Re: [PATCHv10 05/18] x86/kexec: Keep CR4.MCE set during kexec for TDX guest

From: Kirill A. Shutemov
Date: Tue Apr 09 2024 - 11:47:36 EST


On Tue, Apr 09, 2024 at 07:22:24AM -0700, Sean Christopherson wrote:
> On Tue, Apr 09, 2024, Kirill A. Shutemov wrote:
> > Depending on setup, TDX guests might be allowed to clear CR4.MCE.
> > Attempt to clear it leads to #VE.
> >
> > Use alternatives to keep the flag during kexec for TDX guests.
> >
> > The change doesn't affect non-TDX-guest environments.
> >
> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > ---
> > arch/x86/kernel/relocate_kernel_64.S | 8 ++++++++
> > 1 file changed, 8 insertions(+)
> >
> > diff --git a/arch/x86/kernel/relocate_kernel_64.S b/arch/x86/kernel/relocate_kernel_64.S
> > index 56cab1bb25f5..8e2037d78a1f 100644
> > --- a/arch/x86/kernel/relocate_kernel_64.S
> > +++ b/arch/x86/kernel/relocate_kernel_64.S
> > @@ -5,6 +5,8 @@
> > */
> >
> > #include <linux/linkage.h>
> > +#include <linux/stringify.h>
> > +#include <asm/alternative.h>
> > #include <asm/page_types.h>
> > #include <asm/kexec.h>
> > #include <asm/processor-flags.h>
> > @@ -145,11 +147,17 @@ SYM_CODE_START_LOCAL_NOALIGN(identity_mapped)
> > * Set cr4 to a known state:
> > * - physical address extension enabled
> > * - 5-level paging, if it was enabled before
> > + * - Machine check exception on TDX guest, if it was enabled before.
> > + * Clearing MCE might not allowed in TDX guests, depending on setup.
> > */
> > movl $X86_CR4_PAE, %eax
> > testq $X86_CR4_LA57, %r13
> > jz 1f
> > orl $X86_CR4_LA57, %eax
> > +1:
> > + testq $X86_CR4_MCE, %r13
> > + jz 1f
> > + ALTERNATIVE "", __stringify(orl $X86_CR4_MCE, %eax), X86_FEATURE_TDX_GUEST
>
> The TEST+Jcc+OR sequences are rather odd, and require way more instructions and
> thus way more copy+paste than is necessary.
>
> movl $X86_CR4_LA57, %eax
> ALTERNATIVE "", __stringify(orl $X86_CR4_MCE, %eax), X86_FEATURE_TDX_GUEST
> andl %r13d, %eax
> orl $X86_CR4_PAE, %eax
> movq %rax, %cr4
>
> Then preserving new bits unconditionally only requires adding the flag to the
> initial move, and feature-dependent bits only need a single ALTERNATIVE line.

Thanks! It is much better.

> And there's no branches, blazing fast kexec! ;-)

kexec/sec STONKS! :D

Updated patch is below.