Re: [PATCH] Au1x00 8250 uart support (Updated - take #2).

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Wed, Sep 21, 2005 at 07:14:09PM +0300, Pantelis Antoniou wrote:
> > This is also buggy - the problem this has is that although you're
> > unregistering the platform device, references can still remain after
> > the module has been unloaded.  This means you have two options:
> > 
> > 1. never allow code with statically allocated platform devices to be
> >    modules.
> > 2. use something like platform_device_register_simple() (which doesn't
> >    currently exist in a useful form for devices with platform data.)
> > 
> 
> Yes, this is stinking mess. I resorted to just not allowing a module build.
> 
> BTW, the current platform_device_register_simple suffers from the same bug.
> 
> ! struct platform_device *platform_device_register_simple(char *name, unsigned int id,
> !                                                         struct resource *res, unsigned int num)
> !
> 
> later on in the function body we have...
> 
> !         pobj->pdev.name = name;
> 
> It's clearly buggy, since the typical calling sequence passes a constant
> string as a 'name' argument, which is located in the read only segment
> of the module (.text or .rodata).
> 
> To fix this space for the name must be allocated just like the resources...

Is this true?  platform_device_register uses pdev.name to construct the
bus id.  pdev.name is also used when matching against drivers.

However, once a platform device is unregistered, it is no longer available
to be matched against drivers, so pdev.name is unused.  Therefore, I don't
think this is a problem.

> +#ifdef CONFIG_SERIAL_8250_AU1X00_MODULE
> +#error This file can not yet be compiled as a module.
> +#endif

This isn't how we prevent modular builds...

> diff --git a/drivers/serial/Kconfig b/drivers/serial/Kconfig
> --- a/drivers/serial/Kconfig
> +++ b/drivers/serial/Kconfig
> @@ -207,6 +207,14 @@ config SERIAL_8250_ACORN
>  	  system, say Y to this option.  The driver can handle 1, 2, or 3 port
>  	  cards.  If unsure, say N.
>  
> +config SERIAL_8250_AU1X00
> +	tristate "AU1X00 serial port support"

We prevent them by setting this to 'bool'.

> +	depends on SOC_AU1X00 && SERIAL_8250

It doens't really depend on SERIAL_8250 - it can live independently of it.
However, it can be built in when SERIAL_8250 isn't, so this dependency
should be SERIAL_8250 != n.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core