On Wednesday, March 22nd, 2023 at 20:45, Björn Roy Baron <bjorn3_gh@xxxxxxxxxxxxxx> wrote:
> On Friday, March 17th, 2023 at 13:12, Maíra Canal <mcanal@xxxxxxxxxx> wrote:
>
> > A GEMObject, related to struct drm_gem_object, holds a pointer
> > to reservation object associated with the this GEM object. Therefore,
> > expose this reservation object through the DmaResv safe abstraction.
> >
> > Signed-off-by: Maíra Canal <mcanal@xxxxxxxxxx>
> > ---
> > rust/kernel/drm/gem/mod.rs | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/rust/kernel/drm/gem/mod.rs b/rust/kernel/drm/gem/mod.rs
> > index c4a42bb2f718..dae95e5748a7 100644
> > --- a/rust/kernel/drm/gem/mod.rs
> > +++ b/rust/kernel/drm/gem/mod.rs
> > @@ -11,6 +11,7 @@ use alloc::boxed::Box;
> >
> > use crate::{
> > bindings,
> > + dma_resv::DmaResv,
> > drm::{device, drv, file},
> > error::{to_result, Result},
> > prelude::*,
> > @@ -136,6 +137,12 @@ pub trait BaseObject: IntoGEMObject {
> > self.gem_obj().size
> > }
> >
> > + /// Returns the pointer to reservation object associated with this GEM object.
> > + fn resv(&self) -> DmaResv {
> > + // SAFETY: Every GEM object holds a reference to a reservation object
> > + unsafe { DmaResv::from_raw(self.gem_obj().resv) }
> > + }
>
> There is nothing ensuring that DmaResv doesn't outlive self and thus may have been deallocated. You could change DmaResv to be a newtype around UnsafeCell<dma_resv> and then return an &DmaResv here. This way &DmaResv can't outlive &self.
Actually Opaque<dma_resv> may be better as no fields of dma_resv are used on the rust side.
>
> > +
> > /// Sets the exportable flag, which controls whether the object can be exported via PRIME.
> > fn set_exportable(&mut self, exportable: bool) {
> > self.mut_gem_obj().exportable = exportable;
> > --
> > 2.39.2
>
> Cheers,
> Bjorn
Cheers,
Bjorn
