Agreed on both, changed in v4 (patch 4/12):
https://lore.kernel.org/bpf/[email protected]/

Only -ENOMEM from btf_parse_vmlinux() is left uncached with =m; any other
failure is a broken BTF and is remembered as with =y, so it is parsed
once, not again by every later request.  And /sys/kernel/btf/vmlinux now
serves the raw bytes whenever btf_vmlinux.ko supplied them, whether they
parse or not, as with =y.

Jay

Reply via email to