Re: Storing the memory configuration used at link-time into a binary.

[Denis FERRANTI <denis dot ferranti at st dot com>]

Hi Nick,

I see your point. We just thought that it could be helpful for embedded 
systems using the binutils.
Thank you for the time spent on it.

About your proposal to use a linker command line option instead of a 
linker script directive, it's something that we envisaged, but there 
were two reasons to favor the linker script:
- we are dealing with memory configuration that is also described inside 
the linker script
- we wanted to let the programmer decide if this information is located 
in a non-loadable or loadable section. In the first case, he can use it 
only on the host machine (inside a simulator for example that has acces 
to the ELF file). In the second case, he can use the information at 
runtime on the target, but impacting the memory footprint of the 
application.

Cheers,

PS: We have a similar feature under final development to create a table 
of section information, mainly to be used at runtime on a minimal 
embedded system, where the application has no access to ELF headers.
If the meminfo were accepted, we would have submitted this one too ;-)

> From: *Nick Clifton* <nickc@redhat.com <mailto:nickc@redhat.com>>
> Date: Mon, Mar 2, 2009 at 3:52 PM
> Subject: Re: Storing the memory configuration used at link-time into a 
> binary.
> To: Denis FERRANTI <denis.ferranti@st.com <mailto:denis.ferranti@st.com>>
> Cc: bu-sourceware <binutils@sourceware.org 
> <mailto:binutils@sourceware.org>>
>
>
> Hi Denis,
>
>
>
>     Our co-workers at STMicroelectronics develop on many SoCs at a time
>     which sometimes only differ from memory configurations, so they have
>     [to take care] to use a compatible linker-script (among many).  That's
>     why we introduced a new linker-script statement "CREATE_MEMINFO_TABLE"
>     which stores the memory configuration used at link-time into a binary,
>     then the embedded "meminfo" table can be used by loaders to make
>     compatibility checks or by simulators for auto-configuration.
>
>
> Thank you for telling us about this idea.  I am not sure that there 
> would be any need for it outside of your own application, so I do not 
> really think that it is suitable for inclusion into the mainstream 
> sources.
>
> Also I would suggest that it might be better to control this feature 
> via a linker command line option rather than a linker script 
> directive. This would be more flexible because the name of the meminfo 
> table file could be specified on the command line, rather than being 
> fixed by the linker script.
>
> Cheers
>  Nick
--
Denis Ferranti