Hello, devel and arm lists!
We recently ran into a testsuite failure when building glibc for Rawhide on armv7hl: https://bugzilla.redhat.com/show_bug.cgi?id=1951492
The test program experienced a hang after a thread was cancelled.
So far, it appears that either annobin generated notes or the way in which the static linker interprets them lead to incorrect unwind information in 32 bit Arm binaries.
While in this case only a glibc testsuite failure was caught, there is a chance that building with annobin can lead to incorrect unwind information and associated problems for any armv7hl Rawhide package built this way.
With this in mind, I would like to know how the Fedora-Arm developer community feels about disabling annobin when building RPMs only on armv7hl until this issue is investigated and the root cause fixed. This should be possible with a change to the redhat-rpm-config.
Cheers, Arjun
On Tue, May 18, 2021 at 5:52 PM Arjun Shankar arjun@redhat.com wrote:
Hello, devel and arm lists!
We recently ran into a testsuite failure when building glibc for Rawhide on armv7hl: https://bugzilla.redhat.com/show_bug.cgi?id=1951492
The test program experienced a hang after a thread was cancelled.
So far, it appears that either annobin generated notes or the way in which the static linker interprets them lead to incorrect unwind information in 32 bit Arm binaries.
While in this case only a glibc testsuite failure was caught, there is a chance that building with annobin can lead to incorrect unwind information and associated problems for any armv7hl Rawhide package built this way.
With this in mind, I would like to know how the Fedora-Arm developer community feels about disabling annobin when building RPMs only on armv7hl until this issue is investigated and the root cause fixed. This should be possible with a change to the redhat-rpm-config.
Overall not great but you also don't particularly provide alternatives or impact details. I suppose the main question I have is when this regressed, and where?
* Peter Robinson:
On Tue, May 18, 2021 at 5:52 PM Arjun Shankar arjun@redhat.com wrote:
Hello, devel and arm lists!
We recently ran into a testsuite failure when building glibc for Rawhide on armv7hl: https://bugzilla.redhat.com/show_bug.cgi?id=1951492
The test program experienced a hang after a thread was cancelled.
So far, it appears that either annobin generated notes or the way in which the static linker interprets them lead to incorrect unwind information in 32 bit Arm binaries.
While in this case only a glibc testsuite failure was caught, there is a chance that building with annobin can lead to incorrect unwind information and associated problems for any armv7hl Rawhide package built this way.
With this in mind, I would like to know how the Fedora-Arm developer community feels about disabling annobin when building RPMs only on armv7hl until this issue is investigated and the root cause fixed. This should be possible with a change to the redhat-rpm-config.
Overall not great but you also don't particularly provide alternatives or impact details. I suppose the main question I have is when this regressed, and where?
I do not think it is a regression as such. To me it looks like it has always been present since we added annobin on armhfp. (EABI exception handling is very special, different from all the other targets.) Whether that particular glibc test is impacted depends on libc.so.6 layout, so even in glibc, it comes and goes, as changes are made to the code.
Thanks, Florian
-
Arjun Shankar -
Florian Weimer -
Peter Robinson