Re: [Qemu-devel] [PATCH v5] vfio-ccw: support async command subregion

[Cornelia Huck <cohuck@redhat.com>]

On Tue, 11 Jun 2019 15:33:59 -0400
Farhan Ali <alifm@linux.ibm.com> wrote:

> On 06/07/2019 10:53 AM, Cornelia Huck wrote:
> > A vfio-ccw device may provide an async command subregion for
> > issuing halt/clear subchannel requests. If it is present, use
> > it for sending halt/clear request to the device; if not, fall
> > back to emulation (as done today).
> > 
> > Signed-off-by: Cornelia Huck <cohuck@redhat.com>
> > ---
> > 
> > v4->v5:
> > - It seems we need to take the indirection via the class for the
> >    callbacks after all :(
> > - Dropped Eric's R-b: for that reason
> > 
> > ---
> >   hw/s390x/css.c              |  27 +++++++--
> >   hw/s390x/s390-ccw.c         |  20 +++++++
> >   hw/vfio/ccw.c               | 112 +++++++++++++++++++++++++++++++++++-
> >   include/hw/s390x/css.h      |   3 +
> >   include/hw/s390x/s390-ccw.h |   2 +
> >   5 files changed, 158 insertions(+), 6 deletions(-)
> > 

> > @@ -309,11 +400,26 @@ static void vfio_ccw_get_region(VFIOCCWDevice *vcdev, Error **errp)
> >       vcdev->io_region_offset = info->offset;
> >       vcdev->io_region = g_malloc0(info->size);
> >   
> > +    /* check for the optional async command region */
> > +    ret = vfio_get_dev_region_info(vdev, VFIO_REGION_TYPE_CCW,
> > +                                   VFIO_REGION_SUBTYPE_CCW_ASYNC_CMD, &info);
> > +    if (!ret) {
> > +        vcdev->async_cmd_region_size = info->size;
> > +        if (sizeof(*vcdev->async_cmd_region) != vcdev->async_cmd_region_size) {
> > +            error_setg(errp, "vfio: Unexpected size of the async cmd region");
> > +            g_free(info);
> > +            return;
> > +        }
> > +        vcdev->async_cmd_region_offset = info->offset;
> > +        vcdev->async_cmd_region = g_malloc0(info->size);
> > +    }
> > +
> >       g_free(info);
> >   }
> >   
> >   static void vfio_ccw_put_region(VFIOCCWDevice *vcdev)
> >   {
> > +    g_free(vcdev->async_cmd_region);
> >       g_free(vcdev->io_region);
> >   }  
> 
> I think we can have a memory leak given how the code is currently 
> structured and how we call vfio_ccw_get_region.
> 
> vfio_ccw_get_region is called in vfio_ccw_realize. So if we return an 
> error from vfio_ccw_get_region, we would jump to out_region_err in 
> vfio_ccw_realize which would call vfio_ccw_put_device.
> 
> Now we can also return an error from vfio_ccw_get_region for the async 
> region, and so we might never end up freeing the io_region for which we 
> allocated memory successfully.
> 
> I think we would also need to change vfio_ccw_realize, no?

Indeed, you're right. I have the following change on top:

diff --git a/hw/vfio/ccw.c b/hw/vfio/ccw.c
index c9d1c76b4d04..d21ac24f743c 100644
--- a/hw/vfio/ccw.c
+++ b/hw/vfio/ccw.c
@@ -407,6 +407,7 @@ static void vfio_ccw_get_region(VFIOCCWDevice *vcdev, Error **errp)
         vcdev->async_cmd_region_size = info->size;
         if (sizeof(*vcdev->async_cmd_region) != vcdev->async_cmd_region_size) {
             error_setg(errp, "vfio: Unexpected size of the async cmd region");
+            g_free(vcdev->io_region);
             g_free(info);
             return;
         }

Anything else before I respin?