[PATCH v2] drm: Set vm_ops to GEM object's values during mmap

[Daniel Vetter]

On Fri, Jan 15, 2021 at 09:57:24AM +0000, Kieran Bingham wrote:
> Hi Thomas,
> 
> On 15/01/2021 09:30, Thomas Zimmermann wrote:
> > The GEM mmap code relies on the GEM object's mmap callback to set the
> > VMA's vm_ops field. This is easily forgotten and already led to a memory
> > leak in the CMA helpers. Instead set the vm_ops field in the DRM core
> > code to the GEM object's value. Drivers with different needs can override
> > this in their mmap callback.
> > 
> > v2:
> > 	* support (vm_ops == NULL) if mmap is given; required by VRAM
> > 	  helpers

I guess vram helpers need this because ttm has it's own vm_ops struct?
Might be another thing worth unifying (eventually).

> > 
> > Signed-off-by: Thomas Zimmermann <tzimmermann at suse.de>
> > Fixes: f5ca8eb6f9bd ("drm/cma-helper: Implement mmap as GEM CMA object functions")
> > Reported-by: Kieran Bingham <kieran.bingham+renesas at ideasonboard.com>
> > Tested-by: Kieran Bingham <kieran.bingham+renesas at ideasonboard.com>
> 
> Re-tested just fine this side ;-)
>  - https://paste.ubuntu.com/p/Jgz6xMKNJX/

Reviewed-by: Daniel Vetter <daniel.vetter at ffwll.ch>

> 
> Thanks
> 
> Kieran
> 
> > Cc: Maxime Ripard <mripard at kernel.org>
> > Cc: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
> > Cc: David Airlie <airlied at linux.ie>
> > Cc: Daniel Vetter <daniel at ffwll.ch>
> > Cc: Eric Anholt <eric at anholt.net>
> > Cc: dri-devel at lists.freedesktop.org
> > ---
> >  drivers/gpu/drm/drm_gem.c   | 19 ++++++++++---------
> >  drivers/gpu/drm/drm_prime.c |  2 ++
> >  2 files changed, 12 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c
> > index 34b2f111c01c..c2ce78c4edc3 100644
> > --- a/drivers/gpu/drm/drm_gem.c
> > +++ b/drivers/gpu/drm/drm_gem.c
> > @@ -1068,20 +1068,17 @@ int drm_gem_mmap_obj(struct drm_gem_object *obj, unsigned long obj_size,
> >  	drm_gem_object_get(obj);
> >  
> >  	vma->vm_private_data = obj;
> > +	vma->vm_ops = obj->funcs->vm_ops;
> >  
> >  	if (obj->funcs->mmap) {
> >  		ret = obj->funcs->mmap(obj, vma);
> > -		if (ret) {
> > -			drm_gem_object_put(obj);
> > -			return ret;
> > -		}
> > +		if (ret)
> > +			goto err_drm_gem_object_put;
> >  		WARN_ON(!(vma->vm_flags & VM_DONTEXPAND));
> >  	} else {
> > -		if (obj->funcs->vm_ops)
> > -			vma->vm_ops = obj->funcs->vm_ops;
> > -		else {
> > -			drm_gem_object_put(obj);
> > -			return -EINVAL;
> > +		if (!vma->vm_ops) {
> > +			ret = -EINVAL;
> > +			goto err_drm_gem_object_put;
> >  		}
> >  
> >  		vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
> > @@ -1090,6 +1087,10 @@ int drm_gem_mmap_obj(struct drm_gem_object *obj, unsigned long obj_size,
> >  	}
> >  
> >  	return 0;
> > +
> > +err_drm_gem_object_put:
> > +	drm_gem_object_put(obj);
> > +	return ret;
> >  }
> >  EXPORT_SYMBOL(drm_gem_mmap_obj);
> >  
> > diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
> > index 683aa29ecd3b..2a54f86856af 100644
> > --- a/drivers/gpu/drm/drm_prime.c
> > +++ b/drivers/gpu/drm/drm_prime.c
> > @@ -717,6 +717,8 @@ int drm_gem_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
> >  	vma->vm_pgoff += drm_vma_node_start(&obj->vma_node);
> >  
> >  	if (obj->funcs && obj->funcs->mmap) {
> > +		vma->vm_ops = obj->funcs->vm_ops;
> > +

Do you know how much we still need the non-obj->funcs path here? Maybe
time to detele it and wrape the obj->funcs check in a WARN_ON?
-Daniel


> >  		ret = obj->funcs->mmap(obj, vma);
> >  		if (ret)
> >  			return ret;
> > 
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch