Re: [PATCH 2/3] drm/mediatek: mtk_disp_ovl: Enable/disable interrupt on bind/unbind

[AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>]

Il 28/10/25 10:55, CK Hu (胡俊光) ha scritto:
> On Wed, 2025-09-24 at 12:37 +0200, AngeloGioacchino Del Regno wrote:
>> External email : Please do not click links or open attachments until you have verified the sender or the content.
>>
>>
>> The OVL driver is installing an ISR in the probe function but, if
>> the component is not bound yet, the interrupt handler may call the
>> vblank_cb ahead of time (while probing other drivers) or too late
>> (while removing other drivers), possibly accessing memory that it
>> should not try to access by reusing stale pointers.
>>
>> In order to fix this, add a new `irq` member to struct mtk_disp_ovl
>> and then set the NOAUTOEN flag to the irq before installing the ISR
>> to manually call enable_irq() and disable_irq() in the bind and
>> unbind callbacks respectively.
>>
>> Fixes: 119f5173628a ("drm/mediatek: Add DRM Driver for Mediatek SoC MT8173.")
>> Link: https://lore.kernel.org/r/20250402083628.20111-5-angelogioacchino.delregno@collabora.com
>> Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
>> ---
>>   drivers/gpu/drm/mediatek/mtk_disp_ovl.c | 20 +++++++++++++-------
>>   1 file changed, 13 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/mediatek/mtk_disp_ovl.c b/drivers/gpu/drm/mediatek/mtk_disp_ovl.c
>> index e0236353d499..8e20b45411fc 100644
>> --- a/drivers/gpu/drm/mediatek/mtk_disp_ovl.c
>> +++ b/drivers/gpu/drm/mediatek/mtk_disp_ovl.c
>> @@ -161,6 +161,7 @@ struct mtk_disp_ovl {
>>          struct drm_crtc                 *crtc;
>>          struct clk                      *clk;
>>          void __iomem                    *regs;
>> +       int                             irq;
>>          struct cmdq_client_reg          cmdq_reg;
>>          const struct mtk_disp_ovl_data  *data;
>>          void                            (*vblank_cb)(void *data);
>> @@ -594,12 +595,18 @@ void mtk_ovl_bgclr_in_off(struct device *dev)
>>   static int mtk_disp_ovl_bind(struct device *dev, struct device *master,
>>                               void *data)
>>   {
>> +       struct mtk_disp_ovl *priv = dev_get_drvdata(dev);
>> +
>> +       enable_irq(priv->irq);
>>          return 0;
>>   }
>>
>>   static void mtk_disp_ovl_unbind(struct device *dev, struct device *master,
>>                                  void *data)
>>   {
>> +       struct mtk_disp_ovl *priv = dev_get_drvdata(dev);
>> +
>> +       disable_irq(priv->irq);
>>   }
>>
>>   static const struct component_ops mtk_disp_ovl_component_ops = {
>> @@ -611,16 +618,15 @@ static int mtk_disp_ovl_probe(struct platform_device *pdev)
>>   {
>>          struct device *dev = &pdev->dev;
>>          struct mtk_disp_ovl *priv;
>> -       int irq;
>>          int ret;
>>
>>          priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL);
>>          if (!priv)
>>                  return -ENOMEM;
>>
>> -       irq = platform_get_irq(pdev, 0);
>> -       if (irq < 0)
>> -               return irq;
>> +       priv->irq = platform_get_irq(pdev, 0);
>> +       if (priv->irq < 0)
>> +               return priv->irq;
>>
>>          priv->clk = devm_clk_get(dev, NULL);
>>          if (IS_ERR(priv->clk))
>> @@ -640,10 +646,10 @@ static int mtk_disp_ovl_probe(struct platform_device *pdev)
>>          priv->data = of_device_get_match_data(dev);
>>          platform_set_drvdata(pdev, priv);
>>
>> -       ret = devm_request_irq(dev, irq, mtk_disp_ovl_irq_handler,
>> -                              IRQF_TRIGGER_NONE, dev_name(dev), priv);
>> +       ret = devm_request_irq(dev, priv->irq, mtk_disp_ovl_irq_handler,
>> +                              IRQF_NO_AUTOEN, dev_name(dev), priv);
> 
> RDMA driver would clear pending interrupt and disable interrupt before request irq.
> I would like the hardware would be in idle state when probe.
> So OVL should do like RDMA.
> 

As I commented on RDMA - that won't work. After unbinding components, only the
bind() op will be executed, not probe(), so the interrupt will not be reset in
HW, and while re-binding other components, IRQs will come and produce a kernel
panic (in the best scenario, eh!) for use-after-free.

Regards,
Angelo

> Regards,
> CK
> 
>>          if (ret < 0)
>> -               return dev_err_probe(dev, ret, "Failed to request irq %d\n", irq);
>> +               return dev_err_probe(dev, ret, "Failed to request irq %d\n", priv->irq);
>>
>>          pm_runtime_enable(dev);
>>
>> --
>> 2.51.0
>>
>