Crash in [@ jxl::render::stages::upsample::upsample_2x_simd]
Categories
(Core :: Graphics: ImageLib, defect)
Tracking
()
| Tracking | Status | |
|---|---|---|
| firefox-esr140 | --- | unaffected |
| firefox-esr153 | --- | unaffected |
| firefox155 | --- | unaffected |
| firefox156 | --- | unaffected |
| firefox157 | --- | wontfix |
| firefox158 | --- | fixed |
People
(Reporter: clouseau-bot, Assigned: tnikkel)
References
(Blocks 1 open bug, Regression)
Details
(Keywords: crash, regression)
Crash Data
Crash report: https://crash-stats.mozilla.org/report/index/e48dc322-e223-448f-b980-182bc0260909
MOZ_CRASH Reason:
assertion failed: mem.len() >= Self::LEN
Top 10 frames:
0 xul.dll RustMozCrash(char const*, int, char const*) mozglue/static/rust/wrappers.cpp:17
1 xul.dll mozglue_static::panic_hook(std::panic::PanicHookInfo*) mozglue/static/rust/lib.rs:99
2 xul.dll core::ops::function::FnOnce::call_once<void (*)(ref$<std::panic::PanicHookInfo>)...
3 xul.dll std::panicking::panic_with_hook() library/std/src/panicking.rs:833
4 xul.dll std::panicking::panic_handler::closure$0() library/std/src/panicking.rs:691
5 xul.dll std::sys::backtrace::__rust_end_short_backtrace<std::panicking::panic_handler::c... library/std/src/sys/backtrace.rs:182
6 xul.dll std::panicking::panic_handler() library/std/src/panicking.rs:689
7 xul.dll core::panicking::panic_fmt() library/core/src/panicking.rs:80
8 xul.dll core::panicking::panic() library/core/src/panicking.rs:150
9 xul.dll jxl::render::stages::upsample::upsample_2x_simd_dispatch::avx512(jxl_simd::x86_6... third_party/rust/jxl_simd/src/x86_64/mod.rs:92
There is 1 crash in nightly 157 with buildid 20260908042139.
Clouseau analysis (automated, 72% worth investigating — a calibrated estimate that this is worth someone's time, not that the changeset below caused it). The mechanism below fits the evidence but is not proven end-to-end:
Bug 2065753 replaces jxl-rs's strictly-sequential SequentialRunner with a genuinely concurrent PoolRunner backed by the new "JxlDecode" nsThreadPool. This is the first time jxl-rs's low-memory render pipeline group scheduler (InputBuffers::mark_ready's atomic neighbor-completion tracking feeding foreach_ready_rect) can complete groups out of the deterministic order it always had before. That vendored scheduler code, last touched by an unrelated jxl-rs 0.6.0 vendor bump three weeks earlier, appears to assume a chunk width that is always a multiple of the AVX-512 lane width (Self::LEN=16), an assumption sequential execution never violated but concurrent completion-order variability plausibly can — producing the assert!(mem.len() >= Self::LEN) panic in F32VecAvx512::load.
The crashing thread (JxlDecode #2) belongs to a nsThreadPool family that literally did not exist before 6711a24d7e5b created it, and frame #19 (jxl_decoder::runner::Shared::drain, image/rust/jxl/src/runner.rs:165) is glue code that same changeset added — the crash could not have occurred on this thread before this patch landed.
Suspected regressor: 6711a24d7e5b (gh) (bug 2065753) by :tnikkel.
Code references:
- impl JxlParallelRunner for PoolRunner
- assert!(mem.len() >= Self::LEN)
- image/decoders/JxlDecodePool.cpp:64
What the automated skeptic pass checked (its own words — a pass means the check succeeded, which is not always support for the conclusion):
- unverifiable mechanism — The claim that concurrent completion order can actually violate
mark_ready's invariant and shrink a rect below one SIMD lane is architecturally plausible and grounded in real code, but not proven by static analysis alone — the decisive dispatch edge is a Rust dyn-trait call searchfox cannot resolve. - pass candidate —
6711a24d7e5bconfirmed landed 2026-09-05, bug 2065753, 3 days before the crash build; it is what addsJxlDecodePool.cppand thePoolRunner. - pass consistency (thread name) — "JxlDecode" is hardcoded only in this new file; the crashing thread family did not exist before this changeset.
- pass hardware-vs-logic —
int 0x3+panic=abort+ explicitMOZ_CRASH_REASONis a deliberate assertion trap, not a memory fault; this signature's own hardware/bit-flip annotation rate is 0% (small sample) so nothing points at a hardware explanation.
:tnikkel, can you have a look please?
Filed automatically by Clouseau, which analyses nightly crashes with an LLM. Nothing above was written or checked by a human. Please close it as INVALID if it is wrong — that is useful feedback, not a nuisance.
This is not an independent Mozilla discovery: the crash is known here only because somebody submitted the crash report linked at the top. If this duplicates an existing report, please resolve THIS bug as the duplicate and leave the credit with the earlier reporter.
| Assignee | ||
Comment 1•2 months ago
|
||
This code was disabled when landed, only the old serial runner was used, until bug 2069326 flipped the pref on. Otherwise this report all sounds reasonable.
Coincidentally this change https://github.com/libjxl/jxl-rs/commit/5dfeb9e5fddb1ffebb709745f9c4739ec0cc2e82 might have fixed it. Attempting to verify that.
| Assignee | ||
Updated•2 months ago
|
| Assignee | ||
Updated•2 months ago
|
| Assignee | ||
Comment 2•2 months ago
|
||
There's no way for me to confirm, but I think https://github.com/libjxl/jxl-rs/commit/5dfeb9e5fddb1ffebb709745f9c4739ec0cc2e82 would have fixed this crash.
I created a testcase that I think would create a crash like this one and submitted it as a PR to jxl-rs https://github.com/libjxl/jxl-rs/pull/964
I also found a similar issue that I could reproduce and created a PR for jxl-rs with regression test https://github.com/libjxl/jxl-rs/pull/965
| Assignee | ||
Updated•2 months ago
|
Comment 3•2 months ago
|
||
Set release status flags based on info from the regressing bug 2069326
Comment 4•2 months ago
|
||
Set release status flags based on info from the regressing bug 2069326
| Assignee | ||
Comment 5•2 months ago
|
||
Another crash report came in. This one had a url. I was able to fetch the jxl images and to try to recreate what an avx512 machine would see (without avx512 hardware) and it was as expected, so comment 1 seems correct, and the fix for this has landed upstream.
| Assignee | ||
Updated•2 months ago
|
Updated•1 month ago
|
Description
•