Re: [Xen-devel] [PATCH v2] libx86: Work around GCC bug with ebx output constrants

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 19.11.18 at 16:19, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 19/11/2018 15:14, Jan Beulich wrote:
>>>>> On 19.11.18 at 15:45, <andrew.cooper3@xxxxxxxxxx> wrote:
>>> Versions of GCC before 5 can't compile cpuid.c, and fail with the rather 
>>> cryptic:
>>>
>>>   In file included from lib/x86/cpuid.c:3:0:
>>>   lib/x86/cpuid.c: In function ‘x86_cpuid_policy_fill_native’:
>>>   include/xen/lib/x86/cpuid.h:25:5: error: inconsistent operand constraints 
> in an ‘asm’
>>>        asm ( "cpuid"
>>>        ^
>>>
>>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>>> ---
>>> CC: Jan Beulich <JBeulich@xxxxxxxx>
>>> CC: Wei Liu <wei.liu2@xxxxxxxxxx>
>>> CC: Roger Pau Monné <roger.pau@xxxxxxxxxx>
>>>
>>> v2:
>>>  * GCC 5 is fine.  Its cpuid instrinct has none of the PIC workarounds 
>>> thant 
> 4.9 have.
>>>  * Fix 64bit builds with larger models.
>> It is rather odd that 64-bit is also affected - the error gets raised
>> even when there's no use of %rbx for GOT accesses. By when
>> they need a callee-saved register, they indeed use %rbx instead
>> to the ABI-suggested %r15.
>>
>>> --- a/xen/include/xen/lib/x86/cpuid.h
>>> +++ b/xen/include/xen/lib/x86/cpuid.h
>>> @@ -20,21 +20,50 @@ struct cpuid_leaf
>>>      uint32_t a, b, c, d;
>>>  };
>>>  
>>> +/*
>>> + * Versions of GCC before 5 are unable to cope with %rBX output constraints
>>> + * when compiling Position Independent Code, and produce a rather cryptic
>>> + * error:
>>> + *    error: inconsistent operand constraints in an ‘asm’
>>> + *
>>> + * To work around the issue, use a separate register to hold the the %rBX
>>> + * output, and xchg twice to leave %rBX preserved around the asm() 
> statement.
>>> + */
>>> +#if defined(__PIC__) && __GNUC__ < 5 && !defined(__clang__) && 
>>> defined(__i386__)
>>> +# define XCHG_BX "xchg %%ebx, %[bx];"
>>> +# define BX_CON [bx] "=&r"
>>> +#elif defined(__PIC__) && __GNUC__ < 5 && !defined(__clang__) && \
>>> +    defined(__x86_64__) && (defined(__code_model_medium__) || \
>>> +                            defined(__code_model_large__))
>>> +# define XCHG_BX "xchg %%rbx, %q[bx];"
>>> +# define BX_CON [bx] "=&r"
>>> +#else
>>> +# define XCHG_BX ""
>>> +# define BX_CON "=&b"
>> The & is unnecessary here I think. Preferably with it dropped
>> Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>
> 
> That was to match the "=&d" and friends.  I'd prefer to be consistent
> (one way or the other).

Ah, well, yes - that's fine then.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel