Cavium Octeon III处理器运行Valgrind触发Bus Error问题求助

交叉编译配置 #

我尝试在Cavium Octeon III处理器上运行Valgrind，使用以下命令交叉编译Valgrind 3.19.0：

export CROSS_COMPILE=/opt/CAVIUMsdk312/OCTEON-SDK/tools/bin/mips64-octeon-linux-gnu-

./configure --target=mips64-octeon-linux-gnu \
          --host=mips64-octeon-linux-gnu \
          --prefix=/home/prateeka/packages/valgrind-3.19.0/install \
          CFLAGS=-static \
          CC=${CROSS_COMPILE}gcc \
          CPP=${CROSS_COMPILE}cpp \
          CXX=${CROSS_COMPILE}g++ \
          LD=${CROSS_COMPILE}ld \
          AR=${CROSS_COMPILE}ar

运行错误情况 #

我编写了一个带有故意内存泄漏的小型应用程序mem-leak测试Valgrind，但运行时触发Bus Error，命令行日志如下：

-bash-4.3# export VALGRIND_LIB=/etc/valgrind-bins/libexec/valgrind  
-bash-4.3# ./bin/valgrind ./mem-leak  
==20887== Memcheck, a memory error detector  
==20887== Copyright (C) 2002-2022, and GNU GPL'd, by Julian Seward et al.  
==20887== Using Valgrind-3.19.0 and LibVEX; rerun with -h for copyright info  
==20887== Command: ./mem-leak  
==20887==  
==20887== Invalid read of size 8  
==20887==    at 0x40BEEE4: __ctype_init (in /lib64/libc-2.16.so)  
==20887==    by 0x40AFC8C: _init (in /lib64/libc-2.16.so)  
==20887==    by 0x40129D8: call_init (in /lib64/ld-2.16.so)  
==20887==    by 0x4012BC8: _dl_init (in /lib64/ld-2.16.so)  
==20887==    by 0x400328C: _dl_start_user (in /lib64/ld-2.16.so)  
==20887==  Address 0xffffffffffff9000 is not stack'd, malloc'd or (recently) free'd  
==20887==  
==20887==  
==20887== Process terminating with default action of signal 10 (SIGBUS)  
==20887==    at 0x40BEEE4: __ctype_init (in /lib64/libc-2.16.so)  
==20887==    by 0x40AFC8C: _init (in /lib64/libc-2.16.so)  
==20887==    by 0x40129D8: call_init (in /lib64/ld-2.16.so)  
==20887==    by 0x4012BC8: _dl_init (in /lib64/ld-2.16.so)  
==20887==    by 0x400328C: _dl_start_user (in /lib64/ld-2.16.so)  
==20887==  
==20887== HEAP SUMMARY:  
==20887==     in use at exit: 0 bytes in 0 blocks  
==20887==   total heap usage: 0 allocs, 0 frees, 0 bytes allocated  
==20887==  
==20887== All heap blocks were freed -- no leaks are possible  
==20887==  
==20887== For lists of detected and suppressed errors, rerun with: -s  
==20887== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)  
Bus error  

我检查了dmesg内核消息缓冲区，未发现非法虚拟地址访问相关错误，恳请提供调试建议。

调试建议 #

• 验证Valgrind平台支持：Valgrind 3.19对Octeon III的特殊指令集扩展（如多媒体/加密指令）支持可能存在缺陷，尝试升级到Valgrind最新稳定版本，或确认官方文档中MIPS64 Octeon平台的支持状态。
• 调整编译策略：移除CFLAGS=-static选项，改用动态编译Valgrind。静态编译会强制绑定特定版本libc，可能与Valgrind的内存沙箱机制冲突，导致__ctype_init这类初始化函数的内存访问被误判。
• 核对工具链兼容性：确认Cavium SDK提供的mips64-octeon-linux-gnu-*工具链版本与Valgrind 3.19的要求匹配，避免因工具链版本差异导致二进制兼容性问题。
• 启用详细调试日志：运行Valgrind时添加参数-v --trace-syscalls=yes，获取初始化阶段的系统调用、内存映射等详细日志，分析__ctype_init访问0xffffffffffff9000地址的具体原因。
• 简化测试程序：编写仅包含malloc不释放的最小测试程序，排除原测试程序的其他逻辑干扰，确认问题是否源于Valgrind本身。
• 自定义抑制规则：尝试添加Valgrind抑制规则屏蔽__ctype_init的错误，或查看Valgrind源码中MIPS平台的libc初始化处理逻辑，确认是否存在未适配Octeon平台的代码。

内容的提问来源于stack exchange，提问作者theshowstopper