In case there are lingering rxe devices present on the system, adding them again under the infiniband subsystem will simply fail the test. To avoid such scenario, make sure that we start the rxe with a clean slate by removing all soft devices first. This is mostly relevant for systems running jobs which init the rxe but don't run tests which call to revert_soft_roce() for cleanup (see BlobFS-autotest as an example). Change-Id: I12997fbaf7343ae3e9bc0b38f5455f6332c4e6c5 Signed-off-by: Michal Berger <michalx.berger@intel.com> Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3823 Community-CI: Mellanox Build Bot Community-CI: Broadcom CI Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
299 lines
6.4 KiB
Bash
299 lines
6.4 KiB
Bash
NVMF_PORT=4420
|
|
NVMF_SECOND_PORT=4421
|
|
NVMF_THIRD_PORT=4422
|
|
NVMF_IP_PREFIX="192.168.100"
|
|
NVMF_IP_LEAST_ADDR=8
|
|
NVMF_TCP_IP_ADDRESS="127.0.0.1"
|
|
NVMF_TRANSPORT_OPTS=""
|
|
NVMF_SERIAL=SPDK00000000000001
|
|
|
|
function build_nvmf_app_args() {
|
|
if [ $SPDK_RUN_NON_ROOT -eq 1 ]; then
|
|
NVMF_APP=(sudo -u "$USER" "${NVMF_APP[@]}")
|
|
NVMF_APP+=(-i "$NVMF_APP_SHM_ID" -e 0xFFFF)
|
|
else
|
|
NVMF_APP+=(-i "$NVMF_APP_SHM_ID" -e 0xFFFF)
|
|
fi
|
|
}
|
|
|
|
: ${NVMF_APP_SHM_ID="0"}
|
|
export NVMF_APP_SHM_ID
|
|
build_nvmf_app_args
|
|
|
|
have_pci_nics=0
|
|
|
|
function rxe_cfg() {
|
|
"$rootdir/scripts/rxe_cfg_small.sh" "$@"
|
|
}
|
|
|
|
function load_ib_rdma_modules() {
|
|
if [ $(uname) != Linux ]; then
|
|
return 0
|
|
fi
|
|
|
|
modprobe ib_cm
|
|
modprobe ib_core
|
|
# Newer kernels do not have the ib_ucm module
|
|
modprobe ib_ucm || true
|
|
modprobe ib_umad
|
|
modprobe ib_uverbs
|
|
modprobe iw_cm
|
|
modprobe rdma_cm
|
|
modprobe rdma_ucm
|
|
}
|
|
|
|
function detect_soft_roce_nics() {
|
|
rxe_cfg stop # make sure we run tests with a clean slate
|
|
rxe_cfg start
|
|
}
|
|
|
|
# args 1 and 2 represent the grep filters for finding our NICS.
|
|
# subsequent args are all drivers that should be loaded if we find these NICs.
|
|
# Those drivers should be supplied in the correct order.
|
|
function detect_nics_and_probe_drivers() {
|
|
NIC_VENDOR="$1"
|
|
NIC_CLASS="$2"
|
|
|
|
nvmf_nic_bdfs=$(lspci | grep Ethernet | grep "$NIC_VENDOR" | grep "$NIC_CLASS" | awk -F ' ' '{print "0000:"$1}')
|
|
|
|
if [ -z "$nvmf_nic_bdfs" ]; then
|
|
return 0
|
|
fi
|
|
|
|
have_pci_nics=1
|
|
if [ $# -ge 2 ]; then
|
|
# shift out the first two positional arguments.
|
|
shift 2
|
|
# Iterate through the remaining arguments.
|
|
for i; do
|
|
modprobe "$i"
|
|
done
|
|
fi
|
|
}
|
|
|
|
function detect_pci_nics() {
|
|
|
|
if ! hash lspci; then
|
|
return 0
|
|
fi
|
|
|
|
detect_nics_and_probe_drivers "Mellanox" "ConnectX-4" "mlx4_core" "mlx4_ib" "mlx4_en"
|
|
detect_nics_and_probe_drivers "Mellanox" "ConnectX-5" "mlx5_core" "mlx5_ib"
|
|
detect_nics_and_probe_drivers "Intel" "X722" "i40e" "i40iw"
|
|
detect_nics_and_probe_drivers "Chelsio" "Unified Wire" "cxgb4" "iw_cxgb4"
|
|
|
|
if [ "$have_pci_nics" -eq "0" ]; then
|
|
return 0
|
|
fi
|
|
|
|
# Provide time for drivers to properly load.
|
|
sleep 5
|
|
}
|
|
|
|
function detect_rdma_nics() {
|
|
detect_pci_nics
|
|
if [ "$have_pci_nics" -eq "0" ]; then
|
|
detect_soft_roce_nics
|
|
fi
|
|
}
|
|
|
|
function allocate_nic_ips() {
|
|
((count = NVMF_IP_LEAST_ADDR))
|
|
for nic_name in $(get_rdma_if_list); do
|
|
ip="$(get_ip_address $nic_name)"
|
|
if [ -z $ip ]; then
|
|
ip addr add $NVMF_IP_PREFIX.$count/24 dev $nic_name
|
|
ip link set $nic_name up
|
|
((count = count + 1))
|
|
fi
|
|
# dump configuration for debug log
|
|
ip addr show $nic_name
|
|
done
|
|
}
|
|
|
|
function get_available_rdma_ips() {
|
|
for nic_name in $(get_rdma_if_list); do
|
|
get_ip_address $nic_name
|
|
done
|
|
}
|
|
|
|
function get_rdma_if_list() {
|
|
for nic_type in /sys/class/infiniband/*; do
|
|
[[ -e "$nic_type" ]] || break
|
|
for nic_name in /sys/class/infiniband/"$(basename ${nic_type})"/device/net/*; do
|
|
[[ -e "$nic_name" ]] || break
|
|
basename "$nic_name"
|
|
done
|
|
done
|
|
}
|
|
|
|
function get_ip_address() {
|
|
interface=$1
|
|
ip -o -4 addr show $interface | awk '{print $4}' | cut -d"/" -f1
|
|
}
|
|
|
|
function nvmfcleanup() {
|
|
sync
|
|
set +e
|
|
for i in {1..20}; do
|
|
modprobe -v -r nvme-$TEST_TRANSPORT
|
|
if modprobe -v -r nvme-fabrics; then
|
|
set -e
|
|
return 0
|
|
fi
|
|
sleep 1
|
|
done
|
|
set -e
|
|
|
|
# So far unable to remove the kernel modules. Try
|
|
# one more time and let it fail.
|
|
# Allow the transport module to fail for now. See Jim's comment
|
|
# about the nvme-tcp module below.
|
|
modprobe -v -r nvme-$TEST_TRANSPORT || true
|
|
modprobe -v -r nvme-fabrics
|
|
}
|
|
|
|
function nvmftestinit() {
|
|
if [ -z $TEST_TRANSPORT ]; then
|
|
echo "transport not specified - use --transport= to specify"
|
|
return 1
|
|
fi
|
|
if [ "$TEST_MODE" == "iso" ]; then
|
|
$rootdir/scripts/setup.sh
|
|
if [ "$TEST_TRANSPORT" == "rdma" ]; then
|
|
rdma_device_init
|
|
fi
|
|
fi
|
|
|
|
NVMF_TRANSPORT_OPTS="-t $TEST_TRANSPORT"
|
|
if [ "$TEST_TRANSPORT" == "rdma" ]; then
|
|
RDMA_IP_LIST=$(get_available_rdma_ips)
|
|
NVMF_FIRST_TARGET_IP=$(echo "$RDMA_IP_LIST" | head -n 1)
|
|
NVMF_SECOND_TARGET_IP=$(echo "$RDMA_IP_LIST" | tail -n +2 | head -n 1)
|
|
if [ -z $NVMF_FIRST_TARGET_IP ]; then
|
|
echo "no NIC for nvmf test"
|
|
exit 0
|
|
fi
|
|
elif [ "$TEST_TRANSPORT" == "tcp" ]; then
|
|
NVMF_FIRST_TARGET_IP=127.0.0.1
|
|
NVMF_TRANSPORT_OPTS="$NVMF_TRANSPORT_OPTS -o"
|
|
fi
|
|
|
|
# currently we run the host/perf test for TCP even on systems without kernel nvme-tcp
|
|
# support; that's fine since the host/perf test uses the SPDK initiator
|
|
# maybe later we will enforce modprobe to succeed once we have systems in the test pool
|
|
# with nvme-tcp kernel support - but until then let this pass so we can still run the
|
|
# host/perf test with the tcp transport
|
|
modprobe nvme-$TEST_TRANSPORT || true
|
|
}
|
|
|
|
function nvmfappstart() {
|
|
timing_enter start_nvmf_tgt
|
|
"${NVMF_APP[@]}" "$@" &
|
|
nvmfpid=$!
