[PATCH v15 5/8] nptl: Introduce <rseq-access.h> for RSEQ_* accessors

[Michael Jeanson]

On 2025-01-08 04:17, Florian Weimer wrote:
> * Michael Jeanson:
> 
>> In preparation to move the rseq area to the 'extra TLS' block, we need
>> accessors based on the thread pointer and the rseq offset. The VOLATILE
>> variant of the accessors ensures single-copy atomicity for loads and
>> stores which is required for all fields once the registration is active.
> 
> This isn't generally how we use the term “volatile”, though.  Torn reads
> and writes perfectly fine for volatile access.  Maybe it's closer to
> relaxed memory order?  Or if you don't really like that, use ONCE in the
> macro names?

I like "ONCE", I initially used VOLATILE because I derived these macros
from the tcb accessors macros but these new ones do indeed offer
stricter guarantees, I'll update the names.

> 
> The implementation looks okay, but code-generation on Clang is
> suboptimal for x86-64.  It reloads __rseq_offset from memory for each
> RSEQ_GETMEM_VOLATILE usage, while GCC keeps referring to the
> already-loaded value in a register.  We can rework this later if this
> proves to be a problem.

I'll make a note to look into this once the patchset is merged.

> 
> Thanks,
> Florian
>