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

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

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?

>
>> --- /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.

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

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.

I'm not terribly fussed how we do this, but I really do want to reduce
the header tangle.

>
>> +struct cpu_user_regs
>> +{
>> +    DECL_REG_HI(15);
>> +    DECL_REG_HI(14);
>> +    DECL_REG_HI(13);
>> +    DECL_REG_HI(12);
>> +    DECL_REG_LO8(bp);
>> +    DECL_REG_LOHI(b);
>> +    DECL_REG_HI(11);
>> +    DECL_REG_HI(10);
>> +    DECL_REG_HI(9);
>> +    DECL_REG_HI(8);
>> +    DECL_REG_LOHI(a);
>> +    DECL_REG_LOHI(c);
>> +    DECL_REG_LOHI(d);
>> +    DECL_REG_LO8(si);
>> +    DECL_REG_LO8(di);
>> +    uint32_t error_code;
>> +    uint32_t entry_vector;
>> +    DECL_REG_LO16(ip);
>> +    uint16_t cs, _pad0[1];
>> +    uint8_t  saved_upcall_mask;
>> +    uint8_t  _pad1[3];
>> +    DECL_REG_LO16(flags); /* rflags.IF == !saved_upcall_mask */
>> +    DECL_REG_LO8(sp);
>> +    uint16_t ss, _pad2[3];
>> +    uint16_t es, _pad3[3];
>> +    uint16_t ds, _pad4[3];
>> +    uint16_t fs, _pad5[3];
>> +    uint16_t gs, _pad6[3];
> I had to peek ahead at the last patch to figure why you introduce these 4 
> fields
> (plus their padding) here, just to remove them again. Personally I think it 
> would
> be neater if both were folded; nevertheless I'd like to leave this entirely to
> you.

While both patches are reasonably small, I think it's important for
bisection to keep them separate.  They're both complex in separate ways.

~Andrew