Add opt hpa_hugify_sync to hugify synchronously

Linux 6.1 introduced `MADV_COLLAPSE` flag to perform a best-effort
synchronous collapse of the native pages mapped by the memory range into
transparent huge pages.

Synchronous hugification might be beneficial for at least two reasons:
we are not relying on khugepaged anymore and get an instant feedback if
range wasn't hugified.

If `hpa_hugify_sync` option is on, we'll try to perform synchronously
collapse and if it wasn't successful, we'll fallback to asynchronous
behaviour.

src/pages.c

@@ -567,6 +567,30 @@ pages_nohuge_unaligned(void *addr, size_t size) {
 	return pages_nohuge_impl(addr, size, false);
 }
 
+bool
+pages_collapse(void *addr, size_t size) {
+	assert(PAGE_ADDR2BASE(addr) == addr);
+	assert(PAGE_CEILING(size) == size);
+	/*
+	 * There is one more MADV_COLLAPSE precondition that is not easy to
+	 * express with assert statement.  In order to madvise(addr, size,
+	 * MADV_COLLAPSE) call to be successful, at least one page in the range
+	 * must currently be backed by physical memory.  In particularly, this
+	 * means we can't call pages_collapse on freshly mapped memory region.
+	 * See madvise(2) man page for more details.
+	 */
+#if defined(JEMALLOC_HAVE_MADVISE_COLLAPSE) && \
+    (defined(MADV_COLLAPSE) || defined(JEMALLOC_MADV_COLLAPSE))
+#  if defined(MADV_COLLAPSE)
+	return (madvise(addr, size, MADV_COLLAPSE) != 0);
+#  elif defined(JEMALLOC_MADV_COLLAPSE)
+	return (madvise(addr, size, JEMALLOC_MADV_COLLAPSE) != 0);
+#  endif
+#else
+	return true;
+#endif
+}
+
 bool
 pages_dontdump(void *addr, size_t size) {
 	assert(PAGE_ADDR2BASE(addr) == addr);