I have compiled various C and C++ source files in attempt to generate async
unwind tables for better stack trace accuracy with asynchronous signals, adding
-fasynchronous-unwind-tables to compile options. I have so far been completely
unable to come up with any example where the option would produce any
difference in the generated binaries, and unwind tables (.eh_frame)
specifically.
For example please compile the following source file foo.cc:
struct A1 { virtual ~A1(); virtual double foo(void); };
struct A2 { virtual ~A2(); virtual double bar(void); };
struct B : A1, A2 { ~B(); virtual double foo(void); };
A1::~A1() {}
double A1::foo() { return 3.14; }
A2::~A2() {}
double A2::bar() { return 2.17; }
B::~B() {}
double B::foo() { return A1::foo() + A2::bar(); }
with and without -fasynchronous-unwind-tables:
c++ -fasynchronous-unwind-tables -O2 -fPIC -W -Wall -shared -o libfooa.so
foo.cc
c++ -O2 -fPIC -W -Wall -shared -o libfoob.so foo.cc
then compare unwind information:
for x in a b; do
objdump -w -x -d libfoo$x.so > foo$x-disasm.txt
readelf -Wa --debug=frames libfoo$x.so > foo$x-info.txt
done
diff -u foo{a,b}-disasm.txt
--- fooa-disasm.txt 2010-03-23 18:28:46.000000000 +0100
+++ foob-disasm.txt 2010-03-23 18:28:46.000000000 +0100
@@ -1,6 +1,6 @@
-libfooa.so: file format elf64-x86-64
-libfooa.so
+libfoob.so: file format elf64-x86-64
+libfoob.so
architecture: i386:x86-64, flags 0x00000150:
HAS_SYMS, DYNAMIC, D_PAGED
start address 0x0000000000000c90
diff -u foo{a,b}-info.txt
(no output)
note that there is basically no difference at all, and there is
no unwind info for the thunks, disassembly has:
0000000000000df0 <_ZN1B3fooEv>:
df0: 53 push %rbx
df1: 48 89 fb mov %rdi,%rbx
df4: 48 83 c3 08 add $0x8,%rbx
df8: 48 83 ec 10 sub $0x10,%rsp
dfc: e8 7f fe ff ff callq c80 <_zn2a13fo...@plt>
e01: 48 89 df mov %rbx,%rdi
e04: f2 0f 11 44 24 08 movsd %xmm0,0x8(%rsp)
e0a: e8 41 fe ff ff callq c50 <_zn2a23ba...@plt>
e0f: f2 0f 58 44 24 08 addsd 0x8(%rsp),%xmm0
e15: f2 0f 11 44 24 08 movsd %xmm0,0x8(%rsp)
e1b: 48 83 c4 10 add $0x10,%rsp
e1f: 5b pop %rbx
e20: c3 retq
e21: 66 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 nopw
%cs:0x0(%rax,%rax,1)
0000000000000e30 <_ZThn8_N1BD1Ev>:
e30: 48 83 c7 f8 add $0xfffffffffffffff8,%rdi
e34: eb 0a jmp e40 <_ZN1BD1Ev>
e36: 66 2e 0f 1f 84 00 00 00 00 00 nopw %cs:0x0(%rax,%rax,1)
0000000000000e40 <_ZN1BD1Ev>:
e40: 48 89 6c 24 f8 mov %rbp,-0x8(%rsp)
e45: 48 89 5c 24 f0 mov %rbx,-0x10(%rsp)
e4a: 48 83 ec 18 sub $0x18,%rsp
[snip]
and unwind info has a gap at 0xe30, the thunk, and does not describe
the epilogue at address 0xe1b:
000000b0 0000001c 000000b4 FDE cie=00000000 pc=00000df0..00000e21
Augmentation data: 00 00 00 00
DW_CFA_advance_loc: 1 to 00000df1
DW_CFA_def_cfa_offset: 16
DW_CFA_offset: r3 (rbx) at cfa-16
DW_CFA_advance_loc: 11 to 00000dfc
DW_CFA_def_cfa_offset: 32
DW_CFA_nop
DW_CFA_nop
DW_CFA_nop
000000d0 0000001c 000000d4 FDE cie=00000000 pc=00000e40..00000e99
Augmentation data: 83 00 00 00
DW_CFA_advance_loc: 14 to 00000e4e
DW_CFA_def_cfa_offset: 32
DW_CFA_offset: r3 (rbx) at cfa-24
DW_CFA_offset: r6 (rbp) at cfa-16
DW_CFA_nop
DW_CFA_nop
DW_CFA_nop
DW_CFA_nop
I believe the part about missing unwind info for epilogues is a known problem
and patches have been committed to the head. Similarly lack of unwind
information for global ctors and dtors seems to be known. But how about the
missing unwind information for the thunks?
More exactly where should I expect the -fasynchronous-unwind-tables option to
make a difference? Any example code / output available?
The compiler is GCC 4.3.4 with binutils 2.19.1 on RHEL5-derived system.
$ c++ -v
Using built-in specs.
Target: x86_64-unknown-linux-gnu
Configured with: ../configure
--prefix=/build/3jm/slc5_amd64_gcc434/external/gcc/4.3.4
--enable-languages=c,c++,fortran
--with-gmp=/build/3jm/slc5_amd64_gcc434/external/gcc/4.3.4/tmp/gmp
--with-mpfr=/build/3jm/slc5_amd64_gcc434/external/gcc/4.3.4/tmp/mpfr
--enable-shared
Thread model: posix
gcc version 4.3.4 (GCC)
$ as -v /dev/null
GNU assembler version 2.19.1 (x86_64-unknown-linux-gnu) using BFD version (GNU
Binutils) 2.19.1
$ ld -v
GNU ld (GNU Binutils) 2.19.1
--
Summary: -fasynchronous-unwind-tables does not seem to do
anything
Product: gcc
Version: 4.3.4
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: lat at cern dot ch
GCC build triplet: x86_64-unknown-linux-gnu
GCC host triplet: x86_64-unknown-linux-gnu
GCC target triplet: x86_64-unknown-linux-gnu
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43498
