scripts/common: Introduce cache for the pci devices

Expose a cache of pci devices in form of an assoc array that could be
looked up during the runtime of a script like setup.sh.

In case of setup.sh, caching speeds up execution quite visibly:

config run, no caching:
real    0m4.488s
user    0m1.440s
sys     0m1.260s

config run, caching in use:
real    0m2.876s
user    0m0.365s
sys     0m0.420s

Note that for initial config runs, binding controllers to proper
drivers is the actual bottleneck.

status run, no caching:
real    0m1.877s
user    0m1.252s
sys     0m0.984s

status run, caching in use:
real    0m0.371s
user    0m0.242s
sys     0m0.204s

reset run, no caching:
real    0m2.559s
user    0m1.409s
sys     0m1.322s

reset run, caching in use:
real    0m0.960s
user    0m0.432s
sys     0m0.419s

Additionally, in case common tools, e.g. lspci, are missing, fallback to
sysfs to pick all needed devices from the pci bus. Targeted for Linux
systems only.

Change-Id: Ib69ef724b9f09eca0cbb9b88f1c363edc1efd5dc
Signed-off-by: Michal Berger <michalx.berger@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1845
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>

scripts/common.sh

@@ -28,6 +28,108 @@ function pci_can_use() {
 	return 1
 }
 
+cache_pci_init () {
+	local -gA pci_bus_cache
+
+	[[ -z ${pci_bus_cache[*]} || $CMD == reset ]] || return 1
+
+	pci_bus_cache=()
+}
+
+cache_pci () {
+	local pci=$1 class=$2 vendor=$3 device=$4
+
+	if [[ -n $class ]]; then
+		class=0x${class/0x}
+		pci_bus_cache["$class"]="${pci_bus_cache["$class"]:+${pci_bus_cache["$class"]} }$pci"
+	fi
+	if [[ -n $vendor && -n $device ]]; then
+		vendor=0x${vendor/0x} device=0x${device/0x}
+		pci_bus_cache["$vendor"]="${pci_bus_cache["$vendor"]:+${pci_bus_cache["$vendor"]} }$pci"
+		pci_bus_cache["$device"]="${pci_bus_cache["$device"]:+${pci_bus_cache["$device"]} }$pci"
+		pci_bus_cache["$vendor:$device"]="${pci_bus_cache["$vendor:$device"]:+${pci_bus_cache["$vendor:$device"]} }$pci"
+	fi
+}
+
+cache_pci_bus_sysfs () {
+	[[ -e /sys/bus/pci/devices ]] || return 1
+
+	cache_pci_init || return 0
+
+	local pci
+	local class vendor device
+
+	for pci in /sys/bus/pci/devices/*; do
+		class=$(<"$pci/class") vendor=$(<"$pci/vendor") device=$(<"$pci/device")
+		cache_pci "${pci##*/}" "$class" "$vendor" "$device"
+	done
+}
+
+cache_pci_bus_lspci () {
+	hash lspci 2>/dev/null || return 1
+
+	cache_pci_init || return 0
+
+	local dev
+	while read -ra dev; do
+		dev=("${dev[@]//\"/}")
+		# lspci splits ls byte of the class (prog. interface) into a separate
+		# field if it's != 0. Look for it and normalize the value to fit with
+		# what kernel exposes under sysfs.
+		if [[ ${dev[*]} =~ -p([0-9]+) ]]; then
+			dev[1]+=${BASH_REMATCH[1]}
+		else
+			dev[1]+=00
+		fi
+		# pci class vendor device
+		cache_pci "${dev[@]::4}"
+	done < <(lspci -Dnmm)
+}
+
+cache_pci_bus_pciconf () {
+	hash pciconf 2>/dev/null || return 1
+
+	cache_pci_init || return 0
+
+	local class vd vendor device
+	local pci domain bus device function
+
+	while read -r pci class _ vd _; do
+		IFS=":" read -r domain bus device function _ <<<"${pci##*pci}"
+		pci=$(printf '%04x:%02x:%02x:%x' \
+			"$domain" "$bus" "$device" "$function")
+		class=$(printf '0x%06x' $(( class )))
+		vendor=$(printf '0x%04x' $(( vd & 0xffff )))
+		device=$(printf '0x%04x' $(( (vd >> 16) & 0xffff )))
+
+		cache_pci "$pci" "$class" "$vendor" "$device"
+	done < <(pciconf -l)
+}
+
+cache_pci_bus () {
+	case "$(uname -s)" in
+		Linux) cache_pci_bus_lspci || cache_pci_bus_sysfs ;;
+		FreeBSD) cache_pci_bus_pciconf ;;
+	esac
+}
+
+iter_all_pci_sysfs () {
+	cache_pci_bus_sysfs || return 1
+
+	# default to class of the nvme devices
+	local find=${1:-0x010802} findx=$2
+	local pci pcis
+
+	[[ -n ${pci_bus_cache["$find"]} ]] || return 0
+	read -ra pcis <<<"${pci_bus_cache["$find"]}"
+
+	if (( findx )); then
+		printf '%s\n' "${pcis[@]::findx}"
+	else
+		printf '%s\n' "${pcis[@]}"
+	fi
+}
+
 # This function will ignore PCI PCI_WHITELIST and PCI_BLACKLIST
 function iter_all_pci_class_code() {
 	local class
@@ -52,6 +154,8 @@ function iter_all_pci_class_code() {
 		local addr=($(pciconf -l | grep -i "class=0x${class}${subclass}${progif}" | \
 			cut -d$'\t' -f1 | sed -e 's/^[a-zA-Z0-9_]*@pci//g' | tr ':' ' '))
 		printf "%04x:%02x:%02x:%x\n" ${addr[0]} ${addr[1]} ${addr[2]} ${addr[3]}
+	elif iter_all_pci_sysfs "$(printf '0x%06x' $(( 0x$progif | 0x$subclass << 8 | 0x$class << 16 )))"; then
+		:
 	else
 		echo "Missing PCI enumeration utility" >&2
 		exit 1
@@ -72,6 +176,8 @@ function iter_all_pci_dev_id() {
 		local addr=($(pciconf -l | grep -i "chip=0x${dev_id}${ven_id}" | \
 			cut -d$'\t' -f1 | sed -e 's/^[a-zA-Z0-9_]*@pci//g' | tr ':' ' '))
 		printf "%04x:%02x:%02x:%x\n" ${addr[0]} ${addr[1]} ${addr[2]} ${addr[3]}
+	elif iter_all_pci_sysfs "0x$ven_id:0x$dev_id"; then
+		:
 	else
 		echo "Missing PCI enumeration utility" >&2
 		exit 1

