lopper: assists: Split Zephyr DTS generation into zephyr_domain_dts assist by dbingi-amd · Pull Request #779 · devicetree-org/lopper

dbingi-amd · 2026-06-29T06:13:23Z

• Split Zephyr device-tree generation out of gen_domain_dts into a new zephyr_domain_dts assist.
• gen_domain_dts now handles domain pruning and Linux/baremetal output only; Zephyr logic was removed from it.
• Added zephyr_domain_dts.py for Zephyr-specific transforms, board overlay handling, and related cleanup.
• Registered the new assist in lop-load.dts so Lopper loads it automatically.
• Replaced the old gen_domain_dts {proc} zephyr_dt [board.dts] flow with a separate zephyr_domain_dts {proc} [board.dts] step after domain (and imux) generation.
• Moved optional board overlay input to the second argument of zephyr_domain_dts (previously the third argument on gen_domain_dts).
• Preserved --extra-zephyr-comp support in the Zephyr assist (not in gen_domain_dts).
• Updated Zephyr platform docs, Linux domain-tree docs, and the developer guide to describe the new assist and command examples.
• No intended functional change to generated Zephyr DTS — this is primarily a refactor to separate concerns and simplify maintenance.

dbingi-amd · 2026-06-29T06:14:31Z

Hi @kedareswararao,
Please review.

kedareswararao · 2026-06-29T06:15:19Z

@zeddii : Please don't merge this until i approve it may break existing use cases need to cross verify properly before approving

kedareswararao · 2026-07-07T05:23:19Z

@@ -11,7 +11,6 @@
 import sys
 import os
 import glob


unrelated change

The removal of import argparse from gen_domain_dts.py is a direct consequence of this split. It was solely used by _extra_zephyr_comp_paths() (for --extra-zephyr-comp parsing), which has been moved to zephyr_domain_dts.py along with import argparse. No remaining code in gen_domain_dts.py uses argparse, so keeping it would be a dead import.

kedareswararao · 2026-07-07T05:26:40Z

    6. Keeps the status disabled nodes in the final device-tree.
    7. Delete the nodes that have an xlnx,ip-name property value mentioned in the linux_ignore_ip_list.
+
+    Zephyr device-tree generation uses assist zephyr_domain_dts (run after domain DTS).


no need to mention this here need to update the https://github.com/devicetree-org/lopper/tree/master/docs/amd/zephyr/source as per latest assist since it's taken care in the commit 2 no need to add this line here

Agreed, the line has been removed from the xlnx_generate_domain_dts docstring

kedareswararao · 2026-07-07T05:27:05Z

            # Add cpu0_intc (riscv,cpu-intc) as a child of the CPU node,
            # and add the riscv,timer node at root wired to cpu0_intc at IRQ 5.
            # Also wire axi_intc to cpu0_intc at IRQ 9 (IRQ_S_EXT).
-            if linux_dt and match_cpunode.propval('xlnx,ip-name', list)[0] == 'microblaze_riscv':


unrelated change

The explicit linux_dt and was removed because the block was moved inside the existing else: branch (line 529), which already implies linux_dt = 1. The old standalone if linux_dt and microblaze_riscv: and the new else: → if microblaze_riscv: produce identical behavior across all cases — no corner case is affected

there should be one change per commit if you are doing cleanup means it should be in a separate commit

kedareswararao · 2026-07-07T05:27:22Z

-                    if node.propval('xlnx,ip-name', list) == ['axi_intc']:
-                        node + LopperProp("interrupts-extended = <&cpu0_intc 9>")
-                        break
+            for node in sdt.tree['/'].subnodes():


same here unrelated change

This block is inside else: (line 529, = linux_dt) and if microblaze_riscv: — the effective guard is still linux_dt AND microblaze_riscv, same as before

kedareswararao · 2026-07-07T05:28:04Z

@@ -0,0 +1,1740 @@
+#/*
+# * Copyright (C) 2023-2026 Advanced Micro Devices, Inc. All Rights Reserved.


