On 12/03/2013, at 7:57 PM, Alan M. Carroll <a...@network-geographics.com> wrote:
> Unfortunately this commit breaks my build and executable. FC 18.
> --enable-debug and --enable-wccp.
>
> Building:
>
> autoreconf -i
>
> fails.
>
> [amc@yuna ats]$ autoreconf -i
> configure.ac:1115: warning: AC_LANG_CONFTEST: no AC_LANG_SOURCE call detected
> in body
> ../../lib/autoconf/lang.m4:194: AC_LANG_CONFTEST is expanded from...
> ../../lib/autoconf/general.m4:2730: _AC_RUN_IFELSE is expanded from...
> ../../lib/m4sugar/m4sh.m4:606: AS_IF is expanded from...
> ../../lib/autoconf/general.m4:2749: AC_RUN_IFELSE is expanded from...
> configure.ac:1115: the top level
> configure.ac:1115: warning: AC_LANG_CONFTEST: no AC_LANG_SOURCE call detected
> in body
>
> When I tried to run it with just a normal rebuild (I tried autoreconf in case
> there was a configuration problem), I get
>
> Program received signal SIGSEGV, Segmentation fault.
> [Switching to Thread 0x7ffff67fc700 (LWP 14023)]
> 0x00007ffff7dd5a35 in ink_atomic_cas<__int128> (mem=0x7ffff6d05288,
> prev=0x00000000000000000000000000000001,
> next=0x000000000000000000007fffe0015dc1) at ink_atomic.h:153
> 153 return __sync_bool_compare_and_swap(mem, prev, next);
> (gdb) up
> #1 0x00007ffff7dd57b9 in ink_atomiclist_push (l=0x7ffff6d05288,
> item=0x7fffe0015dc0) at ink_queue.cc:481
> 481 result = ink_atomic_cas((__int128_t*) & l->head, head.data,
> item_pair.data);
> (gdb) up
> #2 0x00000000006c62e9 in AtomicSLL<UnixNetVConnection,
> UnixNetVConnection::Link_read_enable_link>::push (this=0x7ffff6d05288,
> c=0x7fffe0015dc0)
> at ../../lib/ts/List.h:477
> 477 void push(C * c) { ink_atomiclist_push(&al, c); }
> (gdb) up
> #3 0x00000000006c40ef in UnixNetVConnection::reenable (this=0x7fffe0015dc0,
> vio=0x7fffe0015ed0) at UnixNetVConnection.cc:721
> 721 nh->read_enable_list.push(this);
>
> I don't see why it segfaults - in the debugger in frame 0 I can look at mem
> and *mem without a problem. Does the 128 int have to be naturally aligned?
I checked the Intel manual ... cmpxchg16b needs 16 byte alignment.
>
> Sunday, March 10, 2013, 10:46:53 AM, you wrote:
>
>> I'm going to commit this patch today unless there are any last minute
>> objections.
>
>> Brian
>
