RDMA: merge iWARP support
cover RDMA: merge iWARP support
Here is a series of patches that adds iWARP (RDMA over IP) support to the InfiniBand support already in the kernel. Since the iWARP RDMA model is quite close to the InfiniBand model, the changes are not that large. The biggest difference is in how connections are established, since iWARP connections are TCP connections, while IB uses a different (native IB) mechanism for establishing a connection.
The first patch in the series adds an iWARP connection manager, which handles establishing and tearing down connections for iWARP devices. The second patch is all the small changes required to hook in the connection manager and make the rest of the IB stuff also work with iWARP devices. The third patch (compressed due to its size) adds the first driver for an iWARP device, the Ammasso 1100 1 Gb/sec RNIC.
My current plan is to merge this stuff for 2.6.19. Please let me know if you see anything (major or minor) that needs to be fixed up.
Thanks,
Roland
PATCH 1/2 RDMA: iWARP connection manager
From: Tom Tucker <[email protected]>
Add an iWARP Connection Manager (CM), which abstracts connection management for iWARP devices (RNICs). It is a logical instance of the xx_cm where xx is the transport type (ib or iw). The symbols exported are used by the transport independent rdma_cm module, and are available also for transport dependent ULPs.
--- drivers/infiniband/core/iwcm.c | 1019 ++++++++++++++++++++++++++++++++++++++++ drivers/infiniband/core/iwcm.h | 62 ++ include/rdma/iw_cm.h | 258 ++++++++++ 3 files changed, 1339 insertions(+), 0 deletions(-)
drivers/infiniband/core/iwcm.c new file +1019 -0
@@ -0,0 +1,1019 @@ +/* +* Copyright (c) 2004, 2005 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. +* Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. +* Copyright (c) 2005 Network Appliance, 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. +* +*/ +#include <linux/dma-mapping.h> +#include <linux/err.h> +#include <linux/idr.h> +#include <linux/interrupt.h> +#include <linux/pci.h> +#include <linux/rbtree.h> +#include <linux/spinlock.h> +#include <linux/workqueue.h> +#include <linux/completion.h> + +#include <rdma/iw_cm.h> +#include <rdma/ib_addr.h> + +#include "iwcm.h" + +MODULE_AUTHOR ("Tom Tucker" ); +MODULE_DESCRIPTION ("iWARP CM" ); +MODULE_LICENSE ("Dual BSD/GPL" ); + +static struct workqueue_struct *iwcm_wq; +struct iwcm_work { +struct work_struct work; +struct iwcm_id_private *cm_id; +struct list_head list; +struct iw_cm_event event; +struct list_head free_list; +}; + +/* +* The following services provide a mechanism for pre-allocating iwcm_work +* elements. The design pre-allocates them based on the cm_id type: +* LISTENING IDS: Get enough elements preallocated to handle the +* listen backlog. +* ACTIVE IDS: 4: CONNECT_REPLY, ESTABLISHED, DISCONNECT, CLOSE +* PASSIVE IDS: 3: ESTABLISHED, DISCONNECT, CLOSE +* +* Allocating them in connect and listen avoids having to deal +* with allocation failures on the event upcall from the provider (which +* is called in the interrupt context). +* +* One exception is when creating the cm_id for incoming connection requests. +* There are two cases: +* 1) in the event upcall, cm_event_handler(), for a listening cm_id. If +* the backlog is exceeded, then no more connection request events will +* be processed. cm_event_handler() returns -ENOMEM in this case. Its up +* to the provider to reject the connectino request. +* 2) in the connection request workqueue handler, cm_conn_req_handler(). +* If work elements cannot be allocated for the new connect request cm_id, +* then IWCM will call the provider reject method. This is ok since +* cm_conn_req_handler() runs in the workqueue thread context. +*/ + +static struct iwcm_work *get_work (struct iwcm_id_private *cm_id_priv) +{ +struct iwcm_work *work; + +if (list_empty (&cm_id_priv->work_free_list)) +return NULL ; + work = list_entry(cm_id_priv->work_free_list.next, struct iwcm_work, + free_list); +list_del_init (&work->free_list); +return work; +} + +static void put_work (struct iwcm_work *work) +{ +list_add (&work->free_list, &work->cm_id->work_free_list); +} + +static void dealloc_work_entries (struct iwcm_id_private *cm_id_priv) +{ +struct list_head *e, *tmp; + +list_for_each_safe (e, tmp, &cm_id_priv->work_free_list) + kfree(list_entry(e, struct iwcm_work, free_list)); +} + +static int alloc_work_entries (struct iwcm_id_private *cm_id_priv,int count) +{ +struct iwcm_work *work; + +BUG_ON (!list_empty (&cm_id_priv->work_free_list)); +while (count--) { + work =kmalloc (sizeof (struct iwcm_work ),GFP_KERNEL ); +if (!work) { +dealloc_work_entries (cm_id_priv); +return -ENOMEM ; + } + work->cm_id = cm_id_priv; +INIT_LIST_HEAD (&work->list); +put_work (work); + } +return 0 ; +} + +/* +* Save private data from incoming connection requests in the +* cm_id_priv so the low level driver doesn't have to. Adjust +* the event ptr to point to the local copy. +*/ +static int copy_private_data (struct iwcm_id_private *cm_id_priv, +struct iw_cm_event *event) +{ +void *p; + + p =kmalloc (event->private_data_len,GFP_ATOMIC ); +if (!p) +return -ENOMEM ; +memcpy (p, event->private_data, event->private_data_len); + event->private_data = p; +return 0 ; +} + +/* +* Release a reference on cm_id. If the last reference is being removed +* and iw_destroy_cm_id is waiting, wake up the waiting thread. +*/ +static int iwcm_deref_id (struct iwcm_id_private *cm_id_priv) +{ +int ret =0 ; + +BUG_ON (atomic_read (&cm_id_priv->refcount)==0 ); +if (atomic_dec_and_test (&cm_id_priv->refcount)) { +BUG_ON (!list_empty (&cm_id_priv->work_list)); +if (waitqueue_active (&cm_id_priv->destroy_comp.wait)) { +BUG_ON (cm_id_priv->state !