Re: [RFC PATCH 00/19] riscv: ifunc support with optimized mem*/str*/cpu_relax routines

["Christoph Müllner" <christoph.muellner@vrull.eu>]

[-- Attachment #1: Type: text/plain, Size: 2172 bytes --]

On Tue, Feb 7, 2023 at 10:14 PM DJ Delorie <dj@redhat.com> wrote:

>
> Philipp Tomsich <philipp.tomsich@vrull.eu> writes:
> > So this patch has always been a stand-in until option #2 is ready.
> > I am strongly opinionated towards a mechanism that uses existing
> > mechanisms in the ELF auxiliary vector to pass information — and tries
> > to avoid the introduction of a new arch-specific syscall. if possible.
>
> If the patch were converted to use tunables, it could be more than a
> standin.  It's the environment variable itself I'm opposed to.
>

Thanks DJ and  Adhemerval for your valuable inputs!

As said in the cover letter, the environment variable approach was not
meant to be merged but represents a starting point for discussions.
It is not what we want but serves as a dirty placeholder, that allows
development of optimized routines.

The IFUNC support and the kernel-userspace API was discussed
multiple times in the past (e.g. on LPC2021 and LPC2022).
There are different opinions on the approaches so the whole process
is regularly getting stuck.

The topic, where most (if not all) in the RISC-V community agree on,
is that we don't want a compile-time-only approach.
Patches that rely on a compile-time-only approach are most likely
written such, because of the absence of ifunc support.

Meanwhile, we have heard from multiple vendors to work on their own
solutions downstream, which results in a duplication of work and not
necessarily in a common solution upstream.

This patchset was sent out to move the discussion from the idea level
to actual code, which can be reviewed, criticized, tested, improved, and
reused (get a common ground for RISC-V vendors).

Both of your comments show how we can move this patchset forward:
- eliminate the first patch
  See also: https://sourceware.org/bugzilla/show_bug.cgi?id=30095
- work on the kernel-userspace interface to query capabilities
- use tunables instead of env variable
- look how we can reuse more of the generic implementation
  to minimize ASM code with little or no benefits

I fully agree with all the mentioned points.

Thanks,
Christoph