Dear Magnus,
In message 59b21cf20905180959p4e736d52g566b0e826e17c07a@mail.gmail.com you wrote:
After having browsed some powerpc code I can find two different ways the struct-thing is used: Variant A, all members declared volatile: struct controller_reg { volatile uint32_t reg1; volatile uint32_t reg2; }
This is probably older code.
struct controller_reg *controller = 0xCAFE0000;
Or variant B: struct controller_reg { uint32_t reg1; uint32_t reg2; }
This is OK.
volatile struct controller_reg *controller = 0xCAFE0000;
Also, is it OK to access the registers using reg = controller->reg1 or
No.
should we use reg = readl(&controller->reg1)?
Yes. All device acesses should use proper I/O accessor calls.
Note though that readl() is not considered to be portable; at least not across architectures. In Linux the ioread() / iowrite() are considered portable and should be used, i. e. ioread16(), ioread16be(), ioread32(), ioread32be() etc. - the plain ioread*() [i. e. without the "be" part] being little-endian on all architectures.
I am aware that we don't have appropriate support for all of this in the standard header files yet (actually I don't think Linu xhas this yet, either).
But that's the theory. Or at least the intersection of what I've been told, what I understood and what I still remember ;-)
I ask this since I don't want to iterate the patches on the mailing list too many times...
Good luck!
Best regards,
Wolfgang Denk