Brijesh-Thakkar opened a new pull request, #20761: URL: https://github.com/apache/nuttx/pull/20761
*Note: Please adhere to [Contributing Guidelines](https://github.com/apache/nuttx/blob/master/CONTRIBUTING.md).* ## Summary Fixes #18393. On Ubuntu 20.04 (gcc 10.5, binutils 2.34), the sim link fails with: /usr/bin/ld: cannot find -lgcc_s /usr/bin/ld: cannot find -lgcc_s collect2: error: ld returned 1 exit status Root cause: the sim build passed `-Wl,-no-pie`. That flag only reaches `ld`, so the gcc driver still sets up a PIE link (`Scrt1.o`, `crtbeginS.o`). On this toolchain, `ld` then searches only for `libgcc_s.a`, which Ubuntu does not ship. NuttX never names `-lgcc_s` itself; the gcc driver adds it. The same failure reproduces outside NuttX: gcc hello.c # ok gcc -no-pie hello.c # ok gcc -Wl,-no-pie hello.c # cannot find -lgcc_s (x2) The make build was already fixed by fd907e6ef4 ("sim: pass no-pie through gcc driver"). The CMake build still used `-Wl,-no-pie`. On focal it fails even earlier, in CMake's compiler check at configure time. This PR has two commits: 1. **arch/sim: resolve host symbols eagerly on Linux in CMake.** Adds `-Wl,-z,now`, which make has had since ad182bcb4b but CMake never got. Without it, once the binary is truly non-PIE, early host calls made with simulated interrupts masked fault inside the lazy binder: _dl_fixup <- _dl_runtime_resolve_xsavec <- up_irq_save <- host_allocheap <- nx_start The same binary boots with `LD_BIND_NOW=1`. This commit comes first so the series stays bisectable. 2. **arch/sim: pass no-pie through the gcc driver in CMake.** `add_link_options(-Wl,-no-pie)` becomes `add_link_options(-no-pie)`, mirroring fd907e6ef4 for make. ## Impact - **Affected:** CMake sim builds on Linux only. The make build is unchanged. - **`-no-pie` change:** reaches only x86_64 Linux non-m32 builds, through the existing `elseif(NOT APPLE AND NOT SIM_HOST_ARM64)` guard. It does not affect macOS, arm64 or m32. - **`-z now`:** applies to all Linux hosts (`SIM_HOST_LINUX`), including aarch64. This matches make's `ifeq ($(CONFIG_HOST_LINUX),y)` exactly. - **No effect on:** hardware targets, documentation or the public API. ## Testing Hosts: - Ubuntu 20.04 container: gcc 10.5.0-1ubuntu1~20.04, binutils 2.34, cmake from pip. NuttX needs cmake 3.20 or later, and focal ships 3.16.3. - Native: Ubuntu 26.04, gcc 15.2, ld 2.46. | Environment | Build | Boot | hello | ostest | poweroff | |---|---|---|---|---|---| | focal gcc 10.5, CMake sim:nsh | pass | pass | pass | exit status 0 | pass | | focal gcc 10.5, make sim:nsh | pass | pass | pass | exit status 0 | pass | | focal gcc 10.5, CMake sim:nsh with CONFIG_SIM_M32=y | pass | pass | pass | exit status 0 | pass | | focal gcc 10.5, CMake sim:romfs | pass | pass | pass | n/a (not in defconfig) | pass | | native gcc 15, CMake sim:nsh | pass | pass | pass | exit status 0 | pass | | native gcc 15, make sim:nsh | pass | pass | pass | exit status 0 | pass | Before (master, focal, CMake configure fails in the compiler check): cc -Wl,-no-pie -Wl,--gc-sections -Wl,-Ttext-segment=0x40000000 -Wl,-z,noexecstack ... -o cmTC_... /usr/bin/ld: cannot find -lgcc_s /usr/bin/ld: cannot find -lgcc_s After (focal, final link succeeds): cc -Wl,--gc-sections -no-pie -Wl,-Ttext-segment=0x40000000 -Wl,-z,noexecstack -Wl,-z,now -T nuttx.ld @CMakeFiles/nuttx.rsp -o nuttx `./tools/checkpatch.sh -c -m -g upstream/master..HEAD` passes on both commits. Not tested: macOS, arm64 and the reporters' WSL2 setup. The existing guards keep the `-no-pie` change away from macOS and arm64. Full nsh/ostest session logs and the gdb backtrace are available on request. ## Known issues / out of scope clang 10 CMake on focal still fails with `-lgcc_s`. On master it fails at configure. With this PR, configure passes, but the `clang -r ... nuttx.rel` partial link then fails with `cannot find -lgcc_s`. That link does not use the flags changed here, so I'll handle it separately. Investigation, reproduction and verification were assisted by Claude Code. I reviewed and tested the final changes. -- 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]
