The MAX_UNTYPED_OBJECTS configuration value is too small for a newer x86 processor DornerWorks has been using. We've run into the issue in the past on a Qualcomm processor as well. Previously, I was just able to allocate 2 pages for the BootInfo struct, but changes in the seL4 kernel seem to hardcode the BootInfo struct to 1 page.

To get things booting on my processor card, I just created a new entry in the BootInfo enum, SEL4_BOOTINFO_HEADER_EXTRA_UNTYPEDS.

Then I added an entry:

    /* populate extra untypeds block */
    {
        seL4_BootInfoHeader header;
        header.id = SEL4_BOOTINFO_HEADER_EXTRA_UNTYPEDS;
        header.len = sizeof(header) + sizeof(seL4_BootInfoExtraUntypeds);
        *(seL4_BootInfoHeader *)(extra_bi_region.start + extra_bi_offset) = header;
        extra_bi_offset += sizeof(header) + sizeof(seL4_BootInfoExtraUntypeds);
    }

In provide_untyped_cap, if the untyped objects overflow, the object is added to the extra untypeds array. If the extra untypeds header doesn't exist, then the existing behavior of printing a warning and returning is true.

Is this a good solution to the problem? I can't just not provide these resources to userspace, as they're some of the largest memory chunks available.