On Mon, 25 Jun 2018 08:44:52 +0200 Gerd Hoffmann <kraxel@xxxxxxxxxx> wrote: > There is no default implementation for dma_buf_ops->unmap. > So add a function unmapping the page, otherwise we'll leak them. > > Signed-off-by: Gerd Hoffmann <kraxel@xxxxxxxxxx> > --- > samples/vfio-mdev/mbochs.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/samples/vfio-mdev/mbochs.c b/samples/vfio-mdev/mbochs.c > index aa25cda21d..85ac603769 100644 > --- a/samples/vfio-mdev/mbochs.c > +++ b/samples/vfio-mdev/mbochs.c > @@ -811,11 +811,18 @@ static void *mbochs_kmap_dmabuf(struct dma_buf *buf, unsigned long page_num) > return kmap(page); > } > > +static void mbochs_kunmap_dmabuf(struct dma_buf *buf, unsigned long page_num, > + void *vaddr) > +{ > + kunmap(vaddr); > +} > + > static struct dma_buf_ops mbochs_dmabuf_ops = { > .map_dma_buf = mbochs_map_dmabuf, > .unmap_dma_buf = mbochs_unmap_dmabuf, > .release = mbochs_release_dmabuf, > .map = mbochs_kmap_dmabuf, > + .unmap = mbochs_kunmap_dmabuf, > .mmap = mbochs_mmap_dmabuf, > }; > Is this a fix for v4.18? AFAICT, the kmap_atomic removal is only in next, not yet upstream and hopefully includes this sample driver before hitting mainline. Should we therefore have a v4.18-rc based patch for this to include before v4.18 to resolve this? Thanks, Alex