Suppress the documented Crossbeam queue-buffer race in ThreadSanitizer.

.github/tsan-suppressions.txt

@@ -0,0 +1,8 @@
+# Crossbeam speculatively reads queue slots before atomically validating the
+# steal. Reads invalidated by concurrent writes are discarded without using or
+# dropping the task. The non-atomic volatile accesses still violate Rust's
+# memory model; this suppression accepts that known dependency limitation.
+# https://github.com/crossbeam-rs/crossbeam/issues/589#issuecomment-720972996
+# Match only the queue buffer accesses, rather than all Crossbeam operations.
+race:crossbeam_deque::deque::Buffer*::read
+race:crossbeam_deque::deque::Buffer*::write

.github/workflows/test.yml

@@ -123,3 +123,4 @@ jobs:
         run: cargo test -Zbuild-std --target x86_64-unknown-linux-gnu --workspace --all-targets --locked -- --skip queue_publication_cannot_leave_a_worker_asleep
         env:
           RUSTFLAGS: -Zsanitizer=${{ matrix.sanitizer }}
+          TSAN_OPTIONS: suppressions=${{ github.workspace }}/.github/tsan-suppressions.txt

context/implementation.md

@@ -104,3 +104,12 @@ and Tokio-only feature combinations. The same TCP/file consumers exercise both
 implementations; Linux tests cover cancellation batches and shutdown races.
 CI covers Linux, macOS, Windows and FreeBSD, with separate Linux io_uring and
 sanitizer jobs; distinguish configured CI from executed results.
+
+ThreadSanitizer loads `.github/tsan-suppressions.txt` to suppress Crossbeam's
+internal queue `Buffer::read` and `Buffer::write` race reports. Crossbeam reads
+slots speculatively and discards values when atomic validation fails. Its
+non-atomic volatile accesses remain a known Rust memory-model limitation;
+the suppression accepts that limitation rather than fixing it. See the
+[upstream discussion](https://github.com/crossbeam-rs/crossbeam/issues/589#issuecomment-720972996).
+Keep suppression patterns scoped to those buffer accesses and reassess them
+when updating Crossbeam. Other race reports continue to fail the sanitizer job.