Project

General

Profile

Actions

Bug #17585

closed

DWARF5 support?

Added by vo.x (Vit Ondruch) about 3 years ago. Updated almost 3 years ago.

Status:
Closed
Assignee:
-
Target version:
-
ruby -v:
ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [armv7hl-linux]
[ruby-core:102256]

Description

Fedora recently switched from DWARF4 to DWARF5 and since that time, I observe test suite errors on ppc64le:

  1) Failure:
TestBugReporter#test_bug_reporter_add [/builddir/build/BUILD/ruby-3.0.0/test/-ext-/bug_reporter/test_bug_reporter.rb:24]:
pid 1691449 killed by SIGSEGV (signal 11) (core dumped)
| -:1: [BUG] Segmentation fault at 0x000003e80019cf39
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [powerpc64le-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0021 s:0006 e:000005 EVAL   -:1 [FINISH]
| c:0001 p:0000 s:0003 E:000480 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -:1:in `<main>'
| -:1:in `kill'
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | Expected /Sample bug reporter: 12345/
   | to match
   |   "-- Control frame information -----------------------------------------------\n"+
   |   "c:0003 p:---- s:0012 e:000011 CFUNC  :kill\n"+
   |   "c:0002 p:0021 s:0006 e:000005 EVAL   -:1 [FINISH]\n"+
   |   "c:0001 p:0000 s:0003 E:000480 (none) [FINISH]\n\n"+
   |   "-- Ruby level backtrace information ----------------------------------------\n"+
   |   "-:1:in `<main>'\n"+
   |   "-:1:in `kill'\n\n"+
   |   "-- C level backtrace information -------------------------------------------\n"
   | after 4 patterns with 119 characters.
  2) Failure:
TestRubyOptions#test_segv_loaded_features [/builddir/build/BUILD/ruby-3.0.0/test/ruby/test_rubyoptions.rb:748]:
pid 1721938 killed by SIGSEGV (signal 11) (core dumped)
| -e:1: [BUG] Segmentation fault at 0x000003e8001a4652
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [powerpc64le-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0016 s:0006 e:000005 BLOCK  -e:1 [FINISH]
| c:0001 p:0000 s:0003 E:000f80 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -e:1:in `block in <main>'
| -e:1:in `kill'
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | <""> expected but was
   | <"-- C level backtrace information -------------------------------------------\n">.
  3) Failure:
TestRubyOptions#test_segv_setproctitle [/builddir/build/BUILD/ruby-3.0.0/test/ruby/test_rubyoptions.rb:762]:
pid 1721957 killed by SIGSEGV (signal 11) (core dumped)
| -e:1: [BUG] Segmentation fault at 0x000003e8001a4665
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [powerpc64le-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0029 s:0006 e:000005 EVAL   -e:1 [FINISH]
| c:0001 p:0000 s:0003 E:0015c0 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -e:1:in `<main>'
| -e:1:in `kill'
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | <""> expected but was
   | <"-- C level backtrace information -------------------------------------------\n">.
  4) Failure:
TestRubyOptions#test_segv_test [/builddir/build/BUILD/ruby-3.0.0/test/ruby/test_rubyoptions.rb:742]:
pid 1721966 killed by SIGSEGV (signal 11) (core dumped)
| -e:1: [BUG] Segmentation fault at 0x000003e8001a466e
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [powerpc64le-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0015 s:0006 e:000005 EVAL   -e:1 [FINISH]
| c:0001 p:0000 s:0003 E:001fa0 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -e:1:in `<main>'
| -e:1:in `kill'
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | <""> expected but was
   | <"-- C level backtrace information -------------------------------------------\n">.
Finished tests in 1111.355067s, 18.8689 tests/s, 2396.5689 assertions/s.
20970 tests, 2663439 assertions, 4 failures, 0 errors, 60 skips
ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [powerpc64le-linux]

And similar issues on armv7hl:

  1) Failure:
TestBugReporter#test_bug_reporter_add [/builddir/build/BUILD/ruby-3.0.0/test/-ext-/bug_reporter/test_bug_reporter.rb:24]:
pid 722 killed by SIGSEGV (signal 11) (core dumped)
| -:1: [BUG] Segmentation fault at 0x000002d2
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [armv7hl-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0021 s:0006 e:000005 EVAL   -:1 [FINISH]
| c:0001 p:0000 s:0003 E:000c98 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -:1:in `<main>'
| -:1:in `kill'
| 
| -- Machine register context ------------------------------------------------
|  "r0: 0x00000000 "r1: 0x0000000b "r2: 0x00000001 "r3: 0x00000001 "r4: 0x000002d2
|  "r5: 0xb689a028 "r6: 0x0000000b "r7: 0x00000025 "r8: 0x00000002 "r9: 0x000002d2
|  "r1: 0x00000001 "sp: 0xbe846d1c "fa: 0x00000000
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | Expected /Sample bug reporter: 12345/
   | to match
   |   "-- Control frame information -----------------------------------------------\n"+
   |   "c:0003 p:---- s:0012 e:000011 CFUNC  :kill\n"+
   |   "c:0002 p:0021 s:0006 e:000005 EVAL   -:1 [FINISH]\n"+
   |   "c:0001 p:0000 s:0003 E:000c98 (none) [FINISH]\n\n"+
   |   "-- Ruby level backtrace information ----------------------------------------\n"+
   |   "-:1:in `<main>'\n"+
   |   "-:1:in `kill'\n\n"+
   |   "-- Machine register context ------------------------------------------------\n"+
   |   " \"r0: 0x00000000 \"r1: 0x0000000b \"r2: 0x00000001 \"r3: 0x00000001 \"r4: 0x000002d2\n"+
   |   " \"r5: 0xb689a028 \"r6: 0x0000000b \"r7: 0x00000025 \"r8: 0x00000002 \"r9: 0x000002d2\n"+
   |   " \"r1: 0x00000001 \"sp: 0xbe846d1c \"fa: 0x00000000\n\n"+
   |   "-- C level backtrace information -------------------------------------------\n"
   | after 4 patterns with 107 characters.
  2) Failure:
TestRubyOptions#test_segv_loaded_features [/builddir/build/BUILD/ruby-3.0.0/test/ruby/test_rubyoptions.rb:748]:
pid 8085 killed by SIGSEGV (signal 11) (core dumped)
| -e:1: [BUG] Segmentation fault at 0x00001f95
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [armv7hl-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0016 s:0006 e:000005 BLOCK  -e:1 [FINISH]
| c:0001 p:0000 s:0003 E:000660 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -e:1:in `block in <main>'
| -e:1:in `kill'
| 
| -- Machine register context ------------------------------------------------
|  "r0: 0x00000000 "r1: 0x0000000b "r2: 0x00000001 "r3: 0x00000001 "r4: 0x00001f95
|  "r5: 0xb68e3028 "r6: 0x0000000b "r7: 0x00000025 "r8: 0x00000002 "r9: 0x00001f95
|  "r1: 0x00000001 "sp: 0xbe8c5ac4 "fa: 0x00000000
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | <""> expected but was
   | <"-- C level backtrace information -------------------------------------------\n">.
  3) Failure:
TestRubyOptions#test_segv_setproctitle [/builddir/build/BUILD/ruby-3.0.0/test/ruby/test_rubyoptions.rb:762]:
pid 8092 killed by SIGSEGV (signal 11) (core dumped)
| -e:1: [BUG] Segmentation fault at 0x00001f9c
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [armv7hl-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0029 s:0006 e:000005 EVAL   -e:1 [FINISH]
| c:0001 p:0000 s:0003 E:000550 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -e:1:in `<main>'
| -e:1:in `kill'
| 
| -- Machine register context ------------------------------------------------
|  "r0: 0x00000000 "r1: 0x0000000b "r2: 0x00000001 "r3: 0x00000001 "r4: 0x00001f9c
|  "r5: 0xb685d028 "r6: 0x0000000b "r7: 0x00000025 "r8: 0x00000002 "r9: 0x00001f9c
|  "r1: 0x00000001 "sp: 0xbef7acfc "fa: 0x00000000
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | <""> expected but was
   | <"-- C level backtrace information -------------------------------------------\n">.
  4) Failure:
TestRubyOptions#test_segv_test [/builddir/build/BUILD/ruby-3.0.0/test/ruby/test_rubyoptions.rb:742]:
pid 8101 killed by SIGSEGV (signal 11) (core dumped)
| -e:1: [BUG] Segmentation fault at 0x00001fa5
| ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [armv7hl-linux]
| 
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC  :kill
| c:0002 p:0015 s:0006 e:000005 EVAL   -e:1 [FINISH]
| c:0001 p:0000 s:0003 E:001f50 (none) [FINISH]
| 
| -- Ruby level backtrace information ----------------------------------------
| -e:1:in `<main>'
| -e:1:in `kill'
| 
| -- Machine register context ------------------------------------------------
|  "r0: 0x00000000 "r1: 0x0000000b "r2: 0x00000001 "r3: 0x00000001 "r4: 0x00001fa5
|  "r5: 0xb68bc028 "r6: 0x0000000b "r7: 0x00000025 "r8: 0x00000002 "r9: 0x00001fa5
|  "r1: 0x00000001 "sp: 0xbee96d3c "fa: 0x00000000
| 
| -- C level backtrace information -------------------------------------------
.
1. [2/2] Assertion for "stderr"
   | <""> expected but was
   | <"-- C level backtrace information -------------------------------------------\n">.
Finished tests in 1428.305544s, 14.6810 tests/s, 1866.1553 assertions/s.
20969 tests, 2665440 assertions, 4 failures, 0 errors, 56 skips
ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [armv7hl-linux]

So is it really due to DWARF5? Why is that just on two platforms?


Files

ruby-addr2line-dwarf5.patch (1.92 KB) ruby-addr2line-dwarf5.patch xtkoba (Tee KOBAYASHI), 02/05/2021 05:12 AM
ruby-dwarf5-debug_line.patch (11 KB) ruby-dwarf5-debug_line.patch xtkoba (Tee KOBAYASHI), 02/10/2021 04:38 PM
ruby-dwarf5-avoid_crash-r1.patch (6.14 KB) ruby-dwarf5-avoid_crash-r1.patch xtkoba (Tee KOBAYASHI), 02/25/2021 09:15 PM

Updated by vo.x (Vit Ondruch) about 3 years ago

Just FTR, I have reported this initially against GCC:

https://bugzilla.redhat.com/show_bug.cgi?id=1920533

Actions #2

Updated by kou (Kouhei Sutou) about 3 years ago

  • Subject changed from DWAR5 support? to DWARF5 support?

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

"A segfault in the segfault handler" occurs also on my x86_64-linux and aarch64-linux when gcc-10.2.0 -gdwarf-5 is used. A backtrace with GDB:

(gdb) run -e 'Process.kill :SEGV, $$'
Starting program: /var/tmp/ruby.build/ruby-3.0.0-x86_64-gcc.dwarf5/miniruby -e 'Process.kill :SEGV, $$'
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff7befdf7 in kill () from /lib64/libc.so.6
(gdb) cont
Continuing.
-e:1: [BUG] Segmentation fault at 0x000003e800000a68
ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-linux]

-- Control frame information -----------------------------------------------
c:0003 p:---- s:0012 e:000011 CFUNC  :kill
c:0002 p:0015 s:0006 e:000005 EVAL   -e:1 [FINISH]
c:0001 p:0000 s:0003 E:000ac0 (none) [FINISH]

-- Ruby level backtrace information ----------------------------------------
-e:1:in `<main>'
-e:1:in `kill'

-- Machine register context ------------------------------------------------
 RIP: 0x00007ffff7befdf7 RBP: 0x00007fffffffc6d0 RSP: 0x00007fffffffc678
 RAX: 0x0000000000000000 RBX: 0x000055555599e0c0 RCX: 0x00007ffff7befdf7
 RDX: 0x000000000000000b RDI: 0x0000000000000a68 RSI: 0x000000000000000b
  R8: 0x00005555557192c4  R9: 0x0000555555a08228 R10: 0xffffffffffffff3f
 R11: 0x0000000000000206 R12: 0x000055555557d2c0 R13: 0x0000000000000000
 R14: 0x0000555555adc358 R15: 0x00007ffff7788fa0 EFL: 0x0000000000000206

