void ib_sa_register_client(struct ib_sa_client *client)
 }
 EXPORT_SYMBOL(ib_sa_register_client);
 
-static inline void ib_sa_client_get(struct ib_sa_client *client)
-{
-       atomic_inc(&client->users);
-}
-
-static inline void ib_sa_client_put(struct ib_sa_client *client)
-{
-       if (atomic_dec_and_test(&client->users))
-               complete(&client->comp);
-}
-
 void ib_sa_unregister_client(struct ib_sa_client *client)
 {
        ib_sa_client_put(client);
@@ -901,7 +890,6 @@ err1:
        kfree(query);
        return ret;
 }
-EXPORT_SYMBOL(ib_sa_mcmember_rec_query);
 
 static void send_handler(struct ib_mad_agent *agent,
                         struct ib_mad_send_wc *mad_send_wc)
@@ -1053,14 +1041,27 @@ static int __init ib_sa_init(void)
        get_random_bytes(&tid, sizeof tid);
 
        ret = ib_register_client(&sa_client);
-       if (ret)
+       if (ret) {
                printk(KERN_ERR "Couldn't register ib_sa client\n");
+               goto err1;
+       }
+
+       ret = mcast_init();
+       if (ret) {
+               printk(KERN_ERR "Couldn't initialize multicast handling\n");
+               goto err2;
+       }
 
+       return 0;
+err2:
+       ib_unregister_client(&sa_client);
+err1:
        return ret;
 }
 
 static void __exit ib_sa_cleanup(void)
 {
+       mcast_cleanup();
        ib_unregister_client(&sa_client);
        idr_destroy(&query_idr);
 }
index fea737f520fdfeae5a8a7b85765799552406b379..b303ce6bc21ee6664d1471a3a99d2455e11aef2e 100644 (file)
@@ -60,14 +60,11 @@ 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_sa_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 +296,22 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
        return 0;
 }
 
-static void
+static int
 ipoib_mcast_sendonly_join_complete(int status,
-                                  struct ib_sa_mcmember_rec *mcmember,
-                                  void *mcast_ptr)
+                                  struct ib_sa_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 +326,10 @@ ipoib_mcast_sendonly_join_complete(int status,
                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 +359,33 @@ static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
        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_sa_join_multicast(&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,
+                                        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_sa_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_sa_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);
 
@@ -395,24 +393,25 @@ static void ipoib_mcast_join_complete(int status,
                        " (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_delayed_work(ipoib_workqueue,
                                           &priv->mcast_task, 0);
                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),
@@ -429,24 +428,18 @@ static void ipoib_mcast_join_complete(int status,
        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_delayed_work(ipoib_workqueue, &priv->mcast_task,
-                                          0);
-               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,
@@ -495,15 +488,14 @@ static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
                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_sa_join_multicast(&ipoib_sa_client, 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_sa_join_multicast failed, status %d\n", ret);
 
                mcast->backoff *= 2;
                if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
@@ -515,8 +507,7 @@ static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast,
                                           &priv->mcast_task,
                                           mcast->backoff * HZ);
                mutex_unlock(&mcast_mutex);
-       } else
-               mcast->query_id = ret;
+       }
 }
 
 void ipoib_mcast_join_task(struct work_struct *work)
@@ -541,7 +532,7 @@ void ipoib_mcast_join_task(struct work_struct *work)
                        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) {
@@ -568,7 +559,8 @@ void ipoib_mcast_join_task(struct work_struct *work)
        }
 
        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;
        }
 
@@ -625,26 +617,9 @@ int ipoib_mcast_start_thread(struct net_device *dev)
        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");
 
@@ -660,52 +635,27 @@ int ipoib_mcast_stop_thread(struct net_device *dev, int flush)
        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_sa_free_multicast(mcast->mc);
 
        return 0;
 }
@@ -758,7 +708,7 @@ void ipoib_mcast_send(struct net_device *dev, void *mgid, struct sk_buff *skb)
                        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))
@@ -916,7 +866,6 @@ void ipoib_mcast_restart_task(struct work_struct *work)
 
        /* 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);
        }
index c094e501286229bbc699bc74534c088235a7a436..c36750ff6ae82caf596aaee36d25de55e9d4316a 100644 (file)
@@ -110,6 +110,12 @@ static inline void ib_addr_set_pkey(struct rdma_dev_addr *dev_addr, u16 pkey)
        dev_addr->broadcast[9] = (unsigned char) pkey;
 }
 
+static inline void ib_addr_get_mgid(struct rdma_dev_addr *dev_addr,
+                                   union ib_gid *gid)
+{
+       memcpy(gid, dev_addr->broadcast + 4, sizeof *gid);
+}
+
 static inline void ib_addr_get_sgid(struct rdma_dev_addr *dev_addr,
                                    union ib_gid *gid)
 {
index 97715b0c20b69f6e50b75b2c03e24fa50758fdb6..5e26b2f53f86e198f1c8893c72d460d624a4c419 100644 (file)
@@ -285,18 +285,6 @@ int ib_sa_path_rec_get(struct ib_sa_client *client,
                       void *context,
                       struct ib_sa_query **query);
 
-int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
-                            struct ib_device *device, u8 port_num,
-                            u8 method,
-                            struct ib_sa_mcmember_rec *rec,
-                            ib_sa_comp_mask comp_mask,
-                            int timeout_ms, gfp_t gfp_mask,
-                            void (*callback)(int status,
-                                             struct ib_sa_mcmember_rec *resp,
-                                             void *context),
-                            void *context,
-                            struct ib_sa_query **query);
-
 int ib_sa_service_rec_query(struct ib_sa_client *client,
                         struct ib_device *device, u8 port_num,
                         u8 method,
@@ -309,93 +297,82 @@ int ib_sa_service_rec_query(struct ib_sa_client *client,
                         void *context,
                         struct ib_sa_query **sa_query);
 
