add Confidential Computing readiness check + collect nvidia-smi conf-compute -q
New read-only "Check" step reports whether this server can run NVIDIA Confidential Computing: CPU TEE support (Intel TDX / AMD SEV-SNP, via dmesg and kvm_amd sysfs params) and GPU firmware CC capability (via `nvidia-smi conf-compute -q`). Also collect that command's output into the techdump export bundle. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Mikhail Chusavitin
@@ -0,0 +1,248 @@
|
||||
package platform
|
||||
|
||||
import (
|
||||
"bytes"
|
||||
"context"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
// ConfidentialComputingStatus summarizes whether this server can run NVIDIA
|
||||
// Confidential Computing: CPU-side TEE support (Intel TDX / AMD SEV-SNP) and
|
||||
// GPU firmware CC capability, as reported by `nvidia-smi conf-compute -q`.
|
||||
type ConfidentialComputingStatus struct {
|
||||
CollectedAt time.Time `json:"collected_at"`
|
||||
|
||||
// GPU-reported fields, parsed from `nvidia-smi conf-compute -q`.
|
||||
NvidiaSMIAvailable bool `json:"nvidia_smi_available"`
|
||||
CCState string `json:"cc_state,omitempty"` // ON / OFF
|
||||
MultiGPUMode string `json:"multi_gpu_mode,omitempty"` // Protected PCIe / ...
|
||||
CPUCCCapability string `json:"cpu_cc_capability,omitempty"` // e.g. "INTEL TDX", "AMD SEV-SNP", "NONE"
|
||||
GPUCCCapability string `json:"gpu_cc_capability,omitempty"` // e.g. "CC Capable", "Not Capable"
|
||||
CCGPUsReadyState string `json:"cc_gpus_ready_state,omitempty"` // Ready / Not Ready
|
||||
|
||||
// Host-side evidence that the CPU's TEE is actually active in the running
|
||||
// kernel (BIOS + kernel cmdline + firmware), independent of what the GPU
|
||||
// driver reports. Used as a fallback when the NVIDIA driver isn't loaded.
|
||||
HostAMDSEVSupported bool `json:"host_amd_sev_supported"`
|
||||
HostAMDSEVESSupported bool `json:"host_amd_sev_es_supported"`
|
||||
HostAMDSEVSNPActive bool `json:"host_amd_sev_snp_active"`
|
||||
HostIntelTDXActive bool `json:"host_intel_tdx_active"`
|
||||
|
||||
// GPUCanRunCC is true when the GPU firmware reports CC-capable.
|
||||
GPUCanRunCC bool `json:"gpu_can_run_cc"`
|
||||
// CPUCanRunCC is true when either the GPU driver or the host kernel
|
||||
// reports an active/available CPU TEE (SEV-SNP or TDX).
|
||||
CPUCanRunCC bool `json:"cpu_can_run_cc"`
|
||||
// Ready is true when both the CPU and the GPU support Confidential
|
||||
// Computing, regardless of whether CC mode is currently enabled.
|
||||
Ready bool `json:"ready"`
|
||||
|
||||
Notes []string `json:"notes,omitempty"`
|
||||
}
|
||||
|
||||
// RunConfidentialComputingCheckPack runs a read-only check of whether this
|
||||
// server can run NVIDIA Confidential Computing: it queries the GPU driver
|
||||
// (`nvidia-smi conf-compute -q`) and inspects host kernel/dmesg evidence of
|
||||
// AMD SEV-SNP / Intel TDX support. It changes nothing on the system.
|
||||
func (s *System) RunConfidentialComputingCheckPack(ctx context.Context, baseDir string, logFunc func(string)) (string, error) {
|
||||
if ctx == nil {
|
||||
ctx = context.Background()
|
||||
}
|
||||
if baseDir == "" {
|
||||
baseDir = "/var/log/bee-sat"
|
||||
}
|
||||
ts := time.Now().UTC().Format("20060102-150405")
|
||||
runDir := filepath.Join(baseDir, "confidential-computing-"+ts)
|
||||
if err := os.MkdirAll(runDir, 0755); err != nil {
|
||||
return "", err
|
||||
}
|
||||
verboseLog := filepath.Join(runDir, "verbose.log")
|
||||
|
||||
status := ConfidentialComputingStatus{CollectedAt: time.Now().UTC()}
|
||||
|
||||
// GPU firmware / driver state.