=IW_CM_STATE_DESTROYING ); +BUG_ON (test_bit (IWCM_F_CALLBACK_DESTROY , + &cm_id_priv->flags)); + ret =1 ; + } +complete (&cm_id_priv->destroy_comp); + } + +return ret; +} + +static void add_ref (struct iw_cm_id *cm_id) +{ +struct iwcm_id_private *cm_id_priv; + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +atomic_inc (&cm_id_priv->refcount); +} + +static void rem_ref (struct iw_cm_id *cm_id) +{ +struct iwcm_id_private *cm_id_priv; + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +iwcm_deref_id (cm_id_priv); +} + +static int cm_event_handler (struct iw_cm_id *cm_id,struct iw_cm_event *event); + +struct iw_cm_id *iw_create_cm_id (struct ib_device *device, +iw_cm_handler cm_handler, +void *context) +{ +struct iwcm_id_private *cm_id_priv; + + cm_id_priv =kzalloc (sizeof (*cm_id_priv),GFP_KERNEL ); +if (!cm_id_priv) +return ERR_PTR (-ENOMEM ); + + cm_id_priv->state =IW_CM_STATE_IDLE ; + cm_id_priv->id.device = device; + cm_id_priv->id.cm_handler = cm_handler; + cm_id_priv->id.context = context; + cm_id_priv->id.event_handler = cm_event_handler; + cm_id_priv->id.add_ref = add_ref; + cm_id_priv->id.rem_ref = rem_ref; +spin_lock_init (&cm_id_priv->lock); +atomic_set (&cm_id_priv->refcount,1 ); +init_waitqueue_head (&cm_id_priv->connect_wait); +init_completion (&cm_id_priv->destroy_comp); +INIT_LIST_HEAD (&cm_id_priv->work_list); +INIT_LIST_HEAD (&cm_id_priv->work_free_list); + +return &cm_id_priv->id; +} +EXPORT_SYMBOL (iw_create_cm_id); + + +static int iwcm_modify_qp_err (struct ib_qp *qp) +{ +struct ib_qp_attr qp_attr; + +if (!qp) +return -EINVAL ; + + qp_attr.qp_state =IB_QPS_ERR ; +return ib_modify_qp (qp, &qp_attr,IB_QP_STATE ); +} + +/* +* This is really the RDMAC CLOSING state. It is most similar to the +* IB SQD QP state. +*/ +static int iwcm_modify_qp_sqd (struct ib_qp *qp) +{ +struct ib_qp_attr qp_attr; + +BUG_ON (qp ==NULL ); + qp_attr.qp_state =IB_QPS_SQD ; +return ib_modify_qp (qp, &qp_attr,IB_QP_STATE ); +} + +/* +* CM_ID <-- CLOSING +* +* Block if a passive or active connection is currenlty being processed. Then +* process the event as follows: +* - If we are ESTABLISHED, move to CLOSING and modify the QP state +* based on the abrupt flag +* - If the connection is already in the CLOSING or IDLE state, the peer is +* disconnecting concurrently with us and we've already seen the +* DISCONNECT event -- ignore the request and return 0 +* - Disconnect on a listening endpoint returns -EINVAL +*/ +int iw_cm_disconnect (struct iw_cm_id *cm_id,int abrupt) +{ +struct iwcm_id_private *cm_id_priv; +unsigned long flags; +int ret =0 ; +struct ib_qp *qp =NULL ; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +/* Wait if we're currently in a connect or accept downcall */ +wait_event (cm_id_priv->connect_wait, + !test_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags)); + +spin_lock_irqsave (&cm_id_priv->lock, flags); +switch (cm_id_priv->state) { +case IW_CM_STATE_ESTABLISHED : + cm_id_priv->state =IW_CM_STATE_CLOSING ; + +/* QP could be <nul> for user-mode client */ +if (cm_id_priv->qp) + qp = cm_id_priv->qp; +else + ret = -EINVAL ; +break ; +case IW_CM_STATE_LISTEN : + ret = -EINVAL ; +break ; +case IW_CM_STATE_CLOSING : +/* remote peer closed first */ +case IW_CM_STATE_IDLE : +/* accept or connect returned !0 */ +break ; +case IW_CM_STATE_CONN_RECV : +/* +* App called disconnect before/without calling accept after +* connect_request event delivered. +*/ +break ; +case IW_CM_STATE_CONN_SENT : +/* Can only get here if wait above fails */ +default : +BUG (); + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + +if (qp) { +if (abrupt) + ret =iwcm_modify_qp_err (qp); +else + ret =iwcm_modify_qp_sqd (qp); + +/* +* If both sides are disconnecting the QP could +* already be in ERR or SQD states +*/ + ret =0 ; + } + +return ret; +} +EXPORT_SYMBOL (iw_cm_disconnect); + +/* +* CM_ID <-- DESTROYING +* +* Clean up all resources associated with the connection and release +* the initial reference taken by iw_create_cm_id. +*/ +static void destroy_cm_id (struct iw_cm_id *cm_id) +{ +struct iwcm_id_private *cm_id_priv; +unsigned long flags; +int ret; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +/* +* Wait if we're currently in a connect or accept downcall. A +* listening endpoint should never block here. +*/ +wait_event (cm_id_priv->connect_wait, + !test_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags)); + +spin_lock_irqsave (&cm_id_priv->lock, flags); +switch (cm_id_priv->state) { +case IW_CM_STATE_LISTEN : + cm_id_priv->state =IW_CM_STATE_DESTROYING ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +/* destroy the listening endpoint */ + ret = cm_id->device->iwcm->destroy_listen (cm_id); +spin_lock_irqsave (&cm_id_priv->lock, flags); +break ; +case IW_CM_STATE_ESTABLISHED : + cm_id_priv->state =IW_CM_STATE_DESTROYING ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +/* Abrupt close of the connection */ + (void )iwcm_modify_qp_err (cm_id_priv->qp); +spin_lock_irqsave (&cm_id_priv->lock, flags); +break ; +case IW_CM_STATE_IDLE : +case IW_CM_STATE_CLOSING : + cm_id_priv->state =IW_CM_STATE_DESTROYING ; +break ; +case IW_CM_STATE_CONN_RECV : +/* +* App called destroy before/without calling accept after +* receiving connection request event notification. +*/ + cm_id_priv->state =IW_CM_STATE_DESTROYING ; +break ; +case IW_CM_STATE_CONN_SENT : +case IW_CM_STATE_DESTROYING : +default : +BUG (); +break ; + } +if (cm_id_priv->qp) { + cm_id_priv->id.device->iwcm->rem_ref (cm_id_priv->qp); + cm_id_priv->qp =NULL ; + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + + (void )iwcm_deref_id (cm_id_priv); +} + +/* +* This function is only called by the application thread and cannot +* be called by the event thread. The function will wait for all +* references to be released on the cm_id and then kfree the cm_id +* object. +*/ +void iw_destroy_cm_id (struct iw_cm_id *cm_id) +{ +struct iwcm_id_private *cm_id_priv; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +BUG_ON (test_bit (IWCM_F_CALLBACK_DESTROY , &cm_id_priv->flags)); + +destroy_cm_id (cm_id); + +wait_for_completion (&cm_id_priv->destroy_comp); + +dealloc_work_entries (cm_id_priv); + +kfree (cm_id_priv); +} +EXPORT_SYMBOL (iw_destroy_cm_id); + +/* +* CM_ID <-- LISTEN +* +* Start listening for connect requests. Generates one CONNECT_REQUEST +* event for each inbound connect request. +*/ +int iw_cm_listen (struct iw_cm_id *cm_id,int backlog) +{ +struct iwcm_id_private *cm_id_priv; +unsigned long flags; +int ret =0 ; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); + + ret =alloc_work_entries (cm_id_priv, backlog); +if (ret) +return ret; + +spin_lock_irqsave (&cm_id_priv->lock, flags); +switch (cm_id_priv->state) { +case IW_CM_STATE_IDLE : + cm_id_priv->state =IW_CM_STATE_LISTEN ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + ret = cm_id->device->iwcm->create_listen (cm_id, backlog); +if (ret) + cm_id_priv->state =IW_CM_STATE_IDLE ; +spin_lock_irqsave (&cm_id_priv->lock, flags); +break ; +default : + ret = -EINVAL ; + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + +return ret; +} +EXPORT_SYMBOL (iw_cm_listen); + +/* +* CM_ID <-- IDLE +* +* Rejects an inbound connection request. No events are generated. +*/ +int iw_cm_reject (struct iw_cm_id *cm_id, +const void *private_data, +u8 private_data_len) +{ +struct iwcm_id_private *cm_id_priv; +unsigned long flags; +int ret; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +set_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); + +spin_lock_irqsave (&cm_id_priv->lock, flags); +if (cm_id_priv->state !=IW_CM_STATE_CONN_RECV ) { +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +wake_up_all (&cm_id_priv->connect_wait); +return -EINVAL ; + } + cm_id_priv->state =IW_CM_STATE_IDLE ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + + ret = cm_id->device->iwcm->reject (cm_id, private_data, + private_data_len); + +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +wake_up_all (&cm_id_priv->connect_wait); + +return ret; +} +EXPORT_SYMBOL (iw_cm_reject); + +/* +* CM_ID <-- ESTABLISHED +* +* Accepts an inbound connection request and generates an ESTABLISHED +* event. Callers of iw_cm_disconnect and iw_destroy_cm_id will block +* until the ESTABLISHED event is received from the provider. +*/ +int iw_cm_accept (struct iw_cm_id *cm_id, +struct iw_cm_conn_param *iw_param) +{ +struct iwcm_id_private *cm_id_priv; +struct ib_qp *qp; +unsigned long flags; +int ret; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +set_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); + +spin_lock_irqsave (&cm_id_priv->lock, flags); +if (cm_id_priv->state !=IW_CM_STATE_CONN_RECV ) { +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +wake_up_all (&cm_id_priv->connect_wait); +return -EINVAL ; + } +/* Get the ib_qp given the QPN */ + qp = cm_id->device->iwcm->get_qp (cm_id->device, iw_param->qpn); +if (!qp) { +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +return -EINVAL ; + } + cm_id->device->iwcm->add_ref (qp); + cm_id_priv->qp = qp; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + + ret = cm_id->device->iwcm->accept (cm_id, iw_param); +if (ret) { +/* An error on accept precludes provider events */ +BUG_ON (cm_id_priv->state !=IW_CM_STATE_CONN_RECV ); + cm_id_priv->state =IW_CM_STATE_IDLE ; +spin_lock_irqsave (&cm_id_priv->lock, flags); +if (cm_id_priv->qp) { + cm_id->device->iwcm->rem_ref (qp); + cm_id_priv->qp =NULL ; + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +wake_up_all (&cm_id_priv->connect_wait); + } + +return ret; +} +EXPORT_SYMBOL (iw_cm_accept); + +/* +* Active Side: CM_ID <-- CONN_SENT +* +* If successful, results in the generation of a CONNECT_REPLY +* event. iw_cm_disconnect and iw_cm_destroy will block until the +* CONNECT_REPLY event is received from the provider. +*/ +int iw_cm_connect (struct iw_cm_id *cm_id,struct iw_cm_conn_param *iw_param) +{ +struct iwcm_id_private *cm_id_priv; +int ret =0 ; +unsigned long flags; +struct ib_qp *qp; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); + + ret =alloc_work_entries (cm_id_priv,4 ); +if (ret) +return ret; + +set_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +spin_lock_irqsave (&cm_id_priv->lock, flags); + +if (cm_id_priv->state !=IW_CM_STATE_IDLE ) { +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +wake_up_all (&cm_id_priv->connect_wait); +return -EINVAL ; + } + +/* Get the ib_qp given the QPN */ + qp = cm_id->device->iwcm->get_qp (cm_id->device, iw_param->qpn); +if (!qp) { +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +return -EINVAL ; + } + cm_id->device->iwcm->add_ref (qp); + cm_id_priv->qp = qp; + cm_id_priv->state =IW_CM_STATE_CONN_SENT ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + + ret = cm_id->device->iwcm->connect (cm_id, iw_param); +if (ret) { +spin_lock_irqsave (&cm_id_priv->lock, flags); +if (cm_id_priv->qp) { + cm_id->device->iwcm->rem_ref (qp); + cm_id_priv->qp =NULL ; + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +BUG_ON (cm_id_priv->state !=IW_CM_STATE_CONN_SENT ); + cm_id_priv->state =IW_CM_STATE_IDLE ; +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +wake_up_all (&cm_id_priv->connect_wait); + } + +return ret; +} +EXPORT_SYMBOL (iw_cm_connect); + +/* +* Passive Side: new CM_ID <-- CONN_RECV +* +* Handles an inbound connect request. The function creates a new +* iw_cm_id to represent the new connection and inherits the client +* callback function and other attributes from the listening parent. +* +* The work item contains a pointer to the listen_cm_id and the event. The +* listen_cm_id contains the client cm_handler, context and +* device. These are copied when the device is cloned. The event +* contains the new four tuple. +* +* An error on the child should not affect the parent, so this +* function does not return a value. +*/ +static void cm_conn_req_handler (struct iwcm_id_private *listen_id_priv, +struct iw_cm_event *iw_event) +{ +unsigned long flags; +struct iw_cm_id *cm_id; +struct iwcm_id_private *cm_id_priv; +int ret; + +/* +* The provider should never generate a connection request +* event with a bad status. +*/ +BUG_ON (iw_event->status); + +/* +* We could be destroying the listening id. If so, ignore this +* upcall. +*/ +spin_lock_irqsave (&listen_id_priv->lock, flags); +if (listen_id_priv->state !