[PATCH] IB: Add RMPP implementation
authorHal Rosenstock <halr@voltaire.com>
Wed, 27 Jul 2005 18:45:37 +0000 (11:45 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Wed, 27 Jul 2005 23:26:13 +0000 (16:26 -0700)
Add RMPP implementation.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Hal Rosenstock <halr@voltaire.com>
Cc: Roland Dreier <rolandd@cisco.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
drivers/infiniband/core/Makefile
drivers/infiniband/core/mad.c
drivers/infiniband/core/mad_priv.h
drivers/infiniband/core/mad_rmpp.c [new file with mode: 0644]
drivers/infiniband/core/mad_rmpp.h [new file with mode: 0644]

index e1a7cf3..96b8eba 100644 (file)
@@ -6,7 +6,7 @@ obj-$(CONFIG_INFINIBAND_USER_VERBS) +=  ib_uverbs.o
 ib_core-y :=                   packer.o ud_header.o verbs.o sysfs.o \
                                device.o fmr_pool.o cache.o
 
-ib_mad-y :=                    mad.o smi.o agent.o
+ib_mad-y :=                    mad.o smi.o agent.o mad_rmpp.o
 
 ib_sa-y :=                     sa_query.o
 
index 26e2b59..b97e210 100644 (file)
@@ -1,5 +1,7 @@
 /*
  * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved.
+ * Copyright (c) 2005 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2005 Mellanox Technologies Ltd.  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
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: mad.c 1389 2004-12-27 22:56:47Z roland $
+ * $Id: mad.c 2817 2005-07-07 11:29:26Z halr $
  */
-
 #include <linux/dma-mapping.h>
 
 #include "mad_priv.h"
+#include "mad_rmpp.h"
 #include "smi.h"
 #include "agent.h"
 
@@ -45,6 +47,7 @@ MODULE_AUTHOR("Sean Hefty");
 
 
 kmem_cache_t *ib_mad_cache;
+
 static struct list_head ib_mad_port_list;
 static u32 ib_mad_client_id = 0;
 
@@ -62,8 +65,6 @@ static struct ib_mad_agent_private *find_mad_agent(
 static int ib_mad_post_receive_mads(struct ib_mad_qp_info *qp_info,
                                    struct ib_mad_private *mad);
 static void cancel_mads(struct ib_mad_agent_private *mad_agent_priv);
-static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
-                                   struct ib_mad_send_wc *mad_send_wc);
 static void timeout_sends(void *data);
 static void local_completions(void *data);
 static int add_nonoui_reg_req(struct ib_mad_reg_req *mad_reg_req,
@@ -195,8 +196,8 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
        if (qpn == -1)
                goto error1;
 
-       if (rmpp_version)
-               goto error1;    /* XXX: until RMPP implemented */
+       if (rmpp_version && rmpp_version != IB_MGMT_RMPP_VERSION)
+               goto error1;
 
        /* Validate MAD registration request if supplied */
        if (mad_reg_req) {
@@ -281,7 +282,7 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
        /* Now, fill in the various structures */
        mad_agent_priv->qp_info = &port_priv->qp_info[qpn];
        mad_agent_priv->reg_req = reg_req;
-       mad_agent_priv->rmpp_version = rmpp_version;
+       mad_agent_priv->agent.rmpp_version = rmpp_version;
        mad_agent_priv->agent.device = device;
        mad_agent_priv->agent.recv_handler = recv_handler;
        mad_agent_priv->agent.send_handler = send_handler;
@@ -341,6 +342,7 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
        INIT_LIST_HEAD(&mad_agent_priv->send_list);
        INIT_LIST_HEAD(&mad_agent_priv->wait_list);
        INIT_LIST_HEAD(&mad_agent_priv->done_list);
+       INIT_LIST_HEAD(&mad_agent_priv->rmpp_list);
        INIT_WORK(&mad_agent_priv->timed_work, timeout_sends, mad_agent_priv);
        INIT_LIST_HEAD(&mad_agent_priv->local_list);
        INIT_WORK(&mad_agent_priv->local_work, local_completions,
@@ -502,6 +504,7 @@ static void unregister_mad_agent(struct ib_mad_agent_private *mad_agent_priv)
        spin_unlock_irqrestore(&port_priv->reg_lock, flags);
 
        flush_workqueue(port_priv->wq);
+       ib_cancel_rmpp_recvs(mad_agent_priv);
 
        atomic_dec(&mad_agent_priv->refcount);
        wait_event(mad_agent_priv->wait,
@@ -786,12 +789,15 @@ struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
        int buf_size;
        void *buf;
 
-       if (rmpp_active)
-               return ERR_PTR(-EINVAL);        /* until RMPP implemented */
        mad_agent_priv = container_of(mad_agent,
                                      struct ib_mad_agent_private, agent);
        buf_size = get_buf_length(hdr_len, data_len);
 
+       if ((!mad_agent->rmpp_version &&
+            (rmpp_active || buf_size > sizeof(struct ib_mad))) ||
+           (!rmpp_active && buf_size > sizeof(struct ib_mad)))
+               return ERR_PTR(-EINVAL);
+
        buf = kmalloc(sizeof *send_buf + buf_size, gfp_mask);
        if (!buf)
                return ERR_PTR(-ENOMEM);
@@ -816,6 +822,18 @@ struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
        send_buf->send_wr.wr.ud.remote_qpn = remote_qpn;
        send_buf->send_wr.wr.ud.remote_qkey = IB_QP_SET_QKEY;
        send_buf->send_wr.wr.ud.pkey_index = pkey_index;
+
+       if (rmpp_active) {
+               struct ib_rmpp_mad *rmpp_mad;
+               rmpp_mad = (struct ib_rmpp_mad *)send_buf->mad;
+               rmpp_mad->rmpp_hdr.paylen_newwin = cpu_to_be32(hdr_len -
+                       offsetof(struct ib_rmpp_mad, data) + data_len);
+               rmpp_mad->rmpp_hdr.rmpp_version = mad_agent->rmpp_version;
+               rmpp_mad->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_DATA;
+               ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr,
+                                 IB_MGMT_RMPP_FLAG_ACTIVE);
+       }
+
        send_buf->mad_agent = mad_agent;
        atomic_inc(&mad_agent_priv->refcount);
        return send_buf;
@@ -839,7 +857,7 @@ void ib_free_send_mad(struct ib_mad_send_buf *send_buf)
 }
 EXPORT_SYMBOL(ib_free_send_mad);
 
