RFC: Harmonised parameter passing

RFC: Harmonised parameter passing

[Henk Vergonet]

Hi,

The current method of parameter passing to drivers build as a module is extremely usefull.
Modules don't have to write there own parsing code, there's a nice macro that can be used to document specifics of the parameter and so on.

Could we extend this method where we use the same methodology for inbound drivers? (Currently a lot of drivers use their own parameter parsing code when it comes to passing values at kernel boot time.)

so we could do the regular:

	insmod mcd io=0x340

for modules, or with kernel boot parameters:

	mcd.io=0x340

for in-kernel drivers.


My proposal would be to introduce something like:

DRIVER_PARM_DESC(variable, description);
DRIVER_PARM(variable, type, scope);

    where scope can be:
	PARM_SCOPE_MODULE	=> This parameter is used in module context.
	PARM_SCOPE_KERNEL	=> This parameter is used in kernel context.
	PARM_SCOPE_MODULE | PARM_SCOPE_KERNEL
				=> This parameter is used in both kernel and module context, which should be the default if scope is omitted.


What do you think?

Re: RFC: Harmonised parameter passing

[Dmitry Torokhov]

On Tue, 8 Mar 2005 15:59:23 +0100, Henk Vergonet
<rememberme@god.dyndns.org> wrote:
> 
> Hi,
> 
> The current method of parameter passing to drivers build as a module is extremely usefull.
> Modules don't have to write there own parsing code, there's a nice macro that can be used to document specifics of the parameter and so on.
> 
> Could we extend this method where we use the same methodology for inbound drivers? (Currently a lot of drivers use their own parameter parsing code when it comes to passing values at kernel boot time.)
> 
> so we could do the regular:
> 
>        insmod mcd io=0x340
> 
> for modules, or with kernel boot parameters:
> 
>        mcd.io=0x340
> 
> for in-kernel drivers.
> 

Umm.. This is already done. For parameters defined with module_param()
you use <paramname>=<value> for modules and
<modulename>.<paramname>=<value> for built-in case.

> My proposal would be to introduce something like:
> 
> DRIVER_PARM_DESC(variable, description);
> DRIVER_PARM(variable, type, scope);
> 
>    where scope can be:
>        PARM_SCOPE_MODULE       => This parameter is used in module context.
>        PARM_SCOPE_KERNEL       => This parameter is used in kernel context.
>        PARM_SCOPE_MODULE | PARM_SCOPE_KERNEL
>                                => This parameter is used in both kernel and module context, which should be the default if scope is omitted.
> 

Why would you want parameters that only work for modules? I'd consider
it a bug, not a feature, when parameter works only when code is
modularized.

-- 
Dmitry

Re: RFC: Harmonised parameter passing

[Henk Vergonet]

On Tue, Mar 08, 2005 at 10:14:32AM -0500, Dmitry Torokhov wrote:
> On Tue, 8 Mar 2005 15:59:23 +0100, Henk Vergonet
> > Could we extend this method where we use the same methodology for inbound drivers? (Currently a lot of drivers use their own parameter parsing code when it comes to passing values at kernel boot time.)
> > 
> > so we could do the regular:
> > 
> >        insmod mcd io=0x340
> > 
> > for modules, or with kernel boot parameters:
> > 
> >        mcd.io=0x340
> > 
> > for in-kernel drivers.
> > 
> 
> Umm.. This is already done. For parameters defined with module_param()
> you use <paramname>=<value> for modules and
> <modulename>.<paramname>=<value> for built-in case.

I did not know, but thats good news! My research was only based on the LKM module howto.

> > My proposal would be to introduce something like:
> > 
> > DRIVER_PARM_DESC(variable, description);
> > DRIVER_PARM(variable, type, scope);
> > 
> >    where scope can be:
> >        PARM_SCOPE_MODULE       => This parameter is used in module context.
> >        PARM_SCOPE_KERNEL       => This parameter is used in kernel context.
> >        PARM_SCOPE_MODULE | PARM_SCOPE_KERNEL
> >                                => This parameter is used in both kernel and module context, which should be the default if scope is omitted.
> > 
> 
> Why would you want parameters that only work for modules? I'd consider
> it a bug, not a feature, when parameter works only when code is
> modularized.
> 
I totally agree! (But I was preparing for the discussion were people
argue that they need a different set of parameters at boot time.)

One question remains though, how do you handle the initialization of
multiple instances of an inbound driver?

mcd0.io=0x340 mcd1.io=0x350 

Thnx,
Henk

Re: RFC: Harmonised parameter passing

[Dmitry Torokhov]

On Tue, 8 Mar 2005 16:47:47 +0100, Henk Vergonet
<rememberme@god.dyndns.org> wrote:
> One question remains though, how do you handle the initialization of
> multiple instances of an inbound driver?
> 
> mcd0.io=0x340 mcd1.io=0x350
> 

I think the most common practice is to specify a list of addresses:
mcd.io=0x340,0x350

module_param_array() helps here.

-- 
Dmitry