Fix lock-free queue bugs, add GitHub Actions CI, and add ABA stress test

.github/workflows/ci.yml

@@ -0,0 +1,21 @@
+name: C/C++ CI
+
+on:
+  push:
+    branches: [ "HP" ]
+  pull_request:
+    branches: [ "HP" ]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+
+    steps:
+    - uses: actions/checkout@v4
+    - name: make
+      run: make
+    - name: make test
+      run: make test

.gitignore

@@ -8,3 +8,13 @@
 /Visual Stdio/.vs/lfqueue/v14
 /Visual Stdio/Release
 /Visual Stdio/Debug
+
+# Build artifacts
+bin/
+*.o
+*.a
+*.so
+*.so.*
+
+# CodeQL artifacts
+_codeql_detected_source_root

Makefile

@@ -3,7 +3,7 @@ CFLAGS=-std=gnu99 -O3 -Wall -Wextra -g
 LDFLAGS=-g
 LOADLIBS=-lpthread
 
-all : bin/test_p1c1 bin/test_p4c4 bin/test_p100c10 bin/test_p10c100 bin/example
+all : bin/test_p1c1 bin/test_p4c4 bin/test_p100c10 bin/test_p10c100 bin/test_aba bin/example
 
 bin/example: example.c liblfq.so.1.0.0
 	gcc $(CFLAGS) $(LDFLAGS) example.c lfq.c -o bin/example -lpthread
@@ -20,17 +20,21 @@ bin/test_p100c10: liblfq.so.1.0.0 test_multithread.c
 bin/test_p10c100: liblfq.so.1.0.0 test_multithread.c
 	gcc $(CFLAGS) $(LDFLAGS) test_multithread.c -o bin/test_p10c100 -L. -Wl,-Bstatic -llfq -Wl,-Bdynamic -lpthread -D MAX_PRODUCER=10 -D MAX_CONSUMER=100
 
+bin/test_aba: liblfq.so.1.0.0 test_aba.c
+	gcc $(CFLAGS) $(LDFLAGS) test_aba.c -o bin/test_aba -L. -Wl,-Bstatic -llfq -Wl,-Bdynamic -lpthread
+
 liblfq.so.1.0.0: lfq.c lfq.h cross-platform.h
 	gcc $(CFLAGS) $(CPPFLAGS) -c lfq.c   # -fno-pie for static linking?
 	ar rcs liblfq.a lfq.o
 	gcc $(CFLAGS) $(CPPFLAGS) -fPIC -c lfq.c
 	gcc $(LDFLAGS) -shared -o liblfq.so.1.0.0 lfq.o
 
-test: bin/test_p1c1 bin/test_p4c4 bin/test_p100c10 bin/test_p10c100
+test: bin/test_p1c1 bin/test_p4c4 bin/test_p100c10 bin/test_p10c100 bin/test_aba
 	$(TESTWRAPPER) bin/test_p1c1
 	$(TESTWRAPPER) bin/test_p4c4
 	$(TESTWRAPPER) bin/test_p100c10
 	$(TESTWRAPPER) bin/test_p10c100
+	$(TESTWRAPPER) bin/test_aba
 
 clean:
 	rm -rf *.o bin/* liblfq.so.1.0.0 liblfq.a

cross-platform.h

@@ -52,7 +52,10 @@
 	#define lmb() asm volatile("":::"memory")   // compiler barrier only.  runtime reordering already impossible on x86
 	#define smb() asm volatile("":::"memory")
        // "mfence" for lmb and smb makes assertion failures rarer, but doesn't eliminate, so it's just papering over the symptoms
-#endif // else no definition
+#else
+	#define lmb() mb()
+	#define smb() mb()
+#endif
 
 	// thread
 	#include <pthread.h>

lfq.c

@@ -91,14 +91,16 @@ int lfq_init(struct lfq_ctx *ctx, int max_consume_thread) {
 		return -errno;
 
 	struct lfq_node * free_pool_node = calloc(1,sizeof(struct lfq_node));
-	if (!free_pool_node) 
+	if (!free_pool_node) {
+		free(tmpnode);
 		return -errno;
+	}
 
 	tmpnode->can_free = free_pool_node->can_free = true;
 	memset(ctx, 0, sizeof(struct lfq_ctx));
 	ctx->MAXHPSIZE = max_consume_thread;
-	ctx->HP = calloc(max_consume_thread,sizeof(struct lfq_node));
-	ctx->tid_map = calloc(max_consume_thread,sizeof(struct lfq_node));
+	ctx->HP = calloc(max_consume_thread,sizeof(struct lfq_node *));
+	ctx->tid_map = calloc(max_consume_thread,sizeof(int));
 	ctx->head = ctx->tail=tmpnode;
 	ctx->fph = ctx->fpt=free_pool_node;
 