-static int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr)
+int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr)
 {
        struct ib_mad_qp_info *qp_info;
        struct ib_send_wr *bad_send_wr;
@@ -940,13 +958,13 @@ int ib_post_send_mad(struct ib_mad_agent *mad_agent,
                        ret = -ENOMEM;
                        goto error2;
                }
+               memset(mad_send_wr, 0, sizeof *mad_send_wr);
 
                mad_send_wr->send_wr = *send_wr;
                mad_send_wr->send_wr.sg_list = mad_send_wr->sg_list;
                memcpy(mad_send_wr->sg_list, send_wr->sg_list,
                       sizeof *send_wr->sg_list * send_wr->num_sge);
-               mad_send_wr->wr_id = mad_send_wr->send_wr.wr_id;
-               mad_send_wr->send_wr.next = NULL;
+               mad_send_wr->wr_id = send_wr->wr_id;
                mad_send_wr->tid = send_wr->wr.ud.mad_hdr->tid;
                mad_send_wr->mad_agent_priv = mad_agent_priv;
                /* Timeout will be updated after send completes */
@@ -964,8 +982,13 @@ int ib_post_send_mad(struct ib_mad_agent *mad_agent,
                              &mad_agent_priv->send_list);
                spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 
-               ret = ib_send_mad(mad_send_wr);
-               if (ret) {
+               if (mad_agent_priv->agent.rmpp_version) {
+                       ret = ib_send_rmpp_mad(mad_send_wr);
+                       if (ret >= 0 && ret != IB_RMPP_RESULT_CONSUMED)
+                               ret = ib_send_mad(mad_send_wr);
+               } else
+                       ret = ib_send_mad(mad_send_wr);
+               if (ret < 0) {
                        /* Fail send request */
                        spin_lock_irqsave(&mad_agent_priv->lock, flags);
                        list_del(&mad_send_wr->agent_list);
@@ -991,31 +1014,25 @@ EXPORT_SYMBOL(ib_post_send_mad);
  */
 void ib_free_recv_mad(struct ib_mad_recv_wc *mad_recv_wc)
 {
-       struct ib_mad_recv_buf *entry;
+       struct ib_mad_recv_buf *mad_recv_buf, *temp_recv_buf;
        struct ib_mad_private_header *mad_priv_hdr;
        struct ib_mad_private *priv;
+       struct list_head free_list;
 
-       mad_priv_hdr = container_of(mad_recv_wc,
-                                   struct ib_mad_private_header,
-                                   recv_wc);
-       priv = container_of(mad_priv_hdr, struct ib_mad_private, header);
+       INIT_LIST_HEAD(&free_list);
+       list_splice_init(&mad_recv_wc->rmpp_list, &free_list);
 
-       /*
-        * Walk receive buffer list associated with this WC
-        * No need to remove them from list of receive buffers
-        */
-       list_for_each_entry(entry, &mad_recv_wc->recv_buf.list, list) {
-               /* Free previous receive buffer */
-               kmem_cache_free(ib_mad_cache, priv);
+       list_for_each_entry_safe(mad_recv_buf, temp_recv_buf,
+                                       &free_list, list) {
+               mad_recv_wc = container_of(mad_recv_buf, struct ib_mad_recv_wc,
+                                          recv_buf);
                mad_priv_hdr = container_of(mad_recv_wc,
                                            struct ib_mad_private_header,
                                            recv_wc);
                priv = container_of(mad_priv_hdr, struct ib_mad_private,
                                    header);
+               kmem_cache_free(ib_mad_cache, priv);
        }
-
-       /* Free last buffer */
-       kmem_cache_free(ib_mad_cache, priv);
 }
 EXPORT_SYMBOL(ib_free_recv_mad);
 
@@ -1524,9 +1541,20 @@ out:
        return valid;
 }
 
-static struct ib_mad_send_wr_private*
-find_send_req(struct ib_mad_agent_private *mad_agent_priv,
-             u64 tid)
+static int is_data_mad(struct ib_mad_agent_private *mad_agent_priv,
+                      struct ib_mad_hdr *mad_hdr)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+
+       rmpp_mad = (struct ib_rmpp_mad *)mad_hdr;
+       return !mad_agent_priv->agent.rmpp_version ||
+               !(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+                                   IB_MGMT_RMPP_FLAG_ACTIVE) ||
+               (rmpp_mad->rmpp_hdr.rmpp_type == IB_MGMT_RMPP_TYPE_DATA);
+}
+
+struct ib_mad_send_wr_private*
+ib_find_send_mad(struct ib_mad_agent_private *mad_agent_priv, u64 tid)
 {
        struct ib_mad_send_wr_private *mad_send_wr;
 
@@ -1542,7 +1570,9 @@ find_send_req(struct ib_mad_agent_private *mad_agent_priv,
         */
        list_for_each_entry(mad_send_wr, &mad_agent_priv->send_list,
                            agent_list) {
-               if (mad_send_wr->tid == tid && mad_send_wr->timeout) {
+               if (is_data_mad(mad_agent_priv,
+                               mad_send_wr->send_wr.wr.ud.mad_hdr) &&
+                   mad_send_wr->tid == tid && mad_send_wr->timeout) {
                        /* Verify request has not been canceled */
                        return (mad_send_wr->status == IB_WC_SUCCESS) ?
                                mad_send_wr : NULL;
@@ -1551,7 +1581,7 @@ find_send_req(struct ib_mad_agent_private *mad_agent_priv,
        return NULL;
 }
 
-static void ib_mark_req_done(struct ib_mad_send_wr_private *mad_send_wr)
+void ib_mark_mad_done(struct ib_mad_send_wr_private *mad_send_wr)
 {
        mad_send_wr->timeout = 0;
        if (mad_send_wr->refcount == 1) {
@@ -1569,12 +1599,23 @@ static void ib_mad_complete_recv(struct ib_mad_agent_private *mad_agent_priv,
        unsigned long flags;
        u64 tid;
 
-       INIT_LIST_HEAD(&mad_recv_wc->recv_buf.list);
+       INIT_LIST_HEAD(&mad_recv_wc->rmpp_list);
+       list_add(&mad_recv_wc->recv_buf.list, &mad_recv_wc->rmpp_list);
+       if (mad_agent_priv->agent.rmpp_version) {
+               mad_recv_wc = ib_process_rmpp_recv_wc(mad_agent_priv,
+                                                     mad_recv_wc);
+               if (!mad_recv_wc) {
+                       if (atomic_dec_and_test(&mad_agent_priv->refcount))
+                               wake_up(&mad_agent_priv->wait);
+                       return;
+               }
+       }
+
        /* Complete corresponding request */
        if (response_mad(mad_recv_wc->recv_buf.mad)) {
                tid = mad_recv_wc->recv_buf.mad->mad_hdr.tid;
                spin_lock_irqsave(&mad_agent_priv->lock, flags);
-               mad_send_wr = find_send_req(mad_agent_priv, tid);
+               mad_send_wr = ib_find_send_mad(mad_agent_priv, tid);
                if (!mad_send_wr) {
                        spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
                        ib_free_recv_mad(mad_recv_wc);
@@ -1582,7 +1623,7 @@ static void ib_mad_complete_recv(struct ib_mad_agent_private *mad_agent_priv,
                                wake_up(&mad_agent_priv->wait);
                        return;
                }
-               ib_mark_req_done(mad_send_wr);
+               ib_mark_mad_done(mad_send_wr);
                spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 
                /* Defined behavior is to complete response before request */
@@ -1787,14 +1828,22 @@ void ib_reset_mad_timeout(struct ib_mad_send_wr_private *mad_send_wr,
 /*
  * Process a send work completion
  */
-static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
-                                   struct ib_mad_send_wc *mad_send_wc)
+void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
+                            struct ib_mad_send_wc *mad_send_wc)
 {
        struct ib_mad_agent_private     *mad_agent_priv;
        unsigned long                   flags;
+       int                             ret;
 
