Re: [patch 3/3] Add tsi108 On Chip Ethernet device driver support

[Jeff Garzik <jgarzik@pobox.com>]

Zang Roy-r61911 wrote:
> On Thu, 2006-09-21 at 12:26, Jeff Garzik wrote:
>> Zang Roy-r61911 wrote:
>>> +#define TSI108_ETH_WRITE_REG(offset, val) \
>>> +     writel(le32_to_cpu(val),data->regs + (offset))
>>> +
>>> +#define TSI108_ETH_READ_REG(offset) \
>>> +     le32_to_cpu(readl(data->regs + (offset)))
>>> +
>>> +#define TSI108_ETH_WRITE_PHYREG(offset, val) \
>>> +     writel(le32_to_cpu(val), data->phyregs + (offset))
>>> +
>>> +#define TSI108_ETH_READ_PHYREG(offset) \
>>> +     le32_to_cpu(readl(data->phyregs + (offset)))
>>
>> NAK:
>>
>> 1) writel() and readl() are defined to be little endian.
>>
>> If your platform is different, then your platform should have its own 
>> foobus_writel() and foobus_readl().
> 
> Tsi108 bridge is designed for powerpc platform. Originally, I use
> out_be32() and in_be32(). While there is no obvious reason to object
> using this bridge in a little endian system. Maybe some extra hardware
> logic needed for the bus interface. le32_to_cpu() can  be aware the
> endian difference.

To restate, readl() should read a little endian value, and return a 
CPU-endian value.  writel() should receive a CPU-endian value, and write 
a little endian value.

If your platform's readl/writel doesn't do that, it's broken.

That's why normal PCI drivers can use readl() and writel() on either 
big-endian or little-endian machines, without needing to use le32_to_cpu().

	Jeff