=IW_CM_STATE_LISTEN ) { +spin_unlock_irqrestore (&listen_id_priv->lock, flags); +return ; + } +spin_unlock_irqrestore (&listen_id_priv->lock, flags); + + cm_id =iw_create_cm_id (listen_id_priv->id.device, + listen_id_priv->id.cm_handler, + listen_id_priv->id.context); +/* If the cm_id could not be created, ignore the request */ +if (IS_ERR (cm_id)) +return ; + + cm_id->provider_data = iw_event->provider_data; + cm_id->local_addr = iw_event->local_addr; + cm_id->remote_addr = iw_event->remote_addr; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); + cm_id_priv->state =IW_CM_STATE_CONN_RECV ; + + ret =alloc_work_entries (cm_id_priv,3 ); +if (ret) { +iw_cm_reject (cm_id,NULL ,0 ); +iw_destroy_cm_id (cm_id); +return ; + } + +/* Call the client CM handler */ + ret = cm_id->cm_handler (cm_id, iw_event); +if (ret) { +set_bit (IWCM_F_CALLBACK_DESTROY , &cm_id_priv->flags); +destroy_cm_id (cm_id); +if (atomic_read (&cm_id_priv->refcount)==0 ) +kfree (cm_id); + } + +if (iw_event->private_data_len) +kfree (iw_event->private_data); +} + +/* +* Passive Side: CM_ID <-- ESTABLISHED +* +* The provider generated an ESTABLISHED event which means that +* the MPA negotion has completed successfully and we are now in MPA +* FPDU mode. +* +* This event can only be received in the CONN_RECV state. If the +* remote peer closed, the ESTABLISHED event would be received followed +* by the CLOSE event. If the app closes, it will block until we wake +* it up after processing this event. +*/ +static int cm_conn_est_handler (struct iwcm_id_private *cm_id_priv, +struct iw_cm_event *iw_event) +{ +unsigned long flags; +int ret =0 ; + +spin_lock_irqsave (&cm_id_priv->lock, flags); + +/* +* We clear the CONNECT_WAIT bit here to allow the callback +* function to call iw_cm_disconnect. Calling iw_destroy_cm_id +* from a callback handler is not allowed. +*/ +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +BUG_ON (cm_id_priv->state !=IW_CM_STATE_CONN_RECV ); + cm_id_priv->state =IW_CM_STATE_ESTABLISHED ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + ret = cm_id_priv->id.cm_handler (&cm_id_priv->id, iw_event); +wake_up_all (&cm_id_priv->connect_wait); + +return ret; +} + +/* +* Active Side: CM_ID <-- ESTABLISHED +* +* The app has called connect and is waiting for the established event to +* post it's requests to the server. This event will wake up anyone +* blocked in iw_cm_disconnect or iw_destroy_id. +*/ +static int cm_conn_rep_handler (struct iwcm_id_private *cm_id_priv, +struct iw_cm_event *iw_event) +{ +unsigned long flags; +int ret =0 ; + +spin_lock_irqsave (&cm_id_priv->lock, flags); +/* +* Clear the connect wait bit so a callback function calling +* iw_cm_disconnect will not wait and deadlock this thread +*/ +clear_bit (IWCM_F_CONNECT_WAIT , &cm_id_priv->flags); +BUG_ON (cm_id_priv->state !=IW_CM_STATE_CONN_SENT ); +if (iw_event->status ==IW_CM_EVENT_STATUS_ACCEPTED ) { + cm_id_priv->id.local_addr = iw_event->local_addr; + cm_id_priv->id.remote_addr = iw_event->remote_addr; + cm_id_priv->state =IW_CM_STATE_ESTABLISHED ; + }else { +/* REJECTED or RESET */ + cm_id_priv->id.device->iwcm->rem_ref (cm_id_priv->qp); + cm_id_priv->qp =NULL ; + cm_id_priv->state =IW_CM_STATE_IDLE ; + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + ret = cm_id_priv->id.cm_handler (&cm_id_priv->id, iw_event); + +if (iw_event->private_data_len) +kfree (iw_event->private_data); + +/* Wake up waiters on connect complete */ +wake_up_all (&cm_id_priv->connect_wait); + +return ret; +} + +/* +* CM_ID <-- CLOSING +* +* If in the ESTABLISHED state, move to CLOSING. +*/ +static void cm_disconnect_handler (struct iwcm_id_private *cm_id_priv, +struct iw_cm_event *iw_event) +{ +unsigned long flags; + +spin_lock_irqsave (&cm_id_priv->lock, flags); +if (cm_id_priv->state ==IW_CM_STATE_ESTABLISHED ) + cm_id_priv->state =IW_CM_STATE_CLOSING ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +} + +/* +* CM_ID <-- IDLE +* +* If in the ESTBLISHED or CLOSING states, the QP will have have been +* moved by the provider to the ERR state. Disassociate the CM_ID from +* the QP, move to IDLE, and remove the 'connected' reference. +* +* If in some other state, the cm_id was destroyed asynchronously. +* This is the last reference that will result in waking up +* the app thread blocked in iw_destroy_cm_id. +*/ +static int cm_close_handler (struct iwcm_id_private *cm_id_priv, +struct iw_cm_event *iw_event) +{ +unsigned long flags; +int ret =0 ; +spin_lock_irqsave (&cm_id_priv->lock, flags); + +if (cm_id_priv->qp) { + cm_id_priv->id.device->iwcm->rem_ref (cm_id_priv->qp); + cm_id_priv->qp =NULL ; + } +switch (cm_id_priv->state) { +case IW_CM_STATE_ESTABLISHED : +case IW_CM_STATE_CLOSING : + cm_id_priv->state =IW_CM_STATE_IDLE ; +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + ret = cm_id_priv->id.cm_handler (&cm_id_priv->id, iw_event); +spin_lock_irqsave (&cm_id_priv->lock, flags); +break ; +case IW_CM_STATE_DESTROYING : +break ; +default : +BUG (); + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + +return ret; +} + +static int process_event (struct iwcm_id_private *cm_id_priv, +struct iw_cm_event *iw_event) +{ +int ret =0 ; + +switch (iw_event->event) { +case IW_CM_EVENT_CONNECT_REQUEST : +cm_conn_req_handler (cm_id_priv, iw_event); +break ; +case IW_CM_EVENT_CONNECT_REPLY : + ret =cm_conn_rep_handler (cm_id_priv, iw_event); +break ; +case IW_CM_EVENT_ESTABLISHED : + ret =cm_conn_est_handler (cm_id_priv, iw_event); +break ; +case