| Message ID | 20240223212404.3709690-1-hsinyi@chromium.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | drm/mediatek: Fix a null pointer crash in mtk_drm_crtc_finish_page_flip | expand |
Hi, Hsin-yi: On Fri, 2024-02-23 at 13:23 -0800, Hsin-Yi Wang wrote: > > External email : Please do not click links or open attachments until > you have verified the sender or the content. > It's possible that mtk_crtc->event is NULL in > mtk_drm_crtc_finish_page_flip(). > > pending_needs_vblank value is set by mtk_crtc->event, but in > mtk_drm_crtc_atomic_flush(), it's is not guarded by the same > lock in mtk_drm_finish_page_flip(), thus a race condition happens. > > Consider the following case: > > CPU1 CPU2 > step 1: > mtk_drm_crtc_atomic_begin() > mtk_crtc->event is not null, > step 1: > mtk_drm_crtc_atomic_flush: > mtk_drm_crtc_update_config( > !!mtk_crtc->event) > step 2: > mtk_crtc_ddp_irq -> > mtk_drm_finish_page_flip: > lock > mtk_crtc->event set to null, > pending_needs_vblank set to false > unlock > pending_needs_vblank set to true, > > step 2: > mtk_crtc_ddp_irq -> > mtk_drm_finish_page_flip called > again, > pending_needs_vblank is still true > //null pointer > > Instead of guarding the entire mtk_drm_crtc_atomic_flush(), it's more > efficient to just check if mtk_crtc->event is null before use. > > Signed-off-by: Hsin-Yi Wang <hsinyi@chromium.org> > Fixes: 119f5173628a ("drm/mediatek: Add DRM Driver for Mediatek SoC > MT8173.") Reviewed-by: CK Hu <ck.hu@mediatek.com> > --- > drivers/gpu/drm/mediatek/mtk_drm_crtc.c | 12 +++++++----- > 1 file changed, 7 insertions(+), 5 deletions(-) > > diff --git a/drivers/gpu/drm/mediatek/mtk_drm_crtc.c > b/drivers/gpu/drm/mediatek/mtk_drm_crtc.c > index db43f9dff912..d645b85f9721 100644 > --- a/drivers/gpu/drm/mediatek/mtk_drm_crtc.c > +++ b/drivers/gpu/drm/mediatek/mtk_drm_crtc.c > @@ -95,11 +95,13 @@ static void mtk_drm_crtc_finish_page_flip(struct > mtk_drm_crtc *mtk_crtc) > struct drm_crtc *crtc = &mtk_crtc->base; > unsigned long flags; > > - spin_lock_irqsave(&crtc->dev->event_lock, flags); > - drm_crtc_send_vblank_event(crtc, mtk_crtc->event); > - drm_crtc_vblank_put(crtc); > - mtk_crtc->event = NULL; > - spin_unlock_irqrestore(&crtc->dev->event_lock, flags); > + if (mtk_crtc->event) { > + spin_lock_irqsave(&crtc->dev->event_lock, flags); > + drm_crtc_send_vblank_event(crtc, mtk_crtc->event); > + drm_crtc_vblank_put(crtc); > + mtk_crtc->event = NULL; > + spin_unlock_irqrestore(&crtc->dev->event_lock, flags); > + } > } > > static void mtk_drm_finish_page_flip(struct mtk_drm_crtc *mtk_crtc) > -- > 2.44.0.rc0.258.g7320e95886-goog
diff --git a/drivers/gpu/drm/mediatek/mtk_drm_crtc.c b/drivers/gpu/drm/mediatek/mtk_drm_crtc.c index db43f9dff912..d645b85f9721 100644 --- a/drivers/gpu/drm/mediatek/mtk_drm_crtc.c +++ b/drivers/gpu/drm/mediatek/mtk_drm_crtc.c @@ -95,11 +95,13 @@ static void mtk_drm_crtc_finish_page_flip(struct mtk_drm_crtc *mtk_crtc) struct drm_crtc *crtc = &mtk_crtc->base; unsigned long flags; - spin_lock_irqsave(&crtc->dev->event_lock, flags); - drm_crtc_send_vblank_event(crtc, mtk_crtc->event); - drm_crtc_vblank_put(crtc); - mtk_crtc->event = NULL; - spin_unlock_irqrestore(&crtc->dev->event_lock, flags); + if (mtk_crtc->event) { + spin_lock_irqsave(&crtc->dev->event_lock, flags); + drm_crtc_send_vblank_event(crtc, mtk_crtc->event); + drm_crtc_vblank_put(crtc); + mtk_crtc->event = NULL; + spin_unlock_irqrestore(&crtc->dev->event_lock, flags); + } } static void mtk_drm_finish_page_flip(struct mtk_drm_crtc *mtk_crtc)
It's possible that mtk_crtc->event is NULL in mtk_drm_crtc_finish_page_flip(). pending_needs_vblank value is set by mtk_crtc->event, but in mtk_drm_crtc_atomic_flush(), it's is not guarded by the same lock in mtk_drm_finish_page_flip(), thus a race condition happens. Consider the following case: CPU1 CPU2 step 1: mtk_drm_crtc_atomic_begin() mtk_crtc->event is not null, step 1: mtk_drm_crtc_atomic_flush: mtk_drm_crtc_update_config( !!mtk_crtc->event) step 2: mtk_crtc_ddp_irq -> mtk_drm_finish_page_flip: lock mtk_crtc->event set to null, pending_needs_vblank set to false unlock pending_needs_vblank set to true, step 2: mtk_crtc_ddp_irq -> mtk_drm_finish_page_flip called again, pending_needs_vblank is still true //null pointer Instead of guarding the entire mtk_drm_crtc_atomic_flush(), it's more efficient to just check if mtk_crtc->event is null before use. Signed-off-by: Hsin-Yi Wang <hsinyi@chromium.org> Fixes: 119f5173628a ("drm/mediatek: Add DRM Driver for Mediatek SoC MT8173.") --- drivers/gpu/drm/mediatek/mtk_drm_crtc.c | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-)