#!/bin/bash
#
# Bring up/down the kernel RDMA stack
#
# chkconfig: - 05 95
# description: Loads/Unloads InfiniBand and iWARP kernel modules
# config:	/etc/rdma/rdma.conf
#
### BEGIN INIT INFO
# Provides:       rdma
# Default-Stop: 0 1 2 3 4 5 6
# Required-Start:
# Required-Stop:
# Short-Description: Loads and unloads the InfiniBand and iWARP kernel modules
# Description: Loads and unloads the InfiniBand and iWARP kernel modules
### END INIT INFO

CONFIG=/etc/rdma/rdma.conf

. /etc/rc.d/init.d/functions

LOAD_ULP_MODULES=""
LOAD_CORE_USER_MODULES="ib_umad ib_uverbs ib_ucm rdma_ucm"
LOAD_CORE_CM_MODULES="iw_cm ib_cm rdma_cm"
LOAD_CORE_MODULES="ib_core ib_mad ib_sa ib_addr"

if [ -f $CONFIG ]; then
    . $CONFIG

    if [ "${RDS_LOAD}" == "yes" ]; then
        IPOIB_LOAD=yes
    fi

    if [ "${IPOIB_LOAD}" == "yes" ]; then
	LOAD_ULP_MODULES="ib_ipoib"
    fi

    if [ "${RDS_LOAD}" == "yes" ]; then
	LOAD_ULP_MODULES="$LOAD_ULP_MODULES rds"
    fi

    if [ "${SRP_LOAD}" == "yes" ]; then
	LOAD_ULP_MODULES="$LOAD_ULP_MODULES ib_srp"
    fi

    if [ "${ISER_LOAD}" == "yes" ]; then
	LOAD_ULP_MODULES="$LOAD_ULP_MODULES ib_iser"
    fi
else
    LOAD_ULP_MODULES="ib_ipoib"
fi

UNLOAD_ULP_MODULES="ib_iser ib_srp rds ib_ipoib"
UNLOAD_HW_MODULES="iw_c2 iw_cxgb3 iw_nes ib_ehca ib_ipath ib_mthca mlx4_ib"
UNLOAD_CORE_USER_MODULES="rdma_ucm ib_ucm ib_uverbs ib_umad"
UNLOAD_CORE_CM_MODULES="rdma_cm ib_cm iw_cm"
UNLOAD_CORE_MODULES="ib_addr ib_sa ib_mad ib_core"

interfaces=`ip link show | grep ": ib[0-9]*" | cut -f 2 -d ' ' | sed -e 'y/\r/ /;s/://'`

# If module $1 is loaded return - 0 else - 1
is_module()
{
    /sbin/lsmod | grep -w "$1" > /dev/null 2>&1
    return $?    
}

load_modules()
{
    local RC=0

    for module in $*; do
	if ! is_module $module; then
	    /sbin/modprobe $module
	    res=$?
	    RC=$[ $RC + $res ]
	    if [ $res -ne 0 ]; then
		echo
		echo -n "Failed to load module $mod"
	    fi
	fi
    done
    return $RC
}

unload_module()
{
    local mod=$1
    # Unload module $1
    if is_module $mod; then
	/sbin/rmmod $mod > /dev/null 2>&1
	if [ $? -ne 0 ]; then
	    echo
	    echo "Failed to unload $mod"
	    return 1
	fi
    fi
    return 0
}

# This function is a horrible hack to work around BIOS authors that should
# be shot.  Specifically, certain BIOSes will map the entire 4GB address
# space as write-back cacheable when the machine has 4GB or more of RAM, and
# then they will exclude the reserved PCI I/O addresses from that 4GB
# cacheable mapping by making on overlapping uncacheable mapping.  However,
# once you do that, it is then impossible to set *any* of the PCI I/O
# address space as write-combining.  This is an absolute death-knell to
# certain IB hardware.  So, we unroll this mapping here.  Instead of
# punching a hole in a single 4GB mapping, we redo the base 4GB mapping as
# a series of discreet mappings that effectively are the same as the 4GB
# mapping minus the hole, and then we delete the uncacheable mappings that
# are used to punch the hole.  This then leaves the PCI I/O address space
# unregistered (which defaults it to uncacheable), but available for
# write-combining mappings where needed.
check_mtrr_registers()
{
    # If we actually change the mtrr registers, then the awk script will
    # return true, and we need to unload the ib_ipath module if it's already
    # loaded.  The udevtrigger in load_hardware_modules will immediately
    # reload the ib_ipath module for us, so there shouldn't be a problem.
    [ -f /proc/mtrr -a -f /etc/rdma/fixup-mtrr.awk ] && 
	awk -f /etc/rdma/fixup-mtrr.awk /proc/mtrr 2>/dev/null &&
	if is_module ib_ipath; then
		/sbin/rmmod ib_ipath
	fi
}

