Handle statics and TLS in raw-dylib for ELF by mati865 · Pull Request #153580 · rust-lang/rust

mati865 · 2026-03-08T22:50:40Z

Follow-up to #153090

rustbot · 2026-03-08T22:50:45Z

rustbot has assigned @jieyouxu.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

Owners of files modified in this PR: compiler
compiler expanded to 69 candidates
Random selection from 15 candidates

mati865 · 2026-03-08T22:50:51Z

cc @bjorn3

jieyouxu · 2026-03-09T08:17:29Z

Probably, r? bjorn3

mati865 · 2026-03-09T08:31:25Z

Shoot, forgot to change the title yesterday.

compiler/rustc_metadata/src/native_libs.rs

bjorn3 · 2026-03-09T09:35:36Z

Actually, for PIE executables we do also need to store the symbol size for statics and maybe thread locals. This because of copy relocations. Basically the linker reserves space in the PIE executable for all statics coming from other DSOs and then at runtime the dynamic linker copies the data from the DSO that the symbol resolves to in the end into the reserved memory. Without the symbol size, the linker can't know how much memory to reserve.

mati865 · 2026-03-09T13:47:10Z

Copy relocations would definitely require known size. I couldn't manage to write Rust code that would make LLVM emit direct access with PIE or PIC, it would use GOT indirection instead which prevents copy relocations. However, I sort of succeeded with static relocation model:

GNU ld didn't create copy relocation, TBH I have no idea what it did but printed value looks like an address in .rodata
LLD refused to link with the error cannot create a copy relocation for symbol global_variable
Wild created a broken copy relocation, when trying to run it, glibc shows message about global_variable size difference and the binary hangs

Dunno about TLS, I'm not familiar enough with how it's handled.

More details

I have used WILD_SAVE_BASE to make debugging easier.
Linking with the test case different linkers:

❯ OUT=bin.ld ./run-with ld.bfd

❯ OUT=bin.lld ./run-with ld.lld
ld.lld: error: cannot create a copy relocation for symbol global_variable

# --sym-info to show what we are dealing with
❯ OUT=bin.wild ./run-with ~/Projects/wild/target/release/wild --sym-info global_variable
Global name `global_variable` refers to: Some(sym-20922)
Definitions / references with name `global_variable`:
  109 -> 20922: Global Undefined: ABSOLUTE
    global_variable
    #31 in File #260 (1/4) ./home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/main.main.d17f5fbe6225cf88-cgu.0.rcgu.o (LOADED)
  20922: Global Data: DYNAMIC | DIRECT | COPY_RELOCATION
    global_variable
    #2 in File #15109 (59/5) /tmp/raw-elf/12/home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/rustchF2qGE/raw-dylibs/lib85yu6bswlwqzp4104wz1sxvsr.so (LOADED)

❯ LD_LIBRARY_PATH=~/Projects/rust/build/host/test/run-make/raw-dylib-elf/rmake_out ./bin.ld
42 4546626 33

❯ LD_LIBRARY_PATH=~/Projects/rust/build/host/test/run-make/raw-dylib-elf/rmake_out ./bin.wild
./bin.wild: Symbol `global_variable' has different size in shared object, consider re-linking

# 4546626 is 0x456042 in hex
❯ readelf -WS bin.ld | rg -C1 456
  [13] .fini             PROGBITS        0000000000455034 055034 00000d 00  AX  0   0  4
  [14] .rodata           PROGBITS        0000000000456000 056000 00ca00 00   A  0   0 16
  [15] .eh_frame_hdr     PROGBITS        0000000000462a00 062a00 001554 00   A  0   0  4

While at it, posting linker-diff for the future reference:

❯ WILD_REFERENCE_LINKER=ld.bfd ./run-with ~/Projects/wild/target/release/wild
wild: ./bin
ref: ./bin.ref-linker
dynsym.global_variable.section
  wild .bss
  ref .data.rel.ro

dynsym.this_is_a_library_function.section
  wild .bss
  ref

dynsym.tls_variable.section
  wild .bss
  ref

rel.R_X86_64_PLT32.R_X86_64_PLT32
  `./home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/main.main.d17f5fbe6225cf88-cgu.0.rcgu.o` .text._RNvCshZ9kMv8z9bM_4main4main _RNvCshZ9kMv8z9bM_4main4main
  ORIG 0x000007: [ e8 00 00 00 00 ] call 0x000000000000000C
                      ^^^^^^^^^^^ R_X86_64_PLT32
  ORIG this_is_a_library_function -4
  wild 0x4194a7: [ e8 34 fe ff ff ] call 0x00000000004192E0
                      ^^^^^^^^^^^ R_X86_64_PLT32 NoOp
  wild PLT->GOT->R_X86_64_GLOB_DAT->this_is_a_library_function@*global*
  wild TRACE: relocation applied flags=DYNAMIC | FUNCTION | GOT | PLT | COPY_RELOCATION,
  wild TRACE: rel_kind=PltRelative,
  wild TRACE: value=0xfffffffffffffe34, symbol_name=this_is_a_library_function
  ref  0x408af7: [ e8 44 95 ff ff ] call 0x0000000000402040
                      ^^^^^^^^^^^ R_X86_64_PLT32 NoOp
  ref  PLT->GOT->R_X86_64_JUMP_SLOT->this_is_a_library_function@*unversioned*

rel.extra-copy-relocation.R_X86_64_64
  `./home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/main.main.d17f5fbe6225cf88-cgu.0.rcgu.o` .text._RNvCshZ9kMv8z9bM_4main4main _RNvCshZ9kMv8z9bM_4main4main
  ORIG 0x000010: [ 48 b8 00 00 00 00 00 00 00 00 ] movabs $0,%rax
                         ^^^^^^^^^^^^^^^^^^^^^^^ R_X86_64_64
  ORIG global_variable
  wild 0x4194b0: [ 48 b8 7c 3b 47 00 00 00 00 00 ] movabs $0x473B7C,%rax
                         ^^^^^^^^^^^^^^^^^^^^^^^ R_X86_64_64 NoOp
  wild COPY(global_variable)
  wild TRACE: relocation applied flags=DYNAMIC | DIRECT | COPY_RELOCATION,
  wild TRACE: rel_kind=Absolute,
  wild TRACE: value=0x473b7c, symbol_name=global_variable
  ref  0x408b00: [ 48 b8 70 c0 46 00 00 00 00 00 ] movabs $0x46C070,%rax
                         ^^^^^^^^^^^^^^^^^^^^^^^ R_X86_64_64 NoOp
  ref  0x46c070 (symbol is undefined)

rel.extra-copy-relocation.R_X86_64_64
  `./home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/main.main.d17f5fbe6225cf88-cgu.0.rcgu.o` .text._RNvCshZ9kMv8z9bM_4main4main _RNvCshZ9kMv8z9bM_4main4main
  ORIG 0x00009b: [ 48 b9 00 00 00 00 00 00 00 00 ] movabs $0,%rcx
                         ^^^^^^^^^^^^^^^^^^^^^^^ R_X86_64_64
  ORIG global_variable
  wild 0x41953b: [ 48 b9 7c 3b 47 00 00 00 00 00 ] movabs $0x473B7C,%rcx
                         ^^^^^^^^^^^^^^^^^^^^^^^ R_X86_64_64 NoOp
  wild COPY(global_variable)
  wild TRACE: relocation applied flags=DYNAMIC | DIRECT | COPY_RELOCATION,
  wild TRACE: rel_kind=Absolute,
  wild TRACE: value=0x473b7c, symbol_name=global_variable
  ref  0x408b8b: [ 48 b9 70 c0 46 00 00 00 00 00 ] movabs $0x46C070,%rcx
                         ^^^^^^^^^^^^^^^^^^^^^^^ R_X86_64_64 NoOp
  ref  0x46c070 (symbol is undefined)

mati865 · 2026-03-09T13:52:27Z

(forgot to write about it in the previous comment)

I have no idea how we can deal with the unknown size. Perhaps we can somehow avoid direct access to raw-dylib statics?

bjorn3 · 2026-03-09T14:02:25Z

We can derive the size of statics from the type layout, right? We don't support unsized statics.

mati865 · 2026-03-09T14:45:39Z

Right, I feel so dumb. For some reason I thought that size of "string" objects would store the length of the underlying string rather than pointer size 🤦🏻‍♂️.

mati865 · 2026-03-09T19:16:42Z

For some reason I thought that size of "string" objects would store the length of the underlying string rather than pointer size

To my defence, this is the case with char foo[] = "foo" strings, not with `char *foo = "foo". I have confused these.

Added size with implementation partially based on i686_arg_list_size and extended test with different relocation models and more types. Static relocation model is just a bombsite.

GNU ld doesn't look half bad, it didn't emit any copy relocations. One big issue standing out is putting data into .data.rel.ro but not producing relocations for them:

❯ readelf -Ws bin.ld | rg '(global|tls)_variable|const_(array|string)|:$'
Symbol table '.dynsym' contains 73 entries:
    39: 0000000000000000     0 TLS     GLOBAL DEFAULT  UND tls_variable
    70: 000000000046d038     8 OBJECT  GLOBAL DEFAULT   25 const_string
    71: 000000000046d040     4 OBJECT  GLOBAL DEFAULT   25 global_variable
    72: 000000000046d044    16 OBJECT  GLOBAL DEFAULT   25 const_array
Symbol table '.symtab' contains 1017 entries:
   749: 000000000046d038     8 OBJECT  GLOBAL DEFAULT   25 const_string
   817: 0000000000000000     0 TLS     GLOBAL DEFAULT  UND tls_variable
   923: 000000000046d040     4 OBJECT  GLOBAL DEFAULT   25 global_variable
   992: 000000000046d044    16 OBJECT  GLOBAL DEFAULT   25 const_array

❯ readelf -WS bin.ld | rg 'data\.rel\.ro|:$'
There are 34 section headers, starting at offset 0x90478:
Section Headers:
  [25] .data.rel.ro      PROGBITS        000000000046d038 06d038 004450 00  WA  0   0  8

❯ readelf -Wr bin.ld | rg '46d0|tls_variable|:$'
Relocation section '.rela.dyn' at offset 0x1100 contains 66 entries:
0000000000471b50  0000002700000012 R_X86_64_TPOFF64       0000000000000000 tls_variable + 0
Relocation section '.rela.plt' at offset 0x1730 contains 3 entries:

Somehow this works, except for const_string which crashes the program.

LLD didn't like our raw-dylib at all. With sizes set, it no longer rejects linkage, but it'd be better if it did rather than creating this abomination:

# Nice, clashing virtual addresses
❯ readelf -Ws bin.lld | rg '(global|tls)_variable|const_(array|string)|:$'
Symbol table '.dynsym' contains 73 entries:
     5: 0000000000000000     0 TLS     GLOBAL DEFAULT  UND tls_variable
    70: 000000000027449a     4 OBJECT  GLOBAL DEFAULT   30 global_variable
    71: 000000000027449a    16 OBJECT  GLOBAL DEFAULT   30 const_array
    72: 000000000027449a     8 OBJECT  GLOBAL DEFAULT   30 const_string
Symbol table '.symtab' contains 1279 entries:
   925: 000000000027449a     4 OBJECT  GLOBAL DEFAULT   30 global_variable
   926: 0000000000000000     0 TLS     GLOBAL DEFAULT  UND tls_variable
   929: 000000000027449a    16 OBJECT  GLOBAL DEFAULT   30 const_array
   930: 000000000027449a     8 OBJECT  GLOBAL DEFAULT   30 const_string

# 0x2743d0 + 0xce = 0x277449e (final offset of .bss)
❯ readelf -WS bin.lld | rg '\.bss|:$'
There are 35 section headers, starting at offset 0x91c38:
Section Headers:
  [30] .bss              NOBITS          00000000002743d0 0713d0 0000ce 00  WA  0   0  8

❯ readelf -Wr bin.lld | rg '449a|tls_variable|:$'
Relocation section '.rela.dyn' at offset 0x1078 contains 67 entries:
0000000000272048  0000000500000012 R_X86_64_TPOFF64       0000000000000000 tls_variable + 0
000000000027449a  0000004600000005 R_X86_64_COPY          000000000027449a global_variable + 0
Relocation section '.rela.plt' at offset 0x16c0 contains 3 entries:
00000000002729d8  0000004500000007 R_X86_64_JUMP_SLOT     000000000027449a this_is_a_library_function + 0

Yeah, this has zero chance of working.

Wild starts off with warnings when linking the same code:

WARNING: wild: Multiple non-weak symbols at the same address have copy relocations: symbol `const_string` (20984 local=5) in file #15365 (60/5) (/tmp/raw-elf/49/home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/rustc8RupL2/raw-dylibs/libbivurdqliqxle18nnha8ay1b9.so) (DYNAMIC | DIRECT | COPY_RELOCATION), symbol `const_array` (20983 local=4) in file #15365 (60/5) (/tmp/raw-elf/49/home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/rustc8RupL2/raw-dylibs/libbivurdqliqxle18nnha8ay1b9.so) (DYNAMIC | DIRECT | COPY_RELOCATION)
WARNING: wild: Multiple non-weak symbols at the same address have copy relocations: symbol `const_array` (20983 local=4) in file #15365 (60/5) (/tmp/raw-elf/49/home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/rustc8RupL2/raw-dylibs/libbivurdqliqxle18nnha8ay1b9.so) (DYNAMIC | DIRECT | COPY_RELOCATION), symbol `global_variable` (20981 local=2) in file #15365 (60/5) (/tmp/raw-elf/49/home/mateusz/Projects/rust/build/x86_64-unknown-linux-gnu/test/run-make/raw-dylib-elf/rmake_out/rustc8RupL2/raw-dylibs/libbivurdqliqxle18nnha8ay1b9.so) (DYNAMIC | DIRECT | COPY_RELOCATION)

And looks like we have a deja vu:

❯ readelf -Ws bin.wild | rg '(global|tls)_variable|const_(array|string)|:$'
Symbol table '.dynsym' contains 72 entries:
    67: 00000000004748dc    16 OBJECT  GLOBAL DEFAULT   29 const_array
    68: 00000000004748dc     8 OBJECT  GLOBAL DEFAULT   29 const_string
    69: 00000000004748dc     4 OBJECT  GLOBAL DEFAULT   29 global_variable
    71: 00000000004748dc     4 TLS     GLOBAL DEFAULT   29 tls_variable
Symbol table '.symtab' contains 958 entries:
   953: 00000000004748dc     4 OBJECT  GLOBAL DEFAULT   29 global_variable
   954: 00000000004748dc     4 TLS     GLOBAL DEFAULT   29 tls_variable
   955: 00000000004748dc    16 OBJECT  GLOBAL DEFAULT   29 const_array
   956: 00000000004748dc     8 OBJECT  GLOBAL DEFAULT   29 const_string

# 0x474838 + 0xa8 = 0x4748e0 (final offset of .bss)
❯ readelf -WS bin.wild | rg '\.bss|:$'
There are 34 section headers, starting at offset 0x350:
Section Headers:
  [29] .bss              NOBITS          0000000000474838 071838 0000a8 00  WA  0   0  8

❯ readelf -Wr bin.wild | rg '48dc|tls_variable|:$'
Relocation section '.rela.dyn' at offset 0x1970 contains 69 entries:
00000000004748dc  0000004500000005 R_X86_64_COPY          00000000004748dc global_variable + 0
0000000000472e58  0000004700000012 R_X86_64_TPOFF64       00000000004748dc tls_variable + 0

How about making static relocation model unsupported for now (with an error emitted) and creating an issue to track fixing it?

mati865 · 2026-03-13T19:13:16Z

I think I have figured it out, but I need to double-check that before marking as ready. Pushed just to see if AArch64 also works fine.

FYI, there are 2 new changes: Put statics in .data section in raw-dylib for ELF
and Set symbols value in raw-dylib for ELF.
Extend raw-dylib test for ELF with reloc models is just split out from 2e52da9.

mati865 · 2026-03-16T01:36:43Z

Sorry for the constant changes but on the second thought (or rather third in this case) I this bumping the value by the unrelated section alignment might be confusing. Changed that to more intuitive 8 bytes.

bjorn3

r=me with or without replacing the unknown symbol kind with a bug!().

View changes since this review

This fixes issues with statics when using static relocation model and linking with GNU ld. Other linkers don't work yet.

With this change LLD finally produces working binaries with static relocation model.

mati865 · 2026-03-17T22:10:32Z

@bors try jobs=aarch64-msvc-1,aarch64-msvc-2,x86_64-mingw-1,x86_64-mingw-2

Handle statics and TLS in raw-dylib for ELF try-job: aarch64-msvc-1 try-job: aarch64-msvc-2 try-job: x86_64-mingw-1 try-job: x86_64-mingw-2

rust-bors · 2026-03-18T01:12:07Z

☀️ Try build successful (CI)
Build commit: 184392a (184392a3b8d1456276e186c1e66e475f8cb3899e, parent: 85e19b8ce869aaef0d5389a050ddee7b258b8029)

mati865 · 2026-03-18T08:41:40Z

@bors r=bjorn3

rust-bors · 2026-03-18T08:41:44Z

📌 Commit 3531ed9 has been approved by bjorn3

It is now in the queue for this repository.

Rollup of 3 pull requests Successful merges: - #153580 (Handle statics and TLS in raw-dylib for ELF) - #154020 (Moved tests/ui/issues/issue-23891.rs to tests/ui/macros/) - #154028 (Rename trait `IntoQueryParam` to `IntoQueryKey`)

matthiaskrgr · 2026-03-19T19:49:02Z

compiler/rustc_metadata/src/native_libs.rs

+                DllImportSymbolType::Static
+            }
+        } else {
+            bug!("Unexpected type for raw-dylib: {}", def_kind.descr(item));


this is reachable via #154111

Ah, I have tried to use type there, but only as a definition. Not as a declaration, I'll look into it tomorrow.

rustbot assigned jieyouxu Mar 8, 2026

rustbot added A-run-make Area: port run-make Makefiles to rmake.rs S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Mar 8, 2026

rustbot assigned bjorn3 and unassigned jieyouxu Mar 9, 2026

mati865 changed the title ~~Elf raw dylib static and tls~~ Handle statics and TLS in raw-dylib for ELF Mar 9, 2026

bjorn3 reviewed Mar 9, 2026

View reviewed changes

compiler/rustc_metadata/src/native_libs.rs Outdated Show resolved Hide resolved

bjorn3 approved these changes Mar 9, 2026

View reviewed changes

mati865 marked this pull request as draft March 9, 2026 14:45

rustbot added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Mar 9, 2026

mati865 force-pushed the elf-raw-dylib-static-and-tls branch 2 times, most recently from 9cb0ca1 to 2e52da9 Compare March 9, 2026 18:24

This comment has been minimized.

Sign in to view

mati865 force-pushed the elf-raw-dylib-static-and-tls branch from 2e52da9 to dc2e08d Compare March 13, 2026 19:05

This comment has been minimized.

Sign in to view

mati865 force-pushed the elf-raw-dylib-static-and-tls branch from dc2e08d to e14cce4 Compare March 13, 2026 19:13

mati865 force-pushed the elf-raw-dylib-static-and-tls branch from e14cce4 to b18def1 Compare March 14, 2026 12:24

mati865 marked this pull request as ready for review March 14, 2026 12:32

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Mar 14, 2026

mati865 force-pushed the elf-raw-dylib-static-and-tls branch from b18def1 to 0d060dd Compare March 16, 2026 01:30

bjorn3 approved these changes Mar 17, 2026

View reviewed changes

mati865 added 6 commits March 17, 2026 21:21

Refactor DllImport to better differentiate types

82727b0

Extend raw-dylib to handle statics and TLS for ELF

2fad36b

Set symbol size in raw-dylib for ELF

f365432

Put statics in .data section in raw-dylib for ELF

3638155

This fixes issues with statics when using static relocation model and linking with GNU ld. Other linkers don't work yet.

Set symbols value in raw-dylib for ELF

ed4afa5

With this change LLD finally produces working binaries with static relocation model.

Extend raw-dylib test for ELF with reloc models

3531ed9

mati865 force-pushed the elf-raw-dylib-static-and-tls branch from 0d060dd to 3531ed9 Compare March 17, 2026 20:31

This comment has been minimized.

Sign in to view

rust-bors bot added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Mar 18, 2026

Zalathar mentioned this pull request Mar 18, 2026

Rollup of 3 pull requests #154039

Merged

rust-bors bot merged commit c0172a3 into rust-lang:main Mar 18, 2026
12 checks passed

rustbot added this to the 1.96.0 milestone Mar 18, 2026

mati865 deleted the elf-raw-dylib-static-and-tls branch March 18, 2026 18:10

mati865 mentioned this pull request Mar 18, 2026

Tracking Issue for raw-dylib on ELF #135694

Open

5 tasks

matthiaskrgr mentioned this pull request Mar 19, 2026

[ICE]: Unexpected type for raw-dylib: foreign type #154111

Open

matthiaskrgr reviewed Mar 19, 2026

View reviewed changes

Uh oh!

Conversation

mati865 commented Mar 8, 2026

Uh oh!

rustbot commented Mar 8, 2026

Uh oh!

mati865 commented Mar 8, 2026

Uh oh!

jieyouxu commented Mar 9, 2026

Uh oh!

mati865 commented Mar 9, 2026

Uh oh!

Uh oh!

bjorn3 commented Mar 9, 2026

Uh oh!

mati865 commented Mar 9, 2026

Uh oh!

mati865 commented Mar 9, 2026

Uh oh!

bjorn3 commented Mar 9, 2026

Uh oh!

mati865 commented Mar 9, 2026

Uh oh!

mati865 commented Mar 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

mati865 commented Mar 13, 2026

Uh oh!

mati865 commented Mar 16, 2026

Uh oh!

bjorn3 left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mati865 commented Mar 17, 2026

Uh oh!

This comment has been minimized.

rust-bors bot commented Mar 18, 2026

Uh oh!

mati865 commented Mar 18, 2026

Uh oh!

rust-bors bot commented Mar 18, 2026

Uh oh!

Uh oh!

matthiaskrgr Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

mati865 Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

mati865 commented Mar 9, 2026 •

edited

Loading

bjorn3 left a comment •

edited by rustbot

Loading