[lld][WebAssembly] Do not relocate ABSOLUTE symbols #153763

yamt · 2025-08-15T08:17:26Z

Fixes #153759

llvmbot · 2025-08-15T08:17:59Z

@llvm/pr-subscribers-lld-wasm

Author: YAMAMOTO Takashi (yamt)

Changes

Fixes #153537

Full diff: https://github.com/llvm/llvm-project/pull/153763.diff

1 Files Affected:

(modified) lld/wasm/SyntheticSections.cpp (+4-1)

diff --git a/lld/wasm/SyntheticSections.cpp b/lld/wasm/SyntheticSections.cpp
index e1192706ea913..6c1c4391ea0d7 100644
--- a/lld/wasm/SyntheticSections.cpp
+++ b/lld/wasm/SyntheticSections.cpp
@@ -442,6 +442,8 @@ void GlobalSection::generateRelocationCode(raw_ostream &os, bool TLS) const {
   for (const Symbol *sym : internalGotSymbols) {
     if (TLS != sym->isTLS())
       continue;
+    if (sym->flags & WASM_SYMBOL_ABSOLUTE)
+      continue;
 
     if (auto *d = dyn_cast<DefinedData>(sym)) {
       // Get __memory_base
@@ -503,7 +505,8 @@ void GlobalSection::writeBody() {
     bool useExtendedConst = false;
     uint32_t globalIdx;
     int64_t offset;
-    if (ctx.arg.extendedConst && ctx.isPic) {
+    if (ctx.arg.extendedConst && ctx.isPic &&
+        (sym->flags & WASM_SYMBOL_ABSOLUTE) == 0) {
       if (auto *d = dyn_cast<DefinedData>(sym)) {
         if (!sym->isTLS()) {
           globalIdx = ctx.sym.memoryBase->getGlobalIndex();

llvmbot · 2025-08-15T08:17:59Z

@llvm/pr-subscribers-lld

Author: YAMAMOTO Takashi (yamt)

Changes

Fixes #153537

Full diff: https://github.com/llvm/llvm-project/pull/153763.diff

1 Files Affected:

(modified) lld/wasm/SyntheticSections.cpp (+4-1)

diff --git a/lld/wasm/SyntheticSections.cpp b/lld/wasm/SyntheticSections.cpp
index e1192706ea913..6c1c4391ea0d7 100644
--- a/lld/wasm/SyntheticSections.cpp
+++ b/lld/wasm/SyntheticSections.cpp
@@ -442,6 +442,8 @@ void GlobalSection::generateRelocationCode(raw_ostream &os, bool TLS) const {
   for (const Symbol *sym : internalGotSymbols) {
     if (TLS != sym->isTLS())
       continue;
+    if (sym->flags & WASM_SYMBOL_ABSOLUTE)
+      continue;
 
     if (auto *d = dyn_cast<DefinedData>(sym)) {
       // Get __memory_base
@@ -503,7 +505,8 @@ void GlobalSection::writeBody() {
     bool useExtendedConst = false;
     uint32_t globalIdx;
     int64_t offset;
-    if (ctx.arg.extendedConst && ctx.isPic) {
+    if (ctx.arg.extendedConst && ctx.isPic &&
+        (sym->flags & WASM_SYMBOL_ABSOLUTE) == 0) {
       if (auto *d = dyn_cast<DefinedData>(sym)) {
         if (!sym->isTLS()) {
           globalIdx = ctx.sym.memoryBase->getGlobalIndex();

sbc100

Can we add a test for this? Perhaps one of the existing tests for internal GOT symbols could have a reference to an absolute symbol (the page size symbol maybe?) added?

sbc100 · 2025-08-15T18:31:59Z

lld/wasm/SyntheticSections.cpp

Maybe do this 2 lines up before the TLS check?

why? just because it can be slightly cheaper?

yamt · 2025-08-18T01:26:06Z

Can we add a test for this? Perhaps one of the existing tests for internal GOT symbols could have a reference to an absolute symbol (the page size symbol maybe?) added?

i added a test.

Fixes llvm#153759

sbc100

lgtm % comment

lld/test/wasm/pie.s

sbc100 · 2025-08-19T16:03:54Z

This causes a bunch of test failures on the emscripten waterfall, mostly lsan tests.

I'm guessing that not all symbols we add with addOptionalDataSymbol and/or addSyntheticDataSymbol should be treated as absolute like this.

I think we should revert until we can figure out the best approach here.

This reverts commit 9247be8.

Reverts #153763 This caused a bunch of failures on the emscripten waterfall, specifically most of the lsan tests started failing.

…s" (#154371) Reverts llvm/llvm-project#153763 This caused a bunch of failures on the emscripten waterfall, specifically most of the lsan tests started failing.

yamt · 2025-08-20T00:17:57Z

This causes a bunch of test failures on the emscripten waterfall, mostly lsan tests.

where can i see them?

sbc100 · 2025-08-20T00:34:02Z

https://ci.chromium.org/ui/p/emscripten-releases/builders/luci.emscripten-releases.ci/linux-test-suites

Looks like all the tests that use lsan + dynamic linking are reporting false leaks:

FAIL: test_dylink_i64_invoke_rtld_local_no_bigint (test_core.asan)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/chrome-bot/.cache/vpython-root.1000/store/cpython+tkj8tpr2n0oj8m0t270p0m71s0/contents/lib/python3.8/unittest/case.py", line 60, in testPartExecutor
    yield
  File "/home/chrome-bot/.cache/vpython-root.1000/store/cpython+tkj8tpr2n0oj8m0t270p0m71s0/contents/lib/python3.8/unittest/case.py", line 676, in run
    self._callTestMethod(testMethod)
  File "/home/chrome-bot/.cache/vpython-root.1000/store/cpython+tkj8tpr2n0oj8m0t270p0m71s0/contents/lib/python3.8/unittest/case.py", line 633, in _callTestMethod
    method()
  File "/b/s/w/ir/x/w/install/emscripten/test/common.py", line 917, in resulting_test
    return func(self, *args)
  File "/b/s/w/ir/x/w/install/emscripten/test/common.py", line 576, in metafunc
    f(self, *args, **kwargs)
  File "/b/s/w/ir/x/w/install/emscripten/test/test_core.py", line 153, in decorated
    return func(self, *args, **kwargs)
  File "/b/s/w/ir/x/w/install/emscripten/test/test_core.py", line 4504, in test_dylink_i64_invoke
    self.dylink_test(r'''\
  File "/b/s/w/ir/x/w/install/emscripten/test/test_core.py", line 4063, in dylink_test
    return self.dylink_testf(main, side, expected, main_module=main_module, **kwargs)
  File "/b/s/w/ir/x/w/install/emscripten/test/test_core.py", line 4118, in dylink_testf
    self.do_runf(main, expected, force_c=force_c, **kwargs)
  File "/b/s/w/ir/x/w/install/emscripten/test/common.py", line 1958, in do_runf
    return self._build_and_run(filename, expected_output, **kwargs)
  File "/b/s/w/ir/x/w/install/emscripten/test/common.py", line 2006, in _build_and_run
    js_output = self.run_js(js_file, engine, args,
  File "/b/s/w/ir/x/w/install/emscripten/test/common.py", line 1586, in run_js
    self.fail('JS subprocess failed (%s): %s (expected=%s).  Output:\n%s' % (error.cmd, error.returncode, assert_returncode, ret))
  File "/home/chrome-bot/.cache/vpython-root.1000/store/cpython+tkj8tpr2n0oj8m0t270p0m71s0/contents/lib/python3.8/unittest/case.py", line 753, in fail
    raise self.failureException(msg)
AssertionError: JS subprocess failed (/b/s/w/ir/cache/builder/emscripten-releases/node-v22.16.0-linux-x64/bin/node --stack-trace-limit=50 --trace-uncaught /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js): 1 (expected=0).  Output:
got 84
got 0

=================================================================
==42==ERROR: LeakSanitizer: detected memory leaks

Indirect leak of 46 byte(s) in 1 object(s) allocated from:
    #0 0x000222dc in calloc+0x222dc (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x222dc)
    #1 0x00009ccf in load_library_start+0x9ccf (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x9ccf)
    #2 0x00009a77 in _dlopen+0x9a77 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x9a77)
    #3 0x00009976 in dlopen+0x9976 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x9976)
    #4 0x00008b68 in main+0x8b68 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x8b68)
    #5 0x800016f9 in callMain /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:5881:15
    #6 0x80001723 in doRun /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:5923:24
    #7 0x8000172d in run /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:5933:5
    #8 0x8000115c in removeRunDependency /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:4444:7
    #9 0x8000028e in receiveInstance /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:654:5
    #10 0x8000029c in receiveInstantiationResult /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:668:12
    #11 0x800002b3 in createWasm /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:691:17

Indirect leak of 16 byte(s) in 1 object(s) allocated from:
    #0 0x000222dc in calloc+0x222dc (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x222dc)
    #1 0x00009570 in new_dlevent+0x9570 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x9570)
    #2 0x00009ed1 in load_library_done+0x9ed1 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x9ed1)
    #3 0x00009a90 in _dlopen+0x9a90 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x9a90)
    #4 0x00009976 in dlopen+0x9976 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x9976)
    #5 0x00008b68 in main+0x8b68 (/b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js+0x8b68)
    #6 0x800016f9 in callMain /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:5881:15
    #7 0x80001723 in doRun /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:5923:24
    #8 0x8000172d in run /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:5933:5
    #9 0x8000115c in removeRunDependency /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:4444:7
    #10 0x8000028e in receiveInstance /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:654:5
    #11 0x8000029c in receiveInstantiationResult /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:668:12
    #12 0x800002b3 in createWasm /b/s/w/ir/x/t/emtest_73mrsv9g/emscripten_test_asan_z7vmenc8/main.js:691:17

SUMMARY: AddressSanitizer: 62 byte(s) leaked in 2 allocation(s).

At least I assume that are false positives, since they go away when this change is reverted.

My guess is that we are using ABSOLUTE for more symbols that just __wasm_first_page_end so we need some way to differentiate these two types of symbols.

yamt · 2025-08-20T09:23:17Z

My guess is that we are using ABSOLUTE for more symbols that just __wasm_first_page_end so we need some way to differentiate these two types of symbols.

#154494

llvmbot added lld lld:wasm labels Aug 15, 2025

yamt mentioned this pull request Aug 15, 2025

[lld][WebAssembly] Allow linker-synthetic symbols to be undefine when… #153537

Open

yamt changed the title ~~[lld][WebAssembly] do not relocate ABOSULTE symbols~~ [lld][WebAssembly] do not relocate ABSOLUTE symbols Aug 15, 2025

yamt force-pushed the wasm-ld-absolute-reloc branch from 0196c11 to 9aa3bf1 Compare August 15, 2025 08:35

sbc100 reviewed Aug 15, 2025

View reviewed changes

sbc100 changed the title ~~[lld][WebAssembly] do not relocate ABSOLUTE symbols~~ [lld][WebAssembly] Do not relocate ABSOLUTE symbols Aug 15, 2025

yamt force-pushed the wasm-ld-absolute-reloc branch 2 times, most recently from af06b91 to 4766075 Compare August 18, 2025 01:25

yamt added 3 commits August 18, 2025 10:33

[lld][WebAssembly] Do not relocate ABSOLUTE symbols

3f65a78

Fixes llvm#153759

add a test

7cdff15

check ABSOLUTE before TLS

08fde96

yamt force-pushed the wasm-ld-absolute-reloc branch from 4766075 to 08fde96 Compare August 18, 2025 01:33

sbc100 reviewed Aug 18, 2025

View reviewed changes

lld/test/wasm/pie.s Show resolved Hide resolved

lld/test/wasm/pie.s: simplify a bit

7c137aa

sbc100 merged commit 9247be8 into llvm:main Aug 19, 2025
9 checks passed

sbc100 added a commit that referenced this pull request Aug 19, 2025

Revert "[lld][WebAssembly] Do not relocate ABSOLUTE symbols (#153763)"

8edcf65

This reverts commit 9247be8.

sbc100 mentioned this pull request Aug 19, 2025

Revert "[lld][WebAssembly] Do not relocate ABSOLUTE symbols" #154371

Merged

sbc100 added a commit that referenced this pull request Aug 19, 2025

Revert "[lld][WebAssembly] Do not relocate ABSOLUTE symbols" (#154371)

28f2fb2

Reverts #153763 This caused a bunch of failures on the emscripten waterfall, specifically most of the lsan tests started failing.

[lld][WebAssembly] Do not relocate ABSOLUTE symbols #153763

[lld][WebAssembly] Do not relocate ABSOLUTE symbols #153763

Uh oh!

Conversation

yamt commented Aug 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Aug 15, 2025

Uh oh!

llvmbot commented Aug 15, 2025

Uh oh!

sbc100 left a comment

Choose a reason for hiding this comment

Uh oh!

sbc100 Aug 15, 2025

Choose a reason for hiding this comment

Uh oh!

yamt Aug 18, 2025

Choose a reason for hiding this comment

Uh oh!

yamt Aug 18, 2025

Choose a reason for hiding this comment

Uh oh!

yamt commented Aug 18, 2025

Uh oh!

sbc100 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

sbc100 commented Aug 19, 2025

Uh oh!

yamt commented Aug 20, 2025

Uh oh!

sbc100 commented Aug 20, 2025

Uh oh!

yamt commented Aug 20, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

yamt commented Aug 15, 2025 •

edited

Loading