Hi Liu, (CC'ing Sam)
Thank you for the patch. On Tue, Jun 16, 2020 at 05:04:52PM +0800, Liu Ying wrote: > It doesn't hurt to add the bridge in the global bridge list also for > platform specific dw-hdmi drivers which are based on the component > framework. This can be achieved by moving the drm_bridge_add() function > call from dw_hdmi_probe() to __dw_hdmi_probe(). Moreover, putting the > drm_bridge_add() function call prior to the interrupt registration and > enablement ensures that the mutex hpd_mutex embedded in the structure > drm_bridge can be initialized in drm_bridge_add() beforehand, which > avoids accessing the uninitialized mutex in case people want to call > function drm_bridge_hpd_notify() with the mutex locked internally to > handle hot plug detection event in the interrupt handler dw_hdmi_irq(). > > Cc: Andrzej Hajda <a.ha...@samsung.com> > Cc: Neil Armstrong <narmstr...@baylibre.com> > Cc: Laurent Pinchart <laurent.pinch...@ideasonboard.com> > Cc: Jonas Karlman <jo...@kwiboo.se> > Cc: Jernej Skrabec <jernej.skra...@siol.net> > Cc: David Airlie <airl...@linux.ie> > Cc: Daniel Vetter <dan...@ffwll.ch> > Cc: Boris Brezillon <boris.brezil...@collabora.com> > Cc: Jerome Brunet <jbru...@baylibre.com> > Cc: Cheng-Yi Chiang <cychi...@chromium.org> > Cc: Dariusz Marcinkiewicz <dar...@google.com> > Cc: Archit Taneja <arch...@codeaurora.org> > Cc: Jose Abreu <joab...@synopsys.com> > Cc: dri-devel@lists.freedesktop.org > Cc: NXP Linux Team <linux-...@nxp.com> > Signed-off-by: Liu Ying <victor....@nxp.com> > --- > Laurent, > > I may see the uninitialized mutex accessing issue with > i.MX dw-hdmi after applying your below patch set[1]. > I think patch '[22/27] drm: bridge: dw-hdmi: Make connector creation optional' > triggers the issue. > > [1] https://patchwork.kernel.org/cover/11569709/ > > drivers/gpu/drm/bridge/synopsys/dw-hdmi.c | 34 > ++++++++++++++----------------- > 1 file changed, 15 insertions(+), 19 deletions(-) > > diff --git a/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c > b/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c > index da84a91..4711700 100644 > --- a/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c > +++ b/drivers/gpu/drm/bridge/synopsys/dw-hdmi.c > @@ -3247,17 +3247,25 @@ __dw_hdmi_probe(struct platform_device *pdev, > > dw_hdmi_init_hw(hdmi); > > + hdmi->bridge.driver_private = hdmi; > + hdmi->bridge.funcs = &dw_hdmi_bridge_funcs; > +#ifdef CONFIG_OF > + hdmi->bridge.of_node = pdev->dev.of_node; > +#endif > + > + drm_bridge_add(&hdmi->bridge); This would introduce a race condition where a display driver could get hold of the bridge before it is fully initialized. I fear the right fix for this may be to add a drm_bridge_init() function to move mutex initialization away from drm_bridge_add(). That's a rather intrusive change I'm afraid :-( > + > irq = platform_get_irq(pdev, 0); > if (irq < 0) { > ret = irq; > - goto err_iahb; > + goto err_irq; > } > > ret = devm_request_threaded_irq(dev, irq, dw_hdmi_hardirq, > dw_hdmi_irq, IRQF_SHARED, > dev_name(dev), hdmi); > if (ret) > - goto err_iahb; > + goto err_irq; > > /* > * To prevent overflows in HDMI_IH_FC_STAT2, set the clk regenerator > @@ -3290,12 +3298,6 @@ __dw_hdmi_probe(struct platform_device *pdev, > hdmi->ddc = NULL; > } > > - hdmi->bridge.driver_private = hdmi; > - hdmi->bridge.funcs = &dw_hdmi_bridge_funcs; > -#ifdef CONFIG_OF > - hdmi->bridge.of_node = pdev->dev.of_node; > -#endif > - > if (hdmi->version >= 0x200a) > hdmi->connector.ycbcr_420_allowed = > hdmi->plat_data->ycbcr_420_allowed; > @@ -3357,6 +3359,8 @@ __dw_hdmi_probe(struct platform_device *pdev, > > return hdmi; > > +err_irq: > + drm_bridge_remove(&hdmi->bridge); > err_iahb: > clk_disable_unprepare(hdmi->iahb_clk); > if (hdmi->cec_clk) > @@ -3371,6 +3375,8 @@ __dw_hdmi_probe(struct platform_device *pdev, > > static void __dw_hdmi_remove(struct dw_hdmi *hdmi) > { > + drm_bridge_remove(&hdmi->bridge); > + > if (hdmi->audio && !IS_ERR(hdmi->audio)) > platform_device_unregister(hdmi->audio); > if (!IS_ERR(hdmi->cec)) > @@ -3396,22 +3402,12 @@ static void __dw_hdmi_remove(struct dw_hdmi *hdmi) > struct dw_hdmi *dw_hdmi_probe(struct platform_device *pdev, > const struct dw_hdmi_plat_data *plat_data) > { > - struct dw_hdmi *hdmi; > - > - hdmi = __dw_hdmi_probe(pdev, plat_data); > - if (IS_ERR(hdmi)) > - return hdmi; > - > - drm_bridge_add(&hdmi->bridge); > - > - return hdmi; > + return __dw_hdmi_probe(pdev, plat_data); > } > EXPORT_SYMBOL_GPL(dw_hdmi_probe); Do we need to keep __dw_hdmi_probe() and dw_hdmi_probe(), can't we rename __dw_hdmi_probe() to dw_hdmi_probe() ? Same for the remove functions. > > void dw_hdmi_remove(struct dw_hdmi *hdmi) > { > - drm_bridge_remove(&hdmi->bridge); > - > __dw_hdmi_remove(hdmi); > } > EXPORT_SYMBOL_GPL(dw_hdmi_remove); -- Regards, Laurent Pinchart _______________________________________________ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
