Roland, Attached are 6 git patches pulled from SVN to queue for 2.6.19. They're from SVN versions:
4578 - include atomic as default QP attribute 8628 - fix reject message if GID is invalid 8434 - add dual-sided RMPP 8826 - remove unnecessary include 8827 - remove unnecessary include 9088 - randomize starting local comm id Let me know if you'd prefer these in another format (such as inline). - Sean
>From d697059a6f69e19c18a50c87df20894d253d3d8f Mon Sep 17 00:00:00 2001 From: Sean Hefty <[EMAIL PROTECTED]> Date: Mon, 28 Aug 2006 15:15:18 -0700 Subject: [PATCH] Randomize the starting local comm ID to avoid getting a rejected connection due to a stale connection after a system reboot or reloading of the ib_cm. Signed-off-by: Sean Hefty <[EMAIL PROTECTED]> --- drivers/infiniband/core/cm.c | 24 ++++++++++++++++-------- 1 files changed, 16 insertions(+), 8 deletions(-) diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c index 1aad33e..47fd03c 100644 --- a/drivers/infiniband/core/cm.c +++ b/drivers/infiniband/core/cm.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2004, 2005 Intel Corporation. All rights reserved. + * Copyright (c) 2004-2006 Intel Corporation. All rights reserved. * Copyright (c) 2004 Topspin Corporation. All rights reserved. * Copyright (c) 2004, 2005 Voltaire Corporation. All rights reserved. * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. @@ -41,6 +41,7 @@ #include <linux/err.h> #include <linux/idr.h> #include <linux/interrupt.h> #include <linux/pci.h> +#include <linux/random.h> #include <linux/rbtree.h> #include <linux/spinlock.h> #include <linux/workqueue.h> @@ -73,6 +74,7 @@ static struct ib_cm { struct rb_root remote_id_table; struct rb_root remote_sidr_table; struct idr local_id_table; + __be32 random_id_operand; struct workqueue_struct *wq; } cm; @@ -299,15 +301,17 @@ static int cm_init_av_by_path(struct ib_ static int cm_alloc_id(struct cm_id_private *cm_id_priv) { unsigned long flags; - int ret; + int ret, id; static int next_id; do { spin_lock_irqsave(&cm.lock, flags); - ret = idr_get_new_above(&cm.local_id_table, cm_id_priv, next_id++, - (__force int *) &cm_id_priv->id.local_id); + ret = idr_get_new_above(&cm.local_id_table, cm_id_priv, + next_id++, &id); spin_unlock_irqrestore(&cm.lock, flags); } while( (ret == -EAGAIN) && idr_pre_get(&cm.local_id_table, GFP_KERNEL) ); + + cm_id_priv->id.local_id = (__force __be32) (id ^ cm.random_id_operand); return ret; } @@ -316,7 +320,8 @@ static void cm_free_id(__be32 local_id) unsigned long flags; spin_lock_irqsave(&cm.lock, flags); - idr_remove(&cm.local_id_table, (__force int) local_id); + idr_remove(&cm.local_id_table, + (__force int) (local_id ^ cm.random_id_operand)); spin_unlock_irqrestore(&cm.lock, flags); } @@ -324,7 +329,8 @@ static struct cm_id_private * cm_get_id( { struct cm_id_private *cm_id_priv; - cm_id_priv = idr_find(&cm.local_id_table, (__force int) local_id); + cm_id_priv = idr_find(&cm.local_id_table, + (__force int) (local_id ^ cm.random_id_operand)); if (cm_id_priv) { if (cm_id_priv->id.remote_id == remote_id) atomic_inc(&cm_id_priv->refcount); @@ -2082,8 +2088,9 @@ static struct cm_id_private * cm_acquire spin_unlock_irqrestore(&cm.lock, flags); return NULL; } - cm_id_priv = idr_find(&cm.local_id_table, - (__force int) timewait_info->work.local_id); + cm_id_priv = idr_find(&cm.local_id_table, (__force int) + (timewait_info->work.local_id ^ + cm.random_id_operand)); if (cm_id_priv) { if (cm_id_priv->id.remote_id == remote_id) atomic_inc(&cm_id_priv->refcount); @@ -3360,6 +3367,7 @@ static int __init ib_cm_init(void) cm.remote_qp_table = RB_ROOT; cm.remote_sidr_table = RB_ROOT; idr_init(&cm.local_id_table); + get_random_bytes(&cm.random_id_operand, sizeof cm.random_id_operand); idr_pre_get(&cm.local_id_table, GFP_KERNEL); cm.wq = create_workqueue("ib_cm"); -- 1.4.2
>From 5cabb459ff4a8f22ab6b44e0704ee8b2e28e4104 Mon Sep 17 00:00:00 2001 From: Sean Hefty <[EMAIL PROTECTED]> Date: Mon, 28 Aug 2006 11:55:52 -0700 Subject: [PATCH] Enable atomic operations along with RDMA reads if a local RDMA read/atomic depth is provided by the user. Signed-off-by: Sean Hefty <[EMAIL PROTECTED]> --- drivers/infiniband/core/cm.c | 3 ++- 1 files changed, 2 insertions(+), 1 deletions(-) diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c index 0de335b..0df1454 100644 --- a/drivers/infiniband/core/cm.c +++ b/drivers/infiniband/core/cm.c @@ -3125,7 +3125,8 @@ static int cm_init_qp_init_attr(struct c qp_attr->qp_access_flags = IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE; if (cm_id_priv->responder_resources) - qp_attr->qp_access_flags |= IB_ACCESS_REMOTE_READ; + qp_attr->qp_access_flags |= IB_ACCESS_REMOTE_READ | + IB_ACCESS_REMOTE_ATOMIC; qp_attr->pkey_index = cm_id_priv->av.pkey_index; qp_attr->port_num = cm_id_priv->av.port->port_num; ret = 0; -- 1.4.2
>From 03138e447d67e0c154823cfd67cad7ed85481ff8 Mon Sep 17 00:00:00 2001 From: Sean Hefty <[EMAIL PROTECTED]> Date: Mon, 28 Aug 2006 11:57:42 -0700 Subject: [PATCH] Set the reject code properly when rejecting a request that contains an invalid GID. A suitable GID is returned by the IB CM in the additional reject information (ARI). This is a spec compliancy issue. Signed-off-by: Sean Hefty <[EMAIL PROTECTED]> --- drivers/infiniband/core/cm.c | 32 +++++++++++++++++++++----------- 1 files changed, 21 insertions(+), 11 deletions(-) diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c index 0df1454..1aad33e 100644 --- a/drivers/infiniband/core/cm.c +++ b/drivers/infiniband/core/cm.c @@ -1354,7 +1354,7 @@ static int cm_req_handler(struct cm_work id.local_id); if (IS_ERR(cm_id_priv->timewait_info)) { ret = PTR_ERR(cm_id_priv->timewait_info); - goto error1; + goto destroy; } cm_id_priv->timewait_info->work.remote_id = req_msg->local_comm_id; cm_id_priv->timewait_info->remote_ca_guid = req_msg->local_ca_guid; @@ -1363,7 +1363,8 @@ static int cm_req_handler(struct cm_work listen_cm_id_priv = cm_match_req(work, cm_id_priv); if (!listen_cm_id_priv) { ret = -EINVAL; - goto error2; + kfree(cm_id_priv->timewait_info); + goto destroy; } cm_id_priv->id.cm_handler = listen_cm_id_priv->id.cm_handler; @@ -1373,12 +1374,22 @@ static int cm_req_handler(struct cm_work cm_format_paths_from_req(req_msg, &work->path[0], &work->path[1]); ret = cm_init_av_by_path(&work->path[0], &cm_id_priv->av); - if (ret) - goto error3; + if (ret) { + ib_get_cached_gid(work->port->cm_dev->device, + work->port->port_num, 0, &work->path[0].sgid); + ib_send_cm_rej(cm_id, IB_CM_REJ_INVALID_GID, + &work->path[0].sgid, sizeof work->path[0].sgid, + NULL, 0); + goto rejected; + } if (req_msg->alt_local_lid) { ret = cm_init_av_by_path(&work->path[1], &cm_id_priv->alt_av); - if (ret) - goto error3; + if (ret) { + ib_send_cm_rej(cm_id, IB_CM_REJ_INVALID_ALT_GID, + &work->path[0].sgid, + sizeof work->path[0].sgid, NULL, 0); + goto rejected; + } } cm_id_priv->tid = req_msg->hdr.tid; cm_id_priv->timeout_ms = cm_convert_to_ms( @@ -1400,12 +1411,11 @@ static int cm_req_handler(struct cm_work cm_deref_id(listen_cm_id_priv); return 0; -error3: atomic_dec(&cm_id_priv->refcount); +rejected: + atomic_dec(&cm_id_priv->refcount); cm_deref_id(listen_cm_id_priv); - cm_cleanup_timewait(cm_id_priv->timewait_info); -error2: kfree(cm_id_priv->timewait_info); - cm_id_priv->timewait_info = NULL; -error1: ib_destroy_cm_id(&cm_id_priv->id); +destroy: + ib_destroy_cm_id(cm_id); return ret; } -- 1.4.2
