Re: [PATCH] [RFC] qemu-upstream: add discard support for xen_disk

[Olaf Hering <olaf@aepfle.de>]

On Fri, Jan 17, Stefano Stabellini wrote:

> Can't we just assume that if the backend can do discard on that file, it
> is simply going to enable feature-discard? Do we really need the
> toolstack driven option too?

If the backing file is fully ppopulated on purpose to avoid
fragmentation of that file, then silently enabling discard means the
unwanted fragmentation will occour over time. If anyone is really doing
that in their setup I certainly dont know. At least kvm/qemu has such
knob, and also an hyperv host offers sparse, fully populated and overlay
as possible option when creating a new virtual disk image.

Now I think that if libxl already writes feature-discard=0, qemu can use
that to clear BDRV_O_UNMAP. Also blkbk could be extended to first check
if the property already exists before blindly enabling discard. No idea
if that makes sense for "phy", but you never know.

Olaf