linux: treat a negative topology id as unavailable, not a parse error

[CodersAcademy006]

The kernel exposes core_id and physical_package_id as -1 when it has no topology information for a processor. Both are parsed with uint32_parser, which rejects any non-digit lead byte, so every such processor produces an error-level log claiming the sysfs file is malformed.

Observed on a native riscv64 board (SiFive U74-class, kernel 5.10.113), which reports core_id = -1 on all four cores:

Error in cpuinfo: failed to parse file /sys/devices/system/cpu/cpu0/topology/core_id: "-1" is not an unsigned number
Error in cpuinfo: failed to parse file /sys/devices/system/cpu/cpu1/topology/core_id: "-1" is not an unsigned number
Error in cpuinfo: failed to parse file /sys/devices/system/cpu/cpu2/topology/core_id: "-1" is not an unsigned number
Error in cpuinfo: failed to parse file /sys/devices/system/cpu/cpu3/topology/core_id: "-1" is not an unsigned number

-1 is a documented sentinel meaning "not populated", not corruption, so this is a false alarm rather than a real parse failure.

Root cause

parse_number (src/linux/processors.c) stops at the first non-digit. For -1 it stops immediately on the -, leaving parsed_end == text_start, which uint32_parser reports as "is not an unsigned number" at error level.

The change

Add a topology_id_parser wrapper used by the two topology getters. A topology id is never legitimately negative, so a leading - is reported as "value unavailable" at debug level instead. The return value is unchanged (false, since the id genuinely is unavailable), so no caller behavior changes. Only the false alarm goes away.

uint32_parser continues to back kernel_max and the three frequency getters, where a negative value would still be a genuine error.

physical_package_id parsed fine on the board above, but is exposed to the same sentinel through the same kernel mechanism, so it is covered for consistency.

Also fixed

The empty-file branch in uint32_parser hardcoded KERNEL_MAX_FILENAME even though the parser is shared by six call sites, so an empty core_id file would have blamed /sys/devices/system/cpu/kernel_max. It now reports the filename it was given.

Test plan

Compiled the real src/linux/processors.c into a harness driving its actual static parsers (stubbing only the sysfs reader), so this exercises the shipped code rather than a copy:

clang -DCPUINFO_LOG_LEVEL=5 -I src -I include -o t t.c src/log.c

Before, reproducing the board output exactly:

Error in cpuinfo: failed to parse file .../topology/core_id: "-1" is not an unsigned number
  core_id "-1"  -> ok=false
  core_id "0"   -> ok=true  value=0

After:

Debug (cpuinfo): file .../topology/core_id reports no topology information for this processor
  core_id "-1"  -> ok=false      (unchanged return, ERROR downgraded to DEBUG)
  core_id "0"   -> ok=true  value=0
  core_id "3"   -> ok=true  value=3
  core_id "-10" -> ok=false

[CodersAcademy006]

@GregoryComer @fbarchard could one of you take a look at this, or route it to whoever owns the Linux topology parsing? It has been open since July 15 with no reviewer assigned, so I suspect it never reached anyone's queue rather than being deliberately parked.

The change is 17 lines in src/linux/processors.c. The kernel exposes core_id and physical_package_id as -1 when it has no topology information for a processor, but both are read with uint32_parser, which rejects any non-digit lead byte. Every such processor therefore produces an error-level log claiming the sysfs file is malformed, on a machine where nothing is actually wrong. This treats a negative id as "unavailable", which is what the kernel means by it, and leaves genuinely malformed input reporting exactly as before.