https://sourceware.org/bugzilla/show_bug.cgi?id=32260
Bug ID: 32260
Summary: regression: /bin/ld: BFD (GNU Binutils for Debian)
2.43.1 assertion fail ../../bfd/merge.c:247
Product: binutils
Version: 2.43.1
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: ld
Assignee: unassigned at sourceware dot org
Reporter: jonny.weir at clearpool dot io
Target Milestone: ---
Hi,
Just to note, this may be related to the following bug I raised previously that
was fixed:
https://sourceware.org/bugzilla/show_bug.cgi?id=31009
To summarize:
I am building a large scale C++23 project now using gcc14 in Debian Linux
(sid).
I have two build environments, one that uses binutils 2.42.90.20240720-2
(Debian package) and another that uses binutils 2.43.1-3, also a Debian
package. One build environment is running slightly behind the other, for
obvious reasons :-).
Aside from the binutils difference, the build environments are identical.
I have been successfully building this project using the 2.42 version of
binutils, however I believe another regression has been introduced in the
2.43.1 version (possibly all 2.43 versions have this issue).
Where I see the failure is as follows using the 2.43.1 version:
g++-14 -o main.o -c -std=c++2b -fexceptions -fconcepts -fcoroutines -flto=auto
-Wall -g -O3 main.cpp
g++-14 -o output -flto=auto -rdynamic -Wl,-rpath=. <object_files> -Xlinker
-Bstatic -lpthread -lrt -lcurl -lssl -lcrypto
If I compile using -O2 everything compiles and links correctly, but when -O3 is
used, the following output is produced during the linking phase:
/bin/ld: BFD (GNU Binutils for Debian) 2.43.1 assertion fail
../../bfd/merge.c:247
/bin/ld: BFD (GNU Binutils for Debian) 2.43.1 assertion fail
../../bfd/merge.c:247
/bin/ld: BFD (GNU Binutils for Debian) 2.43.1 assertion fail
../../bfd/merge.c:247
... etc ...
This is repeated thousands of times, as before.
As per the suggetions in the previous ticket, I have added some printf logging
above line 247 in bfd/merge.c:
printf( "XXX bfdtab->count=%u table->nbuckets=%u\n", bfdtab->count,
table->nbuckets );
BFD_ASSERT (!NEEDS_RESIZE (bfdtab->count + 1, table->nbuckets));
to try and work out what is happening, but I haven't been able to determine why
nbuckets goes to 0.
The output from the build is as follows (duplicate lines have been removed):
XXX bfdtab->count=0 table->nbuckets=8192
XXX bfdtab->count=1 table->nbuckets=8192
...
XXX bfdtab->count=125 table->nbuckets=8192
XXX bfdtab->count=126 table->nbuckets=32768
XXX bfdtab->count=127 table->nbuckets=32768
...
XXX bfdtab->count=488 table->nbuckets=32768
XXX bfdtab->count=489 table->nbuckets=65536
XXX bfdtab->count=490 table->nbuckets=65536
...
XXX bfdtab->count=778 table->nbuckets=65536
XXX bfdtab->count=779 table->nbuckets=262144
XXX bfdtab->count=780 table->nbuckets=262144
XXX bfdtab->count=781 table->nbuckets=262144
...
XXX bfdtab->count=1406 table->nbuckets=262144
XXX bfdtab->count=1407 table->nbuckets=524288
XXX bfdtab->count=1408 table->nbuckets=524288
...
XXX bfdtab->count=1751 table->nbuckets=524288
XXX bfdtab->count=1752 table->nbuckets=0
XXX bfdtab->count=1753 table->nbuckets=0
XXX bfdtab->count=1754 table->nbuckets=0
XXX bfdtab->count=1771 table->nbuckets=0
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
XXX bfdtab->count=1772 table->nbuckets=0
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
...
XXX bfdtab->count=871457 table->nbuckets=0
XXX bfdtab->count=871458 table->nbuckets=0
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: BFD (GNU Binutils) 2.43.1.20241010 assertion fail merge.c:248
/bin/ld: /tmp/cc0M4ieT.ltrans0.ltrans.o: access beyond end of merged section
(20)
/bin/ld: /tmp/cc0M4ieT.ltrans0.ltrans.o: access beyond end of merged section
(33)
/bin/ld: /tmp/cc0M4ieT.ltrans0.ltrans.o: access beyond end of merged section
(41)
/bin/ld: /tmp/cc0M4ieT.ltrans0.ltrans.o: access beyond end of merged section
(48)
...
/bin/ld: final link failed
collect2: error: ld returned 1 exit status
The total number of lines output is around 860,000 (this count excludes my
additional debug logging).
Note the version change, I have built this locally from source from the
binutils-2_43-branch so that I could edit the merge.c file and reproduce this
with the logging.
Finally, during this linking stage, the memory usage for binutils 2.42 was
around 30G for the /bin/ld process. With 2.43.1 the memory usage has increased
dramatically to around 71G. This seems very excessive and I am wondering if
this might help track down the issue with this. Perhaps some new code that was
added around memory reallocation in the 2.43 code.
If you require any more clarity or want me to run the build again with any more
logging / code changes, I am willing to do so.
Thanks,
Jonny
--
You are receiving this mail because:
You are on the CC list for the bug.
