[U-Boot] [PATCH] rpi3: Enable verified boot from FIT image

[Simon Glass]

Hi Heinrich,

On Mon, 2 Sep 2019 at 04:24, Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
>
> On 9/2/19 12:30 PM, Matthias Brugger wrote:
> > +Alex, Lukas, Heinrich, Bin and Simon
> >
> > On 31/07/2019 10:16, Jun Nie wrote:
> >> Matthias Brugger <mbrugger at suse.com> 于2019年7月31日周三 下午4:05写道：
> >>>
> >>>
> >>>
> >>> On 11/07/2019 05:55, Jun Nie wrote:
> >>>> Enable verified boot from FIT image with select configs
> >>>> and specify boot script image node in FIT image, the FIT
> >>>> image is verified before it is run.
> >>>>
> >>>> Code that reusing dtb in firmware is disabled, so that
> >>>> the dtb with pubic key packed in u-boot.bin can be used
> >>>> to verify the signature of next stage FIT image.
> >>>>
> >>>> Signed-off-by: Jun Nie <jun.nie at linaro.org>
> >>>> ---
> >>>>   board/raspberrypi/rpi/rpi.c |  6 ++++++
> >>>>   include/configs/rpi.h       | 15 ++++++++++++++-
> >>>>   2 files changed, 20 insertions(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/board/raspberrypi/rpi/rpi.c b/board/raspberrypi/rpi/rpi.c
> >>>> index 617c892..950ee84 100644
> >>>> --- a/board/raspberrypi/rpi/rpi.c
> >>>> +++ b/board/raspberrypi/rpi/rpi.c
> >>>> @@ -297,6 +297,7 @@ static void set_fdtfile(void)
> >>>>        env_set("fdtfile", fdtfile);
> >>>>   }
> >>>>
> >>>> +#ifndef CONFIG_FIT_SIGNATURE
> >>>>   /*
> >>>>    * If the firmware provided a valid FDT at boot time, let's expose it in
> >>>>    * ${fdt_addr} so it may be passed unmodified to the kernel.
> >>>> @@ -311,6 +312,7 @@ static void set_fdt_addr(void)
> >>>>
> >>>>        env_set_hex("fdt_addr", fw_dtb_pointer);
> >>>>   }
> >>>> +#endif
> >>>>
> >>>>   /*
> >>>>    * Prevent relocation from stomping on a firmware provided FDT blob.
> >>>> @@ -393,7 +395,9 @@ static void set_serial_number(void)
> >>>>
> >>>>   int misc_init_r(void)
> >>>>   {
> >>>> +#ifndef CONFIG_FIT_SIGNATURE
> >>>>        set_fdt_addr();
> >>>> +#endif
> >>>>        set_fdtfile();
> >>>>        set_usbethaddr();
> >>>>   #ifdef CONFIG_ENV_VARS_UBOOT_RUNTIME_CONFIG
> >>>> @@ -470,6 +474,7 @@ int board_init(void)
> >>>>        return bcm2835_power_on_module(BCM2835_MBOX_POWER_DEVID_USB_HCD);
> >>>>   }
> >>>>
> >>>> +#ifndef CONFIG_FIT_SIGNATURE
> >>>>   /*
> >>>>    * If the firmware passed a device tree use it for U-Boot.
> >>>>    */
> >>>> @@ -479,6 +484,7 @@ void *board_fdt_blob_setup(void)
> >>>>                return NULL;
> >>>>        return (void *)fw_dtb_pointer;
> >>>>   }
> >>>> +#endif
> >>>
> >>> Just to get this clear we need this because we want to pass the device tree via
> >>> OF_SEPARATE, correct?
> >>
> >> You are right.  U-boot need to read he signature from dtb.
> >>
> >>>
> >>>>
> >>>>   int ft_board_setup(void *blob, bd_t *bd)
> >>>>   {
> >>>> diff --git a/include/configs/rpi.h b/include/configs/rpi.h
> >>>> index f76c7d1..ba91205 100644
> >>>> --- a/include/configs/rpi.h
> >>>> +++ b/include/configs/rpi.h
> >>>> @@ -180,11 +180,24 @@
> >>>>
> >>>>   #include <config_distro_bootcmd.h>
> >>>>
> >>>> +#ifdef CONFIG_FIT_SIGNATURE
> >>>> +#define FIT_BOOT_CMD                                                 \
> >>>> +     "boot_a_script="                                                \
> >>>> +             "load ${devtype} ${devnum}:${distro_bootpart} "         \
> >>>> +                     "${scriptaddr} ${prefix}${script}; "            \
> >>>> +             "iminfo ${scriptaddr};"                                 \
> >>>> +             "if test $? -eq 1; then reset; fi;"                     \
> >>>> +             "source ${scriptaddr}:bootscr\0"
> >>>> +#else
> >>>> +#define FIT_BOOT_CMD ""
> >>>> +#endif
> >>>> +
> >>>
> >>> Doesn't this overwrite the boot_a_script in distro_bootcmd?
> >>>
> >>> Would it make sense to add FIT booting to the distro boot command?
> >>>
> >>> Regards,
> >>> Matthias
> >>
> >> Yes, it overwrite the boot_a_script in distro_bootcmd. It is make
> >> sense to add this to the distro boot command. I can send another patch
> >> to move these lines to common code later.
> >>
> >
> > Question to the people just added, as you have relevant submission to
> > distroboot. Do you think it makes sense to add FIT_BOOT_CMD to that?
> >
> > Regards,
> > Matthias
>
> The idea of distro-boot was to make it easier for Linux distributions to
> update the information needed by U-Boot to find the right kernel and
> ramdisk.
>
> According to doc/README.distro file extlinux.conf should be used for the
> communication between the distribution and U-Boot. Some distributions
> like Debian still rely on boot.scr.
>
> Many distributions (OpenBSD, FreeBSD, Suse, Fedora) have moved from
> distro-boot to UEFI as booting standard. Unfortunately we have not
> documented our support for this in doc/README.distro (TODO for me).
> Takahiro is working on secure boot using UEFI. Once completed this could
> obsolete FIT images.

Oh please no.

>
> Would we expect Linux distributions to provide FIT images upon kernel
> updates?
> Is there any Linux distribution doing so?
>
> Only if we can answer these questions with yes, adding FIT_BOOT_CMD to
> distro-boot would make sense to me.
>

U-Boot is the main boot loader for ARM. I wish that distros would just
support it (adding what is needed) instead of trying to pretend that
U-Boot is a variant of UEFI :-)

I think having a default secure boot option is useful, but how does
the image get signed?
Regards,
Simon