[PATCH] efi: payload: don't fail postcore initcall when image has no device path

[PATCH] efi: payload: don't fail postcore initcall when image has no device path

[Ahmad Fatoum]

When barebox was loaded from memory or from a device without a device
path, e.g. via QEMU's fw_cfg on top of the barebox EFI loader, the loaded
image has no device handle. efi_postcore_init() currently bails out with
-EINVAL in that case, printing an initcall error and skipping the
registration of the barebox update handler.

There is no partition UUID to export as LoaderDevicePartUUID then, but
that's no reason to abort the initcall. The parent image handle may lack a
device path as well, and device_path_to_filepath() would then crash on the
NULL pointer. Make device_path_next_compatible_node() tolerate a NULL
device path, so both helpers just return NULL and the initcall proceeds
with the defaults.

Fixes: e0e30f636242 ("efi: payload: protect against buggy EFI implementations")
Assisted-by: Claude:fable-5
Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
---
 efi/devicepath.c   | 2 +-
 efi/payload/init.c | 7 ++++---
 2 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/efi/devicepath.c b/efi/devicepath.c
index 131c3df16641..32ddc952cd9b 100644
--- a/efi/devicepath.c
+++ b/efi/devicepath.c
@@ -788,7 +788,7 @@ static const struct efi_device_path *
 device_path_next_compatible_node(const struct efi_device_path *dev_path,
 				 u8 type, u8 subtype)
 {
-	for (; !is_device_path_end(dev_path);
+	for (; dev_path && !is_device_path_end(dev_path);
 	     dev_path = next_device_path_node(dev_path)) {
 
 		if (device_path_type(dev_path) != type)
diff --git a/efi/payload/init.c b/efi/payload/init.c
index 862ca6a3e7d6..1e5f2eb8cf56 100644
--- a/efi/payload/init.c
+++ b/efi/payload/init.c
@@ -246,12 +246,13 @@ static int efi_postcore_init(void)
 	efi_set_variable_uint64_le("LoaderFeatures", &efi_systemd_vendor_guid,
 				   loader_features);
 
+	/*
+	 * The image may have been loaded from memory, e.g. QEMU's fw_cfg.
+	 * There is no partition to report then, but the rest should still run.
+	 */
 	loaded_image_dp = device_path_from_handle(efi_loaded_image->device_handle);
 	pr_debug("loaded-image: %pD\n", loaded_image_dp);
 
-	if (!loaded_image_dp)
-		return -EINVAL;
-
 	uuid = device_path_to_partuuid(loaded_image_dp);
 	if (uuid) {
 		wchar_t *uuid16 = xstrdup_char_to_wchar(uuid);
-- 
2.47.3

Re: [PATCH] efi: payload: don't fail postcore initcall when image has no device path

[Sascha Hauer]

On Wed, 26 Aug 2026 14:15:05 +0200, Ahmad Fatoum wrote:
> When barebox was loaded from memory or from a device without a device
> path, e.g. via QEMU's fw_cfg on top of the barebox EFI loader, the loaded
> image has no device handle. efi_postcore_init() currently bails out with
> -EINVAL in that case, printing an initcall error and skipping the
> registration of the barebox update handler.
> 
> There is no partition UUID to export as LoaderDevicePartUUID then, but
> that's no reason to abort the initcall. The parent image handle may lack a
> device path as well, and device_path_to_filepath() would then crash on the
> NULL pointer. Make device_path_next_compatible_node() tolerate a NULL
> device path, so both helpers just return NULL and the initcall proceeds
> with the defaults.
> 
> [...]

Applied, thanks!

[1/1] efi: payload: don't fail postcore initcall when image has no device path
      https://git.pengutronix.de/cgit/barebox/commit/?id=da47299449f3 (link may not be stable)

Best regards,
-- 
Sascha Hauer <s.hauer@pengutronix.de>