On Tue, Sep 10, 2019 at 11:01:18AM +0000, Patrick DELAUNAY wrote:
Hi,
From: Wolfgang Denk wd@denx.de Sent: samedi 7 septembre 2019 13:52
Dear Patrick,
In message 1567530547-14331-1-git-send-email-patrick.delaunay@st.com you wrote:
Add a new flag CONFIG_ENV_SUPPORT to compile all the environment features in U-Boot (attributes, callbacks and flags). It is the equivalent of the 2 existing flags
I think this is a bda name, as it is misleading. It sounds as if it is used to enable the support of environment vaiables at all, which it does not - instead it only enables / disables a few specific extended features. This must be reflected in the name.
Yes, I use the name than SPL/TPL to use the macro CONFIG_IS_ENABLED(...)
And I agree the name seens not perfect.
- CONFIG_SPL_ENV_SUPPORT for SPL
- CONFIG_TPL_ENV_SUPPORT for TPL
These pre-existing name are defined in common/spl/Kconfig
With the same issue (env/common.o env/env.o are always compiled for SPL/TPL so it is alo bad names)
This scares me. Why are there different settings for SPL, TPL and U-Boot proper? This looks conceptually broken to me - if a system is configured to use a specific set of environment features and extensions, then the exact same settings must be use in all of SPL, TPL and U-Boot proper.
I understand that it may be desirable for example to reduce the size of the SPL by omitting some environment extensions, but provide these in U-Boot proper, but this is broken and dangerous. For example, U-Boot flags are often used to enforce a certain level of security (say, by making environment variables read- only or such).
I agree, that scare me also. For me also ENV_SUPPORT should be always enable for U-Boot.
My preferred proposal was only to solve the regression introduced by my initial patch and always set change_ok for U-Boot proper (when CONFIG_SPL_BUILD is not defined):
struct hsearch_data env_htab = {
- #if CONFIG_IS_ENABLED(ENV_SUPPORT)
/* defined in flags.c, only compile with ENV_SUPPORT */+#if !defined(CONFIG_SPL_BUILD) || CONFIG_IS_ENABLED(ENV_SUPPORT)
/* defined in flags.c, only compile with ENV_SUPPORT for SPL / TPL */ .change_ok = env_flags_validate,#endif };
http://u-boot.10912.n7.nabble.com/U-Boot-Environment-flags-broken-for-U-Boot...
The same test is already done in:
drivers/reset/reset-socfpga.c:47:#if !defined(CONFIG_SPL_BUILD) || CONFIG_IS_ENABLED(ENV_SUPPORT) drivers/input/input.c:656:#if !defined(CONFIG_SPL_BUILD) || CONFIG_IS_ENABLED(ENV_SUPPORT)
The same level of handling and protection must also be maintained in SPL and TPL.
if I understood correctly the proposed dependency is : TPL ENV_SUPPORT select SPL_ENV_SUPPORT SPL ENV_SUPPORT select ENV_SUPPORT
So please reconsider this whole implementation, and make sure that only a single macro ise used everywhere to enable these features.
But, if I use the same CONFIG for the 3 binary SPL,TPL and U-Boot, l increase the size of TPL/SPL for all the platforms when these additional features are not needed.
And I am not the sure of the correct dependency for ENV between binary.
Heiko what is you feedback on Wolfgang remarks....
Do you think that I need to come back to the first/simple proposal ?
My two cents is that I would like to see a regression fix "soon" and for this release, and some corrections / updates to names, what is/isn't possible to enable (keeping in mind what is desirable to allow) for the next release. Thanks all!