Re: [PATCH] isa_{read,write}{w,l} fixed

[Jeff Garzik]

Philipp Rumpf wrote:
> 
> > I do. And the patch is wrong. Physical addresses are unsigned long. Take the
> > errors as a chance to fix up the types in the driver.
> 
> I disagree.  A lot of drivers seem to use structs as a cheap way to get
> separate namespaces for their readl() / sbus_readl() offsets, and inconsistent
> behaviour between readb() and isa_readb()  (as well as isa_readb() on ARM
> and isa_reab() on x86) would be unnecessarily confusing.

Two key points here --

* isa_xxx are only to ease the transition to newstyle ioremap.  if you
are hacking code, go ahead and update it to use
ioremap/readX/writeX/iounmap.

* You can never directly use a struct to access an offset in MMIO
space.  To do so is broken and bad bad bad.  The only acceptable way is

	my_struct = ioremap (...);
	my_reg = readb(&my_struct->foo_register);

I much prefer to avoid structs, and use macros to access registers
instead:

	my_reg = RTL_R8 (FooRegisterConstant);

Regards,

	Jeff

-- 
Jeff Garzik         | "Vegetarian" is the Indian word
Building 1024       | for 'lousy hunter.'
MandrakeSoft, Inc.  |