https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123873
Bug ID: 123873
Summary: UBSAN confuses optimizer
Product: gcc
Version: 15.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: sanitizer
Assignee: unassigned at gcc dot gnu.org
Reporter: peterz at infradead dot org
CC: dodji at gcc dot gnu.org, dvyukov at gcc dot gnu.org,
jakub at gcc dot gnu.org, kcc at gcc dot gnu.org
Target Milestone: ---
Hi,
Some Linux kernel builds with GCC-15.2.0 fail optimizing in the face of UBSAN.
https://lore.kernel.org/all/[email protected]/
Is it considered acceptable for UBSAN to affect optimization quality like this?
I ask because various Distro's are shipping UBSAN=y kernels, and if it is
causing obvious optimization issues like this I'm going to have to tell them to
stop doing that.
The code in question is:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/linux/seqlock.h#n1237
And compilers (clang and GCC) are generally capable of seeing right through all
that and generate tight code. Except with the various sanitizers on; then it
goes off the rails. I've already (as you can see) marked KASAN as 'special'
because nobody should care about performance on a KASAN=y build, but what about
UBSAN?
Do you compiler folks consider UBSAN something that should be viable for a
'release' build -- if so, its failing. If not, I'm going to have to dissuade
Debian and friends from having it enabled.
~ Peter