IW_CM_EVENT_DISCONNECT : +cm_disconnect_handler (cm_id_priv, iw_event); +break ; +case IW_CM_EVENT_CLOSE : + ret =cm_close_handler (cm_id_priv, iw_event); +break ; +default : +BUG (); + } + +return ret; +} + +/* +* Process events on the work_list for the cm_id. If the callback +* function requests that the cm_id be deleted, a flag is set in the +* cm_id flags to indicate that when the last reference is +* removed, the cm_id is to be destroyed. This is necessary to +* distinguish between an object that will be destroyed by the app +* thread asleep on the destroy_comp list vs. an object destroyed +* here synchronously when the last reference is removed. +*/ +static void cm_work_handler (void *arg) +{ +struct iwcm_work *work = arg, lwork; +struct iwcm_id_private *cm_id_priv = work->cm_id; +unsigned long flags; +int empty; +int ret =0 ; + +spin_lock_irqsave (&cm_id_priv->lock, flags); + empty =list_empty (&cm_id_priv->work_list); +while (!empty) { + work =list_entry (cm_id_priv->work_list.next, + struct iwcm_work, list); +list_del_init (&work->list); + empty =list_empty (&cm_id_priv->work_list); + lwork = *work; +put_work (work); +spin_unlock_irqrestore (&cm_id_priv->lock, flags); + + ret =process_event (cm_id_priv, &work->event); +if (ret) { +set_bit (IWCM_F_CALLBACK_DESTROY , &cm_id_priv->flags); +destroy_cm_id (&cm_id_priv->id); + } +BUG_ON (atomic_read (&cm_id_priv->refcount)==0 ); +if (iwcm_deref_id (cm_id_priv)) +return ; + +if (atomic_read (&cm_id_priv->refcount)==0 && +test_bit (IWCM_F_CALLBACK_DESTROY , &cm_id_priv->flags)) { +dealloc_work_entries (cm_id_priv); +kfree (cm_id_priv); +return ; + } +spin_lock_irqsave (&cm_id_priv->lock, flags); + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +} + +/* +* This function is called on interrupt context. Schedule events on +* the iwcm_wq thread to allow callback functions to downcall into +* the CM and/or block. Events are queued to a per-CM_ID +* work_list. If this is the first event on the work_list, the work +* element is also queued on the iwcm_wq thread. +* +* Each event holds a reference on the cm_id. Until the last posted +* event has been delivered and processed, the cm_id cannot be +* deleted. +* +* Returns: +* 0 - the event was handled. +* -ENOMEM - the event was not handled due to lack of resources. +*/ +static int cm_event_handler (struct iw_cm_id *cm_id, +struct iw_cm_event *iw_event) +{ +struct iwcm_work *work; +struct iwcm_id_private *cm_id_priv; +unsigned long flags; +int ret =0 ; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); + +spin_lock_irqsave (&cm_id_priv->lock, flags); + work =get_work (cm_id_priv); +if (!work) { + ret = -ENOMEM ; + goto out; + } + +INIT_WORK (&work->work, cm_work_handler, work); + work->cm_id = cm_id_priv; + work->event = *iw_event; + +if ((work->event.event ==IW_CM_EVENT_CONNECT_REQUEST || + work->event.event ==IW_CM_EVENT_CONNECT_REPLY ) && + work->event.private_data_len) { + ret =copy_private_data (cm_id_priv, &work->event); +if (ret) { +put_work (work); + goto out; + } + } + +atomic_inc (&cm_id_priv->refcount); +if (list_empty (&cm_id_priv->work_list)) { +list_add_tail (&work->list, &cm_id_priv->work_list); +queue_work (iwcm_wq, &work->work); + }else +list_add_tail (&work->list, &cm_id_priv->work_list); +out: +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +return ret; +} + +static int iwcm_init_qp_init_attr (struct iwcm_id_private *cm_id_priv, +struct ib_qp_attr *qp_attr, +int *qp_attr_mask) +{ +unsigned long flags; +int ret; + +spin_lock_irqsave (&cm_id_priv->lock, flags); +switch (cm_id_priv->state) { +case IW_CM_STATE_IDLE : +case IW_CM_STATE_CONN_SENT : +case IW_CM_STATE_CONN_RECV : +case IW_CM_STATE_ESTABLISHED : + *qp_attr_mask =IB_QP_STATE |IB_QP_ACCESS_FLAGS ; + qp_attr->qp_access_flags =IB_ACCESS_LOCAL_WRITE | +IB_ACCESS_REMOTE_WRITE | +IB_ACCESS_REMOTE_READ ; + ret =0 ; +break ; +default : + ret = -EINVAL ; +break ; + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +return ret; +} + +static int iwcm_init_qp_rts_attr (struct iwcm_id_private *cm_id_priv, +struct ib_qp_attr *qp_attr, +int *qp_attr_mask) +{ +unsigned long flags; +int ret; + +spin_lock_irqsave (&cm_id_priv->lock, flags); +switch (cm_id_priv->state) { +case IW_CM_STATE_IDLE : +case IW_CM_STATE_CONN_SENT : +case IW_CM_STATE_CONN_RECV : +case IW_CM_STATE_ESTABLISHED : + *qp_attr_mask =0 ; + ret =0 ; +break ; +default : + ret = -EINVAL ; +break ; + } +spin_unlock_irqrestore (&cm_id_priv->lock, flags); +return ret; +} + +int iw_cm_init_qp_attr (struct iw_cm_id *cm_id, +struct ib_qp_attr *qp_attr, +int *qp_attr_mask) +{ +struct iwcm_id_private *cm_id_priv; +int ret; + + cm_id_priv = container_of(cm_id, struct iwcm_id_private, id); +switch (qp_attr->qp_state) { +case IB_QPS_INIT : +case IB_QPS_RTR : + ret =iwcm_init_qp_init_attr (cm_id_priv, + qp_attr, qp_attr_mask); +break ; +case IB_QPS_RTS : + ret =iwcm_init_qp_rts_attr (cm_id_priv, + qp_attr, qp_attr_mask); +break ; +default : + ret = -EINVAL ; +break ; + } +return ret; +} +EXPORT_SYMBOL (iw_cm_init_qp_attr); + +static int __init iw_cm_init (void ) +{ + iwcm_wq =create_singlethread_workqueue ("iw_cm_wq" ); +if (!iwcm_wq) +return -ENOMEM ; + +return 0 ; +} + +static void __exit iw_cm_cleanup(void) +{ +destroy_workqueue (iwcm_wq); +} + +module_init (iw_cm_init); +module_exit (iw_cm_cleanup);
drivers/infiniband/core/iwcm.h new file +62 -0
62 lines not on this page · raw message
include/rdma/iw_cm.h new file +258 -0
258 lines not on this page · raw message
Signature · 1 line
1.4.1
PATCH 2/2 RDMA: iWARP changes to IB core
From: Tom Tucker <[email protected]>
Modifications to the existing rdma header files, core files, drivers, and ulp files to support iWARP, including:
- Hook iWARP CM into the build system and use it in rdma_cm.
- Convert enum ib_node_type to enum rdma_node_type, which includes the possibility of RDMA_NODE_RNIC, and update everything for this.
