[Cbe-oss-dev] [PATCH] [v3] spidernet : add improved phy support

[Linas Vepstas]

Tested the patch, it works for me.  Thus I'll attach a pre-emptive 
Acked-by: Linas Vepstas <linas at austin.ibm.com>

However, some quibbbles, which I think would be nice to see fixed:

On Mon, Feb 12, 2007 at 09:35:34PM +0100, Jens Osterkamp wrote:
> 
> Index: linux-2.6.20/drivers/net/sungem_phy.c
> ===================================================================
> --- linux-2.6.20.orig/drivers/net/sungem_phy.c
> +++ linux-2.6.20/drivers/net/sungem_phy.c
>  
> +#define BCM5421_MODE_MASK	(1 << 5)

Customary practice is to have these in the heder file ... 

> +	mode = (phy_reg & BCM5421_MODE_MASK) >> 5;
> +
> +	if ( mode == BCM54XX_COPPER)

All this shifting makes the code hard to read and 
hard to verify for correctness.  Part of the problem 
seems to be that you are trying to re-cycle the 
BCM5421_COPPER and BCM5461_COPPER which are in 
different locations.

It would have been clearer to simply have

#define BCM5421_MODE_MASK  (1 << 5)
#define BCM5421_COPPER     0

if (phy_reg & BCM5421_MODE_MASK == BCM5421_COPPER)

> +	if ( (phy_reg & 0x0080) >> 7)

There is no need for the shift. The if statement is 
just as true (or false) with or without the shift.

> +#define BCM5461_FIBER_LINK	(1 << 2)
> +#define BCM5461_MODE_MASK	(3 << 1)
> +
> +	mode = (phy_reg & BCM5461_MODE_MASK ) >> 1;

More confusing shifting ....

> +enum {
> +	BCM54XX_COPPER,
> +	BCM54XX_FIBER,
> +	BCM54XX_GBIC,
> +	BCM54XX_SGMII,

Things that are being used for bitmak tests should probably
not be declared in an enum. For me, at least, there's a 
cognitive dissonance in doing things this way.

-- linas