Re: [PATCH mm] cfag12864b: fix crash when built-in and no parport present

From: Miguel Ojeda
Date: Tue Feb 13 2007 - 01:29:41 EST


On 2/13/07, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
On Sat, 10 Feb 2007 20:51:23 +0100 Miguel Ojeda <maxextreme@xxxxxxxxx>
wrote:

> Andrew, Daniel who reported the crash has tested this patch and says that
it works fine avoiding it, so you can merge.
>
>
drivers-add-lcd-support-fix-crash-when-built-in-and-no-parport-present.patch
> Signed-off-by: Miguel Ojeda Sandonis <maxextreme@xxxxxxxxx>
> ---
> diff --git a/drivers/auxdisplay/cfag12864b.c
b/drivers/auxdisplay/cfag12864b.c
> index 889583d..cb44cb4 100644
> --- a/drivers/auxdisplay/cfag12864b.c
> +++ b/drivers/auxdisplay/cfag12864b.c
> @@ -312,6 +312,17 @@ EXPORT_SYMBOL_GPL(cfag12864b_disable);
> EXPORT_SYMBOL_GPL(cfag12864b_isenabled);
>
> /*
> + * Is the module inited?
> + */
> +
> +static unsigned char cfag12864b_inited;
> +unsigned char cfag12864b_isinited(void)
> +{
> + return cfag12864b_inited;
> +}
> +EXPORT_SYMBOL_GPL(cfag12864b_isinited);
> +
> +/*
> * Module Init & Exit
> */
>
> @@ -319,6 +330,13 @@ static int __init cfag12864b_init(void)
> {
> int ret = -EINVAL;
>
> + /* ks0108_init() must be called first */
> + if (!ks0108_isinited()) {
> + printk(KERN_ERR CFAG12864B_NAME ": ERROR: "
> + "ks0108 is not initialized\n");
> + goto none;
> + }
> +
> if (PAGE_SIZE < CFAG12864B_SIZE) {
> printk(KERN_ERR CFAG12864B_NAME ": ERROR: "
> "page size (%i) < cfag12864b size (%i)\n",
> @@ -354,6 +372,7 @@ static int __init cfag12864b_init(void)
> cfag12864b_clear();
> cfag12864b_on();
>
> + cfag12864b_inited = 1;
> return 0;
>
> cachealloced:
> diff --git a/drivers/auxdisplay/cfag12864bfb.c
b/drivers/auxdisplay/cfag12864bfb.c
> index 94765e7..66fafbb 100644
> --- a/drivers/auxdisplay/cfag12864bfb.c
> +++ b/drivers/auxdisplay/cfag12864bfb.c
> @@ -137,7 +137,14 @@ static struct platform_device *cfag12864
>
> static int __init cfag12864bfb_init(void)
> {
> - int ret;
> + int ret = -EINVAL;
> +
> + /* cfag12864b_init() must be called first */
> + if (!cfag12864b_isinited()) {
> + printk(KERN_ERR CFAG12864BFB_NAME ": ERROR: "
> + "cfag12864b is not initialized\n");
> + goto none;
> + }
>
> if (cfag12864b_enable()) {
> printk(KERN_ERR CFAG12864BFB_NAME ": ERROR: "
> @@ -162,6 +169,7 @@ static int __init cfag12864bfb_init(void
> }
> }
>
> +none:
> return ret;
> }
>
> diff --git a/drivers/auxdisplay/ks0108.c b/drivers/auxdisplay/ks0108.c
> index a637575..e6c3646 100644
> --- a/drivers/auxdisplay/ks0108.c
> +++ b/drivers/auxdisplay/ks0108.c
> @@ -111,6 +111,17 @@ EXPORT_SYMBOL_GPL(ks0108_address);
> EXPORT_SYMBOL_GPL(ks0108_page);
>
> /*
> + * Is the module inited?
> + */
> +
> +static unsigned char ks0108_inited;
> +unsigned char ks0108_isinited(void)
> +{
> + return ks0108_inited;
> +}
> +EXPORT_SYMBOL_GPL(ks0108_isinited);
> +
> +/*
> * Module Init & Exit
> */
>
> @@ -142,6 +153,7 @@ static int __init ks0108_init(void)
> goto registered;
> }
>
> + ks0108_inited = 1;
> return 0;
>
> registered:
> diff --git a/include/linux/cfag12864b.h b/include/linux/cfag12864b.h
> index 0bc45e6..1605dd8 100644
> --- a/include/linux/cfag12864b.h
> +++ b/include/linux/cfag12864b.h
> @@ -73,5 +73,10 @@ extern void cfag12864b_disable(void);
> */
> extern unsigned char cfag12864b_isenabled(void);
>
> +/*
> + * Is the module inited?
> + */
> +extern unsigned char cfag12864b_isinited(void);
> +
> #endif /* _CFAG12864B_H_ */
>
> diff --git a/include/linux/ks0108.h b/include/linux/ks0108.h
> index 8047d4b..a2c54ac 100644
> --- a/include/linux/ks0108.h
> +++ b/include/linux/ks0108.h
> @@ -43,4 +43,7 @@ extern void ks0108_address(unsigned char
> /* Set the controller's current page (0..7) */
> extern void ks0108_page(unsigned char page);
>
> +/* Is the module inited? */
> +extern unsigned char ks0108_isinited(void);
> +
> #endif /* _KS0108_H_ */

This looks hacky.

Your patch didn't include a description of the bug (please always include
such a description) so I'm not sure what's going on here. But I'm


Indeed, sorry. The patch is about "crash on CONFIG_CFAG12864B=y in
2.6.20-rc3-mm1" ( http://lkml.org/lkml/2007/1/7/118 ).

suspecting that module dependencies and appropriate choice of initcall
levels is the right way to fix whatever bug this is.


The problem comes when ks0108/cfag12864b are built-in and any parallel
port is present. ks0108_init() is called first, as it should be, but
fails to load (as there is no parallel port to use).

After that, cfag12864b_init() gets called, without knowing anything
about ks0108 failed, and calls ks0108_writecontrol(), which
dereferences an uninitialized pointer.

Init order is OK, I think. The problem is how to stop
cfag12864b_init() being called if ks0108 failed to load. modprobe does
it for us, but, how when built-in?

As anybody suggested another solution, I send that patch.

Regards,
Miguel

--
Miguel Ojeda
http://maxextreme.googlepages.com/index.htm
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/