Re: [PATCH 2/3] powerpc64le: Check HWCAP bits against compiler build flags

["Lucas A. M. Magalhaes" <lamm@linux.ibm.com>]

Quoting Florian Weimer (2021-05-12 05:27:33)
> * Lucas A. M. Magalhaes:
> 
> > Hi Florian. I've tested this patch on POWER[8..10] with -mcpu=power9 and
> > -mcpu=power10 built with GCC 10.3.1.
> >
> > When built with -mcpu=power10 I got a SIGILL on POWER8 and POWER10.
> 
> Do you mean POWER8 and POWER9?
> 
Yes. Sorry about that.

> > For testing I'm just running ./libc.so.  I try to inspect it with
> > gdb. With -mcpu=power9 I can see it passing through dl_hwcap_check
> > but not with -mcpu=power10.
> >
> > I maybe doing something wrong though.
> 
> I tried to allude in the commit message that this is expected.  Maybe I
> should drop the POWER10 check?
> 
Your commit message is fine, I misunderstood it. IMHO it's better to
drop POWER10 check and add a comment explaining why it's not there.

> I do not think this is fixable in the long term because I expect that
> POWER10 builds won't need the bootstrap relocation in the dynamic loader
> thanks to its PCREL instructions.  (POWER10 should be
> PI_STATIC_AND_HIDDEN.)  We could disable POWER10 code for early parts of
> the dynamic loader, but that will only work until we start to rely on
> PI_STATIC_AND_HIDDEN, which has other benefits.
I'm not familiar with this bootstrap relocation but for what I understood
there is early code using PCREL instructions so it SIGILL before running
this code, right?

I don't oppose this change but still don't get why do we need this code?
What's the benefit of the message instead of just SIGILL?

---
Lucas A. M. Magalhães