On 09/29/2016 08:59 PM, Masahiro Yamada wrote:
2016-09-30 0:47 GMT+09:00 york sun york.sun@nxp.com:
On 09/28/2016 12:17 PM, Simon Glass wrote:
+Masahiro, who may know
Masahiro,
Need some advise on this. How do we deal with duplicated config?
York
On 28 September 2016 at 12:57, york sun york.sun@nxp.com wrote:
On 09/27/2016 10:55 AM, Simon Glass wrote:
Hi York,
On 27 September 2016 at 11:39, york sun york.sun@nxp.com wrote:
Guys,
I want to discuss to rearrange the current CONFIG_* macros used by Freescale Layerscape SoCs. We have been encouraged to use CONFIG_SYS_* macros to define hardware setting, and other CONFIG_* macros for user setting. Clearly this has changed. Kconfig options are used for most user settings from now on. I am OK with it. But for existing CONFIG_SYS_* macros, we need a plan to either move them out of CONFIG_* name space, or move some to Kconfig, if that makes sense. Since most our CONFIG_SYS_* macros are cross platforms (including armv7 and armv8), if using Kconfig option, do we add another mach-fsl folder to host the Kconfig, or somewhere else? We also have macros cross ARM and PowerPC. So we either move them out of arch/, or duplicate them for both arch.
If we move them to another name space, for example SYS_FSL_*, do we want to move them out of config.h?
Any suggestion/comment?
I wonder if some of these (the ones that represent actual values rather than enabling functions) can move to device tree, or tables?
While I try to move some options to Kconfig, I have an issue of duplicated names. For example, the MAX_CPUS is used in arch/x86/Kconfig. I know I can put another config with the same name, for example arch/arm/cpu/armv8/fsl-layerscape/Kconfig. The funny thing is when I search for this config option under menuconfig, it shows the location of x86, but showing defined in arch/arm/cpu/armv8/fsl-layerscape/Kconfig. Same thing happens when I try to move an option which is used by both PowerPC and ARM. This is not related to a common driver so I cannot put the option into a driver Kconfig (for example CONFIG_SYS_HAS_SERDES).
Is there a solution for this?
Right. The "Help" in menuconfig is just to display the path to the first Kconfig entry found. The result looks funny if we have multiple entries for the same option name.
I am not pretty sure about this particular case, but there should be general solutions.
(1) Rename the option to avoid name space conflict, e.g.
MAX_CPUS -> FSL_MAX_CPUS (for Freescale SoCs) MAX_CPUS -> X86_MAX_CPUS (for x86)
It is a solution, but I don't like it for the reason described below[1].
(2) Unify the Kconfig entry
As in Linux, we do not generally create multiple entries for the same option.
If the option is sensible only for some platforms, the general practice is to use "depends on HAVE_..." to hide the option on unrelated platforms.
If X86 and FSL are the only platforms that have interest in this option, we can save HAVE_... and put them directly in "depends on".
config MAX_CPUS int "Maximum number of CPUs permitted" depends on (SMP && X86) || VENDOR_FSLI am not quite sure about "VENDOR_FSL", so please choose your favorite one like SOC_FSL, PLAT_FSL, ARCH_FSL, or whatever.
This sounds like a bandage. I can go with this if there is no better solution.
Checking Linux Kconfig, I noticed the NR_CPUS are used in all archs. The difference is only one arch Kconfig is sourced by
source "arch/$SRCARCH/Kconfig"
I tried to do the same for U-Boot, but that will require we put ARCH at the command line of environmental variable. I think we moved away from this syntax. It also has some trouble to to use aarch64 because we put the actual code under arch/arm.
(3) Deprecate the option?
As Simon suggested Device Tree (or a kind of table?), if it is not really a config option, it may possible to specify it in a different way.
int fsl_get_max_cpus(void) { return 2; }
It is not a direct replace. The MAX_CPUS is used to define array size and others. Using device tree will defeat these codes and requires a lot of rewriting.
Now we have 8400 options in the whitelist, it is a crazy idea to move them as they are. I am wondering how many of them should really be CONFIG.
It is crazy. I am trying to move some to Kconfig and delete them from the white list. Some can be rewritten to avoid using CONFIG_* name space, or restructure code to eliminate the macros.
I admit I didn't pay attention to this change and it caught me by surprise.
[1] We have macros shared between PowerPC and ARM for Freescale QorIQ SoCs because they share similar designs at SoC level. Some options can be moved to driver Kconfig to avoid duplication. But there are many options are tied to hardware and they are not related to a common driver. Using the same macro simplifies the code and is easier to maintain. It is preferred to either PowerPC Kconfig, or ARM Kconfig to avoid duplication if we can.
York