test(hw, adrv9371): flip ensm_mode=radio_on + restore rx-capture check

Root-causes the AD9371 rx-capture gap: the Mykonos driver defaults
the Enable State Machine to ``radio_off`` after probe, so JESD204
RX trains to ``Link status: DATA`` but the ARM's radio FSM is not
delivering samples to the TPL.  ``buf.refill()`` therefore times
out on ``ad_ip_jesd204_tpl_adc`` even though the link looks fine.

ADRV9009 (Talise) doesn't have this problem — either the default
is ``radio_on`` or the post-boot profile load implicitly flips it.
Mykonos needs an explicit write.

Write ``radio_on`` directly to the ``ad9371-phy`` IIO device's
``ensm_mode`` attribute via the labgrid shell (same defensive
pattern the adrv9009 test uses for Talise profile writes — avoids
the libiio TCP socket drop during driver state changes), then
re-verify JESD + dmesg stayed healthy and call
``assert_rx_capture_valid`` for real.

Rules out the alternative approach of runtime ``profile_config``
writes (commits b117949 + 5222208 reverted after proving they
drop JESD and never relock on Mykonos).  Cross-links those commits
from the new code path's docstring so the dead-end is
discoverable.

Author: tfcollins

test/hw/test_adrv9371_zc706_hw.py

@@ -18,6 +18,7 @@
 
 import os
 import shutil
+import time
 from pathlib import Path
 
 import pytest
@@ -35,12 +36,15 @@
 from test.hw.hw_helpers import (  # noqa: E402
     DEFAULT_OUT_DIR,
     acquire_xsa,
+    assert_jesd_links_data,
     assert_no_kernel_faults,
     assert_no_probe_errors,
+    assert_rx_capture_valid,
     collect_dmesg,
     compile_dts_to_dtb,
     deploy_and_boot,
     open_iio_context,
+    shell_out,
 )
 
 
@@ -152,33 +156,49 @@ def test_adrv9371_zc706_xsa_hw(board, built_kernel_image_zynq, tmp_path):
         f"No AD9528 clock device found. Devices: {sorted(found)}"
     )
 
-    # TODO(adrv9371-capture): data-path smoke test is deferred on
-    # AD9371/ZC706 until the out-of-the-box design delivers samples
-    # from the TPL core without runtime intervention.
-    #
-    # Investigated and ruled out (commits b117949, 5222208 — both
-    # reverted):
-    #
-    # - Mirroring the adrv9009 post-boot Talise write with a Mykonos
-    #   profile does *not* unblock DMA here.  The profile file
-    #   (``ad9371_5/profile_TxBW100_ORxBW100_RxBW100.txt``, 2123 bytes,
-    #   deviceClock_kHz=122880 matching the AD9528 VCXO) is accepted
-    #   by the driver at ``/sys/.../iio:device2/profile_config`` — but
-    #   immediately after the write the JESD204 RX link leaves
-    #   ``Link status: DATA`` and does not return within a 30 s poll.
-    #   Talise re-inits and JESD-relocks on ``profile_config`` write;
-    #   Mykonos on AD9371 does not do the same JESD restart, leaving
-    #   the link down after any runtime profile change.
-    #
-    # - On initial boot (before any write), JESD is in DATA but
-    #   ``iio.Buffer.refill()`` on ``ad_ip_jesd204_tpl_adc`` times
-    #   out, i.e. the TPL DMA isn't streaming despite a locked link.
-    #
-    # The next investigation step is a board-specific capture recipe
-    # (possibly via the Mykonos driver's ``ensm_mode`` / calibration
-    # attributes, or via pyadi-iio's ``adi.ad9371`` helpers with
-    # explicit RX enable) — out of scope for the data-capture-smoke
-    # addition in this PR.
+    # --- 7. Switch Mykonos into radio_on before data capture. ---
+    # ad9371-phy defaults to ``ensm_mode=radio_off`` after probe on
+    # this Kuiper design — JESD204 lane training completes (Link
+    # status: DATA) but the ARM's radio FSM is not delivering
+    # samples to the TPL, so ``buf.refill()`` on
+    # ``ad_ip_jesd204_tpl_adc`` times out.  Flip to ``radio_on`` via
+    # sysfs to start sample flow.  (Runtime ``profile_config`` writes
+    # drop the JESD link on Mykonos — commits b117949 + 5222208
+    # reverted after proving that approach doesn't work — so use the
+    # simpler ENSM write here.)
+    ensm_path = shell_out(
+        shell,
+        r"for d in /sys/bus/iio/devices/iio:device*; do "
+        r'[ "$(cat "$d/name" 2>/dev/null)" = "ad9371-phy" ] '
+        r'&& echo "$d/ensm_mode" && break; '
+        r"done",
+    ).strip()
+    assert ensm_path, "Could not locate ad9371-phy/ensm_mode sysfs node"
+    initial_ensm = shell_out(shell, f"cat {ensm_path}").strip()
+    print(f"initial ensm_mode ({ensm_path}): {initial_ensm!r}")
+
+    shell_out(shell, f"echo radio_on > {ensm_path}")
+    time.sleep(1.0)
+    post_ensm = shell_out(shell, f"cat {ensm_path}").strip()
+    assert post_ensm == "radio_on", (
+        f"ensm_mode did not stick: wanted 'radio_on', got {post_ensm!r}"
+    )
+    print(f"post-write ensm_mode: {post_ensm}")
+
+    # ENSM transition should not disturb the link — the Mykonos ARM
+    # just starts streaming.  Re-verify JESD + dmesg stayed clean.
+    assert_jesd_links_data(shell, context="after ensm_mode=radio_on")
+    post_ensm_dmesg = shell_out(shell, "dmesg")
+    assert_no_kernel_faults(post_ensm_dmesg)
+    assert_no_probe_errors(post_ensm_dmesg)
+
+    # --- 8. Data-path smoke test: capture a real AD9371 RX buffer. ---
+    assert_rx_capture_valid(
+        ctx,
+        ("axi-ad9371-rx-hpc", "axi-ad9371-rx-obs-hpc"),
+        n_samples=2**12,
+        context="ad9371 xsa",
+    )
 
 
 # ---------------------------------------------------------------------------