Przejdź do treści
Biblioteka skilli
Programowanie · Budowanie i kod

lock-free-guide

/lock-free-guide

Pomaga pisać i sprawdzać współbieżny kod pamięci współdzielonej: atomowe operacje, struktury bez blokad, porządek pamięci i kolejki.

Repozytorium
platform ↗
Licencja
MIT
Ostatnia zmiana repozytorium
08.10.2026

Ocena i bezpieczeństwo

90/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść25 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
19
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 08.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Pomaga pisać i sprawdzać współbieżny kod pamięci współdzielonej: atomowe operacje, struktury bez blokad, porządek pamięci i kolejki.

Wyzwalacze z opisu

t say
Oryginalny opis (EN)

Use when writing or reviewing shared-memory concurrent code: atomics, lock-free/wait-free data structures, or scalable synchronization. Triggers on prompts about memory ordering (relaxed/acquire/release/seq_cst), CAS/compare-exchange, the ABA problem, SPSC/MPSC/MPMC queues, lock-free stack, hazard pointers / epoch / RCU reclamation, false sharing, spinlocks/backoff, work-stealing, even when the user doesn't say 'lock-free'.

Jak działa

  1. Requirements
  2. Essentials
  3. Memory model
  4. Building blocks
  5. Data structures
  6. Reclamation & scaling
  7. Gotchas
  8. Progressive Disclosure
Sekcje z nagłówków SKILL.md, w oryginale · 66 linii, 591 słów
Podgląd SKILL.md
---
name: lock-free-guide
description: "Use when writing or reviewing shared-memory concurrent code: atomics, lock-free/wait-free data structures, or scalable synchronization. Triggers on prompts about memory ordering (relaxed/acquire/release/seq_cst), CAS/compare-exchange, the ABA problem, SPSC/MPSC/MPMC queues, lock-free stack, hazard pointers / epoch / RCU reclamation, false sharing, spinlocks/backoff, work-stealing, even when the user doesn't say 'lock-free'."
---

# Lock-Free / Wait-Free Concurrency Guidelines

Choose concurrency primitives by their progress guarantees, memory ordering, and reclamation requirements; verify them under contention.

## Requirements

- C11 atomics (`<stdatomic.h>`); applies to systems code with atomic, threading, and job-scheduling primitives.

## Essentials

- **No data races** - Every shared mutable location touched by 2+ threads is atomic or lock-protected, see [references/memory-model.md](references/memory-model.md)
- **CAS loops retry** - `compare_exchange_weak` in a loop; reload the expected value on failure, see [references/atomics-and-cas.md](references/atomics-and-cas.md)
- **Lock-free needs reclamation** - Removing a node from a shared structure is not freeing it, see [references/safe-memory-reclamation.md](references/safe-memory-reclamation.md)
- **Measure, do not assume** - Locks are often faster; profile before going lock-free, see [references/progress-guarantees.md](references/progress-guarantees.md)

## Memory model

- **Progress guarantees** - wait-free ⊃ lock-free ⊃ obstruction-free ⊃ blocking, see [references/progress-guarantees.md](references/progress-guarantees.md)
- **Happens-before** - Acquire/release builds the ordering that makes writes visible, see [references/memory-model.md](references/memory-model.md)
- **Orderings** - relaxed/acquire/release/acq_rel/seq_cst + fences, see [references/memory-ordering.md](references/memory-ordering.md)

## Building blocks

- **Atomics & CAS** - load/store/exchange/fetch\_\*/compare_exchange, weak vs strong, see [references/atomics-and-cas.md](references/atomics-and-cas.md)
- **ABA problem** - CAS can't see a value changed away and back; tag pointers or reclaim safely, see [references/aba-problem.md](references/aba-problem.md)
- **False sharing** - Pad/align hot per-thread and atomic fields to their own cache line, see [references/false-sharing.md](references/false-sharing.md)

## Data structures

- **SPSC ring** - One producer, one consumer, no CAS, see [references/spsc-ring-buffer.md](references/spsc-ring-buffer.md)
- **MPSC queue** - Intrusive: producers XCHG tail, consumer walks next, see [references/mpsc-queue.md](references/mpsc-queue.md)
- **MPMC bounded** - Per-cell sequence numbers, ABA-free, see [references/mpmc-bounded-queue.md](references/mpmc-bounded-queue.md)
- **Lock-free stack** - CAS on head; needs reclamation for pop, see [references/lock-free-stack.md](references/lock-free-stack.md)

## Reclamation & scaling

- **Safe reclamation** - refcount / hazard pointers / epoch (QSBR) / RCU, see [references/safe-memory-reclamation.md](references/safe-memory-reclamation.md)
- **Locks & backoff** - TTAS spinlock, exponential backoff, ticket/MCS/CLH, seqlock, see [references/locks-and-backoff.md](references/locks-and-backoff.md)
- **Scale by not sharing** - Shard, partition, combine, work-steal, NUMA-place, see [references/scalable-architecture.md](references/scalable-architecture.md)

