slub: Replace cmpxchg_double()

Signed-off-by: Peter Zijlstra (Intel) <[e-mail]>
Reviewed-by: Arnd Bergmann <[e-mail]>
Reviewed-by: Mark Rutland <[e-mail]>
Acked-by: Vlastimil Babka <[e-mail]>
Acked-by: Hyeonggon Yoo <[e-mail]>
Tested-by: Mark Rutland <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]

include/linux/slub_def.h

···
 	CPU_PARTIAL_FREE,	/* Refill cpu partial on free */
 	CPU_PARTIAL_NODE,	/* Refill cpu partial from node partial */
 	CPU_PARTIAL_DRAIN,	/* Drain cpu partial to node partial */
-	NR_SLUB_STAT_ITEMS };
+	NR_SLUB_STAT_ITEMS
+};
 
 #ifndef CONFIG_SLUB_TINY
 /*
···
  * with this_cpu_cmpxchg_double() alignment requirements.
  */
 struct kmem_cache_cpu {
-	void **freelist;	/* Pointer to next available object */
-	unsigned long tid;	/* Globally unique transaction id */
+	union {
+		struct {
+			void **freelist;	/* Pointer to next available object */
+			unsigned long tid;	/* Globally unique transaction id */
+		};
+		freelist_aba_t freelist_tid;
+	};
 	struct slab *slab;	/* The slab from which we are allocating */
 #ifdef CONFIG_SLUB_CPU_PARTIAL
 	struct slab *partial;	/* Partially allocated frozen slabs */

mm/slab.h

···
  */
 void __init kmem_cache_init(void);
 
+#ifdef CONFIG_64BIT
+# ifdef system_has_cmpxchg128
+# define system_has_freelist_aba()	system_has_cmpxchg128()
+# define try_cmpxchg_freelist		try_cmpxchg128
+# endif
+#define this_cpu_try_cmpxchg_freelist	this_cpu_try_cmpxchg128
+typedef u128 freelist_full_t;
+#else /* CONFIG_64BIT */
+# ifdef system_has_cmpxchg64
+# define system_has_freelist_aba()	system_has_cmpxchg64()
+# define try_cmpxchg_freelist		try_cmpxchg64
+# endif
+#define this_cpu_try_cmpxchg_freelist	this_cpu_try_cmpxchg64
+typedef u64 freelist_full_t;
+#endif /* CONFIG_64BIT */
+
+#if defined(system_has_freelist_aba) && !defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
+#undef system_has_freelist_aba
+#endif
+
+/*
+ * Freelist pointer and counter to cmpxchg together, avoids the typical ABA
+ * problems with cmpxchg of just a pointer.
+ */
+typedef union {
+	struct {
+		void *freelist;
+		unsigned long counter;
+	};
+	freelist_full_t full;
+} freelist_aba_t;
+
 /* Reuses the bits in struct page */
 struct slab {
 	unsigned long __page_flags;
···
 #endif
 			};
 			/* Double-word boundary */
-			void *freelist;		/* first free object */
 			union {
-				unsigned long counters;
 				struct {
-					unsigned inuse:16;
-					unsigned objects:15;
-					unsigned frozen:1;
+					void *freelist;		/* first free object */
+					union {
+						unsigned long counters;
+						struct {
+							unsigned inuse:16;
+							unsigned objects:15;
+							unsigned frozen:1;
+						};
+					};
 				};
+#ifdef system_has_freelist_aba
+				freelist_aba_t freelist_counter;
+#endif
 			};
 		};
 		struct rcu_head rcu_head;
···
 #endif
 #undef SLAB_MATCH
 static_assert(sizeof(struct slab) <= sizeof(struct page));
-#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && defined(CONFIG_SLUB)
-static_assert(IS_ALIGNED(offsetof(struct slab, freelist), 2*sizeof(void *)));
+#if defined(system_has_freelist_aba) && defined(CONFIG_SLUB)
+static_assert(IS_ALIGNED(offsetof(struct slab, freelist), sizeof(freelist_aba_t)));
 #endif
 
