Re: [Qemu-devel] [PATCH 14/18] xen: add implementations of xen-qdisk connect and disconnect functions...

[Kevin Wolf <kwolf@redhat.com>]

Am 29.11.2018 um 10:33 hat Paul Durrant geschrieben:
> > -----Original Message-----
> > From: Kevin Wolf [mailto:kwolf@redhat.com]
> > Sent: 29 November 2018 09:01
> > To: Paul Durrant <Paul.Durrant@citrix.com>
> > Cc: qemu-block@nongnu.org; qemu-devel@nongnu.org; xen-
> > devel@lists.xenproject.org; Stefano Stabellini <sstabellini@kernel.org>;
> > Anthony Perard <anthony.perard@citrix.com>; Max Reitz <mreitz@redhat.com>
> > Subject: Re: [PATCH 14/18] xen: add implementations of xen-qdisk connect
> > and disconnect functions...
> > 
> > Am 28.11.2018 um 17:40 hat Paul Durrant geschrieben:
> > > > -----Original Message-----
> > > > From: Kevin Wolf [mailto:kwolf@redhat.com]
> > > > Sent: 28 November 2018 16:35
> > > > To: Paul Durrant <Paul.Durrant@citrix.com>
> > > > Cc: qemu-block@nongnu.org; qemu-devel@nongnu.org; xen-
> > > > devel@lists.xenproject.org; Stefano Stabellini
> > <sstabellini@kernel.org>;
> > > > Anthony Perard <anthony.perard@citrix.com>; Max Reitz
> > <mreitz@redhat.com>
> > > > Subject: Re: [PATCH 14/18] xen: add implementations of xen-qdisk
> > connect
> > > > and disconnect functions...
> > > >
> > > > Am 21.11.2018 um 16:12 hat Paul Durrant geschrieben:
> > > > > ...and wire in the dataplane.
> > > > >
> > > > > This patch adds the remaining code to make the xen-qdisk XenDevice
> > > > > functional. The parameters that a block frontend expects to find are
> > > > > populated in the backend xenstore area, and the 'ring-ref' and
> > > > > 'event-channel' values specified in the frontend xenstore area are
> > > > > mapped/bound and used to set up the dataplane.
> > > > >
> > > > > Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
> > > > > ---
> > > > > Cc: Stefano Stabellini <sstabellini@kernel.org>
> > > > > Cc: Anthony Perard <anthony.perard@citrix.com>
> > > > > Cc: Kevin Wolf <kwolf@redhat.com>
> > > > > Cc: Max Reitz <mreitz@redhat.com>
> > > > > ---
> > > > >  hw/block/xen-qdisk.c       | 140
> > > > +++++++++++++++++++++++++++++++++++++++++++++
> > > > >  hw/xen/xen-bus.c           |  12 ++--
> > > > >  include/hw/xen/xen-bus.h   |   8 +++
> > > > >  include/hw/xen/xen-qdisk.h |  12 ++++
> > > > >  4 files changed, 166 insertions(+), 6 deletions(-)
> > > > >
> > > > > diff --git a/hw/block/xen-qdisk.c b/hw/block/xen-qdisk.c
> > > > > index 35f7b70480..8c88393832 100644
> > > > > --- a/hw/block/xen-qdisk.c
> > > > > +++ b/hw/block/xen-qdisk.c
> > > > > @@ -9,6 +9,10 @@
> > > > >  #include "qapi/visitor.h"
> > > > >  #include "hw/hw.h"
> > > > >  #include "hw/xen/xen-qdisk.h"
> > > > > +#include "sysemu/blockdev.h"
> > > > > +#include "sysemu/block-backend.h"
> > > > > +#include "sysemu/iothread.h"
> > > > > +#include "dataplane/xen-qdisk.h"
> > > > >  #include "trace.h"
> > > > >
> > > > >  static char *xen_qdisk_get_name(XenDevice *xendev, Error **errp)
> > > > > @@ -23,6 +27,11 @@ static void xen_qdisk_realize(XenDevice *xendev,
> > > > Error **errp)
> > > > >  {
> > > > >      XenQdiskDevice *qdiskdev = XEN_QDISK_DEVICE(xendev);
> > > > >      XenQdiskVdev *vdev = &qdiskdev->vdev;
> > > > > +    BlockConf *conf = &qdiskdev->conf;
> > > > > +    DriveInfo *dinfo;
> > > > > +    bool is_cdrom;
> > > > > +    unsigned int info;
> > > > > +    int64_t size;
> > > > >
> > > > >      if (!vdev->valid) {
> > > > >          error_setg(errp, "vdev property not set");
> > > > > @@ -30,13 +39,134 @@ static void xen_qdisk_realize(XenDevice
> > *xendev,
> > > > Error **errp)
> > > > >      }
> > > > >
> > > > >      trace_xen_qdisk_realize(vdev->disk, vdev->partition);
> > > > > +
> > > > > +    if (!conf->blk) {
> > > > > +        error_setg(errp, "drive property not set");
> > > > > +        return;
> > > > > +    }
> > > > > +
> > > > > +    if (!blk_is_inserted(conf->blk)) {
> > > > > +        error_setg(errp, "device needs media, but drive is empty");
> > > > > +        return;
> > > > > +    }
> > > >
> > > > Hm, the code below suggests that you support CD-ROMs. Don't you want
> > to
> > > > support media change as well then? Which would mean that you need to
> > > > support empty drives.
> > >
> > > Yes, that's a good point. I should get rid of that check.
> > 
> > Or rather apply it only to hard disks. And for empty CDs, you'll
> > probably need to create an empty BlockBackend (the !conf->blk case).
> > Just check the IDE and/or SCSI code for comparison.
> > 
> > > >
> > > > > +    if (!blkconf_apply_backend_options(conf, blk_is_read_only(conf-
> > > > >blk),
> > > > > +                                       false, errp)) {
> > > > > +        return;
> > > > > +    }
> > > > > +
> > > > > +    if (!blkconf_geometry(conf, NULL, 65535, 255, 255, errp)) {
> > > > > +        return;
> > > > > +    }
> > > > > +
> > > > > +    dinfo = blk_legacy_dinfo(conf->blk);
> > > > > +    is_cdrom = (dinfo && dinfo->media_cd);
> > > >
> > > > It's called legacy for a reason. Don't use this in new devices.
> > > >
> > > > The proper way is to have two different devices for hard disks and CDs
> > > > (like scsi-hd and scsi-cd).
> > >
> > > ...or presumably I could have a property? The legacy init code could
> > > then set it based on the drive info.
> > 
> > Technically yes, but why would that be a good way to model things? I
> > mean, it's true that xen-qdisk is not real hardware, but I've never seen
> > any hardware that has a switch to decide whether it should behave as a
> > CD drive or a hard disk.
> > 
> > Both have very different characteristics (read-only with removable
> > media, or a single read-write disk), and the existing implementations
> > use two separate devices. So even if you're not convinced that users
> > will consider them different concepts (I am; and if they weren't
> > different concepts, you wouldn't need an is_cdrom variable), consistency
> > is still a good thing.
> 
> Ok. I'll split the device as you suggest... it may mean duplicated
> code, but the datapath can still be common.

If you have a look at IDE and SCSI, they don't really duplicate a lot of
code. Basically it's just a second QOM class definition, the rest is
shared. Even the realize functions are essentially shared, with just two
small wrappers for each device type around the common code.

Kevin