Merge pull request #3377 from wineway/main

libct/cg: support SCHED_IDLE for runc cgroupfs

contrib/completions/bash/runc

@@ -743,6 +743,7 @@ _runc_update() {
 	   --pids-limit
 	   --l3-cache-schema
 	   --mem-bw-schema
+	   --cpu-idle
 	"
 
 	case "$prev" in

libcontainer/cgroups/fs/cpu.go

@@ -94,6 +94,14 @@ func (s *CpuGroup) Set(path string, r *configs.Resources) error {
 			}
 		}
 	}
+
+	if r.CPUIdle != nil {
+		idle := strconv.FormatInt(*r.CPUIdle, 10)
+		if err := cgroups.WriteFile(path, "cpu.idle", idle); err != nil {
+			return err
+		}
+	}
+
 	return s.SetRtSched(path, r)
 }
 

libcontainer/cgroups/fs2/cpu.go

@@ -11,14 +11,20 @@ import (
 )
 
 func isCpuSet(r *configs.Resources) bool {
-	return r.CpuWeight != 0 || r.CpuQuota != 0 || r.CpuPeriod != 0
+	return r.CpuWeight != 0 || r.CpuQuota != 0 || r.CpuPeriod != 0 || r.CPUIdle != nil
 }
 
 func setCpu(dirPath string, r *configs.Resources) error {
 	if !isCpuSet(r) {
 		return nil
 	}
 
+	if r.CPUIdle != nil {
+		if err := cgroups.WriteFile(dirPath, "cpu.idle", strconv.FormatInt(*r.CPUIdle, 10)); err != nil {
+			return err
+		}
+	}
+
 	// NOTE: .CpuShares is not used here. Conversion is the caller's responsibility.
 	if r.CpuWeight != 0 {
 		if err := cgroups.WriteFile(dirPath, "cpu.weight", strconv.FormatUint(r.CpuWeight, 10)); err != nil {

libcontainer/configs/cgroup_linux.go

@@ -84,6 +84,9 @@ type Resources struct {
 	// MEM to use
 	CpusetMems string `json:"cpuset_mems"`
 
+	// cgroup SCHED_IDLE
+	CPUIdle *int64 `json:"cpu_idle,omitempty"`
+
 	// Process limit; set <= `0' to disable limit.
 	PidsLimit int64 `json:"pids_limit"`
 

libcontainer/specconv/spec_linux.go

@@ -748,6 +748,7 @@ func CreateCgroupConfig(opts *CreateOpts, defaultDevs []*devices.Device) (*confi
 				}
 				c.Resources.CpusetCpus = r.CPU.Cpus
 				c.Resources.CpusetMems = r.CPU.Mems
+				c.Resources.CPUIdle = r.CPU.Idle
 			}
 			if r.Pids != nil {
 				c.Resources.PidsLimit = r.Pids.Limit

tests/integration/cgroups.bats

@@ -187,6 +187,18 @@ function setup() {
 	[[ "$weights" == *"$major:$minor 444"* ]]
 }
 
+@test "runc run (cpu.idle)" {
+	requires cgroups_cpu_idle
+	[ $EUID -ne 0 ] && requires rootless_cgroup
+
+	set_cgroups_path
+	update_config '.linux.resources.cpu.idle = 1'
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_cgroups_unified
+	[ "$status" -eq 0 ]
+	check_cgroup_value "cpu.idle" "1"
+}
+
 @test "runc run (cgroup v2 resources.unified only)" {
 	requires root cgroups_v2
 

tests/integration/helpers.bash

@@ -413,6 +413,15 @@ function requires() {
 				skip_me=1
 			fi
 			;;
+		cgroups_cpu_idle)
+			local p
+			init_cgroup_paths
+			[ -v CGROUP_V1 ] && p="$CGROUP_CPU_BASE_PATH"
+			[ -v CGROUP_V2 ] && p="$CGROUP_BASE_PATH"
+			if [ -z "$(find "$p" -name cpu.idle -print -quit)" ]; then
+				skip_me=1
+			fi
+			;;
 		cgroupns)
 			if [ ! -e "/proc/self/ns/cgroup" ]; then
 				skip_me=1

tests/integration/update.bats

@@ -425,6 +425,41 @@ EOF
 	check_cpu_quota 3000 10000 "300ms"
 }
 
+@test "update cgroup cpu.idle" {
+	requires cgroups_cpu_idle
+	[ $EUID -ne 0 ] && requires rootless_cgroup
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_update
+	[ "$status" -eq 0 ]
+
+	check_cgroup_value "cpu.idle" "0"
+
+	local val
+	for val in 1 0 1; do
+		runc update -r - test_update <<EOF
+{
+  "cpu": {
+    "idle": $val
+  }
+}
+EOF
+		[ "$status" -eq 0 ]
+		check_cgroup_value "cpu.idle" "$val"
+	done
+
+	for val in 1 0 1; do
+		runc update --cpu-idle "$val" test_update
+
+		[ "$status" -eq 0 ]
+		check_cgroup_value "cpu.idle" "$val"
+	done
+
+	# test update other option won't impact on cpu.idle
+	runc update --cpu-period 10000 test_update
+	[ "$status" -eq 0 ]
+	check_cgroup_value "cpu.idle" "1"
+}
+
 @test "update cgroup v2 resources via unified map" {
 	[ $EUID -ne 0 ] && requires rootless_cgroup
 	requires cgroups_v2

update.go

@@ -48,7 +48,8 @@ The accepted format is as follow (unchanged values can be omitted):
     "realtimeRuntime": 0,
     "realtimePeriod": 0,
     "cpus": "",
-    "mems": ""
+    "mems": "",
+    "idle": 0
   },
   "blockIO": {
     "weight": 0
@@ -106,6 +107,10 @@ other options are ignored.
 			Name:  "memory",
 			Usage: "Memory limit (in bytes)",
 		},
+		cli.StringFlag{
+			Name:  "cpu-idle",
+			Usage: "set cgroup SCHED_IDLE or not, 0: default behavior, 1: SCHED_IDLE",
+		},
 		cli.StringFlag{
 			Name:  "memory-reservation",
 			Usage: "Memory reservation or soft_limit (in bytes)",
@@ -192,6 +197,13 @@ other options are ignored.
 			if val := context.String("cpuset-mems"); val != "" {
 				r.CPU.Mems = val
 			}
+			if val := context.String("cpu-idle"); val != "" {
+				idle, err := strconv.ParseInt(val, 10, 64)
+				if err != nil {
+					return fmt.Errorf("invalid value for cpu-idle: %w", err)
+				}
+				r.CPU.Idle = i64Ptr(idle)
+			}
 
 			for _, pair := range []struct {
 				opt  string
@@ -294,6 +306,7 @@ other options are ignored.
 		config.Cgroups.Resources.CpusetCpus = r.CPU.Cpus
 		config.Cgroups.Resources.CpusetMems = r.CPU.Mems
 		config.Cgroups.Resources.Memory = *r.Memory.Limit
+		config.Cgroups.Resources.CPUIdle = r.CPU.Idle
 		config.Cgroups.Resources.MemoryReservation = *r.Memory.Reservation
 		config.Cgroups.Resources.MemorySwap = *r.Memory.Swap
 		config.Cgroups.Resources.MemoryCheckBeforeUpdate = *r.Memory.CheckBeforeUpdate