Go version

go version go1.25-prerelease-b062eb46e8 linux/amd64

Output of go env in your module/workspace:

AR='ar'
CC='gcc'
CGO_CFLAGS='-O2 -g'
CGO_CPPFLAGS=''
CGO_CXXFLAGS='-O2 -g'
CGO_ENABLED='1'
CGO_FFLAGS='-O2 -g'
CGO_LDFLAGS='-O2 -g'
CXX='g++'
GCCGO='gccgo'
GO111MODULE='auto'
GOAMD64='v1'
GOARCH='amd64'
GOAUTH='netrc'
GOBIN=''
GOCACHE='/home/jaten/.cache/go-build'
GOCACHEPROG=''
GODEBUG=''
GOENV='/home/jaten/.config/go/env'
GOEXE=''
GOEXPERIMENT=''
GOFIPS140='off'
GOFLAGS=''
GOGCCFLAGS='-fPIC -m64 -pthread -Wl,--no-gc-sections -fmessage-length=0 -ffile-prefix-map=/tmp/go-build2866629629=/tmp/go-build -gno-record-gcc-switches'
GOHOSTARCH='amd64'
GOHOSTOS='linux'
GOINSECURE=''
GOMOD=''
GOMODCACHE='/home/jaten/go/pkg/mod'
GONOPROXY='github.com/glycerine/*'
GONOSUMDB='github.com/glycerine/*'
GOOS='linux'
GOPATH='/home/jaten/go'
GOPRIVATE='github.com/glycerine/*'
GOPROXY='https://proxy.golang.org,direct'
GOROOT='/mnt/oldrog/usr/local/dev-go/go'
GOSUMDB='sum.golang.org'
GOTELEMETRY='off'
GOTELEMETRYDIR='/home/jaten/.config/go/telemetry'
GOTMPDIR=''
GOTOOLCHAIN='auto'
GOTOOLDIR='/mnt/oldrog/usr/local/dev-go/go/pkg/tool/linux_amd64'
GOVCS=''
GOVERSION='go1.25-prerelease-b062eb46e8'
GOWORK=''
PKG_CONFIG='pkg-config'

What did you do?

I was trying to characterize the race detector/tsan bugs found in

https://github.com/golang/go/issues/74019 . While originally found in go1.24.3, these issues have persisted into go1.25 pre-release (b062eb46e8).

See also https://github.com/glycerine/rr_binary_for_issue74019?tab=readme-ov-file#checking-on-go125-prerelease-at-b062eb46e8).

So I tried to build the latest tsan thread-sanitizer from the llvm tree. The thought was, perhaps there have been fixes in the last 20 months to tsan. If bugs have already been fixed in tsan, then maybe that will help with #74019.

https://github.com/golang/go/blob/master/src/runtime/race/README says that the linux amd64 .syso is from 

internal/amd64v1/race_linux.syso built with LLVM 51bfeff0e4b0757ff773da6882f4d538996c9b04 and Go e7d582b55dda36e76ce4d0ce770139ca0915b7c5.

which is from https://github.com/llvm/llvm-project/commit/51bfeff0e4b0757ff773da6882f4d538996c9b04 -- committed 2023 Oct 20; quite a while back.

I executed the same commands that "racebuild" build does in the embedded shell script, since my attempt to run racebuild itself, per the instructions in src/runtime/race/README did not work (it uses gomote, and that needs a special gmail account maybe--I signed in to gmail but do not have permissions); thus the identical commands, as read from https://cs.opensource.google/go/x/build/+/master:cmd/racebuild/racebuild.go for the linux amd64 target, are done manually below.

The attempt to build the race detector .syso for linux amd64 failed with a race itself.