scripts/setup.sh

@@ -211,7 +211,7 @@ function configure_linux_pci {
 	fi
 
 	# NVMe
-	for bdf in $(iter_all_pci_class_code 01 08 02); do
+	for bdf in ${pci_bus_cache["0x010802"]}; do
 		blknames=()
 		if ! pci_can_use $bdf; then
 			pci_dev_echo "$bdf" "Skipping un-whitelisted NVMe controller at $bdf"
@@ -244,7 +244,7 @@ function configure_linux_pci {
 
 	while IFS= read -r dev_id
 	do
-		for bdf in $(iter_all_pci_dev_id 8086 $dev_id); do
+		for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
 			if ! pci_can_use $bdf; then
 				pci_dev_echo "$bdf" "Skipping un-whitelisted I/OAT device"
 				continue
@@ -263,7 +263,7 @@ function configure_linux_pci {
 
         while IFS= read -r dev_id
         do
-                for bdf in $(iter_all_pci_dev_id 8086 $dev_id); do
+                for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
                         if ! pci_can_use $bdf; then
                                 pci_dev_echo "$bdf" "Skipping un-whitelisted IDXD device"
                                 continue
@@ -282,7 +282,7 @@ function configure_linux_pci {
 
 	while IFS= read -r dev_id
 	do
-		for bdf in $(iter_all_pci_dev_id 1af4 $dev_id); do
+		for bdf in ${pci_bus_cache["0x1af4:0x$dev_id"]}; do
 			if ! pci_can_use $bdf; then
 				pci_dev_echo "$bdf" "Skipping un-whitelisted Virtio device at $bdf"
 				continue
@@ -309,7 +309,7 @@ function configure_linux_pci {
 
