Re: [PATCH 15/34] image: Add a new type for extlinux

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 4646 bytes --]

On Fri, Oct 18, 2024 at 12:19:26PM -0600, Simon Glass wrote:
> Hi Tom,
> 
> On Fri, 18 Oct 2024 at 11:54, Tom Rini <trini@konsulko.com> wrote:
> >
> > On Fri, Oct 18, 2024 at 11:20:45AM -0600, Simon Glass wrote:
> > > Hi Tom,
> > >
> > > On Fri, 18 Oct 2024 at 09:17, Tom Rini <trini@konsulko.com> wrote:
> > > >
> > > > On Fri, Oct 18, 2024 at 08:57:02AM -0600, Simon Glass wrote:
> > > > > Hi Tom,
> > > > >
> > > > > On Thu, 17 Oct 2024 at 21:14, Tom Rini <trini@konsulko.com> wrote:
> > > > > >
> > > > > > On Thu, Oct 17, 2024 at 05:23:54PM -0600, Simon Glass wrote:
> > > > > >
> > > > > > > Add an image type for the extlinux.cfg file, which U-Boot supports
> > > > > > > reading.
> > > > > > >
> > > > > > > Signed-off-by: Simon Glass <sjg@chromium.org>
> > > > > > > ---
> > > > > > >
> > > > > > >  boot/image.c    | 1 +
> > > > > > >  include/image.h | 1 +
> > > > > > >  2 files changed, 2 insertions(+)
> > > > > > >
> > > > > > > diff --git a/boot/image.c b/boot/image.c
> > > > > > > index abac254e026..cf16fb7fbf7 100644
> > > > > > > --- a/boot/image.c
> > > > > > > +++ b/boot/image.c
> > > > > > > @@ -183,6 +183,7 @@ static const table_entry_t uimage_type[] = {
> > > > > > >       {       IH_TYPE_FDT_LEGACY, "fdt_legacy", "legacy Image with Flat Device Tree ", },
> > > > > > >       {       IH_TYPE_RENESAS_SPKG, "spkgimage", "Renesas SPKG Image" },
> > > > > > >       {       IH_TYPE_STARFIVE_SPL, "sfspl", "StarFive SPL Image" },
> > > > > > > +     {       IH_TYPE_EXTLINUX_CFG, "extlinux_cfg", "Extlinux configuration" },
> > > > > > >       {       -1,                 "",           "",                   },
> > > > > > >  };
> > > > > >
> > > > > > This whole part of the series feels like we're abusing IH_TYPE_ far past
> > > > > > what it's used for. Especially the command line string one. It seems
> > > > > > like the main use is to be able to later on print something human
> > > > > > friendly. Please try and figure out some other way to do that. Thanks.
> > > > >
> > > > > Yes I had the same thought, particularly with cmdline, as you say.
> > > > >
> > > > > The obvious alternative is to just have a string which communicates
> > > > > the type. Then I can use IH_TYPE_INVALID, perhaps, with an extra
> > > > > string indicating what it really is.
> > > > >
> > > > > But one advantage of the approach in this patch is that, for bootmeths
> > > > > which include a cmdline, it can be identified and added as a file.
> > > > > That includes zimage and ChromiumOS, but not EFI. It will allow me to
> > > > > get rid of the cmdline in struct bootflow, perhaps.
> > > >
> > > > Yeah, this all sounds like IH_TYPE is the wrong direction to expand.
> > > > Where you're grabbing these things from is where you might get a hint as
> > > > to what these things are and that's where to take the store the
> > > > knowledge for later.
> > >
> > > They are grabbed from several places:
> > > - extlinux comes from disk, with the extlinux bootmeth, but can also
> > > come from 'include' files within that scripts, also pxe
> > > - EFI apps from from disk with the EFI bootmeth
> > > - kernel, ramdisk and fdt some from most bootmeths
> > > - cmdline comes from extlinux (with zImage only) and ChromeOS,
> > > probably Android but I haven't looked
> > > - logo comes just from script so far, but I suppose UKI and FIT will
> > > have it eventually
> > >
> > > Can you be more specific with your suggestion? I am not seeing an
> > > alternative right now.
> >
> > I'm just saying that this is not what IH_TYPE is. As you said maybe you
> > can store this string information elsewhere? And I'm suggesting that
> > perhaps when you load a thing you'll then also know what the thing is.
> 
> Yes, I agree. 'Type' is really about the actual file that you are
> reading, not the data types of the bits within it, for example. That's
> the problem with my approach.
> 
> The other idea I had was to add a new thing to image.h (IH_DATA_xxx).
> But I wasn't sure it was worth it. Also, if you look at
> fit_image_load() it reads an image of a certain type out of an
> existing (containing) FIT. So we already have images within images.
> 
> How about I just put a data type in bootflow.h ? Then it will be
> separate from image.h and I'm not blurring the line any further
> between an image and its component parts? We can always change it
> later if we find a better way.

Certainly somewhere else, and then lets move this to the part of the
conversion on patch 23 where I suggest another way to handle these
issues.

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]