lib: don't fall back to /dev/mem for ACPI tables exposed via sysfs

[rucoder]

Problem

Closes #307.

On kernels built with CONFIG_STRICT_DEVMEM=y, initializing libpqos (or running the pqos CLI) emits:

ERROR: Memory map failed, address=... size=...
ERROR: Memory mapping failed!
ERROR: Failed to obtain RSDP table!
WARN: Could not obtain ERDT table

acpi_get_sig() first looks for a table under /sys/firmware/acpi/tables/, and if it is not there falls back to acpi_get_mmap(), which opens /dev/mem and maps/reads physical memory (RSDP/BIOS region). CONFIG_STRICT_DEVMEM=y restricts /dev/mem, so the mmap fails — or, for the low-BIOS scan region, the mmap can succeed and the subsequent read faults. For a library caller this is worse than log noise: a fault in this path takes down the process (e.g. a CGO host such as the Go bindings).

Why the memory fallback is unnecessary here

The kernel exposes every ACPI table advertised in the RSDT/XSDT under /sys/firmware/acpi/tables/, keyed by signature, regardless of whether the kernel has a driver for it (drivers/acpi/sysfs.c walks acpi_gbl_root_table_list with no signature filtering). Therefore:

• If ERDT/MRRM/IRDT exist, firmware lists them in the XSDT and they are already present in sysfs — no /dev/mem needed.
• If they are not in sysfs, they are not obtainable from /dev/mem either — the fallback finds nothing and only risks the strict-devmem fault.

The fallback is only meaningful on legacy kernels that do not expose /sys/firmware/acpi/tables/ at all.

Fix

In acpi_get_sig(), when /sys/firmware/acpi/tables/ exists but the requested table is not present there, return NULL instead of scanning /dev/mem. The memory fallback is preserved for kernels that do not expose that directory.

Test

• make -C lib builds clean (no new warnings; acpi_get_mmap is still referenced for the legacy path).
• On a CONFIG_STRICT_DEVMEM=y host where /sys/firmware/acpi/tables/ exists without an ERDT entry, init no longer touches /dev/mem and the errors above are gone.

[Copilot]

Pull request overview

This PR updates libpqos ACPI table discovery to avoid falling back to /dev/mem when the kernel exposes ACPI tables via sysfs, preventing failures (and potential process faults) on CONFIG_STRICT_DEVMEM=y systems.

Changes:

• In acpi_get_sig(), if /sys/firmware/acpi/tables/ exists but the requested table is not found there, return NULL instead of scanning physical memory via /dev/mem.
• Preserve the legacy /dev/mem fallback only for kernels that do not expose the ACPI tables sysfs directory.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rkanagar]

HI @rucoder ,
Please address the review comments. Thanks

[rucoder]

@rkanagar addressed the review comments in 9fbbbc3: reworded the comment and debug message in acpi_get_sig() to say the table is "not available via sysfs" (covering both the absent case and the present-but-unreadable/invalid case), since the early return also triggers when acpi_read_fs() fails. No behavior change. Thanks!

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 1 comment.