[PATCH 06/16] board: ti: Convert cape detection to use UCLASS framework

[Simon Glass]

Hi Tom,

On Mon, 6 Oct 2025 at 15:44, Tom Rini <trini at konsulko.com> wrote:
>
> On Mon, Oct 06, 2025 at 02:38:23PM -0600, Simon Glass wrote:
> > Hi Kory,
> >
> > On Fri, 3 Oct 2025 at 10:34, Kory Maincent <kory.maincent at bootlin.com> wrote:
> > >
> > > Migrate TI board cape detection from legacy extension support to the
> > > new UCLASS-based extension board framework.
> > >
> > > Signed-off-by: Kory Maincent <kory.maincent at bootlin.com>
> > > ---
> > >  arch/arm/mach-omap2/am33xx/Kconfig |  2 +-
> > >  arch/arm/mach-omap2/omap5/Kconfig  |  2 +-
> > >  board/ti/common/cape_detect.c      | 24 +++++++++++-------------
> > >  3 files changed, 13 insertions(+), 15 deletions(-)
> > >
> > > diff --git a/arch/arm/mach-omap2/am33xx/Kconfig b/arch/arm/mach-omap2/am33xx/Kconfig
> > > index dff4f1cf202..77f7938305b 100644
> > > --- a/arch/arm/mach-omap2/am33xx/Kconfig
> > > +++ b/arch/arm/mach-omap2/am33xx/Kconfig
> > > @@ -15,7 +15,7 @@ config TARGET_AM335X_EVM
> > >         select DM_GPIO
> > >         select DM_SERIAL
> > >         select TI_I2C_BOARD_DETECT
> > > -       select SUPPORT_EXTENSION_SCAN
> > > +       select SUPPORT_DM_EXTENSION_SCAN
> > >         imply CMD_DM
> > >         imply SPL_DM
> > >         imply SPL_DM_SEQ_ALIAS
> > > diff --git a/arch/arm/mach-omap2/omap5/Kconfig b/arch/arm/mach-omap2/omap5/Kconfig
> > > index 5394529658b..819490a8cf8 100644
> > > --- a/arch/arm/mach-omap2/omap5/Kconfig
> > > +++ b/arch/arm/mach-omap2/omap5/Kconfig
> > > @@ -38,7 +38,7 @@ config TARGET_AM57XX_EVM
> > >         select CMD_DDR3
> > >         select DRA7XX
> > >         select TI_I2C_BOARD_DETECT
> > > -       select SUPPORT_EXTENSION_SCAN
> > > +       select SUPPORT_DM_EXTENSION_SCAN
> > >         imply DM_THERMAL
> > >         imply SCSI
> > >         imply SPL_THERMAL
> > > diff --git a/board/ti/common/cape_detect.c b/board/ti/common/cape_detect.c
> > > index 7786bdda5d1..23c93964dc3 100644
> > > --- a/board/ti/common/cape_detect.c
> > > +++ b/board/ti/common/cape_detect.c
> > > @@ -22,7 +22,7 @@ static void sanitize_field(char *text, size_t size)
> > >         }
> > >  }
> > >
> > > -int extension_board_scan(struct list_head *extension_list)
> > > +static int ti_extension_board_scan(struct alist *extension_list)
> > >  {
> > >         unsigned char addr;
> > >         int num_capes = 0;
> > > @@ -31,7 +31,7 @@ int extension_board_scan(struct list_head *extension_list)
> > >                 struct am335x_cape_eeprom_id eeprom_header;
> > >                 char process_cape_part_number[17] = {'0'};
> > >                 char process_cape_version[5] = {'0'};
> > > -               struct extension *cape;
> > > +               struct extension cape = {0}, *_cape;
> > >                 struct udevice *dev;
> > >                 u8 cursor = 0;
> > >                 int ret, i;
> > > @@ -78,22 +78,20 @@ int extension_board_scan(struct list_head *extension_list)
> >
> > I suppose that a method called scan() could be added to the uclass and
> > implemented by this driver.
> >
> > >
> > >                 printf("BeagleBone Cape: %s (0x%x)\n", eeprom_header.board_name, addr);
> > >
> > > -               cape = calloc(1, sizeof(struct extension));
> > > -               if (!cape) {
> > > -                       printf("Error in memory allocation\n");
> > > -                       return num_capes;
> > > -               }
> > > -
> > > -               snprintf(cape->overlay, sizeof(cape->overlay), "%s-%s.dtbo",
> > > +               snprintf(cape.overlay, sizeof(cape.overlay), "%s-%s.dtbo",
> > >                          process_cape_part_number, process_cape_version);
> > > -               strlcpy(cape->name, eeprom_header.board_name,
> > > +               strlcpy(cape.name, eeprom_header.board_name,
> > >                         sizeof(eeprom_header.board_name));
> > > -               strlcpy(cape->version, process_cape_version,
> > > +               strlcpy(cape.version, process_cape_version,
> > >                         sizeof(process_cape_version));
> > > -               strlcpy(cape->owner, eeprom_header.manufacturer,
> > > +               strlcpy(cape.owner, eeprom_header.manufacturer,
> > >                         sizeof(eeprom_header.manufacturer) + 1);
> > > -               list_add_tail(&cape->list, extension_list);
> > > +               _cape = alist_add(extension_list, cape);
> > > +               if (!_cape)
> > > +                       return -ENOMEM;
> > >                 num_capes++;
> > >         }
> > >         return num_capes;
> > >  }
> > > +
> > > +U_BOOT_EXTENSION(cape, ti_extension_board_scan);
> >
> > Can the extension information go in the devicetree, so boards can
> > select their scheme that way?
>
> Since we're scanning to see what extension boards are present and so
> what device tree overlays to apply, I don't think that can work?

My understanding is that the scanning process is board-specific. For
example we would not want the Beagle mechanism to be compiled into
rpi. The idea of scanning things in a board-specific way seems like a
driver to me. Note there is also the SYSINFO uclass which can do
similar things.

The overlays being applied here are additions to what should already
be a fairly complete devicetree.

If this isn't possible, then where is the information known?

Regards,
SImon