 /**

mm/slub.c

···
 /* Poison object */
 #define __OBJECT_POISON		((slab_flags_t __force)0x80000000U)
 /* Use cmpxchg_double */
+
+#ifdef system_has_freelist_aba
 #define __CMPXCHG_DOUBLE	((slab_flags_t __force)0x40000000U)
+#else
+#define __CMPXCHG_DOUBLE	((slab_flags_t __force)0U)
+#endif
 
 /*
  * Tracking user of a slab.
···
 	__bit_spin_unlock(PG_locked, &page->flags);
 }
 
+static inline bool
+__update_freelist_fast(struct slab *slab,
+		      void *freelist_old, unsigned long counters_old,
+		      void *freelist_new, unsigned long counters_new)
+{
+#ifdef system_has_freelist_aba
+	freelist_aba_t old = { .freelist = freelist_old, .counter = counters_old };
+	freelist_aba_t new = { .freelist = freelist_new, .counter = counters_new };
+
+	return try_cmpxchg_freelist(&slab->freelist_counter.full, &old.full, new.full);
+#else
+	return false;
+#endif
+}
+
+static inline bool
+__update_freelist_slow(struct slab *slab,
+		      void *freelist_old, unsigned long counters_old,
+		      void *freelist_new, unsigned long counters_new)
+{
+	bool ret = false;
+
+	slab_lock(slab);
+	if (slab->freelist == freelist_old &&
+	    slab->counters == counters_old) {
+		slab->freelist = freelist_new;
+		slab->counters = counters_new;
+		ret = true;
+	}
+	slab_unlock(slab);
+
+	return ret;
+}
+
 /*
  * Interrupts must be disabled (for the fallback code to work right), typically
  * by an _irqsave() lock variant. On PREEMPT_RT the preempt_disable(), which is
···
  * allocation/ free operation in hardirq context. Therefore nothing can
  * interrupt the operation.
  */
-static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
+static inline bool __slab_update_freelist(struct kmem_cache *s, struct slab *slab,
 		void *freelist_old, unsigned long counters_old,
 		void *freelist_new, unsigned long counters_new,
 		const char *n)
 {
+	bool ret;
+
 	if (USE_LOCKLESS_FAST_PATH())
 		lockdep_assert_irqs_disabled();
-#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
-    defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
+
 	if (s->flags & __CMPXCHG_DOUBLE) {
-		if (cmpxchg_double(&slab->freelist, &slab->counters,
-				   freelist_old, counters_old,
-				   freelist_new, counters_new))
-			return true;
-	} else
-#endif
-	{
-		slab_lock(slab);
-		if (slab->freelist == freelist_old &&
-					slab->counters == counters_old) {
-			slab->freelist = freelist_new;
-			slab->counters = counters_new;
-			slab_unlock(slab);
-			return true;
-		}
-		slab_unlock(slab);
+		ret = __update_freelist_fast(slab, freelist_old, counters_old,
+				            freelist_new, counters_new);
+	} else {
+		ret = __update_freelist_slow(slab, freelist_old, counters_old,
+				            freelist_new, counters_new);
 	}
+	if (likely(ret))
+		return true;
 
 	cpu_relax();
 	stat(s, CMPXCHG_DOUBLE_FAIL);
···
 	return false;
 }
 
-static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab,
+static inline bool slab_update_freelist(struct kmem_cache *s, struct slab *slab,
 		void *freelist_old, unsigned long counters_old,
 		void *freelist_new, unsigned long counters_new,
 		const char *n)
 {
-#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
-    defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
+	bool ret;
+
 	if (s->flags & __CMPXCHG_DOUBLE) {
-		if (cmpxchg_double(&slab->freelist, &slab->counters,
-				   freelist_old, counters_old,
-				   freelist_new, counters_new))
-			return true;
-	} else
-#endif
-	{
+		ret = __update_freelist_fast(slab, freelist_old, counters_old,
+				            freelist_new, counters_new);
+	} else {
 		unsigned long flags;
 
