On Fri, Jul 5, 2024 at 6:22 PM Jürgen Groß <[email protected]> wrote: > > On 05.07.24 16:59, Andrei Stan wrote: > > Currently only shared libraries build with USE_PTHREAD enabled. > > > > This patch adds the macro also to xs.o. > > > > Backport: 4.18+ # maybe older > > Signed-off-by: Andrei Stan <[email protected]> > > Commit dde3e02b7068 did that explicitly, stating that phtreads are > not compatible with static linking. > > Was that reasoning wrong? > > Juergen > --- > > --- > > tools/libs/store/Makefile | 1 + > > 1 file changed, 1 insertion(+) > > > > diff --git a/tools/libs/store/Makefile b/tools/libs/store/Makefile > > index 0649cf8307..c6f147c72f 100644 > > --- a/tools/libs/store/Makefile > > +++ b/tools/libs/store/Makefile > > @@ -20,6 +20,7 @@ CFLAGS += -include $(XEN_ROOT)/tools/config.h > > CFLAGS += $(CFLAGS_libxentoolcore) > > > > xs.opic: CFLAGS += -DUSE_PTHREAD > > +xs.o: CFLAGS += -DUSE_PTHREAD > > ifeq ($(CONFIG_Linux),y) > > xs.opic: CFLAGS += -DUSE_DLSYM > > endif
It was a pretty nasty situation, giving some context, this is for a Go based CLI tool and some things in there are multithreaded, but i don't think calling in the bindings happens anywhere in parallel. Without this patch there would be some sort of race conditions (or so I assume from the debugging i've done) happening withing the xen/tools code, making it impossible to start a domain. In this case there were no issues in linking pthreads statically. I was not even aware of that being a possible issue. Is it really? Also I am not too sure how much that code path is actually tested, probably the majority of the testing is done with dynamic builds. Andrei