--- drivers/infiniband/core/Makefile | 4 drivers/infiniband/core/addr.c | 18 + drivers/infiniband/core/cache.c | 5 drivers/infiniband/core/cm.c | 3 drivers/infiniband/core/cma.c | 355 +++++++++++++++++++++++--- drivers/infiniband/core/device.c | 4 drivers/infiniband/core/mad.c | 7 - drivers/infiniband/core/sa_query.c | 5 drivers/infiniband/core/smi.c | 16 + drivers/infiniband/core/sysfs.c | 11 - drivers/infiniband/core/ucm.c | 3 drivers/infiniband/core/user_mad.c | 5 drivers/infiniband/core/verbs.c | 17 + drivers/infiniband/hw/ehca/ehca_main.c | 2 drivers/infiniband/hw/ipath/ipath_verbs.c | 2 drivers/infiniband/hw/mthca/mthca_provider.c | 2 drivers/infiniband/ulp/ipoib/ipoib_main.c | 8 + drivers/infiniband/ulp/srp/ib_srp.c | 2 include/rdma/ib_addr.h | 17 + include/rdma/ib_verbs.h | 25 ++ 20 files changed, 430 insertions(+), 81 deletions(-)
drivers/infiniband/core/Makefile +3 -1
@@ -1,7 +1,7 @@ infiniband-$(CONFIG_INFINIBAND_ADDR_TRANS) := ib_addr.o rdma_cm.o obj-$(CONFIG_INFINIBAND) += ib_core.o ib_mad.o ib_sa.o \ - ib_cm.o $(infiniband-y) + ib_cm.o iw_cm.o $(infiniband-y) obj-$(CONFIG_INFINIBAND_USER_MAD) += ib_umad.o obj-$(CONFIG_INFINIBAND_USER_ACCESS) += ib_uverbs.o ib_ucm.o
@@ -14,6 +14,8 @@ ib_sa-y := sa_query.o ib_cm-y := cm.o +iw_cm-y := iwcm.o + rdma_cm-y := cma.o ib_addr-y := addr.o
drivers/infiniband/core/addr.c +11 -7
@@ -61,12 +61,15 @@ static LIST_HEAD(req_list); static DECLARE_WORK(work, process_req, NULL);static struct workqueue_struct *addr_wq; -static int copy_addr (struct rdma_dev_addr *dev_addr,struct net_device *dev, -unsigned char *dst_dev_addr) +int rdma_copy_addr (struct rdma_dev_addr *dev_addr,struct net_device *dev, +const unsigned char *dst_dev_addr) {switch (dev->type) {case ARPHRD_INFINIBAND : - dev_addr->dev_type =IB_NODE_CA ; + dev_addr->dev_type =RDMA_NODE_IB_CA ; +break ; +case ARPHRD_ETHER : + dev_addr->dev_type =RDMA_NODE_RNIC ;break ;default :return -EADDRNOTAVAIL ;
@@ -78,6 +81,7 @@ static int copy_addr(struct rdma_dev_addmemcpy (dev_addr->dst_dev_addr, dst_dev_addr,MAX_ADDR_LEN );return 0 ; } +EXPORT_SYMBOL (rdma_copy_addr);int rdma_translate_ip (struct sockaddr *addr,struct rdma_dev_addr *dev_addr) {
@@ -89,7 +93,7 @@ int rdma_translate_ip(struct sockaddr *aif (!dev)return -EADDRNOTAVAIL ; - ret =copy_addr (dev_addr, dev,NULL ); + ret =rdma_copy_addr (dev_addr, dev,NULL );dev_put (dev);return ret; }
@@ -161,7 +165,7 @@ static int addr_resolve_remote(struct so/* If the device does ARP internally, return 'done' */ if (rt->idev->dev->flags &IFF_NOARP ) { -copy_addr (addr, rt->idev->dev,NULL ); +rdma_copy_addr (addr, rt->idev->dev,NULL ); goto put; }
@@ -181,7 +185,7 @@ static int addr_resolve_remote(struct so src_in->sin_addr.s_addr = rt->rt_src; } - ret =copy_addr (addr, neigh->dev, neigh->ha); + ret =rdma_copy_addr (addr, neigh->dev, neigh->ha); release:neigh_release (neigh); put:
@@ -245,7 +249,7 @@ static int addr_resolve_local(struct socif (ZERONET (src_ip)) { src_in->sin_family = dst_in->sin_family; src_in->sin_addr.s_addr = dst_ip; - ret =copy_addr (addr, dev, dev->dev_addr); + ret =rdma_copy_addr (addr, dev, dev->dev_addr); }else if (LOOPBACK (src_ip)) { ret =rdma_translate_ip ((struct sockaddr *)dst_in, addr);if (!ret)
drivers/infiniband/core/cache.c +3 -2
@@ -62,12 +62,13 @@ struct ib_update_work {static inline int start_port (struct ib_device *device) { -return device->node_type ==IB_NODE_SWITCH ?0 :1 ; +return (device->node_type ==RDMA_NODE_IB_SWITCH ) ?0 :1 ; }static inline int end_port (struct ib_device *device) { -return device->node_type ==IB_NODE_SWITCH ?0 : device->phys_port_cnt; +return (device->node_type ==RDMA_NODE_IB_SWITCH ) ? +0 : device->phys_port_cnt; } int ib_get_cached_gid(struct ib_device *device,
drivers/infiniband/core/cm.c +3 -0
@@ -3280,6 +3280,9 @@ static void cm_add_one(struct ib_deviceint ret;u8 i; +if (rdma_node_get_transport (device->node_type) !=RDMA_TRANSPORT_IB ) +return ; + cm_dev =kmalloc (sizeof (*cm_dev) +sizeof (*port) * device->phys_port_cnt,GFP_KERNEL );if (!cm_dev)
drivers/infiniband/core/cma.c +314 -41
538 lines not on this page · raw message
drivers/infiniband/core/device.c +2 -2
@@ -505,7 +505,7 @@ int ib_query_port(struct ib_device *devi u8 port_num, struct ib_port_attr *port_attr) { -if (device->node_type ==IB_NODE_SWITCH ) { +if (device->node_type ==RDMA_NODE_IB_SWITCH ) {if (port_num)return -EINVAL ; }else if (port_num <1 || port_num > device->phys_port_cnt)
@@ -580,7 +580,7 @@ int ib_modify_port(struct ib_device *dev u8 port_num, int port_modify_mask, struct ib_port_modify *port_modify) { - if (device->node_type == IB_NODE_SWITCH) { + if (device->node_type == RDMA_NODE_IB_SWITCH) { if (port_num) return -EINVAL; } else if (port_num < 1 || port_num > device->phys_port_cnt)
drivers/infiniband/core/mad.c +5 -2
