rpms/rdma/F-8 rdma-90-rdma.rules, NONE, 1.1 rdma-fixup-mtrr.awk, NONE, 1.1 rdma.conf, NONE, 1.1 rdma.init, NONE, 1.1 rdma.spec, NONE, 1.1

[Doug Ledford (dledford)]

Author: dledford

Update of /cvs/extras/rpms/rdma/F-8
In directory cvs-int.fedora.redhat.com:/tmp/cvs-serv21869

Added Files:
	rdma-90-rdma.rules rdma-fixup-mtrr.awk rdma.conf rdma.init 
	rdma.spec 
Log Message:
* Mon Jun 09 2008 Doug Ledford <dledford at redhat.com> - 1.0-2
- Attempt to use --subsystem-match=infiniband in the rdma init script use
  of udevtrigger so we don't trigger the whole system
- Add a requirement to stop opensm to the init script



--- NEW FILE rdma-90-rdma.rules ---
KERNEL=="umad*", NAME="infiniband/%k"
KERNEL=="issm*", NAME="infiniband/%k"
KERNEL=="ucm*", NAME="infiniband/%k", MODE="0666"
KERNEL=="uverbs*", NAME="infiniband/%k", MODE="0666"
KERNEL=="uat", NAME="infiniband/%k", MODE="0666"
KERNEL=="ucma", NAME="infiniband/%k", MODE="0666"
KERNEL=="rdma_cm", NAME="infiniband/%k", MODE="0666"


--- NEW FILE rdma-fixup-mtrr.awk ---
# This is a simple script that checks the contents of /proc/mtrr to see if
# the BIOS maker for the computer took the easy way out in terms of
# specifying memory regions when there is a hole below 4GB for PCI access
# and the machine has 4GB or more of RAM.  When the contents of /proc/mtrr
# show a 4GB mapping of write-back cached RAM, minus punch out hole(s) of
# uncacheable regions (the area reserved for PCI access), then it becomes
# impossible for the ib_ipath driver to set write_combining on its PIO
# buffers.  To correct the problem, remap the lower memory region in various
# chunks up to the start of the punch out hole(s), then delete the punch out
# hole(s) entirely as they aren't needed any more.  That way, ib_ipath will
# be able to set write_combining on its PIO memory access region.

BEGIN {
	regs = 0
}

function check_base(mem)
{
	printf "Base memory data: base=0x%08x, size=0x%x\n", base[mem], size[mem] > "/dev/stderr"
	if (size[mem] < (512 * 1024 * 1024))
		return 0
	if (type[mem] != "write-back")
		return 0
	if (base[mem] >= (4 * 1024 * 1024 * 1024))
		return 0
	return 1
}

function check_hole(hole)
{
	printf "Hole data: base=0x%08x, size=0x%x\n", base[hole], size[hole] > "/dev/stderr"
	if (size[hole] > (1 * 1024 * 1024 * 1024))
		return 0
	if (type[hole] != "uncachable")
		return 0
	if ((base[hole] + size[hole]) > (4 * 1024 * 1024 * 1024))
		return 0
	return 1
}

function build_entries(start, end,     new_base, new_size, tmp_base)
{
	# mtrr registers require alignment of blocks, so a 256MB chunk must
	# be 256MB aligned.  Additionally, all blocks must be a power of 2
	# in size.  So, do the largest power of two size that we can and
	# still have start + block <= end, rinse and repeat.
	tmp_base = start
	do {
		new_base = tmp_base
		new_size = 4096
		while (((new_base + new_size) < end) &&
		       ((new_base % new_size) == 0))
			new_size = lshift(new_size, 1)
		if (((new_base + new_size) > end) ||
		    ((new_base % new_size) != 0))
			new_size = rshift(new_size, 1)
		printf "base=0x%x size=0x%x type=%s\n",
			new_base, new_size, type[mem] > "/dev/stderr"
		printf "base=0x%x size=0x%x type=%s\n",
			new_base, new_size, type[mem] > "/proc/mtrr"
		fflush("")
		tmp_base = new_base + new_size
	} while (tmp_base < end)
}

