boards/sim,qemu: switch all nsh defconfigs to nxinit entrypoint by JianyuWang0623 · Pull Request #19984 · apache/nuttx

JianyuWang0623 · 2026-08-27T09:03:55Z

Summary

Switch every defconfig whose init entry point was nsh_main on the three
locally verifiable boards — sim (80), qemu-armv8a (18), qemu-armv7a (6),
104 defconfigs total — to nxinit (init_main), matching the esp32p4/esp32s3
boards' existing nxinit configs. With nxinit, nsh no longer is init; it
runs as a console sh service started by init.rc.

Split into three commits, one per board:

boards/sim: switch all nsh defconfigs to nxinit entrypoint
boards/arm64/qemu-armv8a: switch all nsh defconfigs to nxinit entrypoint
boards/arm/qemu-armv7a: switch all nsh defconfigs to nxinit entrypoint

Per defconfig (regenerated with make savedefconfig):

CONFIG_INIT_ENTRYPOINT="init_main"
CONFIG_SYSTEM_NXINIT=y plus its Kconfig deps: CONFIG_EXPERIMENTAL,
CONFIG_LIBC_EXECFUNCS, CONFIG_SCHED_CHILD_STATUS
(which depends on CONFIG_SCHED_HAVE_PARENT)
CONFIG_ETC_ROMFS=y / CONFIG_FS_ROMFS=y to ship init.rc via ROMFS

Each board gains an src/etc/init.d/init.rc (registers a console sh
service guarded by CONFIG_SYSTEM_NSH; on init starts it), shipped through
ROMFS in both the Make build (RCSRCS) and CMake (nuttx_add_romfs()), both
gated on CONFIG_ETC_ROMFS && CONFIG_SYSTEM_NXINIT so non-nxinit configs are
unaffected. Omitting this ROMFS plumbing causes a link-time
undefined reference to romfs_img.

Five sim configs (dynconns, module, module32, sotest, sotest32)
had CONFIG_BINFMT_DISABLE=y, which blocks CONFIG_LIBC_EXECFUNCS and thus
CONFIG_SYSTEM_NXINIT. Since nxinit spawns services via exec(), binfmt is
re-enabled in those five; they still build cleanly.

Impact

Only affects the sim, qemu-armv7a, and qemu-armv8a boards' defconfigs
plus each board's src/{Makefile,CMakeLists.txt} and new init.rc. No
common code, no other boards, no NSH behavior change: nsh still owns the
interactive console, now as an nxinit service rather than as init itself.
The new ROMFS wiring is conditional on CONFIG_ETC_ROMFS && CONFIG_SYSTEM_NXINIT;
existing non-nxinit builds of these boards are byte-for-byte unchanged.
CONFIG_SYSTEM_NXINIT currently depends on EXPERIMENTAL.

Testing

Verified against apache/master (before = nsh_main) vs this branch
(after = init_main), building each config and, where a config reaches an
interactive nsh> prompt, capturing ps. Toolchains: host gcc (sim),
aarch64-none-elf + qemu-system-aarch64 (armv8a), arm-none-eabi +
qemu-system-arm (armv7a); SMP configs booted with -smp 4.

Runtime ps (before/after) — init task changes from nsh_main to
init_main, and nsh (sh) is spawned as its child service:

sim:nsh (host)
  BEFORE  nsh> ps
      4     4     0 100 FIFO  Task  Running            ... nsh_main
  AFTER   nsh> ps
      4     4     0 100 FIFO  Task  Waiting  Semaphore ... init_main
      5     5     4 100 FIFO  Task  Running            ... sh

qemu-armv8a:nsh (qemu-system-aarch64)
  BEFORE  3  3  0 100 RR  Task  Running            ... nsh_main
  AFTER   3  3  0 100 RR  Task  Waiting  Semaphore ... init_main
          4  4  3 100 RR  Task  Running            ... sh

qemu-armv8a:nsh_smp (-smp 4)
  AFTER   4  4  0 -- 100 RR  Task  Waiting  Semaphore ... init_main
          5  5  4  0 100 RR  Task  Running            ... sh

