blob: 781a3e2eaf07c1fa46cdfdab9dddd882a5321277 [file] [log] [blame]
/*
* Copyright (C) 2009 Michael Brown <mbrown@fensystems.co.uk>.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation; either version 2 of the
* License, or any later version.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
* 02110-1301, USA.
*
* You can also choose to distribute this program under the terms of
* the Unmodified Binary Distribution Licence (as given in the file
* COPYING.UBDL), provided that you have satisfied its requirements.
*/
FILE_LICENCE ( GPL2_OR_LATER_OR_UBDL );
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>
#include <stdio.h>
#include <unistd.h>
#include <byteswap.h>
#include <ipxe/infiniband.h>
#include <ipxe/iobuf.h>
#include <ipxe/ib_mi.h>
/**
* @file
*
* Infiniband management interfaces
*
*/
/** Management interface number of send WQEs
*
* This is a policy decision.
*/
#define IB_MI_NUM_SEND_WQES 4
/** Management interface number of receive WQEs
*
* This is a policy decision.
*/
#define IB_MI_NUM_RECV_WQES 2
/** Management interface number of completion queue entries
*
* This is a policy decision
*/
#define IB_MI_NUM_CQES 8
/** TID magic signature */
#define IB_MI_TID_MAGIC ( ( 'i' << 24 ) | ( 'P' << 16 ) | ( 'X' << 8 ) | 'E' )
/** TID to use for next MAD */
static unsigned int next_tid;
/**
* Handle received MAD
*
* @v ibdev Infiniband device
* @v mi Management interface
* @v mad Received MAD
* @v av Source address vector
* @ret rc Return status code
*/
static int ib_mi_handle ( struct ib_device *ibdev,
struct ib_mad_interface *mi,
union ib_mad *mad,
struct ib_address_vector *av ) {
struct ib_mad_hdr *hdr = &mad->hdr;
struct ib_mad_transaction *madx;
struct ib_mad_agent *agent;
/* Look for a matching transaction by TID */
list_for_each_entry ( madx, &mi->madx, list ) {
if ( memcmp ( &hdr->tid, &madx->mad.hdr.tid,
sizeof ( hdr->tid ) ) != 0 )
continue;
/* Found a matching transaction */
madx->op->complete ( ibdev, mi, madx, 0, mad, av );
return 0;
}
/* If there is no matching transaction, look for a listening agent */
for_each_table_entry ( agent, IB_MAD_AGENTS ) {
if ( ( ( agent->mgmt_class & IB_MGMT_CLASS_MASK ) !=
( hdr->mgmt_class & IB_MGMT_CLASS_MASK ) ) ||
( agent->class_version != hdr->class_version ) ||
( agent->attr_id != hdr->attr_id ) )
continue;
/* Found a matching agent */
agent->handle ( ibdev, mi, mad, av );
return 0;
}
/* Otherwise, ignore it */
DBGC ( mi, "MI %p RX TID %08x%08x ignored\n",
mi, ntohl ( hdr->tid.high ), ntohl ( hdr->tid.low ) );
return -ENOTSUP;
}
/**
* Complete receive via management interface
*
*
* @v ibdev Infiniband device
* @v qp Queue pair
* @v dest Destination address vector
* @v source Source address vector
* @v iobuf I/O buffer
* @v rc Completion status code
*/
static void ib_mi_complete_recv ( struct ib_device *ibdev,
struct ib_queue_pair *qp,
struct ib_address_vector *dest __unused,
struct ib_address_vector *source,
struct io_buffer *iobuf, int rc ) {
struct ib_mad_interface *mi = ib_qp_get_ownerdata ( qp );
union ib_mad *mad;
struct ib_mad_hdr *hdr;
/* Ignore errors */
if ( rc != 0 ) {
DBGC ( mi, "MI %p RX error: %s\n", mi, strerror ( rc ) );
goto out;
}
/* Sanity checks */
if ( iob_len ( iobuf ) != sizeof ( *mad ) ) {
DBGC ( mi, "MI %p RX bad size (%zd bytes)\n",
mi, iob_len ( iobuf ) );
DBGC_HDA ( mi, 0, iobuf->data, iob_len ( iobuf ) );
goto out;
}
mad = iobuf->data;
hdr = &mad->hdr;
if ( hdr->base_version != IB_MGMT_BASE_VERSION ) {
DBGC ( mi, "MI %p RX unsupported base version %x\n",
mi, hdr->base_version );
DBGC_HDA ( mi, 0, mad, sizeof ( *mad ) );
goto out;
}
DBGC ( mi, "MI %p RX TID %08x%08x (%02x,%02x,%02x,%04x) status "
