src: start the .text section with an asm symbol

We create an object file in assembly which introduces the symbol
`__node_text_start` into the .text section and place the resulting
object file as the first file the linker encounters. We do this to
ensure that we can recognize the boundaries of the .text section when
attempting to establish the address range to map to large pages.

Additionally, we rename the section containing the remapping code from
`.lpstub` to `lpstub` so as to take advantage of the linker's feature
whereby it inserts the symbol `__start_lpstub` when the section's name
can be rendered as a valid C variable. We need this symbol in order to
avoid self-mapping the remapping code to large pages, because doing so
would cause the process to crash.

Backport-PR-URL: #32092
PR-URL: #31981
Reviewed-By: Franziska Hinkelmann <franziska.hinkelmann@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: David Carlier <devnexen@gmail.com>
Signed-off-by: Gabriel Schulhof <gabriel.schulhof@intel.com>

node.gyp

@@ -311,6 +311,19 @@
   },
 
   'targets': [
+    {
+      'target_name': 'node_text_start',
+      'type': 'none',
+      'conditions': [
+        [ 'OS=="linux" and '
+          'target_arch=="x64"', {
+          'type': 'static_library',
+          'sources': [
+            'src/large_pages/node_text_start.S'
+          ]
+        }],
+      ]
+    },
     {
       'target_name': '<(node_core_target_name)',
       'type': 'executable',
@@ -496,6 +509,13 @@
             'src/node_snapshot_stub.cc'
           ],
         }],
+        [ 'OS=="linux" and '
+          'target_arch=="x64"', {
+          'dependencies': [ 'node_text_start' ],
+          'ldflags+': [
+            '<(PRODUCT_DIR)/obj.target/node_text_start/src/large_pages/node_text_start.o'
+          ]
+        }],
       ],
     }, # node_core_target_name
     {

src/large_pages/node_large_page.cc

@@ -71,7 +71,11 @@
 
 #if defined(__linux__)
 extern "C" {
-extern char __executable_start;
+// This symbol must be declared weak because this file becomes part of all
+// Node.js targets (like node_mksnapshot, node_mkcodecache, and cctest) and
+// those files do not supply the symbol.
+extern char __attribute__((weak)) __node_text_start;
+extern char __start_lpstub;
 }  // extern "C"
 #endif  // defined(__linux__)
 
@@ -121,6 +125,8 @@ struct text_region FindNodeTextRegion() {
   std::string dev;
   char dash;
   uintptr_t start, end, offset, inode;
+  uintptr_t node_text_start = reinterpret_cast<uintptr_t>(&__node_text_start);
+  uintptr_t lpstub_start = reinterpret_cast<uintptr_t>(&__start_lpstub);
 
   ifs.open("/proc/self/maps");
   if (!ifs) {
@@ -144,21 +150,15 @@ struct text_region FindNodeTextRegion() {
     std::string pathname;
     iss >> pathname;
 
-    if (start != reinterpret_cast<uintptr_t>(&__executable_start))
+    if (permission != "r-xp")
       continue;
 
-    // The next line is our .text section.
-    if (!std::getline(ifs, map_line))
-      break;
-
-    iss = std::istringstream(map_line);
-    iss >> std::hex >> start;
-    iss >> dash;
-    iss >> std::hex >> end;
-    iss >> permission;
+    if (node_text_start < start || node_text_start >= end)
+      continue;
 
-    if (permission != "r-xp")
-      break;
+    start = node_text_start;
+    if (lpstub_start > start && lpstub_start <= end)
+      end = lpstub_start;
 
     char* from = reinterpret_cast<char*>(hugepage_align_up(start));
     char* to = reinterpret_cast<char*>(hugepage_align_down(end));
@@ -318,7 +318,7 @@ static bool IsSuperPagesEnabled() {
 // d. If successful copy the code there and unmap the original region
 int
 #if !defined(__APPLE__)
-__attribute__((__section__(".lpstub")))
+__attribute__((__section__("lpstub")))
 #else
 __attribute__((__section__("__TEXT,__lpstub")))
 #endif

src/large_pages/node_text_start.S

@@ -0,0 +1,5 @@
+.text
+.align 0x2000
+.global __node_text_start
+.hidden __node_text_start
+__node_text_start: