如何调试“程序接收信号SIGSEGV:分段故障”错误

Question

我正在运行一个Fortran exe，我得到了错误：

 set_nml_output Echo NML values to log file only
 Trying to open namelist log dart_log.nml
 PE 0: initialize_mpi_utilities:  Running with            8  MPI processes.

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:

然后我尝试使用gdb来查找一些内容，它报告说

[New LWP 9883]
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Failed to read a valid object file image from memory.
Core was generated by `./filter'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x00002af8e021390c in netcdf::nf90_open (
    path=<error reading variable: value requires 57959040 bytes, which is more than max-value-size>, mode=0, 
    ncid=<error reading variable: Cannot access memory at address 0x7ffe439346b0>, 
    chunksize=<error reading variable: Cannot access memory at address 0x0>, 
    cache_size=<error reading variable: Cannot access memory at address 0x7ffe43934530>, 
    cache_nelems=<error reading variable: Cannot access memory at address 0x7ffe43934528>, 
    cache_preemption=<error reading variable: Cannot access memory at address 0x7ffe439345a0>, 
---Type <return> to continue, or q <return> to quit---
    comm=<error reading variable: Cannot access memory at address 0x7ffe439345a8>, 
    info=<error reading variable: Cannot access memory at address 0x7ffe439345b0>, 
    _path=<error reading variable: Cannot access memory at address 0x7ffe439345b8>) at netcdf4_file.f90:39
39  netcdf4_file.f90: No such file or directory.
(gdb) bt
#0  0x00002af8e021390c in netcdf::nf90_open (
    path=<error reading variable: value requires 57959040 bytes, which is more than max-value-size>, mode=0, 
    ncid=<error reading variable: Cannot access memory at address 0x7ffe439346b0>, 
    chunksize=<error reading variable: Cannot access memory at address 0x0>, 
    cache_size=<error reading variable: Cannot access memory at address 0x7ffe43934530>, 
    cache_nelems=<error reading variable: Cannot access memory at address 0x7ffe43934528>, 
    cache_preemption=<error reading variable: Cannot access memory at address 0x7ffe439345a0>, 
    comm=<error reading variable: Cannot access memory at address 0x7ffe439345a8>, 
    info=<error reading variable: Cannot access memory at address 0x7ffe439345b0>, 
    _path=<error reading variable: Cannot access memory at address 0x7ffe439345b8>) at netcdf4_file.f90:39
Backtrace stopped: Cannot access memory at address 0x7ffe43934598

netcdf4_file.f90:39如下图所示：

if (present(cache_size) .or. present(cache_nelems) .or. &
       present(cache_preemption)) then
     ret = nf_get_chunk_cache(size_in, nelems_in, preemption_in)
     if (ret .ne. nf90_noerr) then
        nf90_open = ret
        return
     end if
     if (present(cache_size)) then
        size_out = cache_size     #### line 39
     else
        size_out = size_in
     end if
     if (present(cache_nelems)) then
        nelems_out = cache_nelems
     else
        nelems_out = nelems_in
     end if

netcdf的版本是否与问题相关，或者应该修改某些设置？

谁能给我一些建议，如何解决这个问题，因为我不是很熟悉这些。提前谢谢。

Answer 1

分段错误真的很难调试，但下面是我要做的几件事：

使用调试符号和运行时检查进行编译。这些标志依赖于编译器，但以下是用于gfortran与英特尔Fortran的标志：

gfortran     ifort         effect
------------------------------------------------------
-g           -g            Stores the code inside the binary
-O0          -O0           Disables optimisation
-fbacktrace  -traceback    More informative stack trace
-Wall        -warn all     Enable all compile time warnings
-fcheck=all  -check all    Enable run time checks

幸运的是，当您的程序在以这种方式编译后崩溃时，将更容易推断出哪里出了问题。