Hi, On Mon, Dec 9, 2019 at 8:38 PM Dmitry Sepp <dmitry.sepp@xxxxxxxxxxxxxxx> wrote: > > Hi, > > On Montag, 9. Dezember 2019 11:46:15 CET Gerd Hoffmann wrote: > > Hi, > > > > > > For (1) you'll simply do a QUEUE_BUFFER. The command carries references > > > > to the buffer pages. No resource management needed. > > > > > > > > For (2) you'll have RESOURCE_CREATE + RESOURCE_DESTROY + QUEUE_RESOURCE, > > > > where RESOURCE_CREATE passes the scatter list of buffer pages to the > > > > host and QUEUE_RESOURCE will carry just the resource id. > > > > > > > > For (3) you'll have RESOURCE_IMPORT + RESOURCE_DESTROY + QUEUE_RESOURCE. > > > > > > Thanks for the clarification. > > > On second thought, however, I'm wondering if we should keep all > > > RESOURCE controls. > > > This should be an option (4). > > > > Well, that's actually (2), aka "we use RESOURCE_* commands to manage > > resources". With the commands in the latest draft that would be: > > > > (1) RESOURCE_CREATE > > (2) RESOURCE_ATTACH_BACKING > > (3) RESOURCE_QUEUE (resource can be used multiple times here) > > (4) RESOURCE_DETACH_BACKING > > (5) RESOURCE_DESTROY > > > > I'm not sure we need the separate *_BACKING commands. I think we could > > also have RESOURCE_CREATE pass the buffer pages scatter list instead. > > > > Why it is done this way? Just because the design was copied from > > virtio-gpu? Or is there some specific reason? > > Yes, the design was just derived from virtio-gpu at early stages. > > I do agree we should merge the two steps. Thanks for the explanation. Merging them sounds good to me. -Keiichi > > > > > cheers, > > Gerd > >
