On Sun, Dec 17, 2017 at 12:17:26PM -0800, Mike Larkin wrote: > On Sun, Dec 17, 2017 at 08:34:52PM +0100, Patrick Wildt wrote: > > Hi, > > > > I had been looking into booting a 32-bit kernel with efiboot and the > > first thing I stumbled upon is that when we jump back into 32-bit mode > > we don't disable the Long Mode Extension. Thus when the i386 kernel > > turns on paging, LME goes active as well and "weird things happen". > > > > Not sure this path is worth pursuing, but this diff at least resets > > EFER correctly. > > > > Opinions? ok? > > > > Patrick > > > > diff --git a/sys/arch/amd64/stand/efiboot/run_i386.S > > b/sys/arch/amd64/stand/efiboot/run_i386.S > > index 1c70f8d4610..63d6f1f1dca 100644 > > --- a/sys/arch/amd64/stand/efiboot/run_i386.S > > +++ b/sys/arch/amd64/stand/efiboot/run_i386.S > > @@ -88,6 +88,12 @@ start32a: > > andl $(~CR4_PAE), %eax > > movl %eax, %cr4 > > > > + /* Disable LME */ > > + movl $MSR_EFER, %ecx > > + rdmsr > > + xor %eax, %eax > > + wrmsr > > + > > jmp start32b > > start32b: > > .code32 > > > > Could we ever have the case where EFI set up EFER_NXE? Does that > need to be preserved (even in i386 mode)? > > -ml >
Actually looking a few lines earlier, PAE is being disabled, so perhaps this is not even a valid question in the first place. Any reason you're doing the rdmsr there? If you just want to zero it, xor both the arguments to wrmsr and be done with it, right? -ml
