https://sourceware.org/bugzilla/show_bug.cgi?id=32096
Bug ID: 32096 Summary: UBSAN issues in gprofng Product: binutils Version: unspecified Status: NEW Severity: normal Priority: P2 Component: gprofng Assignee: vladimir.mezentsev at oracle dot com Reporter: sam at gentoo dot org Blocks: 32092 Target Milestone: --- Created attachment 15673 --> https://sourceware.org/bugzilla/attachment.cgi?id=15673&action=edit gprofng.log When looking at PR32092, I tried building with UBSAN in case it showed anything. It showed a lot! Build with C{,XX}FLAGS="-O2 -flto=jobserver -fsanitize=undefined" and LDFLAGS="-fsanitize=undefined". I used unreleased GCC 15 (gcc-15 (Gentoo Hardened 15.0.9999 p, commit 70f5adef33e0620d934fc7fb0822e592e3ff04a1) 15.0.0 20240814 (experimental) 10972e6fb59cf83e12dcf7d5d6db01aa8e38dc52) but I can reproduce with 14 too. I didn't get any UBSAN issues without LTO. As a horrible local hack to workaround PR32095, because I only needed to run one test (for PR32092), I did: ``` --- a/gprofng/testsuite/gprofng.display/display.exp +++ b/gprofng/testsuite/gprofng.display/display.exp @@ -51,8 +51,6 @@ if { "$CHECK_TARGET" == "check-extra" || "$CHECK_TARGET" == "check-install" } { } else { # "$CHECK_TARGET" is "check" or "check-DEJAGNU" # dir cflags gprofflags tflags set table { - {"jsynprog" "-g -Wall" "-p on -j on" "-j add_int"} - {"mttest" "-g -O0" "-p on" "-j calladd"} {"synprog" "-g -O0" "-p on" "cpu"} } } `` and ``` $ export UBSAN_OPTIONS="print_stacktrace=1:halt_on_error=0" $ LD_PRELOAD=/usr/lib/gcc/x86_64-pc-linux-gnu/15/libasan.so:/usr/lib/gcc/x86_64-pc-linux-gnu/15/libubsan.so make RUNTESTFLAGS="display.exp=synprog" check ```` Referenced Bugs: https://sourceware.org/bugzilla/show_bug.cgi?id=32092 [Bug 32092] gprofng testsuite fails with -flto (ERROR: comparison of results in synprog failed) -- You are receiving this mail because: You are on the CC list for the bug.