Skip to content

OCP Recovery streaming boot: open design questions #705

Open

Description

@chrysh

Design: #680. Related: #48

#680 proposes OCP Recovery streaming boot for the eRoT. The eRoT serves the OCP Recovery interface as the Device, and the BMC drives the interface as the Recovery Agent. The eRoT verifies the streamed image, resets, verifies the image again at boot and boots the image. #680 leaves open where the streamed image lands (question 1). The transport binding is out of scope in #680, and #744 collects the questions on I3C read timing, which include how the BMC repeats an I3C read the eRoT NACKs.

@FerralCoder, can you describe scenarios for when OCP Recovery starts on the eRoT and how the platform then uses OCP Recovery? Several concrete cases help the design more than a single case.

The questions below need an answer before implementation starts. Please reply by number.

  1. Where does the eRoT put the streamed image: in RAM that keeps its contents across a reset, or in a flash slot? With RAM, the eRoT writes no flash, no trial boot runs, the SVN floor keeps the SVN it has, and a power cycle clears the image. With flash, the image survives a power cycle, and a trial boot and an SVN floor advance on confirm apply as they do for an update. A flash slot makes OCP Recovery a second install path next to PLDM firmware update.
  2. Must streaming boot cover the case where the operational image does not run at all, without physical access? In docs: Add the OCP Recovery streaming boot design proposal #680, when the ROM rejects a corrupt image, a power loss interrupts an update, or every slot holds an image below the SVN floor, a technician has to set the hardware strap. A yes means a second signed image: a recovery copy in a write-protected flash region, which the hardware boots as a fallback.
  3. If question 2 is a yes: is the recovery copy a full firmware with the degraded OCP mode, a minimal recovery-only image, or does each board choose?
  4. If question 2 is a yes: does the default OpenPRoT build produce the recovery copy, or a separate build?
  5. Who may rewrite the recovery copy, and with what authorization? Rewriting the recovery copy lifts the flash write protection.
  6. How does the BMC reset the eRoT when the streamed image that booted stops serving the interface? DEVICE_RESET needs a serving image, so the board needs a BMC-driven pin that fires the boot fallback or a reset line to the eRoT. Which does the board provide, the pin or the reset line?
  7. Where does streaming boot sit in the boot order? docs: Add the OCP Recovery streaming boot design proposal #680 puts streaming boot below every local slot and the golden partition and above the hardware strap. A board that trusts the BMC more than a stale golden copy would put streaming boot above the golden partition. Some platforms put streaming boot first, as the primary boot path. Which order is the default, and which failure should the design name as the case where streaming boot runs?
  8. May a platform that has latched LockdownLatch serve the interface? The latch is one-way, so a streamed image that runs after the platform latched undoes the safe state.
  9. Does the platform need forced recovery, where the BMC writes the forced-recovery value so a healthy eRoT enters recovery mode at the next reset? docs: Add the OCP Recovery streaming boot design proposal #680 leaves forced recovery out of scope.
  10. Does the platform need multi-stage activation, where the Recovery Agent streams several stages over OCP Recovery? docs: Add the OCP Recovery streaming boot design proposal #680 streams a single image and boots that image.
  11. Does the platform boot in 2 steps, a small first image over OCP Recovery and the rest over PLDM firmware update, or from the OCP Recovery image alone?
  12. Which push mechanism does the eRoT serve? In docs: Add the OCP Recovery streaming boot design proposal #680 the eRoT serves the FIFO (0x2d to 0x2f) and refuses the indirect memory window (0x29 to 0x2b). The OpenBMC Recovery Agent proposed under the ocp-fw-recovery topic on gerrit.openbmc.org pushes through the memory window and refuses a Device that advertises only the FIFO. Either the eRoT serves the memory window as well, or the OpenBMC Recovery Agent adds FIFO support.
  13. What maximum boot time must a streamed image meet? The maximum boot time decides which transports and image sizes work.
  14. For how long, and with how many retries, does the BMC poll RECOVERY_STATUS after the BMC activates the streamed image? The verify before the reset, the reset, the verify at boot and the boot have to finish inside the time the BMC keeps polling RECOVERY_STATUS, and BMC reads fail during the reset.
  15. Which reason code does the eRoT report on entering recovery? docs: Add the OCP Recovery streaming boot design proposal #680 reports FlashlessStreamingBoot (0x12), which OCP Secure Firmware Recovery 1.1 lists as Flashless/Streaming Boot. Section 7.3 of the spec describes flashless boot as a mode the Recovery Agent commands with a write to the RESET register. An eRoT whose local images failed enters recovery without that write and still holds local images, so docs: Add the OCP Recovery streaming boot design proposal #680 leaves open which reason code fits that eRoT.

cc @FerralCoder @rusty1968

Activity

  1. rusty1968 commented on Nov 15, 2026

    @rusty1968
    Collaborator
  2. chrysh commented on Nov 18, 2026

    @chrysh
    ContributorAuthor

    I updated the description to match the current design in #680. Question 1 is now about where the streamed image lands, in RAM that keeps its contents across a reset or in a flash slot, because #680 no longer fixes a boot from RAM. Questions 12 to 15 are new: the push mechanism, the maximum boot time, the RECOVERY_STATUS poll and the reason code the eRoT reports on entering recovery. Questions 6 and 7 have small wording changes. Questions 2 to 11 keep their numbers. The questions on I3C read timing are in #744.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions