On Thu, Mar 20, 2025 at 09:34:50AM -0700, Bart Van Assche wrote: > The fact that this change is proposed because there are device > manufacturers that want to produce devices with block sizes larger than > 64 KiB would be useful information for the commit message. I think it's just the proposal that is very confused. Keith pointed out that number of segments mostly matters when you have 4k non-contiguous pages and that's where the quoted limit comes from. The LBA size is a lower bound to the folio size in the page cache, so the limit automatically increases with that, although in practice the file system will usually allocate larger folios even with a smaller block size, and I assume most systems (or at least the ones that care about performance) actually use transparent huge pages/folios for anonymous memory as well.
