hujun260 opened a new pull request, #22082:
URL: https://github.com/apache/nuttx/pull/22082
# arm64/imx95: enable SMP on i.MX95, expand DRAM to 512 MB, and align
nsh_smp defconfig
## Summary
Five commits preparing `imx95-a55-evk` for a real SMP workload (FastDDS / C++
runtime, basePlateAPP, SQLite):
1. `arm64/cmake` — set default LLVM arch/ABI so downstream WASM AOT tooling
can
generate native code for arm64 targets.
2. `boards/arm64/imx95` — add the first `nsh_smp` defconfig for the board.
3. `imx95-a55-evk` — grow NuttX DRAM from 128 MB to 512 MB, keeping i.MX93 at
128 MB.
4. `imx95-a55-evk` — align `nsh_smp` with the `usbdev` config: eMMC, USB/ADB,
FATFS, large files.
5. `imx95-a55-evk` — align `nsh_smp` for the FastDDS/C++ runtime: locale,
`atexit`, file-descriptor limits, `pthread_atfork`, larger stacks, more
TLS
slots.
## Commits
```text
e8a307c9e51 arm64/cmake: set default LLVM arch and ABI type for arm64 targets
953a1842840 boards/arm64/imx95: add smp
fa69c1a2ebd imx95-a55-evk: expand NuttX RAM from 128MB to 512MB
1d9a34dfade boards/imx95-a55-evk: align nsh_smp defconfig with usbdev
504a6c3b132 boards/imx95-a55-evk: align nsh_smp defconfig for FastDDS runtime
```
```text
arch/arm64/include/imx9/chip.h | 5 +-
arch/arm64/src/cmake/Toolchain.cmake | 4 +
boards/arm64/imx9/imx95-a55-evk/configs/nsh/defconfig | 2 +-
boards/arm64/imx9/imx95-a55-evk/configs/nsh_smp/defconfig | 123
++++++++++++++++++++--
boards/arm64/imx9/imx95-a55-evk/scripts/dramboot.ld | 4 +-
5 files changed, 130 insertions(+), 8 deletions(-)
```
## Change detail
### 1. LLVM arch/ABI defaults (`arch/arm64/src/cmake/Toolchain.cmake`)
```diff
+# Set default LLVM arch/abi for all arm64 targets
+set(LLVM_ARCHTYPE aarch64)
+set(LLVM_ABITYPE aapcs64)
```
These are consumed by wamrc when AOT-compiling WASM to native arm64. Without
them the tooling has no way to infer the target triple from the
Makefile-driven
build, so every arm64 board had to set it locally.
### 2. NuttX DRAM 128 MB → 512 MB (`arch/arm64/include/imx9/chip.h`)
`CONFIG_RAMBANK1_SIZE` was a single `MB(128)` shared by both i.MX93 and
i.MX95.
It is now selected per chip so i.MX93 keeps its 128 MB:
```diff
#if defined(CONFIG_ARCH_CHIP_IMX93)
#define CONFIG_RAMBANK1_ADDR 0x80000000
+#define CONFIG_RAMBANK1_SIZE MB(128)
#elif defined(CONFIG_ARCH_CHIP_IMX95)
#define CONFIG_RAMBANK1_ADDR 0xa0100000
+#define CONFIG_RAMBANK1_SIZE MB(512)
#endif
-#define CONFIG_RAMBANK1_SIZE MB(128)
```
Because the PCI DMA window used to sit immediately after the old 128 MB
bank, it
moves out of the way:
```diff
-#define CONFIG_PCI_DMA_ADDR 0xa8100000
+#define CONFIG_PCI_DMA_ADDR 0xc0100000
```
`0xa0100000 + 512 MB = 0xc0100000`, so the DMA window now starts exactly
where
DRAM ends and no region overlaps. The linker script and both defconfigs
follow:
```diff
- dram (rwx) : ORIGIN = 0xa0100000, LENGTH = 128M
+ dram (rwx) : ORIGIN = 0xa0100000, LENGTH = 512M
-CONFIG_RAM_SIZE=134217728
+CONFIG_RAM_SIZE=536870912
```
### 3–5. `nsh_smp` defconfig
The `nsh_smp` config starts from the existing `nsh` config and adds, in
order:
- **SMP**: `CONFIG_SMP=y`, `CONFIG_SMP_NCPUS=6` (all six Cortex-A55 cores).
- **Storage / USB**: `CONFIG_IMX9_USDHC1` (eMMC), `CONFIG_USBDEV_DWC3` + ADB,
`FS_FATFS` instead of the legacy `FS_FAT`, `FS_LARGEFILE`.
- **C++ runtime**: `LIBC_LOCALE` + `LIBC_LOCALE_GETTEXT`, `LIBC_LOCALTIME`,
`LIBC_MAX_EXITFUNS=1`, `LIBC_OPEN_MAX=255`, `PTHREAD_ATFORK`,
`PTHREAD_STACK_DEFAULT=65536`, `TLS_NELEM=16`, `TLS_TASK_NELEM=32`.
`FS_LARGEFILE` is not cosmetic: MBR partition parsing multiplies `blkcnt_t`
by
512, which overflows 32 bits for partitions beyond 4 GB, so the switch to
FATFS
is what makes the eMMC usable at all on a 512 MB device image.
## Verification
### Build
```text
$ export PATH=<aarch64-none-elf>/bin:$PATH
$ tools/configure.sh -a ../apps imx95-a55-evk:nsh_smp
$ make -j$(nproc)
...
LD: nuttx
Memory region Used Size Region Size %age Used
dram: 680160 B 512 MB 0.13%
CP: nuttx.hex
CP: nuttx.bin
```
The 512 MB region is visible in the linker report. The other two configs of
the
same board also build clean:
```text
imx95-a55-evk:nsh PASS dram: 483136 B / 512 MB
imx95-a55-evk:emmc PASS dram: 540528 B / 512 MB
imx95-a55-evk:nsh_smp PASS dram: 680160 B / 512 MB
```
### checkpatch
```text
$ tools/checkpatch.sh -c -u -m -g ed122374690..HEAD
Used config files:
1: .codespellrc
✔️ All checks pass.
```
### Note: a build break fixed in this series
The `usbdev` alignment commit originally replaced `CONFIG_SCHED_HPWORK=y`
with
the new MMCSD/USDHC options instead of adding to them. That broke
`imx95-a55-evk:nsh_smp` outright:
```text
./chip/imx9_usdhc.c:93:4: error: #error "Callback support requires
CONFIG_SCHED_WORKQUEUE and CONFIG_SCHED_HPWORK"
```
`imx9_usdhc.c` requires both symbols whenever the driver and `MMCSD_SDIO` are
enabled. The fix is folded into that commit:
```diff
CONFIG_SCHED_HAVE_PARENT=y
+CONFIG_SCHED_HPWORK=y
CONFIG_SCHED_HPWORKPRIORITY=192
CONFIG_SCHED_LPWORK=y
CONFIG_SCHED_LPWORKPRIORITY=50
+CONFIG_SCHED_WORKQUEUE=y
CONFIG_SDIO_BLOCKSETUP=y
```
Note that `CONFIG_SCHED_HPWORKPRIORITY=192` was already present without
`CONFIG_SCHED_HPWORK=y`, which is why the break was not obvious from the
priority setting alone.
## Honest notes
- **No runtime verification.** These changes target `imx95-a55-evk` hardware,
which is not available here. Everything above is build- and
tooling-verified; the SMP boot path itself was not exercised on a board and
no `nsh` session output is included.
- **`CONFIG_SMP_NCPUS=6` vs. `CONFIG_NCPUS`.** This series predates the
`CONFIG_NCPUS` unification, so the defconfig uses `CONFIG_SMP_NCPUS`. If
it is
rebased onto a tree that renamed the symbol, that line needs updating.
- **The 512 MB figure is the NuttX bank, not total DRAM.**
`CONFIG_RAMBANK1_ADDR`
is `0xa0100000` (after the u-boot reserved area), so the mapped region
ends at
`0xc0100000`. Boards whose u-boot layout differs will need their own value.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]