2026 is sufficient since it's a new file

Updated the copyright year

kedareswararao · 2026-07-07T05:29:05Z

+    Zephyr domain DT assist. Invoked after pruned domain DT is available.
+    Args: options['args'][0] = processor instance (same as gen_domain_dts);
+          options['args'][1] = optional Zephyr board overlay .dts path.
+          Optional flag --extra-zephyr-comp <path> may be repeated to merge


cross check this comment and fix properly

Updated the doc string

dbingi-amd · 2026-07-10T05:20:57Z

Hi @kedareswararao,

Please review the updated change

kedareswararao · 2026-07-10T07:05:47Z

                    mapped_children_nodes.append(node)
                    continue
-                elif xlnx_openamp_keep_node(linux_dt, zephyr_dt, node, sdt.tree):
+                elif xlnx_openamp_keep_node(linux_dt, False, node, sdt.tree):


@bentheredonethat : will this change effect open amp assists?
@dbingi-amd : Please check with Ben and test the open amp dts generation use case as well so that nothing will broke due to this change

@dbingi-amd @kedareswararao NAK - keep zephyr_dt flag there.

kedareswararao · 2026-07-10T07:06:05Z

                intc_node['#interrupt-cells'] = 1
                intc_node + LopperProp("interrupt-controller")
-                match_cpunode.add(intc_node)	
+                match_cpunode.add(intc_node)


unrelated change

kedareswararao · 2026-07-10T07:06:19Z

                timer_node.name = "timer"
                timer_node.label = "int_timer"
-                timer_node ["compatible"] = "riscv,timer"
+                timer_node["compatible"] = "riscv,timer"


same here unrelated change

zeddii · 2026-07-24T19:35:33Z

Bumping this. A few things are outstanding before this can be considered for merge:

There's an unaddressed NAK from @bentheredonethat ("keep zephyr_dt flag there"), and @kedareswararao's request to check with Ben and test the OpenAMP DTS generation use case — neither has had a follow-up yet.
The PR now conflicts with master and needs a rebase.

@dbingi-amd — could you rebase against the latest master and respond to the outstanding review comments? Note @kedareswararao has also asked to hold merge until they approve.

zeddii · 2026-07-29T21:10:37Z

This has drifted to CONFLICTING since my last note, so it needs a rebase on master before it can move forward.

Recapping what's still outstanding (unchanged):

the NAK from @bentheredonethat (keep the zephyr_dt flag) hasn't been addressed;
@kedareswararao is holding pending verification that the split doesn't break existing use cases.

@dbingi-amd — a rebase plus resolving the NAK would get this back to a reviewable state.

Remove the zephyr_dt CLI path and Zephyr-specific generation helpers from gen_domain_dts so the assist handles domain pruning and Linux/baremetal output only. linux_dt, baremetal, and zynqmp_fsbl behaviour is unchanged. Rename _zephyr_node_must_be_preserved() to _node_must_be_preserved() and keep it in the domain prune loop. OpenAMP pipelines run gen_domain_dts before zephyr_domain_dts, so nodes already referenced from /chosen must survive generic domain pruning — including ipc shm set by the OpenAMP assist. Signed-off-by: Bingi Dinesh kumar <[email protected]>

Move Zephyr device-tree generation into a dedicated assist that calls gen_domain_dts when the input is a full SDT, then applies Zephyr transforms. Platform lops such as imux remain external -i invocations. Register the assist in lop-load.dts. Reuse _node_must_be_preserved from gen_domain_dts during Zephyr-phase pruning instead of maintaining a separate copy of the preserve logic. Adjust zephyr_board_dt overlay discovery for the zephyr_domain_dts CLI. Carry forward the ZynqMP IPI mailbox child handling, MicroBlaze RISC-V RV32E kconfig generation, and other Zephyr-specific transforms recently present in gen_domain_dts. Signed-off-by: Bingi Dinesh kumar <[email protected]>