-- C level backtrace information -------------------------------------------

Program received signal SIGSEGV, Segmentation fault.
uleb128 (p=0x555555a965f8) at addr2line.c:198
198             unsigned char b = *(unsigned char *)(*p)++;
(gdb) bt
#0  uleb128 (p=0x555555a965f8) at addr2line.c:198
#1  0x000055555582f870 in di_read_die (reader=0x555555a965a0, die=0x555555a96390) at addr2line.c:1327
#2  0x000055555582ff4b in debug_info_read (reader=0x555555a965a0, num_traces=23, traces=0x555555998900 <trace>, lines=0x555555adf7c0, offset=0) at addr2line.c:1557
#3  0x0000555555830cc8 in fill_lines (num_traces=23, traces=0x555555998900 <trace>, check_debuglink=1, objp=0x555555a96ea0, lines=0x555555adf7c0, offset=0) at addr2line.c:1813
#4  0x0000555555831554 in rb_dump_backtrace_with_lines (num_traces=23, traces=0x555555998900 <trace>) at addr2line.c:2207
#5  0x000055555582443b in rb_print_backtrace () at vm_dump.c:760
#6  0x00005555558248b2 in rb_vm_bugreport (ctx=0x555555a97240) at vm_dump.c:998
#7  0x0000555555609c11 in rb_bug_for_fatal_signal (default_sighandler=0x0, sig=11, ctx=0x555555a97240, fmt=0x55555587718b "Segmentation fault at %p") at error.c:786
#8  0x0000555555774bc7 in sigsegv (sig=11, info=0x555555a97370, ctx=0x555555a97240) at signal.c:960
#9  <signal handler called>
#10 0x00007ffff7befdf7 in kill () from /lib64/libc.so.6
#11 0x000055555577439d in rb_f_kill (argc=2, argv=0x7ffff7689048) at signal.c:481
(...snip...)
(gdb) p *p
$1 = 0x401 <error: Cannot access memory at address 0x401>

I propose the following change as a workaround:

--- a/addr2line.c
+++ b/addr2line.c
@@ -1471,7 +1471,7 @@
     }
     reader->cu_end = reader->p + unit_length;
     version = read_uint16(&reader->p);
-    if (version > 5) {
+    if (version > 4) {
         return -1;
     }
     else if (version == 5) {

Updated by mame (Yusuke Endoh) about 3 years ago

@xtkoba (Tee KOBAYASHI) Awesome!

@vo.x (Vit Ondruch) Can you check the patch on your Fedora machines?

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

It seems that GCC emits version 3 .debug_line sections even when -gdwarf-5 is provided, in contrast to Clang/LLVM emitting version 5, which addr2line.c simply skips currently.

Not causing crashes with DWARF-5 is one thing, and supporting DWARF-5 is another. It will require a fair amount of work to fully support version 5 .debug_line (or other) sections.

I tried making a minimal modification to addr2line.c so that it prints file:line infomation with Clang/LLVM's DWARF-5 sections. A patch is attached for that. It works with my x86_64 Linux + Clang 11.0.1, but probably will not with others.

Updated by vo.x (Vit Ondruch) about 3 years ago

xtkoba (Tee KOBAYASHI) wrote in #note-5:

A patch is attached to avoid segfaults with DWARF5.

I can't see any change with this patch. It fails on ppc64le and armv7hl. Will try the other one.

Updated by vo.x (Vit Ondruch) about 3 years ago

With the later patch, it crashes even on my x86_64 :(

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

Excuse me, but the patch in #note-6 is just a scratch and is not expected to work as it is.

And it is highly probable that there exist new attributes from DWARF-5 that are not covered by the patch in #note-5.

Once I thought it would be a workaround to add -gdwarf-4 to debugflags, but it will be insufficient and will still cause a crash when a shared library linked from ruby (e.g. libc.so) uses DWARF-5.

Now might be the time to consider moving from Ruby's own addr2line to Ian Lance Taylor's libbacktrace:
https://github.com/ianlancetaylor/libbacktrace

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

For the libbacktrace library, I made a feature request (Feature #17638).

What I feel strange is that the failures were reported to happen only on two platforms, contrary to my experiment in which a crash occurred with x86_64-linux + gcc-10.2.0 -gdwarf-5 -ggdb3.

It might not be improbable that in the reported situations crashes happened within backtrace(3). I know that it can cause a crash on QEMU-user-emulated ppc64le Linux. It is true that if so the version of DWARF would be irrelevant, but it might be worth investigating using GDB or so exactly where a crash happens.

Actions #11

Updated by vo.x (Vit Ondruch) about 3 years ago

Just FTR, there are long time outstanding platform specific issues with SIGSEV handler such as #16492 so this might be just coincidence. Not sure though

Actions #12

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

With the patch in #note-5 applied and compiled using gcc-10.2.0 -O0 -gdwarf-5 -ggdb3, addr2line.c appears to print incorrect information on some Linux environments.

