[PATCH] drm/bridge: adv7511: fix crash on irq during probe

[PATCH] drm/bridge: adv7511: fix crash on irq during probe

[Alvin Šipraga]

From: Mads Bligaard Nielsen <bli@bang-olufsen.dk>

Moved IRQ registration down to end of adv7511_probe().

If an IRQ already is pending during adv7511_probe
(before adv7511_cec_init) then cec_received_msg_ts
could crash using uninitialized data:

    Unable to handle kernel read from unreadable memory at virtual address 00000000000003d5
    Internal error: Oops: 96000004 [#1] PREEMPT_RT SMP
    Call trace:
     cec_received_msg_ts+0x48/0x990 [cec]
     adv7511_cec_irq_process+0x1cc/0x308 [adv7511]
     adv7511_irq_process+0xd8/0x120 [adv7511]
     adv7511_irq_handler+0x1c/0x30 [adv7511]
     irq_thread_fn+0x30/0xa0
     irq_thread+0x14c/0x238
     kthread+0x190/0x1a8

Fixes: 3b1b975003e4 ("drm: adv7511/33: add HDMI CEC support")
Signed-off-by: Mads Bligaard Nielsen <bli@bang-olufsen.dk>
Signed-off-by: Alvin Šipraga <alsi@bang-olufsen.dk>
---
 drivers/gpu/drm/bridge/adv7511/adv7511_drv.c | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c b/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c
index d518de88b5c3..71022cb8abe4 100644
--- a/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c
+++ b/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c
@@ -1291,17 +1291,6 @@ static int adv7511_probe(struct i2c_client *i2c)
 
 	INIT_WORK(&adv7511->hpd_work, adv7511_hpd_work);
 
-	if (i2c->irq) {
-		init_waitqueue_head(&adv7511->wq);
-
-		ret = devm_request_threaded_irq(dev, i2c->irq, NULL,
-						adv7511_irq_handler,
-						IRQF_ONESHOT, dev_name(dev),
-						adv7511);
-		if (ret)
-			goto err_unregister_cec;
-	}
-
 	adv7511_power_off(adv7511);
 
 	i2c_set_clientdata(i2c, adv7511);
@@ -1325,6 +1314,17 @@ static int adv7511_probe(struct i2c_client *i2c)
 
 	adv7511_audio_init(dev, adv7511);
 
+	if (i2c->irq) {
+		init_waitqueue_head(&adv7511->wq);
+
+		ret = devm_request_threaded_irq(dev, i2c->irq, NULL,
+						adv7511_irq_handler,
+						IRQF_ONESHOT, dev_name(dev),
+						adv7511);
+		if (ret)
+			goto err_unregister_audio;
+	}
+
 	if (adv7511->type == ADV7533 || adv7511->type == ADV7535) {
 		ret = adv7533_attach_dsi(adv7511);
 		if (ret)

---
base-commit: a48e2cc92835fa1d9b373b804b2173c779387b8e
change-id: 20231014-adv7511-cec-irq-crash-fix-6fdd9093dc7a

Re: [PATCH] drm/bridge: adv7511: fix crash on irq during probe

[Laurent Pinchart]

Hello Alvin,

On Sat, Oct 14, 2023 at 08:46:12PM +0200, Alvin Šipraga wrote:
> From: Mads Bligaard Nielsen <bli@bang-olufsen.dk>
> 
> Moved IRQ registration down to end of adv7511_probe().
> 
> If an IRQ already is pending during adv7511_probe
> (before adv7511_cec_init) then cec_received_msg_ts
> could crash using uninitialized data:
> 
>     Unable to handle kernel read from unreadable memory at virtual address 00000000000003d5
>     Internal error: Oops: 96000004 [#1] PREEMPT_RT SMP
>     Call trace:
>      cec_received_msg_ts+0x48/0x990 [cec]
>      adv7511_cec_irq_process+0x1cc/0x308 [adv7511]
>      adv7511_irq_process+0xd8/0x120 [adv7511]
>      adv7511_irq_handler+0x1c/0x30 [adv7511]
>      irq_thread_fn+0x30/0xa0
>      irq_thread+0x14c/0x238
>      kthread+0x190/0x1a8
> 
> Fixes: 3b1b975003e4 ("drm: adv7511/33: add HDMI CEC support")

Isn't the issue older than that ?

