[erlang-questions] Erlang in Debian on m68k architecture

[Mikael Pettersson]

Sergei Golovan writes:
 > On 1/25/07, Mikael Pettersson <mikpe@REDACTED> wrote:
 > > Definitely smells like a toolchain bug to me. In this case, since you've
 > > enabled threads, erts_mtx_t is a properly declared struct type containing
 > > among other things a pthread_mutex_t, an int, and a couple of pointers.
 > > I don't remember how strict M68K is about alignment, but presumably it wants
 > > non-arbitrary alignment for word-sized fields, especially perhaps if they're
 > > used for atomic accesses.
 > >
 > > The error could be in gcc, gas, ld, or even libc.
 > >
 > > As GDB session on the core dump should tell us more about the location
 > > of the crash, the failing instruction, and what data it tried to access.
 > 
 > Here are two backtraces, without and with DEBUG defined. If you need
 > some additional info, I'll try to give it also
...
 > With debug.
 > -----------------------------------------------------
 > (gdb) bt
 > #0  0xc0103522 in kill () from /lib/libc.so.6
 > #1  0xc004a87e in pthread_kill () from /lib/libpthread.so.0
 > #2  0xc004a8b6 in raise () from /lib/libpthread.so.0
 > #3  0xc0103380 in raise () from /lib/libc.so.6
 > #4  0xc0104266 in abort () from /lib/libc.so.6
 > #5  0x8009b534 in et_abort (expr=0x800fb643 "_is_aligned(x)",
 > file=0x800fa37f "beam/erl_lock_check.c", line=1018) at
 > beam/erl_term.c:45
 > #6  0x8009c1b4 in checked_make_boxed (x=0x0, file=0x800fa37f
 > "beam/erl_lock_check.c", line=1018) at beam/erl_term.c:123
 > #7  0x80092bca in erts_lc_init_lock (lck=0x8015abe2, name=0x800f0178
 > "binary_alloc", flags=0) at beam/erl_lock_check.c:1018
 > #8  0x800504c0 in erts_mtx_init (mtx=0x8015abba, name=0x800f0178
 > "binary_alloc") at beam/erl_threads.h:249
 > #9  0x800731e6 in erts_init_binary () at beam/binary.c:40
 > #10 0x8002d85c in erl_init () at beam/erl_init.c:211
 > #11 0x8002e568 in erl_start (argc=37, argv=0xeffff58c) at beam/erl_init.c:907
 > #12 0x8001a6fe in main (argc=37, argv=0xeffff584) at sys/unix/erl_main.c:28
 > -----------------------------------------------------

What's happening is that an embedded Eterm field in the mutex,
placed there and used by the lock checker, isn't word-aligned,
which triggers an abort() in make_boxed(). In your non-debug
GDB trace this error went unnoticed but later on _seriously_
broke the GC.

The problem is still one of alignment. It seems that your M68K
toolchain doesn't align

struct { int x; } var;

to a 32-bit boundary. Whether that is the bug, or whether the
lock checker's alignment-of-static-data assumption is the bug,
I cannot tell until someone sends me a pointer to the M68K Linux
(or SVR4) ABI document.

/Mikael