[Build] OpenBLAS getarch fails on aarch64 with LSAN/ASAN/ASAN_UT/UBSAN/TSAN sanitizer modes

[heguanhui]

Version

trunk

What's Wrong?

On aarch64, building Doris BE with LSAN, ASAN, ASAN_UT, or UBSAN sanitizer modes fails.

1. LSAN/ASAN/ASAN_UT - CMake configuration failure

The OpenBLAS getarch build-time tool inherits sanitizer flags from the parent CMake context, which causes LeakSanitizer to detect memory leaks in OpenBLAS's detect() function (strdup without free on aarch64). This causes getarch to exit with non-zero code, and CMake configuration fails.

Error example (LSAN):

==641==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 480 byte(s) in 96 object(s) allocated from:
    #0 malloc
    #1 strdup
    #2 detect (getarch)
    #3 get_corename (getarch)
    #4 main (getarch)

SUMMARY: LeakSanitizer: 972 byte(s) leaked in 194 allocation(s).

Additionally, __lsan_ignore_object is only enabled under ADDRESS_SANITIZER but LEAK_SANITIZER also provides this symbol. Under LSAN mode, an empty stub is used instead, causing link errors:

undefined reference to '__lsan_ignore_object'

2. UBSAN - Compilation failure

-mcmodel=medium is x86_64-specific and unsupported on aarch64. It is hardcoded in CXX_FLAGS_UBSAN and causes compilation failure:

clang: error: unsupported argument 'medium' to option '-mcmodel=' for target 'aarch64-unknown-linux-gnu'

After fixing -mcmodel=medium, UBSAN build on aarch64 also fails due to -Wunused-macros errors (promoted to errors by -Werror):

2a. -Wno-unused-macros suppression not working for flex/bison generated files — the flag was incorrectly gated behind the -Wno-unused-but-set-variable check in be/src/exprs/CMakeLists.txt, so the suppression never took effect:

be/ut_build_UBSAN/__/gensrc/build/geo/wkt_lex.l.cpp:7:9: error: macro is not used [-Werror,-Wunused-macros]
    7 | #define FLEX_SCANNER
      | ^~~~~~~~~~~~~
be/ut_build_UBSAN/__/gensrc/build/geo/wkt_yacc.y.cpp:80:9: error: macro is not used [-Werror,-Wunused-macros]
   80 | #define YYCOPY(Dst, Src, Count) \
      | ^

2b. Unused #define PMR macro in function_string_misc.cpp:

be/src/exprs/function/function_string_misc.cpp:82:9: error: macro is not used [-Werror,-Wunused-macros]
   82 | #define PMR std::pmr
      | ^~~

2c. Unused ALIGN_CACHE_LINE and PURE macros in compiler_util.h:

be/src/common/compiler_util.h:51:9: error: macro is not used [-Werror,-Wunused-macros]
   51 | #define ALIGN_CACHE_LINE __attribute__((aligned(CACHE_LINE_SIZE)))
      | ^~~~~~~~~~~~~~~~
be/src/common/compiler_util.h:53:9: error: macro is not used [-Werror,-Wunused-macros]
   53 | #define PURE __attribute__((pure))
      | ^~~~

3. TSAN - CMake configuration failure

The OpenBLAS getarch build-time tool inherits -fsanitize=thread flag, causing getarch to produce incorrect output. This leads to SGEMM_DEFAULT_UNROLL_M and other macros being undefined when compiling getarch_2nd, resulting in CMake configuration failure.

Error example:

error: use of undeclared identifier 'SGEMM_DEFAULT_UNROLL_M'
   22 |     printf("SGEMM_UNROLL_M=%d\n", SGEMM_DEFAULT_UNROLL_M);

Note: TSAN was verified to work on aarch64 without additional changes after the sanitizer flag stripping fix.

What You Expected?

BE should build successfully with LSAN/ASAN/ASAN_UT/UBSAN/TSAN sanitizer modes on aarch64, and BE UT should run normally.

How to Reproduce?

On an aarch64 machine:

1. Build Doris BE with LSAN mode: BUILD_TYPE=LSAN bash build.sh --be
2. Or with ASAN_UT mode: BUILD_TYPE=ASAN_UT bash build.sh --be
3. Or with ASAN mode: BUILD_TYPE=ASAN bash build.sh --be
4. Or with UBSAN mode: BUILD_TYPE=UBSAN bash build.sh --be
5. Or with TSAN mode: BUILD_TYPE=TSAN bash build.sh --be
6. CMake configuration or compilation fails

Are you willing to submit PR?

Yes

[zclllyybb]

Breakwater-GitHub-Analysis-Slot: slot_83de95694b4d

Initial read: this looks like a real build-system isolation bug, not an OpenBLAS runtime issue.

What I could confirm from current upstream/master:

• Doris sets sanitizer build flags globally in be/CMakeLists.txt: LSAN uses -fsanitize=leak, ASAN / ASAN_UT use -fsanitize=address, and UBSAN currently includes -mcmodel=medium.
• The same file then copies the C++ flags into CMAKE_C_FLAGS, so these options also affect C-only configure helpers.
• be/src/storage/index/ann/cmake-protect/CMakeLists.txt adds only -w before importing contrib/openblas and contrib/faiss; it does not clear sanitizer flags or the UBSAN -mcmodel=medium flag before add_subdirectory().
• OpenBLAS CMake builds and runs the getarch helper during configure via try_compile() / execute_process().
• The checked OpenBLAS aarch64 detector (cpuid_arm64.c) allocates cpu_part / cpu_implementer with strdup() in detect() and returns through get_corename() without freeing those buffers, so an LSAN-enabled build-time getarch can report leaks exactly in this path.

So the likely failure chain is:

1. Doris sanitizer mode sets global BE C/C++ flags.
2. The ANN CMake wrapper imports OpenBLAS/FAISS without isolating third-party configure/build flags.
3. OpenBLAS builds getarch with those inherited sanitizer flags.
4. On aarch64, getarch runs CPU detection code that leaks small strdup() allocations.
5. LeakSanitizer can make the build-time helper fail during CMake configure, blocking the BE build before Doris code is built.

For UBSAN, the separate -mcmodel=medium report is also consistent with the current CMake wiring: the flag is set in Doris' UBSAN build flags and then inherited by OpenBLAS/FAISS. That flag should not be propagated into aarch64 third-party sub-builds.

Suggested fix direction:

• Isolate the ANN third-party sub-builds in be/src/storage/index/ann/cmake-protect/CMakeLists.txt: save the Doris CMAKE_C_FLAGS / CMAKE_CXX_FLAGS, strip sanitizer options/defines and x86-only options such as -mcmodel=medium before adding OpenBLAS and FAISS, then restore the original flags for Doris targets.
• Prefer this over globally disabling LSAN leak detection, because global detect_leaks=0 would hide real Doris leaks and would not address the UBSAN -mcmodel=medium leakage.
• A smaller OpenBLAS-side fix to free the aarch64 strdup() buffers would also be reasonable, but Doris should still avoid instrumenting build-time third-party configure helpers with Doris' sanitizer flags.

Useful validation artifacts for the PR:

• Full CMake configure logs for LSAN, ASAN_UT, ASAN, and UBSAN on the aarch64 host, including the printed C Flags / CXX Flags lines.
• Compiler and sanitizer runtime versions (gcc/clang, libc, OS image).
• The first fatal CMake error for the UBSAN case, especially whether it comes from OpenBLAS getarch, OpenBLAS compilation, or FAISS compilation.