Re: [PATCH 7/8] x86/public: Split the struct cpu_user_regs type

[Jan Beulich <jbeulich@xxxxxxxx>]

On 21.03.2025 16:11, Andrew Cooper wrote:
> On 17/03/2025 12:15 pm, Jan Beulich wrote:
>> On 11.03.2025 22:10, Andrew Cooper wrote:
>>> In order to support FRED, we're going to have to remove the {ds..gs} fields
>>> from struct cpu_user_regs, meaning that it is going to have to become a
>>> different type to the structure embedded in vcpu_guest_context_u.
>>>
>>> struct cpu_user_regs is a name used in common Xen code (i.e. needs to stay
>>> using this name), so renaming the public struct to be guest_user_regs in 
>>> Xen's
>>> view only.
>>>
>>> Introduce a brand hew cpu-user-regs.h, currently containing a duplicate
>>> structure.  This removes the need for current.h to include public/xen.h, and
>>> highlights a case where the emulator was picking up cpu_user_regs
>>> transitively.
>>>
>>> No functional change.
>>>
>>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>> Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>
> 
> Thanks.
> 
>>> cpu_user_regs_t and the guest handle don't seem to be used anywhere.  I'm
>>> tempted to exclude them from Xen builds.
>> I concur. We can always re-expose them should they be needed somewhere.
> 
> It's actually a little ugly to do.
> 
> #ifdef __XEN__
> #undef cpu_user_regs
> #else
> typedef struct cpu_user_regs cpu_user_regs_t;
> DEFINE_XEN_GUEST_HANDLE(cpu_user_regs_t);
> #endif
> 
> and I don't particularly like it, given the complexity of #ifdef-ary
> around it.  Thoughts?

It's not really pretty, but I'd be okay with this.

>>> --- /dev/null
>>> +++ b/xen/arch/x86/include/asm/cpu-user-regs.h
>>> @@ -0,0 +1,69 @@
>>> +/* SPDX-License-Identifier: GPL-2.0-or-later */
>>> +#ifndef X86_CPU_USER_REGS_H
>>> +#define X86_CPU_USER_REGS_H
>>> +
>>> +#define DECL_REG_LOHI(which) union { \
>>> +    uint64_t r ## which ## x; \
>>> +    uint32_t e ## which ## x; \
>>> +    uint16_t which ## x; \
>>> +    struct { \
>>> +        uint8_t which ## l; \
>>> +        uint8_t which ## h; \
>>> +    }; \
>>> +}
>>> +#define DECL_REG_LO8(name) union { \
>>> +    uint64_t r ## name; \
>>> +    uint32_t e ## name; \
>>> +    uint16_t name; \
>>> +    uint8_t name ## l; \
>>> +}
>>> +#define DECL_REG_LO16(name) union { \
>>> +    uint64_t r ## name; \
>>> +    uint32_t e ## name; \
>>> +    uint16_t name; \
>>> +}
>>> +#define DECL_REG_HI(num) union { \
>>> +    uint64_t r ## num; \
>>> +    uint32_t r ## num ## d; \
>>> +    uint16_t r ## num ## w; \
>>> +    uint8_t r ## num ## b; \
>>> +}
>> Can we try to avoid repeating these here? The #undef-s in the public header 
>> are
>> to keep external consumers' namespaces reasonably tidy. In Xen, since we 
>> don't
>> otherwise use identifiers of these names, can't we simply #ifdef-out those
>> #undef-s, and then not re-introduce the same (less the two underscores) here?
>> Granted we then need to include the public header here, but I think that's a
>> fair price to pay to avoid the redundancy.
> 
> Breaking the connection between asm/current.h and public/xen.h is very
> important IMO.  Right now, the public interface/types/defines are in
> every TU, and they absolutely shouldn't be.

Hmm, that's a good point. Nevertheless I wonder if we still couldn't avoid the
unhelpful redundancy. E.g. by introducing a separate, small public header with
just these. Which we'd then pull in here as well.

> Sadly, the compiler isn't happy when including public/xen.h after
> asm/current.h, hence the dropping of the underscores.

Even if the ones here are #undef-ed after use?

> I did have half a mind to expand them fully.  I find them unintuitive,
> but I also didn't think I'd successfully argue that change in.

Roger - do you have an opinion here? I like these wrappers, yet then I also
understand this is code that's pretty unlikely to ever change again. Hence
fully expanding them is an option.

Jan