BPI-R64 hardpps soak — patches 1-4 only (old pps_get_ts), tickless, 1Hz pulses

Cortex-A53 (BPI-R64), 12.5 MHz arch counter, u-blox GNSS timepulse into pps-gpio, in-kernel hardpps() disciplining CLOCK_REALTIME with STA_PPSTIME|STA_PPSFREQ|STA_PLL. Kernel: ONLY patches 1-4 of the posted series (rodolfo-14-test branch) on 7.3-rc1 — the old pps_get_ts() path, no snapshot ntp_error correction — plus the NTP_PPS Kconfig change to allow tickless — plus silent bench instrumentation (per-pulse ring captured in the PPS hardirq, drained to disk once a minute in the pulse's shadow). No distro userspace: a single static PID 1 sets the date from NMEA, binds hardpps, and sleeps. All periodic wakeup sources were disabled (vmstat, runtime-PM autosuspend, fair dl_server, ethernet not probed). Tickless cells boot NO_HZ_IDLE; tickful cells add nohz=off on the same kernel binary.

Per-pulse snapshot ntp_error

Pulse arrival phase (corrected ts_real)

Discipline state

Disciplined frequency vs room temperature

Room air temperature (sensor near the board, not on it) against the PLL-disciplined frequency. Context only: the crystal's own temperature is not measured.

Methodology

This run is the test asked for on the list: hardpps on a tickless kernel driven by a real PPS source, with its real jitter and IRQ latency, and the CPU genuinely idle between pulses.

• Hardware: Banana Pi R64 (MT7622, 2×Cortex-A53), 12.5 MHz arch counter, u-blox GNSS HAT. Timepulse period set per cell via UBX-CFG-TP5 (GNSS-locked, aligned to top of second). PPS on a GPIO via pps-gpio; timestamp taken in the hardirq handler.
• Kernel: the timekeeping series on 7.3-rc1 (unmodified STA_PPSTIME delivery). While a phase slew or ntp_error drain is active, the timekeeping CPU's idle sleep is bounded to the current second so the rate is recomputed on schedule — the fix that makes multi-second pulse intervals stable on a tickless kernel, part of the series. Tickful cells run the same binary with nohz=off.
• Quiescence: no distro userspace — a single static PID 1 binary sleeps between once-a-minute dumps. All periodic wakeup sources found by tracing were disabled: vmstat interval raised, runtime-PM autosuspend clamped, per-CPU fair dl_server off, ethernet (mtk_eth_soc + mt7530 1 s pollers) not probed. The per-minute log dump and its mmc/console I/O are issued immediately after a pulse edge so they never perturb a timestamp. The gap histogram (above) verifies the result per cell: on tickless cells the timekeeper advances in multi-second bursts bounded only by the pulse and the per-second slew wakeups; on tickful cells it ticks at HZ=250 throughout.
• Instrumentation: at each pulse the hardirq records snapshot_ntp_error() — the divergence between the ideal NTP-disciplined time and the sanitized clock_gettime() line that the series corrects in pps_get_ts() — plus the corrected ts_real phase against the second boundary. The kernel discipline state (hardpps corrections, second_overflow decisions, every mult step) is logged via printk to the kmsg ring and drained to disk per dump.
• Caveats: the ∼30 µs correction stripe in the slew plot is bench self-noise (a recurring IRQ-latency event, one-sided: latency can only make a pulse look late), and this kernel still carries the trace printks; a quiet-kernel confirmation run and the tickful (nohz=off) A/B comparison on identical hardware are the follow-ups.

Data

• pulses.csv — per-pulse snapshot ntp_error, tk ledger, cycle_delta, corrected ts_real phase, err_drain
• adjtimex.csv — PLL state, 1/min
• hist.csv — per-minute log2 histogram of timekeeping-advance gap sizes + max
• sec_rows.csv — per-second ledger (time_offset, skew, ntp_error, err_drain)
• drain_counters.csv — drain engagements / disengagements / overshoot drops, 1/min
• bench.log.gz — the raw on-board log