|
||||
ccOut, ccErr := runSATCommandCtx(ctx, verboseLog, "nvidia-smi-conf-compute-q", []string{"nvidia-smi", "conf-compute", "-q"}, nil, logFunc)
|
||||
_ = os.WriteFile(filepath.Join(runDir, "01-nvidia-smi-conf-compute-q.log"), ccOut, 0644)
|
||||
if ccErr == nil {
|
||||
status.NvidiaSMIAvailable = true
|
||||
fields := parseConfComputeFields(ccOut)
|
||||
status.CCState = fields["CC State"]
|
||||
status.MultiGPUMode = fields["Multi-GPU Mode"]
|
||||
status.CPUCCCapability = fields["CPU CC Capabilities"]
|
||||
status.GPUCCCapability = fields["GPU CC Capabilities"]
|
||||
status.CCGPUsReadyState = fields["CC GPUs Ready State"]
|
||||
} else {
|
||||
status.Notes = append(status.Notes, "nvidia-smi conf-compute -q unavailable (no NVIDIA driver, or GPU not present): "+firstLine(string(ccOut)))
|
||||
}
|
||||
|
||||
// Host kernel evidence, independent of the GPU driver.
|
||||
dmesgOut, _ := runSATCommandCtx(ctx, verboseLog, "dmesg", []string{"dmesg"}, nil, nil)
|
||||
ccDmesgLines := filterConfComputeDmesgLines(dmesgOut)
|
||||
_ = os.WriteFile(filepath.Join(runDir, "02-dmesg-cc-relevant.log"), []byte(strings.Join(ccDmesgLines, "\n")+"\n"), 0644)
|
||||
|
||||
lowerDmesg := strings.ToLower(strings.Join(ccDmesgLines, "\n"))
|
||||
status.HostAMDSEVSNPActive = strings.Contains(lowerDmesg, "sev-snp enabled")
|
||||
status.HostIntelTDXActive = strings.Contains(lowerDmesg, "tdx module") && strings.Contains(lowerDmesg, "module initialized") ||
|
||||
strings.Contains(lowerDmesg, "virt/tdx: module initialized")
|
||||
|
||||
for i, path := range []string{
|
||||
"/sys/module/kvm_amd/parameters/sev",
|
||||
"/sys/module/kvm_amd/parameters/sev_es",
|
||||
"/sys/module/kvm_amd/parameters/sev_snp",
|
||||
} {
|
||||
name := fmt.Sprintf("sysfs-%s", filepath.Base(path))
|
||||
out, err := runSATCommandCtx(ctx, verboseLog, name, []string{"cat", path}, nil, nil)
|
||||
_ = os.WriteFile(filepath.Join(runDir, fmt.Sprintf("03-%02d-%s.log", i+1, name)), out, 0644)
|
||||
if err != nil {
|
||||
continue
|
||||
}
|
||||
val := strings.TrimSpace(string(out))
|
||||
switch filepath.Base(path) {
|
||||
case "sev":
|
||||
status.HostAMDSEVSupported = strings.EqualFold(val, "Y")
|
||||
case "sev_es":
|
||||
status.HostAMDSEVESSupported = strings.EqualFold(val, "Y")
|
||||
case "sev_snp":
|
||||
if strings.EqualFold(val, "Y") {
|
||||
status.HostAMDSEVSNPActive = true
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
status.GPUCanRunCC = strings.EqualFold(strings.TrimSpace(status.GPUCCCapability), "CC Capable")
|
||||
cpuCapReported := strings.TrimSpace(status.CPUCCCapability)
|
||||
status.CPUCanRunCC = status.HostAMDSEVSNPActive || status.HostIntelTDXActive ||
|
||||
(cpuCapReported != "" && !strings.EqualFold(cpuCapReported, "NONE"))
|
||||
status.Ready = status.CPUCanRunCC && status.GPUCanRunCC
|
||||
|
||||
if !status.NvidiaSMIAvailable {
|
||||
status.Notes = append(status.Notes, "GPU CC capability unknown — install the NVIDIA driver to query it with `nvidia-smi conf-compute -q`.")
