[FischLu]

Summary

Fix a NULL pointer dereference in ARM64 getarch when some CPUs are offline, such as on NVIDIA Jetson devices in low-power modes.

Root cause

detect() reads the CPU count from /sys/devices/system/cpu/possible, but falls back to /proc/cpuinfo, which only lists online CPUs.

When possible > online, the fallback loop can reach EOF and call strstr() with a NULL cpu_implementer, causing SIGSEGV.

Reproduction

NVIDIA Jetson AGX Orin, 30W power mode:

Possible CPUs: 12 (0-11)
Online CPUs:    8 (0-7)
Offline CPUs:   4 (8-11)

getarch 0 → SIGSEGV (exit 139)
getarch 1 → SIGSEGV (exit 139)

GDB confirms the crash in cpuid_arm64.c due to strstr(NULL, "0x41").

This also causes downstream CMake configuration and compilation errors.

Fix

Stop the /proc/cpuinfo fallback loop when the required fields are missing:

if (cpu_implementer == NULL || cpu_part == NULL) {
    break;
}

Only 3 lines added. Existing CPU-count behavior remains unchanged.

Testing

• Before fix: SIGSEGV reproduced on Jetson AGX Orin.
• Unaffected host: Jetson AGX Thor (14 CPUs online) works normally.
• After fix: Not independently tested on ARM64.

The issue also affects v0.3.29 and v0.3.33. A backport may be appropriate.

[martin-frbg]

Thank you - interesting trap I set for myself there.
(Alpine CI failure can be ignored - they closed down access to their repository while addressing urgent bugs)