Fix vision CPU throttling with offline cores

starpilot/common/cpu_throttle.py

@@ -1,6 +1,8 @@
 from __future__ import annotations
 
+from functools import lru_cache
 import math
+from pathlib import Path
 import time
 
 
@@ -11,9 +13,34 @@ DEVICE_BUSY_HOT_CORE_COUNT = 4
 _throttle_state: dict[str, dict[str, float]] = {}
 
 
+@lru_cache(maxsize=1)
+def _online_cpu_count() -> int | None:
+  try:
+    spec = Path("/sys/devices/system/cpu/online").read_text(encoding="utf-8").strip()
+    count = 0
+    for group in spec.split(","):
+      bounds = group.split("-", maxsplit=1)
+      first = int(bounds[0])
+      last = int(bounds[-1])
+      if last < first:
+        return None
+      count += last - first + 1
+    return count or None
+  except (OSError, ValueError):
+    return None
+
+
+def _online_cpu_usage(cpu_usage) -> list[float]:
+  usage = [float(value) for value in cpu_usage]
+  online_count = _online_cpu_count()
+  if online_count is not None and online_count < len(usage):
+    return usage[:online_count]
+  return usage
+
+
 def device_cpu_throttle_factor(cpu_usage, name="vision"):
   """Return a process-local, low-pass-filtered CPU throttle factor."""
-  usage = list(cpu_usage)
+  usage = _online_cpu_usage(cpu_usage)
   if not usage:
     return 1.0
 
@@ -34,7 +61,7 @@ def device_cpu_throttle_factor(cpu_usage, name="vision"):
   state["factor"] = min(max(target_factor * alpha + state["factor"] * (1.0 - alpha), 1.0), 4.0)
 
   if state["factor"] >= state["last_logged_factor"] + 0.6:
-    print(f"[{name}] CPU throttle factor={state['factor']:.1f}x (avg={average:.0f}%, hot={hot_cores})")
+    print(f"[{name}] CPU throttle factor={state['factor']:.1f}x (avg={average:.0f}%, hot={hot_cores}/{len(usage)})")
     state["last_logged_factor"] = state["factor"]
   elif state["factor"] <= 1.05 and state["last_logged_factor"] > 1.05:
     print(f"[{name}] CPU recovered")

starpilot/common/tests/test_cpu_throttle.py

@@ -0,0 +1,26 @@
+import pytest
+
+from openpilot.starpilot.common import cpu_throttle
+
+
+def test_offline_cpu_placeholders_do_not_hide_sustained_load(monkeypatch):
+  monkeypatch.setattr(cpu_throttle, "_online_cpu_count", lambda: 4)
+
+  usage = cpu_throttle._online_cpu_usage([80, 80, 80, 80, 0, 0, 0, 0])
+  average = sum(usage) / len(usage)
+
+  assert usage == [80.0, 80.0, 80.0, 80.0]
+  assert cpu_throttle._compute_throttle_factor(average, 0) == pytest.approx(1.75)
+
+
+@pytest.mark.parametrize(("online_spec", "expected"), (
+  ("0-3", 4),
+  ("0-3,6-7", 6),
+))
+def test_online_cpu_count_parser(monkeypatch, tmp_path, online_spec, expected):
+  online_path = tmp_path / "online"
+  online_path.write_text(online_spec, encoding="utf-8")
+  monkeypatch.setattr(cpu_throttle, "Path", lambda _: online_path)
+  cpu_throttle._online_cpu_count.cache_clear()
+
+  assert cpu_throttle._online_cpu_count() == expected