        mad_agent_priv = mad_send_wr->mad_agent_priv;
        spin_lock_irqsave(&mad_agent_priv->lock, flags);
+       if (mad_agent_priv->agent.rmpp_version) {
+               ret = ib_process_rmpp_send_wc(mad_send_wr, mad_send_wc);
+               if (ret == IB_RMPP_RESULT_CONSUMED)
+                       goto done;
+       } else
+               ret = IB_RMPP_RESULT_UNHANDLED;
+
        if (mad_send_wc->status != IB_WC_SUCCESS &&
            mad_send_wr->status == IB_WC_SUCCESS) {
                mad_send_wr->status = mad_send_wc->status;
@@ -1806,8 +1855,7 @@ static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
                    mad_send_wr->status == IB_WC_SUCCESS) {
                        wait_for_response(mad_send_wr);
                }
-               spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
-               return;
+               goto done;
        }
 
        /* Remove send from MAD agent and notify client of completion */
@@ -1817,14 +1865,18 @@ static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
 
        if (mad_send_wr->status != IB_WC_SUCCESS )
                mad_send_wc->status = mad_send_wr->status;
-       mad_agent_priv->agent.send_handler(&mad_agent_priv->agent,
-                                           mad_send_wc);
+       if (ret != IB_RMPP_RESULT_INTERNAL)
+               mad_agent_priv->agent.send_handler(&mad_agent_priv->agent,
+                                                  mad_send_wc);
 
        /* Release reference on agent taken when sending */
        if (atomic_dec_and_test(&mad_agent_priv->refcount))
                wake_up(&mad_agent_priv->wait);
 
        kfree(mad_send_wr);
+       return;
+done:
+       spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 }
 
 static void ib_mad_send_done_handler(struct ib_mad_port_private *port_priv,
@@ -2036,7 +2088,9 @@ find_send_by_wr_id(struct ib_mad_agent_private *mad_agent_priv, u64 wr_id)
 
        list_for_each_entry(mad_send_wr, &mad_agent_priv->send_list,
                            agent_list) {
-               if (mad_send_wr->wr_id == wr_id)
+               if (is_data_mad(mad_agent_priv,
+                               mad_send_wr->send_wr.wr.ud.mad_hdr) &&
+                   mad_send_wr->wr_id == wr_id)
                        return mad_send_wr;
        }
        return NULL;
@@ -2118,7 +2172,9 @@ static void local_completions(void *data)
                        local->mad_priv->header.recv_wc.wc = &wc;
                        local->mad_priv->header.recv_wc.mad_len =
                                                sizeof(struct ib_mad);
-                       INIT_LIST_HEAD(&local->mad_priv->header.recv_wc.recv_buf.list);
+                       INIT_LIST_HEAD(&local->mad_priv->header.recv_wc.rmpp_list);
+                       list_add(&local->mad_priv->header.recv_wc.recv_buf.list,
+                                &local->mad_priv->header.recv_wc.rmpp_list);
                        local->mad_priv->header.recv_wc.recv_buf.grh = NULL;
                        local->mad_priv->header.recv_wc.recv_buf.mad =
                                                &local->mad_priv->mad.mad;
@@ -2166,7 +2222,21 @@ static int retry_send(struct ib_mad_send_wr_private *mad_send_wr)
        mad_send_wr->timeout = msecs_to_jiffies(mad_send_wr->send_wr.
                                                wr.ud.timeout_ms);
 
-       ret = ib_send_mad(mad_send_wr);
+       if (mad_send_wr->mad_agent_priv->agent.rmpp_version) {
+               ret = ib_retry_rmpp(mad_send_wr);
+               switch (ret) {
+               case IB_RMPP_RESULT_UNHANDLED:
+                       ret = ib_send_mad(mad_send_wr);
+                       break;
+               case IB_RMPP_RESULT_CONSUMED:
+                       ret = 0;
+                       break;
+               default:
+                       ret = -ECOMM;
+                       break;
+               }
+       } else
+               ret = ib_send_mad(mad_send_wr);
 
        if (!ret) {
                mad_send_wr->refcount++;
@@ -2724,3 +2794,4 @@ static void __exit ib_mad_cleanup_module(void)
 
 module_init(ib_mad_init_module);
 module_exit(ib_mad_cleanup_module);
+
index e5e37b5..568da10 100644 (file)
@@ -1,5 +1,7 @@
 /*
  * Copyright (c) 2004, 2005, Voltaire, Inc. All rights reserved.
+ * Copyright (c) 2005 Intel Corporation. All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -29,7 +31,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: mad_priv.h 1980 2005-03-11 22:33:53Z sean.hefty $
+ * $Id: mad_priv.h 2730 2005-06-28 16:43:03Z sean.hefty $
  */
 
 #ifndef __IB_MAD_PRIV_H__
@@ -97,10 +99,10 @@ struct ib_mad_agent_private {
        unsigned long timeout;
        struct list_head local_list;
        struct work_struct local_work;
+       struct list_head rmpp_list;
 
        atomic_t refcount;
        wait_queue_head_t wait;
-       u8 rmpp_version;
 };
 
 struct ib_mad_snoop_private {
@@ -125,6 +127,14 @@ struct ib_mad_send_wr_private {
        int retry;
        int refcount;
        enum ib_wc_status status;
+
+       /* RMPP control */
+       int last_ack;
+       int seg_num;
+       int newwin;
+       int total_seg;
+       int data_offset;
+       int pad;
 };
 
 struct ib_mad_local_private {
@@ -135,7 +145,6 @@ struct ib_mad_local_private {
        struct ib_sge sg_list[IB_MAD_SEND_REQ_MAX_SG];
        u64 wr_id;                      /* client WR ID */
        u64 tid;
-       int retries;
 };
 
 struct ib_mad_mgmt_method_table {
@@ -198,4 +207,17 @@ struct ib_mad_port_private {
 
 extern kmem_cache_t *ib_mad_cache;
 