@@ -2876,7 +2876,10 @@ static void ib_mad_init_device(struct ib {int start, end, i; -if (device->node_type ==IB_NODE_SWITCH ) { +if (rdma_node_get_transport (device->node_type) !=RDMA_TRANSPORT_IB ) +return ; + +if (device->node_type ==RDMA_NODE_IB_SWITCH ) { start =0 ; end =0 ; }else {
@@ -2923,7 +2926,7 @@ static void ib_mad_remove_device(struct {int i, num_ports, cur_port; -if (device->node_type ==IB_NODE_SWITCH ) { +if (device->node_type ==RDMA_NODE_IB_SWITCH ) { num_ports =1 ; cur_port =0 ; }else {
drivers/infiniband/core/sa_query.c +4 -1
@@ -919,7 +919,10 @@ static void ib_sa_add_one(struct ib_devistruct ib_sa_device *sa_dev;int s, e, i; -if (device->node_type ==IB_NODE_SWITCH ) +if (rdma_node_get_transport (device->node_type) !=RDMA_TRANSPORT_IB ) +return ; + +if (device->node_type ==RDMA_NODE_IB_SWITCH ) s = e =0 ;else { s =1 ;
drivers/infiniband/core/smi.c +8 -8
@@ -64,7 +64,7 @@ int smi_handle_dr_smp_send(struct ib_smp/* C14-9:2 */ if (hop_ptr && hop_ptr < hop_cnt) { -if (node_type !=IB_NODE_SWITCH ) +if (node_type !=RDMA_NODE_IB_SWITCH )return 0 ;/* smp->return_path set when received */
@@ -77,7 +77,7 @@ int smi_handle_dr_smp_send(struct ib_smpif (hop_ptr == hop_cnt) {/* smp->return_path set when received */ smp->hop_ptr++; -return (node_type ==IB_NODE_SWITCH || +return (node_type ==RDMA_NODE_IB_SWITCH || smp->dr_dlid ==IB_LID_PERMISSIVE ); }
@@ -95,7 +95,7 @@ int smi_handle_dr_smp_send(struct ib_smp/* C14-13:2 */ if (2 <= hop_ptr && hop_ptr <= hop_cnt) { -if (node_type !=IB_NODE_SWITCH ) +if (node_type !=RDMA_NODE_IB_SWITCH )return 0 ; smp->hop_ptr--;
@@ -107,7 +107,7 @@ int smi_handle_dr_smp_send(struct ib_smpif (hop_ptr ==1 ) { smp->hop_ptr--;/* C14-13:3 -- SMPs destined for SM shouldn't be here */ -return (node_type ==IB_NODE_SWITCH || +return (node_type ==RDMA_NODE_IB_SWITCH || smp->dr_slid ==IB_LID_PERMISSIVE ); }
@@ -142,7 +142,7 @@ int smi_handle_dr_smp_recv(struct ib_smp/* C14-9:2 -- intermediate hop */ if (hop_ptr && hop_ptr < hop_cnt) { -if (node_type !=IB_NODE_SWITCH ) +if (node_type !=RDMA_NODE_IB_SWITCH )return 0 ; smp->return_path[hop_ptr] = port_num;
@@ -156,7 +156,7 @@ int smi_handle_dr_smp_recv(struct ib_smp smp->return_path[hop_ptr] = port_num;/* smp->hop_ptr updated when sending */ -return (node_type ==IB_NODE_SWITCH || +return (node_type ==RDMA_NODE_IB_SWITCH || smp->dr_dlid ==IB_LID_PERMISSIVE ); }
@@ -175,7 +175,7 @@ int smi_handle_dr_smp_recv(struct ib_smp/* C14-13:2 */ if (2 <= hop_ptr && hop_ptr <= hop_cnt) { -if (node_type !=IB_NODE_SWITCH ) +if (node_type !=RDMA_NODE_IB_SWITCH )return 0 ;/* smp->hop_ptr updated when sending */
@@ -190,7 +190,7 @@ int smi_handle_dr_smp_recv(struct ib_smpreturn 1 ; }/* smp->hop_ptr updated when sending */ -return (node_type ==IB_NODE_SWITCH ); +return (node_type ==RDMA_NODE_IB_SWITCH ); }/* C14-13:4 -- hop_ptr = 0 -> give to SM */
drivers/infiniband/core/sysfs.c +6 -5
@@ -589,10 +589,11 @@ static ssize_t show_node_type(struct clareturn -ENODEV ;switch (dev->node_type) { -case IB_NODE_CA :return sprintf (buf,"%d: CA\n" , dev->node_type); -case IB_NODE_SWITCH :return sprintf (buf,"%d: switch\n" , dev->node_type); -case IB_NODE_ROUTER :return sprintf (buf,"%d: router\n" , dev->node_type); -default :return sprintf (buf,"%d: <unknown>\n" , dev->node_type); +case RDMA_NODE_IB_CA :return sprintf (buf,"%d: CA\n" , dev->node_type); +case RDMA_NODE_RNIC :return sprintf (buf,"%d: RNIC\n" , dev->node_type); +case RDMA_NODE_IB_SWITCH :return sprintf (buf,"%d: switch\n" , dev->node_type); +case RDMA_NODE_IB_ROUTER :return sprintf (buf,"%d: router\n" , dev->node_type); +default :return sprintf (buf,"%d: <unknown>\n" , dev->node_type); } }
@@ -708,7 +709,7 @@ int ib_device_register_sysfs(struct ib_dif (ret) goto err_put; -if (device->node_type ==IB_NODE_SWITCH ) { +if (device->node_type ==RDMA_NODE_IB_SWITCH ) { ret =add_port (device,0 );if (ret) goto err_put;
drivers/infiniband/core/ucm.c +2 -1
@@ -1247,7 +1247,8 @@ static void ib_ucm_add_one(struct ib_dev {struct ib_ucm_device *ucm_dev; -if (!device->alloc_ucontext) +if (!device->alloc_ucontext || +rdma_node_get_transport (device->node_type) !=RDMA_TRANSPORT_IB )return ; ucm_dev =kzalloc (sizeof *ucm_dev,GFP_KERNEL );
drivers/infiniband/core/user_mad.c +4 -1
@@ -1032,7 +1032,10 @@ static void ib_umad_add_one(struct ib_destruct ib_umad_device *umad_dev;int s, e, i; -if (device->node_type ==IB_NODE_SWITCH ) +if (rdma_node_get_transport (device->node_type) !=RDMA_TRANSPORT_IB ) +return ; + +if (device->node_type ==RDMA_NODE_IB_SWITCH ) s = e =0 ;else { s =1 ;
drivers/infiniband/core/verbs.c +17 -0
@@ -79,6 +79,23 @@ enum ib_rate mult_to_ib_rate(int mult) }EXPORT_SYMBOL (mult_to_ib_rate); +enum rdma_transport_type +rdma_node_get_transport (enum rdma_node_type node_type) +{ +switch (node_type) { +case RDMA_NODE_IB_CA : +case RDMA_NODE_IB_SWITCH : +case RDMA_NODE_IB_ROUTER : +return RDMA_TRANSPORT_IB ; +case RDMA_NODE_RNIC : +return RDMA_TRANSPORT_IWARP ; +default : +BUG (); +return 0 ; + } +} +EXPORT_SYMBOL (rdma_node_get_transport); +/* Protection domains */ struct ib_pd *ib_alloc_pd (struct ib_device *device)