|
||||
}
|
||||
|
||||
summary := renderConfidentialComputingSummary(status)
|
||||
if err := os.WriteFile(filepath.Join(runDir, "summary.txt"), []byte(summary), 0644); err != nil {
|
||||
return "", err
|
||||
}
|
||||
report := renderConfidentialComputingReport(status)
|
||||
if err := os.WriteFile(filepath.Join(runDir, "confidential-computing-report.txt"), []byte(report), 0644); err != nil {
|
||||
return "", err
|
||||
}
|
||||
|
||||
return runDir, nil
|
||||
}
|
||||
|
||||
// parseConfComputeFields parses the indented "Key : Value" block emitted by
|
||||
// `nvidia-smi conf-compute -q`, e.g.:
|
||||
//
|
||||
// CC State : OFF
|
||||
// Multi-GPU Mode : Protected PCIe
|
||||
// CPU CC Capabilities : INTEL TDX
|
||||
// GPU CC Capabilities : CC Capable
|
||||
// CC GPUs Ready State : Not Ready
|
||||
func parseConfComputeFields(out []byte) map[string]string {
|
||||
fields := map[string]string{}
|
||||
for _, line := range strings.Split(string(out), "\n") {
|
||||
idx := strings.Index(line, ":")
|
||||
if idx < 0 {
|
||||
continue
|
||||
}
|
||||
key := strings.TrimSpace(line[:idx])
|
||||
val := strings.TrimSpace(line[idx+1:])
|
||||
if key == "" || val == "" {
|
||||
continue
|
||||
}
|
||||
fields[key] = val
|
||||
}
|
||||
return fields
|
||||
}
|
||||
|
||||
// filterConfComputeDmesgLines returns the dmesg lines relevant to CPU
|
||||
// Confidential Computing support (AMD SEV/SEV-ES/SEV-SNP, Intel TDX).
|
||||
func filterConfComputeDmesgLines(dmesgOut []byte) []string {
|
||||
var lines []string
|
||||
scanner := bytes.Split(dmesgOut, []byte("\n"))
|
||||
for _, raw := range scanner {
|
||||
lower := strings.ToLower(string(raw))
|
||||
if strings.Contains(lower, "sev") || strings.Contains(lower, "tdx") {
|
||||
lines = append(lines, string(raw))
|
||||
}
|
||||
}
|
||||
return lines
|
||||
}
|
||||
|
||||
func renderConfidentialComputingSummary(status ConfidentialComputingStatus) string {
|
||||
var b strings.Builder
|
||||
fmt.Fprintf(&b, "run_at_utc=%s\n", status.CollectedAt.Format(time.RFC3339))
|
||||
fmt.Fprintf(&b, "nvidia_smi_available=%t\n", status.NvidiaSMIAvailable)
|
||||
fmt.Fprintf(&b, "cc_state=%s\n", status.CCState)
|
||||
fmt.Fprintf(&b, "multi_gpu_mode=%s\n", status.MultiGPUMode)
|
||||
fmt.Fprintf(&b, "cpu_cc_capability=%s\n", status.CPUCCCapability)
|
||||
fmt.Fprintf(&b, "gpu_cc_capability=%s\n", status.GPUCCCapability)
|
||||
fmt.Fprintf(&b, "cc_gpus_ready_state=%s\n", status.CCGPUsReadyState)
|
||||
fmt.Fprintf(&b, "host_amd_sev_supported=%t\n", status.HostAMDSEVSupported)
|
||||
fmt.Fprintf(&b, "host_amd_sev_es_supported=%t\n", status.HostAMDSEVESSupported)
|
||||
fmt.Fprintf(&b, "host_amd_sev_snp_active=%t\n", status.HostAMDSEVSNPActive)
|
||||
fmt.Fprintf(&b, "host_intel_tdx_active=%t\n", status.HostIntelTDXActive)
|
||||
fmt.Fprintf(&b, "cpu_can_run_cc=%t\n", status.CPUCanRunCC)
|
||||
fmt.Fprintf(&b, "gpu_can_run_cc=%t\n", status.GPUCanRunCC)
|
||||
fmt.Fprintf(&b, "ready=%t\n", status.Ready)
|
||||
if status.Ready {
|
||||
fmt.Fprintln(&b, "overall_status=OK")
|
