On Mon, 05 Oct 2026, Thomas Zimmermann <[email protected]> wrote: > I have serious doubts about these failure codes. The core issue to me is > that when the kernel driver detects an impossible commit, it probably > knows best how to fix it. Yet the failure codes often lack actionable > meaning.
>From a different perspective, one of my concerns is that not only the error codes become ABI, but also the behaviour becomes ABI. When you're doing atomic check, you usually bail out at the first problem you can't wiggle around. It's hard or impossible to figure out the next problem or a bigger problem, because you've just detected you have a configuration you can't support anyway. You hit the first problem and you report that. Now, let's assume userspace adapts to an error code, retries with different settings, and things work. What if we want to do atomic checks in a different order in the driver, because it makes sense or is better or whatever. Is it a regression to report a different error code because we now do the checks in a different order? Is that painting ourselves in the corner? BR, Jani. -- Jani Nikula, Intel