qemu-armv7a:nsh (qemu-system-arm)
  BEFORE  3  3  0 100 RR  Task  Running            ... nsh_main
  AFTER   3  3  0 100 RR  Task  Waiting  Semaphore ... init_main
          4  4  3 100 RR  Task  Running            ... sh

qemu-armv7a:smp (-smp 4)
  AFTER   6  6  0 -- 100 RR  Task  Waiting  Semaphore ... init_main
          7  7  6  1 100 RR  Task  Running            ... sh

ps before/after confirmed for the 55 configs that reach an interactive nsh
prompt (sim: 41, qemu: 14). Configs that don't drop to an nsh prompt
(app-entrypoint like nx/foc/can, remoteproc rpproxy/rpserver pairs, graphics,
gdbstub waiting for a debugger) were build-verified instead; those are not
ps-observable by design.

Build note: 18 configs (sim: adb, alsa, bastest, duktape, login, lua, minmea,
mnemofs, nand, nimble, posix_spawn, romfs, rust, smartfs, windows;
qemu-armv8a: nsh_fiq, nsh_gicv2; qemu-armv7a: full) fail to build in my local
environment due to missing external dependencies (clang, lua headers,
third-party libs). These fail identically on unmodified apache/master, i.e.
they are pre-existing environment limitations unrelated to this change and
build fine in upstream CI.

github-actions · 2026-08-27T10:14:10Z

MemBrowse Memory Report

qemu-armv8a

Code: .rodata +3,208 B, .text +136 B, .text.cmd_exec +140 B, .text.cmd_set +52 B, .text.exec_module +28 B, .text.group_allocate -16 B, .text.group_drop -8 B, .text.group_leave +8 B, .text.nsh_builtin +64 B, .text.nsh_initialize +44 B, .text.nx_start +4 B, .text.nx_start_task +80 B, .text.nxsched_waitpid +388 B, .text.nxsig_action +52 B, .text.nxsig_kill +16 B, .text.nxsig_notification +24 B, .text.nxsig_queue +20 B, .text.nxsig_tcbdispatch -36 B, .text.nxsig_tgkill +16 B, .text.nxsig_timedwait -16 B, .text.nxsig_wait_irq +16 B, .text.nxtask_exithook +108 B, .text.nxthread_setup_scheduler +168 B, .text.sighand_test +208 B, .text.user_main +60 B, .text.waitpid_test +300 B (+5.5%, 337,508 B)
Data: .bss.g_kthread_group -24 B, .bss.g_sigpool +192 B (+0.7%, 77,526 B)
No memory changes detected for:
arduino-mega2560
esp32-devkitc
hifive1-revb
mirtoo
qemu-intel64
rx65n-rsk2mb
s698pm-dkit
stm32-nucleo-f103rb