 	while IFS= read -r dev_id
 	do
-		for bdf in $(iter_pci_dev_id 8086 $dev_id); do
+		for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
 			if [[ -z "$PCI_WHITELIST" ]] || ! pci_can_use $bdf; then
 				echo "Skipping un-whitelisted VMD device at $bdf"
 				continue
@@ -448,7 +448,7 @@ function reset_linux_pci {
 	check_for_driver nvme
 	driver_loaded=$?
 	set -e
-	for bdf in $(iter_all_pci_class_code 01 08 02); do
+	for bdf in ${pci_bus_cache["0x010802"]}; do
 		if ! pci_can_use $bdf; then
 			pci_dev_echo "$bdf" "Skipping un-whitelisted NVMe controller $blkname"
 			continue
@@ -472,7 +472,7 @@ function reset_linux_pci {
 	set -e
 	while IFS= read -r dev_id
 	do
-		for bdf in $(iter_all_pci_dev_id 8086 $dev_id); do
+		for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
 			if ! pci_can_use $bdf; then
 				pci_dev_echo "$bdf" "Skipping un-whitelisted I/OAT device"
 				continue
@@ -497,7 +497,7 @@ function reset_linux_pci {
         set -e
         while IFS= read -r dev_id
         do
-                for bdf in $(iter_all_pci_dev_id 8086 $dev_id); do
+                for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
                         if ! pci_can_use $bdf; then
                                 pci_dev_echo "$bdf" "Skipping un-whitelisted IDXD device"
                                 continue
@@ -524,7 +524,7 @@ function reset_linux_pci {
 	modprobe virtio-pci || true
 	while IFS= read -r dev_id
 	do
-		for bdf in $(iter_all_pci_dev_id 1af4 $dev_id); do
+		for bdf in ${pci_bus_cache["0x1af4:0x$dev_id"]}; do
 			if ! pci_can_use $bdf; then
 				pci_dev_echo "$bdf" "Skipping un-whitelisted Virtio device at"
 				continue
@@ -546,7 +546,7 @@ function reset_linux_pci {
 	set -e
 	while IFS= read -r dev_id
 	do
-		for bdf in $(iter_pci_dev_id 8086 $dev_id); do
+		for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
 			if ! pci_can_use $bdf; then
 				echo "Skipping un-whitelisted VMD device at $bdf"
 				continue
@@ -605,7 +605,7 @@ function status_linux {
 	echo "NVMe devices"
 
 	echo -e "BDF\t\tVendor\tDevice\tNUMA\tDriver\t\tDevice name"
-	for bdf in $(iter_all_pci_class_code 01 08 02); do
+	for bdf in ${pci_bus_cache["0x010802"]}; do
 		driver=$(grep DRIVER /sys/bus/pci/devices/$bdf/uevent |awk -F"=" '{print $2}')
 		if [ "$numa_nodes" = "0" ]; then
 			node="-"
@@ -630,7 +630,7 @@ function status_linux {
 	| awk -F"x" '{print $2}')
 	echo -e "BDF\t\tVendor\tDevice\tNUMA\tDriver"
 	for dev_id in $TMP; do
-		for bdf in $(iter_all_pci_dev_id 8086 $dev_id); do
+		for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
 			driver=$(grep DRIVER /sys/bus/pci/devices/$bdf/uevent |awk -F"=" '{print $2}')
 			if [ "$numa_nodes" = "0" ]; then
 				node="-"
@@ -651,7 +651,7 @@ function status_linux {
         | awk -F"x" '{print $2}')
         echo -e "BDF\t\tVendor\tDevice\tNUMA\tDriver"
         for dev_id in $TMP; do
-                for bdf in $(iter_all_pci_dev_id 8086 $dev_id); do
+                for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
                         driver=$(grep DRIVER /sys/bus/pci/devices/$bdf/uevent |awk -F"=" '{print $2}')
                         if [ "$numa_nodes" = "0" ]; then
                                 node="-"
@@ -672,7 +672,7 @@ function status_linux {
 	| awk -F"x" '{print $2}')
 	echo -e "BDF\t\tVendor\tDevice\tNUMA\tDriver\t\tDevice name"
 	for dev_id in $TMP; do
-		for bdf in $(iter_all_pci_dev_id 1af4 $dev_id); do
+		for bdf in ${pci_bus_cache["0x1af4:0x$dev_id"]}; do
 			driver=$(grep DRIVER /sys/bus/pci/devices/$bdf/uevent |awk -F"=" '{print $2}')
 			if [ "$numa_nodes" = "0" ]; then
 				node="-"
@@ -694,7 +694,7 @@ function status_linux {
 	| awk -F"x" '{print $2}')
 	echo -e "BDF\t\tNuma Node\tDriver Name"
 	for dev_id in $TMP; do
-		for bdf in $(iter_pci_dev_id 8086 $dev_id); do
+		for bdf in ${pci_bus_cache["0x8086:0x$dev_id"]}; do
 			driver=$(grep DRIVER /sys/bus/pci/devices/$bdf/uevent |awk -F"=" '{print $2}')
 			node=$(cat /sys/bus/pci/devices/$bdf/numa_node);
 			echo -e "$bdf\t$node\t\t$driver"
@@ -703,44 +703,31 @@ function status_linux {
 }
 