"%04x\n", mi, ntohl ( hdr->tid.high ), ntohl ( hdr->tid.low ),
hdr->mgmt_class, hdr->class_version, hdr->method,
ntohs ( hdr->attr_id ), ntohs ( hdr->status ) );
DBGC2_HDA ( mi, 0, mad, sizeof ( *mad ) );
/* Handle MAD */
if ( ( rc = ib_mi_handle ( ibdev, mi, mad, source ) ) != 0 )
goto out;
out:
free_iob ( iobuf );
}
/** Management interface completion operations */
static struct ib_completion_queue_operations ib_mi_completion_ops = {
.complete_recv = ib_mi_complete_recv,
};
/** Management interface queue pair operations */
static struct ib_queue_pair_operations ib_mi_queue_pair_ops = {
.alloc_iob = alloc_iob,
};
/**
* Transmit MAD
*
* @v ibdev Infiniband device
* @v mi Management interface
* @v mad MAD
* @v av Destination address vector
* @ret rc Return status code
*/
int ib_mi_send ( struct ib_device *ibdev, struct ib_mad_interface *mi,
union ib_mad *mad, struct ib_address_vector *av ) {
struct ib_mad_hdr *hdr = &mad->hdr;
struct io_buffer *iobuf;
int rc;
/* Set common fields */
hdr->base_version = IB_MGMT_BASE_VERSION;
if ( ( hdr->tid.high == 0 ) && ( hdr->tid.low == 0 ) ) {
hdr->tid.high = htonl ( IB_MI_TID_MAGIC );
hdr->tid.low = htonl ( ++next_tid );
}
DBGC ( mi, "MI %p TX TID %08x%08x (%02x,%02x,%02x,%04x) status "
"%04x\n", mi, ntohl ( hdr->tid.high ), ntohl ( hdr->tid.low ),
hdr->mgmt_class, hdr->class_version, hdr->method,
ntohs ( hdr->attr_id ), ntohs ( hdr->status ) );
DBGC2_HDA ( mi, 0, mad, sizeof ( *mad ) );
/* Construct directed route portion of response, if necessary */
if ( hdr->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE ) {
struct ib_mad_smp *smp = &mad->smp;
unsigned int hop_pointer;
unsigned int hop_count;
smp->mad_hdr.status |= htons ( IB_SMP_STATUS_D_INBOUND );
hop_pointer = smp->mad_hdr.class_specific.smp.hop_pointer;
hop_count = smp->mad_hdr.class_specific.smp.hop_count;
assert ( hop_count == hop_pointer );
if ( hop_pointer < ( sizeof ( smp->return_path.hops ) /
sizeof ( smp->return_path.hops[0] ) ) ) {
smp->return_path.hops[hop_pointer] = ibdev->port;
} else {
DBGC ( mi, "MI %p TX TID %08x%08x invalid hop pointer "
"%d\n", mi, ntohl ( hdr->tid.high ),
ntohl ( hdr->tid.low ), hop_pointer );
return -EINVAL;
}
}
/* Construct I/O buffer */
iobuf = alloc_iob ( sizeof ( *mad ) );
if ( ! iobuf ) {
DBGC ( mi, "MI %p could not allocate buffer for TID "
"%08x%08x\n",
mi, ntohl ( hdr->tid.high ), ntohl ( hdr->tid.low ) );
return -ENOMEM;
}
memcpy ( iob_put ( iobuf, sizeof ( *mad ) ), mad, sizeof ( *mad ) );
/* Send I/O buffer */
if ( ( rc = ib_post_send ( ibdev, mi->qp, av, iobuf ) ) != 0 ) {
DBGC ( mi, "MI %p TX TID %08x%08x failed: %s\n",
mi, ntohl ( hdr->tid.high ), ntohl ( hdr->tid.low ),
strerror ( rc ) );
free_iob ( iobuf );
return rc;
}
return 0;
}
/**
* Handle management transaction timer expiry
*
* @v timer Retry timer
* @v expired Failure indicator
*/
static void ib_mi_timer_expired ( struct retry_timer *timer, int expired ) {
struct ib_mad_transaction *madx =
container_of ( timer, struct ib_mad_transaction, timer );
struct ib_mad_interface *mi = madx->mi;
struct ib_device *ibdev = mi->ibdev;
struct ib_mad_hdr *hdr = &madx->mad.hdr;
/* Abandon transaction if we have tried too many times */
if ( expired ) {
DBGC ( mi, "MI %p abandoning TID %08x%08x\n",
mi, ntohl ( hdr->tid.high ), ntohl ( hdr->tid.low ) );
madx->op->complete ( ibdev, mi, madx, -ETIMEDOUT, NULL, NULL );
return;
}
/* Restart retransmission timer */
start_timer ( timer );
/* Resend MAD */
ib_mi_send ( ibdev, mi, &madx->mad, &madx->av );
}
/**
* Create management transaction
*
* @v ibdev Infiniband device
* @v mi Management interface
* @v mad MAD to send
* @v av Destination address, or NULL to use SM's GSI
* @v op Management transaction operations
* @ret madx Management transaction, or NULL
*/
struct ib_mad_transaction *
ib_create_madx ( struct ib_device *ibdev, struct ib_mad_interface *mi,
union ib_mad *mad, struct ib_address_vector *av,
struct ib_mad_transaction_operations *op ) {
struct ib_mad_transaction *madx;
/* Allocate and initialise structure */
madx = zalloc ( sizeof ( *madx ) );
if ( ! madx )
return NULL;
timer_init ( &madx->timer, ib_mi_timer_expired, NULL );
madx->mi = mi;
madx->op = op;
/* Determine address vector */
if ( av ) {
memcpy ( &madx->av, av, sizeof ( madx->av ) );
} else {
madx->av.lid = ibdev->sm_lid;
madx->av.sl = ibdev->sm_sl;
madx->av.qpn = IB_QPN_GSI;
madx->av.qkey = IB_QKEY_GSI;
}
/* Copy MAD */
memcpy ( &madx->mad, mad, sizeof ( madx->mad ) );
/* Add to list and start timer to send initial MAD */
list_add ( &madx->list, &mi->madx );
start_timer_nodelay ( &madx->timer );
return madx;
}
/**
* Destroy management transaction
*
* @v ibdev Infiniband device
* @v mi Management interface
* @v madx Management transaction
*/
void ib_destroy_madx ( struct ib_device *ibdev __unused,
struct ib_mad_interface *mi __unused,
struct ib_mad_transaction *madx ) {
/* Stop timer and remove from list */
stop_timer ( &madx->timer );
list_del ( &madx->list );
/* Free transaction */
free ( madx );
}
/**
* Create management interface
*
* @v ibdev Infiniband device
* @v type Queue pair type
* @v new_mi New management interface to fill in
* @ret rc Return status code
*/
int ib_create_mi ( struct ib_device *ibdev, enum ib_queue_pair_type type,
struct ib_mad_interface **new_mi ) {
struct ib_mad_interface *mi;
const char *name;
int rc;
/* Allocate and initialise fields */
mi = zalloc ( sizeof ( *mi ) );
if ( ! mi ) {
rc = -ENOMEM;
goto err_alloc;
}
mi->ibdev = ibdev;
INIT_LIST_HEAD ( &mi->madx );
/* Create completion queue */
if ( ( rc = ib_create_cq ( ibdev, IB_MI_NUM_CQES, &ib_mi_completion_ops,
&mi->cq ) ) != 0 ) {
DBGC ( mi, "MI %p could not create completion queue: %s\n",
mi, strerror ( rc ) );
goto err_create_cq;
}
/* Create queue pair */
name = ( ( type == IB_QPT_SMI ) ? "SMI" : "GSI" );
if ( ( rc = ib_create_qp ( ibdev, type, IB_MI_NUM_SEND_WQES, mi->cq,
IB_MI_NUM_RECV_WQES, mi->cq,
&ib_mi_queue_pair_ops, name, &mi->qp ) )!=0){
DBGC ( mi, "MI %p could not create queue pair: %s\n",
mi, strerror ( rc ) );
goto err_create_qp;
}
ib_qp_set_ownerdata ( mi->qp, mi );
DBGC ( mi, "MI %p (%s) running on QPN %#lx\n",
mi, mi->qp->name, mi->qp->qpn );
/* Set queue key */
mi->qp->qkey = ( ( type == IB_QPT_SMI ) ? IB_QKEY_SMI : IB_QKEY_GSI );
if ( ( rc = ib_modify_qp ( ibdev, mi->qp ) ) != 0 ) {
DBGC ( mi, "MI %p could not set queue key: %s\n",
mi, strerror ( rc ) );
goto err_modify_qp;
}
/* Fill receive ring */
ib_refill_recv ( ibdev, mi->qp );
*new_mi = mi;
return 0;
err_modify_qp:
ib_destroy_qp ( ibdev, mi->qp );
err_create_qp:
ib_destroy_cq ( ibdev, mi->cq );
err_create_cq:
free ( mi );
err_alloc:
return rc;
}
/**
* Destroy management interface
*
* @v mi Management interface
*/
void ib_destroy_mi ( struct ib_device *ibdev, struct ib_mad_interface *mi ) {
struct ib_mad_transaction *madx;
struct ib_mad_transaction *tmp;
/* Flush any outstanding requests */
list_for_each_entry_safe ( madx, tmp, &mi->madx, list ) {
DBGC ( mi, "MI %p destroyed while TID %08x%08x in progress\n",
mi, ntohl ( madx->mad.hdr.tid.high ),
ntohl ( madx->mad.hdr.tid.low ) );
madx->op->complete ( ibdev, mi, madx, -ECANCELED, NULL, NULL );
}
ib_destroy_qp ( ibdev, mi->qp );
ib_destroy_cq ( ibdev, mi->cq );
free ( mi );
}