On 12/05/2021 18:59, Julien Grall wrote:
> Hi,
>
> On 11/05/2021 07:37, Michal Orzel wrote:
>> On 05.05.2021 10:00, Jan Beulich wrote:
>>> On 05.05.2021 09:43, Michal Orzel wrote:
>>>> --- a/xen/include/public/arch-arm.h
>>>> +++ b/xen/include/public/arch-arm.h
>>>> @@ -267,10 +267,10 @@ struct vcpu_guest_core_regs
>>>> /* Return address and mode */
>>>> __DECL_REG(pc64, pc32); /* ELR_EL2 */
>>>> - uint32_t cpsr; /* SPSR_EL2 */
>>>> + uint64_t cpsr; /* SPSR_EL2 */
>>>> union {
>>>> - uint32_t spsr_el1; /* AArch64 */
>>>> + uint64_t spsr_el1; /* AArch64 */
>>>> uint32_t spsr_svc; /* AArch32 */
>>>> };
>>>
>>> This change affects, besides domctl, also default_initialise_vcpu(),
>>> which Arm's arch_initialise_vcpu() calls. I realize do_arm_vcpu_op()
>>> only allows two unrelated VCPUOP_* to pass, but then I don't
>>> understand why arch_initialise_vcpu() doesn't simply return e.g.
>>> -EOPNOTSUPP. Hence I suspect I'm missing something.
>
> I think it is just an overlooked when reviewing the following commit:
>
> commit 192df6f9122ddebc21d0a632c10da3453aeee1c2
> Author: Roger Pau Monné <roger....@citrix.com>
> Date: Tue Dec 15 14:12:32 2015 +0100
>
> x86: allow HVM guests to use hypercalls to bring up vCPUs
>
> Allow the usage of the VCPUOP_initialise, VCPUOP_up, VCPUOP_down,
> VCPUOP_is_up, VCPUOP_get_physid and VCPUOP_send_nmi hypercalls
> from HVM
> guests.
>
> This patch introduces a new structure (vcpu_hvm_context) that
> should be used
> in conjuction with the VCPUOP_initialise hypercall in order to
> initialize
> vCPUs for HVM guests.
>
> Signed-off-by: Roger Pau Monné <roger....@citrix.com>
> Signed-off-by: Andrew Cooper <andrew.coop...@citrix.com>
> Reviewed-by: Jan Beulich <jbeul...@suse.com>
> Acked-by: Ian Campbell <ian.campb...@citrix.com>
>
> On Arm, the structure vcpu_guest_context is not exposed outside of Xen
> and the tools. Interestingly vcpu_guest_core_regs is but it should
> only be used within vcpu_guest_context.
>
> So as this is not used by stable ABI, it is fine to break it.
>
>>>
>> I agree that do_arm_vcpu_op only allows two VCPUOP* to pass and
>> arch_initialise_vcpu being called in case of VCPUOP_initialise
>> has no sense as VCPUOP_initialise is not supported on arm.
>> It makes sense that it should return -EOPNOTSUPP.
>> However do_arm_vcpu_op will not accept VCPUOP_initialise and will return
>> -EINVAL. So arch_initialise_vcpu for arm will not be called.
>> Do you think that changing this behaviour so that
>> arch_initialise_vcpu returns
>> -EOPNOTSUPP should be part of this patch?
>
> I think this change is unrelated. So it should be handled in a
> follow-up patch.
>
> If you are taking care of this, would you mind to also look to move
> struct vcpu_guest_core_regs within the #if defined(__XEN__) ||
> defined(__XEN_TOOLS__)?
+1. Fairly sure this is the conclusion of a discussion a year or so
back where I noted the same peculiarity, and tried to untangle the mess
which is the common vs arch specific code. (which is still outstanding,
and I don't immediately recall why.)
~Andrew
