On Tue, 23 Jan 2018, Borislav Petkov wrote: > > + mode = retp_compiler() ? SPECTRE_V2_RETPOLINE_GENERIC : > > + SPECTRE_V2_RETPOLINE_MINIMAL; > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ > > but that might not always be an option.
I think we should start recording CFLAGS the kernel has been compiled with anyway; doesn't hurt and might come handy when debugging. /proc/version is probably not the best place ... /proc/cflags? > And it probably should be a more reliable method which we probably could > use to detect !retpolined modules too. That's the vermagic stuff Andi pushed. But that's not really acceptable for distros. Distros have always been in the situation "we let the external modules to load, as it'll work when it comes to functionality, but then it's our duty/responsibility to explain to 3rd parties that they *really* should recompile". Mostly because of security fixes to static inlines, but not only that. So that vermagic patch doesn't really help anything in real world (FWIW I've just dropped it from SLE kernel). "Potentially insecure" doesn't mean it shouldn't be loaded if the user wishes so. Only "functionally incorrect" (which is the kernel ABI compatibility check) should be the show stopper. -- Jiri Kosina SUSE Labs