+int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr);
+
+struct ib_mad_send_wr_private *
+ib_find_send_mad(struct ib_mad_agent_private *mad_agent_priv, u64 tid);
+
+void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
+                            struct ib_mad_send_wc *mad_send_wc);
+
+void ib_mark_mad_done(struct ib_mad_send_wr_private *mad_send_wr);
+
+void ib_reset_mad_timeout(struct ib_mad_send_wr_private *mad_send_wr,
+                         int timeout_ms);
+
 #endif /* __IB_MAD_PRIV_H__ */
diff --git a/drivers/infiniband/core/mad_rmpp.c b/drivers/infiniband/core/mad_rmpp.c
new file mode 100644 (file)
index 0000000..8f1eb80
--- /dev/null
@@ -0,0 +1,765 @@
+/*
+ * Copyright (c) 2005 Intel Inc. All rights reserved.
+ * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mad_rmpp.c 1921 2005-03-02 22:58:44Z sean.hefty $
+ */
+
+#include <linux/dma-mapping.h>
+
+#include "mad_priv.h"
+#include "mad_rmpp.h"
+
+enum rmpp_state {
+       RMPP_STATE_ACTIVE,
+       RMPP_STATE_TIMEOUT,
+       RMPP_STATE_COMPLETE
+};
+
+struct mad_rmpp_recv {
+       struct ib_mad_agent_private *agent;
+       struct list_head list;
+       struct work_struct timeout_work;
+       struct work_struct cleanup_work;
+       wait_queue_head_t wait;
+       enum rmpp_state state;
+       spinlock_t lock;
+       atomic_t refcount;
+
+       struct ib_ah *ah;
+       struct ib_mad_recv_wc *rmpp_wc;
+       struct ib_mad_recv_buf *cur_seg_buf;
+       int last_ack;
+       int seg_num;
+       int newwin;
+
+       u64 tid;
+       u32 src_qp;
+       u16 slid;
+       u8 mgmt_class;
+       u8 class_version;
+       u8 method;
+};
+
+static void destroy_rmpp_recv(struct mad_rmpp_recv *rmpp_recv)
+{
+       atomic_dec(&rmpp_recv->refcount);
+       wait_event(rmpp_recv->wait, !atomic_read(&rmpp_recv->refcount));
+       ib_destroy_ah(rmpp_recv->ah);
+       kfree(rmpp_recv);
+}
+
+void ib_cancel_rmpp_recvs(struct ib_mad_agent_private *agent)
+{
+       struct mad_rmpp_recv *rmpp_recv, *temp_rmpp_recv;
+       unsigned long flags;
+
+       spin_lock_irqsave(&agent->lock, flags);
+       list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
+               cancel_delayed_work(&rmpp_recv->timeout_work);
+               cancel_delayed_work(&rmpp_recv->cleanup_work);
+       }
+       spin_unlock_irqrestore(&agent->lock, flags);
+
+       flush_workqueue(agent->qp_info->port_priv->wq);
+
+       list_for_each_entry_safe(rmpp_recv, temp_rmpp_recv,
+                                &agent->rmpp_list, list) {
+               list_del(&rmpp_recv->list);
+               if (rmpp_recv->state != RMPP_STATE_COMPLETE)
+                       ib_free_recv_mad(rmpp_recv->rmpp_wc);
+               destroy_rmpp_recv(rmpp_recv);
+       }
+}
+
+static void recv_timeout_handler(void *data)
+{
+       struct mad_rmpp_recv *rmpp_recv = data;
+       struct ib_mad_recv_wc *rmpp_wc;
+       unsigned long flags;
+
+       spin_lock_irqsave(&rmpp_recv->agent->lock, flags);
+       if (rmpp_recv->state != RMPP_STATE_ACTIVE) {
+               spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
+               return;
+       }
+       rmpp_recv->state = RMPP_STATE_TIMEOUT;
+       list_del(&rmpp_recv->list);
+       spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
+
+       /* TODO: send abort. */
+       rmpp_wc = rmpp_recv->rmpp_wc;
+       destroy_rmpp_recv(rmpp_recv);
+       ib_free_recv_mad(rmpp_wc);
+}
+
+static void recv_cleanup_handler(void *data)
+{
+       struct mad_rmpp_recv *rmpp_recv = data;
+       unsigned long flags;
+
+       spin_lock_irqsave(&rmpp_recv->agent->lock, flags);
+       list_del(&rmpp_recv->list);
+       spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
+       destroy_rmpp_recv(rmpp_recv);
+}
+
+static struct mad_rmpp_recv *
+create_rmpp_recv(struct ib_mad_agent_private *agent,
+                struct ib_mad_recv_wc *mad_recv_wc)
+{
+       struct mad_rmpp_recv *rmpp_recv;
+       struct ib_mad_hdr *mad_hdr;
+
+       rmpp_recv = kmalloc(sizeof *rmpp_recv, GFP_KERNEL);
+       if (!rmpp_recv)
+               return NULL;
+
+       rmpp_recv->ah = ib_create_ah_from_wc(agent->agent.qp->pd,
+                                            mad_recv_wc->wc,
+                                            mad_recv_wc->recv_buf.grh,
+                                            agent->agent.port_num);
+       if (IS_ERR(rmpp_recv->ah))
+               goto error;
+
+       rmpp_recv->agent = agent;
+       init_waitqueue_head(&rmpp_recv->wait);
+       INIT_WORK(&rmpp_recv->timeout_work, recv_timeout_handler, rmpp_recv);
+       INIT_WORK(&rmpp_recv->cleanup_work, recv_cleanup_handler, rmpp_recv);
+       spin_lock_init(&rmpp_recv->lock);
+       rmpp_recv->state = RMPP_STATE_ACTIVE;
+       atomic_set(&rmpp_recv->refcount, 1);
+
+       rmpp_recv->rmpp_wc = mad_recv_wc;
+       rmpp_recv->cur_seg_buf = &mad_recv_wc->recv_buf;
+       rmpp_recv->newwin = 1;
+       rmpp_recv->seg_num = 1;
+       rmpp_recv->last_ack = 0;
+
+       mad_hdr = &mad_recv_wc->recv_buf.mad->mad_hdr;
+       rmpp_recv->tid = mad_hdr->tid;
+       rmpp_recv->src_qp = mad_recv_wc->wc->src_qp;
+       rmpp_recv->slid = mad_recv_wc->wc->slid;
+       rmpp_recv->mgmt_class = mad_hdr->mgmt_class;
+       rmpp_recv->class_version = mad_hdr->class_version;
+       rmpp_recv->method  = mad_hdr->method;
+       return rmpp_recv;
+
+error: kfree(rmpp_recv);
+       return NULL;
+}
+
+static inline void deref_rmpp_recv(struct mad_rmpp_recv *rmpp_recv)
+{
+       if (atomic_dec_and_test(&rmpp_recv->refcount))
+               wake_up(&rmpp_recv->wait);
+}
+
+static struct mad_rmpp_recv *
+find_rmpp_recv(struct ib_mad_agent_private *agent,
+              struct ib_mad_recv_wc *mad_recv_wc)
