Re: ioremap return type

[Philipp Rumpf]

On Thu, Aug 10, 2000 at 02:14:13PM +0200, Jamie Lokier wrote:
> Many drivers use a struct to define the offsets in their I/O or iomem
> space.  E.g. acenic.c: readl(&regs->CpuCtrl).  We know that's not
> portable in the sense that struct layout is not guaranteed.  However,
> provided all the fields are suitably padded & aligned, it works in
> practice.  (So far..)

Not only does it work in practice, it's used by most drivers and file systems
when they describe on-disk or dma structures.  Currently the rules are:

 - naturally align everything
 - all struct sizes might be aligned to a multiple of four bytes

By now, I believe it is more likely that gcc would be fixed for a new
architecture that violates this assumption, so it's not worth it to go
around and "fix" those drivers.

As for using two arguments to readl, that would break the &regs->RegName
trick, unless people are happy with a global #define Regs ((struct foo *)0)
and read32(regs, &Regs->RegName);  maybe it's best to just stop doing that
though.

	Philipp Rumpf