Hi Felix,
On Wednesday 11 July 2018 06:59 PM, Felix Brack wrote:
Hi Lokesh, many thanks for the feedback!
On 10.07.2018 08:42, Lokesh Vutla wrote:
Hi Felix,
On Thursday 24 May 2018 02:32 PM, Felix Brack wrote:
Hello,
I am working on a patch to synchronize the DTS files of the am33xx SoC with those from Linux 4.16.11 (current stable).
After some tiny modifications to the boards am335x-pdu001, am335x-evm, am335x-rut, am437x-gp-evm and am43x-epos-evm buildman passes without any warnings on the 46 am33xx based boards. This is required but not sufficient at all.
As I'm the maintainer of the am335x-pdu001 board I will use it to illustrate a sample problem. The am335x-pdu001 board uses the ns16550 driver which reads the property <reg-shift> from the DT. This property is required and must be set to a value of 2, otherwise it would default to 0. The am33xx.dtsi file currently used by U-Boot sets this property correctly for all 6 uarts. The am33xx.dtsi file from Linux 4.16.11 however does not define this property anymore. For the am335x-pdu001 board the fix is trivial: just add the property <reg-shift> to the board and U-Boot specific am335xx-pdu001-u-boot.dtsi file. But this has a major drawback: only the am335x-pdu001 board gets fixed. What about other boards requiring the <reg-shift> property?
I agree this is a problem. Also not all kernel drivers need this property as there are separate driver files for some of the compatibles. And few drivers has this property hard-coded.
The first idea that would probably come into mind is to put the property <reg-shift> into the SoC specific am33xx-u-boot.dtsi file. But this file is ignored if a board specific file already exists. Hence this can not be done.
One solution I can think of is add a Kconfig entry CONFIG_SYS_NS16550_REG_SHIFT. A default value of 0 can be assigned for this and can be updated in as necessary.
So when doing dev_read_u32_default for reg-shift, CONFIG_SYS_NS16550_REG_SHIFT can be passed as default instead of 0. This way no DTS files needs to be updated.
This sounds like a very good solution for the <reg-shift> property. I'm thinking about sending in a patch ...
That would be great.
Thanks and regards, Lokesh