[lld][COFF] Restore `lto-embed-bitcode` and `-fembed-bitcode` Bitcode Embedding Features by realoriginal · Pull Request #188398 · llvm/llvm-project

realoriginal · 2026-03-25T02:22:53Z

Removes the patches introduced by #150897 which broke LTO embed documented features for creating whole-program-bitcode representations of executables, used in production analysis/rewriting toolsets. This was a documented feature available up until 21.1.8 broken by 22.x release.

This previously allowed the users to have a whole-program-bitcode section .llvmbc embedded inside of the final executable.

This reverts commit 41f3332.

)" This reverts commit 07d396b.

realoriginal · 2026-03-25T02:25:49Z

@mstorsjo any way we can get this merged?

HaohaiWen

This will always keep .llvmbc and .llvmcmd in the final binary if -fembed-bitcode is presented.
As I mentioned in the #150897 (comment). Sometimes, we just want it exist in the object/bitcode but not in the final binary to avoid 2GB PE limitation.

HaohaiWen · 2026-03-25T03:30:44Z

llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp

      Name == getInstrProfSectionName(IPSK_covname, Triple::COFF,
-                                      /*AddSegmentInfo=*/false) ||
-      Name == ".llvmbc" || Name == ".llvmcmd")
+                                      /*AddSegmentInfo=*/false))


They're metadata, right?

If it has to be metadata, I'm fine with that, as long it ends up in the final binary as per lto-embed-bitcode's documented feature allows so that myself and others whom build tools on this documented assumption can extract it

Setting is to metadata does not mean lld will drop them automatically.
I think they should be metadata (IMAGE_SCN_MEM_DISCARDABLEj).

I'll wait for what @MaskRay says, so we can get this merged in and fixed quickly

Reverting this also affect our production, as bitcode/cmdline sections are usually too big to be embedded in. Sometimes lld even failed for large application.

llvm-project/lld/COFF/Writer.cpp

Lines 800 to 802 in 1904867

if (fileSize > UINT32_MAX)

Fatal(ctx) << "image size (" << fileSize << ") "

<< "exceeds maximum allowable size (" << UINT32_MAX << ")";

-lto-embed-bitcode is indeed an issue that I did not aware of. I aggress to revert the lld code, and I suggest to add an /strip-embedded-bitcode option so that both products can work.
Regarding metadata, I don't think this will affect existence of those sections. ELF also set them as metadata sections.

Do we really need these sections to remain in the .EXE file? I think it could be straightforward to move them in a PDB named stream? For the reasons @HaohaiWen mentionned, ie. PE has a hard size limit, whereas PDB has a 16 TB limit (when using /PDBPAGESIZE:xxx).

It’s just kind of strange to completely break a documented linker flag that is fully optional and only affects people who explicitly set it.

For ‘-fembed-bitcode` it’s a different story, I wouldn’t expect that to end up in the final executable since enabling it signals you are processing the objects individually anyway.

realoriginal · 2026-03-25T03:39:22Z

@HaohaiWen I would agree with you if the case was it was always embedded, however as far as I've seen there is no code path I've seen that leads to this case. Theres alternatives if you just want the bitcode by itself without it being in the final binary, and the linker flag -lto-embed-bitcode is supposed to embed the bitcode in the final binary. If you don't specify that, it shouldn't end up in the final binary. As far as I've ever seen anyway.

HaohaiWen · 2026-03-25T06:34:12Z

How about adding a new option to explicitly drop it in lld? e.g.
/strip-embedded-bitcode

…0879)" This reverts commit 7f0a008.

realoriginal · 2026-03-25T13:58:26Z

I've reverted the metadata patch so its re-applied, at this point, does at least restoring the original behavior and taking care of an additional flag fine as a seperate PR seem sufficient so this can be merged? @HaohaiWen / @aganea / @MaskRay / @mstorsjo

llvmbot · 2026-03-25T17:47:33Z

@llvm/pr-subscribers-lld-coff

Author: Austin Hudson (realoriginal)

Changes

Removes the patches introduced by #150879 and #150897 which broke LTO embed documented features for creating whole-program-bitcode representations of executables, used in production analysis/rewriting toolsets. This was a documented feature available up until 21.1.8 broken by 22.x release.

This previously allowed the users to have a whole-program-bitcode section .llvmbc embedded inside of the final executable.

If there is anything else necessary to getting this merged, please let me know as this feature is critical to myself and others.

