On Sun, 20 Nov 2016, Borislav Petkov wrote: > We will have set (or not) the X86_FEATURE_CPUID bit at > early_identify_cpu() time. Looking at the code, we do call sync_core() > pretty early. :-\
Hmm, watch out for the early microcode update driver for Intel processors should something get changed in the implementation, or in the behavior of sync_core(). That driver absolutely needs to issue a cpuid (with EAX = 1) before each rdmsr(MSR_IA32_UCODE_REV). And it uses sync_core() calls to do it. A CR2 access just won't do in this extremely specific case. This kind of pitfall is why I wanted to replace all use of sync_core() in arch/x86/kernel/cpu/microcode/intel.c with an explicit use of an inconditional cpuid(eax = 1)... (note: this protocol to read MSR_IA32_UCODE_REV was made an architectural requirement a while ago -- it was once considered an erratum workaround. It is documented in the "Intel 64 and IA‐32 Architectures Software Developer's Manual", Volume 3A: System Programming Guide, Part 1, section 9.11). -- Henrique Holschuh