{
	gsub("^reg", "")
	gsub(": base=", " ")
	gsub(" [(].*), size=", " ")
	gsub(": ", " ")
	gsub(", count=.*$", "")
	register[regs] = strtonum($1)
	base[regs] = strtonum($2)
	size[regs] = strtonum($3)
	human_size[regs] = size[regs]
	if (match($3, "MB")) { size[regs] *= 1024*1024; mult[regs] = "MB" }
	else { size[regs] *= 1024; mult[regs] = "KB" }
	type[regs] = $4
	enabled[regs] = 1
	end[regs] = base[regs] + size[regs]
	regs++
}

END {
	# First we need to find our base memory region.  We only care about
	# the memory register that starts at base 0.  This is the only one
	# that we can reliably know is our global memory region, and the
	# only one that we can reliably check against overlaps.  It's entirely
	# possible that any memory region not starting at 0 and having an
	# overlap with another memory region is in fact intentional and we
	# shouldn't touch it.
	for(i=0; i<regs; i++)
		if (base[i] == 0)
			break
	# Did we get a valid base register?
	if (i == regs)
		exit 1
	mem = i
	if (!check_base(mem))
		exit 1

	cur_hole = 0
	for(i=0; i<regs; i++) {
		if (i == mem)
			continue
		if (base[i] < end[mem] && check_hole(i))
			holes[cur_hole++] = i
	}
	if (cur_hole == 0) {
		print "Nothing to do" > "/dev/stderr"
		exit 1
	}
	printf "Found %d punch-out holes\n", cur_hole > "/dev/stderr"

	# We need to sort the holes according to base address
	for(j = 0; j < cur_hole - 1; j++) {
		for(i = cur_hole - 1; i > j; i--) {
			if(base[holes[i]] < base[holes[i-1]]) {
				tmp = holes[i]
				holes[i] = holes[i-1]
				holes[i-1] = tmp
			}
		}
	}
	# OK, the common case would be that the BIOS is mapping holes out
	# of the 4GB memory range, and that our hole(s) are consecutive and
	# that our holes and our memory region end at the same place.  However,
	# things like machines with 8GB of RAM or more can foul up these
	# common traits.
	#
	# So, our modus operandi is to disable all of the memory/hole regions
	# to start, then build new base memory zones that in the end add
	# up to the same as our original zone minus the holes.  We know that
	# we will never have a hole listed here that belongs to a valid
	# hole punched in a write-combining memory region because you can't
	# overlay write-combining on top of write-back and we know our base
	# memory region is write-back, so in order for this hole to overlap
	# our base memory region it can't be also overlapping a write-combining
	# region.
	printf "disable=%d\n", register[mem] > "/dev/stderr"
	printf "disable=%d\n", register[mem] > "/proc/mtrr"
	fflush("")
	enabled[mem] = 0
	for(i=0; i < cur_hole; i++) {
		printf "disable=%d\n", register[holes[i]] > "/dev/stderr"
		printf "disable=%d\n", register[holes[i]] > "/proc/mtrr"
		fflush("")
		enabled[holes[i]] = 0
	}
	build_entries(base[mem], base[holes[0]])
	for(i=0; i < cur_hole - 1; i++)
		if (base[holes[i+1]] > end[holes[i]])
			build_entries(end[holes[i]], base[holes[i+1]])
	if (end[mem] > end[holes[i]])
		build_entries(end[holes[i]], end[mem])
	# We changed up the mtrr regs, so signal to the rdma script to
	# reload modules that need the mtrr regs to be right.
	exit 0
}



--- NEW FILE rdma.conf ---
# Load IPoIB
IPOIB_LOAD=yes
# Load SRP module
SRP_LOAD=no
# Load iSER module
ISER_LOAD=no
# Load QLogic VNIC module
QLGC_VNIC_LOAD=no
# Should we modify the system mtrr registers?  We may need to do this if you
# get messages from the ib_ipath driver saying that it couldn't enable
# write combining for the PIO buffs on the card.
FIXUP_MTRR_REGS=no


--- NEW FILE rdma.init ---
#!/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-Stop: $network $srpd $opensm
# 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 [ "${IPOIB_LOAD}" == "yes" ]; then
	LOAD_ULP_MODULES="ib_ipoib"
    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 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=`/sbin/ifconfig | grep "^ib[0-9]*" | cut -f 1 -d ' ' | sed -e 'y/\r/ /'`

