Message ID | E1Xd5f4-0005K2-VO@rmk-PC.arm.linux.org.uk (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Sun, Oct 12, 2014 at 12:01:18AM +0100, Russell King wrote: > A refcounting leak was found of the original frame buffer attached to > the CRTC when using the page_flip ioctl, resulting in the frame buffer > never being freed. > > This was not obvious initially, as if the page flip subsequently > re-attaches the original frame buffer, the refcounts will be balanced. > However, if the original frame buffer is freed, then it will be leaked. > > Fix this by ensuring that we take a reference on the incoming fb, but > rely on the queued work to drop that ref count. > > Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Yeah, looks like what I've expected, so Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> > --- > drivers/gpu/drm/armada/armada_crtc.c | 13 +++++-------- > 1 file changed, 5 insertions(+), 8 deletions(-) > > diff --git a/drivers/gpu/drm/armada/armada_crtc.c b/drivers/gpu/drm/armada/armada_crtc.c > index 1f0875c26dc5..ac2d73f4af45 100644 > --- a/drivers/gpu/drm/armada/armada_crtc.c > +++ b/drivers/gpu/drm/armada/armada_crtc.c > @@ -945,18 +945,15 @@ static int armada_drm_crtc_page_flip(struct drm_crtc *crtc, > armada_reg_queue_end(work->regs, i); > > /* > - * Hold the old framebuffer for the work - DRM appears to drop our > - * reference to the old framebuffer in drm_mode_page_flip_ioctl(). > + * Ensure that we hold a reference on the new framebuffer. > + * This has to match the behaviour in mode_set. > */ > - drm_framebuffer_reference(work->old_fb); > + drm_framebuffer_reference(fb); > > ret = armada_drm_crtc_queue_frame_work(dcrtc, work); > if (ret) { > - /* > - * Undo our reference above; DRM does not drop the reference > - * to this object on error, so that's okay. > - */ > - drm_framebuffer_unreference(work->old_fb); > + /* Undo our reference above */ > + drm_framebuffer_unreference(fb); > kfree(work); > return ret; > } > -- > 1.8.3.1 >
diff --git a/drivers/gpu/drm/armada/armada_crtc.c b/drivers/gpu/drm/armada/armada_crtc.c index 1f0875c26dc5..ac2d73f4af45 100644 --- a/drivers/gpu/drm/armada/armada_crtc.c +++ b/drivers/gpu/drm/armada/armada_crtc.c @@ -945,18 +945,15 @@ static int armada_drm_crtc_page_flip(struct drm_crtc *crtc, armada_reg_queue_end(work->regs, i); /* - * Hold the old framebuffer for the work - DRM appears to drop our - * reference to the old framebuffer in drm_mode_page_flip_ioctl(). + * Ensure that we hold a reference on the new framebuffer. + * This has to match the behaviour in mode_set. */ - drm_framebuffer_reference(work->old_fb); + drm_framebuffer_reference(fb); ret = armada_drm_crtc_queue_frame_work(dcrtc, work); if (ret) { - /* - * Undo our reference above; DRM does not drop the reference - * to this object on error, so that's okay. - */ - drm_framebuffer_unreference(work->old_fb); + /* Undo our reference above */ + drm_framebuffer_unreference(fb); kfree(work); return ret; }
A refcounting leak was found of the original frame buffer attached to the CRTC when using the page_flip ioctl, resulting in the frame buffer never being freed. This was not obvious initially, as if the page flip subsequently re-attaches the original frame buffer, the refcounts will be balanced. However, if the original frame buffer is freed, then it will be leaked. Fix this by ensuring that we take a reference on the incoming fb, but rely on the queued work to drop that ref count. Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> --- drivers/gpu/drm/armada/armada_crtc.c | 13 +++++-------- 1 file changed, 5 insertions(+), 8 deletions(-)