load_hardware_modules()
{
    local -i RC=0

    [ "$FIXUP_MTRR_REGS" = "yes" ] && check_mtrr_registers
    # WARNING!!  If you are using this script to take down and bring up
    # your IB interfaces on a machine that uses more than one low level
    # Infiniband hardware driver, then there is no guarantee that the
    # ordering of rdma interfaces after you take down and bring up the
    # stack will be the same as the ordering of the interfaces on a
    # clean boot.
    #
    # We match both class NETWORK and class INFINIBAND devices since our
    # iWARP hardware is listed under class NETWORK.  The side effect of
    # this is that we might cause a non-iWARP network driver to be loaded.
    udevadm trigger --subsystem-match=pci --attr-nomatch=driver --attr-match=class=0x020000 --attr-match=class=0x0c0600
    udevadm settle
    if [ -r /proc/device-tree ]; then
	if [ -n "`ls /proc/device-tree | grep lhca`" ]; then
	    if ! is_module ib_ehca; then
		load_modules ib_ehca
		RC+=$?
	    fi
	fi
    fi
    if is_module cxgb3 -a ! is_module iw_cxgb3; then
	load_modules iw_cxgb3
	RC+=$?
    fi
    if is_module mlx4_core -a ! is_module mlx4_ib; then
	load_modules mlx4_ib
	RC+=$?
    fi
    return $RC
}

errata_58()
{
    # Check AMD chipset issue Errata #58
    if test -x /sbin/lspci && test -x /sbin/setpci; then
	if ( /sbin/lspci -nd 1022:1100 | grep "1100" > /dev/null ) &&
	   ( /sbin/lspci -nd 1022:7450 | grep "7450" > /dev/null ) &&
	   ( /sbin/lspci -nd 15b3:5a46 | grep "5a46" > /dev/null ); then
	    CURVAL=`/sbin/setpci -d 1022:1100 69`
	    for val in $CURVAL
	    do
		if [ "${val}" != "c0" ]; then
		    /sbin/setpci -d 1022:1100 69=c0
		    if [ $? -eq 0 ]; then
			break
		    else
			echo "Failed to apply AMD-8131 Errata #58 workaround"
		    fi
		fi
	    done
	fi
    fi
}

errata_56()
{
    # Check AMD chipset issue Errata #56
    if test -x /sbin/lspci && test -x /sbin/setpci; then
	if ( /sbin/lspci -nd 1022:1100 | grep "1100" > /dev/null ) &&
	   ( /sbin/lspci -nd 1022:7450 | grep "7450" > /dev/null ) &&
	   ( /sbin/lspci -nd 15b3:5a46 | grep "5a46" > /dev/null ); then
	    bus=""
	    # Look for devices AMD-8131
	    for dev in `/sbin/setpci -v -f -d 1022:7450 19 | cut -d':' -f1,2`
	    do
		bus=`/sbin/setpci -s $dev 19`
		rev=`/sbin/setpci -s $dev 8`
		# Look for Tavor attach to secondary bus of this devices
		for device in `/sbin/setpci -f -s $bus: -d 15b3:5a46 19`
		do
		    if [ $rev -lt 13 ]; then
			/sbin/setpci -d 15b3:5a44 72=14
			if [ $? -eq 0 ]; then
			    break
			else
			    echo
			    echo "Failed to apply AMD-8131 Errata #56 workaround"
			fi
		    else
			continue
		    fi
		    # If more than one device is on the bus the issue a
		    # warning
		    num=`/sbin/setpci -f -s $bus: 0 | wc -l |  sed 's/\ *//g'`
		    if [ $num -gt 1 ]; then
			echo "Warning: your current PCI-X configuration might be incorrect."
			echo "see AMD-8131 Errata 56 for more details."
		    fi
		done
	    done
	fi
    fi
}

start()
{
    local RC=0
    local loaded=0

    echo -n "Loading OpenIB kernel modules:"

    load_hardware_modules
    RC=$[ $RC + $? ]
    load_modules $LOAD_CORE_MODULES
    RC=$[ $RC + $? ]
    load_modules $LOAD_CORE_CM_MODULES
    RC=$[ $RC + $? ]
    load_modules $LOAD_CORE_USER_MODULES
    RC=$[ $RC + $? ]
    load_modules $LOAD_ULP_MODULES
    RC=$[ $RC + $? ]
   
    # Add node description to sysfs
    IBSYSDIR="/sys/class/infiniband"
    if [ -d ${IBSYSDIR} ]; then
	declare -i hca_id=1
	for hca in ${IBSYSDIR}/*
	do
	    if [ -w ${hca}/node_desc ]; then
	    	echo -n "$(hostname | cut -f 1 -d .) HCA-${hca_id}" >> ${hca}/node_desc 2> /dev/null
	    fi
	    let hca_id++
	done
    fi
   
    errata_58
    errata_56
    
    touch /var/lock/subsys/rdma
    [ $RC -eq 0 ] && echo_success || echo_failure
    echo
    return $RC    
}

stop()
{
    # Check if applications which use infiniband are running
    local apps="opensm osmtest srp_daemon"
    local pid
    local RC=0
    
    echo -n "Unloading OpenIB kernel modules:"

    for app in $apps
    do
    	if ( ps -ef | grep $app | grep -v grep > /dev/null 2>&1 ); then
	    echo
	    echo "Found $app running."
	    echo "Please stop all RDMA applications before downing the stack."
	    echo_failure
	    echo
	    return 1
	fi
    done

    if is_module svcrdma; then
    	echo "NFSoRDMA support is still enabled."
	echo "Please stop the nfs-rdma service before stopping the rdma service."
	echo_failure
	echo
	return 1
    fi

    if ! is_module ib_core; then
	# Nothing to do, make sure lock file is gone and return
	rm -f /var/lock/subsys/rdma
	echo_success
	echo
	return 0
    fi