Version:

ruby 3.1.0dev (2021-02-18 master 07ab172ebe) (patched)

Command:

./miniruby -e 'Process.kill :SEGV, $$'

The top 3 lines from C level backtrace information for each environment:

[aarch64-linux]
./miniruby(rb_print_backtrace+0x18) [0x55002dc6b4] vm_dump.c:822
./miniruby(rb_vm_bugreport+0x124) [0x55002dcc6c] vm_dump.c:1074
./miniruby(rb_bug_for_fatal_signal+0xf8) [0x55000b8504] error.c:801

[armv7a-linux-eabi]
./miniruby(rb_print_backtrace+0x20) [0x40348d80] vm_dump.c:824
./miniruby(rb_vm_bugreport+0x13c) [0x4034928c] vm_dump.c:1083
./miniruby(rb_bug_for_fatal_signal+0xb0) [0x400c02b4] error.c:848

[powerpc64le-linux]
./miniruby(rb_print_backtrace+0x24) [0x400039aa50] vm_dump.c:1024
./miniruby(rb_vm_bugreport+0x15c) [0x400039abf0] vm_dump.c:1074
./miniruby(rb_bug_for_fatal_signal+0xc4) [0x40000e0258] error.c:801

[riscv64-linux]
./miniruby(rb_print_backtrace+0x1c) [0x40002a9be8] vm_dump.c:770
./miniruby(rb_vm_bugreport+0x110) [0x40002aa154] vm_dump.c:1054
./miniruby(rb_bug_for_fatal_signal+0x9e) [0x40000a7afc] error.c:801

[s390x-linux]
./miniruby(rb_print_backtrace+0x1e) [0x40003dedee] vm_dump.c:770
./miniruby(rb_vm_bugreport+0x170) [0x40003def90] vm_dump.c:1054
./miniruby(rb_bug_for_fatal_signal+0x10a) [0x40000f2792] error.c:801

[x86_64-linux]
./miniruby(rb_print_backtrace+0x19) [0x560766410e42] vm_dump.c:770
./miniruby(rb_vm_bugreport+0x14b) [0x56076641135c] vm_dump.c:1054
./miniruby(rb_bug_for_fatal_signal+0x123) [0x5607661f4084] error.c:801

In the above outputs, the line numbers seem to be correct on riscv64, s390x and x86_64, but not on aarch64, armv7a or powerpc64le. In the latter three architectures, it also happens that some function names are not printed (in another case of segfault):

[aarch64-linux]
(...)
./miniruby(rb_load_with_builtin_functions+0x1c) [0x550002ba04] mini_builtin.c:47
./miniruby(0x55000ee0f0) [0x55000ee0f0]
./miniruby(rb_call_builtin_inits+0xc) [0x55000fa5a4] inits.c:88
(...)

[armv7a-linux-eabi]
(...)
./miniruby(rb_builtin_ast+0x7c8) [0x4001c210] miniprelude.c:3022
./miniruby(0x4001c568) [0x4001c568]
./miniruby(0x4001c65c) [0x4001c65c]
./miniruby(0x40216e88) [0x40216e88]
./miniruby(rb_call_builtin_inits+0x10) [0x4010d384] inits.c:89
(...)

[powerpc64le-linux]
(...)
./miniruby(rb_load_with_builtin_functions+0x28) [0x400002b57c] mini_builtin.c:47
./miniruby(0x4000124b58) [0x4000124b58]
./miniruby(rb_call_builtin_inits+0x18) [0x4000135c18] inits.c:88
(...)

This shows that the patch in #note-5 is nothing but incomplete, and SHOULD NOT BE APPLIED as it is. I am sorry for my poor work.

The patch in #note-3 still remains to be the best workaround by me for now, except for using libbacktrace.

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

A (revised) patch is attached to avoid segfaults with GCC's DWARF 5.

In summary, there are three changes:

(1) correct the interpretation of DW_LNS_advance_pc statements when the minimum instruction length is not equal to 1 (which is the case for aarch64, arm and powerpc64le),

(2) skip DW_FORM_implicit_const attributes properly, and

(3) read and use the .debug_rnglists section if available.

When applied together with the patch for Bug #17656, this patch will make addr2line.c correctly extract function+offset and file:line information from GCC's DWARF 5 sections, except that some "inlined by" information is not shown.

Note that in the above changes, (3) is minimal and only DW_RLE_base_address and DW_RLE_offset_pair are interpreted, because other range list entries seem unnecessary for now. It is highly probable that other entries become necessary in the future.