## Gotchas

- A successful CAS proves only that the bit pattern matched, not that nothing happened in between (ABA); pointers reused after free defeat it.
- `volatile` is not atomic and gives no ordering. It never makes concurrent C correct; use `_Atomic`/`atomic_*`.
- `compare_exchange_weak` may fail spuriously even when the value matches, only ever use it inside a retry loop.
- Removing a node from a lock-free list does not mean no other thread is still dereferencing it: freeing too early is a use-after-free.
- Two unrelated atomics on the same cache line still contend (false sharing): alignment, not correctness, but it can erase all your scaling.

## Progressive Disclosure

- Read [references/progress-guarantees.md](references/progress-guarantees.md) - Load when deciding between locks, lock-free, or wait-free, or explaining the trade-offs
- Read [references/memory-model.md](references/memory-model.md) - Load when reasoning about data races, visibility, or happens-before
- Read [references/memory-ordering.md](references/memory-ordering.md) - Load when choosing or reviewing a `memory_order_*` argument or a fence
- Read [references/atomics-and-cas.md](references/atomics-and-cas.md) - Load when writing atomic loads/stores/RMW or a compare-exchange loop
- Read [references/aba-problem.md](references/aba-problem.md) - Load when a CAS-based structure reuses memory or pointers
- Read [references/false-sharing.md](references/false-sharing.md) - Load when per-thread counters or atomics scale badly under contention
- Read [references/spsc-ring-buffer.md](references/spsc-ring-buffer.md) - Load when building a single-producer/single-consumer queue
- Read [references/mpsc-queue.md](references/mpsc-queue.md) - Load when many threads feed one consumer (logging, job results)
- Read [references/mpmc-bounded-queue.md](references/mpmc-bounded-queue.md) - Load when building a general bounded multi-producer/multi-consumer queue
- Read [references/lock-free-stack.md](references/lock-free-stack.md) - Load when implementing a lock-free stack or freelist
- Read [references/safe-memory-reclamation.md](references/safe-memory-reclamation.md) - Load when freeing nodes that other threads might still read
- Read [references/locks-and-backoff.md](references/locks-and-backoff.md) - Load when a lock is the right tool and you need a scalable one
- Read [references/scalable-architecture.md](references/scalable-architecture.md) - Load when synchronization itself is the bottleneck
- Read [references/testing-and-verification.md](references/testing-and-verification.md) - Load when validating concurrent code before merge
Pliki w paczce (19)
  • SKILL.md 6 KB
  • SOURCES.md 4 KB
  • eval-queries.json 5 KB
  • evals.json 3 KB
  • moon.yml 1 KB
  • references/aba-problem.md 3 KB
  • references/atomics-and-cas.md 3 KB
  • references/false-sharing.md 2 KB
  • references/lock-free-stack.md 3 KB
  • references/locks-and-backoff.md 4 KB
  • references/memory-model.md 3 KB
  • references/memory-ordering.md 2 KB
  • references/mpmc-bounded-queue.md 5 KB
  • references/mpsc-queue.md 4 KB
  • references/progress-guarantees.md 3 KB
  • references/safe-memory-reclamation.md 4 KB
  • references/scalable-architecture.md 4 KB
  • references/spsc-ring-buffer.md 3 KB
  • references/testing-and-verification.md 3 KB

Wersje

W bazie nie ma innych skilli o nazwie lock-free-guide.

Migawka bazy z 13.11.2026 · wersja paczki b3a7a09dedf3 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100memory-model
    Wyjaśnia model pamięci C++ i Rust dla programowania współbieżnego: kolejność dostępu, acquire-release, seq_cst, happens-before, wyścigi danych i struktury lock-free.
  • 100memory-model
    Pomaga dobierać uporządkowanie pamięci dla atomików w C++ i Rust, rozumować o happens-before, pisać wzorce lock-free i diagnozować wyścigi danych.
  • 100lock-free-ring-buffers
    Wyjaśnia komunikację między wątkami bez blokad przez bufory pierścieniowe i projekt uwzględniający pamięć podręczną.
  • 95memory-model
    Dobiera porządki pamięci dla atomów w C++ i Rust, rozumuje o zachodzeniu przed (happens-before) i diagnozuje wyścigi danych.
  • 93wk-spinlocks
    Chroni dane używane na poziomie DISPATCH_LEVEL lub wyższym przez spinlocki jądra, z zapisem i przywracaniem IRQL oraz diagnozą blokad.
  • 90concurrency-correctness
    Wykrywa w kodzie wielowątkowym blokady trzymane podczas operacji blokujących, ryzyko zakleszczeń i błędne kolejności pamięci.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.