Created attachment 470112 [details, diff] kexec-tools-2.0.13-r1-sscanf.patch For version 2.0.13-r1, the problem looks like this: $ /etc/init.d/kexec start * Caching service dependencies ... [ ok ] * Using kernel image /boot/vmlinuz (with /boot/initrd) for kexec ... Cannot load /boot/vmlinuz [ !! ] * ERROR: kexec failed to start For version 2.0.14, the problem looks like this: $ /etc/init.d/kexec start * Caching service dependencies ... [ ok ] * Using kernel image /boot/vmlinuz (with /boot/initrd) for kexec ... Cannot get kernel page_offset_base symbol address Cannot load /boot/vmlinuz [ !! ] * ERROR: kexec failed to start Note that the "Cannot get kernel page_offset_base symbol address" error message is unrelated. Neither version I tried gives any indication of why /boot/vmlinuz could not be loaded, and the 2.0.14 one gives us a red herring. Requisite conditions: - Be using musl libc. Or anything strictly adhering to ISO C should do ;) - Have sys-apps/kexec-tools-2.0.14 or earlier installed (as of this writing, this is the latest version that isn't upstream's git HEAD). - Use a crash kernel to catch kernel panics: More specifically, pass -p to kexec; I set KEXEC_OPT_ARGS="-p" in /etc/conf.d/kexec to get my bug on. Solution: This problem was caused by kexec using %Lx specifiers in sscanf format strings when %llx was intended. The bug is fixed in this commit: https://github.com/horms/kexec-tools/commit/47cc70157c6610d4c01a7ab4f8f0d24ab054b43b The "Cannot get kernel page_offset_base symbol address" error is another bug that upstream is aware of, and it seems to be benign. Details: https://bugzilla.redhat.com/show_bug.cgi?id=1428246 https://bugzilla.redhat.com/show_bug.cgi?id=1432322 The attached patch is based off of the above commit that fixes the sscanf format strings. With this patch my kexec kernel is no longer homeless :) I tested it on versions 2.0.13-r1 and 2.0.14. It might work for earlier versions. (I wish my google searches would have brought up that commit in the beginning of this... instead I ended up spending some hours figuring out that the crash_reserved_mem array always described one region with 0 as its start and end address, then figuring out /why/ the crash_reserved_mem array was so affected. Once I learned that it was about format specifiers, the upstream commit was easy to find, of course. o.O ) Hope that helps.
This was included with kexec-tools-2.0.15
