Function toggleCudaProcs

pkg/sentry/control/state_cuda.go:321–449

(sctx context.Context, k *kernel.Kernel, cudaCheckpointPath string, cudaProcs []*kernel.ThreadGroup, timeline *timing.Timeline, sequential bool)

}

func toggleCudaProcs(sctx context.Context, k *kernel.Kernel, cudaCheckpointPath string, cudaProcs []*kernel.ThreadGroup, timeline *timing.Timeline, sequential bool) error {
	start := time.Now()

	// Open /dev/null once for the stdin of all cuda-checkpoint processes.
	nullVD := k.VFS().NewAnonVirtualDentry("null")
	defer nullVD.DecRef(sctx)
	nullFD, err := memdev.NewNullFD(sctx, nullVD.Mount(), nullVD.Dentry(), vfs.OpenOptions{})
	if err != nil {
		log.Warningf("Failed to open /dev/null for cuda-checkpoint stdin: %v", err)
	} else {
		defer nullFD.DecRef(sctx)
	}

	// Call cuda-checkpoint for each CUDA PID.
	ckptTimerNames := make([]string, len(cudaProcs))
	for i, cudaProc := range cudaProcs {
		ckptTimerNames[i] = fmt.Sprintf("cuda-ckpt %s", cudaProc.ID())
	}
	var ckptTimings []*timing.Lease
	if !sequential {
		ckptTimelines := timeline.MultiFork(ckptTimerNames)
		ckptTimings = make([]*timing.Lease, len(cudaProcs))
		for i := range cudaProcs {
			ckptTimings[i] = ckptTimelines[i].Lease()
		}
	}
	defer func() {
		for _, t := range ckptTimings {
			t.End()
		}
	}()
	proc := &Proc{Kernel: k}
	ckptProcs := make(map[*kernel.ThreadGroup]checkpointProc)
	var errs []error
	for i, cudaProc := range cudaProcs {
		var ckptTiming *timing.Lease
		if sequential {
			ckptTiming = timeline.Fork(ckptTimerNames[i]).Lease()
		} else {
			ckptTiming = ckptTimings[i]
		}
		ckptProc, cleanup, err := invokeCudaCheckpoint(sctx, k, proc, cudaCheckpointPath, cudaProc, "--toggle", nullFD)
		if err != nil {
			ckptTiming.Reached("invoke error")
			errs = append(errs, err)
			break
		}
		if ckptProc.tg == nil {
			ckptTiming.Reached("tg nil")
			continue
		}
		ckptProcs[cudaProc] = ckptProc
		ckptTimeline := ckptTiming.Transfer()
		if sequential {
			ckptProc.tg.WaitExited()
			if status := ckptProc.tg.ExitStatus(); status != 0 {
				ckptTimeline.Reached("exec error")
			}