#!/bin/bash

# Clean: shellcheck -e 2001 ./tcwg-benchmark.sh

set -eux

scripts=$(dirname "$0")
# shellcheck source=jenkins-helpers.sh
. $scripts/jenkins-helpers.sh

convert_args_to_variables "$@"
obligatory_variables \
    boardname \
    image_arch \
    toolchain_url \
    bench_list \
    cflags \
    ldflags \
    extension \
    testmode \
    iterations \
    run_profile \
    sysroot \
    forceinstall \
    builder \
    results_id \
    BUILD_NUMBER \
    WORKSPACE \
    reboot \
    ignore_errors \
    clean_older_than
declare -g \
	boardname \
	image_arch \
	toolchain_url \
	bench_list \
	cflags \
	ldflags \
	extension \
	testmode \
	iterations \
	run_profile \
	sysroot \
	forceinstall \
	builder \
	results_id \
	BUILD_NUMBER \
	WORKSPACE \
	reboot \
	ignore_errors \
	clean_older_than

# Make shellcheck happy and workaround Jenkins not defining variables
# for empty arguments.
bench_container_tag="${bench_container_tag-bionic}"
build_container_tag="${build_container_tag-bionic}"
toolchain_type="${toolchain_type-auto}"
prepare_board="${prepare_board-true}"

if echo "$builder" | grep -q ".*-[0-9]\+"; then
    docker_host_opt="--arch amd64 --node $builder"
else
    docker_host_opt="--label $builder"
fi

# shellcheck source=jenkins-helpers.sh
. $scripts/jenkins-helpers.sh

# If $toolchain_url is of ssh:// type, don't create a remote build
# container, just use the ssh command as provided.
build_container_host=
build_container_port=
case "$toolchain_url" in
    "ssh://"*)
	ccprefix="${toolchain_url##ssh://}"

	# Extract host:port: specification from ccprefix, we don't
	# need to care about :parallelize here, just pass it to run.sh
	# if present.
	build=${ccprefix%:*}
	build_container_host="$(echo $build | cut -d: -f 1)"
	case ${ccprefix} in
	    *:*:*)
		build_container_port="$(echo $build | cut -s -d: -f 2)"
		;;
	    *:*)
		# If no port is specified, use 22 (ssh default port)
		build_container_port=22
		;;
	esac

	if [ "x$build_container_host" = "x" ]; then
	    echo "ERROR: ssh:// toolchain_url lacks a host: $toolchain_url."
	    exit 1
	fi
	if [ "x$build_container_port" = "x" ]; then
	    echo "ERROR: ssh:// toolchain_url lacks a port: $toolchain_url."
	    exit 1
	fi
	;;
    *)
	# Make sure to cleanup build container if something goes
	# wrong when preparing the test environment
	trap "cleanup_all_containers" EXIT
	$scripts/start-container-docker.sh $docker_host_opt --distro "$build_container_tag" --task build --prefix build_ > build-container.sh
	. ./build-container.sh
	;;
esac

case "$toolchain_url" in
    "ssh://"*)
	if [ x"$sysroot" = x"tarball" ]; then
            echo "ERROR: Unsupported sysroot $sysroot for toolchain_url $toolchain_url"
            exit 1
	fi
	# Last component of ccprefix is the path, keep it
	toolchaindir="$(dirname ${ccprefix##*:})"
	;;
    "http://"*".tar.xz"|"https://"*".tar.xz")
	toolchaindir=$(untar_url "$toolchain_url" "$WORKSPACE" "--strip-components 1")
	;;
    "rsync://"*)
	ccprefix="${toolchain_url##rsync://}"
	# We want to access the remote toolchain via a container, to
	# avoid problems with the hosts's ssh server restrictions on the
	# number of simulaneous connexions.
	# We copy it to the build container (assuming it uses the same
	# architecture as the machine pointed to by $toolchain_url).
	# Assume ccprefix looks like /path/bin/target-triplet-, and
	# compute 'path'.
	src_toolchaindir=$(dirname "$(dirname ${ccprefix})")
	toolchaindir="${WORKSPACE}/toolchain-${BUILD_NUMBER}"
	rsync -az --delete "$src_toolchaindir/" "$toolchaindir/"
	;;
    *)
	echo "ERROR: Cannot handle toolchain_url: $toolchain_url"
	exit 1
	;;
esac

