T1: 8814AU path-C/D mirrors in canary mask sets

tests/canary_diff.py

@@ -36,11 +36,14 @@
     chip temperature so each capture shows a slightly different
     value. The thermal value is also the input to phydm's TX
     BB-swing tracking (see BB 0xc1c[31:21] below).
-  - BB 0xc1c bits 31:21 / 0xe1c bits 31:21: TX BB-swing
+  - BB 0xc1c / 0xe1c / 0x181c / 0x1a1c bits 31:21: TX BB-swing
     `tx_scaling_table_jaguar` index, written by `PowerTracking8812a`
     (and the kernel's phydm watchdog) based on the thermal-meter
-    sample. Same drift class as RF[A] 0x42. Other bits of 0xc1c
-    (AGC table select [11:8], static base bits) ARE checked.
+    sample. Same drift class as RF[A] 0x42. Devourer's pwrtrk is
+    gated to CHIP_8812 only, so on 8814 only the kernel walks
+    path C/D — masking is still correct for the diff. Other bits
+    of these registers (AGC table select [11:8], static base
+    bits) ARE checked.
   - BB 0xc50, 0xe50, 0x1850, 0x1a50 bits 7:0: DIG IGI (path
     A/B/C/D Initial Gain). The kernel's phydm DIG watchdog walks
     the IGI value up and down each interrupt cycle based on
@@ -56,10 +59,11 @@
 There's also a known capture-state asymmetry: the kernel iface is
 long-lived (CCK regs at 5G retain values written during prior 2.4G
 activity), while devourer captures from a fresh process per run
-(BB-init defaults). At 5G channels we therefore skip `BB 0xc20`
-(rTxAGC_A_CCK11_CCK1_JAguar) because it's CCK-only — never written
-at 5G by either side, but reflects different histories. Add more
-to `CAPTURE_STATE_5G_ARTIFACTS` if new ones surface.
+(BB-init defaults). At 5G channels we therefore skip the per-path
+CCK TX-AGC registers — `rTxAGC_*_CCK11_CCK1_*` for paths A/B (8812 /
+8821) and paths C/D (8814) — they're CCK-only, never written at
+5G by either side, but reflect different histories. Add more to
+`CAPTURE_STATE_5G_ARTIFACTS` if new ones surface.
 """
 
 from __future__ import annotations
@@ -82,9 +86,14 @@
     ("RF[A]", 0x42): 0xFFFFFFFF,
     ("RF[B]", 0x42): 0xFFFFFFFF,
     # BB TX-swing thermal pwrtrk — only bits 31:21 (the
-    # tx_scaling_table_jaguar index) are thermal-tracked.
-    ("BB", 0xc1c): 0xFFE00000,
-    ("BB", 0xe1c): 0xFFE00000,
+    # tx_scaling_table_jaguar index) are thermal-tracked. Devourer's
+    # PowerTracking8812a is gated to CHIP_8812 only, so on 8814 the
+    # kernel walks the path-C/D swing while devourer leaves it static
+    # — same divergence class, same masking applies.
+    ("BB", 0xc1c): 0xFFE00000,    # path A
+    ("BB", 0xe1c): 0xFFE00000,    # path B
+    ("BB", 0x181c): 0xFFE00000,   # path C (8814 only)
+    ("BB", 0x1a1c): 0xFFE00000,   # path D (8814 only)
     # DIG IGI (bits 7:0) — kernel's phydm DIG watchdog continuously
     # walks the path-IGI based on RX noise floor; devourer writes
     # the 0x1c floor once and doesn't update.
@@ -112,8 +121,10 @@
 # iface, while devourer captures from a fresh process. Skip
 # entirely when --channel > 14.
 CAPTURE_STATE_5G_ARTIFACTS: set[tuple[str, int]] = {
-    ("BB", 0xc20),  # rTxAGC_A_CCK11_CCK1_JAguar
-    ("BB", 0xe20),  # path-B mirror
+    ("BB", 0xc20),    # rTxAGC_A_CCK11_CCK1_JAguar      (path A)
+    ("BB", 0xe20),    # rTxAGC_B_CCK11_CCK1_JAguar      (path B)
+    ("BB", 0x1820),   # rTxAGC_C_CCK11_CCK1_Jaguar2     (path C, 8814)
+    ("BB", 0x1a20),   # rTxAGC_D_CCK11_CCK1_Jaguar2     (path D, 8814)
 }
 
 LINE_RE = re.compile(

tests/test_canary_diff.py

@@ -157,6 +157,35 @@ def test_5g_capture_state_artifact_masked(tmp_path: Path) -> None:
     assert "0xc20" in res.stdout
 
 
+def test_5g_capture_state_artifact_8814_path_c_d_masked(tmp_path: Path) -> None:
+    """8814 path-C/D CCK TX-AGC mirrors (0x1820 / 0x1a20) — same
+    asymmetry as path-A/B, should also be masked at 5G."""
+    body_k = "BB 0x1820 = 0x2A2A2A2A\nBB 0x1a20 = 0x3B3B3B3B"
+    body_d = "BB 0x1820 = 0x00000000\nBB 0x1a20 = 0x00000000"
+    res = run_diff(wrap(body_k), wrap(body_d),
+                   "--channel", "100", tmp_path=tmp_path)
+    assert res.returncode == 0, res.stdout + res.stderr
+    assert "0x1820" in res.stdout
+    assert "0x1a20" in res.stdout
+
+
+def test_8814_pathcd_bb_swing_thermal_masked(tmp_path: Path) -> None:
+    """8814 path-C/D TX scaling (0x181c / 0x1a1c) bits 31:21 are
+    the same pwrtrk-tracked field as path-A/B 0xc1c / 0xe1c."""
+    kernel = wrap("BB 0x181c = 0x47C00003\nBB 0x1a1c = 0x47C00003")
+    devourer = wrap("BB 0x181c = 0x40000003\nBB 0x1a1c = 0x40000003")
+    res = run_diff(kernel, devourer, tmp_path=tmp_path)
+    assert res.returncode == 0, res.stdout + res.stderr
+
+
+def test_8814_pathcd_bb_swing_lower_bits_still_diffed(tmp_path: Path) -> None:
+    """Lower bits of 0x181c / 0x1a1c are checked."""
+    kernel = wrap("BB 0x181c = 0x47C00003")
+    devourer = wrap("BB 0x181c = 0x47C00103")  # bit 8 differs
+    res = run_diff(kernel, devourer, tmp_path=tmp_path)
+    assert res.returncode == 1, res.stdout + res.stderr
+
+
 def test_5g_capture_state_artifact_NOT_masked_at_2g(tmp_path: Path) -> None:
     """Same divergence at ch6 is treated as a real difference (CCK is
     active at 2.4G; if it diverges there it's a real init drift)."""