1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
|
[[!meta copyright="Copyright © 2010, 2012 Free Software Foundation, Inc."]]
[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license
is included in the section entitled [[GNU Free Documentation
License|/fdl]]."]]"""]]
[[!tag open_issue_glibc]]
# IRC, unknown channel, unknown date
<azeem> tschwinge: ext2fs.static: thread-cancel.c:55: hurd_thread_cancel: Assertion `! __spin_lock_locked (&ss->critical_section_lock)' failed.
<youpi> it'd be great if we could have backtraces in such case
<youpi> at least just the function names
<youpi> and in this case (static), just addresses would be enough
# IRC, freenode, #hurd, 2012-07-19
In context of the [[ext2fs_libports_reference_counting_assertion]].
<braunr> pinotree: tschwinge: do you know if our packages are built with
-rdynamic ?
<pinotree> braunr: debian's cflags don't include it, so unless the upstream
build systems do, -rdynamic is not added
<braunr> i doubt glibc' backtrace() is able to find debugging symbol files
on its own
<pinotree> what do you mean?
<braunr> the port reference bug youpi noticed is rare
<pinotree> even on linux, a program compiled with normal optimizations (eg
-O2 -g) can give just pointer values in backtrace()'s output
<braunr> core dumps are unreliable at best
[[crash_server]].
<braunr> uh, no, backtrace does give names
<braunr> but not with -fomit-frame-pointer
<braunr> unless the binary is built with -rdynamic
<braunr> at least it used to
<pinotree> not really, when being optimized some steps can be optimized
away (eg inlines)
<braunr> that's ok
<braunr> anyway, the point is i'd like a way that can give us as much
information as possible when the problem happens
<braunr> the stack trace being the most useful imo
<pinotree> do you face issues currently with backtrace()?
<braunr> not tried yet
<braunr> i guess i could make the application trap in the kernel, and fault
there, so we can attach gdb while still in the pager address space :>
<pinotree> that would imply the need for interactivity when the fault
happens, wouldn't it?
<braunr> no
<braunr> it would remain this way until someone comes, hours, days later
<braunr> pinotree: well ok, it would require interactivity, but not *when*
it happens ;p
<braunr> pinotree: right, it needs -rdynamic
## IRC, freenode, #hurd, 2012-07-21
<braunr> tschwinge: my current "approach" is to introduce an infinite loop
<braunr> it makes the faulting task mapped in often enough to use gdb
through qemu
<braunr> ... :)
<tschwinge> My understanding is that glibc already does have some mechanism
for that: I have seen it print backtraces whendetecting malloc
inconsistencies (double free and the lite).
<braunr> yes, i thought it used the backtrace functions internally though
<braunr> that is, execinfo
<braunr> but this does require -rdynamic
# GCC's libbacktrace
Introduced in GCC commit ecd3459e7bb829202601e3274411135a15c64dde.
|
