https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84439
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |iant at google dot com
Version|4.8.5 |7.3.1
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. With glibc 2.22 I get into
99999 calls to backtrace()
Could not print backtrace: mmap: Cannot allocate memory
#0 0x7f7a4717268a
#1 0x4008c8
#2 0x40082f
#3 0x400814
#4 0x4007f9
#5 0x4007de
#6 0x400907
#7 0x40094a
#8 0x7f7a466816d4
#9 0x400718
#10 0xffffffffffffffff
***** calls to backtrace()
Could not print backtrace: mmap: Cannot allocate memory
#0 0x7f7a4717268a
#1 0x4008c8
#2 0x40082f
#3 0x400814
#4 0x4007f9
#5 0x4007de
#6 0x400907
#7 0x40094a
#8 0x7f7a466816d4
#9 0x400718
#10 0xffffffffffffffff
while with glibc 2.26 I get
...
Could not print backtrace: mmap: Cannot allocate memory
#0 0x7f928d0a750a
#1 0x4008a8
#2 0x40080f
#3 0x4007f4
#4 0x4007d9
#5 0x4007be
#6 0x4008e4
#7 0x400927
#8 0x7f928c54ff49
#9 0x4006f9
#10 0xffffffffffffffff
4857 calls to backtrace()
Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
I think the issue lies within libgfortran show_backtrace which does
lbstate = backtrace_create_state (NULL, __gthread_active_p (),
error_callback, NULL);
for each call to backtrace (). There doesn't seem to be a corresponding
backtrace_free_state API so we leak whatever is allocated here. Ian?
I suppose a workaround would be to make lbstate static and only initialize
it if NULL.
Index: libgfortran/runtime/backtrace.c
===================================================================
--- libgfortran/runtime/backtrace.c (revision 258030)
+++ libgfortran/runtime/backtrace.c (working copy)
@@ -135,11 +135,12 @@ full_callback (void *data, uintptr_t pc,
void
show_backtrace (bool in_signal_handler)
{
- struct backtrace_state *lbstate;
+ static struct backtrace_state *lbstate;
struct mystate state = { 0, false, in_signal_handler };
- lbstate = backtrace_create_state (NULL, __gthread_active_p (),
- error_callback, NULL);
+ if (!lbstate)
+ lbstate = backtrace_create_state (NULL, __gthread_active_p (),
+ error_callback, NULL);
if (lbstate == NULL)
return;
note the above is wrong when the __gthread_active_p () state changes.
Ian, I suppose there should be an API to deallocate the state?
