On Mon, 22 May 2017, David Daney wrote: > > I suspect it will affect more than just `show_cpuinfo', e.g. some inline > > asm, and you could have included a justification as to why this patch is > > correct, such as by referring to how `set_isa' sets flags in `isa_level'. > > That is correct, and exactly what I said in my reply to the patch when it was > posted. When the OCTEON code was merged, different code paths were taken in > the kernel, and there was a deliberate decision to structure > mach-cavium-octeon/cpu-feature-overrides.h the way we did it. > > I also noted that the information exposed to userspace via /proc/cpuinfo > should be represented in the kernel by a distinct mechanism from how the > kernel makes internal decisions about CPU features. Explicit checks for Octeon should then be used instead in the individual pieces of code affected, e.g.: if (cpu_has_mips32r1 && !cpu_mach_octeon) or suchlike, possibly with an explanatory comment as to why such an exclusion has been made. The `cpu_has_mips32r1', etc. macros are supposed to be generic architectural checks. Also any design decisions should have been noted in the description of the original commit. Maciej