# 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

    [ "$FIXUP_MTRR_REGS" = "yes" ] && check_mtrr_registers
    # WARNING!!  If you are using this script to take down and bring up
    # your IB interfaces on a machine that uses more than one low level
    # Infiniband hardware driver, then there is no guarantee that the
    # ordering of rdma interfaces after you take down and bring up the
    # stack will be the same as the ordering of the interfaces on a
    # clean boot.
    #
    # We match both class NETWORK and class INFINIBAND devices since our
    # iWARP hardware is listed under class NETWORK.  The side effect of
    # this is that we might cause a non-iWARP network driver to be loaded.
    udevtrigger --subsystem-match=pci --attr-nomatch=driver --attr-match=class=0x020000 --attr-match=class=0x0c0600
    udevsettle
    if [ -r /proc/device-tree ]; then
	if [ -n "`ls /proc/device-tree | grep lhca`" ]; then
	    if ! is_module ib_ehca; then
		load_modules ib_ehca
		RC+=$?
	    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 qlgc_vnic; then
	echo
	echo "The qlgc_vnic service is still running."
	echo "Please stop all RDMA applications before downing the stack."
	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
	    ifdown $i > /dev/null 2>&1
	done    
    fi
	
    # 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
	    ifconfig $i | grep 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 $1 in
    start) start; RC=$? ;;
    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


--- NEW FILE rdma.spec ---
#  Copyright (c) 2008 Red Hat, Inc.

#  There is no URL or upstream source entry as this package constitutes
#  upstream for itself.

Summary: Infiniband/iWARP Kernel Module Initializer
Name: rdma
Version: 1.0
Release: 2%{?dist}
License: GPLv2+
Group: System Environment/Base
Source0: rdma.conf
Source1: rdma.init
Source2: rdma-fixup-mtrr.awk
Source3: rdma-90-rdma.rules
BuildRoot: %{_tmppath}/%{name}-%{version}-%{release}-root-%(%{__id_u} -n)
BuildArch: noarch
Requires(post): chkconfig
Requires(preun): chkconfig
Requires: udev >= 095
%description 
User space initialization scripts for the kernel InfiniBand/iWARP drivers

%prep

%build

%install
rm -rf ${RPM_BUILD_ROOT}
install -d ${RPM_BUILD_ROOT}%{_initrddir}
install -d ${RPM_BUILD_ROOT}%{_sysconfdir}/%{name}
install -d ${RPM_BUILD_ROOT}%{_sysconfdir}/udev/rules.d
	
install -m 0644 %{SOURCE0} ${RPM_BUILD_ROOT}%{_sysconfdir}/%{name}/%{name}.conf
install -m 0755 %{SOURCE1} ${RPM_BUILD_ROOT}%{_initrddir}/%{name}
install -m 0644 %{SOURCE2} ${RPM_BUILD_ROOT}%{_sysconfdir}/%{name}/fixup-mtrr.awk
install -m 0644 %{SOURCE3} ${RPM_BUILD_ROOT}%{_sysconfdir}/udev/rules.d/90-%{name}.rules

%clean
rm -rf ${RPM_BUILD_ROOT}

%post
if [ $1 = 1 ]; then
    /sbin/chkconfig --add %{name} 
fi

%preun
if [ $1 = 0 ]; then
    /sbin/chkconfig --del %{name}
fi

%files
%defattr(-,root,root,-)
%dir %{_sysconfdir}/%{name}
%config(noreplace) %{_sysconfdir}/%{name}/%{name}.conf
%{_sysconfdir}/%{name}/fixup-mtrr.awk
%{_initrddir}/%{name}
%{_sysconfdir}/udev/rules.d/90-%{name}.rules

%changelog
* Mon Jun 09 2008 Doug Ledford <dledford at redhat.com> - 1.0-2
- Attempt to use --subsystem-match=infiniband in the rdma init script use
  of udevtrigger so we don't trigger the whole system
- Add a requirement to stop opensm to the init script

* Sun Jun 08 2008 Doug Ledford <dledford at redhat.com> - 1.0-1
- Create an initial package for Fedora review