 		local_irq_save(flags);
-		slab_lock(slab);
-		if (slab->freelist == freelist_old &&
-					slab->counters == counters_old) {
-			slab->freelist = freelist_new;
-			slab->counters = counters_new;
-			slab_unlock(slab);
-			local_irq_restore(flags);
-			return true;
-		}
-		slab_unlock(slab);
+		ret = __update_freelist_slow(slab, freelist_old, counters_old,
+				            freelist_new, counters_new);
 		local_irq_restore(flags);
 	}
+	if (likely(ret))
+		return true;
 
 	cpu_relax();
 	stat(s, CMPXCHG_DOUBLE_FAIL);
···
 	VM_BUG_ON(new.frozen);
 	new.frozen = 1;
 
-	if (!__cmpxchg_double_slab(s, slab,
+	if (!__slab_update_freelist(s, slab,
 			freelist, counters,
 			new.freelist, new.counters,
 			"acquire_slab"))
···
 	}
 
 
-	if (!cmpxchg_double_slab(s, slab,
+	if (!slab_update_freelist(s, slab,
 				old.freelist, old.counters,
 				new.freelist, new.counters,
 				"unfreezing slab")) {
···
 
 			new.frozen = 0;
 
-		} while (!__cmpxchg_double_slab(s, slab,
+		} while (!__slab_update_freelist(s, slab,
 				old.freelist, old.counters,
 				new.freelist, new.counters,
 				"unfreezing slab"));
···
 }
 
 #ifndef CONFIG_SLUB_TINY
+static inline bool
+__update_cpu_freelist_fast(struct kmem_cache *s,
+			   void *freelist_old, void *freelist_new,
+			   unsigned long tid)
+{
+	freelist_aba_t old = { .freelist = freelist_old, .counter = tid };
+	freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) };
+
+	return this_cpu_try_cmpxchg_freelist(s->cpu_slab->freelist_tid.full,
+					     &old.full, new.full);
+}
+
 /*
  * Check the slab->freelist and either transfer the freelist to the
  * per cpu freelist or deactivate the slab.
···
 		new.inuse = slab->objects;
 		new.frozen = freelist != NULL;
 
-	} while (!__cmpxchg_double_slab(s, slab,
+	} while (!__slab_update_freelist(s, slab,
 		freelist, counters,
 		NULL, new.counters,
 		"get_freelist"));
···
 		 * against code executing on this cpu *not* from access by
 		 * other cpus.
 		 */
-		if (unlikely(!this_cpu_cmpxchg_double(
-				s->cpu_slab->freelist, s->cpu_slab->tid,
-				object, tid,
-				next_object, next_tid(tid)))) {
-
+		if (unlikely(!__update_cpu_freelist_fast(s, object, next_object, tid))) {
 			note_cmpxchg_failure("slab_alloc", s, tid);
 			goto redo;
 		}
···
 			}
 		}
 
-	} while (!cmpxchg_double_slab(s, slab,
+	} while (!slab_update_freelist(s, slab,
 		prior, counters,
 		head, new.counters,
 		"__slab_free"));
···
 
 		set_freepointer(s, tail_obj, freelist);
 
-		if (unlikely(!this_cpu_cmpxchg_double(
-				s->cpu_slab->freelist, s->cpu_slab->tid,
-				freelist, tid,
-				head, next_tid(tid)))) {
-
+		if (unlikely(!__update_cpu_freelist_fast(s, freelist, head, tid))) {
 			note_cmpxchg_failure("slab_free", s, tid);
 			goto redo;
 		}
···
 		}
 	}
 
-#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
-    defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
-	if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
+#ifdef system_has_freelist_aba
+	if (system_has_freelist_aba() && !(s->flags & SLAB_NO_CMPXCHG)) {
 		/* Enable fast mode */
 		s->flags |= __CMPXCHG_DOUBLE;
+	}
 #endif
 
 	/*