This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[RFA/alpha-osf] another next frame confusion in signal unwinder?


Hello,

Using the following C program:

        #include <signal.h>
        #include <stdio.h>
        #include <unistd.h>
        
        void blow_up (int x)
        {
          *((int*) (x|1)) = 3;
        }
        
        
        void handler (int sig) {
          blow_up (sig);
          printf ("This is a test.\n");
        }
        
        main ()
        {
          signal (SIGTERM, handler);
        
          kill (getpid (), SIGTERM);
        }

Compiled on Tru64 5.1a with GCC:

        % gcc -g -o cause_signal cause_signal.c

The following transcript reveals that GDB is unable to unwind
the stack past the signal handler frame:

        (gdb) run
        Starting program: /usr/prague.a/brobecke/53-63/ex/cause_signal

        Program received signal SIGTERM, Terminated.
        0x000003ff800e8a78 in kill () from /usr/shlib/libc.so
        (gdb) cont
        Continuing.

        Program received signal SIGSEGV, Segmentation fault.
        0x0000000120001248 in blow_up (x=15) at cause_signal.c:7
        7         *((int*) (x|1)) = 3;
        (gdb) bt
        #0  0x0000000120001248 in blow_up (x=15) at cause_signal.c:7
        #1  0x0000000120001290 in handler (sig=15) at cause_signal.c:12
        #2  <signal handler called>
        #3  0xa79db5f0b43c0000 in ?? ()
        warning: Hit heuristic-fence-post without finding
        warning: enclosing function for address 0xa79db5f0b43c0000
        This warning occurs if you are debugging a function without any symbols
        (for example, in a stripped executable).  In that case, you may wish to
        increase the size of the search with the `set heuristic-fence-post'
+command.

        Otherwise, you told GDB there was a function where there isn't one, or
        (more likely) you have encountered a bug in GDB.
        #4  0x2c9000006bfa8001 in ?? ()
        warning: Hit heuristic-fence-post without finding
        warning: enclosing function for address 0x2c9000006bfa8001
        warning: Previous frame identical to this frame (corrupt stack?)

The expected output for the "bt" command is:

        (gdb) bt
        #0  0x0000000120001248 in blow_up (x=15) at cause_signal.c:7
        #1  0x0000000120001290 in handler (sig=15) at cause_signal.c:12
        #2  <signal handler called>
        #3  0x000003ff800e8a78 in kill () from /usr/shlib/libc.so
        #4  0x0000000120001318 in main () at cause_signal.c:20

When trying to compute the frame previous to the signal handler one,
GDB computes the frame cache, and as a consequence ends up calling

        static struct alpha_sigtramp_unwind_cache *
        alpha_sigtramp_frame_unwind_cache (struct frame_info *next_frame,
                                           void **this_prologue_cache)
        {   
          [...]
          info->sigcontext_addr = tdep->sigcontext_addr (next_frame);
        
          return info;
        }   

This results in a call to:

        static CORE_ADDR
        alpha_osf1_sigcontext_addr (struct frame_info *frame)
        {
          struct frame_info *next_frame = get_next_frame (frame);
        
          if (next_frame != NULL)
            return (read_memory_integer (get_frame_base (next_frame), 8));
          else
            return (read_memory_integer (get_frame_base (frame), 8));
        }

It looks to me that we already have the next frame, so computing
the next frame of that frame in alpha_osf1_sigcontext_addr() is
making us using the wrong frame when getting the base address for
the signal context area. So I changed this function to:

        static CORE_ADDR
        alpha_osf1_sigcontext_addr (struct frame_info *next_frame)
        {      
          return (read_memory_integer (get_frame_base (next_frame), 8));
        } 

This fixes the problem above, and I obtain the expected callstack.

2004-12-01  Joel Brobecker  <brobecker@gnat.com>

        * alpha-osf1-tdep.c (alpha_osf1_sigcontext_addr): Change
        parameter name to make it clear that we already have a next
        frame.  Return the sigcontext from that next frame instead
        of the frame following it.
        Fixes [DB30-017].

Tested on alpha-tru64 5.1a. No regression.

OK to apply?
-- 
Joel

Attachment: signal.diff
Description: Text document


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]