PowerPC iotable_init equivalent?

PowerPC iotable_init equivalent?

[Eddie Dawydiuk]

Hello,

In the past I've worked with ARM architectures where I could setup virtual / 
physical address mappings so I don't have to ioremap then pass around pointers. 
Does PowerPC have an equivalent abstraction? If not whats the recommended approach?

That is, is there a better approach to the following...

volatile static unsigned char *my_reg = NULL;

static inline void read_func() {

	if (!my_reg)
         	my_reg = (unsigned char *)
			ioremap(REG_PHYS_BASE, REG_SIZE);
//do something with the reg
}


static inline void write_func() {

	if (!my_reg)
         	my_reg = (unsigned char *)
			ioremap(REG_PHYS_BASE, REG_SIZE);
//do something with the reg
}

-- 
Best Regards,
________________________________________________________________
  Eddie Dawydiuk, Technologic Systems | voice:  (480) 837-5200
  16525 East Laser Drive 	     | fax:    (480) 837-5300
  Fountain Hills, AZ 85268            | web: www.embeddedARM.com

Re: PowerPC iotable_init equivalent?

[Roderick Colenbrander]

On Fri, Apr 17, 2009 at 11:21 PM, Eddie Dawydiuk <eddie@embeddedarm.com> wr=
ote:
> Hello,
>
> In the past I've worked with ARM architectures where I could setup virtua=
l /
> physical address mappings so I don't have to ioremap then pass around
> pointers. Does PowerPC have an equivalent abstraction? If not whats the
> recommended approach?
>
> That is, is there a better approach to the following...
>
> volatile static unsigned char *my_reg =3D NULL;
>
> static inline void read_func() {
>
> =A0 =A0 =A0 =A0if (!my_reg)
> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0my_reg =3D (unsigned char *)
> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0ioremap(REG_PHYS_BASE, REG=
_SIZE);
> //do something with the reg
> }
>
>
> static inline void write_func() {
>
> =A0 =A0 =A0 =A0if (!my_reg)
> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0my_reg =3D (unsigned char *)
> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0ioremap(REG_PHYS_BASE, REG=
_SIZE);
> //do something with the reg
> }
>
> --
> Best Regards,
> ________________________________________________________________
> =A0Eddie Dawydiuk, Technologic Systems | voice: =A0(480) 837-5200
> =A016525 East Laser Drive =A0 =A0 =A0 =A0 =A0 =A0 =A0| fax: =A0 =A0(480) =
837-5300
> =A0Fountain Hills, AZ 85268 =A0 =A0 =A0 =A0 =A0 =A0| web: www.embeddedARM=
.com
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-dev
>

In general you set devices up a dts file. You can reach its contents
using openfirmware functions (of_*). In general you use an of_*
function to look up your periperhal in the device tree. Once you have
its node you would able to do e.g. of_iomap (which uses ioremap with
the right offsets behind its back) to map the device. So you don't
hard code physical addresses anymore in the code.

Regards,
Roderick Colenbrander

Re: PowerPC iotable_init equivalent?

[Kumar Gala]

On Apr 17, 2009, at 4:21 PM, Eddie Dawydiuk wrote:

> Hello,
>
> In the past I've worked with ARM architectures where I could setup  
> virtual / physical address mappings so I don't have to ioremap then  
> pass around pointers. Does PowerPC have an equivalent abstraction?  
> If not whats the recommended approach?
>
> That is, is there a better approach to the following...
>
> volatile static unsigned char *my_reg = NULL;
>
> static inline void read_func() {
>
> 	if (!my_reg)
>        	my_reg = (unsigned char *)
> 			ioremap(REG_PHYS_BASE, REG_SIZE);
> //do something with the reg
> }
>
>
> static inline void write_func() {
>
> 	if (!my_reg)
>        	my_reg = (unsigned char *)
> 			ioremap(REG_PHYS_BASE, REG_SIZE);
> //do something with the reg
> }

Yes have a struct that keeps track of the virt addr and do the mapping  
once at init time.  This is what most drivers do.

- k

Re: PowerPC iotable_init equivalent?

[Grant Likely]

On Fri, Apr 17, 2009 at 3:21 PM, Eddie Dawydiuk <eddie@embeddedarm.com> wrote:
> Hello,
>
> In the past I've worked with ARM architectures where I could setup virtual /
> physical address mappings so I don't have to ioremap then pass around
> pointers. Does PowerPC have an equivalent abstraction? If not whats the
> recommended approach?

No.  Predefining virt->phys mappings is fragile since it makes
assumptions about how the kernel is going to carve up the virtual
address space.  Better to let the kernel allocate virtual ranges as it
needs them.

As Kumar says, do your ioremap() (or, even better: of_iomap()) in your
driver's probe function and store it in the driver's private data
structure.

g.

-- 
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.