Fix i64-using function pointers in upstream dynamic linking (emscripten-core#9810)

The dynamic loader needs to place the original wasm functions in the table, so
that other modules can call them. The JS-legalized versions will have the wrong
types for indirect calls.

To implement this, binaryen's legalization pass exports illegal functions twice,
once legalized for JS, and once as orig$ plus the original name. This PR makes
the loader use those.

Fixes emscripten-core#9562

Author: kripken

emscripten.py

@@ -1798,18 +1798,26 @@ def create_fp_accessors(metadata):
                  'perhaps a side module was not linked in? if this symbol was expected to arrive '
                  'from a system library, try to build the MAIN_MODULE with '
                  'EMCC_FORCE_STDLIBS=XX in the environment");')
+    # the name of the original function is generally the normal function
+    # name, unless it is legalized, in which case the export is the legalized
+    # version, and the original provided by orig$X
+    if shared.Settings.LEGALIZE_JS_FFI and not shared.JS.is_legal_sig(sig):
+      name = 'orig$' + name
 
     accessors.append('''
 Module['%(full)s'] = function() {
   %(assert)s
-  var func = Module['%(mangled)s'];
-  if (!func)
-    func = %(mangled)s;
+  // Use the wasm function itself, for the table.
+  var func = Module['asm']['%(original)s'];
+  // If there is no wasm function, this may be a JS library function or
+  // something from another module.
+  if (!func) func = Module['%(mangled)s'];
+  if (!func) func = %(mangled)s;
   var fp = addFunction(func, '%(sig)s');
   Module['%(full)s'] = function() { return fp };
   return fp;
 }
-''' % {'full': asmjs_mangle(fullname), 'mangled': mangled, 'assert': assertion, 'sig': sig})
+''' % {'full': asmjs_mangle(fullname), 'mangled': mangled, 'original': name, 'assert': assertion, 'sig': sig})
 
   return '\n'.join(accessors)
 
@@ -2319,6 +2327,11 @@ def debug_copy(src, dst):
     args.append('--check-stack-overflow')
   if shared.Settings.STANDALONE_WASM:
     args.append('--standalone-wasm')
+  # When we dynamically link our JS loader adds functions from wasm modules to
+  # the table. It must add the original versions of them, not legalized ones,
+  # so that indirect calls have the right type, so export those.
+  if shared.Settings.RELOCATABLE:
+    args.append('--pass-arg=legalize-js-interface-export-originals')
   stdout = shared.Building.run_binaryen_command('wasm-emscripten-finalize',
                                                 infile=base_wasm,
                                                 outfile=wasm,

src/support.js

@@ -436,7 +436,10 @@ function loadWebAssemblyModule(binary, flags) {
     // b) Symbols from loaded modules are not always added to the global Module.
     var moduleLocal = {};
 
-    var resolveSymbol = function(sym, type) {
+    var resolveSymbol = function(sym, type, legalized) {
+      if (legalized) {
+        sym = 'orig$' + sym;
+      }
 #if WASM_BACKEND
       sym = '_' + sym;
 #endif
@@ -496,10 +499,11 @@ function loadWebAssemblyModule(binary, flags) {
           assert(parts.length == 3)
           var name = parts[1];
           var sig = parts[2];
+          var legalized = sig.indexOf('j') >= 0; // check for i64s
           var fp = 0;
           return obj[prop] = function() {
             if (!fp) {
-              var f = resolveSymbol(name, 'function');
+              var f = resolveSymbol(name, 'function', legalized);
               fp = addFunction(f, sig);
             }
             return fp;

