nvidia: Thor provisioning leaves RCM, fTPM node, upstream flash helper, Jetson getty mask - #317
Merged
Merged
Conversation
adbd64 in NVIDIA's T264 flashing initramfs implements 'adb reboot' by
setting the Android sys.powerctl property, which nothing serves there, so
it always fails ('reboot (reboot,adb) failed') and the board sits in the
initramfs after a successful flash. Run 'reboot -f' over adb shell instead,
which issues the reboot syscall directly.
…sh-helper.sh Our tegra264-flash-helper.sh was a wholesale snapshot of an R38-era meta-tegra split helper (added in 609512e, never edited). Diffing it against the pinned meta-tegra's unified tegra-flash-helper.sh shows no avocado logic in it -- every difference is upstream being newer: the CHPID typo, the RAMCODE-from-board fallback when RAMCODE is unset, get_active_key for T26x PKC keylists, --cpubl_rcm, the guarded $odmdata_arg in the RCM flashcmd (ours still passed a raw --odmdata $ODMDATA, the same empty-value hazard ac23aef fixed on t234), and BCTFILE for the SDRAM config. The --no-flash --rcm-boot path initrd-flash depends on (rcmboot_blob + membct_ramcode_file rewrite + rcmdump_blob.tar) is unchanged. Point FLASH_HELPER_SCRIPT:tegra264 at upstream's helper, stop installing the copy from tegra-helper-scripts and nativesdk-tegraflash-tools, and have the SDK recipe copy tegra-flash-helper.sh from the native sysroot instead. tegraflash-tools-deploy already copies the whole native tegra-flash tree, so the runtime path needs nothing extra. tegra234-flash-helper.sh stays: it carries the ODMDATA env-precedence fix upstream does not have.
Our machine conf carried the Orin-style value ('--minratchet_config <dts>').
meta-tegra's Thor includes define it as the bare .dts, and the unified
tegra-flash-helper.sh (now used on Thor) passes it unquoted as
brbct_cfg/minratchet=$MINRATCHET_CONFIG, so the value split into two args and
nvbct-config died on the flag-less second one:
ValueError: not enough values to unpack (expected 2, got 1)
ERR: could not update BCT configuration
Our former tegra264-flash-helper.sh snapshot never passed minratchet to
nvbct-config, which is why this stayed hidden. Every other TEGRA_FLASHVAR_*
in the conf matches upstream agx-thor-t5000.inc.
…node On Orin the firmware/ftpm node comes from tegra-optee.dtbo via TEGRA_PLUGIN_MANAGER_OVERLAYS. On Thor L4T ships it in tegra264-ftpm.dtbo and applies it at flash time via OVERLAY_DTB_FILE (t264.conf.common), which meta-tegra does not carry for tegra264. Result on an AGX Thor devkit with var.encrypt: the TOS has the fTPM TA and UEFI measures into it, the initramfs ships tpm_ftpm_tee.ko, but /proc/device-tree has no ftpm node, the tee bus only enumerates the RNG TA, no /dev/tpm0 appears, and cryptsetup-var opens /var with the Argon2id recovery key instead of a PCR-7 sealed keyslot.
It was only in MACHINE_ESSENTIAL_EXTRA_RDEPENDS, which packagegroup-avocado-rootfs.bb intentionally does not expand, so every avocado-built Jetson rootfs still boots degraded with getty@getty.service.
Two pieces, both in this layer (no vendor-layer patches):
- tegra-target-flash-scripts_%.bbappend: meta-tegra's sed on NVIDIA's /init
drops the /sbin/reboot wrapper ('busybox reboot -f') together with the mount
block. PID 1 there is /bin/bash, so bare busybox reboot -- which only signals
init -- is a silent no-op from init's error/timeout paths, the debug prompt
and 'adb shell reboot'. Reinstate it as an update-alternative above
busybox's priority.
- initrd-flash.sh: after 'adb shell reboot -f', T264 keeps the RCM boot mode
across the warm reset and comes back as the boot-ROM APX device (0955:7026)
rather than cold-booting the flashed system (observed twice on an AGX Thor
devkit). NVIDIA's bootburn carries Mb2AppletReset() for exactly this but
never calls it. Wait up to 60 s for the device to re-enter RCM with
find-jetson-usb --wait (it matches only RCM PIDs, not the adb gadget) and
issue 'tegrarcm_v2 --chip 0x26 0 --reboot coldboot'; if it never reappears
it booted on its own.
First run on the AGX Thor devkit confirmed the device re-enters RCM after 'adb shell reboot -f' ([found]) but the kick failed: File rcm_state open failed ERROR: failed to read rcm_state tegrarcm_v2 keeps its RCM session in an rcm_state file in the cwd, written by --new_session; NVIDIA's generated rcmbootcmd.txt only ever issues '--reboot recovery' after such a session. Do the same: --new_session --uid in a scratch dir, then --reboot coldboot.
