[qcom-next] efi_loader: capsule: report spec-compliant ImageIndex for dynamic fw_images[] by balajiselvanathan · Pull Request #131 · qualcomm-linux/u-boot

balajiselvanathan · 2026-08-28T08:50:06Z

Platforms that build their fw_images[] array at runtime from the partition
table (Qualcomm Snapdragon) assign each descriptor a fixed, per-component
image_index. A board only carries a subset of components, so those indices
can fall outside 1 <= ImageIndex <= DescriptorCount — the range UEFI 2.9A
§23.1 requires GetImageInfo to report.

This series makes such platforms assign image_index sequentially and resolves
capsule payloads by image type GUID instead of by index, so the reported
descriptor array and the ESRT stay in range. Platforms with a static 1:1
fw_images[] are unaffected.

This reverts commit c699c6d. With OSIndications now set from Kernel, OsIndications is taken into account during capsule update in U-Boot.

varada-qcom · 2026-08-28T10:59:27Z

+Such platforms set num_images and dfu_alt_num_valid = true. When
+dfu_alt_num_valid is false (the default) behaviour is unchanged: the capsule's
+UpdateImageIndex is used as-is and dfu_alt_num is ignored.
+


Not sure if validated and tested code should be changed to handle snapdragon not aligning with the expectation. If GUID overrides the index for flashing, why cant Qualcomm use an index that is inline with EFI expectation?

Not sure if validated and tested code should be changed to handle snapdragon not aligning with the expectation. If GUID overrides the index for flashing, why cant Qualcomm use an index that is inline with EFI expectation?

Hi, index varies across boards, depending the on the partitions present on a given board.
Hence, we cant use index for identifying dfu string (and so we relyon GUID).
Also, testing for capsule update is yet to start, hence, doing the change now, so testing can validate these changes also.

varada-qcom · 2026-08-31T09:13:25Z


 	struct efi_capsule_update_info {
 		const char *dfu_string;
+		int num_images;


varada-qcom · 2026-08-31T09:15:25Z

+/**
+ * efi_capsule_resolve_image_index - map a capsule to its platform image index
+ * @image_type:		Image type GUID from the capsule (UpdateImageTypeId)
+ * @capsule_index:	Image index from the capsule (UpdateImageIndex)
+ *
+ * Return the image_index to use for efi_fmp_find() and set_image(). The weak
+ * default trusts the capsule's own UpdateImageIndex, so behaviour is unchanged
+ * for platforms with a static 1:1 fw_images[]. A platform whose fw_images[] is
+ * built dynamically (so the capsule's UpdateImageIndex cannot be relied on)
+ * overrides this to resolve @image_type to its own image_index.
+ *
+ * Return: the image_index to use for efi_fmp_find() and set_image()
+ */


The info in the include/efi_loader.h for the prototype is better. This can be dropped.

varada-qcom · 2026-08-31T09:19:27Z

+/**
+ * efi_capsule_resolve_image_index - map a capsule to its platform image index
+ * @image_type:		Image type GUID from the capsule (UpdateImageTypeId)
+ * @capsule_index:	Image index from the capsule (UpdateImageIndex)
+ *
+ * Return the image_index to use for efi_fmp_find() and set_image(). The weak
+ * default trusts the capsule's own UpdateImageIndex, so behaviour is unchanged
+ * for platforms with a static 1:1 fw_images[]. A platform whose fw_images[] is
+ * built dynamically (so the capsule's UpdateImageIndex cannot be relied on)
+ * overrides this to resolve @image_type to its own image_index.
+ *
+ * Return: the image_index to use for efi_fmp_find() and set_image()
+ */


Drop. Comments in the header is enough and it is better

varada-qcom · 2026-08-31T09:19:45Z


 	struct efi_capsule_update_info {
 		const char *dfu_string;
+		int num_images;


