Re: [Xen-devel] [PATCH 3 of 3] blktap3/libxl: Handles blktap3 device in libxl

[Thanos Makatos <thanos.makatos@xxxxxxxxxx>]

> ...
> > +            case LIBXL_DISK_BACKEND_TAP3:
> > +                flexarray_append(back, "params");
> > +                flexarray_append(back, libxl__sprintf(gc, "%s:%s",
> > +                    libxl__device_disk_string_of_format(disk-
> >format),
> > +                    disk->pdev_path));
> > +                break;
> >              case LIBXL_DISK_BACKEND_QDISK:
> >                  flexarray_append(back, "params");
> >                  flexarray_append(back, libxl__sprintf(gc, "%s:%s",
> 
> This seems to duplicate much of the entry for LIBXL_DISK_BACKEND_QDISK.
> I think this could be profitably reorganised to avoid clone-and-hack.

You're right, though I think I missed something: tapback spawns tapdisk when 
the front-end starts running the Xenbus protocol, so if it takes too long for 
tapback to spawn the tapdisk (e.g. the host is over-loaded) the domain creation 
will time out. To address this, libxl could tell tapback it's going to need a 
tapdisk on a particular file/disk/whatever (it will also have to explicitly 
tell tapback that it doesn't need the tapdisk anymore). Does this sound 
plausible?

> 
> > diff -r dd63f2992e71 -r 6b54db4abe12 tools/libxl/libxl_device.c
> > --- a/tools/libxl/libxl_device.c    Fri Feb 08 17:23:25 2013 +0000
> > +++ b/tools/libxl/libxl_device.c    Fri Feb 08 17:23:26 2013 +0000
> > @@ -190,6 +190,23 @@ static int disk_try_backend(disk_try_bac
> >          }
> >          return backend;
> >
> > +    case LIBXL_DISK_BACKEND_TAP3:
> > +        /* TODO What's that script thing? */
> > +        if (a->disk->script) goto bad_script;
> 
> Is this series supposed to be an RFC ?  It still has the odd TODO and
> FIXME in it.

No it's not an RFC. Some TODOs are there as guides for possible 
improvements/optimisations.

> 
> A disk script is an external script which arranges to provision a block
> device for use by guests (and dom0).  It's an alternative to
> TAP3 so I think you can just delete the comment.

I'll update the comment.

> 
> Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel