pxa-dt and multiple platforms support
Hi Daniel,
When you added support for device-tree for pxa3xx, in commit "ARM: pxa3xx: add generic DT machine code" (e7749a2) you wrote : "This file can be extended to also support pxa2xx".
I was wondering is that is really possible with the current pxa status, because :
- you have to include mach/pxa3xx.h
=> this includes mach/pxa3xx-regs.h
=> this is specific to pxa3xx architecture
If I add pxa27x support into the same file, I'll have to include mach/pxa27x.h (because the initialization functions are there, same reason as yours for mach/pxa3xx-regs.h).
=> this includes mach/pxa2xx-regs.h
=> this is specific to pxa2xx architecture
Of course there will be a lot of defines conflicts, as both platforms share the same register names ...
So did you have something specific in mind, or should I rename pxa-dt.c into pxa3xx-dt.c, and create pxa27-dt.c (this last option is my current choice unless a better ideas emerges) ?
Cheers.
Signature · 1 line
Robert
Hi Robert,
Quoted 16 lines · Robert Jarzmik
When you added support for device-tree for pxa3xx, in commit "ARM: pxa3xx: add generic DT machine code" (e7749a2) you wrote : "This file can be extended to also support pxa2xx".
I was wondering is that is really possible with the current pxa status, because :
- you have to includemach/pxa3xx.h
=> this includesmach/pxa3xx-regs.h
=> this is specific to pxa3xx architectureIf I add pxa27x support into the same file, I'll have to include
mach/pxa27x.h(because the initialization functions are there, same reason as yours formach/pxa3xx-regs.h).
=> this includesmach/pxa2xx-regs.h
=> this is specific to pxa2xx architecture
Quoted 6 lines
Of course there will be a lot of defines conflicts, as both platforms share the same register names ...
So did you have something specific in mind, or should I rename
pxa-dt.cintopxa3xx-dt.c, and createpxa27-dt.c(this last option is my current choice unless a better ideas emerges) ?
I think the latter is fine for now. I certainly didn't work on pxa27x DT bits yet.
Thanks,
Daniel
Quoted 21 lines · On Saturday 27 September 2014 20:08:51 Daniel Mack wrote:
Quoted 19 lines · Robert Jarzmik
When you added support for device-tree for pxa3xx, in commit "ARM: pxa3xx: add generic DT machine code" (e7749a2) you wrote : "This file can be extended to also support pxa2xx".
I was wondering is that is really possible with the current pxa status, because :
- you have to includemach/pxa3xx.h
=> this includesmach/pxa3xx-regs.h
=> this is specific to pxa3xx architectureIf I add pxa27x support into the same file, I'll have to include
mach/pxa27x.h(because the initialization functions are there, same reason as yours formach/pxa3xx-regs.h).
=> this includesmach/pxa2xx-regs.h
=> this is specific to pxa2xx architectureOf course there will be a lot of defines conflicts, as both platforms share the same register names ...
Quoted 6 lines
Quoted 3 lines
So did you have something specific in mind, or should I rename
pxa-dt.cintopxa3xx-dt.c, and createpxa27-dt.c(this last option is my current choice unless a better ideas emerges) ?I think the latter is fine for now. I certainly didn't work on pxa27x DT bits yet.
I think it should be easy enough to move the function declarations you need out of mach/pxa3xx.h and mach/pxa27x.h into a file that is shared between the platforms, and that is included by both. The DT file won't need the register definitions or a lot of the other things that are provided by mach/pxa3xx.h.
Arnd
Quoted 4 lines · Arnd Bergmann
I think it should be easy enough to move the function declarations you need out of
mach/pxa3xx.handmach/pxa27x.hinto a file that is shared between the platforms, and that is included by both. The DT file won't need the register definitions or a lot of the other things that are provided bymach/pxa3xx.h.
Yes, I had thought of that, and generic.h would be a good candidate.
I have not mentioned it because there is one thing I'm afraid of : should we discover that any pxa variant initialization requires one register to be accessed, then the pxa-dt should be split into several pxaXXX-dt.c anyway.
It is not the case AFAIK. I'm pondering the risk against having 2 split apart files ...
Signature · 1 line
Robert
Quoted 7 lines · On Saturday 27 September 2014 21:43:40 Robert Jarzmik wrote:
Quoted 4 lines · Arnd Bergmann
I think it should be easy enough to move the function declarations you need out of
mach/pxa3xx.handmach/pxa27x.hinto a file that is shared between the platforms, and that is included by both. The DT file won't need the register definitions or a lot of the other things that are provided bymach/pxa3xx.h.
Quoted 8 lines
Yes, I had thought of that, and
generic.hwould be a good candidate.I have not mentioned it because there is one thing I'm afraid of : should we discover that any pxa variant initialization requires one register to be accessed, then the pxa-dt should be split into several
pxaXXX-dt.canyway.It is not the case AFAIK. I'm pondering the risk against having 2 split apart files ...
The register definitions are total crap and you shouldn't use them for the DT file anyway. If you get to the point where you have to touch some registers, I think the right answer would be to have a DT node for the register block and use of_iomap with a platform specific offset, or a proper device driver.
It's more likely that you won't actually need this though.
Arnd