Hello, On Wed, Aug 30, 2023 at 2:16 AM Samuel Thibault <samuel.thiba...@gnu.org> wrote: > Just to make sure, may your work be affected by this? > > https://lists.gnu.org/archive/html/bug-hurd/2017-10/msg00011.html
Yes, I learned about that patch just after having sent mine. > We had a patch in Debian that was disabling the coalescing because it > could potentially coalesce areas not to be swappable (such as IDE write > buffers) with swappable areas. Not that I have any idea how IDE drivers work, but that patch sounds rather pointless, since that very 'if' condition already checks for the vm_entry being wired (= not swappable), and doesn't coalesce in that case. Also, according to [0], that patch fixed nothing at the time. So it should be safe to just drop it, which I see you have already done. [0]: https://mail.gnu.org/archive/html/bug-hurd/2017-10/msg00013.html Related: I hacked up vm-object-create (the ramdisk replacement that exposes the original pages instead of copying to kmalloc memory). I haven't quite tested it because libstore doesn't support specifying memobj port name on the command line, and I haven't gotten around to implementing that. Would you still be interested in reviewing / testing those patches, and eventually using it in Debian instead of ramdisk? It would be great if Richard could review it too, since it touches some of the vm_page things that I don't really understand. Sergey