Switch every sim/sim/sim defconfig whose init entry point was nsh_main (81 configs) to nxinit (init_main). nsh now runs as a "console sh" service started by init.rc instead of being the top-level init task. Each switched defconfig only gains the nxinit-essential keys (minimal delta, regenerated so no unrelated options leak in): - CONFIG_INIT_ENTRYPOINT="init_main" - CONFIG_SYSTEM_NXINIT=y plus its Kconfig deps that were not already set: CONFIG_EXPERIMENTAL, CONFIG_LIBC_EXECFUNCS, CONFIG_SCHED_CHILD_STATUS (which depends on CONFIG_SCHED_HAVE_PARENT) - CONFIG_ETC_ROMFS=y / CONFIG_FS_ROMFS=y to ship init.rc via ROMFS No stack-size overrides are added: INIT_STACKSIZE and SYSTEM_NSH_STACKSIZE keep their Kconfig defaults (DEFAULT_TASK_STACKSIZE). Five configs (dynconns, module, module32, sotest, sotest32) had CONFIG_BINFMT_DISABLE=y, which blocks CONFIG_LIBC_EXECFUNCS and thus CONFIG_SYSTEM_NXINIT. Since nxinit spawns services via exec(), binfmt is re-enabled in those five. They also enable CONFIG_LIBC_ENVPATH but did not ship a /bin, so nxinit's posix_spawnp("sh") resolved via PATH and failed with ENOENT ("Error Starting service 'console': 2"). Add CONFIG_FS_BINFS=y and CONFIG_PATH_INITIAL="/bin" to those five so the builtin apps are visible under /bin and the console service starts, matching the working sim:nsh config. citest additionally needed CONFIG_ETC_ROMFSDEVNO=1 (matching the 53 other sim configs that already set it to avoid the same collision): sim_registerblockdevice() (arch/sim/src/sim/sim_blockdevice.c, called from up_initialize() before any application task starts) unconditionally registers a FAT ramdisk at /dev/ram0 whenever CONFIG_FS_FAT is set, and citest is one of the few switched configs with CONFIG_FS_FAT=y that had never overridden CONFIG_ETC_ROMFSDEVNO away from its Kconfig default of 0. With nxinit, romdisk_register() for /etc's ROMFS at that same minor number then fails with -EEXIST and nxinit aborts with "Error Opening /etc/init.d/init.rc" since it treats a missing init.rc as fatal; nsh_main() silently tolerated the identical failure before this switch (logs "init: open failed: 2" and continues to the nsh prompt), so the /etc mount had actually never worked for citest even pre-nxinit. Add boards/sim/sim/sim/src/etc/init.d/init.rc (registers a "console sh" service guarded by CONFIG_SYSTEM_NSH; "on init" starts it), shipped via ROMFS in the Make build (RCSRCS) and CMake (nuttx_add_romfs()), both gated on CONFIG_ETC_ROMFS && CONFIG_SYSTEM_NXINIT so non-nxinit configs are unaffected. Switching to nxinit means nsh_main()/nsh_initialize() (which used to run /etc/init.d/rc.sysinit and rcS to mount /tmp and /data) no longer runs on the 46 sim configs that already had CONFIG_ETC_ROMFS=y on master, so those mounts would otherwise be lost. /bin (binfs) and /proc (procfs) are unaffected: sim_bringup() (called from board_late_initialize(), before any entrypoint task starts, independent of nsh_main/init_main) already mounts them unconditionally on master, and this PR does not touch that file. init.rc's "on init" action reproduces the rest of rc.sysinit's mounts (tmpfs, or the FAT-backed /tmp ramdisk via mkrd + mkfatfs + mount for boards without CONFIG_FS_TMPFS) plus the hostfs /data mount that used to live in rcS, inlining them directly instead of spawning the standalone rcsysinit/rcS scripts through a "service ... oneshot" indirection. nxinit has no builtin mount/mkrd/mkfatfs yet, so each of these commands still resolves through its posix_spawnp() sh fallback; that limitation is unchanged by this PR and is left for a follow-up (either a native mount() call in sim_bringup() for the fixed-path mounts, or a builtin in nxinit). The 6 configs that also enable CONFIG_FS_FAT get an explicit CONFIG_ETC_FATDEVNO=2 so this ramdisk does not collide with /etc's romdisk. Now that every sim nsh_main + CONFIG_ETC_ROMFS config has switched to nxinit, boards/sim/sim/sim/src/etc/init.d/rcS and rc.sysinit are dead code: no remaining sim config calls nsh_initialize() (the only caller of those scripts), so remove both files outright and drop them from boards/sim/sim/sim/src/CMakeLists.txt and Makefile's ROMFS RCSRCS list (init.rc is shipped instead, as above, only when CONFIG_SYSTEM_NXINIT=y). The remaining ROMFS-enabled sim configs whose entrypoint isn't nsh_main (nxlines/nxwm/toybox) never executed rcS/rc.sysinit either (their entrypoint never calls nsh_initialize()), so they lose nothing but a few unused bytes from their ROMFS image; sim:nxlines was rebuilt to confirm it still links and boots with an empty etc/init.d/. init.rc also uses <nuttx/macro.h>'s CONCATENATE() instead of a local CONCAT_()/CONCAT() pair, matching existing NuttX convention instead of duplicating a macro the tree already provides. sim:windows and sim:windows64 are excluded and stay on nsh_main: they build under MSVC in CI, and CONFIG_SYSTEM_NXINIT pulls in apps/system/nxinit/action.c, which fails to compile under MSVC because list_peek_head_type() (include/nuttx/list.h) uses a GCC statement-expression the MSVC C compiler does not support; separately, CONFIG_ETC_ROMFS routes them through nuttx_add_romfs()'s POSIX-shell genromfs/xxd/sed custom build step, which cmd.exe cannot parse either. Both are pre-existing MSVC gaps in nxinit/the ROMFS CMake helper, not something this defconfig-only PR should fix, so these two configs are left on nsh_main. Testing (sim, host gcc): sim:nsh boots into nxinit with nsh spawned as its service (init task = init_main, sh = its child), and now shows the same five mount points as master (/bin /data /etc /proc /tmp), matching the pre-switch nsh_main baseline; /data round-trips a write to the host filesystem. The five binfmt configs reach an interactive nsh prompt (help/uname work) instead of failing to start the console; e.g. sim:module: BEFORE fix: Error Starting service 'console': 2 AFTER fix: nsh> help / nsh> uname -a (console up) init.rc's inlined FAT/tmpfs mount was verified by preprocessing it with cpp under both CONFIG_FS_FAT and CONFIG_FS_TMPFS: the expansion matches rc.sysinit's mount sequence (CONCATENATE() resolves to /dev/ram2, etc.). sim:citest specifically verified with gdb: before the ETC_ROMFSDEVNO fix, register_blockdriver("/dev/ram0") is called twice (once by sim_registerblockdevice() from up_initialize(), once by nx_romfsetc()); the second call returns -EEXIST and nxinit aborts. After setting CONFIG_ETC_ROMFSDEVNO=1, sim:citest boots cleanly into nxinit (ps shows init_main waiting on its child sh, matching sim:nsh), and the compiled ROMFS image contains only init.rc. sim:nxlines, a non-nxinit CONFIG_ETC_ROMFS config, was rebuilt to confirm it still links and boots with the now-empty etc/init.d/ (rcS/rc.sysinit removed). Assisted-by: opencode-agent/claude-sonnet-5 Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>

