Re: [PATCH] x86/oprof: fix !HVM && !PV32 build

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 23/04/2021 10:50, Roger Pau Monné wrote:
> On Fri, Apr 16, 2021 at 04:20:59PM +0200, Jan Beulich wrote:
>> On 16.04.2021 15:41, Andrew Cooper wrote:
>>> On 16/04/2021 09:16, Jan Beulich wrote:
>>>> clang, at the very least, doesn't like unused inline functions, unless
>>>> their definitions live in a header.
>>>>
>>>> Fixes: d23d792478 ("x86: avoid building COMPAT code when !HVM && !PV32")
>>>> Reported-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>>>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
>>> I agree this will fix the build.  However, looking at the code, I'm not
>>> sure the original CONFIG_COMPAT was correct.  In particular, ...
>>>
>>>> --- a/xen/arch/x86/oprofile/backtrace.c
>>>> +++ b/xen/arch/x86/oprofile/backtrace.c
>>>> @@ -43,6 +43,7 @@ dump_hypervisor_backtrace(struct vcpu *v
>>>>      return head->ebp;
>>>>  }
>>>>  
>>>> +#ifdef CONFIG_COMPAT
>>>>  static inline int is_32bit_vcpu(struct vcpu *vcpu)
>>>>  {
>>>>      if (is_hvm_vcpu(vcpu))
>>> ... this chunk of logic demonstrates that what oprofile is doing isn't
>>> related to the Xen ABI in the slightest.
>>>
>>> I think OProfile is misusing the guest handle infrastructure, and
>>> shouldn't be using it for this task.
>> I'm afraid I consider this something for another day. Both the
>> original #ifdef and the one getting added here are merely
>> measures to get things to build.
> Acked-by: Roger Pau Monné <roger.pau@xxxxxxxxxx>
>
> Without entering on the debate whether CONFIG_COMPAT is the correct
> conditional to use it's not making the issue any worse, and it will
> allow to unblock the build. We can discuss about the CONFIG_COMPAT
> stuff later.

I disagree.  Fixing this less effort than the time wasted arguing about
fixing it.

But if you are going to insist on not fixing it, and putting in a patch
like this, then at a minimum, it needs to include a TODO comment stating
that the use of CONFIG_COMPAT is bogus and needs fixing.

~Andrew