    # Down all IPoIB interfaces
    if is_module ib_ipoib; then
	for i in $interfaces
	do
	    . /etc/sysconfig/network-scripts/ifcfg-$i
	    [ "${SLAVE}" = yes -a "${MASTER}" ] && ifdown ${MASTER} >/dev/null 2>&1
	    ifdown $i > /dev/null 2>&1
	done    
    fi
    # Small sleep to let the ifdown settle before we remove any modules
    sleep 1
	
    # Unload OpenIB modules
    MODULES="$UNLOAD_ULP_MODULES $UNLOAD_CORE_USER_MODULES"
    MODULES="$MODULES $UNLOAD_CORE_CM_MODULES"
    for mod in $MODULES
    do
	unload_module $mod
	RC=$[ $RC + $? ]
    done
    # Insert a sleep here for all the ULP modules to have been fully removed
    # before proceeding to unload the driver modules
    sleep 1
    MODULES="$UNLOAD_HW_MODULES $UNLOAD_CORE_MODULES" 
    for mod in $MODULES
    do
	unload_module $mod
	RC=$[ $RC + $? ]
    done

    rm -f /var/lock/subsys/rdma
    [ $RC -eq 0 ] && echo_success || echo_failure
    echo
    return $RC
}

status()
{
    local -i cnt=0
    local -i modules=0
    local module=""

    echo -ne "Low level hardware support loaded:\n\t"
    for module in $UNLOAD_HW_MODULES; do
	if is_module $module; then
	    echo -n "$module "
	    let cnt++
	fi
    done
    [ $cnt -eq 0 ] && echo -n "none found"
    modules+=cnt
    echo
    echo

    echo -ne "Upper layer protocol modules:\n\t"
    cnt=0
    for module in $UNLOAD_ULP_MODULES; do
	if is_module $module; then
	    echo -n "$module "
	    let cnt++
	fi
    done
    [ $cnt -eq 0 ] && echo -n "none found"
    modules+=cnt
    echo
    echo

    echo -ne "User space access modules:\n\t"
    cnt=0
    for module in $UNLOAD_CORE_USER_MODULES; do
	if is_module $module; then
	    echo -n "$module "
	    let cnt++
	fi
    done
    [ $cnt -eq 0 ] && echo -n "none found"
    modules+=cnt
    echo
    echo

    echo -ne "Connection management modules:\n\t"
    cnt=0
    for module in $UNLOAD_CORE_CM_MODULES; do
	if is_module $module; then
	    echo -n "$module "
	    let cnt++
	fi
    done
    [ $cnt -eq 0 ] && echo -n "none found"
    modules+=cnt
    echo
    echo

    for module in $UNLOAD_CORE_MODULES; do
	if is_module $module; then
	    let modules++
	fi
    done

    if is_module ib_ipoib; then
	echo -n "Configured IPoIB interfaces: "
	cnt=0
	for i in /etc/sysconfig/network-scripts/ifcfg-ib*
	do
	    if [ -f $i ]; then
		. $i
		echo -n "$DEVICE "
		let cnt++
	    fi
	done
	[ $cnt -eq 0 ] && echo -n "none"
	echo
	echo -n "Currently active IPoIB interfaces: "
	cnt=0
	for i in $interfaces
	do
	    ip link show $i | grep -w UP > /dev/null 2>&1
	    [ $? -eq 0 ] && echo -n "$i " && let cnt++
	done
	[ $cnt -eq 0 ] && echo -n "none"
	echo
    fi
    
    if [ $modules -eq 0 ]; then
	if [ -f /var/lock/subsys/rdma ]; then
	    return 2
	else
	    return 3
	fi
    else
	return 0
    fi
}

restart ()
{
    stop
    start
}

condrestart ()
{
    [ -e /var/lock/subsys/rdma ] && restart || return 0
}

usage ()
{
    echo
    echo "Usage: `basename $0` {start|stop|restart|condrestart|try-restart|force-reload|status}"
    echo
    return 2
}

case $1 in
    start|stop|restart|condrestart|try-restart|force-reload)
	[ `id -u` != "0" ] && exit 4 ;;
esac

case $1 in
    start) start; RC=$? ;;
    stop) stop; RC=$? ;;
    restart) restart; RC=$? ;;
    reload) RC=3 ;;
    condrestart) condrestart; RC=$? ;;
    try-restart) condrestart; RC=$? ;;
    force-reload) condrestart; RC=$? ;;
    status) status; RC=$? ;;
    *) usage; RC=$? ;;
esac

exit $RC