>From 69fd025f17b20b2fa3de8bf435f4601f9feeb0fa Mon Sep 17 00:00:00 2001 From: Sean Hefty <[EMAIL PROTECTED]> Date: Mon, 28 Aug 2006 15:10:32 -0700 Subject: [PATCH] Add support for dual-sided RMPP transfers. The implementation assumes that any RMPP request that requires a response uses DS RMPP. Based on the RMPP start-up scenarios defined by the spec, this should be a valid assumption. That is, there is no start-up scenario defined where an RMPP request is followed by a non-RMPP response. By having this assumption we avoid any API changes. In order for a node that supports DS RMPP to communicate with one that does not, RMPP responses assume a new window size of 1 if a DS ACK has not been received. (By DS ACK, I'm referring to the turn-around ACK after the final ACK of the request.) This is a slight spec deviation, but is necessary to allow communication with nodes that do not generate the DS ACK. It also handles the case when a response is sent after the request state has been discarded. Signed-off-by: Sean Hefty <[EMAIL PROTECTED]> --- drivers/infiniband/core/mad_rmpp.c | 90 +++++++++++++++++++++++++++++++++++- 1 files changed, 87 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/core/mad_rmpp.c b/drivers/infiniband/core/mad_rmpp.c index ebcd5b1..26469c1 100644 --- a/drivers/infiniband/core/mad_rmpp.c +++ b/drivers/infiniband/core/mad_rmpp.c @@ -60,6 +60,7 @@ struct mad_rmpp_recv { int last_ack; int seg_num; int newwin; + int repwin; __be64 tid; u32 src_qp; @@ -170,6 +171,32 @@ static struct ib_mad_send_buf *alloc_res return msg; } +static void ack_ds_ack(struct ib_mad_agent_private *agent, + struct ib_mad_recv_wc *recv_wc) +{ + struct ib_mad_send_buf *msg; + struct ib_rmpp_mad *rmpp_mad; + int ret; + + msg = alloc_response_msg(&agent->agent, recv_wc); + if (IS_ERR(msg)) + return; + + rmpp_mad = msg->mad; + memcpy(rmpp_mad, recv_wc->recv_buf.mad, msg->hdr_len); + + rmpp_mad->mad_hdr.method ^= IB_MGMT_METHOD_RESP; + ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE); + rmpp_mad->rmpp_hdr.seg_num = 0; + rmpp_mad->rmpp_hdr.paylen_newwin = cpu_to_be32(1); + + ret = ib_post_send_mad(msg, NULL); + if (ret) { + ib_destroy_ah(msg->ah); + ib_free_send_mad(msg); + } +} + void ib_rmpp_send_handler(struct ib_mad_send_wc *mad_send_wc) { struct ib_rmpp_mad *rmpp_mad = mad_send_wc->send_buf->mad; @@ -271,6 +298,7 @@ create_rmpp_recv(struct ib_mad_agent_pri rmpp_recv->newwin = 1; rmpp_recv->seg_num = 1; rmpp_recv->last_ack = 0; + rmpp_recv->repwin = 1; mad_hdr = &mad_recv_wc->recv_buf.mad->mad_hdr; rmpp_recv->tid = mad_hdr->tid; @@ -591,6 +619,16 @@ static inline void adjust_last_ack(struc break; } +static void process_ds_ack(struct ib_mad_agent_private *agent, + struct ib_mad_recv_wc *mad_recv_wc, int newwin) +{ + struct mad_rmpp_recv *rmpp_recv; + + rmpp_recv = find_rmpp_recv(agent, mad_recv_wc); + if (rmpp_recv && rmpp_recv->state == RMPP_STATE_COMPLETE) + rmpp_recv->repwin = newwin; +} + static void process_rmpp_ack(struct ib_mad_agent_private *agent, struct ib_mad_recv_wc *mad_recv_wc) { @@ -616,8 +654,18 @@ static void process_rmpp_ack(struct ib_m spin_lock_irqsave(&agent->lock, flags); mad_send_wr = ib_find_send_mad(agent, mad_recv_wc); - if (!mad_send_wr) - goto out; /* Unmatched ACK */ + if (!mad_send_wr) { + if (!seg_num) + process_ds_ack(agent, mad_recv_wc, newwin); + goto out; /* Unmatched or DS RMPP ACK */ + } + + if ((mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count) && + (mad_send_wr->timeout)) { + spin_unlock_irqrestore(&agent->lock, flags); + ack_ds_ack(agent, mad_recv_wc); + return; /* Repeated