I hope that the workaround in #note-3 would be no longer needed.

#note-5 is deleted to avoid confusion.

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

A minor mod to the patch in #note-13 (due to an issue similar to Bug #17645):

--- a/addr2line.c
+++ b/addr2line.c
@@ -1463,7 +1463,7 @@
                     }
                     break;
                   case DW_RLE_base_address:
-                    base_address = read_dw_form_addr(reader, &p);
+                    base_address = (uintptr_t)read_dw_form_addr(reader, &p);
                     break;
                   case DW_RLE_start_end:
                     read_dw_form_addr(reader, &p);

Updated by vo.x (Vit Ondruch) about 3 years ago

xtkoba (Tee KOBAYASHI) wrote in #note-13:

A (revised) patch is attached to avoid segfaults with GCC's DWARF 5.

I have tested this patch (without any others you have referenced) and after 10 build in Fedora build system (that is 60 builds on various platforms supported by Fedora) and it seems to fix the original issue. I think I'll keep it applied in Fedora, unless you foresee some further issues.

Thx a lot for working on this!

Updated by xtkoba (Tee KOBAYASHI) about 3 years ago

Glad to hear that :-)

What I am still (probably too) anxious about is that the current addr2line.c calls a lot of async-signal unsafe functions in it. It is true that practically they seem to cause no problem 99% of the time. But theoretically any chaos can happen as long as these functions are used.

Updated by mame (Yusuke Endoh) about 3 years ago

Thank you @xtkoba (Tee KOBAYASHI) and @vo.x (Vit Ondruch) !

I've created a PR https://github.com/ruby/ruby/pull/4240 to test the patch on the GitHub Actions CI. If @xtkoba (Tee KOBAYASHI) can send a PR yourself, please ignore my PR and create yours (with commit message).

Actions #18

Updated by mame (Yusuke Endoh) about 3 years ago

  • Status changed from Open to Closed

Applied in changeset git|9e5105ca451d6d38eb2d03a2ffc904074f0333b9.


Support GCC's DWARF 5 [Bug #17585] (#4240)

Co-Authored-By: xtkoba (Tee KOBAYASHI)

Updated by vo.x (Vit Ondruch) almost 3 years ago

  • Backport changed from 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN to 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: REQUIRED

Setting the backport flag, because it would be useful for Fedora/RHEL and we could drop one patch from our package. However, I understand this might be considered as a new feature and therefore not backported.

Updated by xtkoba (Tee KOBAYASHI) almost 3 years ago

GCC 11 is released yesterday, and in Changes, New Features, and Fixes it is said that:

To take full advantage of DWARF version 5 GCC needs to be build against binutils version 2.35.2 or higher. When GCC is build against earlier versions of binutils GCC will still emit DWARF version 5 for most debuginfo data, but will generate version 4 debug line tables (even when explicitly given `-gdwarf-5`).

Though I have not yet tested the released version of GCC 11, there will be (at least) three cases according to the version of GCC and binutils when GCC is supplied with the -gdwarf-5 option (which is the default for GCC 11).

  1. GCC 10 (+ binutils with any version?): version 5 .debug_info, version 3 .debug_line (as noted in #note-6)
  • This case is what I aimed at in the main issue.
  1. GCC 11 + binutils with version >= 2.35.2: version 5 .debug_info, version 5 .debug_line
  • As I wrote in #note-6, version 5 .debug_line is just skipped in the current addr2line.c, and so I suppose there will be no new issue in this case other than that filename:lineno information is not shown in C backtrace.
  1. GCC 11 + binutils with version < 2.35.2: version 5 .debug_info, version <= 4 .debug_line
  • The binutils in my environment is up to date, and so I will not be able to test this case. Hopefully the same as case 1.

To summarize, I hope there will be no new issue with the released version of GCC 11. Anyway, feel free to reopen this ticket whenever a segfault is observed in addr2line.c with DWARF 5.

Updated by vo.x (Vit Ondruch) almost 3 years ago

Fedora Rawhide is using GCC11 since ~beginning of the year and I am not aware of any issues. Actually, Fedora 34 was released yesterday, so should there been any issues, we will soon find out :)

Updated by nagachika (Tomoyuki Chikanaga) almost 3 years ago

Honestly, I cannot judge did this change keep backward compatibility.
I mean, is the current master works fine with -gdwarf-4 option or old GCC?

Actions

Also available in: Atom PDF

Like0
Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0