The pages backing the GEM objects are kept pinned in place as long as they are alive, so they must not be allocated from the MOVABLE zone. Blocking page migration for too long will cause the VM subsystem headaches and will outright break CMA, as a few pinned pages in CMA will lead to failure to find the required large contiguous regions. Signed-off-by: Lucas Stach <l.stach@xxxxxxxxxxxxxx> --- drivers/gpu/drm/msm/msm_gem.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 51a95da694d8..dc4860ae3171 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -1022,6 +1022,13 @@ static struct drm_gem_object *_msm_gem_new(struct drm_device *dev, ret = drm_gem_object_init(dev, obj, size); if (ret) goto fail; + /* + * Our buffers are kept pinned, so allocating them from the + * MOVABLE zone is a really bad idea, and conflicts with CMA. + * See comments above new_inode() why this is required _and_ + * expected if you're going to pin these pages. + */ + mapping_set_gfp_mask(obj->filp->f_mapping, GFP_HIGHUSER); } return obj; -- 2.20.1