Full diff: https://github.com/llvm/llvm-project/pull/188398.diff

2 Files Affected:

(modified) lld/COFF/InputFiles.cpp (-5)
(removed) lld/test/COFF/embed-bitcode.test (-30)

diff --git a/lld/COFF/InputFiles.cpp b/lld/COFF/InputFiles.cpp
index 492b2ad80166d..4320a39644616 100644
--- a/lld/COFF/InputFiles.cpp
+++ b/lld/COFF/InputFiles.cpp
@@ -404,11 +404,6 @@ SectionChunk *ObjFile::readSection(uint32_t sectionNumber,
     return nullptr;
   }
 
-  // Those sections are generated by -fembed-bitcode and do not need to be kept
-  // in executable files.
-  if (name == ".llvmbc" || name == ".llvmcmd")
-    return nullptr;
-
   // Object files may have DWARF debug info or MS CodeView debug info
   // (or both).
   //
diff --git a/lld/test/COFF/embed-bitcode.test b/lld/test/COFF/embed-bitcode.test
deleted file mode 100644
index 10f88c5c0117d..0000000000000
--- a/lld/test/COFF/embed-bitcode.test
+++ /dev/null
@@ -1,30 +0,0 @@
-# RUN: yaml2obj %s -o %t.obj
-# RUN: lld-link /entry:main /subsystem:console /out:%t.exe %t.obj
-# RUN: llvm-readobj -S %t.exe | FileCheck %s
-
-# CHECK-NOT: Name: .llvmbc
-# CHECK-NOT: Name: .llvmcmd
-
---- !COFF
-header:
-  Machine:         IMAGE_FILE_MACHINE_AMD64
-
-sections:
-  - Name:            .text
-    Characteristics: [ IMAGE_SCN_CNT_CODE, IMAGE_SCN_MEM_EXECUTE, IMAGE_SCN_MEM_READ ]
-    SectionData:     "C3"
-  - Name:           .llvmbc
-    Characteristics: [ IMAGE_SCN_MEM_DISCARDABLE ]
-    SectionData:     "4243C0DE"
-  - Name:           .llvmcmd
-    Characteristics: [ IMAGE_SCN_MEM_DISCARDABLE ]
-    SectionData:     "2D63633100"
-
-symbols:
-  - Name:            main
-    Value:           0
-    SectionNumber:   1
-    SimpleType:      IMAGE_SYM_TYPE_NULL
-    ComplexType:     IMAGE_SYM_DTYPE_FUNCTION
-    StorageClass:    IMAGE_SYM_CLASS_EXTERNAL
-...

llvmbot · 2026-03-25T17:47:33Z

@llvm/pr-subscribers-platform-windows

Author: Austin Hudson (realoriginal)

Changes

Removes the patches introduced by #150879 and #150897 which broke LTO embed documented features for creating whole-program-bitcode representations of executables, used in production analysis/rewriting toolsets. This was a documented feature available up until 21.1.8 broken by 22.x release.

This previously allowed the users to have a whole-program-bitcode section .llvmbc embedded inside of the final executable.

If there is anything else necessary to getting this merged, please let me know as this feature is critical to myself and others.

Full diff: https://github.com/llvm/llvm-project/pull/188398.diff

2 Files Affected:

(modified) lld/COFF/InputFiles.cpp (-5)
(removed) lld/test/COFF/embed-bitcode.test (-30)

diff --git a/lld/COFF/InputFiles.cpp b/lld/COFF/InputFiles.cpp
index 492b2ad80166d..4320a39644616 100644
--- a/lld/COFF/InputFiles.cpp
+++ b/lld/COFF/InputFiles.cpp
@@ -404,11 +404,6 @@ SectionChunk *ObjFile::readSection(uint32_t sectionNumber,
     return nullptr;
   }
 
-  // Those sections are generated by -fembed-bitcode and do not need to be kept
-  // in executable files.
-  if (name == ".llvmbc" || name == ".llvmcmd")
-    return nullptr;
-
   // Object files may have DWARF debug info or MS CodeView debug info
   // (or both).
   //
diff --git a/lld/test/COFF/embed-bitcode.test b/lld/test/COFF/embed-bitcode.test
deleted file mode 100644
index 10f88c5c0117d..0000000000000
--- a/lld/test/COFF/embed-bitcode.test
+++ /dev/null
@@ -1,30 +0,0 @@
-# RUN: yaml2obj %s -o %t.obj
-# RUN: lld-link /entry:main /subsystem:console /out:%t.exe %t.obj
-# RUN: llvm-readobj -S %t.exe | FileCheck %s
-
-# CHECK-NOT: Name: .llvmbc
-# CHECK-NOT: Name: .llvmcmd
-
---- !COFF
-header:
-  Machine:         IMAGE_FILE_MACHINE_AMD64
-
-sections:
-  - Name:            .text
-    Characteristics: [ IMAGE_SCN_CNT_CODE, IMAGE_SCN_MEM_EXECUTE, IMAGE_SCN_MEM_READ ]
-    SectionData:     "C3"
-  - Name:           .llvmbc
-    Characteristics: [ IMAGE_SCN_MEM_DISCARDABLE ]
-    SectionData:     "4243C0DE"
-  - Name:           .llvmcmd
-    Characteristics: [ IMAGE_SCN_MEM_DISCARDABLE ]
-    SectionData:     "2D63633100"
-
-symbols:
-  - Name:            main
-    Value:           0
-    SectionNumber:   1
-    SimpleType:      IMAGE_SYM_TYPE_NULL
-    ComplexType:     IMAGE_SYM_DTYPE_FUNCTION
-    StorageClass:    IMAGE_SYM_CLASS_EXTERNAL
-...

llvmbot · 2026-03-25T17:47:34Z

@llvm/pr-subscribers-lld

Author: Austin Hudson (realoriginal)

Changes

Removes the patches introduced by #150879 and #150897 which broke LTO embed documented features for creating whole-program-bitcode representations of executables, used in production analysis/rewriting toolsets. This was a documented feature available up until 21.1.8 broken by 22.x release.

This previously allowed the users to have a whole-program-bitcode section .llvmbc embedded inside of the final executable.

If there is anything else necessary to getting this merged, please let me know as this feature is critical to myself and others.

Full diff: https://github.com/llvm/llvm-project/pull/188398.diff

2 Files Affected:

(modified) lld/COFF/InputFiles.cpp (-5)
(removed) lld/test/COFF/embed-bitcode.test (-30)

diff --git a/lld/COFF/InputFiles.cpp b/lld/COFF/InputFiles.cpp
index 492b2ad80166d..4320a39644616 100644
--- a/lld/COFF/InputFiles.cpp
+++ b/lld/COFF/InputFiles.cpp
@@ -404,11 +404,6 @@ SectionChunk *ObjFile::readSection(uint32_t sectionNumber,
     return nullptr;
   }
 
-  // Those sections are generated by -fembed-bitcode and do not need to be kept
-  // in executable files.
-  if (name == ".llvmbc" || name == ".llvmcmd")
-    return nullptr;
-
   // Object files may have DWARF debug info or MS CodeView debug info
   // (or both).
   //
diff --git a/lld/test/COFF/embed-bitcode.test b/lld/test/COFF/embed-bitcode.test
deleted file mode 100644
index 10f88c5c0117d..0000000000000
--- a/lld/test/COFF/embed-bitcode.test
+++ /dev/null
@@ -1,30 +0,0 @@
-# RUN: yaml2obj %s -o %t.obj
-# RUN: lld-link /entry:main /subsystem:console /out:%t.exe %t.obj
-# RUN: llvm-readobj -S %t.exe | FileCheck %s
-
-# CHECK-NOT: Name: .llvmbc
-# CHECK-NOT: Name: .llvmcmd
-
---- !COFF
-header:
-  Machine:         IMAGE_FILE_MACHINE_AMD64
-
-sections:
-  - Name:            .text
-    Characteristics: [ IMAGE_SCN_CNT_CODE, IMAGE_SCN_MEM_EXECUTE, IMAGE_SCN_MEM_READ ]
-    SectionData:     "C3"
-  - Name:           .llvmbc
-    Characteristics: [ IMAGE_SCN_MEM_DISCARDABLE ]
-    SectionData:     "4243C0DE"
-  - Name:           .llvmcmd
-    Characteristics: [ IMAGE_SCN_MEM_DISCARDABLE ]
-    SectionData:     "2D63633100"
-
-symbols:
-  - Name:            main
-    Value:           0
-    SectionNumber:   1
-    SimpleType:      IMAGE_SYM_TYPE_NULL
-    ComplexType:     IMAGE_SYM_DTYPE_FUNCTION
-    StorageClass:    IMAGE_SYM_CLASS_EXTERNAL
-...

mstorsjo · 2026-03-27T09:54:40Z

If @HaohaiWen is ok with doing this (and looking into adding an option for this afterwards), then I guess we can go ahead and merge this (after a couple of CI checks run complete).

mstorsjo · 2026-03-27T10:07:53Z

@realoriginal For merging, I noticed that you email address on github is set to private; see https://github.com/llvm/llvm-project/blob/main/.github/workflows/email-check.yaml#L34-L37 for our policy on the matter.

realoriginal · 2026-03-27T22:50:06Z

Good to go, I've made it public @mstorsjo

MaskRay · 2026-03-28T18:54:33Z

lld/test/COFF/embed-bitcode.test

-# RUN: lld-link /entry:main /subsystem:console /out:%t.exe %t.obj
-# RUN: llvm-readobj -S %t.exe | FileCheck %s
-
-# CHECK-NOT: Name: .llvmbc


Instead of deleting the test, you can check that they are still present

mstorsjo · 2026-03-30T13:07:25Z

As the regression was in 22.x, I think this should be backported as well.

mstorsjo · 2026-03-30T13:07:53Z

/cherry-pick 1e99c9e

llvmbot · 2026-03-30T13:15:05Z

/pull-request #189375

Add a -strip-embedded-bitcode option so that users who compile with -fembed-bitcode but don't want these sections in the final binary can explicitly opt out, while preserving the default behavior. The .llvmbc and .llvmcmd sections was previously stripped from the final binary unconditionally (llvm#150897). However, this broke the workflow of -lto-embed-bitcode and llvm#188398 reverted it. The test of this PR is from llvm#150897.

… Embedding Features (llvm#188398) Removes the patches introduced by llvm#150897 which broke LTO embed documented features for creating whole-program-bitcode representations of executables, used in production analysis/rewriting toolsets. This was a documented feature available up until 21.1.8 broken by 22.x release. This previously allowed the users to have a whole-program-bitcode section `.llvmbc` embedded inside of the final executable. (cherry picked from commit 1e99c9e)

realoriginal added 2 commits March 24, 2026 19:17

Revert "[LLD][COFF] Discard .llvmbc and .llvmcmd sections (llvm#150897)"

08bccbf

This reverts commit 41f3332.

Revert "[COFF] Set .llvmbc and .llvmcmd to metadata section (llvm#150879

7f0a008

)" This reverts commit 07d396b.

realoriginal changed the title ~~[lld][COFF] Restore lto-embed-bitcode and -fembed-bitcode Bitcode Embedding Features From~~ [lld][COFF] Restore lto-embed-bitcode and -fembed-bitcode Bitcode Embedding Features Mar 25, 2026

HaohaiWen reviewed Mar 25, 2026

View reviewed changes

HaohaiWen requested a review from MaskRay March 25, 2026 03:33

realoriginal mentioned this pull request Mar 25, 2026

[LLD][COFF] Discard .llvmbc and .llvmcmd sections #150897

Merged

Reapply "[COFF] Set .llvmbc and .llvmcmd to metadata section (llvm#15…

16b841f

…0879)" This reverts commit 7f0a008.

realoriginal force-pushed the llvm-22.x-restore-embed-bitcode-feature branch from 951486c to 16b841f Compare March 25, 2026 13:27

Merge branch 'main' into llvm-22.x-restore-embed-bitcode-feature

6c06b7f

HaohaiWen approved these changes Mar 25, 2026

View reviewed changes

Merge branch 'main' into llvm-22.x-restore-embed-bitcode-feature

29bc199

llvmbot added lld lld:COFF platform:windows labels Mar 25, 2026

MaskRay approved these changes Mar 28, 2026

View reviewed changes

MaskRay reviewed Mar 28, 2026

View reviewed changes

mstorsjo merged commit 1e99c9e into llvm:main Mar 30, 2026
13 checks passed

mstorsjo added this to the LLVM 22.x Release milestone Mar 30, 2026

github-project-automation bot added this to LLVM Release Status Mar 30, 2026

github-project-automation bot moved this to Needs Triage in LLVM Release Status Mar 30, 2026

github-project-automation bot moved this from Needs Triage to Done in LLVM Release Status Mar 30, 2026

HaohaiWen mentioned this pull request Mar 31, 2026

[lld][COFF] Add -strip-embedded-bitcode option #189542

Open

	if (fileSize > UINT32_MAX)
	Fatal(ctx) << "image size (" << fileSize << ") "
	<< "exceeds maximum allowable size (" << UINT32_MAX << ")";

Conversation

realoriginal commented Mar 25, 2026 • edited by mstorsjo Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

realoriginal commented Mar 25, 2026

Uh oh!

HaohaiWen left a comment

Choose a reason for hiding this comment

Uh oh!

HaohaiWen Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

realoriginal Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HaohaiWen Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

realoriginal Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HaohaiWen Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

aganea Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mrexodia Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

realoriginal commented Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

HaohaiWen commented Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

realoriginal commented Mar 25, 2026

Uh oh!

llvmbot commented Mar 25, 2026

Uh oh!

llvmbot commented Mar 25, 2026

Uh oh!

llvmbot commented Mar 25, 2026

Uh oh!

mstorsjo commented Mar 27, 2026

Uh oh!

mstorsjo commented Mar 27, 2026

Uh oh!

realoriginal commented Mar 27, 2026

Uh oh!

MaskRay Mar 28, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

mstorsjo commented Mar 30, 2026

Uh oh!

mstorsjo commented Mar 30, 2026

Uh oh!

llvmbot commented Mar 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

realoriginal commented Mar 25, 2026 •

edited by mstorsjo

Loading

realoriginal Mar 25, 2026 •

edited

Loading

realoriginal Mar 25, 2026 •

edited

Loading

HaohaiWen Mar 25, 2026 •

edited

Loading

aganea Mar 25, 2026 •

edited

Loading

realoriginal commented Mar 25, 2026 •

edited

Loading

HaohaiWen commented Mar 25, 2026 •

edited

Loading