You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Bun 1.3.14 segfault (SIGILL) on Linux x86_64 #33890
changed the title [-]Bun 1.3.14 segfault (SIGILL) on Linux x86_64 — AMD EPYC Zen4 (AVX-512)[/-][+]Bun 1.3.14 segfault (SIGILL) on Linux x86_64[/+]on Jun 25, 2026
This is a deliberate panic/abort trap (ud2) on the Bun/JSC GC thread (HeapHelper), reached just after a memset/allocation helper — not a CPU-incompatible SIMD instruction. The binary embeds Rust panic machinery (std::panicking::begin_panic_handler, std::panicking::rust_panic_with_hook); ud2 is the standard Rust/Zig panic→abort trap.
Triage note for the broader "invalid opcode" bucket (incl. #33595): a Bun/Zig ud2 panic produces the identical kernel trap invalid opcode / fatal signal 4 message as a genuine missing-AVX illegal instruction. The two are only distinguishable by disassembling the faulting IP. This instance is unambiguously a GC/runtime panic (this issue / #22084 bucket), not an AVX-baseline mismatch — the CPU here has no AVX-512 yet the binary is the baseline build, and the trapping byte is 0f 0b ud2, not an EVEX/zmm op.
Reproduced on the latest 1.17.11, so the current release does not fix it. Crash recovered cleanly via process auto-restart; application state intact. Aggravating context: high concurrency (3+ primary sessions in one project + several vite dev servers) → elevated GC pressure on the HeapHelper thread.
Activity
github-actions commentedon Jun 25, 2026
This issue might be a duplicate of existing issues. Please check:
[-]Bun 1.3.14 segfault (SIGILL) on Linux x86_64 — AMD EPYC Zen4 (AVX-512)[/-][+]Bun 1.3.14 segfault (SIGILL) on Linux x86_64[/+]jjiangweilan commentedon Jun 25, 2026
same thing can happen on Windows 11
dmancilla85 commentedon Jun 25, 2026
Same on Windows 11 24GB RAM i7-11
Downgrading to v1.17.9 solve the issue in my case
simonklee commentedon Jun 25, 2026
These windows issues are a different thing AFAIK. For that please try installing the last version that we just released v1.17.11
cloph-dsp commentedon Jun 25, 2026
@simonklee can i safely update? will it work now?
JRedeker commentedon Jun 27, 2026
Additional data point: faulting IP disassembles to a deliberate
ud2panic, not an AVX/illegal-SIMD instructionBun v1.3.14 (Linux x64 baseline)6.6.114.1-microsoft-standard-WSL2, Intel i7-12700K (avx2, no avx512), 41 GiB RAM (not OOM; 19 GiB free, load 1.68 at crash)The binary is
ET_EXEC, so the runtime IP maps 1:1 to the file.objdump -dat0x2607064:This is a deliberate panic/abort trap (
ud2) on the Bun/JSC GC thread (HeapHelper), reached just after a memset/allocation helper — not a CPU-incompatible SIMD instruction. The binary embeds Rust panic machinery (std::panicking::begin_panic_handler,std::panicking::rust_panic_with_hook);ud2is the standard Rust/Zig panic→abort trap.Triage note for the broader "invalid opcode" bucket (incl. #33595): a Bun/Zig
ud2panic produces the identical kerneltrap invalid opcode/fatal signal 4message as a genuine missing-AVX illegal instruction. The two are only distinguishable by disassembling the faulting IP. This instance is unambiguously a GC/runtime panic (this issue / #22084 bucket), not an AVX-baseline mismatch — the CPU here has no AVX-512 yet the binary is thebaselinebuild, and the trapping byte is0f 0b ud2, not an EVEX/zmm op.Reproduced on the latest 1.17.11, so the current release does not fix it. Crash recovered cleanly via process auto-restart; application state intact. Aggravating context: high concurrency (3+ primary sessions in one project + several vite dev servers) → elevated GC pressure on the
HeapHelperthread.github-actions commentedon Aug 27, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.