[U-Boot] Setting ipaddr via setenv() - H_INTERACTIVE vs H_PROGRAMMATIC

[Stefan Roese]

Hi Joe,

(adding Tom to Cc as this seems to be a fundamental issue)

On 01.03.2016 20:40, Joe Hershberger wrote:
> On Tue, Mar 1, 2016 at 4:35 AM, Stefan Roese <sr at denx.de> wrote:
>> Hi Joe,
>>
>> I'm currently stumbling over a problem in some board specific code
>> (not in mainline yet), that sets the "ipaddr" env variable via
>> setenv(). But as I now noticed, the callback on_ipaddr() doesn't
>> set net_ip to this new value. As flags has H_PROGRAMMATIC set and
>> op_ipaddr() skips setting this value in this case.
>>
>> I fail to see why we need this check in on_ipaddr(). Could you
>> please explain why this is needed? Why would someone use setenv()
>> for the "ipaddr" and not want to also update the net_ip value.
>> This results in the "ipaddr" value being updated but net_ip
>> still being configured to the old value.
> 
> The purpose is that programmatic accesses may write directly to the
> net_ip variable, but the user on the console cannot. This is because
> the "programmatic" accesses to these variables is expected to be
> things like the dhcp command and the linklocal command after
> successful netloop.

Please correct me, but it seems to be implemented the other way around.
If H_PROGRAMMATIC is set (called via setenv()) the net variables
(net_ip...) will *not* be set:

static int on_ipaddr(const char *name, const char *value, enum env_op op,
	int flags)
{
	if (flags & H_PROGRAMMATIC)
		return 0;

	net_ip = string_to_ip(value);

	return 0;
}

> You can see this call in the netboot_update_env()
> function in cmd/net.c.

netboot_update_env() does not update the internal variables. At least
not in my tests. op_ipaddr() returns directly after the "flags" check
(see above).

I've checked how "dhcp" works. And here the variables are updated
in store_net_params() via explicit net_copy_ip() calls.
 
>> Or if this is really needed, what is the correct way to update
>> the ipaddr (env variable and net_ip value) from the U-Boot code?
> 
> You can work around it in a similar way to what I did in the initial
> change (94b467b14ed908c89a0780256e89b375aa3cf3ef - env: Distinguish
> finer between source of env change). The do_env_edit function used to
> call setenv, but I changed it to call as though it was interactive.
> You could do the same thing in your board driver.

Again, my findings are that this is implemented the other way around.
setenv from the user via the command "setenv ipaddr ..." will update
the net_ip variable. And code that calls setenv() will not.

I still fails to see why one of both should not update the net_ip
(and others) variable. From my understanding setenv() from code and
"setenv ipaddr ..." from the prompt should update the internal
variables.

Thanks,
Stefan