[PATCH v2 08/10] slub: Replace cmpxchg_double()
From: Peter Zijlstra
Date: Thu Feb 02 2023 - 10:31:47 EST
Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
---
include/linux/slub_def.h | 12 ++-
mm/slab.h | 45 +++++++++++++-
mm/slub.c | 142 ++++++++++++++++++++++++++++-------------------
3 files changed, 135 insertions(+), 64 deletions(-)
--- a/include/linux/slub_def.h
+++ b/include/linux/slub_def.h
@@ -39,7 +39,8 @@ enum stat_item {
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
/*
@@ -47,8 +48,13 @@ enum stat_item {
* 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 */
--- a/mm/slab.h
+++ b/mm/slab.h
@@ -5,6 +5,34 @@
* Internal slab definitions
*/
+/*
+ * 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;
+ };
+#ifdef CONFIG_64BIT
+ u128 full;
+#else
+ u64 full;
+#endif
+} freelist_aba_t;
+
+#ifdef CONFIG_64BIT
+# ifdef system_has_cmpxchg128
+# define system_has_freelist_aba() system_has_cmpxchg128()
+# define try_cmpxchg_freelist try_cmpxchg128
+# endif
+#else /* CONFIG_64BIT */
+# ifdef system_has_cmpxchg64
+# define system_has_freelist_aba() system_has_cmpxchg64()
+# define try_cmpxchg_freelist try_cmpxchg64
+# endif
+#endif /* CONFIG_64BIT */
+
/* Reuses the bits in struct page */
struct slab {
unsigned long __page_flags;
@@ -37,14 +65,21 @@ struct slab {
#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;
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -292,7 +292,13 @@ static inline bool kmem_cache_has_cpu_pa
/* Poison object */
#define __OBJECT_POISON ((slab_flags_t __force)0x80000000U)
/* Use cmpxchg_double */
+
+#if defined(system_has_freelist_aba) && \
+ defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
#define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U)
+#else
+#define __CMPXCHG_DOUBLE ((slab_flags_t __force)0U)
+#endif
/*
* Tracking user of a slab.
@@ -512,6 +518,43 @@ static __always_inline void slab_unlock(
__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)
+{
+
+ bool ret = false;
+
+#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 };
+
+ ret = try_cmpxchg_freelist(&slab->freelist_counter.full, &old.full, new.full);
+#endif /* system_has_freelist_aba */
+
+ return ret;
+}
+
+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
@@ -519,33 +562,25 @@ static __always_inline void slab_unlock(
* 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);
@@ -557,36 +592,26 @@ static inline bool __cmpxchg_double_slab
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);
@@ -2229,7 +2254,7 @@ static inline void *acquire_slab(struct
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"))
@@ -2555,7 +2580,7 @@ static void deactivate_slab(struct kmem_
}
- if (!cmpxchg_double_slab(s, slab,
+ if (!slab_update_freelist(s, slab,
old.freelist, old.counters,
new.freelist, new.counters,
"unfreezing slab")) {
@@ -2612,7 +2637,7 @@ static void __unfreeze_partials(struct k
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"));
@@ -3009,6 +3034,18 @@ static inline bool pfmemalloc_match(stru
}
#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_cmpxchg(s->cpu_slab->freelist_tid.full,
+ old.full, new.full) == old.full;
+}
+
/*
* Check the slab->freelist and either transfer the freelist to the
* per cpu freelist or deactivate the slab.
@@ -3035,7 +3072,7 @@ static inline void *get_freelist(struct
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"));
@@ -3360,11 +3397,7 @@ static __always_inline void *__slab_allo
* 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;
}
@@ -3632,7 +3665,7 @@ static void __slab_free(struct kmem_cach
}
}
- } while (!cmpxchg_double_slab(s, slab,
+ } while (!slab_update_freelist(s, slab,
prior, counters,
head, new.counters,
"__slab_free"));
@@ -3737,11 +3770,7 @@ static __always_inline void do_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;
}
@@ -4505,11 +4534,12 @@ static int kmem_cache_open(struct kmem_c
}
}
-#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
+#if defined(system_has_freelist_aba) && \
defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
- if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
+ if (system_has_freelist_aba() && !(s->flags & SLAB_NO_CMPXCHG)) {
/* Enable fast mode */
s->flags |= __CMPXCHG_DOUBLE;
+ }
#endif
/*