+{
+       struct mad_rmpp_recv *rmpp_recv;
+       struct ib_mad_hdr *mad_hdr = &mad_recv_wc->recv_buf.mad->mad_hdr;
+
+       list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
+               if (rmpp_recv->tid == mad_hdr->tid &&
+                   rmpp_recv->src_qp == mad_recv_wc->wc->src_qp &&
+                   rmpp_recv->slid == mad_recv_wc->wc->slid &&
+                   rmpp_recv->mgmt_class == mad_hdr->mgmt_class &&
+                   rmpp_recv->class_version == mad_hdr->class_version &&
+                   rmpp_recv->method == mad_hdr->method)
+                       return rmpp_recv;
+       }
+       return NULL;
+}
+
+static struct mad_rmpp_recv *
+acquire_rmpp_recv(struct ib_mad_agent_private *agent,
+                 struct ib_mad_recv_wc *mad_recv_wc)
+{
+       struct mad_rmpp_recv *rmpp_recv;
+       unsigned long flags;
+
+       spin_lock_irqsave(&agent->lock, flags);
+       rmpp_recv = find_rmpp_recv(agent, mad_recv_wc);
+       if (rmpp_recv)
+               atomic_inc(&rmpp_recv->refcount);
+       spin_unlock_irqrestore(&agent->lock, flags);
+       return rmpp_recv;
+}
+
+static struct mad_rmpp_recv *
+insert_rmpp_recv(struct ib_mad_agent_private *agent,
+                struct mad_rmpp_recv *rmpp_recv)
+{
+       struct mad_rmpp_recv *cur_rmpp_recv;
+
+       cur_rmpp_recv = find_rmpp_recv(agent, rmpp_recv->rmpp_wc);
+       if (!cur_rmpp_recv)
+               list_add_tail(&rmpp_recv->list, &agent->rmpp_list);
+
+       return cur_rmpp_recv;
+}
+
+static int data_offset(u8 mgmt_class)
+{
+       if (mgmt_class == IB_MGMT_CLASS_SUBN_ADM)
+               return offsetof(struct ib_sa_mad, data);
+       else if ((mgmt_class >= IB_MGMT_CLASS_VENDOR_RANGE2_START) &&
+                (mgmt_class <= IB_MGMT_CLASS_VENDOR_RANGE2_END))
+               return offsetof(struct ib_vendor_mad, data);
+       else
+               return offsetof(struct ib_rmpp_mad, data);
+}
+
+static void format_ack(struct ib_rmpp_mad *ack,
+                      struct ib_rmpp_mad *data,
+                      struct mad_rmpp_recv *rmpp_recv)
+{
+       unsigned long flags;
+
+       memcpy(&ack->mad_hdr, &data->mad_hdr,
+              data_offset(data->mad_hdr.mgmt_class));
+
+       ack->mad_hdr.method ^= IB_MGMT_METHOD_RESP;
+       ack->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_ACK;
+       ib_set_rmpp_flags(&ack->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
+
+       spin_lock_irqsave(&rmpp_recv->lock, flags);
+       rmpp_recv->last_ack = rmpp_recv->seg_num;
+       ack->rmpp_hdr.seg_num = cpu_to_be32(rmpp_recv->seg_num);
+       ack->rmpp_hdr.paylen_newwin = cpu_to_be32(rmpp_recv->newwin);
+       spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+}
+
+static void ack_recv(struct mad_rmpp_recv *rmpp_recv,
+                    struct ib_mad_recv_wc *recv_wc)
+{
+       struct ib_mad_send_buf *msg;
+       struct ib_send_wr *bad_send_wr;
+       int hdr_len, ret;
+
+       hdr_len = sizeof(struct ib_mad_hdr) + sizeof(struct ib_rmpp_hdr);
+       msg = ib_create_send_mad(&rmpp_recv->agent->agent, recv_wc->wc->src_qp,
+                                recv_wc->wc->pkey_index, rmpp_recv->ah, 1,
+                                hdr_len, sizeof(struct ib_rmpp_mad) - hdr_len,
+                                GFP_KERNEL);
+       if (!msg)
+               return;
+
+       format_ack((struct ib_rmpp_mad *) msg->mad,
+                  (struct ib_rmpp_mad *) recv_wc->recv_buf.mad, rmpp_recv);
+       ret = ib_post_send_mad(&rmpp_recv->agent->agent, &msg->send_wr,
+                              &bad_send_wr);
+       if (ret)
+               ib_free_send_mad(msg);
+}
+
+static inline int get_last_flag(struct ib_mad_recv_buf *seg)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+
+       rmpp_mad = (struct ib_rmpp_mad *) seg->mad;
+       return ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_LAST;
+}
+
+static inline int get_seg_num(struct ib_mad_recv_buf *seg)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+
+       rmpp_mad = (struct ib_rmpp_mad *) seg->mad;
+       return be32_to_cpu(rmpp_mad->rmpp_hdr.seg_num);
+}
+
+static inline struct ib_mad_recv_buf * get_next_seg(struct list_head *rmpp_list,
+                                                   struct ib_mad_recv_buf *seg)
+{
+       if (seg->list.next == rmpp_list)
+               return NULL;
+
+       return container_of(seg->list.next, struct ib_mad_recv_buf, list);
+}
+
+static inline int window_size(struct ib_mad_agent_private *agent)
+{
+       return max(agent->qp_info->recv_queue.max_active >> 3, 1);
+}
+
+static struct ib_mad_recv_buf * find_seg_location(struct list_head *rmpp_list,
+                                                 int seg_num)
+{
+        struct ib_mad_recv_buf *seg_buf;
+       int cur_seg_num;
+
+       list_for_each_entry_reverse(seg_buf, rmpp_list, list) {
+               cur_seg_num = get_seg_num(seg_buf);
+               if (seg_num > cur_seg_num)
+                       return seg_buf;
+               if (seg_num == cur_seg_num)
+                       break;
+       }
+       return NULL;
+}
+
+static void update_seg_num(struct mad_rmpp_recv *rmpp_recv,
+                          struct ib_mad_recv_buf *new_buf)
+{
+       struct list_head *rmpp_list = &rmpp_recv->rmpp_wc->rmpp_list;
+
+       while (new_buf && (get_seg_num(new_buf) == rmpp_recv->seg_num + 1)) {
+               rmpp_recv->cur_seg_buf = new_buf;
+               rmpp_recv->seg_num++;
+               new_buf = get_next_seg(rmpp_list, new_buf);
+       }
+}
+
+static inline int get_mad_len(struct mad_rmpp_recv *rmpp_recv)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+       int hdr_size, data_size, pad;
+
