atproto relay implementation in zig
zlay.waow.tech
add fiber GPF issue writeup
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+65
+65
scripts/fiber_gpf_issue.md
+65
scripts/fiber_gpf_issue.md
···
1
+
# Io.Evented + ReleaseSafe: GPF in fiber.zig contextSwitch on x86_64-linux
2
+
3
+
`Io.Evented` crashes with a general protection fault under ReleaseSafe on x86_64-linux. Debug, ReleaseFast, and ReleaseSmall all pass. Tested on unpatched zig — no modifications to the stdlib.
4
+
5
+
## Reproduction
6
+
7
+
```zig
8
+
const std = @import("std");
9
+
const Io = std.Io;
10
+
11
+
var evented: Io.Evented = undefined;
12
+
13
+
var debug_threaded_io: Io.Threaded = undefined;
14
+
pub const std_options_debug_threaded_io: ?*Io.Threaded = &debug_threaded_io;
15
+
16
+
fn fiberReturn(_: Io) void {}
17
+
18
+
pub fn main() !void {
19
+
const allocator = std.heap.c_allocator;
20
+
debug_threaded_io = Io.Threaded.init(allocator, .{});
21
+
try Io.Evented.init(&evented, allocator, .{});
22
+
const io = evented.io();
23
+
24
+
var f = try io.concurrent(fiberReturn, .{io});
25
+
io.sleep(Io.Duration.fromMilliseconds(10), .awake) catch {};
26
+
f.cancel(io);
27
+
}
28
+
```
29
+
30
+
```
31
+
$ zig build-exe -OReleaseSafe repro.zig -lc && ./repro
32
+
General protection exception (no address available)
33
+
lib/std/Io/fiber.zig:30:20: 0x1079589 in contextSwitch
34
+
lib/std/Io/Uring.zig:1142:12: 0x109cc86 in mainIdle
35
+
```
36
+
37
+
| Mode | Result |
38
+
|------|--------|
39
+
| Debug | pass |
40
+
| ReleaseFast | pass |
41
+
| ReleaseSmall | pass |
42
+
| ReleaseSafe | GPF |
43
+
44
+
## Environment
45
+
46
+
- zig 0.16.0-dev.3059+42e33db9d, unmodified
47
+
- x86_64-linux, kernel 6.8.0-101-generic (Debian bookworm, glibc)
48
+
49
+
## What we've observed
50
+
51
+
The crash is at the inline asm in `fiber.zig` contextSwitch (x86_64 path, line 244). This gets inlined into `Uring.idle`.
52
+
53
+
Comparing the disassembly of `Uring.idle` between ReleaseFast and ReleaseSafe, the difference we see is in how `%rsi` is set up before the inline asm block. The asm uses `"{rsi}" (s)` as an input constraint:
54
+
55
+
**ReleaseFast** — there's a `lea -0x80(%rbp),%rsi` that loads the SwitchMessage address into `%rsi` before the asm block.
56
+
57
+
**ReleaseSafe** — the corresponding `lea` appears to be absent. `%rsi` seems to hold a stale value from a prior function call (`Thread.current`, which is caller-saved). The ReleaseSafe prologue includes `__zig_probe_stack` and a stack canary load (`fs:0x28`) that aren't present in ReleaseFast.
58
+
59
+
We're not certain this is the full picture — we may be misreading the disassembly or missing something about how the register constraint interacts with the surrounding code. But the crash is consistent and the repro is deterministic.
60
+
61
+
## Context
62
+
63
+
We're building an AT Protocol relay ([zlay](https://tangled.sh/zzstoatzz.io/zlay)) that runs ~2,800 concurrent fibers on `Io.Evented`. We've been working around this by building with ReleaseFast, but recently a separate bug (a websocket off-by-one) manifested as a silent SIGSEGV under ReleaseFast for weeks. ReleaseSafe would have caught it immediately as a bounds-check panic with a stack trace.
64
+
65
+
We understand Evented is experimental. Happy to provide more information or test patches.