# Sanity check that toolchain_type is supported
case "$toolchain_type" in
    gnu|llvm) ;;
    *)
	echo "ERROR: Unsupported toolchain type: $toolchain_type"
	exit 1
	;;
esac

case "$toolchain_url" in
    "http://"*|"https://"*|"rsync://"*|"ssh://"*)

	# In the ssh:// case, we have to perform the 'find' operations
	# remotely.
	case "$toolchain_url" in
	    "ssh://"*)
		maybe_remote="ssh -p $build_container_port $build_container_host"
		;;
	    *)
		maybe_remote=""
		;;
	esac

	case "$toolchain_type" in
	    "gnu"|"llvm") ;;
	    "auto")
		if [ x"$($maybe_remote find "$toolchaindir" -path "*bin/*gcc" | wc -l)" != x"0" ]; then
		    toolchain_type="gnu"
		elif [ x"$($maybe_remote find "$toolchaindir" -path "*bin/*clang" | wc -l)" != x"0" ]; then
		    toolchain_type="llvm"
		else
		    echo "ERROR: Cannot autodetect toolchain type"
		    exit 1
		fi
		;;
	esac

	case "$toolchain_type" in
	    "gnu") ccname="gcc" ;;
	    "llvm") ccname="clang" ;;
	esac
	ccpath=$($maybe_remote find "$toolchaindir" -path "*bin/*$ccname")
	if [ "$(echo "$ccpath" | wc -w)" -ne 1 ]; then
	    echo "ERROR: found more than one compiler: $ccpath"
	    exit 1
	fi

	# Non-ssh:// cases have to copy the just-copied toolchain to
	# the remote build container. For ssh://, we'll access the
	# toolchain remotely.
	case "$toolchain_url" in
	    "ssh://"*) ;;
	    *)
		ccprefix=$(echo "$ccpath" | sed -e "s/$ccname\$//")
		# Copy toolchain to the build container.
		rsync -a --delete -e "ssh -p$build_container_port" "$toolchaindir/" "$build_container_host:$toolchaindir/"
		ccprefix="$build_container_host:$build_container_port:$ccprefix"
		;;
	esac
	;;
esac

case "$sysroot" in
    "tarball")
	sysroot="$build_container_host:$build_container_port:$(find "$toolchaindir" -name "libc")"
	;;
    "http://"*|"https://"*)
	sysrootdir=$(untar_url "$sysroot" "$WORKSPACE" "--strip-components 1")
	# Copy toolchain to the build container.
	rsync -a --delete -e "ssh -p$build_container_port" "$sysrootdir/" "$build_container_host:$sysrootdir/"
	sysroot="$build_container_host:$build_container_port:$sysrootdir"
	;;
    "ssh://"*)
	sysroot="${sysroot##ssh://}"

	# Check host:port specification from sysroot.
	case ${sysroot} in
	    *:*) ;;
	    *)
		echo "ERROR: ssh:// sysroot lacks a host: $sysroot"
		exit 1
		;;
	esac
	;;

    "")
	# Use system sysroot.
	;;
    *)
	echo "ERROR: Cannot handle sysroot: $sysroot"
	exit 1
	;;
esac

if echo "$results_id" | grep -q "\.\."; then
    echo "ERROR: results_id should not escape /home/tcwg-benchmark/results* hierarchy; do not use \"..\""
    exit 1
fi

hw_tag="${results_id%%/*}"
case "$hw_tag:$boardname:$image_arch" in
    sq_32:*-sq-*:armhf) ;;
    sq_64:*-sq-*:arm64) ;;
    tk1_32:*-tk1-*:armhf) ;;
    tx1_64:*-tx1-*:arm64) ;;
    tx1_32:*-tx1-*:armhf) ;;
    *)
	echo "ERROR: results_id does not start with a valid hw_tag: $hw_tag"
	exit 1
	;;
esac

