On Tue, 24 Feb 2026, "Murthy, Arun R" <[email protected]> wrote: > On 24-02-2026 14:58, Jani Nikula wrote: >> On Tue, 24 Feb 2026, Ville Syrjälä <[email protected]> wrote: >>> Although I kinda doubt its actual usefulness to drive useful >>> fallback logic because often the restrictions might be a combination >>> of many things, and the kernel can only realistically report one of >>> those things. >> Yeah, this is my main concern as well. The drivers will have to bail out >> on the first issue they hit, whatever it is. The drivers may choose to >> do the checks in different orders, resulting in different failure modes >> for different drivers. And finally, accidentally making the order of the >> checks part of the ABI contract is a scary prospect. Imagine user space >> depending on certain checks happening first in order for the fallback >> logic to work properly. Is it a kernel regression to change the order of >> the checks then? > We are just reporting the 1st error that we see in the KMD and return > from there.
Yes. But we can't guarantee all drivers will report the *same* first error in the same circumstances. We can't guarantee we will maintain the *same* first error over time, we can't make that promise without painting ourselves in the corner wrt driver maintenance. BR, Jani. -- Jani Nikula, Intel
