Edit report at https://bugs.php.net/bug.php?id=60078&edit=1
ID: 60078 User updated by: odou...@php.net Reported by: odou...@php.net Summary: SIGSEGV in xhprof.c Status: Open Type: Bug Package: xhprof Operating System: - PHP Version: Irrelevant Block user comment: N Private report: N New Comment: I created a patch for this (tested successfully) : https://github.com/olivierd/xhprof/commit/2e74533746bf14b0bcfc9a6fae08e1bf9b4f724b Previous Comments: ------------------------------------------------------------------------ [2011-10-19 17:45:05] odou...@php.net System is Linux 64 x64 (kernel 2.6.36) Bi CPU Intel(R) Xeon(R) CPU L5630 @ 2.13GHz I found this bug on a particular machine where some CPUs are deactivated on purpose (sorry, this is a major information but I only detected it now). Command used to deactivate a thread: echo 0 > /sys/devices/system/cpu/cpu1/online function bind_to_cpu failed for cpu 1, and now I can see why. Do you have any idea how to handle this on xhprof ? Maybe not resetting the whole hp_globals.cpu_frequencies array if bind_ failed ? ------------------------------------------------------------------------ [2011-10-19 17:39:26] scott...@php.net Any more information about the OS or version of PHP? I have this working fine on OS X with PHP 5.3 and PHP 5.4. ------------------------------------------------------------------------ [2011-10-18 13:22:27] odou...@php.net More debugging : it seems bug is happening in get_cpu_frequency() that returned 0 on line 1335 so array hp_globals.cpu_frequencies is wiped out by function clear_frequencies(); Just before, we have an error ("setaffinity: Invalid argument") thrown by line 1228, so my guess is that function bind_to_cpu() failed, and at the end program is segfaulting because this has an impact on an array. ------------------------------------------------------------------------ [2011-10-17 16:51:21] odou...@php.net Description: ------------ I'll try to be as precise as possible : This happens in a special case that can be reproduced 100%, but I cannot provide a test script (it is using 20MB of closed customer code). This happens only whith xhprof_enable(). No problem is encountered when the module is just loaded with no call to xhprof_enable() In latest clone from git (commit a6bae51236 for file xhprof.c) Program received signal SIGSEGV, Segmentation fault. 0x00007ffff3575f49 in hp_mode_shared_endfn_cb (top=0xef0210, symbol=<value optimized out>) at /usr/src/xhprof/extension/xhprof.c:1553 bt #0 hp_mode_shared_endfn_cb (top=0xef0210, symbol=<value optimized out>) at /usr/src/xhprof/extension/xhprof.c:1553 #1 0x00007ffff357609e in hp_mode_hier_endfn_cb (entries=<value optimized out>) at /usr/src/xhprof/extension/xhprof.c:1573 #2 0x00007ffff3576e66 in hp_compile_file (file_handle=<value optimized out>, type=8) at /usr/src/xhprof/extension/xhprof.c:1721 #3 0x00000000007218a4 in ?? () #4 0x000000000071f294 in execute () #5 0x00000000006faf7b in zend_execute_scripts () #6 0x00000000006b573a in php_execute_script () #7 0x0000000000772287 in main () Ok so problem is in the function "hp_mode_shared_endfn_cb" Let's try to see what is the value of each variable here : print /f hp_globals.cpu_frequencies[hp_globals.cur_cpu_id] Cannot access memory at address 0x0 ok so problem is in this expression. print hp_globals.cpu_frequencies $8 = (double *) 0x0 (gdb) print /f hp_globals.cur_cpu_id $9 = 0 Ok so I can see that hp_globals.cpu_frequencies equals NULL (right ?), and we attempt to access it as an array. I read the source code quickly, and I can see that this array should be filled at some point. Seems it is not. I made a dirty patch just to avoid the SIGSEGV, but all my timings in xhprof reports are inaccurate now. ------------------------------------------------------------------------ -- Edit this bug report at https://bugs.php.net/bug.php?id=60078&edit=1
