docs(sphinx): note merged-DTB IIO names + ADRV9009 ZC706 fixes

Two narrow Sphinx updates so future contributors don't have to
rediscover the merged-DTB realities:

* `xsa.rst` — generalise the "Hardware test note" beyond FMCDAQ2:
  spell out *why* sdtgen-built XSAs land different IIO names than
  Kuiper-built DTs (`cf_axi_adc` reads the IIO name from
  `of_node->name`).  Call out the ADRV9009 ZC706 case where two
  IIO devices share the name `ad_ip_jesd204_tpl_adc` (RX +
  OBS) and tests need to disambiguate by `of_node` reg, plus the
  pyadi-iio `_rxadc = None` partial-init pattern.

* `xsa_developer.rst` — add an "ADRV9009 ZC706 specifics"
  subsection documenting the two builder workarounds: `sampl_clk`
  on TPL cores (cf_axi_adc reads its rate to size the DMA buffer)
  and the `zc706`-only DMA-done IRQ override
  (`PCW_IRQ_F2P_MODE = REVERSE` in the XSA vs the
  DIRECT-mode SPIs the bitstream actually fires on).

Author: tfcollins

doc/source/xsa.rst

@@ -833,11 +833,33 @@ the parser implementation.
 Hardware test note
 ~~~~~~~~~~~~~~~~~~
 
-Some XSA-derived FMCDAQ2 designs expose tpl-core IIO names instead of legacy
-core names. Hardware tests therefore accept either set:
-
-- legacy: ``axi-ad9680-hpc``, ``axi-ad9144-hpc``
-- tpl-core: ``ad_ip_jesd204_tpl_adc``, ``ad_ip_jesd204_tpl_dac``
+The Kuiper kernel's ``cf_axi_adc_core.c`` sets the IIO device name from
+``pdev->dev.of_node->name``, so the IIO name follows the DT *node-name*.
+Production Kuiper DTs spell the buffered ADC node ``axi-<chip>-rx-hpc``;
+sdtgen-built XSAs spell the same node ``ad_ip_jesd204_tpl_<adc|dac>``.
+Hardware tests therefore accept either set, e.g.:
+
+- FMCDAQ2: ``axi-ad9680-hpc`` / ``axi-ad9144-hpc`` *or*
+  ``ad_ip_jesd204_tpl_adc`` / ``ad_ip_jesd204_tpl_dac``
+- ADRV9009 ZC706: ``axi-adrv9009-rx-hpc`` /
+  ``axi-adrv9009-rx-obs-hpc`` *or* both ADC frontends as
+  ``ad_ip_jesd204_tpl_adc``
+
+ADRV9009 ZC706 in particular binds two IIO devices both named
+``ad_ip_jesd204_tpl_adc`` — RX (``@44a00000``, via ``cf_axi_adc``) and
+OBS (``@44a08000``, via ``ad_adc.c`` matching
+``adi,axi-adrv9009-obs-1.0``).  ``ctx.find_device(name)`` returns
+whichever probed first.  Tests that need the RX path (the one Talise
+``radio_on`` actually streams) disambiguate by ``of_node`` reg
+address; ``test/hw/xsa/test_adrv9009_zc706_overlay.py::test_dma_loopback``
+shows the pattern.
+
+``pyadi-iio`` looks up the buffered ADC by the production name
+(``axi-adrv9009-rx-hpc``).  On the merged DTB that name is absent, so
+``adi.adrv9009(uri=...)`` returns successfully but with
+``_rxadc = None``.  Tests that drop down to ``pyadi-iio`` therefore
+need to gate on ``getattr(dev, "_rxadc", None) is not None`` and skip
+the high-level path when the merged-DTB names are in play.
 
 Profile validation
 ~~~~~~~~~~~~~~~~~~

doc/source/xsa_developer.rst

@@ -525,6 +525,33 @@ Template catalogue
    * - ``axi_ad9081.tmpl``
      - AXI AD9081 MXFE PL core overlay.
 
+ADRV9009 ZC706 specifics
+~~~~~~~~~~~~~~~~~~~~~~~~
+
+The ADRV9009 builder applies two ZC706-only fixes to make
+buffered RX capture work on the sdtgen-merged DTB:
+
+1. **``sampl_clk`` on TPL cores.**  ``cf_axi_adc`` calls
+   ``clk_get_rate(sampl_clk)`` to size the DMA buffer.  Without it,
+   the driver still binds and the JESD link reaches ``DATA``, but
+   ``iio_buffer_refill`` never triggers and capture hangs.  The
+   builder wires ``sampl_clk`` in two passes: pass 1 points at the
+   PS reference (``clkc``); pass 2 redirects to the ADRV9009 chip's
+   per-direction clock-provider outputs (``trx0_adrv9009 0/1/2``)
+   once the phy label is resolvable.
+
+2. **DMA-done IRQ override.**  The Kuiper ZC706 ADRV9009 XSA
+   declares ``PCW_IRQ_F2P_MODE = REVERSE`` on the PS, and sdtgen
+   follows that — emitting DT cell 31 (SPI 63) for ``In13`` /
+   ``rx_dma``.  The bitstream actually fires DMA-done on the
+   *DIRECT*-mode SPIs the production
+   ``zynq-zc706-adv7511-adrv9009.dts`` uses (DT cells 57/56/55
+   for rx/tx/rx-obs).  With the sdtgen vectors,
+   ``/proc/interrupts`` shows zero counts during refill and capture
+   stalls.  The builder overrides ``interrupts`` on the rx / tx /
+   rx-obs ``axi_dmac`` nodes for ``platform == "zc706"`` only;
+   ZynqMP boards keep their sdtgen interrupts unchanged.
+
 Context builders
 ----------------