Re: [PATCH v2 2/2] riscv: Resolve symbols directly for symbols with STO_RISCV_VARIANT_CC.

[Palmer Dabbelt <palmer@dabbelt.com>]

On Wed, 19 Jan 2022 18:38:35 PST (-0800), H.J. Lu wrote:
> On Wed, Jan 19, 2022 at 6:22 PM Palmer Dabbelt <palmer@dabbelt.com> wrote:

[snip]

>> Given the complexity around this psABI spec deviation and how close we
>> are to release I'd prefer to wait and see if we can come up with a
>> better solution, though -- for example, I'd been kicking around some
>> ideas related to ELF object attributes saying "this follows the
>> psABI-1.0" vs "this follows the legacy GNU psABI extensions".  That way
>> we could at least tag binaries that explicitly rely on this new behavior
>> as such, which would give us a shot at eventually getting rid of them.
>
> If you want to go this route, I suggest you use GNU property for this.
> ld.so supports GNU property.

Makes sense.

I'd been thinking of essentially just defining one bit for each of these 
incompatibilites, with an absence of them meaning the legacy behavior 
and then a true/false meaning that users have explicitly opted into to 
the spec'd or legacy behavior.  I hadn't gotten as far as actually 
figuring out where to put those bits -- I got hung up on the pc-relative 
vs position-independent one, and that's clearly for after this round of 
releases so I kind of just put it on the backburner.

Given that pretty much all of these are going to need to drive runtime 
behavior, though, it sounds like a GNU property is a reasonable way to 
go.  There's going to be a lot of edge cases here, though, so happy to 
hear if anyone has ideas ;)

Thanks!