Re: [PATCH v3 6/9] xen/arm64: bpi: Add missing code symbol annotations

[Jan Beulich <jbeulich@xxxxxxxx>]

On 07.05.2024 19:37, Julien Grall wrote:
> 
> 
> On 07/05/2024 17:55, Edgar E. Iglesias wrote:
>> On Tue, May 7, 2024 at 11:57 AM Julien Grall <julien@xxxxxxx> wrote:
>> Hi Julien,
> 
> Hi Edgar,
> 
>>
>> The reason I choose FUNC for the start of the symbol is because these
>> symbols contain
>> executable code (not only a table of pointers to code somewhere else)
>> and the ELF spec
>> says that STT_FUNC means the symbol contains functions or other executable
>> code (not only callable functions IIUC):
>>
>> "STT_FUNC The symbol is associated with a function or other executable code."
>> https://refspecs.linuxbase.org/elf/elf.pdf
>> (Symbol Table 1-20).
> 
> Thanks for the pointer. I originally did intend to suggest the change, 
> but then I saw the use of LABEL in x86 (such as svm_stgi_label). There 
> are a few others example with LABEL_LOCAL.
> 
> AFAICT, this is also executable code which the only difference that it 
> is not meant to be called by someone else. Furthermore, LABEL is using 
> DO_CODE_ALIGN(...) for the alignment which imply that it is intended to 
> be used by executable code. So I thought the only difference was whether 
> the label was intended to be used as a function.

No. See below.

>> I think using LABEL instead of GLOBAL for the _end labels of these
>> code sequences makes sense.
>> I'm happy to change the _start labels to LABEL too if you guys feel
>> that's better.
> 
> I have to admit I am little confused with the difference between LABEL 
> vs FUNC. I think I will need some guidance from Jan (he introduced 
> linkage.h).

For annotations the question is what is a "unit" of code. That wants to
be enclosed in FUNC() / END(). Any "inner" entry points or markers would
use LABEL(). On x86 I think it's mainly markers (i.e. addresses pointing
into code which we need e.g. for comparison operations on what Arm would
call PC) where we use LABEL().

Jan