# Check that we can ssh to the board and rsync scripts.  This ensures that
# the board is online and filesystem is good condition.  Try to reboot and/or
# power-cycle the board as needed.
if $reboot; then
    # 1. Try access after soft reboot
    # 2. Try access after power-cycle
    tries_left=2
else
    # 1. Try access without rebooting
    # 2. Try access after soft reboot
    # 3. Try access after power-cycle
    tries_left=3
fi
force_power_cycle=false
while [ $tries_left != 0 ]; do
    tries_left=$(($tries_left-1))

    if $reboot; then
	if ! ssh "$boardname" true || $force_power_cycle; then
	    echo "Trying to power-cycle $boardname"
	    (
		pdu_name=$(echo "${boardname%.tcwglab}" \
			       | sed -e 's/^tcwg-bmk-/tcwg-/')
		nvidia-power-cycle.sh "$pdu_name"
		wait_for_ssh_server "$boardname" 22 100
	    ) &
	    wait $! || exit $EXTERNAL_FAIL
	    echo "Successfully powered-cycled $boardname"
	else
	    # Reboot the board.
	    # Ping board every second (ServerAliveInterval=1) to avoid
	    # waiting [default] 5min for ssh to break connection.
	    ssh -Snone -oServerAliveInterval=1 $boardname sudo /sbin/reboot \
		|| true
	    # Wait until the ssh server is ready
	    sleep 30 # Give time to the board to shutdown
	    ret=0
	    wait_for_ssh_server $boardname 22 100 || ret=$?
	    if [ $ret != 0 ]; then
		echo "SSH server did not respond after reboot, exiting."
		exit $EXTERNAL_FAIL
	    fi
	fi
    fi

    rsync -az --delete bmk-scripts/ "$boardname:bmk-scripts/" &
    res=0 && wait $! || res=$?
    if [ $res = 0 ]; then
	break
    else
	reboot=true
	if [ x"$tries_left" = x"1" ]; then
	    force_power_cycle=true
	fi
    fi
done

if [ $res != 0 ]; then
    echo "ERROR: Could not get board online"
    exit $EXTERNAL_FAIL
fi

case "$testmode" in
    build|verify) input_size="test" ;;
    benchmark) input_size="ref" ;;
esac

if $prepare_board; then
    # FIXME: Implement more configurations and checks:
    # disable swap
    # set interrupt affinity
    # check that there are no stray processes
    # test that taskset works
    remote_exec "$boardname:::-t -Snone" \
		sudo /usr/local/bin/benchmark.sh --hw_tag "$hw_tag" \
		--action start_board --verbose \
		--image "linaro/ci-$image_arch-tcwg-build-ubuntu:$bench_container_tag" &

    res=0 && wait $! || res=$?
    if [ $res != 0 ]; then
	echo "ERROR: Could not prepare board for benchmarking"
	exit $EXTERNAL_FAIL
    fi
fi

# Start a container to run the benchmarks in.
# We install SPEC in /home/tcwg-benchmark, so bind-mount it as $WORKSPACE.
WORKSPACE=$HOME $scripts/start-container-docker.sh --session-host "$boardname" --arch "$image_arch" --distro "$bench_container_tag" --task bench --docker_opts "--privileged" --prefix run_ > run-container.sh &

res=0 && wait $! || res=$?
if [ $res != 0 ]; then
    echo "ERROR: Could not start benchmarking container"
    exit $EXTERNAL_FAIL
fi

trap "cleanup_all_containers" EXIT
. ./run-container.sh

# vars are from run-container.sh sourced above
# shellcheck disable=SC2154
remote_exec "$run_container_host:$run_container_port::-t -Snone" \
	    bmk-scripts/run.sh \
	    --bench "$bench_list" \
	    --config "${BUILD_NUMBER}-$run_profile" \
	    --cflags "$cflags" \
	    --ldflags "$ldflags" \
	    --ccprefix "$ccprefix" \
	    --extension "$extension" \
	    --hw_tag "$hw_tag" \
	    --ignore_errors "$ignore_errors" \
	    --input_size "$input_size" \
	    --iterations "$iterations" \
	    --run_profile "$run_profile" \
	    ${sysroot:+--sysroot "$sysroot"} \
	    --toolchain "$toolchain_type" \
	    --resultsdest "bkp-01.tcwglab:/home/tcwg-benchmark/results-${results_id}/$boardname" \
	    --nodename "$boardname" \
	    --forceinstall "${forceinstall}" \
	    ${clean_older_than:+--clean_older_than "$clean_older_than"} \
	    --verbose true

if $prepare_board; then
    remote_exec "$boardname:::-t -Snone" \
		sudo /usr/local/bin/benchmark.sh --action stop_board --verbose &
    res=0 && wait $! || res=$?
    if [ $res != 0 ]; then
	echo "Warning: prepare-board.sh did not finish cleanly"
    fi
fi