Re: [PATCH v2 19/25] binman: Keep a separate list of entries for fit

[Simon Glass <sjg@chromium.org>]

Hi Alper,

On Thu, 3 Mar 2022 at 14:17, Alper Nebi Yasak <alpernebiyasak@gmail.com> wrote:
>
> On 24/02/2022 02:00, Simon Glass wrote:
> > The current implementation sets up the FIT entries but then deletes the
> > 'generator' ones so they don't appear in the final image.
>
> They still show up in the fdtmap if I add one to rockchip-u-boot.dtsi:
>
>   $ binman ls -i u-boot-rockchip.bin
>   Name                Image-pos  Size     Entry-type    Offset
>   -------------------------------------------------------------
>   main-section                0   103520  section             0
>     mkimage                   0    1a000  mkimage             0
>     fit                   1a000    e8d74  fit             1a000
>       u-boot              1a644    b1898  section           644
>         u-boot-nodtb      1a644    b1898  u-boot-nodtb        0
>       @atf-SEQ                0        0  section             0
>         atf-bl31              0        0  atf-bl31            0
>       @tee-SEQ                0        0  section             0
>         tee-os                0        0  tee-os              0
>       @fdt-SEQ                0        0  section             0
>     fdtmap               102d74      79c  fdtmap         102d74
>
> But not simple-bin.map:
>
>   ImagePos    Offset      Size  Name
>   00000000  00000000  00103520  main-section
>   00000000   00000000  0001a000  mkimage
>   0001a000   0001a000  000e8d74  fit
>   0001a644    00000644  000b1898  u-boot
>   0001a644     00000000  000b1898  u-boot-nodtb
>   00102d74   00102d74  0000079c  fdtmap
>
> This happens in v1 as well, but I hadn't checked it then.
>
> (The "fit" size and "fdtmap" offset are off by 0x10 too, but I'm not
> sure why yet.)

Well let's fix this after the series. We need tests for the map and so on.

>
> > This is a bit clumsy. We cannot build the image more than once, since the
> > generator entries are lost during the first build. Binman requires that
> > calling BuildSectionData() multiple times returns a valid result each
> > time.
>
> I think the generator entries should be turned into concrete entries and
> be removed in _gen_entries(), so BuildSectionData() should work on the
> entries that were generated. This way the individual entries would show
> up in fdtmap and could be binman-extracted/replaced as well.

That makes sense, but I'm not sure how to implement it. The split-elf
thing needs the contents of the entries which is not available at the
start. It is likely available before BuildSectionData() is called, but
not necessarily. So the template needs to hang around at least as long
as that.

I think there is something we could do here, but it isn't quite clear
to me. Perhaps we need a expand-nodes-based-on-contents phase in
control.py ? Eek...

>
> For FDTs we can generate blob entries for each file, for ELFs maybe we
> can split them into files like the script used to do (bl31_0x<addr>.bin)
> and do the same?

I'm not a big fan of adding files. Binman should be able to hold
everything in memory. It does generate files for later inspection
though, so yes we could add this feature.

>
> Maybe some new entry types, "data" for arbitrary data unrelated to a
> file whose contents we can set, or "elf-segment" that can extract a
> segment from an ELF file (but I don't like the information loss there).

Yes I quite like the idea of new entry types. In fact I was hoping to
turn split-elf into an entry type (one that generated multiple
entries). But I decided to stop before doing that since we really need
to gets the code in there, fix the bugs /move forward with some of the
ideas you and others have.

>
> > Keep a separate, private list which includes the generator nodes and use
> > that where needed, to correct this problem. Ensure that the missing list
> > includes removed generator entries too.
> >
> > Signed-off-by: Simon Glass <sjg@chromium.org>
> > ---
> >
> > (no changes since v1)
>
> I'm more and more convinced that generator nodes needs a thorough
> redesign, but the patch is consistent with status quo, so:
>
> Reviewed-by: Alper Nebi Yasak <alpernebiyasak@gmail.com>
>
> >  tools/binman/etype/fit.py | 33 ++++++++++++++++++++++++++++-----
> >  1 file changed, 28 insertions(+), 5 deletions(-)
> >
> > diff --git a/tools/binman/etype/fit.py b/tools/binman/etype/fit.py
> > index e1b056f56e..30b20a07a2 100644
> > --- a/tools/binman/etype/fit.py
> > +++ b/tools/binman/etype/fit.py
> > @@ -163,10 +163,15 @@ class Entry_fit(Entry_section):
> >                  key: relative path to entry Node (from the base of the FIT)
> >                  value: Entry_section object comprising the contents of this
> >                      node
> > +            _priv_entries: Internal copy of _entries which includes 'generator'
> > +                entries which are used to create the FIT, but should not be
> > +                processed as real entries. This is set up once we have the
> > +                entries
>
> Maybe this could be "_templates" or "_entry_generators" and only keep
> track of the generator entries.

Again I think we can revisit that. I feel it is a bit messy right now, too.

>
> >          """
> >          super().__init__(section, etype, node)
> >          self._fit = None
> >          self._fit_props = {}
> > +        self._priv_entries = {}
> >
> >          for pname, prop in self._node.props.items():
> >              if pname.startswith('fit,'):
> >
> > [...]

Regards,
Simon