+       rmpp_mad = (struct ib_rmpp_mad *)rmpp_recv->cur_seg_buf->mad;
+
+       hdr_size = data_offset(rmpp_mad->mad_hdr.mgmt_class);
+       data_size = sizeof(struct ib_rmpp_mad) - hdr_size;
+       pad = data_size - be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
+       if (pad > data_size || pad < 0)
+               pad = 0;
+
+       return hdr_size + rmpp_recv->seg_num * data_size - pad;
+}
+
+static struct ib_mad_recv_wc * complete_rmpp(struct mad_rmpp_recv *rmpp_recv)
+{
+       struct ib_mad_recv_wc *rmpp_wc;
+
+       ack_recv(rmpp_recv, rmpp_recv->rmpp_wc);
+       if (rmpp_recv->seg_num > 1)
+               cancel_delayed_work(&rmpp_recv->timeout_work);
+
+       rmpp_wc = rmpp_recv->rmpp_wc;
+       rmpp_wc->mad_len = get_mad_len(rmpp_recv);
+       /* 10 seconds until we can find the packet lifetime */
+       queue_delayed_work(rmpp_recv->agent->qp_info->port_priv->wq,
+                          &rmpp_recv->cleanup_work, msecs_to_jiffies(10000));
+       return rmpp_wc;
+}
+
+void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc, void *buf)
+{
+       struct ib_mad_recv_buf *seg_buf;
+       struct ib_rmpp_mad *rmpp_mad;
+       void *data;
+       int size, len, offset;
+       u8 flags;
+
+       len = mad_recv_wc->mad_len;
+       if (len <= sizeof(struct ib_mad)) {
+               memcpy(buf, mad_recv_wc->recv_buf.mad, len);
+               return;
+       }
+
+       offset = data_offset(mad_recv_wc->recv_buf.mad->mad_hdr.mgmt_class);
+
+       list_for_each_entry(seg_buf, &mad_recv_wc->rmpp_list, list) {
+               rmpp_mad = (struct ib_rmpp_mad *)seg_buf->mad;
+               flags = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr);
+
+               if (flags & IB_MGMT_RMPP_FLAG_FIRST) {
+                       data = rmpp_mad;
+                       size = sizeof(*rmpp_mad);
+               } else {
+                       data = (void *) rmpp_mad + offset;
+                       if (flags & IB_MGMT_RMPP_FLAG_LAST)
+                               size = len;
+                       else
+                               size = sizeof(*rmpp_mad) - offset;
+               }
+
+               memcpy(buf, data, size);
+               len -= size;
+               buf += size;
+       }
+}
+EXPORT_SYMBOL(ib_coalesce_recv_mad);
+
+static struct ib_mad_recv_wc *
+continue_rmpp(struct ib_mad_agent_private *agent,
+             struct ib_mad_recv_wc *mad_recv_wc)
+{
+       struct mad_rmpp_recv *rmpp_recv;
+       struct ib_mad_recv_buf *prev_buf;
+       struct ib_mad_recv_wc *done_wc;
+       int seg_num;
+       unsigned long flags;
+
+       rmpp_recv = acquire_rmpp_recv(agent, mad_recv_wc);
+       if (!rmpp_recv)
+               goto drop1;
+
+       seg_num = get_seg_num(&mad_recv_wc->recv_buf);
+
+       spin_lock_irqsave(&rmpp_recv->lock, flags);
+       if ((rmpp_recv->state == RMPP_STATE_TIMEOUT) ||
+           (seg_num > rmpp_recv->newwin))
+               goto drop3;
+
+       if ((seg_num <= rmpp_recv->last_ack) ||
+           (rmpp_recv->state == RMPP_STATE_COMPLETE)) {
+               spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+               ack_recv(rmpp_recv, mad_recv_wc);
+               goto drop2;
+       }
+
+       prev_buf = find_seg_location(&rmpp_recv->rmpp_wc->rmpp_list, seg_num);
+       if (!prev_buf)
+               goto drop3;
+
+       done_wc = NULL;
+       list_add(&mad_recv_wc->recv_buf.list, &prev_buf->list);
+       if (rmpp_recv->cur_seg_buf == prev_buf) {
+               update_seg_num(rmpp_recv, &mad_recv_wc->recv_buf);
+               if (get_last_flag(rmpp_recv->cur_seg_buf)) {
+                       rmpp_recv->state = RMPP_STATE_COMPLETE;
+                       spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+                       done_wc = complete_rmpp(rmpp_recv);
+                       goto out;
+               } else if (rmpp_recv->seg_num == rmpp_recv->newwin) {
+                       rmpp_recv->newwin += window_size(agent);
+                       spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+                       ack_recv(rmpp_recv, mad_recv_wc);
+                       goto out;
+               }
+       }
+       spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+out:
+       deref_rmpp_recv(rmpp_recv);
+       return done_wc;
+
+drop3: spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+drop2: deref_rmpp_recv(rmpp_recv);
+drop1: ib_free_recv_mad(mad_recv_wc);
+       return NULL;
+}
+
+static struct ib_mad_recv_wc *
+start_rmpp(struct ib_mad_agent_private *agent,
+          struct ib_mad_recv_wc *mad_recv_wc)
+{
+       struct mad_rmpp_recv *rmpp_recv;
+       unsigned long flags;
+
+       rmpp_recv = create_rmpp_recv(agent, mad_recv_wc);
+       if (!rmpp_recv) {
+               ib_free_recv_mad(mad_recv_wc);
+               return NULL;
+       }
+
+       spin_lock_irqsave(&agent->lock, flags);
+       if (insert_rmpp_recv(agent, rmpp_recv)) {
+               spin_unlock_irqrestore(&agent->lock, flags);
+               /* duplicate first MAD */
+               destroy_rmpp_recv(rmpp_recv);
+               return continue_rmpp(agent, mad_recv_wc);
+       }
+       atomic_inc(&rmpp_recv->refcount);
+
+       if (get_last_flag(&mad_recv_wc->recv_buf)) {
+               rmpp_recv->state = RMPP_STATE_COMPLETE;
+               spin_unlock_irqrestore(&agent->lock, flags);
+               complete_rmpp(rmpp_recv);
+       } else {
+               spin_unlock_irqrestore(&agent->lock, flags);
+               /* 40 seconds until we can find the packet lifetimes */
+               queue_delayed_work(agent->qp_info->port_priv->wq,
+                                  &rmpp_recv->timeout_work,
+                                  msecs_to_jiffies(40000));