Ignoring the massive irony, fixing the build of tsan could be a good starting point for addressing the

74019 bugs.

jaten@rog /usr $ cd local/dev-go/go/
jaten@rog /usr/local/dev-go/go (go1.25-goj) $ ls
api             CONTRIBUTING.md  lib      PATENTS    SECURITY.md  VERSION
bin             doc              LICENSE  pkg        src          VERSION~
codereview.cfg  go.env           misc     README.md  test
jaten@rog /usr/local/dev-go/go (go1.25-goj) $ cd ..
jaten@rog /usr/local/dev-go $ mkdir tmp
jaten@rog /usr/local/dev-go $ cd tmp
jaten@rog /usr/local/dev-go/tmp $ curl -L -o llvm.zip https://github.com/llvm/llvm-project/archive/46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d.zip
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0
100  318M    0  318M    0     0  6697k      0 --:--:--  0:00:48 --:--:-- 6865k
jaten@rog /usr/local/dev-go/tmp $ unzip -q llvm.zip llvm-project-46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d/compiler-rt/* 
jaten@rog /usr/local/dev-go/tmp $ ls
llvm-project-46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d  llvm.zip
jaten@rog /usr/local/dev-go/tmp $ cd llvm-project-46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d/
jaten@rog /usr/local/dev-go/tmp/llvm-project-46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d $ cd compiler-rt/lib/tsan/go
jaten@rog /usr/local/dev-go/tmp/llvm-project-46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d/compiler-rt/lib/tsan/go $ ls
build.bat  buildgo.sh  test.c  tsan_go.cpp
jaten@rog /usr/local/dev-go/tmp/llvm-project-46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d/compiler-rt/lib/tsan/go $ ./buildgo.sh 
In file included from ../rtl/tsan_sync.h:19,
                 from ../rtl/tsan_rtl.h:47,
                 from tsan_go.cpp:13:
../rtl/tsan_dense_alloc.h: In instantiation of ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Drain(Cache*) [with T = __tsan::MBlock; long unsigned int kL1Size = 262144; long unsigned int kL2Size = 4096; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache]’:
../rtl/tsan_dense_alloc.h:77:7:   required from ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Free(Cache*, IndexT) [with T = __tsan::MBlock; long unsigned int kL1Size = 262144; long unsigned int kL2Size = 4096; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache; IndexT = unsigned int]’
../rtl/tsan_sync.cpp:84:26:   required from here
../rtl/tsan_dense_alloc.h:165:46: warning: suggest parentheses around arithmetic in operand of ‘|’ []8;;https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#index-Wparentheses-Wparentheses]8;;]
  165 |       xchg = head_idx | (cmp & kCounterMask) + kCounterInc;
      |                         ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~
../rtl/tsan_dense_alloc.h: In instantiation of ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Drain(Cache*) [with T = __tsan::SyncVar; long unsigned int kL1Size = 1048576; long unsigned int kL2Size = 1024; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache]’:
../rtl/tsan_dense_alloc.h:77:7:   required from ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Free(Cache*, IndexT) [with T = __tsan::SyncVar; long unsigned int kL1Size = 1048576; long unsigned int kL2Size = 1024; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache; IndexT = unsigned int]’
../rtl/tsan_sync.cpp:92:25:   required from here
../rtl/tsan_dense_alloc.h:165:46: warning: suggest parentheses around arithmetic in operand of ‘|’ []8;;https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#index-Wparentheses-Wparentheses]8;;]
cc1plus: note: unrecognized command-line option ‘-Wno-unknown-warning-option’ may have been intended to silence earlier diagnostics
gcc gotsan.cpp -c -o ./race_linux_amd64.syso -I../rtl -I../.. -I../../sanitizer_common -I../../../include -std=c++17 -Wall -fno-exceptions -fno-rtti -DSANITIZER_GO=1 -DSANITIZER_DEADLOCK_DETECTOR_VERSION=2 -fPIC -Wno-maybe-uninitialized -ffreestanding -Wno-unused-const-variable -Wno-unknown-warning-option -m64 -msse3 -DSANITIZER_DEBUG=0 -O3 -fomit-frame-pointer
In file included from ../rtl/tsan_sync.h:19,
                 from ../rtl/tsan_rtl.h:47,
                 from tsan_go.cpp:13:
../rtl/tsan_dense_alloc.h: In instantiation of ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Drain(Cache*) [with T = __tsan::MBlock; long unsigned int kL1Size = 262144; long unsigned int kL2Size = 4096; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache]’:
../rtl/tsan_dense_alloc.h:77:7:   required from ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Free(Cache*, IndexT) [with T = __tsan::MBlock; long unsigned int kL1Size = 262144; long unsigned int kL2Size = 4096; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache; IndexT = unsigned int]’
../rtl/tsan_sync.cpp:84:26:   required from here
../rtl/tsan_dense_alloc.h:165:46: warning: suggest parentheses around arithmetic in operand of ‘|’ []8;;https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#index-Wparentheses-Wparentheses]8;;]
  165 |       xchg = head_idx | (cmp & kCounterMask) + kCounterInc;
      |                         ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~
../rtl/tsan_dense_alloc.h: In instantiation of ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Drain(Cache*) [with T = __tsan::SyncVar; long unsigned int kL1Size = 1048576; long unsigned int kL2Size = 1024; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache]’:
../rtl/tsan_dense_alloc.h:77:7:   required from ‘void __tsan::DenseSlabAlloc< <template-parameter-1-1>, <anonymous>, <anonymous>, <anonymous> >::Free(Cache*, IndexT) [with T = __tsan::SyncVar; long unsigned int kL1Size = 1048576; long unsigned int kL2Size = 1024; long long unsigned int kReserved = 3221225472; Cache = __tsan::DenseSlabAllocCache; IndexT = unsigned int]’
../rtl/tsan_sync.cpp:92:25:   required from here
../rtl/tsan_dense_alloc.h:165:46: warning: suggest parentheses around arithmetic in operand of ‘|’ []8;;https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#index-Wparentheses-Wparentheses]8;;]
cc1plus: note: unrecognized command-line option ‘-Wno-unknown-warning-option’ may have been intended to silence earlier diagnostics
==================
WARNING: DATA RACE
Read at 0x00c011110000 by goroutine 2:
  <null>()
      <null>:0 +0x0