test_aba.c

@@ -0,0 +1,166 @@
+/*
+ * ABA stress test for lfqueue
+ *
+ * The ABA problem: a CAS on a pointer can succeed spuriously when:
+ *   1. Thread 1 reads head = node_A
+ *   2. Thread 2 dequeues A, frees it; malloc() returns the same address for
+ *      a new node B; B is enqueued, making head cycle back to address A
+ *   3. Thread 1's CAS(&head, A, A->next) succeeds even though the node at
+ *      address A is now logically B — A->next is stale/garbage.
+ *
+ * To maximize ABA probability this test uses:
+ *   - All threads as both producers AND consumers (tight enq+deq loops)
+ *   - No thread_yield between enqueue and dequeue (maximises racing)
+ *   - Very small queue depth (each thread keeps at most 1 item in-flight)
+ *   - Many threads competing on the same head pointer
+ *   - Magic values in each node to detect data corruption caused by ABA
+ */
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+#include <inttypes.h>
+#include "lfq.h"
+#include "cross-platform.h"
+
+#if defined __GNUC__ || defined __CYGWIN__ || defined __MINGW32__
+#include <unistd.h>
+#include <pthread.h>
+#else
+#include <windows.h>
+#endif
+
+#ifndef ABA_THREADS
+#define ABA_THREADS 16
+#endif
+
+#ifndef ABA_ITERATIONS
+#define ABA_ITERATIONS 500000
+#endif
+
+/* Two distinct magic values — LIVE (in queue) and DEAD (already freed).
+ * A read of DEAD_MAGIC on a dequeued item means use-after-free / ABA
+ * corruption. */
+#define LIVE_MAGIC 0xAB1AB1EFUL
+#define DEAD_MAGIC 0xDEADBEEFUL
+
+struct aba_item {
+	volatile uint32_t magic;
+};
+
+static volatile uint64_t total_enqueued = 0;
+static volatile uint64_t total_dequeued = 0;
+static volatile int      errors = 0;
+static volatile int      cn_t   = 0;   /* tid allocator */
+
+static struct lfq_ctx ctx;
+
+/*
+ * Each thread alternates between enqueue and spin-dequeue in a tight loop.
+ * This keeps the queue depth near zero, so the same pointer addresses cycle
+ * through head rapidly — exactly the condition that triggers ABA.
+ */
+THREAD_FN aba_thread(void *arg) {
+	(void)arg;
+	/* Allocate a dedicated hazard-pointer slot for this thread */
+	int tid = ATOMIC_ADD(&cn_t, 1) - 1;
+
+	uint64_t local_enq = 0, local_deq = 0;
+
+	for (int i = 0; i < ABA_ITERATIONS && !errors; i++) {
+		/* --- Enqueue a fresh item --- */
+		struct aba_item *item = malloc(sizeof(struct aba_item));
+		if (!item) {
+			ATOMIC_ADD(&errors, 1);
+			break;
+		}
+		item->magic = LIVE_MAGIC;
+
+		if (lfq_enqueue(&ctx, item) != 0) {
+			free(item);
+			ATOMIC_ADD(&errors, 1);
+			break;
+		}
+		local_enq++;
+
+		/*
+		 * --- Spin-dequeue (no yield) ---
+		 * By immediately spinning for a result with no sleep we keep
+		 * maximum pressure on the head CAS, creating the window for ABA:
+		 *   our enqueued node may be stolen and its memory recycled before
+		 *   we manage to dequeue it ourselves.
+		 */
+		struct aba_item *got;
+		do {
+			got = lfq_dequeue_tid(&ctx, tid);
+		} while (got == NULL);
+
+		/* Validate: corruption means ABA or use-after-free occurred */
+		if (got->magic != LIVE_MAGIC) {
+			printf("ABA corruption detected! "
+			       "magic=0x%08X (expected 0x%08X) iter=%d tid=%d\n",
+			       got->magic, (uint32_t)LIVE_MAGIC, i, tid);
+			ATOMIC_ADD(&errors, 1);
+		}
+		/* Poison before free to detect future use-after-free reads */
+		got->magic = DEAD_MAGIC;
+		free(got);
+		local_deq++;
+	}
+
+	ATOMIC_ADD64(&total_enqueued, local_enq);
+	ATOMIC_ADD64(&total_dequeued, local_deq);
+	return 0;
+}
+
+int main(void) {
+	printf("ABA stress test: %d threads x %d iterations each\n",
+	       ABA_THREADS, ABA_ITERATIONS);
+	printf("(tight enq+deq loops, no yield — maximises head-pointer reuse)\n");
+
+	if (lfq_init(&ctx, ABA_THREADS) != 0) {
+		fprintf(stderr, "lfq_init failed\n");
+		return 1;
+	}
+
+	THREAD_TOKEN threads[ABA_THREADS];
+	for (int i = 0; i < ABA_THREADS; i++) {
+#if defined __GNUC__ || defined __CYGWIN__ || defined __MINGW32__
+		pthread_create(&threads[i], NULL, aba_thread, NULL);
+#else
+#pragma warning(disable:4133)
+		threads[i] = CreateThread(NULL, 0, aba_thread, NULL, 0, 0);
+#endif
+	}
+
+	for (int i = 0; i < ABA_THREADS; i++)
+		THREAD_WAIT(threads[i]);
+
+	/* Drain any items left in the queue after all threads exit */
+	struct aba_item *item;
+	int drain_tid = 0; /* safe: all thread tids have been released */
+	while ((item = lfq_dequeue_tid(&ctx, drain_tid)) != NULL) {
+		if (item->magic != LIVE_MAGIC) {
+			printf("ABA corruption in drain! magic=0x%08X\n", item->magic);
+			errors++;
+		}
+		item->magic = DEAD_MAGIC;
+		free(item);
+		total_dequeued++;
+	}
+
+	long freecount = lfg_count_freelist(&ctx);
+	int  clean     = lfq_clean(&ctx);
+
+	printf("Enqueued=%" PRId64 " Dequeued=%" PRId64
+	       " freelist=%ld clean=%d\n",
+	       total_enqueued, total_dequeued, freecount, clean);
+
+	if (errors)
+		printf("ABA Test FAILED!! (%d corruption(s) detected)\n", errors);
+	else
+		printf("ABA Test PASS!!\n");
+
+	return errors != 0;
+}

test_multithread.c

@@ -59,7 +59,7 @@ THREAD_FN addq( void * data ) {
 THREAD_FN delq(void * data) {
 	struct lfq_ctx * ctx = data;
 	struct user_data * p;
-	int tid = ATOMIC_ADD(&cn_t, 1);
+	int tid = ATOMIC_ADD(&cn_t, 1) - 1;
 
 	long deleted = 0;
 	while(1) {