Repository navigation
perf(runtime): per-thread route-pin cache for foreign frees (B4 R1) - #79
Merged
Merged
Conversation
PHPCraftdream
added
the
gate-run
A foreign (cross-thread) small free resolved its segment route through the global RouteDirectory shard lock on every call. R1 keeps a small per-thread cache of RoutePin handles keyed by segment base and incarnation, so repeated foreign frees into the same segment publish through the already-pinned sidecar without the directory lookup. - route_pin_cache.rs: 4-slot TLS cache; publish_cached is the unsafe transfer-once boundary; unlinked pins are swept on miss/insert and released at thread exit; loom::thread_local! seam under cfg(loom, test). - directory.rs: Entry gains `linked`, stored false (Release) on unlink under the shard lock, so a cached pin never publishes after unlink. - routing.rs: cached fast path, uncached slow path, shared tail counters; publish_foreign stays #[cold] so the own-thread free is unchanged. - Tests: b4_r1_pin_cache (hit/insert/release, re-registration, retention, thread exit, totals64), b4_r1_pin_cache_miri (SB/TB, CI only), loom_b4_r1_pin_cache (O2 no publish after unlink, O4 release on exit; linked-gate and drop mutants go red on the assert oracles). iai A/B run 2 (39222595039, main 3f0eab5 vs 811bfba): r19_prf01_joined_producer_same_word -969909 / 1445928 = -67.08 %; own free cells +1 Ir per cell. callgrind T=1: larson 319 = 319, mstress 494 = 494 Ir/op. See docs/perf/B4_R1_PIN_CACHE.md §7. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e merge) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PHPCraftdream
force-pushed
the
b4-r1-pin-cache
branch
from
October 19, 2026 11:00
811bfba to
35a1d7d
Compare
…0 PASS, run 39241457236) Gate configs removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PHPCraftdream
removed
the
gate-run
# Conflicts: # docs/evidence/REGISTRY.md
…t Box The owner-side drain loop dropped a fresh Box<Block> per cycle. On the non-fastbin miri rows that own free reaches dealloc_small, which writes the freelist link while the by-value Box::drop frame is live: the accepted P1-box defect (card 164), not the pin-cache behaviour this test pins. CI run 39264902142 went red there (miri-plain and miri (core), UB in Node::write_next from dealloc_small at the loop's drop). The loop now cycles through GlobalAlloc::alloc/dealloc directly, which keeps the same own alloc/free drain trigger without a protected Box argument. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On the non-fastbin miri rows the scenario itself passed, then libtest's teardown dropped a by-value Box on the main thread through the installed allocator, which hits the accepted P1-box defect (card 164: dealloc_small writes the link while the Box frame is live). The cell now follows the miri_global_box_acceptance pattern: harness = false, test = false, one scenario in main, and a COMPLETE marker printed after the last assertion; CI and the miri matrix grep that marker. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # .github/workflows/ci.yml # Cargo.toml # README.md # docs/ARCHITECTURE.md # docs/evidence/REGISTRY.md
# Conflicts: # README.md # docs/ARCHITECTURE.md # docs/evidence/REGISTRY.md
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Per-thread cache of route pins for cross-thread (foreign) small frees, so repeated foreign frees into the same segment skip the global
RouteDirectoryshard-lock lookup.What changes
src/registry/heap_core_xthread/route_pin_cache.rs(new): 4-slot TLS cache keyed by segment base + incarnation.publish_cachedis the unsafe transfer-once boundary. Unlinked pins are swept on miss/insert and released at thread exit. Undercfg(all(loom, test))the cache usesloom::thread_local!.src/registry/segment_route/directory.rs:Entry::linked, storedfalsewith Release under the shard lock on unlink. A cached pin never publishes after unlink.src/registry/heap_core_xthread/routing.rs: cached fast path, uncached slow path, shared tail for theFOREIGN_OR_UNROUTABLE_FREES/B4_FOREIGN_PUBLISH_OKcounters and the Miri publication gate.Evidence
3f0eab59: B4/route tests pass on 3 feature sets × debug/release; clippy-D warningsis clean on all 8 CI rows; structural tests pass.tests/loom_b4_r1_pin_cache.rs: baseline 4/4. The linked-gate mutant goes red on O2 (1 ≠ 0). The drop-at-exit mutant goes red on O4 (3 ≠ 2).tests/b4_r1_pin_cache_miri.rs) runs on this PR's CI only. It cannot run on Windows (GetModuleHandleW)..github/gate-run.json, removed before merge.Costs
B4_FOREIGN_PUBLISH_OK(bench-internals diagnostic) now also counts a synchronous fallback-heap free.Report:
docs/perf/B4_R1_PIN_CACHE.md.🤖 Generated with Claude Code