Previous write at 0x00c011110000 by main goroutine:
  <null>()
      <null>:0 +0x0
  <null>()
      <null>:0 +0x0

Goroutine 2 (running) created at:
  <null>()
      <null>:0 +0x0
==================
Found 1 data race(s)
jaten@rog /usr/local/dev-go/tmp/llvm-project-46e3ec0244c4d75a57cd635a28a9d5cbaee67c3d/compiler-rt/lib/tsan/go $

What did you see happen?

see above.

What did you expect to see?

Ability to re-build the race detector with the latest from the llvm tree, so I could evaluate whether that helps with the #74019 bugs.

Originally queried in https://groups.google.com/g/golang-nuts/c/Tuo829iE0q0

Comment From: gabyhelp

Related Issues

Related Code Changes

(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)

Comment From: glycerine

Ah. I see now what happened. The build script runs a test after the build is finished, to check that a race is detected. I do see linux .syso files being created, so what I took to be a build failure was actually a successful test of the newly built race detector library.

I'm almost ready to close this out as it actually does seem to be building. I'll try the new .syso next.

One last question though:

why does the race detector in use by the latest Go release lag 20 months behind that in llvm?

@dvyukov is there some concern about compatibility or something where we don't expect that latest tsan in llvm tree to be compatible with the latest Go?

Thanks!

Comment From: randall77

It is just that it is work to build a new race detector for all the platforms, so we generally don't do it unless there is a race detector feature/bugfix that we want.

Comment From: glycerine

@randall77 Thanks Keith. So the expectation is that the latest tsan in llvm should work, right?

I'm seeing some different errors with the latest tsan under the #74019 reproducers, which may be useful, especially as some of them give new/(maybe) more useful error messages like

ThreadSanitizer: CHECK failed: tsan_rtl_thread.cpp:261 "((parts->Queued(trace.local_head))) != (0)" (0x0, 0x0) (tid=342934)

Comment From: randall77

@randall77 Thanks Keith. So the expectation is that the latest tsan in llvm should work, right?

That is the expectation. In the past I've always had a bit of friction pulling in a new tsan version. Something doesn't compile on our builders, some assumption about our runtime no longer holds, etc. But generally it should work.

Comment From: glycerine

Thanks Keith. I skimmed through the 200 git log changes since 2023-Oct-20 to the files the race detector uses.

I'm no expert but nothing really stood out as a critical fix.

There are some things for Go related to https://github.com/golang/go/issues/64345; riscv support related, but it doesn't look like that issue is ready to ship anyway--once it does they will need 76910f914cdd4b86b28e0d5852155244ee47dc53 from Feb 28 18:27:48 2025.

So I don't see any (glaring) cause to upgrade either at this point. I just prefer to not waste time just in case there is something subtle. I'll close this issue since its really just due to my not groking the race detector build script. Thanks for your advice.