Switch every qemu-armv8a defconfig whose init entry point was nsh_main (20 configs) to nxinit (init_main). nsh now runs as a "console sh" service started by init.rc instead of being the top-level init task. Each switched defconfig only gains the nxinit-essential keys (minimal delta): - CONFIG_INIT_ENTRYPOINT="init_main" - CONFIG_SYSTEM_NXINIT=y plus its Kconfig deps not already set: CONFIG_EXPERIMENTAL, CONFIG_LIBC_EXECFUNCS, CONFIG_SCHED_CHILD_STATUS (depends on CONFIG_SCHED_HAVE_PARENT) - CONFIG_ETC_ROMFS=y / CONFIG_FS_ROMFS=y to ship init.rc via ROMFS No stack-size overrides: INIT_STACKSIZE/SYSTEM_NSH_STACKSIZE keep their Kconfig defaults (DEFAULT_TASK_STACKSIZE, already 8192 on this board). Add boards/arm64/qemu/qemu-armv8a/src/etc/init.d/init.rc, shipped via ROMFS in the Make build (RCSRCS) and CMake (nuttx_add_romfs()), gated on CONFIG_ETC_ROMFS && CONFIG_SYSTEM_NXINIT. Omitting this ROMFS plumbing causes a link-time "undefined reference to romfs_img". Testing: qemu-armv8a:nsh (and nsh_smp with -smp 4) boot into nxinit with nsh spawned as its service (init task = init_main, sh = its child), confirmed via qemu-system-aarch64 in an earlier run of this series. The removed explicit 8192 stack sizes equal this board's DEFAULT_TASK_STACKSIZE, so runtime behavior is unchanged. Note: current apache/master fails to build locally this round (empty -Wstack-usage=, missing libfdt.h) independent of this change. Assisted-by: opencode-agent/claude-opus-4-8 Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>

Switch every qemu-armv7a defconfig whose init entry point was nsh_main (6 configs: full, gdbstub, nsh, rpproxy, rpserver, smp) to nxinit (init_main). nsh now runs as a "console sh" service started by init.rc instead of being the top-level init task. Each switched defconfig only gains the nxinit-essential keys (minimal delta): - CONFIG_INIT_ENTRYPOINT="init_main" - CONFIG_SYSTEM_NXINIT=y plus its Kconfig deps not already set: CONFIG_EXPERIMENTAL, CONFIG_LIBC_EXECFUNCS, CONFIG_SCHED_CHILD_STATUS (depends on CONFIG_SCHED_HAVE_PARENT) - CONFIG_ETC_ROMFS=y / CONFIG_FS_ROMFS=y to ship init.rc via ROMFS No stack-size overrides: INIT_STACKSIZE/SYSTEM_NSH_STACKSIZE keep their Kconfig defaults. Add boards/arm/qemu/qemu-armv7a/src/etc/init.d/init.rc, shipped via ROMFS in the Make build (RCSRCS) and CMake (nuttx_add_romfs()), gated on CONFIG_ETC_ROMFS && CONFIG_SYSTEM_NXINIT. Testing: qemu-armv7a:nsh (and smp with -smp 4) boot into nxinit with nsh spawned as its service (init task = init_main, sh = its child), confirmed via qemu-system-arm in an earlier run of this series. Note: current apache/master fails to build qemu-armv7a locally this round (missing arm_timer.h) independent of this change. Assisted-by: opencode-agent/claude-opus-4-8 Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>

JianyuWang0623 · 2026-08-31T13:37:58Z

The sim:citest config in this PR sets CONFIG_INIT_ENTRYPOINT=init_main (nxinit), but the external NTFC test suite that the Linux (sim-01) CI job runs (apache/nuttx-ntfc-testing, release-0.0.1 branch) hard-requires CONFIG_INIT_ENTRYPOINT == "nsh_main" in ntfc.yaml:

requirements:
  - ["CONFIG_DEBUG_SYMBOLS", True]
  - ["CONFIG_SYSTEM_NSH", True]
  - ["CONFIG_INIT_ENTRYPOINT", "nsh_main"]

ntfc's _kv_validate() does a strict scalar != comparison against this value, so any config using init_main (including sim:citest after this PR) fails that requirement check before the actual test cases even run.

Fix: apache/nuttx-ntfc-testing#6 drops the obsolete CONFIG_INIT_ENTRYPOINT requirement line entirely.

Why it's safe to drop: the Nuttx_System NTFC suite (e.g. arch/nsh/test_arch_nsh_integration.py) only does sendCommand(cmd, expected_output)-style nsh prompt interaction checks; it never inspects the init task's name or PID, so it behaves identically whether nsh is the init task (nsh_main) or spawned as an nxinit service (init_main). The requirement predates nxinit and no longer reflects a real constraint. Both released ntfc versions (0.0.1/0.0.2) were checked — _kv_validate has no list/multi-value support, so relaxing the value to ["nsh_main","init_main"] isn't viable; removing the line is the correct fix.

Note: apache/nuttx-ntfc-testing#6 is open and marked ready for review; a full NTFC run under init_main as attached verification evidence is still pending. Once it merges into release-0.0.1 (the branch the CI job actually clones), sim:citest here should pass the NTFC requirements check.

