ptrace PEEKTEXT IO error?!

[David Lecomber <david@allinea.com>]

Dear all,

I have a multithreaded code which is misbehaving in GDB (verified on
latest CVS).  After threads are created, the target suddenly seems to
become unwriteable, and even unreadable in some bits.

(gdb) n
Single stepping until exit from function sleep,
which has no line number information.
main (argc=1, argv=0x7ffffffff3d8, environ=0x7ffffffff3e8) at hello.c:89
89              MPI_Init(&argc, &argv);
(gdb) n
[New Thread 32769 (LWP 18096)]
[New Thread 16386 (LWP 18097)]
[New Thread 32771 (LWP 18098)]
90              MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
(gdb) info threads
  4 Thread 32771 (LWP 18098)  0x00002aaaab5f8bc9 in ioctl ()
from /lib64/libc.so.6
  3 Thread 16386 (LWP 18097)  0x00002aaaab5f8bc9 in ioctl ()
from /lib64/libc.so.6
  2 Thread 32769 (LWP 18096)  0x00002aaaab5f7829 in poll ()
from /lib64/libc.so.6
* 1 Thread 16384 (LWP 17994)  main (argc=1, argv=0x7ffffffff3d8,
environ=0x7ffffffff3e8) at hello.c:90
(gdb) info address hello.c:90
No symbol "hello.c:90" in current context.
(gdb) info line
Line 90 of "hello.c" starts at address 0x402042 <main+316> and ends at
0x402056 <main+336>.
(gdb) n
Warning:
Cannot insert breakpoint 0.
Error accessing memory address 0x402056: Input/output error.
 

If I then strace the running GDB, I see the following errors (and a few
more ptrace errors).. 

ptrace(PTRACE_PEEKTEXT, 17994, 0x407480, [0]) = -1 EIO (Input/output
error)
ptrace(PTRACE_POKEDATA, 17994, 0x407480, 0xffffffffffffccff) = -1 EIO
(Input/output error)


Anyone know what could be going on here?  It's a  2.6.11  kernel, x86_64
based on Debian.

Cheers
David