Re: [PATCH v3] rs6000: Rework ELFv2 support for -fpatchable-function-entry* [PR99888]

["Kewen.Lin" <linkw@linux.ibm.com>]

Hi Segher,

on 2022/8/23 22:33, Segher Boessenkool wrote:
> On Fri, Aug 19, 2022 at 10:40:10AM +0800, Kewen.Lin wrote:
>> Since you proposed to update the documentation, I'm thinking if we can
>> reconsider Fangrui's proposal in the PR which Alan seconded: Put preceding
>> nops before GEP and succeeding nops after LEP.  Previously I had the concern
>> that the nops inserted doesn't respect to a same function entry, it looks
>> inconsistent to the documentation, and you also noted that "The nops have
>> to be consecutive".  If we want to update the documentation, could we reword
>> it for PowerPC ELFv2 ABI?
>>
>> What's your opinion?
> 
> I'm not sure what the question is, sorry.
> 

Sorry for confusion.  The question is that if we can consider the proposal
in [1], by noting the particularity on ppc64le in documentation.

btw, I did some searching on why the feature supports preceding nops, and
commented it in [2].

> If you want different semantics for ELFv2 (which might well be useful),
> we need some new command line option for that.
> 

Not sure if we really needs one new command line option, dual entries on
ppc64le is special comparing with a normal unique function entry, couldn't
it be a special case in the documentation?

> I suggested here to just describe in the existing doc what is done for
> global and local entry points on ELFv2.
> 

Yeah, you have suggested nice wordings "For PowerPC with the ELFv2 ABI,
there will be M nops before the local entry point, and N-M after".

I thought if we can consider [1] and updated the documentation similarly
like "For PowerPC with the ELFv2 ABI, there will be M nops before the global
entry point, and N-M after the local entry point".

[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99888#c5
[2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99888#c10

BR,
Kewen