Hi all, On 27/10/2026 21:54, Pavel Begunkov wrote: > Hey Li, > > On 10/27/26 09:08, Li Wang wrote: ... > > xfs support it a it orthogonal to the uapi question, but it should be > implemented in xfs and not side stepped. There is an RFC on the mailing > list from Anuj and Nitesh that is trying to do that. > Thanks for the feedback! As far as we know, Pavel's current work only supports direct data transfer between GPU memory and NVMe raw block devices. As Pavel pointed out, I tended to prefer an implementation that was as non-intrusive to the filesystem as possible. My rationale was that GPU-NVME direct data transfer would ideally not require filesystem opt-in, since there was no readahead, no caching, and no complex metadata operations involved here. In most cases, the filesystem merely helps translate file offsets into disk block addresses. Additionally, adopting an ioctl interface had a smaller impact scope and provided a simpler userspace synchronous operation interface.
We will track the community progress, and we appreciate being let know if there is anything we can help contribute to, as we are active users of this feature. Thanks, Li Wang
