Re: [PATCH v4 07/28] arm64: KVM: Allow the main HYP code to use the init hyp stub implementation

[Marc Zyngier <marc.zyngier@arm.com>]

On 24/03/17 14:33, Christoffer Dall wrote:
> On Tue, Mar 21, 2017 at 07:20:37PM +0000, Marc Zyngier wrote:
>> We now have a full hyp-stub implementation in the KVM init code,
>> but the main KVM code only supports HVC_GET_VECTORS, which is not
>> enough.
>>
>> Instead of reinventing the wheel, let's reuse the init implementation
>> by branching to the idmap page when called with a hyp-stub hypercall.
>>
>> Reviewed-by: James Morse <james.morse@arm.com>
>> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
>> ---
>>  arch/arm64/kvm/hyp/hyp-entry.S | 24 +++++++++++++++++++-----
>>  1 file changed, 19 insertions(+), 5 deletions(-)
>>
>> diff --git a/arch/arm64/kvm/hyp/hyp-entry.S b/arch/arm64/kvm/hyp/hyp-entry.S
>> index d8ef788646c6..4f34c5996f86 100644
>> --- a/arch/arm64/kvm/hyp/hyp-entry.S
>> +++ b/arch/arm64/kvm/hyp/hyp-entry.S
>> @@ -87,10 +87,24 @@ alternative_endif
>>  	/* Here, we're pretty sure the host called HVC. */
>>  	ldp	x0, x1, [sp], #16
>>  
>> -	cmp	x0, #HVC_GET_VECTORS
>> -	b.ne	1f
>> -	mrs	x0, vbar_el2
>> -	b	2f
>> +	/* Check for a stub HVC call */
>> +	cmp	x0, #HVC_STUB_HCALL_NR
>> +	b.hs	1f
>> +
>> +	/*
>> +	 * Compute the idmap address of __kvm_handle_stub_hvc and
>> +	 * jump there. Since we use kimage_voffset, do not use the
>> +	 * HYP VA for __kvm_handle_stub_hvc, but the kernel VA instead
>> +	 * (by loading it from the constant pool).
>> +	 *
>> +	 * Preserve x0-x4, which may contain stub parameters.
>> +	 */
>> +	ldr	x5, =__kvm_handle_stub_hvc
>> +	ldr_l	x6, kimage_voffset
> 
> Isn't it a bit dodgy to just overwrite x5 and x6 in something which is
> not a function?  I know that in practice this always gets called through
> a function call and we can rely on the calling convention, but this can
> break if you issue a hypercall to KVM's HVC sub implementation using
> inline assembly, I think.
> 
> Am I missing something here?

I don't think you're missing anything. We're definitely relying on
getting there via a function call which is going to preserve the
caller-saved registers.

The only case where we issue a naked HVC call is when we use
HVC_SOFT_RESTART. At this point, corrupted registers is not much of a worry.

Now, I see two ways of dealing with this:
- We save x5/x6 on the stack, restoring them in the stub handler , right
before the eret
- Or we simply document the behaviour in asm/virt.h so that people
adding new stub methods know the restrictions.

Thoughts?

	M.
-- 
Jazz is not dead. It just smells funny...