+               rmpp_recv->newwin += window_size(agent);
+               ack_recv(rmpp_recv, mad_recv_wc);
+               mad_recv_wc = NULL;
+       }
+       deref_rmpp_recv(rmpp_recv);
+       return mad_recv_wc;
+}
+
+static inline u64 get_seg_addr(struct ib_mad_send_wr_private *mad_send_wr)
+{
+       return mad_send_wr->sg_list[0].addr + mad_send_wr->data_offset +
+              (sizeof(struct ib_rmpp_mad) - mad_send_wr->data_offset) *
+              (mad_send_wr->seg_num - 1);
+}
+
+static int send_next_seg(struct ib_mad_send_wr_private *mad_send_wr)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+       int timeout;
+
+       rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+       ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
+       rmpp_mad->rmpp_hdr.seg_num = cpu_to_be32(mad_send_wr->seg_num);
+
+       if (mad_send_wr->seg_num == 1) {
+               rmpp_mad->rmpp_hdr.rmpp_rtime_flags |= IB_MGMT_RMPP_FLAG_FIRST;
+               rmpp_mad->rmpp_hdr.paylen_newwin =
+                       cpu_to_be32(mad_send_wr->total_seg *
+                                   (sizeof(struct ib_rmpp_mad) -
+                                      offsetof(struct ib_rmpp_mad, data)));
+               mad_send_wr->sg_list[0].length = sizeof(struct ib_rmpp_mad);
+       } else {
+               mad_send_wr->send_wr.num_sge = 2;
+               mad_send_wr->sg_list[0].length = mad_send_wr->data_offset;
+               mad_send_wr->sg_list[1].addr = get_seg_addr(mad_send_wr);
+               mad_send_wr->sg_list[1].length = sizeof(struct ib_rmpp_mad) -
+                                                mad_send_wr->data_offset;
+               mad_send_wr->sg_list[1].lkey = mad_send_wr->sg_list[0].lkey;
+       }
+
+       if (mad_send_wr->seg_num == mad_send_wr->total_seg) {
+               rmpp_mad->rmpp_hdr.rmpp_rtime_flags |= IB_MGMT_RMPP_FLAG_LAST;
+               rmpp_mad->rmpp_hdr.paylen_newwin =
+                       cpu_to_be32(sizeof(struct ib_rmpp_mad) -
+                                   offsetof(struct ib_rmpp_mad, data) -
+                                   mad_send_wr->pad);
+       }
+
+       /* 2 seconds for an ACK until we can find the packet lifetime */
+       timeout = mad_send_wr->send_wr.wr.ud.timeout_ms;
+       if (!timeout || timeout > 2000)
+               mad_send_wr->timeout = msecs_to_jiffies(2000);
+       mad_send_wr->seg_num++;
+       return ib_send_mad(mad_send_wr);
+}
+
+static void process_rmpp_ack(struct ib_mad_agent_private *agent,
+                            struct ib_mad_recv_wc *mad_recv_wc)
+{
+       struct ib_mad_send_wr_private *mad_send_wr;
+       struct ib_rmpp_mad *rmpp_mad;
+       unsigned long flags;
+       int seg_num, newwin, ret;
+
+       rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
+       if (rmpp_mad->rmpp_hdr.rmpp_status)
+               return;
+
+       seg_num = be32_to_cpu(rmpp_mad->rmpp_hdr.seg_num);
+       newwin = be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
+
+       spin_lock_irqsave(&agent->lock, flags);
+       mad_send_wr = ib_find_send_mad(agent, rmpp_mad->mad_hdr.tid);
+       if (!mad_send_wr)
+               goto out;       /* Unmatched ACK */
+
+       if ((mad_send_wr->last_ack == mad_send_wr->total_seg) ||
+           (!mad_send_wr->timeout) || (mad_send_wr->status != IB_WC_SUCCESS))
+               goto out;       /* Send is already done */
+
+       if (seg_num > mad_send_wr->total_seg)
+               goto out;       /* Bad ACK */
+
+       if (newwin < mad_send_wr->newwin || seg_num < mad_send_wr->last_ack)
+               goto out;       /* Old ACK */
+
+       if (seg_num > mad_send_wr->last_ack) {
+               mad_send_wr->last_ack = seg_num;
+               mad_send_wr->retries = mad_send_wr->send_wr.wr.ud.retries;
+       }
+       mad_send_wr->newwin = newwin;
+       if (mad_send_wr->last_ack == mad_send_wr->total_seg) {
+               /* If no response is expected, the ACK completes the send */
+               if (!mad_send_wr->send_wr.wr.ud.timeout_ms) {
+                       struct ib_mad_send_wc wc;
+
+                       ib_mark_mad_done(mad_send_wr);
+                       spin_unlock_irqrestore(&agent->lock, flags);
+
+                       wc.status = IB_WC_SUCCESS;
+                       wc.vendor_err = 0;
+                       wc.wr_id = mad_send_wr->wr_id;
+                       ib_mad_complete_send_wr(mad_send_wr, &wc);
+                       return;
+               }
+               if (mad_send_wr->refcount == 1)
+                       ib_reset_mad_timeout(mad_send_wr, mad_send_wr->
+                                            send_wr.wr.ud.timeout_ms);
+       } else if (mad_send_wr->refcount == 1 &&
+                  mad_send_wr->seg_num < mad_send_wr->newwin &&
+                  mad_send_wr->seg_num <= mad_send_wr->total_seg) {
+               /* Send failure will just result in a timeout/retry */
+               ret = send_next_seg(mad_send_wr);
+               if (ret)
+                       goto out;
+
+               mad_send_wr->refcount++;
+               list_del(&mad_send_wr->agent_list);
+               list_add_tail(&mad_send_wr->agent_list,
+                             &mad_send_wr->mad_agent_priv->send_list);
+       }
+out:
+       spin_unlock_irqrestore(&agent->lock, flags);
+}
+
+struct ib_mad_recv_wc *
+ib_process_rmpp_recv_wc(struct ib_mad_agent_private *agent,
+                       struct ib_mad_recv_wc *mad_recv_wc)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+
+       rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
+       if (!(rmpp_mad->rmpp_hdr.rmpp_rtime_flags & IB_MGMT_RMPP_FLAG_ACTIVE))
+               return mad_recv_wc;
+
+       if (rmpp_mad->rmpp_hdr.rmpp_version != IB_MGMT_RMPP_VERSION)
+               goto out;
+
+       switch (rmpp_mad->rmpp_hdr.rmpp_type) {
+       case IB_MGMT_RMPP_TYPE_DATA:
+               if (rmpp_mad->rmpp_hdr.seg_num == __constant_htonl(1))
+                       return start_rmpp(agent, mad_recv_wc);
+               else
+                       return continue_rmpp(agent, mad_recv_wc);
+       case IB_MGMT_RMPP_TYPE_ACK:
+               process_rmpp_ack(agent, mad_recv_wc);
+               break;
+       case IB_MGMT_RMPP_TYPE_STOP:
+       case IB_MGMT_RMPP_TYPE_ABORT:
+               /* TODO: process_rmpp_nack(agent, mad_recv_wc); */
+               break;
+       default:
+               break;
+       }
+out:
+       ib_free_recv_mad(mad_recv_wc);
+       return NULL;
+}
+
+int ib_send_rmpp_mad(struct ib_mad_send_wr_private *mad_send_wr)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+       int i, total_len, ret;
+
+       rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+       if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+             IB_MGMT_RMPP_FLAG_ACTIVE))
+               return IB_RMPP_RESULT_UNHANDLED;
+
+       if (rmpp_mad->rmpp_hdr.rmpp_type != IB_MGMT_RMPP_TYPE_DATA)
+               return IB_RMPP_RESULT_INTERNAL;
+
+       if (mad_send_wr->send_wr.num_sge > 1)
+               return -EINVAL;         /* TODO: support num_sge > 1 */
+
+       mad_send_wr->seg_num = 1;
+       mad_send_wr->newwin = 1;
+       mad_send_wr->data_offset = data_offset(rmpp_mad->mad_hdr.mgmt_class);
+
+       total_len = 0;
+       for (i = 0; i < mad_send_wr->send_wr.num_sge; i++)
+               total_len += mad_send_wr->send_wr.sg_list[i].length;
+
+        mad_send_wr->total_seg = (total_len - mad_send_wr->data_offset) /
+                       (sizeof(struct ib_rmpp_mad) - mad_send_wr->data_offset);
+       mad_send_wr->pad = total_len - offsetof(struct ib_rmpp_mad, data) -
+                          be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
+
+       /* We need to wait for the final ACK even if there isn't a response */
+       mad_send_wr->refcount += (mad_send_wr->timeout == 0);
+       ret = send_next_seg(mad_send_wr);
+       if (!ret)
+               return IB_RMPP_RESULT_CONSUMED;
+       return ret;
+}
+
+int ib_process_rmpp_send_wc(struct ib_mad_send_wr_private *mad_send_wr,
+                           struct ib_mad_send_wc *mad_send_wc)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+       struct ib_mad_send_buf *msg;
+       int ret;
+
+       rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+       if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+             IB_MGMT_RMPP_FLAG_ACTIVE))
+               return IB_RMPP_RESULT_UNHANDLED; /* RMPP not active */
+
+       if (rmpp_mad->rmpp_hdr.rmpp_type != IB_MGMT_RMPP_TYPE_DATA) {
+               msg = (struct ib_mad_send_buf *) (unsigned long)
+                     mad_send_wc->wr_id;
+               ib_free_send_mad(msg);
+               return IB_RMPP_RESULT_INTERNAL;  /* ACK, STOP, or ABORT */
+       }
+
+       if (mad_send_wc->status != IB_WC_SUCCESS ||
+           mad_send_wr->status != IB_WC_SUCCESS)
+               return IB_RMPP_RESULT_PROCESSED; /* Canceled or send error */
+
+       if (!mad_send_wr->timeout)
+               return IB_RMPP_RESULT_PROCESSED; /* Response received */
+
+       if (mad_send_wr->last_ack == mad_send_wr->total_seg) {
+               mad_send_wr->timeout =
+                       msecs_to_jiffies(mad_send_wr->send_wr.wr.ud.timeout_ms);
+               return IB_RMPP_RESULT_PROCESSED; /* Send done */
+       }
+
+       if (mad_send_wr->seg_num > mad_send_wr->newwin ||
+           mad_send_wr->seg_num > mad_send_wr->total_seg)
+               return IB_RMPP_RESULT_PROCESSED; /* Wait for ACK */
+
+       ret = send_next_seg(mad_send_wr);
+       if (ret) {
+               mad_send_wc->status = IB_WC_GENERAL_ERR;
+               return IB_RMPP_RESULT_PROCESSED;
+       }
+       return IB_RMPP_RESULT_CONSUMED;
+}
+
+int ib_retry_rmpp(struct ib_mad_send_wr_private *mad_send_wr)
+{
+       struct ib_rmpp_mad *rmpp_mad;
+       int ret;
+
+       rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+       if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+             IB_MGMT_RMPP_FLAG_ACTIVE))
+               return IB_RMPP_RESULT_UNHANDLED; /* RMPP not active */
+
+       if (mad_send_wr->last_ack == mad_send_wr->total_seg)
+               return IB_RMPP_RESULT_PROCESSED;
+
+       mad_send_wr->seg_num = mad_send_wr->last_ack + 1;
+       ret = send_next_seg(mad_send_wr);
+       if (ret)
+               return IB_RMPP_RESULT_PROCESSED;
+
+       return IB_RMPP_RESULT_CONSUMED;
+}
diff --git a/drivers/infiniband/core/mad_rmpp.h b/drivers/infiniband/core/mad_rmpp.h
new file mode 100644 (file)
index 0000000..c4924df
--- /dev/null
@@ -0,0 +1,58 @@
+/*
+ * Copyright (c) 2005 Intel Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * $Id: mad_rmpp.h 1921 2005-02-25 22:58:44Z sean.hefty $
+ */
+
+#ifndef __MAD_RMPP_H__
+#define __MAD_RMPP_H__
+
+enum {
+       IB_RMPP_RESULT_PROCESSED,
+       IB_RMPP_RESULT_CONSUMED,
+       IB_RMPP_RESULT_INTERNAL,
+       IB_RMPP_RESULT_UNHANDLED
+};
+
+int ib_send_rmpp_mad(struct ib_mad_send_wr_private *mad_send_wr);
+
+struct ib_mad_recv_wc *
+ib_process_rmpp_recv_wc(struct ib_mad_agent_private *agent,
+                       struct ib_mad_recv_wc *mad_recv_wc);
+
+int ib_process_rmpp_send_wc(struct ib_mad_send_wr_private *mad_send_wr,
+                           struct ib_mad_send_wc *mad_send_wc);
+
+void ib_cancel_rmpp_recvs(struct ib_mad_agent_private *agent);
+
+int ib_retry_rmpp(struct ib_mad_send_wr_private *mad_send_wr);
+
+#endif /* __MAD_RMPP_H__ */