Re: [PATCH 4/4] x86/kexec: Disable FRED earlier in kexec_reloc()

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 24/03/2026 1:19 pm, Jan Beulich wrote:
> On 19.03.2026 13:25, Andrew Cooper wrote:
>> With FRED just as with IDT, it's unsafe to run the exception handlers after
>> switching stack.
>>
>> To remove this unsafe window, %cr4 needs clearing earlier.  In turn, we may
>> need to switch to PCID 0 earlier too in order to be able to clear CR4.PCIDE.
>>
>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>

Thanks.

>
>> The pagetable switch cannot easily be moved to be earlier, as that leaves a
>> period of time where FRED is enabled but all stack pointers point to
>> non-existent mappings.
> But the clearing of PCID could be moved earlier, I guess.

Yes but then it's even further away from it's directly-related logic.

>
>> --- a/xen/arch/x86/x86_64/kexec_reloc.S
>> +++ b/xen/arch/x86/x86_64/kexec_reloc.S
>> @@ -54,6 +54,21 @@ FUNC(kexec_reloc, PAGE_SIZE)
>>          lidt    (%rsp)
>>          add     $10, %rsp
>>  
>> +        /* Move to PCID 0 if necessary, as a prerequisite to clearing 
>> CR4.PCIDE */
>> +        mov     %cr3, %rax
>> +        test    $0xfff, %eax
>> +        jz      1f
> This conditional looks not overly relevant on this path, and we'd get
> more consistent overall behavior ...
>
>> +        and     $~0xfff, %rax
>> +        mov     %rax, %cr3
> ... if the implied TLB flush would happen in all cases. (Of course the
> clearing of ...
>
>> +1:
>> +
>> +        /*
>> +         * Set CR4 to PAE only.  This may disable FRED, which must happen
>> +         * before switching off Xen's stack.
>> +         */
>> +        mov     $X86_CR4_PAE, %eax
>> +        mov     %rax, %cr4
> ... PCIDE will flush again anyway.)

Also PGE getting cleared.  There's also a full flush just out of context
lower when we switch off Xen's pagetables onto the identity pagetables.  

We don't need a 3rd unconditional flush.

~Andrew