Re: [PATCHv3, RESEND] x86/trampoline: Bypass compat mode in trampoline_start64() if not needed

[Kirill A. Shutemov]

On Thu, Jan 25, 2024 at 01:57:18AM -0800, Andi Kleen wrote:
> > +	/* Paging mode is correct proceed in 64-bit mode */
> > +
> > +	LOCK_AND_LOAD_REALMODE_ESP lock_rip=1
> > +
> > +	movw	$__KERNEL_DS, %dx
> > +	movl	%edx, %ss
> > +	addl	$pa_real_mode_base, %esp
> > +	movl	%edx, %ds
> > +	movl	%edx, %es
> > +	movl	%edx, %fs
> > +	movl	%edx, %gs
> > +
> > +	movl	$pa_trampoline_pgd, %eax
> > +	movq	%rax, %cr3
> > +
> > +	jmpq	*tr_start(%rip)
> 
> Still think we should add a far jump here so that we run on a defined
> code segment. It probably doesn't matter since there are likely no
> IRETs before reloading anyways, but it seems cleaner.

I think it is cleaner to switch to IRET here. Does this work for you?

diff --git a/arch/x86/realmode/rm/trampoline_64.S b/arch/x86/realmode/rm/trampoline_64.S
index 608f108cba7d..14d9c7daf90f 100644
--- a/arch/x86/realmode/rm/trampoline_64.S
+++ b/arch/x86/realmode/rm/trampoline_64.S
@@ -243,7 +243,9 @@ SYM_CODE_START(trampoline_start64)
 	movl	$pa_trampoline_pgd, %eax
 	movq	%rax, %cr3
 
-	jmpq	*tr_start(%rip)
+	pushq	$__KERNEL_CS
+	pushq	tr_start(%rip)
+	lretq
 .L_switch_paging:
 	/*
 	 * To switch between 4- and 5-level paging modes, it is necessary
-- 
  Kiryl Shutsemau / Kirill A. Shutemov