…oardctl Tested on the AGX Thor devkit: after 'adb shell reboot -f' the SoC returns as the boot-ROM APX device, and 'tegrarcm_v2 --new_session --uid' + '--reboot coldboot' is accepted (BR_CID printed) but leaves it in RCM; the reset button is what boots the flashed system. Drop the tegrarcm kick and press SYS_RESET through the devkit's TOPO debug MCU instead: 'boardctl -t $BOARDCTL_TARGET [-s $BOARDCTL_SERIAL] reset', defaulting the target to thor-jetson-devkit when a 0955:7045 Operator is on the bus and CHIPID is 0x26. Without a TOPO, print ACTION NEEDED and ask for the button. nativesdk-boardctl: boardctl was unusable in the SDK -- it scans _targets/ next to itself (FileNotFoundError), and the symlink made os.path.abspath(__file__) resolve there. Install _targets/ and __init__.py and replace the symlink with a wrapper.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes found while bringing up encrypted-/var (#310) on a Jetson AGX Thor devkit; verified end-to-end on Thor (T264, L4T 6.8) and re-verified on AGX Orin (T234, linux-yocto 6.18).
Thor: provisioning now ends with the board booting the flashed system
adb shell reboot -finstead ofadb reboot— the device-sideadbd64implementsadb rebootvia the Androidsys.powerctlproperty, which nothing serves in the flashing initramfs (reboot (reboot,adb) failed).rebootrestored (tegra-target-flash-scripts_%.bbappend) — meta-tegra's sed on NVIDIA's/initdrops the/sbin/reboot→busybox reboot -fwrapper; with bash as PID 1, bare busyboxrebootis a silent no-op. Reinstated as an update-alternative (priority 100 > busybox 50). Kept in this layer, no vendor patch.boardctl reset— T264 keeps the RCM boot mode across the warm reset, so the SoC re-enumerates as boot-ROM APX (0955:7026) instead of cold-booting.tegrarcm_v2 --new_session … --reboot coldbootis accepted but does nothing (tested); NVIDIA's bootburn never resets either.initrd-flash.shnow waits ≤60 s for the APX device and presses SYS_RESET through the TOPO debug MCU (boardctl -t $BOARDCTL_TARGET reset, defaulting tothor-jetson-devkitwhen a0955:7045Operator is on the bus); without a TOPO it printsACTION NEEDED … press RESET.nativesdk-boardctlwas unusable — never installed_targets/, and the${bindir}symlink madeos.path.abspath(__file__)resolve to${bindir}/_targets(FileNotFoundError). Installs_targets/+__init__.py, ships a wrapper instead of a symlink.Thor: fTPM device node
OVERLAY_DTB_FILE = "tegra264-ftpm.dtbo"— Orin getsfirmware/ftpmfromtegra-optee.dtboviaTEGRA_PLUGIN_MANAGER_OVERLAYS; Thor's L4T ships it integra264-ftpm.dtboviaOVERLAY_DTB_FILE(t264.conf.common), which meta-tegra doesn't carry. Without it the TOS has the TA and UEFI measures into it, the initramfs shipstpm_ftpm_tee.ko, but Linux has no node → no/dev/tpm0and cryptsetup-var falls back to the recovery key. Now:TPM2 PCR-7 keyslot enrolled.Thor: drop our stale
tegra264-flash-helper.shtegra-flash-helper.shwas upstream being newer (CHPID typo, RAMCODE-from-board fallback,get_active_key,--cpubl_rcm, guarded$odmdata_arg).FLASH_HELPER_SCRIPT:tegra264now points at upstream's helper;tegra234-flash-helper.shstays (carries the ODMDATA env-precedence fix upstream lacks).MINRATCHET_CONFIGis a bare filename on tegra264 — our conf had the Orin--minratchet_config <dts>form, which the unified helper splits into two args →nvbct-configValueError: not enough values to unpack. Every otherTEGRA_FLASHVAR_*matches upstreamagx-thor-t5000.inc.All Jetsons
systemd-jetson-masksin the avocado rootfs — it only rodeMACHINE_ESSENTIAL_EXTRA_RDEPENDS, whichpackagegroup-avocado-rootfsdeliberately doesn't expand, so every Jetson still booteddegradedwithgetty@getty.service. Added viaRDEPENDS:${PN}:append:tegraon the existing nvidia bbappend.Verification
[found]→pressing SYS_RESET via boardctl (target=thor-jetson-devkit)→ board boots unattended;systemctl is-system-running=running,/dev/tpm0present,/varLUKS2 withsystemd-tpm2token.running,/dev/tpm0, PCR-7 keyslot enrolled, grow-var completes, no failed units.Companion: avocado-linux/ext-dev#3 (grow-var: resolve the whole disk with
lsblk -d; read sizes from sysfs) — the old grow-var wrote a GPT over the LUKS header on dm-backed/var.