On Fri, 27 Jun 2025 07:13:26 GMT, Matthias Baesken <mbaes...@openjdk.org> wrote:
>> There are a number of :tier3 HS jtreg tests using libjsig. Unfortunately >> they clash with asan, so it should be avoided to run them if asan is >> configured. >> >> Examples : >> runtime/signal/TestSigalrm.java >> runtime/signal/TestSigbus.java >> >> They run into errors like this >> >> >> stdout: []; >> stderr: [==59482==ASan runtime does not come first in initial library list; >> you should either link runtime to your application or manually preload it >> with LD_PRELOAD. >> >> >> One option would be to avoid running those tests when asan is enabled. >> Another option is to avoid the so called 'link order check' . >> https://github.com/google/sanitizers/wiki/addresssanitizerflags >> >> verify_asan_link_order - Check position of ASan runtime in library list >> (needs to be disabled when other library has to be preloaded system-wide) > > Matthias Baesken has updated the pull request incrementally with one > additional commit since the last revision: > > Readjust COPYRIGHT year Saw this too late. I am unsure about this change, because it disables a rather necessary warning. ASAN interposes a number of APIs, e.g. malloc/free. Bad things can happen if that interposition does not cover the full JVM process. E.g. you preload library1, which comes before ASAN interposition is in place, so it uses libc malloc. You then load ASAN, then the rest of JVM. Anything loaded after ASAN will route through ASAN's version of malloc/free. So different parts of the process could use different versions of malloc/free. That could be trouble since we don't know whether pointers returned from raw libc malloc are compatible with ASAN's version of free, and vice versa. Even if nothing like that happens, it would quietly disable ASAN for a part of the libjvm, e.g. in case of libjsig for the libjsig. Unless we fully understand what would happen in these cases, I would not switch off these warning for the full process. Rather just switch off tests on ASAN builds that rely on LD_PRELOAD. We don't have that many (should only be libjsig, no?) This is just my 5 cents. I think ASAN is really useful, and the bugs @MBaesken already found proves that. ------------- PR Comment: https://git.openjdk.org/jdk/pull/25978#issuecomment-3113258471
