Re: [RFC PATCH] x86_64,entry,xen: Do not invoke espfix64 on Xen

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 07/15/2014 01:22 PM, Andy Lutomirski wrote:
> On Tue, Jul 15, 2014 at 10:19 AM, Boris Ostrovsky
> <boris.ostrovsky@oracle.com> wrote:
>> On 07/15/2014 12:23 PM, Andy Lutomirski wrote:
>>>
>>> diff --git a/arch/x86/kernel/paravirt_patch_64.c
>>> b/arch/x86/kernel/paravirt_patch_64.c
>>> index 3f08f34..a1da673 100644
>>> --- a/arch/x86/kernel/paravirt_patch_64.c
>>> +++ b/arch/x86/kernel/paravirt_patch_64.c
>>> @@ -6,7 +6,6 @@ DEF_NATIVE(pv_irq_ops, irq_disable, "cli");
>>>    DEF_NATIVE(pv_irq_ops, irq_enable, "sti");
>>>    DEF_NATIVE(pv_irq_ops, restore_fl, "pushq %rdi; popfq");
>>>    DEF_NATIVE(pv_irq_ops, save_fl, "pushfq; popq %rax");
>>> -DEF_NATIVE(pv_cpu_ops, iret, "iretq");
>>>    DEF_NATIVE(pv_mmu_ops, read_cr2, "movq %cr2, %rax");
>>>    DEF_NATIVE(pv_mmu_ops, read_cr3, "movq %cr3, %rax");
>>>    DEF_NATIVE(pv_mmu_ops, write_cr3, "movq %rdi, %cr3");
>>> @@ -50,7 +49,6 @@ unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
>>>                  PATCH_SITE(pv_irq_ops, save_fl);
>>>                  PATCH_SITE(pv_irq_ops, irq_enable);
>>>                  PATCH_SITE(pv_irq_ops, irq_disable);
>>> -               PATCH_SITE(pv_cpu_ops, iret);
>>
>>
>> Does this mean that we are no longer patching IRET with a jump to a
>> hypercall?
>>
> IIUC this means that, on native, we are no longer patching
> INTERRUPT_RETURN with an "iretq" instruction, so INTERRUPT_RETURN will
> remain a "jmp native_iret".  I'm not sure why this patch was there in
> the first place.  On Xen, it should still get patched with the
> hypercall (although someone should verify this).

Right, I missed the fact that this is native_patch.

I did some light testing and it appears to work. Are you targeting this 
for 3.16?

One way or the other we need to disable espfix64 on PV --- I discovered 
that one of Peter's tests crashes the hypervisor.


-boris