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

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 23/04/2021 12:08, Jan Beulich wrote:
> On 23.04.2021 13:04, Andrew Cooper wrote:
>> On 23/04/2021 11:58, Jan Beulich wrote:
>>> On 23.04.2021 12:51, Andrew Cooper wrote:
>>>> 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.
>>> I disagree: It is (for now) just you saying this is bogus. The (ab)use
>>> of the handle infrastructure was there before. You could have sent a
>>> fix long ago, therefore, if you were thinking this needs fixing.
>> I only know it needed fixing because you didn't build test your change
>> in CI.  Don't make it out to be my fault I didn't spot this 6 months ago.
>>
>>> I can
>>> see that you have good intentions, but orthogonal issues shouldn't be
>>> used to block necessary adjustments (and this applies to other pending
>>> build fixes as well).
>> You genuinely regressed things for 32bit HVM guests, with the
>> CONFIG_COMPAT change.
>>
>> The code may have been using inappropriate interfaces to perform its job
>> before, but its actually broken now.
> In which way? COMPAT gets selected by both PV32 and HVM.

Hmm ok - with the select in place, I accept that it is only a problem in
principle.

~Andrew