Point existing Zephyr unit tests at zephyr_domain_dts instead of gen_domain_dts zephyr_dt. Add IPI mailbox child conversion and promotion coverage for the VCK190 and ZCU102 cases handled by the new assist. Update the OpenAMP Zephyr pipeline sanity step to call zephyr_domain_dts. Signed-off-by: Bingi Dinesh kumar <[email protected]>

Replace gen_domain_dts <proc> zephyr_dt examples with zephyr_domain_dts <proc> [board.dts] in developer-guide.md and the AMD Zephyr platform guides for Cortex-A78, Cortex-R52, and MicroBlaze RISC-V. Note in the Linux device-tree chapter that gen_domain_dts remains the entry point for Linux domain trees. Signed-off-by: Bingi Dinesh kumar <[email protected]>

dbingi-amd · 2026-08-26T10:26:55Z

Hi @kedareswararao @bentheredonethat ,

Updated the branch per review feedback — please re-review.

@bentheredonethat — on your NAK: the zephyr_dt flag removed from gen_domain_dts CLI is intentional (step 5 is now zephyr_domain_dts). The openamp … zephyr_dt assist is unchanged. The xlnx_openamp_keep_node(linux_dt, False, …) change in domain prune is correct — that second arg is not the OpenAMP assist flag; it only affects UIO retention during Linux/baremetal prune.

Per your advice:
• Renamed _zephyr_node_must_be_preserved() → _node_must_be_preserved() and kept it in gen_domain_dts (step 4 domain prune), so /chosen refs like zephyr,ipc_shm survive before zephyr_domain_dts runs.
• VEK385 OpenAMP multidomain flow re-tested — pass after this fix.

OpenAMP test flow used (VEK385):

YAML merge — merge OpenAMP/libmetal YAML into SDT
Domain chosen — apply lop-domain-chosen for RPU Zephyr domain
OpenAMP assist — openamp cortexr52_0 zephyr_dt (unchanged by this PR)
Domain prune + imux — gen_domain_dts cortexr52_0 + R52 imux lop (unchanged)
Zephyr DTS — zephyr_domain_dts cortexr52_0 + board overlay (was gen_domain_dts zephyr_dt)

Also ran pytest and lopper_sanity.py --openamp.

OpenAMP plugin extension for mmio-sram ipc nodes can follow separately if you still want that.

@kedareswararao — your feedback addressed: unrelated Linux changes reverted, Zephyr line removed from gen_domain_dts docstring, docs updated under docs/amd/zephyr/source.
Please re-review the updated commits.

Thanks,
Dinesh

kedareswararao reviewed Jul 7, 2026

View reviewed changes

dbingi-amd force-pushed the zephyr-lopper branch 3 times, most recently from e1b60d6 to e566888 Compare July 10, 2026 05:18

kedareswararao reviewed Jul 10, 2026

View reviewed changes

dbingi-amd force-pushed the zephyr-lopper branch from e566888 to f77c5c6 Compare August 26, 2026 10:12

dbingi-amd added 4 commits August 26, 2026 15:51

dbingi-amd force-pushed the zephyr-lopper branch from f77c5c6 to 097a0f2 Compare August 26, 2026 10:25

		@@ -0,0 +1,1740 @@
		#/*
		# * Copyright (C) 2023-2026 Advanced Micro Devices, Inc. All Rights Reserved.

Uh oh!

Conversation

dbingi-amd commented Jun 29, 2026

Uh oh!

dbingi-amd commented Jun 29, 2026

Uh oh!

kedareswararao commented Jun 29, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dbingi-amd commented Jul 10, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

zeddii commented Jul 24, 2026

Uh oh!

zeddii commented Jul 29, 2026

Uh oh!

dbingi-amd commented Aug 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

dbingi-amd commented Aug 26, 2026 •

edited

Loading