drivers/infiniband/hw/ehca/ehca_main.c +1 -1
@@ -268,7 +268,7 @@ int ehca_register_device(struct ehca_shc (1ull <<IB_USER_VERBS_CMD_ATTACH_MCAST ) | (1ull <<IB_USER_VERBS_CMD_DETACH_MCAST ); - shca->ib_device.node_type =IB_NODE_CA ; + shca->ib_device.node_type =RDMA_NODE_IB_CA ; shca->ib_device.phys_port_cnt = shca->num_ports; shca->ib_device.dma_device = &shca->ibmebus_dev->ofdev.dev; shca->ib_device.query_device = ehca_query_device;
drivers/infiniband/hw/ipath/ipath_verbs.c +1 -1
@@ -1538,7 +1538,7 @@ int ipath_register_ib_device(struct ipat (1ull <<IB_USER_VERBS_CMD_QUERY_SRQ ) | (1ull <<IB_USER_VERBS_CMD_DESTROY_SRQ ) | (1ull <<IB_USER_VERBS_CMD_POST_SRQ_RECV ); - dev->node_type =IB_NODE_CA ; + dev->node_type =RDMA_NODE_IB_CA ; dev->phys_port_cnt =1 ; dev->dma_device = &dd->pcidev->dev; dev->class_dev.dev = dev->dma_device;
drivers/infiniband/hw/mthca/mthca_provider.c +1 -1
@@ -1288,7 +1288,7 @@ int mthca_register_device(struct mthca_d (1ull <<IB_USER_VERBS_CMD_DESTROY_QP ) | (1ull <<IB_USER_VERBS_CMD_ATTACH_MCAST ) | (1ull <<IB_USER_VERBS_CMD_DETACH_MCAST ); - dev->ib_dev.node_type =IB_NODE_CA ; + dev->ib_dev.node_type =RDMA_NODE_IB_CA ; dev->ib_dev.phys_port_cnt = dev->limits.num_ports; dev->ib_dev.dma_device = &dev->pdev->dev; dev->ib_dev.class_dev.dev = &dev->pdev->dev;
drivers/infiniband/ulp/ipoib/ipoib_main.c +7 -1
@@ -1111,13 +1111,16 @@ static void ipoib_add_one(struct ib_devistruct ipoib_dev_priv *priv;int s, e, p; +if (rdma_node_get_transport (device->node_type) !=RDMA_TRANSPORT_IB ) +return ; + dev_list =kmalloc (sizeof *dev_list,GFP_KERNEL );if (!dev_list)return ;INIT_LIST_HEAD (dev_list); -if (device->node_type ==IB_NODE_SWITCH ) { +if (device->node_type ==RDMA_NODE_IB_SWITCH ) { s =0 ; e =0 ; }else {
@@ -1141,6 +1144,9 @@ static void ipoib_remove_one(struct ib_dstruct ipoib_dev_priv *priv, *tmp;struct list_head *dev_list; +if (rdma_node_get_transport (device->node_type) !=RDMA_TRANSPORT_IB ) +return ; + dev_list =ib_get_client_data (device, &ipoib_client);list_for_each_entry_safe (priv, tmp, dev_list, list) {
drivers/infiniband/ulp/srp/ib_srp.c +1 -1
@@ -1913,7 +1913,7 @@ static void srp_add_one(struct ib_deviceif (IS_ERR (srp_dev->fmr_pool)) srp_dev->fmr_pool =NULL ; -if (device->node_type ==IB_NODE_SWITCH ) { +if (device->node_type ==RDMA_NODE_IB_SWITCH ) { s =0 ; e =0 ; }else {
include/rdma/ib_addr.h +16 -1
@@ -40,7 +40,7 @@ struct rdma_dev_addr {unsigned char src_dev_addr[MAX_ADDR_LEN ];unsigned char dst_dev_addr[MAX_ADDR_LEN ];unsigned char broadcast[MAX_ADDR_LEN ]; -enum ib_node_type dev_type; +enum rdma_node_type dev_type; };/**
@@ -72,6 +72,9 @@ int rdma_resolve_ip(struct sockaddr *srcvoid rdma_addr_cancel (struct rdma_dev_addr *addr); +int rdma_copy_addr (struct rdma_dev_addr *dev_addr,struct net_device *dev, +const unsigned char *dst_dev_addr); +static inline int ip_addr_size (struct sockaddr *addr) { return addr->sa_family == AF_INET6 ?
@@ -113,4 +116,16 @@ static inline void ib_addr_set_dgid(stru memcpy(dev_addr->dst_dev_addr + 4, gid, sizeof *gid); } +static inline void iw_addr_get_sgid (struct rdma_dev_addr *dev_addr, +union ib_gid *gid) +{ +memcpy (gid, dev_addr->src_dev_addr,sizeof *gid); +} + +static inline void iw_addr_get_dgid (struct rdma_dev_addr *dev_addr, +union ib_gid *gid) +{ +memcpy (gid, dev_addr->dst_dev_addr,sizeof *gid); +} +#endif /* IB_ADDR_H */
include/rdma/ib_verbs.h +21 -4
@@ -56,12 +56,22 @@ union ib_gid { } global; }; -enum ib_node_type { -IB_NODE_CA =1 , -IB_NODE_SWITCH , -IB_NODE_ROUTER +enum rdma_node_type { +/* IB values map to NodeInfo:NodeType. */ +RDMA_NODE_IB_CA =1 , +RDMA_NODE_IB_SWITCH , +RDMA_NODE_IB_ROUTER , +RDMA_NODE_RNIC }; +enum rdma_transport_type { +RDMA_TRANSPORT_IB , +RDMA_TRANSPORT_IWARP +}; + +enum rdma_transport_type +rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__; +enum ib_device_cap_flags {IB_DEVICE_RESIZE_MAX_WR =1 ,IB_DEVICE_BAD_PKEY_CNTR = (1 <<1 ),
@@ -78,6 +88,9 @@ enum ib_device_cap_flags {IB_DEVICE_RC_RNR_NAK_GEN = (1 <<12 ),IB_DEVICE_SRQ_RESIZE = (1 <<13 ),IB_DEVICE_N_NOTIFY_CQ = (1 <<14 ), +IB_DEVICE_ZERO_STAG = (1 <<15 ), +IB_DEVICE_SEND_W_INV = (1 <<16 ), +IB_DEVICE_MEM_WINDOW = (1 <<17 ) }; enum ib_atomic_cap {
@@ -835,6 +848,8 @@ struct ib_cache {u8 *lmc_cache; }; +struct iw_cm_verbs ; +struct ib_device {struct device *dma_device;
@@ -851,6 +866,8 @@ struct ib_device {u32 flags; +struct iw_cm_verbs *iwcm; +int (*query_device)(struct ib_device *device,struct ib_device_attr *device_attr); int (*query_port)(struct ib_device *device,
Signature · 1 line
1.4.1
[PATCH 3/2] RDMA: Ammasso 1100 RNIC driver
Here's the compressed patch adding the amso1100 driver. You can also find this in my git tree at
git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband.git
in the for-2.6.19 branch.