On 11.01.2016, at 14:38, Arnd Bergmann arnd@arndb.de wrote:
On Sunday 10 January 2016 13:17:17 Martin Sperl wrote:
What you propose is a major change to the clock framework, so I would hope that Eric (the original author of this clock-driver) would address it.
Maybe someone has a better idea for a pattern to use to achieve the required while maintaining “synchronization” between defines inside the dt-binding and the driver.
It's funny to look at the register list:
#define CM_VPUCTL 0x008 #define CM_VPUDIV 0x00c #define CM_SYSCTL 0x010 #define CM_SYSDIV 0x014 #define CM_PERIACTL 0x018 #define CM_PERIADIV 0x01c #define CM_PERIICTL 0x020 #define CM_PERIIDIV 0x024 #define CM_H264CTL 0x028 #define CM_H264DIV 0x02c
This one seems completely regular, there is always a pair of CTL and DIV registers, so I would have guessed that we could just take the index of that to completely avoid the need for the header file and just have a lookup table of each index.
I can see two possible ways forward:
a) deprecate the existing binding and write a new simpler driver to replace it with one that does not need the header. We probably need to keep both drivers around indefinitely though to deal with people that have their own dtb files, so this is a bit awkward.
I was thinking about this as well, but am concerned about the dt-changes.
Also initialization order may be an issue, so I have avoided that.
b) look at all the holes in the table and define new numbers for them now, then remove the count as the driver can simply hardcode the maximum number as it knows it will never change.
There are clocks for most of them - See my patch I sent some time ago.
Right now I have got a working version with the following patches: 817176d clk: bcm2835: add missing clocks 8f74701 clk: bcm2835: enable management of PCM clock 62a7d94 clk: bcm2835: move to a different initialization scheme.
There is one more thing that I would like to do before submitting those as a patch-set: enable the fractual divider, because right now we calculate the fractual divider value correctly, but we do not set the flag to enable it, so we are actually running too fast most of the times...