fix fdpic regression in dynamic linker with overly smart compilers

at least gcc 9 broke execution of DT_INIT/DT_FINI for fdpic archs
(presently only sh) by recognizing that the stores to the
compound-literal function descriptor constructed to call them were
dead stores. there's no way to make a "may_alias function", so instead
launder the descriptor through an asm-statement barrier. in practice
just making the compound literal volatile seemed to have worked too,
but this should be less of a hack and more accurately convey the
semantics of what transformations are not valid.

ldso/dynlink.c

@@ -185,8 +185,14 @@ static void *laddr_pg(const struct dso *p, size_t v)
 	}
 	return (void *)(v - p->loadmap->segs[j].p_vaddr + p->loadmap->segs[j].addr);
 }
-#define fpaddr(p, v) ((void (*)())&(struct funcdesc){ \
-	laddr(p, v), (p)->got })
+static void (*fdbarrier(void *p))()
+{
+	void (*fd)();
+	__asm__("" : "=r"(fd) : "0"(p));
+	return fd;
+}
+#define fpaddr(p, v) fdbarrier((&(struct funcdesc){ \
+	laddr(p, v), (p)->got }))
 #else
 #define laddr(p, v) (void *)((p)->base + (v))
 #define laddr_pg(p, v) laddr(p, v)