[PATCH] media: dvb_frontend: dvb_unregister_frontend must not call dvb_detach for fe->ops.release

[PATCH] media: dvb_frontend: dvb_unregister_frontend must not call dvb_detach for fe->ops.release

[Wolfgang Rohdewald]

because ops.release was never dvb_attached.
Which makes sense because f->ops.release does not attach anything.

Now, rmmod dvb_usb_pctv452e correctly sets counters for
stb6100/stb0899 to 0.

Before, stb0899 got a counter -1, and for my 4 receivers I got 3 OOPses
like

Nov 24 14:40:41 s5 kernel: [  194.211014] WARNING: CPU: 6 PID: 3055 at
   module_put.part.45+0x132/0x1a0
Call Trace:
 ? _stb0899_read_reg+0x100/0x100 [stb0899]
 ? _stb0899_read_reg+0x100/0x100 [stb0899]
 symbol_put_addr+0x38/0x60
 dvb_frontend_put+0x42/0x60 [dvb_core]
 ? stb0899_sleep+0x50/0x50 [stb0899]
 dvb_frontend_detach+0x7c/0x90 [dvb_core]
 dvb_usb_adapter_frontend_exit+0x57/0x80 [dvb_usb]
 dvb_usb_exit+0x39/0xb0 [dvb_usb]
 dvb_usb_device_exit+0x3f/0x60 [dvb_usb]
 pctv452e_usb_disconnect+0x6f/0x80 [dvb_usb_pctv452e]
 usb_unbind_interface+0x75/0x290
 ? _raw_spin_unlock_irqrestore+0x4a/0x80
 device_release_driver_internal+0x160/0x210
 driver_detach+0x40/0x80
 bus_remove_driver+0x5c/0xd0
 driver_unregister+0x2c/0x40
 usb_deregister+0x6c/0xf0
 pctv452e_usb_driver_exit+0x10/0xec0 [dvb_usb_pctv452e]

Signed-off-by: Wolfgang Rohdewald <wolfgang@rohdewald.de>
---
 drivers/media/dvb-core/dvb_frontend.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/media/dvb-core/dvb_frontend.c b/drivers/media/dvb-core/dvb_frontend.c
index 9139d01ba7ed..c2cc794299c9 100644
--- a/drivers/media/dvb-core/dvb_frontend.c
+++ b/drivers/media/dvb-core/dvb_frontend.c
@@ -150,7 +150,8 @@ static void __dvb_frontend_free(struct dvb_frontend *fe)
 
 	dvb_free_device(fepriv->dvbdev);
 
-	dvb_frontend_invoke_release(fe, fe->ops.release);
+	if (fe->ops.release)
+		fe->ops.release(fe);
 
 	kfree(fepriv);
 	fe->frontend_priv = NULL;
-- 
2.11.0

Re: [PATCH] media: dvb_frontend: dvb_unregister_frontend must not call dvb_detach for fe->ops.release

[Daniel Scheller]

Am Fri, 24 Nov 2017 15:02:30 +0100
schrieb Wolfgang Rohdewald <wolfgang@rohdewald.de>:

> because ops.release was never dvb_attached.
> Which makes sense because f->ops.release does not attach anything.
> 
> Now, rmmod dvb_usb_pctv452e correctly sets counters for
> stb6100/stb0899 to 0.
> 
> Before, stb0899 got a counter -1, and for my 4 receivers I got 3
> OOPses like
> 
> Nov 24 14:40:41 s5 kernel: [  194.211014] WARNING: CPU: 6 PID: 3055 at
>    module_put.part.45+0x132/0x1a0
> Call Trace:
>  ? _stb0899_read_reg+0x100/0x100 [stb0899]
>  ? _stb0899_read_reg+0x100/0x100 [stb0899]
>  symbol_put_addr+0x38/0x60
>  dvb_frontend_put+0x42/0x60 [dvb_core]
>  ? stb0899_sleep+0x50/0x50 [stb0899]
>  dvb_frontend_detach+0x7c/0x90 [dvb_core]
>  dvb_usb_adapter_frontend_exit+0x57/0x80 [dvb_usb]
>  dvb_usb_exit+0x39/0xb0 [dvb_usb]
>  dvb_usb_device_exit+0x3f/0x60 [dvb_usb]
>  pctv452e_usb_disconnect+0x6f/0x80 [dvb_usb_pctv452e]
>  usb_unbind_interface+0x75/0x290
>  ? _raw_spin_unlock_irqrestore+0x4a/0x80
>  device_release_driver_internal+0x160/0x210
>  driver_detach+0x40/0x80
>  bus_remove_driver+0x5c/0xd0
>  driver_unregister+0x2c/0x40
>  usb_deregister+0x6c/0xf0
>  pctv452e_usb_driver_exit+0x10/0xec0 [dvb_usb_pctv452e]
> 
> Signed-off-by: Wolfgang Rohdewald <wolfgang@rohdewald.de>
> ---
>  drivers/media/dvb-core/dvb_frontend.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/media/dvb-core/dvb_frontend.c
> b/drivers/media/dvb-core/dvb_frontend.c index
> 9139d01ba7ed..c2cc794299c9 100644 ---
> a/drivers/media/dvb-core/dvb_frontend.c +++
> b/drivers/media/dvb-core/dvb_frontend.c @@ -150,7 +150,8 @@ static
> void __dvb_frontend_free(struct dvb_frontend *fe) 
>  	dvb_free_device(fepriv->dvbdev);
>  
> -	dvb_frontend_invoke_release(fe, fe->ops.release);
> +	if (fe->ops.release)
> +		fe->ops.release(fe);
>  
>  	kfree(fepriv);
>  	fe->frontend_priv = NULL;

FWIW, this patch breaks unloading for all DVB cards in my dev system,
in that attached drivers like demod and tuner drivers won't usecount
to zero anymore ("rmmod ddbridge" keeps the usecount of stv0367 and
cxd2841er up, and "rmmod budget_av" doesn't usecount the tda10023 back
to zero).

Best regards,
Daniel Scheller
-- 
https://github.com/herrnst