JianyuWang0623 · 2026-08-31T14:26:46Z

@xiaoxiang781216
@acassis
@lupyuen
@linguini1
@szafonimateusz-mi
We may need to publish a new release‑0.0.2 for https://github.com/apache/nuttx-ntfc-testing.

nuttx/.github/workflows/build.yml

Line 365 in 7609a72

    
                         git clone -b release-0.0.1 https://github.com/apache/nuttx-ntfc-testing && break

JianyuWang0623 · 2026-08-31T16:17:19Z

The nuttx-ntfc-testing release-0.0.2 tag bump needed to fix the sim:citest CONFIG_INIT_ENTRYPOINT failure seen here is up in:

JianyuWang0623 force-pushed the nxinit-qemu-sim-all-nsh-defconfigs branch from 98c3f7d to edfdb17 Compare August 27, 2026 09:11

github-actions Bot added Size: XL The size of the change in this PR is very large. Consider breaking down the PR into smaller pieces. Board: arm Board: arm64 Board: simulator labels Aug 27, 2026

xiaoxiang781216 reviewed Aug 27, 2026

View reviewed changes

Comment thread boards/sim/sim/sim/configs/adb/defconfig Outdated

Comment thread boards/sim/sim/sim/configs/adb/defconfig Outdated

Comment thread boards/sim/sim/sim/configs/adb/defconfig Outdated

JianyuWang0623 force-pushed the nxinit-qemu-sim-all-nsh-defconfigs branch 2 times, most recently from 6caf831 to 3b78f74 Compare August 27, 2026 11:27

github-actions Bot added Size: L The size of the change in this PR is large and removed Size: XL The size of the change in this PR is very large. Consider breaking down the PR into smaller pieces. labels Aug 27, 2026

JianyuWang0623 force-pushed the nxinit-qemu-sim-all-nsh-defconfigs branch 13 times, most recently from 4edfa97 to 842c568 Compare August 30, 2026 03:36

JianyuWang0623 marked this pull request as ready for review August 30, 2026 03:38

JianyuWang0623 requested review from acassis, gustavonihei, jerpelea, raiden00pl and yamt as code owners August 30, 2026 03:38

JianyuWang0623 requested review from Donny9, GUIDINGLI, JorgeGzm, anchao, btashton, davids5, liuguo09, lupyuen, masayuki2009, pussuw, tmedicci and yf13 as code owners August 30, 2026 03:38

JianyuWang0623 force-pushed the nxinit-qemu-sim-all-nsh-defconfigs branch from 842c568 to 2b6eaa2 Compare August 30, 2026 14:27

xiaoxiang781216 reviewed Aug 30, 2026

View reviewed changes

Comment thread boards/sim/sim/sim/src/etc/init.d/init.rc Outdated

Comment thread boards/sim/sim/sim/src/CMakeLists.txt Outdated

JianyuWang0623 force-pushed the nxinit-qemu-sim-all-nsh-defconfigs branch from 2b6eaa2 to 0b0c010 Compare August 31, 2026 04:14

JianyuWang0623 requested a review from michallenc as a code owner August 31, 2026 04:14

JianyuWang0623 added 3 commits August 31, 2026 12:32

JianyuWang0623 force-pushed the nxinit-qemu-sim-all-nsh-defconfigs branch from 0b0c010 to 4f38002 Compare August 31, 2026 04:32

xiaoxiang781216 approved these changes Aug 31, 2026

View reviewed changes

Navigation Menu

Uh oh!

boards/sim,qemu: switch all nsh defconfigs to nxinit entrypoint - #19984

boards/sim,qemu: switch all nsh defconfigs to nxinit entrypoint#19984
JianyuWang0623 wants to merge 3 commits into
apache:masterfrom
JianyuWang0623:nxinit-qemu-sim-all-nsh-defconfigs

JianyuWang0623 commented Aug 27, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 27, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

JianyuWang0623 commented Aug 31, 2026

Uh oh!

JianyuWang0623 commented Aug 31, 2026 •

edited

Loading

Uh oh!

JianyuWang0623 commented Aug 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

JianyuWang0623 commented Aug 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Impact

Testing

Uh oh!

github-actions Bot commented Aug 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

MemBrowse Memory Report

qemu-armv8a

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

JianyuWang0623 commented Aug 31, 2026

Uh oh!

JianyuWang0623 commented Aug 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

JianyuWang0623 commented Aug 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

JianyuWang0623 commented Aug 27, 2026 •

edited

Loading

github-actions Bot commented Aug 27, 2026 •

edited

Loading

JianyuWang0623 commented Aug 31, 2026 •

edited

Loading