|
|
trap 'process_shm --id $NVMF_APP_SHM_ID; nvmftestfini; exit 1' SIGINT SIGTERM EXIT
|
|
waitforlisten $nvmfpid
|
|
timing_exit start_nvmf_tgt
|
|
}
|
|
|
|
function nvmftestfini() {
|
|
nvmfcleanup || :
|
|
if [ -n "$nvmfpid" ]; then
|
|
killprocess $nvmfpid
|
|
fi
|
|
if [ "$TEST_MODE" == "iso" ]; then
|
|
$rootdir/scripts/setup.sh reset
|
|
if [ "$TEST_TRANSPORT" == "rdma" ]; then
|
|
rdma_device_init
|
|
fi
|
|
fi
|
|
}
|
|
|
|
function rdma_device_init() {
|
|
load_ib_rdma_modules
|
|
detect_rdma_nics
|
|
allocate_nic_ips
|
|
}
|
|
|
|
function revert_soft_roce() {
|
|
rxe_cfg stop
|
|
}
|
|
|
|
function check_ip_is_soft_roce() {
|
|
if [ "$TEST_TRANSPORT" != "rdma" ]; then
|
|
return 0
|
|
fi
|
|
rxe_cfg status rxe | grep -wq "$1"
|
|
}
|
|
|
|
function nvme_connect() {
|
|
local init_count
|
|
init_count=$(nvme list | wc -l)
|
|
|
|
if ! nvme connect "$@"; then return $?; fi
|
|
|
|
for i in $(seq 1 10); do
|
|
if [ $(nvme list | wc -l) -gt $init_count ]; then
|
|
return 0
|
|
else
|
|
sleep 1s
|
|
fi
|
|
done
|
|
return 1
|
|
}
|
|
|
|
function get_nvme_devs() {
|
|
local dev rest
|
|
|
|
nvmes=()
|
|
while read -r dev rest; do
|
|
if [[ $dev == /dev/nvme* ]]; then
|
|
nvmes+=("$dev")
|
|
fi
|
|
if [[ $1 == print ]]; then
|
|
echo "$dev $rest"
|
|
fi
|
|
done < <(nvme list)
|
|
((${#nvmes[@]})) || return 1
|
|
echo "${#nvmes[@]}" >&2
|
|
}
|
|
|
|
function gen_nvmf_target_json() {
|
|
local subsystem config=()
|
|
|
|
for subsystem in "${@:-1}"; do
|
|
config+=(
|
|
"$(
|
|
cat <<- EOF
|
|
{
|
|
"params": {
|
|
"name": "Nvme$subsystem",
|
|
"trtype": "$TEST_TRANSPORT",
|
|
"traddr": "$NVMF_FIRST_TARGET_IP",
|
|
"adrfam": "ipv4",
|
|
"trsvcid": "$NVMF_PORT",
|
|
"subnqn": "nqn.2016-06.io.spdk:cnode$subsystem"
|
|
},
|
|
"method": "bdev_nvme_attach_controller"
|
|
}
|
|
EOF
|
|
)"
|
|
)
|
|
done
|
|
jq . <<- JSON
|
|
{
|
|
"subsystems": [
|
|
{
|
|
"subsystem": "bdev",
|
|
"config": [
|
|
$(
|
|
IFS=","
|
|
printf '%s\n' "${config[*]}"
|
|
)
|
|
]
|
|
}
|
|
]
|
|
}
|
|
JSON
|
|
}
|