Hardware
Every subsystem against stock L4T R21.x — and exactly where mainline stands.
SMP boot, 4× Cortex-A15
LIVEQuad Cortex-A15 @ 2.2 GHz + LP shadow core; clean SMP bring-up, re-verified on Kit A.
CPU DVFS (cpufreq)
LIVE22 OPP entries, 204 MHz–2.32 GHz, ondemand governor — synthesized from the hardcoded CVB table, not a DT OPP list (a DT table would collide and corrupt cpufreq on this SoC family).
Display / HDMI
LIVEtegra-drm KMS drives HDMI 1.4a at native panel-free resolution.
Audio — analog + HDMI
LIVERT5640 codec on i2c-0 @ 0x1c; both the 3.5mm jack and HDMI audio paths landed with the r33 BCLK-PLL fix.
USB 2.0 + 3.0
LIVEMicro-B USB2 gadget (ci_hdrc) and USB3 host (tegra-xusb) both enumerate.
PCIe + Gigabit Ethernet
LIVEOnboard Realtek RTL8111GS behind PCIe via R8169; eth0-dhcp brings the board up headless — the devkit's WiFi-free access path.
microSD + eMMC
LIVEBoth boot paths verified. Rootfs is always pinned by UUID — mmcblk numbering is non-deterministic across boots.
Thermal sensing
LIVEFour zones — cpu / gpu / pll / mem — all report; idle around 29°C.
Watchdog
LIVEKernel watchdog armed, re-verified on hardware.
zram swap
LIVEA mainline-only improvement — L4T 3.x shipped no compressed swap at all.
GPU acceleration (GK20A)
THE WALLnouveau is built =m and blacklisted: modprobe hard-hangs the GK20A probe outright — a kernel-side wedge, not a missing-userspace gap. All rendering today is llvmpipe software (glmark2 Score 17). The fleet's shared Kepler wall — root-caused here first because this board has a real console.
EMC memory DVFS
PORTINGtegra30_devfreq is blacklisted — tegra_actmon_resume NULL-derefs during udev coldplug on r24+ kernels. Low functional impact (EMC pins at boot clock) but it's now a cross-device boot blocker: Shield Tablet already carries the same blacklist.
Suspend / resume (SC7)
PORTINGA kworker hangs on suspend entry — the same upstream Tegra K1 issue tracked as pmaports #53 against the Nexus 9.
AS3722 PMIC interrupt
PORTINGIRQ 84 storms roughly 200k unhandled firings at boot; the kernel auto-disables the line around t≈60s. RTC-alarm wake and PMIC fault reporting are dead as a result — the power button is unaffected (separate GPIO IRQ 132).
Software poweroff
PORTINGNo AS3722 power-off handler is wired up. poweroff halts the CPU but the rails stay hot — the only way down is pulling the 12V barrel.
Fan PWM control
VERIFYFour thermal zones and cooling_device0 are registered; the actual PWM response to the cooling map under sustained load hasn't been driven yet.
SATA
VERIFYThe Tegra AHCI driver is present in mainline; no drive has been connected to the bench to confirm enumeration.
Beyond the vendor L4T
What a living mainline kernel gets over a frozen 2014 Ubuntu/3.10 vendor BSP — plus the one thing only this board does for the fleet.
A living kernel
Mainline 6.18.29-grate where L4T froze at Ubuntu 14.04 / kernel 3.10 in 2014 — over a decade of upstream security and driver work this board keeps for free.
The fleet's GK20A reference
The probe hang blocking GPU accel here (see Live sprint) is the exact same Kepler wall sitting in front of the Nexus 9, Shield Tablet, and Project Tango — three battery-powered devices sharing this GPU.
Cross-device fixes already shipped
The tegra30_devfreq actmon NULL-deref root-caused on this board is already carried as the identical blacklist on the Shield Tablet.
Felt parity: N/A by design
Track R has no touch panel, no battery, and no gesture shell to feel snappy or vibrate on commit — the fleet-wide felt-parity baseline (docs/FLEET_TRACKS.md) applies to Tracks 1–4 only.
Road to RO-OS
The fleet-wide phase ladder (P0 boots → P7 full parity). This board's phase reads P4 — but Track R doesn't climb this ladder for itself.
Live sprint
What's on the bench right now.
GK20A probe hang — the shared Kepler wall
Boot, SMP, CPU DVFS, display, audio, USB, PCIe/GbE, storage — all re-confirmed on real hardware at r36. Ten of seventeen rows on the parity checklist are closed.
tegra_actmon_resume NULL-derefs on udev coldplug — root-caused here first. The same blacklist-actmon.conf workaround is already carried on the Shield Tablet (device-nvidia-shieldtablet-grate, commit 361fdf9): the reference role already paid for itself once, before the GPU even lands.
modprobe nouveau hard-hangs the board outright — a kernel-side probe wedge, not a missing-firmware gap. UART + drm.debug=0x1e capture is in progress to pin the wedge to GPU power-domain ungate, GK20A clock setup, or GR-falcon FECS/GPCCS ucode load.
Once the probe succeeds: bring up mesa's nouveau nvc0 driver, confirm renderD128 is a real render node, and re-run glmark2 off the Score 17 / llvmpipe baseline. That delta is the exact fix that promotes N9, Shield Tablet, and Tango out of software rendering too.