Re: [RFC v1 14/25] genirq: Kill the first parameter 'irq' of irq_flow_handler_t

[Julia Lawall <julia.lawall@lip6.fr>]

On Wed, 20 May 2015, Thomas Gleixner wrote:

> B1;2802;0cOn Wed, 20 May 2015, Jiang Liu wrote:
> > On 2015/5/20 23:25, Thomas Gleixner wrote:
> > > On Wed, 20 May 2015, Jiang Liu wrote:
> > >> --- a/include/linux/irqdesc.h
> > >> +++ b/include/linux/irqdesc.h
> > >> @@ -99,6 +99,11 @@ static inline struct irq_desc *irq_data_to_desc(struct irq_data *data)
> > >>  	return container_of(data->common, struct irq_desc, irq_common_data);
> > >>  }
> > >>
> > >> +static inline unsigned int irq_desc_get_irq(struct irq_desc *desc)
> > >> +{
> > >> +	return desc->irq_data.irq;
> > >> +}
> > >> +
> > >
> > > Does not apply either.
> > Hi Thomas,
> > 	Will check the base again.
> > 	Please do not apply patch after [14/25]. There are here to
> > ask for comments, there are still building issues with patch after
> > [14/25]. Just help to comment whether it's on the right direction.
> > If so, I will send another version for formal review.
>
> I think you are doing that conversion wrong. You should first change
> all handlers which use the irq argument to:
>
>     handler(unsigned __irq, *desc)
>
> and add the local variable
>
>     unsigned irq;
>
> or
>
>     unsigned irq = irq_desc_get_irq(desc);
>
> for those.
>
> So that results in a pretty small series of patches.
>
> Once that is done, you can remove the irq argument in one go treewide.
>
> And you should do that with scripting aid. Coccinelle is the proper
> tool for this. If you have problems with cocci, you can politely ask
> Julia (Cc'ed) for help. She's the master of coccinelle :)
>
> The purpose of the script is to:
>
>     1) Identify all handlers
>
>        This can be done by analyzing the arguments of
>
>        	    __irq_set_handler()
> 	    irq_set_handler()
> 	    irq_set_chained_handler()
> 	    irq_alloc_generic_chip()
> 	    irq_alloc_domain_generic_chips()
> 	    irq_set_chip_and_handler_name()
> 	    irq_set_chip_and_handler()
> 	    __irq_set_handler_locked()
> 	    __irq_set_chip_handler_name_locked()
> 	    __irq_set_preflow_handler()
>
> 	If the argument is a function local to the file step 2 can
> 	take place.
>
> 	If it's a extern function, then the script should at least
> 	print out the function names. Those are in most of the cases
> 	the generic flow handler functions.

Looks reasonable.

>     2) Check the handlers whether they use the irq argument
>
>        For those which do:
>
>           Check whether they just use the argument instead of a local
>        	  variable, i.e. the first usage is an assignment. In that
>        	  case rename the argument to __irq and add a local variable
>
> 	      unsigned int irq;
>
> 	  If they use the irq argument for some computation or as
>        	  argument to another function, then you need to rename the
>        	  argument as well and add:
>
>        	      unsigned int irq = irq_get_desc_irq(desc);

Scripting this may be a little complex, because the variable can be used
in one way in one execution path (eg if branch) and another way (or not at
all) in another.  Let me know if help is needed.

julia

> So this results in a series of patches which can be merged right
> away. If that lot has hit Linus tree and right after -rc1, we run the
> script again and find and fix stuff which got added throughout the
> merge window. That might result in 0 to n patches.
>
> The last step is another script which does:
>
>     1) Identify all handlers
>
>     2) Remove the irq argument
>
> We can send that right after -rc1 so no new users can trickle in.
>
> It's really important to do that with scripts. It seems you try to do
> it via compile testing. But that will fail as you CANNOT execute all
> possible config combinations.
>
> Thanks,
>
> 	tglx
>
>