On 02.10.2024 11:07, Andrew Cooper wrote:
> On 02/10/2024 8:41 am, Jan Beulich wrote:
>> Further use the generic framework from xen/linkage.h. While there drop
>> excess alignment and move to .bss.
>>
>> Requested-by: Andrew Cooper <[email protected]>
>> Signed-off-by: Jan Beulich <[email protected]>
>> ---
>> Of course alongside ASM_INT() we could introduce ASM_QUAD() and
>> ASM_QUAD_LOCAL() (only the latter needed right here) to aid readability.
>> Thoughts?
> 
> Honestly, ASM_INT() hiding a .long is confusing enough already.
> 
> ASM_C_{INT,LONG}() wouldn't be as bad.  At least they're clear about
> being a particular type in another language.

I don't think the _C_ would add much; we all know C is the language Xen
is written in. ASM_LONG() / ASM_C_LONG() would not be generalizable, i.e.
couldn't be put in xen/linkage.h without arch customization, as that can
neither expand uniformly to .long nor to .quad. It's all solvable, but
would be getting involved.

>> --- a/xen/arch/x86/acpi/wakeup_prot.S
>> +++ b/xen/arch/x86/acpi/wakeup_prot.S
>> @@ -1,3 +1,5 @@
>> +#define DATA_FILL 0 /* For the .bss contributions at the bottom. */
>> +
> 
> I really feel that here is the wrong place for this to live.
> 
> Why isn't it in xen/linkage.h?  When is data typically padded with
> anything other than 0's?

As per what we currently have, the default data padding is ~0. Personally
I consider this marginally better than 0, but it could of course be
changed.

Jan

Reply via email to