device_t vs. struct device*

[Myles Watson]

We seem to have a mix of usage here.  Is there a reason we shouldn't
just use one?  Which one is preferred?
> From src/include/device/device.h:

typedef struct device * device_t;

Thanks,
Myles

[Carl-Daniel Hailfinger]

On 11.05.2009 15:43, Myles Watson wrote:
> We seem to have a mix of usage here.  Is there a reason we shouldn't
> just use one?  Which one is preferred?
> > From src/include/device/device.h:
>
> typedef struct device * device_t;

It's a bit complicated.
In early (CAR/ROMCC) code, device_t is u32 for stack size reasons.
Later, device_t is struct device *.

Using device_t allows to share some header files and/or code. However,
it gets confusing very fast. If you can untangle the mess, please go
ahead. I will ack that.
Maybe renaming the u32 variants of PCI access functions to *_early or so
is one way to proceed.

Regards,
Carl-Daniel

[Myles Watson]

On Mon, May 11, 2009 at 7:53 AM, Carl-Daniel Hailfinger
 wrote:
> On 11.05.2009 15:43, Myles Watson wrote:
> > We seem to have a mix of usage here.  Is there a reason we shouldn't
> > just use one?  Which one is preferred?
> > > From src/include/device/device.h:
> >
> > typedef struct device * device_t;
>
> It's a bit complicated.
> In early (CAR/ROMCC) code, device_t is u32 for stack size reasons.
> Later, device_t is struct device *.

Right.  I think I knew that but was repressing it. :)
> Using device_t allows to share some header files and/or code. However,
> it gets confusing very fast. If you can untangle the mess, please go
> ahead. I will ack that.

:)
> Maybe renaming the u32 variants of PCI access functions to *_early or so
> is one way to proceed.

So to avoid as much confusion as possible, should we use struct device
* in functions that should never be called early, or should we just
always use device_t because it is shorter and prettier?

Thanks,
Myles

[Carl-Daniel Hailfinger]

On 11.05.2009 16:04, Myles Watson wrote:
> On Mon, May 11, 2009 at 7:53 AM, Carl-Daniel Hailfinger
>  wrote:
> > Maybe renaming the u32 variants of PCI access functions to *_early or so
> > is one way to proceed.
>
> So to avoid as much confusion as possible, should we use struct device
> in functions that should never be called early,

Yes please. It helps a lot when reading the code.
> or should we just
> always use device_t because it is shorter and prettier?

Please don't. The v3 decision to avoid typedefs helped me a lot when I
was trying to understand the device code.

Regards,
Carl-Daniel

[ron minnich]

struct device * whereever possible is a good idea.

ron