rayon crashes

[bjorn3]

https://github.com/xiph/rav1e

All of them are like

---- api::test::output_frameno_incremental_reorder_scene_change_at_3 stdout ----
---- api::test::output_frameno_incremental_reorder_scene_change_at_3 stderr ----
thread '<unnamed>' panicked at 'index out of bounds: the len is 140710327231184 but the index is 128', /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:541:24
stack backtrace:
   0: rust_begin_unwind
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panicking.rs:593:5
   1: core::panicking::panic_fmt
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/core/src/panicking.rs:67:14
   2: core::panicking::panic_bounds_check
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/core/src/panic.rs:61:9
   3: rayon_core::registry::Registry::in_worker_cold::{{closure}}::{{closure}}
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:541:24
   4: rayon_core::job::JobResult<T>::call::{{closure}}
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/job.rs:218:41
   5: <core::panic::unwind_safe::AssertUnwindSafe<F> as core::ops::function::FnOnce<()>>::call_once
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/core/src/panic/unwind_safe.rs:265:9
   6: std::panicking::try::do_call
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panicking.rs:500:40
   7: std::panicking::try
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panicking.rs:464:19
   8: std::panic::catch_unwind
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panic.rs:142:14
   9: rayon_core::unwind::halt_unwinding
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/unwind.rs:17:5
  10: rayon_core::job::JobResult<T>::call
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/job.rs:217:5
  11: <rayon_core::job::StackJob<L,F,R> as rayon_core::job::Job>::execute
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/job.rs:120:32
  12: rayon_core::job::JobRef::execute
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/job.rs:64:9
  13: rayon_core::registry::WorkerThread::execute
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:874:9
  14: rayon_core::registry::WorkerThread::wait_until_cold
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:820:17
  15: rayon_core::registry::WorkerThread::wait_until
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:803:13
  16: rayon_core::registry::main_loop
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:948:5
  17: rayon_core::registry::ThreadBuilder::run
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:54:18
  18: <rayon_core::registry::DefaultSpawn as rayon_core::registry::ThreadSpawn>::spawn::{{closure}}
             at /home/gh-bjorn3/.cargo/registry/src/index.crates.io-6f17d22bba15001f/rayon-core-1.11.0/src/registry.rs:99:20
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.

[bjorn3]

The same crash happens with wasmtime. Running rayons test suite also results in this error.

[bjorn3]

Looks like cg_clif accidentally used PanicBoundsCheck for AssertKind::MisalignedPointerDereference causing a wrong panic message. The correct panic message is:

thread '<unnamed>' panicked at 'misaligned pointer dereference: address must be a multiple of 0x80 but is 0x7f29244ed1c0', /home/gh-bjorn3/cg_clif/rayon/rayon-core/src/registry.rs:523:24
stack backtrace:
   0: rust_begin_unwind
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panicking.rs:593:5
   1: core::panicking::panic_nounwind_fmt
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/core/src/panicking.rs:96:14
   2: core::panicking::panic_misaligned_pointer_dereference
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/core/src/panicking.rs:175:5
   3: rayon_core::registry::Registry::in_worker_cold::{{closure}}::{{closure}}
             at ./rayon-core/src/registry.rs:523:24
   4: rayon_core::job::JobResult<T>::call::{{closure}}
             at ./rayon-core/src/job.rs:218:41
   5: <core::panic::unwind_safe::AssertUnwindSafe<F> as core::ops::function::FnOnce<()>>::call_once
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/core/src/panic/unwind_safe.rs:265:9
   6: std::panicking::try::do_call
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panicking.rs:500:40
   7: std::panicking::try
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panicking.rs:464:19
   8: std::panic::catch_unwind
             at /home/gh-bjorn3/cg_clif/build/stdlib/library/std/src/panic.rs:142:14
   9: rayon_core::unwind::halt_unwinding
             at ./rayon-core/src/unwind.rs:17:5
  10: rayon_core::job::JobResult<T>::call
             at ./rayon-core/src/job.rs:218:15
  11: <rayon_core::job::StackJob<L,F,R> as rayon_core::job::Job>::execute
             at ./rayon-core/src/job.rs:120:32
  12: rayon_core::job::JobRef::execute
             at ./rayon-core/src/job.rs:64:9
  13: rayon_core::registry::WorkerThread::execute
             at ./rayon-core/src/registry.rs:855:9
  14: rayon_core::registry::WorkerThread::wait_until_cold
             at ./rayon-core/src/registry.rs:802:21
  15: rayon_core::registry::WorkerThread::wait_until
             at ./rayon-core/src/registry.rs:777:13
  16: rayon_core::registry::main_loop
             at ./rayon-core/src/registry.rs:926:5
  17: rayon_core::registry::ThreadBuilder::run
             at ./rayon-core/src/registry.rs:54:18
  18: <rayon_core::registry::DefaultSpawn as rayon_core::registry::ThreadSpawn>::spawn::{{closure}}
             at ./rayon-core/src/registry.rs:99:20

[bjorn3]

Looks like WorkerThread is stack allocated and aligned to the cacheline size. Cranelift doesn't support aligning stack values to more than 16 bytes yet. If I heap allocate WorkerThread all tests pass except for a single test which requires panic=unwind.

Edit: Opened bytecodealliance/wasmtime#6716

[bjorn3]

Fixed in 55380a5. The entire rayon test suite now passes with the exception of a single test which tests the behavior of rayon when user code panics as cg_clif doesn't yet support unwinding on panics. In addition the entire test suite of rav1e passes on aarch64, but on x86_64 it requires _mm_avg_epu16.