||||
} else {
|
||||
fmt.Fprintln(&b, "overall_status=NOT_READY")
|
||||
}
|
||||
return b.String()
|
||||
}
|
||||
|
||||
func renderConfidentialComputingReport(status ConfidentialComputingStatus) string {
|
||||
var b strings.Builder
|
||||
line := strings.Repeat("=", 80)
|
||||
b.WriteString(line + "\n")
|
||||
b.WriteString("Confidential Computing Readiness\n")
|
||||
b.WriteString(line + "\n\n")
|
||||
|
||||
verdict := "NOT READY"
|
||||
if status.Ready {
|
||||
verdict = "READY"
|
||||
}
|
||||
fmt.Fprintf(&b, "Verdict: %s\n\n", verdict)
|
||||
|
||||
b.WriteString("-- CPU ----------------------------------------------------------------------\n")
|
||||
fmt.Fprintf(&b, " Reported by GPU driver : %s\n", nonEmptyOr(status.CPUCCCapability, "unknown"))
|
||||
fmt.Fprintf(&b, " AMD SEV supported : %t\n", status.HostAMDSEVSupported)
|
||||
fmt.Fprintf(&b, " AMD SEV-ES supported : %t\n", status.HostAMDSEVESSupported)
|
||||
fmt.Fprintf(&b, " AMD SEV-SNP active : %t\n", status.HostAMDSEVSNPActive)
|
||||
fmt.Fprintf(&b, " Intel TDX active : %t\n", status.HostIntelTDXActive)
|
||||
fmt.Fprintf(&b, " Can run CC : %t\n\n", status.CPUCanRunCC)
|
||||
|
||||
b.WriteString("-- GPU ----------------------------------------------------------------------\n")
|
||||
fmt.Fprintf(&b, " nvidia-smi available : %t\n", status.NvidiaSMIAvailable)
|
||||
fmt.Fprintf(&b, " GPU CC Capabilities : %s\n", nonEmptyOr(status.GPUCCCapability, "unknown"))
|
||||
fmt.Fprintf(&b, " CC State (current) : %s\n", nonEmptyOr(status.CCState, "unknown"))
|
||||
fmt.Fprintf(&b, " Multi-GPU Mode : %s\n", nonEmptyOr(status.MultiGPUMode, "unknown"))
|
||||
fmt.Fprintf(&b, " CC GPUs Ready State : %s\n", nonEmptyOr(status.CCGPUsReadyState, "unknown"))
|
||||
fmt.Fprintf(&b, " Can run CC : %t\n\n", status.GPUCanRunCC)
|
||||
|
||||
if len(status.Notes) > 0 {
|
||||
b.WriteString("-- Notes ----------------------------------------------------------------------\n")
|
||||
for _, n := range status.Notes {
|
||||
fmt.Fprintf(&b, " - %s\n", n)
|
||||
}
|
||||
b.WriteString("\n")
|
||||
}
|
||||
|
||||
fmt.Fprintf(&b, "Collected : %s\n", status.CollectedAt.Format("2006-01-02 15:04:05 UTC"))
|
||||
b.WriteString(line + "\n")
|
||||
return b.String()
|
||||
}
|
||||
|
||||
func nonEmptyOr(v, fallback string) string {
|
||||
if strings.TrimSpace(v) == "" {
|
||||
return fallback
|
||||
}
|
||||
return v
|
||||
}
|
||||
@@ -38,6 +38,7 @@ var techDumpNvidiaCommands = []struct {
|
||||
}{
|
||||
{Name: "nvidia-smi", Args: []string{"-q"}, File: "nvidia-smi-q.txt"},
|
||||
{Name: "nvidia-smi", Args: []string{"--query-gpu=index,pci.bus_id,serial,vbios_version,temperature.gpu,power.draw,ecc.errors.uncorrected.aggregate.total,ecc.errors.corrected.aggregate.total,clocks_throttle_reasons.hw_slowdown", "--format=csv,noheader,nounits"}, File: "nvidia-smi-query.csv"},
|
||||
{Name: "nvidia-smi", Args: []string{"conf-compute", "-q"}, File: "nvidia-smi-conf-compute-q.txt"},
|
||||
}
|
||||
|
||||
type lsblkDumpRoot struct {
|
||||
|
||||
Reference in New Issue
Block a user