Re: [PATCH v2 2/2] panthor: use drm_gpuva_unlink_defer()

[Alice Ryhl <aliceryhl@google.com>]

On Thu, Sep 11, 2025 at 12:15:37PM +0200, Boris Brezillon wrote:
> On Tue, 09 Sep 2025 13:36:23 +0000
> Alice Ryhl <aliceryhl@google.com> wrote:
> 
> >  static void panthor_vma_init(struct panthor_vma *vma, u32 flags)
> > @@ -2084,12 +2010,12 @@ static int panthor_gpuva_sm_step_map(struct drm_gpuva_op *op, void *priv)
> >  	if (ret)
> >  		return ret;
> >  
> > -	/* Ref owned by the mapping now, clear the obj field so we don't release the
> > -	 * pinning/obj ref behind GPUVA's back.
> > -	 */
> >  	drm_gpuva_map(&vm->base, &vma->base, &op->map);
> >  	panthor_vma_link(vm, vma, op_ctx->map.vm_bo);
> > +
> > +	drm_gpuvm_bo_put_deferred(op_ctx->map.vm_bo);
> >  	op_ctx->map.vm_bo = NULL;
> 
> Hm, I don't see why we need a drm_gpuvm_bo_put_deferred() here. The
> original idea was to delegate the vm_bo ownership to the VA being added
> to the VM tree, so if we put it here, we have a UAF situation, don't we?

The vm_bo refcount goes like this:

incr vm_bo_obtain()
incr vma_link()
decr vm_bo_put()

There is no decrement in panthor_vm_cleanup_op_ctx() due to this line:

	op_ctx->map.vm_bo = NULL

So when everything is done, it is linked once and the refcount is
incremented by one, which is correct.

Alice