tests/other/metadce/hello_world_O3_MAIN_MODULE.exports

@@ -1115,6 +1115,215 @@ opterr
 optind
 optopt
 optreset
+orig$__addtf3
+orig$__ashldi3
+orig$__ashlti3
+orig$__ashrdi3
+orig$__ashrti3
+orig$__atomic_compare_exchange_16
+orig$__atomic_compare_exchange_8
+orig$__atomic_exchange_16
+orig$__atomic_exchange_8
+orig$__atomic_fetch_add_16
+orig$__atomic_fetch_add_8
+orig$__atomic_fetch_and_16
+orig$__atomic_fetch_and_8
+orig$__atomic_fetch_or_16
+orig$__atomic_fetch_or_8
+orig$__atomic_fetch_sub_16
+orig$__atomic_fetch_sub_8
+orig$__atomic_fetch_xor_16
+orig$__atomic_fetch_xor_8
+orig$__atomic_load_8
+orig$__atomic_store_16
+orig$__atomic_store_8
+orig$__clzti2
+orig$__cosl
+orig$__divdi3
+orig$__divmoddi4
+orig$__divtf3
+orig$__divti3
+orig$__eqtf2
+orig$__fixdfdi
+orig$__fixtfdi
+orig$__fixtfsi
+orig$__fixtfti
+orig$__fixunsdfdi
+orig$__fixunstfdi
+orig$__fixunstfsi
+orig$__fixunstfti
+orig$__floatdidf
+orig$__floatditf
+orig$__floattidf
+orig$__floattisf
+orig$__floattitf
+orig$__floatunditf
+orig$__floatuntidf
+orig$__floatuntisf
+orig$__fpclassifyl
+orig$__fseeko
+orig$__fseeko_unlocked
+orig$__ftello
+orig$__ftello_unlocked
+orig$__getf2
+orig$__gttf2
+orig$__intscan
+orig$__invtrigl_R
+orig$__letf2
+orig$__lgammal_r
+orig$__lshrdi3
+orig$__lshrti3
+orig$__lttf2
+orig$__mmap
+orig$__moddi3
+orig$__modti3
+orig$__muldi3
+orig$__multc3
+orig$__multf3
+orig$__multi3
+orig$__netf2
+orig$__p1evll
+orig$__polevll
+orig$__rand48_step
+orig$__rem_pio2l
+orig$__shlim
+orig$__signbitl
+orig$__sinl
+orig$__stdio_seek
+orig$__strtoimax_internal
+orig$__strtoll_internal
+orig$__strtoull_internal
+orig$__strtoumax_internal
+orig$__subtf3
+orig$__tanl
+orig$__trunctfdf2
+orig$__trunctfsf2
+orig$__udivdi3
+orig$__udivmoddi4
+orig$__udivmodti4
+orig$__udivti3
+orig$__umoddi3
+orig$__umodti3
+orig$__unordtf2
+orig$__uremdi3
+orig$_emscripten_atomic_fetch_and_add_u64
+orig$_emscripten_atomic_fetch_and_and_u64
+orig$_emscripten_atomic_fetch_and_or_u64
+orig$_emscripten_atomic_fetch_and_sub_u64
+orig$_emscripten_atomic_fetch_and_xor_u64
+orig$acoshl
+orig$acosl
+orig$asinhl
+orig$asinl
+orig$atan2l
+orig$atanhl
+orig$atanl
+orig$atoll
+orig$cbrtl
+orig$copysignl
+orig$coshl
+orig$cosl
+orig$emscripten_atomic_add_u64
+orig$emscripten_atomic_and_u64
+orig$emscripten_atomic_cas_u64
+orig$emscripten_atomic_exchange_u64
+orig$emscripten_atomic_load_u64
+orig$emscripten_atomic_or_u64
+orig$emscripten_atomic_store_u64
+orig$emscripten_atomic_sub_u64
+orig$emscripten_atomic_xor_u64
+orig$erfcl
+orig$erfl
+orig$exp10l
+orig$exp2l
+orig$expl
+orig$expm1l
+orig$fdiml
+orig$ffsll
+orig$fmal
+orig$fmaxl
+orig$fminl
+orig$fmodl
+orig$frexpl
+orig$fseeko
+orig$fseeko64
+orig$ftello
+orig$ftello64
+orig$ftruncate
+orig$ftruncate64
+orig$hypotl
+orig$ilogbl
+orig$imaxabs
+orig$imaxdiv
+orig$ldexpl
+orig$lgammal
+orig$lgammal_r
+orig$llabs
+orig$lldiv
+orig$llrint
+orig$llrintf
+orig$llrintl
+orig$llround
+orig$llroundf
+orig$llroundl
+orig$lockf
+orig$lockf64
+orig$log10l
+orig$log1pl
+orig$log2l
+orig$logbl
+orig$logl
+orig$lrintl
+orig$lroundl
+orig$lseek
+orig$lseek64
+orig$mmap
+orig$mmap64
+orig$modfl
+orig$nearbyintl
+orig$nextafterl
+orig$nexttoward
+orig$nexttowardf
+orig$nexttowardl
+orig$posix_fadvise
+orig$posix_fadvise64
+orig$posix_fallocate
+orig$posix_fallocate64
+orig$pow10l
+orig$powl
+orig$pread
+orig$pread64
+orig$preadv
+orig$preadv64
+orig$pwrite
+orig$pwrite64
+orig$pwritev
+orig$pwritev64
+orig$remainderl
+orig$remquol
+orig$rintl
+orig$roundl
+orig$scalblnl
+orig$scalbnl
+orig$sincosl
+orig$sinhl
+orig$sinl
+orig$strtoimax
+orig$strtoll
+orig$strtoll_l
+orig$strtoull
+orig$strtoull_l
+orig$strtoumax
+orig$tanhl
+orig$tanl
+orig$tgammal
+orig$truncate
+orig$truncate64
+orig$truncl
+orig$wcstoimax
+orig$wcstoll
+orig$wcstoull
+orig$wcstoumax
 pause
 pclose
 perror

tests/test_core.py

@@ -3893,9 +3893,15 @@ def test_dylink_i64_b(self):
       #include <stdio.h>
       #include <stdint.h>
       extern int64_t sidey();
+      int64_t testAdd(int64_t a) {
+        return a + 1;
+      }
+      typedef int64_t (*testAddHandler)(int64_t);
+      testAddHandler h = &testAdd;
       int main() {
         printf("other says %lld.\n", sidey());
-        return 0;
+        int64_t r = h(42);
+        printf("my fp says: %lld.\n", r);
       }
     ''', '''
       #include <stdint.h>
@@ -3905,7 +3911,7 @@ def test_dylink_i64_b(self):
         x = 18 - x;
         return x;
       }
-    ''', 'other says -1311768467750121224.')
+    ''', 'other says -1311768467750121224.\nmy fp says: 43.')
 
   @needs_dlfcn
   def test_dylink_class(self):

tests/test_other.py

@@ -8171,7 +8171,7 @@ def test_metadce_cxx_fastcomp(self, *args):
     # don't compare the # of functions in a main module, which changes a lot
     # TODO(sbc): Investivate why the number of exports is order of magnitude
     # larger for wasm backend.
-    'main_module_1': (['-O3', '-s', 'MAIN_MODULE=1'],  174, [], [], 517336, None, 1520, None), # noqa
+    'main_module_1': (['-O3', '-s', 'MAIN_MODULE=1'],  174, [], [], 517336, None, 1729, None), # noqa
     'main_module_2': (['-O3', '-s', 'MAIN_MODULE=2'],   12, [], [],  10770,   12,   10, None), # noqa
   })
   @no_fastcomp()

tools/shared.py

@@ -3025,6 +3025,10 @@ def legalize_sig(sig):
         ret.append('i')
     return ''.join(ret)
 
+  @staticmethod
+  def is_legal_sig(sig):
+    return sig == JS.legalize_sig(sig)
+
   @staticmethod
   def make_extcall(sig, named=True):
     args = ','.join(['a' + str(i) for i in range(1, len(sig))])