> Signed-off-by: Mads Bligaard Nielsen <bli@bang-olufsen.dk>
> Signed-off-by: Alvin Šipraga <alsi@bang-olufsen.dk>

With the Fixes: tag updated,

Reviewed-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

> ---
>  drivers/gpu/drm/bridge/adv7511/adv7511_drv.c | 22 +++++++++++-----------
>  1 file changed, 11 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c b/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c
> index d518de88b5c3..71022cb8abe4 100644
> --- a/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c
> +++ b/drivers/gpu/drm/bridge/adv7511/adv7511_drv.c
> @@ -1291,17 +1291,6 @@ static int adv7511_probe(struct i2c_client *i2c)
>  
>  	INIT_WORK(&adv7511->hpd_work, adv7511_hpd_work);
>  
> -	if (i2c->irq) {
> -		init_waitqueue_head(&adv7511->wq);
> -
> -		ret = devm_request_threaded_irq(dev, i2c->irq, NULL,
> -						adv7511_irq_handler,
> -						IRQF_ONESHOT, dev_name(dev),
> -						adv7511);
> -		if (ret)
> -			goto err_unregister_cec;
> -	}
> -
>  	adv7511_power_off(adv7511);
>  
>  	i2c_set_clientdata(i2c, adv7511);
> @@ -1325,6 +1314,17 @@ static int adv7511_probe(struct i2c_client *i2c)
>  
>  	adv7511_audio_init(dev, adv7511);
>  
> +	if (i2c->irq) {
> +		init_waitqueue_head(&adv7511->wq);
> +
> +		ret = devm_request_threaded_irq(dev, i2c->irq, NULL,
> +						adv7511_irq_handler,
> +						IRQF_ONESHOT, dev_name(dev),
> +						adv7511);
> +		if (ret)
> +			goto err_unregister_audio;
> +	}
> +
>  	if (adv7511->type == ADV7533 || adv7511->type == ADV7535) {
>  		ret = adv7533_attach_dsi(adv7511);
>  		if (ret)
> 

-- 
Regards,

Laurent Pinchart

Re: [PATCH] drm/bridge: adv7511: fix crash on irq during probe

[Alvin Šipraga]

Hi Laurent,

Thanks for the quick review!

On Mon, Oct 16, 2023 at 11:14:44AM +0300, Laurent Pinchart wrote:
> Hello Alvin,
> 
> On Sat, Oct 14, 2023 at 08:46:12PM +0200, Alvin Šipraga wrote:
> > From: Mads Bligaard Nielsen <bli@bang-olufsen.dk>
> > 
> > Moved IRQ registration down to end of adv7511_probe().
> > 
> > If an IRQ already is pending during adv7511_probe
> > (before adv7511_cec_init) then cec_received_msg_ts
> > could crash using uninitialized data:
> > 
> >     Unable to handle kernel read from unreadable memory at virtual address 00000000000003d5
> >     Internal error: Oops: 96000004 [#1] PREEMPT_RT SMP
> >     Call trace:
> >      cec_received_msg_ts+0x48/0x990 [cec]
> >      adv7511_cec_irq_process+0x1cc/0x308 [adv7511]
> >      adv7511_irq_process+0xd8/0x120 [adv7511]
> >      adv7511_irq_handler+0x1c/0x30 [adv7511]
> >      irq_thread_fn+0x30/0xa0
> >      irq_thread+0x14c/0x238
> >      kthread+0x190/0x1a8
> > 
> > Fixes: 3b1b975003e4 ("drm: adv7511/33: add HDMI CEC support")
> 
> Isn't the issue older than that ?

I don't think so. The stacktrace shows the crash is in CEC handling code, which
was added in the blamed commit. A static analysis of adv7511_irq_process()
suggests that the only other place where data could be uninitialized is if the
hpd_work is scheduled:

	if (process_hpd && irq0 & ADV7511_INT0_HPD && adv7511->bridge.encoder)
		schedule_work(&adv7511->hpd_work);

... but this has a check on bridge.encoder, which seems to have been introduced
in a similar fix here:

| commit a1d0503d26ea2ef04f3f013d379e8f4d29c27127
| Author: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
| Date:   Thu May 14 00:31:07 2015 +0300
| 
|     drm: adv7511: Fix crash in IRQ handler when no encoder is associated
|     
|     The ADV7511 is probed before its slave encoder init function associates
|     it with an encoder. This creates a time window during which hot plug
|     detection interrupts can occur with an encoder, resulting in a crash in
|     the IRQ handler.
|     
|     Fix this by ignoring hot plug detection IRQs when no encoder is
|     associated yet.
|     
|     Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
|     Acked-by: Lars-Peter Clausen <lars@metafoo.de>
| 
| diff --git a/drivers/gpu/drm/i2c/adv7511.c b/drivers/gpu/drm/i2c/adv7511.c
| index b728523e194f..2aaa3c88999e 100644
| --- a/drivers/gpu/drm/i2c/adv7511.c
| +++ b/drivers/gpu/drm/i2c/adv7511.c
| @@ -438,7 +438,7 @@ static int adv7511_irq_process(struct adv7511 *adv7511)
|         regmap_write(adv7511->regmap, ADV7511_REG_INT(0), irq0);
|         regmap_write(adv7511->regmap, ADV7511_REG_INT(1), irq1);
|  
| -       if (irq0 & ADV7511_INT0_HDP)
| +       if (irq0 & ADV7511_INT0_HDP && adv7511->encoder)
|                 drm_helper_hpd_irq_event(adv7511->encoder->dev);
|  
|         if (irq0 & ADV7511_INT0_EDID_READY || irq1 & ADV7511_INT1_DDC_ERROR) {

So assuming that is the case, I am not sure which Fixes: tag I ought to
otherwise use. What do you think?

> 
> > Signed-off-by: Mads Bligaard Nielsen <bli@bang-olufsen.dk>
> > Signed-off-by: Alvin Šipraga <alsi@bang-olufsen.dk>
> 
> With the Fixes: tag updated,
> 
> Reviewed-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>

Kind regards,
Alvin

Re: [PATCH] drm/bridge: adv7511: fix crash on irq during probe

[Alvin Šipraga]

Hi Laurent,

This is a friendly ping to get your feedback on my reply below. I don't think
the Fixes tag is incorrect here. Please could you take another look and let me
know if I can resend with your Reviewed-by?

Kind regards,
Alvin

On Mon, Oct 16, 2023 at 10:42:27AM +0200, Alvin Šipraga wrote:
> Hi Laurent,
> 
> Thanks for the quick review!
> 
> On Mon, Oct 16, 2023 at 11:14:44AM +0300, Laurent Pinchart wrote:
> > Hello Alvin,
> > 
> > On Sat, Oct 14, 2023 at 08:46:12PM +0200, Alvin Šipraga wrote:
> > > From: Mads Bligaard Nielsen <bli@bang-olufsen.dk>
> > > 
> > > Moved IRQ registration down to end of adv7511_probe().
> > > 
> > > If an IRQ already is pending during adv7511_probe
> > > (before adv7511_cec_init) then cec_received_msg_ts
> > > could crash using uninitialized data:
> > > 
> > >     Unable to handle kernel read from unreadable memory at virtual address 00000000000003d5
> > >     Internal error: Oops: 96000004 [#1] PREEMPT_RT SMP
> > >     Call trace:
> > >      cec_received_msg_ts+0x48/0x990 [cec]
> > >      adv7511_cec_irq_process+0x1cc/0x308 [adv7511]
> > >      adv7511_irq_process+0xd8/0x120 [adv7511]
> > >      adv7511_irq_handler+0x1c/0x30 [adv7511]
> > >      irq_thread_fn+0x30/0xa0
> > >      irq_thread+0x14c/0x238
> > >      kthread+0x190/0x1a8
> > > 
> > > Fixes: 3b1b975003e4 ("drm: adv7511/33: add HDMI CEC support")
> > 
> > Isn't the issue older than that ?
> 
> I don't think so. The stacktrace shows the crash is in CEC handling code, which
> was added in the blamed commit. A static analysis of adv7511_irq_process()
> suggests that the only other place where data could be uninitialized is if the
> hpd_work is scheduled:
> 
> 	if (process_hpd && irq0 & ADV7511_INT0_HPD && adv7511->bridge.encoder)
> 		schedule_work(&adv7511->hpd_work);
> 
> ... but this has a check on bridge.encoder, which seems to have been introduced
> in a similar fix here:
> 
> | commit a1d0503d26ea2ef04f3f013d379e8f4d29c27127
> | Author: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
> | Date:   Thu May 14 00:31:07 2015 +0300
> | 
> |     drm: adv7511: Fix crash in IRQ handler when no encoder is associated
> |     
> |     The ADV7511 is probed before its slave encoder init function associates
> |     it with an encoder. This creates a time window during which hot plug
> |     detection interrupts can occur with an encoder, resulting in a crash in
> |     the IRQ handler.
> |     
> |     Fix this by ignoring hot plug detection IRQs when no encoder is
> |     associated yet.
> |     
> |     Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
> |     Acked-by: Lars-Peter Clausen <lars@metafoo.de>
> | 
> | diff --git a/drivers/gpu/drm/i2c/adv7511.c b/drivers/gpu/drm/i2c/adv7511.c
> | index b728523e194f..2aaa3c88999e 100644
> | --- a/drivers/gpu/drm/i2c/adv7511.c
> | +++ b/drivers/gpu/drm/i2c/adv7511.c
> | @@ -438,7 +438,7 @@ static int adv7511_irq_process(struct adv7511 *adv7511)
> |         regmap_write(adv7511->regmap, ADV7511_REG_INT(0), irq0);
> |         regmap_write(adv7511->regmap, ADV7511_REG_INT(1), irq1);
> |  
> | -       if (irq0 & ADV7511_INT0_HDP)
> | +       if (irq0 & ADV7511_INT0_HDP && adv7511->encoder)
> |                 drm_helper_hpd_irq_event(adv7511->encoder->dev);
> |  
> |         if (irq0 & ADV7511_INT0_EDID_READY || irq1 & ADV7511_INT1_DDC_ERROR) {
> 
> So assuming that is the case, I am not sure which Fixes: tag I ought to
> otherwise use. What do you think?
> 
> > 
> > > Signed-off-by: Mads Bligaard Nielsen <bli@bang-olufsen.dk>
> > > Signed-off-by: Alvin Šipraga <alsi@bang-olufsen.dk>
> > 
> > With the Fixes: tag updated,
> > 
> > Reviewed-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
> 
> Kind regards,
> Alvin