[U-Boot] [PATCH RFC 0/4] Reworking of env

Gerlando Falauto gerlando.falauto at keymile.com
Fri Sep 30 14:04:54 CEST 2011


Hi all,

trying to follow Wolfgang Denk's suggestion in
http://lists.denx.de/pipermail/u-boot/2011-September/102118.html:

>I agree that it makes sense to generalize and clean up this interface.
>It makes sense to select individual variables, and it makes sense to
>unify the "-f" handling to enforce actions on protected variables
>(while without "-f" only actions on the "normal" variables should be
>done).

>I can even imagine introducing a new variable that contains the name
>of the write-protected variables (and probably other properties, like
>being excluded from saveenv, etc.) - this has been discussed a number
>of times before, now we have the code base in place to actually
>implement it.

>All we need to do is extend the struct entry (in "include/search.h")
>by an "int flags"), and we can there register properties like
>read-only, don't-save etc.  In a first step this could be added
>transparently - so we could remove all the special handling of
>"ethaddr", "serial#" etc. in common/cmd_nvedit.c; then we could unify
>this to include "eth1addr" etc as well; then  we could extend it to
>read the names of such variables and their properties from a variable,
>etc.

here I am proposing a set of changes in the behaviour of the environment 
import/set_to_default functions.

Since a PATCH-er is worth a thousand words, I decided to take the risk 
and implement it first so to get your feedback on code rather than words...

=======
PATCH 1
=======
The most shocking change lies in the "new" himport_ex() function, which
has 3 new arguments:

1-2) "nvars", "vars", number and list of variables to take into account (0 means ALL)

3) "apply" callback function which is in charge of checking whether a 
variable can be overwritten, and possibly immediately apply the changes. 
This parameter would be either set to NULL (in which case nothing should 
change wrt to the past -- i.e. environment is blindly imported) or to 
"env_check_apply()" function, whose code was taken away from _do_env_set().
This would be useful, for instance, for "baudrate" or "stdin,stderr,stdout", 
whose changes would not otherwise be effective until the next reboot.

The idea is that there should be a single place where all the checks are
to be performed. So the same function env_check_apply() is called from 
_do_env_set() as well (thus keeping the previous behavior).

I know we're talking about heavy changes here... this is a just a draft -- 
I would just like some feedback about the whole idea, before putting more 
time into it.

=======
PATCH 2
=======
Here I added the same behaviour when deleting variables.
Not quite sure whether this makes sense or not.

=======
PATCH 3
=======
Here I added the code for setting to default individual variables

=======
PATCH 4
=======
Here I implemented the code for importing individual variables
(and also taking into account "-f" for forcing).

Thank you for your patience,
Gerlando Falauto

Gerlando Falauto (4):
  Groundwork for generalization of env interface
  env: check and apply changes on delete/destroy
  env: implement selective "env default"
  env: implement "env import -n var[,var...]"

 README                   |    2 +
 common/cmd_nvedit.c      |  221 +++++++++++++++++++++++++++++++--------------
 common/env_common.c      |   28 ++++++-
 include/config_cmd_all.h |    1 +
 include/environment.h    |    7 ++
 include/search.h         |   19 ++++-
 lib/hashtable.c          |   67 +++++++++++++--
 7 files changed, 266 insertions(+), 79 deletions(-)



More information about the U-Boot mailing list