Skip to content

Tracing JIT: intermittent segfault in JIT buffer in long-running CLI command (PHP 8.5.12, opcache.jit=tracing) #25551

Open

Description

@phuongvmhut

Description

A Laravel scheduled CLI command (php artisan schedule:run, run every minute by cron) intermittently segfaults inside JIT-generated code.

It happens in bursts: once it starts, nearly every hourly run of the same scheduled workload crashes for 1–2 days (36 crashes on one day, 21 the next), then it stops on its own after unrelated code/data changes and does not recur for days.

Kernel log at the time of a crash:

php[3844339]: segfault at 0 ip 00005648b9e003eb sp 00007ffcd6eb38b0 error 4 in zero (deleted)[5648b9e00000+8000000]

The faulting instruction is inside an anonymous 0x8000000 (128M) mapping, which matches opcache.jit_buffer_size=128M, i.e. JIT-generated code.

No core dump or backtrace is available yet (the crashes stopped before one could be captured). I will add a backtrace and the failing command when the next burst happens.

Opcache settings (CLI):

opcache.enable=1
opcache.enable_cli=1
opcache.jit=tracing
opcache.jit_buffer_size=128M

Resulted in this output:

Segmentation fault (exit code 139)

But I expected this output instead:

The command completes normally.

PHP Version

PHP 8.5.12 (cli) (NTS), LiteSpeed lsphp85 build

Operating System

AlmaLinux 9, x86_64 (OpenLiteSpeed)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions