zig stuck at llvm emit object on riscv64-linux sifive_u74

[RossComputerGuy]

Zig Version

0.12.0-dev.2058+04ac028a2

Steps to Reproduce and Observed Behavior

Run zig build -Dcpu=sifive_u74 in this repo on a Sifive U74 system such as a VisionFive 2. This can also occur with zig-bootstrap (even if updated to use zig master).

After waiting 3 hours or longer, nothing happens during the LLVM Emit Object stage for building the zig executable itself. On the other hand, the build runnable builds in 6 minutes on the VisionFive 2 so this likely isn't just a LLVM is entirely broken.

Expected Behavior

Zig should finish.

[Vexu]

Are you also setting -Dtarget=riscv64-linux? Without it I get a panic on this line (which is a bug in itself #18876).

With riscv32 I get error: unsupported arch and with riscv64 error: root struct of file 'os.linux.riscv64' has no member named 'ucontext_t'.

[RossComputerGuy]

I have a few PR's merged, the RISC-V glibc, CPU model, and the other one. I don't set the target when I run on the VisionFive2 but I do when zig-bootstrap or cross compiling.

[Vexu]

By the mention of the glibc PR I'm guessing that you're linking libc?

[RossComputerGuy]

No, I'm not linking libc. I just have the patch.

[Vexu]

Then I'm not sure how you're getting to LLVM Emit Object unless you're talking about building the build script? Does zig build -Dno-lib -Dcpu=sifive_u74 -Dtarget=riscv64-linux --help print usage for you?

[RossComputerGuy]

It's not with the builder, it's with building a zig test or zig executable itself. It just gets stuck. Tried getting the tests for std itself to run but it's been stuck for like 14 hours.

$ ~/zig/zig test -ODebug -I /home/user/ExpidusOS/zig/test -Mroot=/home/user/ExpidusOS/zig/lib/std/std.zig --cache-dir /home/user/ExpidusOS/zig/zig-cache --global-cache-dir /home/user/.cache/zig --name test --zig-lib-dir /home/user/ExpidusOS/zig/lib -target riscv64-linux-musl -lc -mcpu baseline_rv64 --verbose-cc
LLVM Emit Object..

~/zig is an rsync copy of a baseline_rv64 zig-bootstrap build of Zig. You can try reproducing this with building zig-bootstrap with riscv64-linux sifive_u74.

[RossComputerGuy]

I've found the problem:

AttemptToFoldSymbolOffsetDifference(llvm::MCAssembler const*, llvm::MCAsmLayout const*, llvm::DenseMap<llvm::MCSection const*, unsigned long, llvm::DenseMapInfo<llvm::MCSection const*, void>, llvm::detail::DenseMapPair<llvm::MCSection const*, unsigned long> > const*, bool, llvm::MCSymbolRefExpr const*&, llvm::MCSymbolRefExpr const*&, long&)

This is where LLVM gets stuck.

Backtrace:

Thread 1 "zig" received signal SIGINT, Interrupt.
0x0000ffffeea65b20 in AttemptToFoldSymbolOffsetDifference(llvm::MCAssembler const*, llvm::MCAsmLayout const*, llvm::DenseMap<llvm::MCSection const*, unsigned long, llvm::DenseMapInfo<llvm::MCSection const*, void>, llvm::detail::DenseMapPair<llvm::MCSection const*, unsigned long> > const*, bool, llvm::MCSymbolRefExpr const*&, llvm::MCSymbolRefExpr const*&, long&) ()
   from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
(gdb) bt
#0  0x0000ffffeea65b20 in AttemptToFoldSymbolOffsetDifference(llvm::MCAssembler const*, llvm::MCAsmLayout const*, llvm::DenseMap<llvm::MCSection const*, unsigned long, llvm::DenseMapInfo<llvm::MCSection const*, void>, llvm::detail::DenseMapPair<llvm::MCSection const*, unsigned long> > const*, bool, llvm::MCSymbolRefExpr const*&, llvm::MCSymbolRefExpr const*&, long&) ()
   from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#1  0x0000ffffeea66044 in EvaluateSymbolicAdd(llvm::MCAssembler const*, llvm::MCAsmLayout const*, llvm::DenseMap<llvm::MCSection const*, unsigned long, llvm::DenseMapInfo<llvm::MCSection const*, void>, llvm::detail::DenseMapPair<llvm::MCSection const*, unsigned long> > const*, bool, llvm::MCValue const&, llvm::MCSymbolRefExpr const*, llvm::MCSymbolRefExpr const*, long, llvm::MCValue&) () from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#2  0x0000ffffeea650d8 in llvm::MCExpr::evaluateAsRelocatableImpl(llvm::MCValue&, llvm::MCAssembler const*, llvm::MCAsmLayout const*, llvm::MCFixup const*, llvm::DenseMap<llvm::MCSection const*, unsigned long, llvm::DenseMapInfo<llvm::MCSection const*, void>, llvm::detail::DenseMapPair<llvm::MCSection const*, unsigned long> > const*, bool) const ()
   from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#3  0x0000ffffeea39e5c in llvm::MCAssembler::evaluateFixup(llvm::MCAsmLayout const&, llvm::MCFixup const&, llvm::MCFragment const*, llvm::MCValue&, unsigned long&, bool&) const ()
   from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#4  0x0000ffffeea3de88 in llvm::MCAssembler::layout(llvm::MCAsmLayout&) () from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#5  0x0000ffffeea3e050 in llvm::MCAssembler::Finish() () from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#6  0x0000ffffeda6e600 in llvm::AsmPrinter::doFinalization(llvm::Module&) () from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#7  0x0000ffffed1b5410 in llvm::FPPassManager::doFinalization(llvm::Module&) () from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#8  0x0000ffffed1c0180 in llvm::legacy::PassManagerImpl::run(llvm::Module&) () from /nix/store/y0pcvaby42h59svn6xzf08ivcs60py6s-llvm-17.0.6-lib/lib/libLLVM-17.so
#9  0x0000000003e09974 in ZigLLVMTargetMachineEmitToFile (targ_machine_ref=0x4446930, module_ref=0x43f8970, error_message=0xfffffffe7400, is_debug=true, is_small=false,
    time_report=false, tsan=false, lto=false, asm_filename=0x0, bin_filename=0x439889e3 "zig.o", llvm_ir_filename=0x0, bitcode_filename=0x0)
    at /home/ross/ExpidusOS/zig/src/zig_llvm.cpp:357
#10 0x000000000194fe94 in codegen.llvm.Object.emit (self=0x442a390, options=...) at /home/ross/ExpidusOS/zig/src/codegen/llvm.zig:1273
#11 0x000000000195ac40 in Compilation.emitLlvmObject (comp=0x4428b90, arena=..., default_emit=..., bin_emit_loc=..., llvm_object=0x442a390, prog_node=0xfffffffeccf8)
    at Compilation.zig:2661
#12 0x000000000244e6cc in link.File.emitLlvmObject (base=..., arena=..., llvm_object=0x442a390, prog_node=0xfffffffeccf8) at link.zig:1001
#13 0x000000000201cc60 in link.Elf.flushModule (self=0x4429ca8, arena=..., prog_node=0xfffffffeccf8) at /home/ross/ExpidusOS/zig/src/link/Elf.zig:1045
#14 0x0000000002013da0 in link.Elf.linkWithLLD (self=0x4429ca8, arena=..., prog_node=0xfffffffeccf8) at /home/ross/ExpidusOS/zig/src/link/Elf.zig:2333
#15 0x0000000001c93588 in link.Elf.flush (self=0x4429ca8, arena=..., prog_node=0xfffffffeccf8) at /home/ross/ExpidusOS/zig/src/link/Elf.zig:1032
#16 0x000000000199bbf8 in link.File.flush (base=0x4429f60, arena=..., prog_node=0xfffffffeccf8) at link.zig:588
#17 0x000000000199b218 in Compilation.flush (comp=0x4428b90, arena=..., prog_node=0xfffffffeccf8) at Compilation.zig:2322
#18 0x000000000199ea10 in Compilation.update (comp=0x4428b90, main_progress_node=0xfffffffeccf8) at Compilation.zig:2313
#19 0x00000000019cab8c in main.updateModule (comp=0x4428b90, color=auto) at main.zig:4502
#20 0x00000000019e7648 in main.buildOutputType (gpa=..., arena=..., all_args=..., arg_mode=...) at main.zig:3355
#21 0x00000000017eff20 in main.mainArgs (gpa=..., arena=..., args=...) at main.zig:277
#22 0x00000000017ece10 in main.main () at main.zig:223

[Vexu]

I was able to reproduce it.

[RossComputerGuy]

That's great, I thought I was going crazy yesterday when Zig was just constantly getting stuck. There's been times where zig-bootstrap work with baseline but there's times where it doesn't work. Should an issue be opened on LLVM's side reporting this bug?

[Vexu]

Yes but it'd be good to investigate a bit first. You happen to have a debug build of LLVM handy? I'll start the build otherwise.