[Slackbuilds-users] Segmentation Fault help needed

[414N]

Hello Antonio!

TL;DR: the error seems to be caused by the MAP_FIXED mmap flag used in 
bbccon.c:1728.
You can find a patch attached that solves that issue, but I cannot 
guarantee that the program will still work as intended with it. The 
program seems to really want to allocate memory at a specific address 
and removing that flag breaks that assumption.

The steps I followed to determine the source of the issue:

  1. modify the src/console/makefile file to change every optimization 
flag (i.e. gcc "-O2" or "-Os") to debug flags ("-O0 -g" for gcc, "-g" 
for nasm) and removing any stripping instruction ("-Wl,-s")
  2. recompile and run via gdb:
    cd console/linux
    make -f makefile
    gdb ./bbcbasic
    (gdb) run
    Starting program: /tmp/BBCSDL/console/linux/bbcbasic
    [Thread debugging using libthread_db enabled]
    Using host libthread_db library "/lib64/libthread_db.so.1".

    Program received signal SIGSEGV, Segmentation fault.
    0x000
    0x00000001003fffff in ?? ()
    (gdb) bt
    #0  0x00000001003fffff in ?? ()
    #1  0x0000000000000001 in ?? ()
    #2  0x0000000103ae75f6 in ?? ()
    #3  0x0000000000000000 in ?? ()
    (gdb)
  3. as can be seen from the gdb backtrace (bt) output, no useful 
information can be discerned
  4. another way to catch the problem is by stepping one instruction at 
a time from the main() function entrypoint (hoping that the error spawns 
early in the program code ^_^), so, back in gdb (starting another 
session)[1]:
    (gdb) run
    Starting program: /tmp/BBCSDL/console/linux/bbcbasic
    [Thread debugging using libthread_db enabled]
    Using host libthread_db library "/lib64/libthread_db.so.1".

    Breakpoint 1, main (argc=1, argv=0x7fffffffdf68) at ../../src 
/bbccon.c:1683
    1683    int exitcode = 0 ;
    (gdb) next
    1684    void *immediate = NULL ;
    (gdb)
    1715    pthread_t hThread = 0 ;
    (gdb)
    1717            platform = 1 ;
    (gdb)
    1719            void *base = NULL ;
    (gdb)
    1721            while ((MaximumRAM >= MINIMUM_RAM) && (NULL == (base 
= mymap (MaximumRAM))))
    (gdb)
    1726            if (base != NULL)
    (gdb)
    1727                    userRAM = mmap (base, MaximumRAM, PROT_EXEC 
| PROT_READ | PROT_WRITE,
    (gdb)

    Program received signal SIGSEGV, Segmentation fault.
    0x00000001003fffff in ?? ()
    (gdb)

As can be seen, the SIGSEGV occurred after trying to step over the mmap 
function call in bbccon.c at line 1727. By inspecting it in the source 
file and by looking at the "mmap" man page, one can see that:

   - the MAP_FIXED flag needs some care to be correctly used;
   - the SIGSEGV signal can be the result of "Attempted write into a 
region mapped as read-only."

I think this means that the "base" address returned by the "mymap" 
function does not work well in the following "mmap" call with the 
MAP_FIXED flag, probably due to some memory management kernel settings 
that are different between Slackware and Debian, like 
"/proc/sys/vm/mmap_min_addr" (if I remember correctly).

Hope to be helpful!

-- 
Alan Alberghini

SBo clone: https://github.com/414n/slackbuilds.org


[1]: after the first "next" instruction no command is echoed in the 
following (gdb) prompts because I just pressed ENTER on the keyboard to 
repeat the last command entered



-------------- next part --------------
A non-text attachment was scrubbed...
Name: bbcsdl-no-SIGSEGV.patch
Type: text/x-patch
Size: 330 bytes
Desc: not available
URL: <http://lists.slackbuilds.org/pipermail/slackbuilds-users/attachments/20240318/febea89a/attachment.bin>