 function configure_freebsd_pci {
-	TMP=$(mktemp)
+	local devs ids id
+	local BDFS
 
-	# NVMe
+	devs=PCI_DEVICE_ID_INTEL_IOAT
-	GREP_STR="class=0x010802"
+	devs+="|PCI_DEVICE_ID_INTEL_IDXD"
+	devs+="|PCI_DEVICE_ID_INTEL_VMD"
 
-	# IOAT
+	ids=($(grep -E "$devs" "$rootdir/include/spdk/pci_ids.h" | awk '{print $3}'))
-	grep "PCI_DEVICE_ID_INTEL_IOAT" $rootdir/include/spdk/pci_ids.h \
-	| awk -F"x" '{print $2}' > $TMP
-	while IFS= read -r dev_id
-	do
-		GREP_STR="${GREP_STR}\|chip=0x${dev_id}8086"
-	done < $TMP
 
-        # IDXD
+	if [[ -n ${pci_bus_cache["0x010802"]} ]]; then
-        grep "PCI_DEVICE_ID_INTEL_IDXD" $rootdir/include/spdk/pci_ids.h \
+		BDFS+=(${pci_bus_cache["0x010802"]})
-        | awk -F"x" '{print $2}' > $TMP
+	fi
-        while IFS= read -r dev_id
-        do
-                GREP_STR="${GREP_STR}\|chip=0x${dev_id}8086"
-        done < $TMP
 
-	# VMD
+	for id in "${ids[@]}"; do
-	grep "PCI_DEVICE_ID_INTEL_VMD" $rootdir/include/spdk/pci_ids.h \
+		[[ -n ${pci_bus_cache["0x8086:$id"]} ]] || continue
-	| awk -F"x" '{print $2}' > $TMP
+		BDFS+=(${pci_bus_cache["0x8086:$id"]})
-	while IFS= read -r dev_id
+	done
-	do
-		GREP_STR="${GREP_STR}\|chip=0x${dev_id}8086"
-	done < $TMP
 
-	AWK_PROG=("{if (count > 0) printf \",\"; printf \"%s:%s:%s\",\$2,\$3,\$4; count++}")
+	# Drop the domain part from all the addresses
-	echo "${AWK_PROG[*]}" > $TMP
+	BDFS=("${BDFS[@]#*:}")
-
-	BDFS=$(pciconf -l | grep "${GREP_STR}" | awk -F: -f $TMP)
 
+	local IFS=","
 	kldunload nic_uio.ko || true
-	kenv hw.nic_uio.bdfs=$BDFS
+	kenv hw.nic_uio.bdfs="${BDFS[*]}"
 	kldload nic_uio.ko
-	rm $TMP
 }
 
 function configure_freebsd {
@@ -764,6 +751,8 @@ function reset_freebsd {
 	kldunload nic_uio.ko || true
 }
 
+CMD=reset cache_pci_bus
+
 mode=$1
 
 if [ -z "$mode" ]; then