Platforms that build their fw_images[] array dynamically from the partition table (e.g. Qualcomm Snapdragon) assign each descriptor a fixed, per-component image_index. Because a given board only carries a subset of components, those values can fall outside the range 1 <= ImageIndex <= DescriptorCount that UEFI 2.9A section 23.1 requires GetImageInfo to report. Such a platform must instead assign image_index sequentially (1..num_images), after which the capsule's UpdateImageIndex can no longer be trusted to select the right image. Add a __weak efi_capsule_resolve_image_index() that maps a capsule to the image_index used for efi_fmp_find() and set_image(). The default returns the capsule's UpdateImageIndex unchanged, so platforms with a static 1:1 fw_images[] are byte-for-byte unchanged. A platform with a dynamically built fw_images[] overrides it to resolve the payload by image_type_id (GUID) to its own sequential image_index; the generic code carries no platform-specific policy. Signed-off-by: Balaji Selvanathan <balaji.selvanathan@oss.qualcomm.com>

Build on the generic weak hook and make Snapdragon's dynamically built fw_images[] spec-compliant: image_index is assigned sequentially (i + 1) in qcom_build_fw_images(), so GetImageInfo reports ImageIndex values in 1..DescriptorCount as required by UEFI 2.9A section 23.1. Once image_index is sequential it depends on which components a board carries, so it can no longer match the capsule's frozen UpdateImageIndex. Override efi_capsule_resolve_image_index() to map an incoming payload to its image_index by image_type_id (GUID); the generic capsule flow uses that resolved index for both efi_fmp_find() and set_image(). No flag on the shared struct is needed - platforms that don't override keep the weak default and are unaffected. Signed-off-by: Balaji Selvanathan <balaji.selvanathan@oss.qualcomm.com>

Add the multiimgqti component (partition multiimgqti_a/_b, image type GUID 846c6f05-eb46-4c0a-a1a3-3648ef3f9d0e) to qcom_image_map[] so boards that carry it can capsule-update it. It is appended with a new authoring index (17) so the frozen indices of already-authored capsules are preserved. Drop the apdp component from the map. Removing a row leaves a gap in the authoring-index sequence (15); this is harmless because the index is only compared by value (rule-1 dedup) and is never used as an array subscript, and the runtime fw_images[].image_index is assigned sequentially regardless. Signed-off-by: Balaji Selvanathan <balaji.selvanathan@oss.qualcomm.com>

Revert "configs: qcom: ignore OsIndications for capsule-on-disk"

fb4253d

This reverts commit c699c6d. With OSIndications now set from Kernel, OsIndications is taken into account during capsule update in U-Boot.

varada-qcom reviewed Aug 28, 2026

View reviewed changes

varada-qcom requested changes Aug 28, 2026

View reviewed changes

balajiselvanathan force-pushed the capsule-seq-image-index branch from b484d4a to 347d6a9 Compare August 31, 2026 08:42

varada-qcom reviewed Aug 31, 2026

View reviewed changes

Comment thread doc/develop/uefi/uefi.rst Outdated

varada-qcom reviewed Aug 31, 2026

View reviewed changes

varada-qcom requested changes Aug 31, 2026

View reviewed changes

balajiselvanathan added 3 commits August 31, 2026 15:23

balajiselvanathan force-pushed the capsule-seq-image-index branch from 347d6a9 to bc25e9b Compare August 31, 2026 10:04

varada-qcom approved these changes Aug 31, 2026

View reviewed changes

Navigation Menu

Uh oh!

[qcom-next] efi_loader: capsule: report spec-compliant ImageIndex for dynamic fw_images[] - #131

[qcom-next] efi_loader: capsule: report spec-compliant ImageIndex for dynamic fw_images[]#131
balajiselvanathan wants to merge 4 commits into
qualcomm-linux:qcom-nextfrom
balajiselvanathan:capsule-seq-image-index

balajiselvanathan commented Aug 28, 2026

Uh oh!

varada-qcom Aug 28, 2026

Uh oh!

balajiselvanathan Aug 31, 2026

Uh oh!

Uh oh!

varada-qcom Aug 31, 2026

Uh oh!

balajiselvanathan Aug 31, 2026

Uh oh!

varada-qcom Aug 31, 2026

Uh oh!

balajiselvanathan Aug 31, 2026

Uh oh!

varada-qcom Aug 31, 2026

Uh oh!

balajiselvanathan Aug 31, 2026

Uh oh!

varada-qcom Aug 31, 2026

Uh oh!

balajiselvanathan Aug 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

balajiselvanathan commented Aug 28, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants