On 07.05.2021 10:21, Roger Pau Monné wrote:
> On Fri, May 07, 2021 at 08:22:38AM +0200, Jan Beulich wrote:
>> In this case compat headers don't get generated (and aren't needed).
>> The changes made by 527922008bce ("x86: slim down hypercall handling
>> when !PV32") also weren't quite sufficient for this case.
>>
>> Try to limit #ifdef-ary by introducing two "fallback" #define-s.
>>
>> Fixes: d23d792478db ("x86: avoid building COMPAT code when !HVM && !PV32")
>> Reported-by: Andrew Cooper <[email protected]>
>> Signed-off-by: Jan Beulich <[email protected]>
>>
>> --- a/xen/arch/x86/pv/shim.c
>> +++ b/xen/arch/x86/pv/shim.c
>> @@ -34,8 +34,6 @@
>> #include <public/arch-x86/cpuid.h>
>> #include <public/hvm/params.h>
>>
>> -#include <compat/grant_table.h>
>> -
>> #undef virt_to_mfn
>> #define virt_to_mfn(va) _mfn(__virt_to_mfn(va))
>>
>> @@ -300,8 +298,10 @@ static void write_start_info(struct doma
>> &si->console.domU.mfn) )
>> BUG();
>>
>> +#ifdef CONFIG_PV32
>> if ( compat )
>> xlat_start_info(si, XLAT_start_info_console_domU);
>> +#endif
>
> Would it help the compiler logic if the 'compat' local variable was
> made const?
No, because XLAT_start_info_console_domU is undeclared when there are
no compat headers.
> I'm wondering if there's a way we can force DCE from the compiler and
> avoid the ifdefs around the usage of compat.
The issue isn't with DCE - I believe the compiler does okay in that
regard. The issue is with things simply lacking a declaration /
definition. That's no different from e.g. struct fields living
inside an #ifdef - all uses then need to as well, no matter whether
the compiler is capable of otherwise recognizing the code as dead.
Jan
