cmd/compile: OOM killed on linux/arm64 [1.25 backport]

[j-modernc-org]

Go version

go version go1.25.2 linux/arm64

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=''
GOARCH='arm64'
GOARM64='v8.0'
GOAUTH='netrc'
GOBIN=''
GOCACHE='/home/jnml/.cache/go-build'
GOCACHEPROG=''
GODEBUG=''
GOENV='/home/jnml/.config/go/env'
GOEXE=''
GOEXPERIMENT=''
GOFIPS140='off'
GOFLAGS=''
GOGCCFLAGS='-fPIC -pthread -Wl,--no-gc-sections -fmessage-length=0 -ffile-prefix-map=/tmp/go-build2613515218=/tmp/go-build -gno-record-gcc-switches'
GOHOSTARCH='arm64'
GOHOSTOS='linux'
GOINSECURE=''
GOMOD='/dev/null'
GOMODCACHE='/home/jnml/pkg/mod'
GONOPROXY=''
GONOSUMDB=''
GOOS='linux'
GOPATH='/home/jnml'
GOPRIVATE=''
GOPROXY='https://proxy.golang.org,direct'
GOROOT='/usr/local/go'
GOSUMDB='sum.golang.org'
GOTELEMETRY='local'
GOTELEMETRYDIR='/home/jnml/.config/go/telemetry'
GOTMPDIR=''
GOTOOLCHAIN='auto'
GOTOOLDIR='/usr/local/go/pkg/tool/linux_arm64'
GOVCS=''
GOVERSION='go1.25.2'
GOWORK=''
PKG_CONFIG='pkg-config'

What did you do?

me@rpi5:~/tmp/bug $ ls -l
total 12
-rw-r--r-- 1 me me 9491 Oct 10 11:34 ccgo.zip
me@rpi5:~/tmp/bug $ unzip -q ccgo.zip && cd ccgo && ls -l
/home/me/tmp/bug/ccgo
total 40
-rw-r--r-- 1 me me   540 Oct 10 10:28 go.mod
-rw-r--r-- 1 me me  3707 Oct 10 10:28 go.sum
-rw-r--r-- 1 me me 31853 Oct 10 11:23 main.go
me@rpi5:~/tmp/bug/ccgo $ time go build
example.com/ccgo: /usr/local/go/pkg/tool/linux_arm64/compile: signal: killed

real	0m23.135s
user	0m33.775s
sys	0m1.539s
me@rpi5:~/tmp/bug/ccgo $ 

Found by CSmith+ccgo. Code only partially minimized, refuses to reproduce the bug when I tried more.

ccgo.zip

What did you see happen?

gc compiler OOM killed.

What did you expect to see?

go build works normally. Like it does with the same program on linux/amd64:

me@e5-1650:~/tmp/bug$ go version
go version go1.25.2 linux/amd64
me@e5-1650:~/tmp/bug$ ls -l
total 12
-rw-r--r-- 1 me me 9491 Oct 10 11:39 ccgo.zip
me@e5-1650:~/tmp/bug$ unzip -q ccgo.zip && cd ccgo && ls -l
/home/me/tmp/bug/ccgo
total 40
-rw-r--r-- 1 me me   540 Oct 10 10:28 go.mod
-rw-r--r-- 1 me me  3707 Oct 10 10:28 go.sum
-rw-r--r-- 1 me me 31853 Oct 10 11:23 main.go
me@e5-1650:~/tmp/bug/ccgo$ time go build

real	0m0.370s
user	0m0.573s
sys	0m0.082s
me@e5-1650:~/tmp/bug/ccgo$ 

[gabyhelp]

Related Issues

• cmd/compile: ccxt oom killed on compilation #73365
• cmd/compile: out of memory building github.com/cosmos72/gomacro with go1.11beta2 on 386/linux, arm/linux #26546 (closed)
• cmd/compile: SIGILL when building golang.org/x/sys/unix on openbsd/arm64 #42517 (closed)
• runtime: docker build failure with go-1.6.2/musl libc on armhf #16081 (closed)
• cmd/compile: out of memory with large slice and map composite literals #22010 (closed)
• cmd/compile: runs out of memory allocating 12k + items in a map at init, works as literal map #13554 (closed)
• cmd/compile/internal/ssa: out of memory error when boostrapping 1.12 #30981 (closed)
• cmd/compile: 1.12 compiler needs double the memory than 1.11 #32089 (closed)
• cmd/compile: Go 1.9 oom killed on compilation #21713 (closed)
• cgo-gcc-export-header-prolog:31:14: error: size of array ‘_check_for_64_bit_pointer_matching_GoInt’ is negative #64190 (closed)

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

[randall77]

This looks like it was fixed by https://go-review.googlesource.com/c/go/+/696035

I guess we should backport that.

[gopherbot]

Change https://go.dev/cl/710875 mentions this issue: [release-branch.go1.25] cmd/compile: during regalloc, fixedreg values are always available

[prattmic]

Is this a regression in 1.25?

[randall77]

Yes, this problem does not exist in 1.24.

[dr2chase]

Decided not to backport because it is a fuzzer bug, not "real" code, and also fixed on tip.