+struct ib_sa_multicast {
+       struct ib_sa_mcmember_rec rec;
+       ib_sa_comp_mask         comp_mask;
+       int                     (*callback)(int status,
+                                           struct ib_sa_multicast *multicast);
+       void                    *context;
+};
+
 /**
- * ib_sa_mcmember_rec_set - Start an MCMember set query
- * @client:SA client
- * @device:device to send query on
- * @port_num: port number to send query on
- * @rec:MCMember Record to send in query
- * @comp_mask:component mask to send in query
- * @timeout_ms:time to wait for response
- * @gfp_mask:GFP mask to use for internal allocations
- * @callback:function called when query completes, times out or is
- * canceled
- * @context:opaque user context passed to callback
- * @sa_query:query context, used to cancel query
+ * ib_sa_join_multicast - Initiates a join request to the specified multicast
+ *   group.
+ * @client: SA client
+ * @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_sa_multicast structure.
  *
- * Send an MCMember Set query to the SA (eg to join a multicast
- * group).  The callback function will be called when the query
- * completes (or fails); status is 0 for a successful response, -EINTR
- * if the query is canceled, -ETIMEDOUT is the query timed out, or
- * -EIO if an error occurred sending the query.  The resp parameter of
- * the callback is only valid if status is 0.
+ * 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_sa_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.)
  *
- * If the return value of ib_sa_mcmember_rec_set() is negative, it is
- * an error code.  Otherwise it is a query ID that can be used to
- * cancel the query.
+ * If the join operation fails; status will be non-zero, with the following
+ * failures possible:
+ * -ETIMEDOUT: The request timed out.
+ * -EIO: An error occurred sending the query.
+ * -EINVAL: The MCMemberRecord values differed from the existing group's.
+ * -ENETRESET: Indicates that an fatal error has occurred on the multicast
+ *   group, and the user must rejoin the group to continue using it.
  */
-static inline int
-ib_sa_mcmember_rec_set(struct ib_sa_client *client,
-                      struct ib_device *device, u8 port_num,
-                      struct ib_sa_mcmember_rec *rec,
-                      ib_sa_comp_mask comp_mask,
-                      int timeout_ms, gfp_t gfp_mask,
-                      void (*callback)(int status,
-                                       struct ib_sa_mcmember_rec *resp,
-                                       void *context),
-                      void *context,
-                      struct ib_sa_query **query)
-{
-       return ib_sa_mcmember_rec_query(client, device, port_num,
-                                       IB_MGMT_METHOD_SET,
-                                       rec, comp_mask,
-                                       timeout_ms, gfp_mask, callback,
-                                       context, query);
-}
+struct ib_sa_multicast *ib_sa_join_multicast(struct ib_sa_client *client,
+                                            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_sa_multicast
+                                                                   *multicast),
+                                            void *context);
 
 /**
- * ib_sa_mcmember_rec_delete - Start an MCMember delete query
- * @client:SA client
- * @device:device to send query on
- * @port_num: port number to send query on
- * @rec:MCMember Record to send in query
- * @comp_mask:component mask to send in query
- * @timeout_ms:time to wait for response
- * @gfp_mask:GFP mask to use for internal allocations
- * @callback:function called when query completes, times out or is
- * canceled
- * @context:opaque user context passed to callback
- * @sa_query:query context, used to cancel query
- *
- * Send an MCMember Delete query to the SA (eg to leave a multicast
- * group).  The callback function will be called when the query
- * completes (or fails); status is 0 for a successful response, -EINTR
- * if the query is canceled, -ETIMEDOUT is the query timed out, or
- * -EIO if an error occurred sending the query.  The resp parameter of
- * the callback is only valid if status is 0.
+ * 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.
  *
- * If the return value of ib_sa_mcmember_rec_delete() is negative, it
- * is an error code.  Otherwise it is a query ID that can be used to
- * cancel the query.
+ * This call blocks until the multicast 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_sa_free_multicast(struct ib_sa_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: MGID of multicast group.
+ * @rec: Location to copy SA multicast member record.
  */
-static inline int
-ib_sa_mcmember_rec_delete(struct ib_sa_client *client,
-                         struct ib_device *device, u8 port_num,
-                         struct ib_sa_mcmember_rec *rec,
-                         ib_sa_comp_mask comp_mask,
-                         int timeout_ms, gfp_t gfp_mask,
-                         void (*callback)(int status,
-                                          struct ib_sa_mcmember_rec *resp,
-                                          void *context),
-                         void *context,
-                         struct ib_sa_query **query)
-{
-       return ib_sa_mcmember_rec_query(client, device, port_num,
-                                       IB_SA_METHOD_DELETE,
-                                       rec, comp_mask,
-                                       timeout_ms, gfp_mask, callback,
-                                       context, query);
-}
+int ib_sa_get_mcmember_rec(struct ib_device *device, u8 port_num,
+                          union ib_gid *mgid, struct ib_sa_mcmember_rec *rec);
+
+/**
+ * ib_init_ah_from_mcmember - Initialize address handle attributes based on
+ * an SA multicast member record.
+ */
+int ib_init_ah_from_mcmember(struct ib_device *device, u8 port_num,
+                            struct ib_sa_mcmember_rec *rec,
+                            struct ib_ah_attr *ah_attr);
 
 /**
  * ib_init_ah_from_path - Initialize address handle attributes based on an SA