On Thursday, 27 November 2025 11:46:05 CET Jani Nikula wrote:
On Thu, 27 Nov 2025, Janusz Krzysztofik janusz.krzysztofik@linux.intel.com wrote:
To my taste, zeroing on allocation would be a more clean solution.
IIUC there are micro optimizations to not clear on allocation when you don't strictly have to...
I'm not advocating one or the other approach, just stating what I believe is the reason.
OK, good to hear there is still someone who is able to recall what the reason could be when no hints can be found in git history nor inline comments.
If that's the case, but we agree on pre-zeroing only the sub-area dedicated to the vma table rather than doing that on failure and limited to one element that follows the one that failed, as Krzysztof initially proposed, then I'd vote for restoring memset() that was dropped with commit 170fa29b14fad ("drm/ i915: Simplify eb_lookup_vmas()"). In any case, a clarification (in commit description or inline comment) on why we chose one solutions and not the another wouldn't hurt.
Thanks, Janusz
BR, Jani.