[PATCH 00/12] S390: Implement __fentry__

[Florian Weimer]

On 08/02/2018 12:43 PM, Ilya Leoshkevich wrote:
> 
>> Am 02.08.2018 um 11:17 schrieb Florian Weimer <fweimer@redhat.com>:
>>
>> On 08/02/2018 09:57 AM, Ilya Leoshkevich wrote:
>>> This patch series adds the runtime support in glibc for the -mfentry
>>> gcc feature introduced in [1] and [2].
>>> Patches 1-9 deal with avoiding clobbering %r0 when calling lazily bound
>>> functions, so that the new __fentry__ symbol could be called with return
>>> address in that register.
>>
>> Should GCC arrange for suitable ABI markup if -mfentry is used?  I'm worried that unexpected clobbers of r0 could be quite difficult to figure out otherwise.
> 
> I don’t think we need to do anything ABI-related at this point.
> 
> __fentry__ itself is being newly introduced, so there is no old ABI to
> comply with. Instrumented functions’ ABI does not change, particularly,
> %r0, being a volatile register, used to be clobbered anyway.
> 
> Are there other cases that I’m missing?

Agreed, with the __fentry__ addition in the same release as the 
trampoline change, this should be safe and avoid the issue I was worried 
about.

Thanks,
Florian