On 11/28/19 8:20 AM, Heinrich Schuchardt wrote:
On 11/27/19 8:45 PM, Cristian Ciocaltea wrote:
On Tue, Nov 26, 2019 at 07:31:39PM +0100, Heinrich Schuchardt wrote:
On 11/24/19 9:11 PM, Cristian Ciocaltea wrote:
Currently the only way to run an EFI binary like GRUB2 is via the
'bootefi' command, which cannot be used in a verified boot scenario.

The obvious solution to this limitation is to add support for
booting FIT images containing those EFI binaries.

The implementation relies on a new image type - IH_OS_EFI - which
can be created by using 'os = "efi"' inside an ITS file:

/ {
      #address-cells = <1>;

      images {
          efi-grub {
              description = "GRUB EFI";
              data = /incbin/("EFI/BOOT/bootarm.efi");
              type = "kernel_noload";
              arch = "arm";
              os = "efi";
              compression = "none";
              load = <0x0>;
              entry = <0x0>;
              hash-1 {
                  algo = "sha256";
              };
          };
      };

      configurations {
          default = "config-grub";
          config-grub {
              kernel = "efi-grub";
              signature-1 {
                  algo = "sha256,rsa2048";
                  sign-images = "kernel";
              };
          };
      };
};

The bootm command has been extended to handle the IH_OS_EFI images.
To enable this feature, a new configuration option has been added:
BOOTM_EFI

I tested the solution using the 'qemu_arm' board:

=> load scsi 0:1 ${kernel_addr_r} efi-image.fit
=> bootm ${kernel_addr_r}#config-grub

Thanks a lot for the patch series which makes good sense to me.

I think we should pass addresses and not strings to cmd/bootefi.c. This
will need a bit of refactoring as already addressed in a comment to
patch 2/2.

Additionally the documentation in doc/uefi/u-boot_on_efi.rst and
doc/uImage.FIT/howto.txt should be updated.

I cc the contributors given by
scripts/get_maintainer.pl -f common/bootm_os.c

Best regards

Heinrich


Thanks for the feedback, Heinrich!

Instead of creating new function(s), I think we could simply extend
   int do_bootefi_image(const char *image_opt)
with a new parameter to hold the fdt address and move here the call
to 'efi_install_fdt()', which is now performed by 'do_bootefi()'.

efi_install_fdt() has to be called for the 'bootefi bootmgr' command too
so the refactoring is a bit more complicated. I have started on that.

The first step is to change efi_install_fdt() to expect the argument as
address instead of a string.

https://github.com/xypron/u-boot-patches/blob/efi-next/0001-efi_loader-pass-address-to-efi_install_fdt.patch


fdt_addr==NULL indicates no device tree supplied by user.

Best regards

Heinrich


However, I'm not sure about changing the data types, i.e. from
'char *' to ulong, for the following reasons:
1. image_opt may have a different meaning in addition to efi address
2. fdt address may not be provided, so we need somehow to detect an >> invalid 
value

Kind regards,
Cristian


Hello Christian,

patch series
efi_loader: prepare for FIT images
https://lists.denx.de/pipermail/u-boot/2019-December/393192.html
is now available. It offers these functions:

/* Install device tree */
efi_status_t efi_install_fdt(uintptr_t fdt_addr);
/* Run loaded UEFI image */
efi_status_t efi_run_image(void *source_buffer, efi_uintn_t source_size);

Could you, please, rebase your patches on this patch series.

Please, call efi_install_fdt with EFI_FDT_USE_INTERNAL if no device tree
is supplied by the FIT image.

The patch series is also available in branch efi-2020-04 at
https://gitlab.denx.de/u-boot/custodians/u-boot-efi.git

Best regards

Heinrich

Reply via email to