ACK for DS RMPP transaction */ + } if ((mad_send_wr->last_ack == mad_send_wr->send_buf.seg_count) || (!mad_send_wr->timeout) || (mad_send_wr->status != IB_WC_SUCCESS)) @@ -656,6 +704,9 @@ static void process_rmpp_ack(struct ib_m if (mad_send_wr->refcount == 1) ib_reset_mad_timeout(mad_send_wr, mad_send_wr->send_buf.timeout_ms); + spin_unlock_irqrestore(&agent->lock, flags); + ack_ds_ack(agent, mad_recv_wc); + return; } else if (mad_send_wr->refcount == 1 && mad_send_wr->seg_num < mad_send_wr->newwin && mad_send_wr->seg_num < mad_send_wr->send_buf.seg_count) { @@ -772,6 +823,39 @@ out: return NULL; } +static int init_newwin(struct ib_mad_send_wr_private *mad_send_wr) +{ + struct ib_mad_agent_private *agent = mad_send_wr->mad_agent_priv; + struct ib_mad_hdr *mad_hdr = mad_send_wr->send_buf.mad; + struct mad_rmpp_recv *rmpp_recv; + struct ib_ah_attr ah_attr; + unsigned long flags; + int newwin = 1; + + if (!(mad_hdr->method & IB_MGMT_METHOD_RESP)) + goto out; + + spin_lock_irqsave(&agent->lock, flags); + list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) { + if (rmpp_recv->tid != mad_hdr->tid || + rmpp_recv->mgmt_class != mad_hdr->mgmt_class || + rmpp_recv->class_version != mad_hdr->class_version || + (rmpp_recv->method & IB_MGMT_METHOD_RESP)) + continue; + + if (ib_query_ah(mad_send_wr->send_buf.ah, &ah_attr)) + continue; + + if (rmpp_recv->slid == ah_attr.dlid) { + newwin = rmpp_recv->repwin; + break; + } + } + spin_unlock_irqrestore(&agent->lock, flags); +out: + return newwin; +} + int ib_send_rmpp_mad(struct ib_mad_send_wr_private *mad_send_wr) { struct ib_rmpp_mad *rmpp_mad; @@ -787,7 +871,7 @@ int ib_send_rmpp_mad(struct ib_mad_send_ return IB_RMPP_RESULT_INTERNAL; } - mad_send_wr->newwin = 1; + mad_send_wr->newwin = init_newwin(mad_send_wr); /* We need to wait for the final ACK even if there isn't a response */ mad_send_wr->refcount += (mad_send_wr->timeout == 0); -- 1.4.2
>From e13596c5861d2e8371bfbb4f8dc980134234f26d Mon Sep 17 00:00:00 2001 From: Sean Hefty <[EMAIL PROTECTED]> Date: Mon, 28 Aug 2006 15:12:04 -0700 Subject: [PATCH] The ib_mad module does not use a kthread function, but mad_priv.h includes kthread.h. Remove the include. Signed-off-by: James Lentini <[EMAIL PROTECTED]> Signed-off-by: Sean Hefty <[EMAIL PROTECTED]> --- drivers/infiniband/core/mad_priv.h | 1 - 1 files changed, 0 insertions(+), 1 deletions(-) diff --git a/drivers/infiniband/core/mad_priv.h b/drivers/infiniband/core/mad_priv.h index d147f3b..1da9adb 100644 --- a/drivers/infiniband/core/mad_priv.h +++ b/drivers/infiniband/core/mad_priv.h @@ -39,7 +39,6 @@ #define __IB_MAD_PRIV_H__ #include <linux/completion.h> #include <linux/pci.h> -#include <linux/kthread.h> #include <linux/workqueue.h> #include <rdma/ib_mad.h> #include <rdma/ib_smi.h> -- 1.4.2
>From 526b67f066a3afafc6459217d8e5820641d3af26 Mon Sep 17 00:00:00 2001 From: Sean Hefty <[EMAIL PROTECTED]> Date: Mon, 28 Aug 2006 15:13:07 -0700 Subject: [PATCH] Remove unnecessary include. Signed-off-by: James Lentini <[EMAIL PROTECTED]> Signed-off-by: Sean Hefty <[EMAIL PROTECTED]> --- drivers/infiniband/core/mad_rmpp.c | 2 -- 1 files changed, 0 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/core/mad_rmpp.c b/drivers/infiniband/core/mad_rmpp.c index 26469c1..f2becdf 100644 --- a/drivers/infiniband/core/mad_rmpp.c +++ b/drivers/infiniband/core/mad_rmpp.c @@ -33,8 +33,6 @@ * $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" -- 1.4.2
_______________________________________________ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general