Re: [PATCHv9 05/17] x86/kexec: Keep CR4.MCE set during kexec for TDX guest

[Kirill A. Shutemov]

On Mon, Mar 17, 2025 at 11:32:42AM +0000, David Woodhouse wrote:
> > And we don't want MCE to be enabled during kexec for !TDX_GUEST:
> > 
> > https://lore.kernel.org/all/1144340e-dd95-ee3b-dabb-579f9a65b3c7@xxxxxxxxxx/
> 
> Actually now I've added proper exception handling in relocate_kernel
> perhaps we could rethink that. But that's for the future.

There still going to be a gap once the new kernel is started until it gets
proper exception handling in place.

> > I think we should patch AND instruction to include X86_CR4_MCE on
> > TDX_GUEST:
> > ...
> > -	andl	$(X86_CR4_PAE | X86_CR4_LA57), %r13d
> > -	ALTERNATIVE "", __stringify(orl $X86_CR4_MCE, %r13d), X86_FEATURE_TDX_GUEST
> > +	ALTERNATIVE __stringify(andl	$(X86_CR4_PAE | X86_CR4_LA57), %r13d), \
> > +		    __stringify(andl	$(X86_CR4_PAE | X86_CR4_LA57 | X86_CR4_MCE), %r13d), X86_FEATURE_TDX_GUEST
> 
> Yeah... although the reason I'm looking at this is because I want to
> kill the ALTERNATIVE so that I can move the relocate_kernel() function
> into a data section:
> https://lore.kernel.org/all/20241218212326.44qff3i5n6cxuu5d@jpoimboe/
> 
> So I think I'll do it like this instead:

Looks good to me. Thanks for the cleanup you are doing!

-- 
  Kiryl Shutsemau / Kirill A. Shutemov