Thanks for the review. Yes, this patch was developed with AI assistance. I should have included the required Assisted-by trailer, and I apologize for the omission.
The related issue is pertinent. Lorenzo's follow-up shows that the underlying problem is broader than the fork/mremap path: secretmem accounting follows the VMA lifetime, while the folios remain unevictable until inode eviction. My VM_DONTEXPAND approach only blocks one path and is not an appropriate fix. Please drop this patch. I will test Lorenzo's follow-up patch and report the results on that thread. Thanks, Daehyeon
