https://bugs.kde.org/show_bug.cgi?id=492663
Bug ID: 492663
Summary: Valgrind ignores debug info for some binaries
Classification: Developer tools
Product: valgrind
Version: 3.24 GIT
Platform: Other
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: general
Assignee: jsew...@acm.org
Reporter: m...@glandium.org
Target Milestone: ---
Created attachment 173332
--> https://bugs.kde.org/attachment.cgi?id=173332&action=edit
Test case archive (.tar.bz2)
SUMMARY
Some binary layouts make it so that valgrind ignores their debug info.
STEPS TO REPRODUCE
1. Download and unpack the attached test case archive
2. run `valgrind --leak-check=full testcase/firefox-bad` on Linux x86_64.
OBSERVED RESULT
A leak is reported with the second frame non-symbolicated
EXPECTED RESULT
The second frame should be symbolicated
ADDITIONAL INFORMATION
The archive also contains a `firefox-good` binary that does get the second
frame symbolicated.
The only difference in the source code between `firefox-bad` and `firefox-good`
is the content and length of a single string. It's enough to push the
boundaries of PT_LOADs to a place that makes `check_elf_and_get_rw_loads` not
agree with how coregrind handles segments-merges on mmap, leading
`di_notify_mmap` to bail with `no dinfo loaded firefox-bad (no rx or no rw
mapping)` on the last firefox-bad segment, and the debug info not being loaded,
ultimately leading to the lack of symbolication.
The following patch "fixes" this problem:
```
diff --git a/coregrind/m_debuginfo/readelf.c b/coregrind/m_debuginfo/readelf.c
index 735f83044..326a1c59d 100644
--- a/coregrind/m_debuginfo/readelf.c
+++ b/coregrind/m_debuginfo/readelf.c
@@ -3954,7 +3954,6 @@ Bool ML_(check_elf_and_get_rw_loads) ( Int fd, const
HChar* filename, Int * rw_l
* with or without rounding up.
* */
if (previous_rw_a_phdr.p_memsz > 0 &&
- ehdr_m.e_type == ET_EXEC &&
previous_rw_a_phdr.p_vaddr + previous_rw_a_phdr.p_filesz
== a_phdr.p_vaddr)
{
```
but the whole ordeal is incredibly fragile and making assumptions that I'm sure
future linkers and/or dynamic loaders will keep breaking.
--
You are receiving this mail because:
You are watching all bug changes.
