Re: [Xen-devel] [PATCH v3 18/24] x86/shadow: Avoid raising faults behind the emulators back

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 01/12/16 13:00, Jan Beulich wrote:
>>>> On 30.11.16 at 14:50, <andrew.cooper3@xxxxxxxxxx> wrote:
>> --- a/xen/arch/x86/mm/shadow/multi.c
>> +++ b/xen/arch/x86/mm/shadow/multi.c
>> @@ -3373,18 +3373,35 @@ static int sh_page_fault(struct vcpu *v,
>>  
>>      r = x86_emulate(&emul_ctxt.ctxt, emul_ops);
>>  
>> -    /*
>> -     * The previous lack of inject_{sw,hw}*() hooks caused exceptions raised
>> -     * by the emulator itself to become X86EMUL_UNHANDLEABLE.  Such 
>> exceptions
>> -     * now set event_pending instead.  Exceptions raised behind the back of
>> -     * the emulator don't yet set event_pending.
>> -     *
>> -     * For now, cause such cases to return to the X86EMUL_UNHANDLEABLE path,
>> -     * for no functional change from before.  Future patches will fix this
>> -     * properly.
>> -     */
>>      if ( r == X86EMUL_EXCEPTION && emul_ctxt.ctxt.event_pending )
>> -        r = X86EMUL_UNHANDLEABLE;
>> +    {
>> +        /*
>> +         * This emulation covers writes to shadow pagetables.  We tolerate 
>> #PF
>> +         * (from hitting adjacent pages) and #GP/#SS (from segmentation
>> +         * errors).  Anything else is an emulation bug, or a guest playing
>> +         * with the instruction stream under Xen's feet.
>> +         */
> Same comment here regarding "adjacent".

In this case, the answer is different.  A misaligned write across the
end of a shadow pagetable may legitimately trigger a #PF.

>
>> +        if ( emul_ctxt.ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
>> +             (emul_ctxt.ctxt.event.vector < 32) &&
>> +             ((1u << emul_ctxt.ctxt.event.vector) &
>> +              ((1u << TRAP_stack_error) | (1u << TRAP_gp_fault) |
>> +               (1u << TRAP_page_fault))) )
> May I suggest to also demand an error code of zero for #GP/#SS?

Ok.

>
>> +        {
>> +            if ( is_hvm_vcpu(v) )
> has_hvm_container_domain()?

Very good point.  Will fix.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel