Re: [XEN][PATCH v3] xen/x86: guest_access: optimize raw_x_guest() for PV and HVM combinations

[Jan Beulich <jbeulich@xxxxxxxx>]

On 12.11.2025 12:27, Grygorii Strashko wrote:
> 
> 
> On 12.11.25 08:38, Jan Beulich wrote:
>> On 11.11.2025 18:52, Grygorii Strashko wrote:
>>> On 10.11.25 09:11, Jan Beulich wrote:
>>>> On 07.11.2025 19:17, Grygorii Strashko wrote:
>>>>> --- a/xen/arch/x86/include/asm/guest_access.h
>>>>> +++ b/xen/arch/x86/include/asm/guest_access.h
>>>>> @@ -13,26 +13,64 @@
>>>>>    #include <asm/hvm/guest_access.h>
>>>>>      /* Raw access functions: no type checking. */
>>>>> -#define raw_copy_to_guest(dst, src, len)        \
>>>>> -    (is_hvm_vcpu(current) ?                     \
>>>>> -     copy_to_user_hvm((dst), (src), (len)) :    \
>>>>> -     copy_to_guest_pv(dst, src, len))
>>>>> -#define raw_copy_from_guest(dst, src, len)      \
>>>>> -    (is_hvm_vcpu(current) ?                     \
>>>>> -     copy_from_user_hvm((dst), (src), (len)) :  \
>>>>> -     copy_from_guest_pv(dst, src, len))
>>>>> -#define raw_clear_guest(dst,  len)              \
>>>>> -    (is_hvm_vcpu(current) ?                     \
>>>>> -     clear_user_hvm((dst), (len)) :             \
>>>>> -     clear_guest_pv(dst, len))
>>>>> -#define __raw_copy_to_guest(dst, src, len)      \
>>>>> -    (is_hvm_vcpu(current) ?                     \
>>>>> -     copy_to_user_hvm((dst), (src), (len)) :    \
>>>>> -     __copy_to_guest_pv(dst, src, len))
>>>>> -#define __raw_copy_from_guest(dst, src, len)    \
>>>>> -    (is_hvm_vcpu(current) ?                     \
>>>>> -     copy_from_user_hvm((dst), (src), (len)) :  \
>>>>> -     __copy_from_guest_pv(dst, src, len))
>>>>> +static inline bool raw_use_hvm_access(const struct vcpu *v)
>>>>> +{
>>>>> +    return IS_ENABLED(CONFIG_HVM) && (!IS_ENABLED(CONFIG_PV) || 
>>>>> is_hvm_vcpu(v));
>>>>> +}
>>>>
>>>> Without a full audit (likely tedious and error prone) this still is a
>>>> behavioral change for some (likely unintended) use against a system domain
>>>> (likely the idle one): With HVM=y PV=n we'd suddenly use the HVM accessor
>>>> there. IOW imo the "system domains are implicitly PV" aspect wants
>>>> retaining, even if only "just in case". It's okay not to invoke the PV
>>>> accessor (but return "len" instead), but it's not okay to invoke the HVM
>>>> one.
>>>
>>> This patch is subset of "constify is_hvm_domain() for PV=n case" attempts.
>>>
>>> It was made under assumption that:
>>> "System domains do not have Guests running, so can't initiate hypecalls and
>>>   can not be users of copy_to/from_user() routines. There are no Guest and 
>>> no user memory".
>>> [IDLE, COW, IO, XEN]
>>>
>>> If above assumption is correct - this patch was assumed safe.
>>>
>>> if not - it all make no sense, probably.
>>
>> I wouldn't go as far as saying that. It can be arranged to avid the corner
>> case I mentioned, I think.
> 
> do you mean adding "&& !is_system_domain(v->domain)" in raw_use_hvm_access()?

No, we want to avoid adding any new any runtime checks.

> Hm, I see that vcpu(s) are not even created for system domains in 
> domain_create().
> So seems !is_system_domain(v->domain) == true always here.

"always" in what sense? It _should_ be always true, but in the unlikely event we
have a path where it isn't (which we could be sure of only after a full audit),
behavior there shouldn't change in the described problematic way.

> Am I missing smth?
> Or you meant smth. else?

I was thinking of something along the lines of

    if ( is_hvm_vcpu(current) )
        return ..._hvm();

    if ( !IS_ENABLED(CONFIG_PV) )
        return len;

    return ..._pv();

Jan