From: Peter Zijlstra <peterz@infradead.org>
To: torvalds@linux-foundation.org
Cc: corbet@lwn.net, will@kernel.org, peterz@infradead.org,
boqun.feng@gmail.com, mark.rutland@arm.com,
catalin.marinas@arm.com, dennis@kernel.org, tj@kernel.org,
cl@linux.com, hca@linux.ibm.com, gor@linux.ibm.com,
agordeev@linux.ibm.com, borntraeger@linux.ibm.com,
svens@linux.ibm.com, tglx@linutronix.de, mingo@redhat.com,
bp@alien8.de, dave.hansen@linux.intel.com, x86@kernel.org,
hpa@zytor.com, joro@8bytes.org, suravee.suthikulpanit@amd.com,
robin.murphy@arm.com, dwmw2@infradead.org,
baolu.lu@linux.intel.com, Arnd Bergmann <arnd@arndb.de>,
Herbert Xu <herbert@gondor.apana.org.au>,
davem@davemloft.net, penberg@kernel.org, rientjes@google.com,
iamjoonsoo.kim@lge.com, Andrew Morton <akpm@linux-foundation.org>,
vbabka@suse.cz, roman.gushchin@linux.dev, 42.hyeyoo@gmail.com,
linux-doc@vger.kernel.org, linux-kernel@vger.kernel.org,
linux-mm@kvack.org, linux-s390@vger.kernel.org,
iommu@lists.linux.dev, linux-arch@vger.kernel.org,
linux-crypto@vger.kernel.org, sfr@canb.auug.org.au,
mpe@ellerman.id.au, James.Bottomley@hansenpartnership.com,
deller@gmx.de, linux-parisc@vger.kernel.org
Subject: [PATCH 10/12] slub: Replace cmpxchg_double()
Date: Wed, 31 May 2023 15:08:43 +0200 [thread overview]
Message-ID: <20230531132323.924677086@infradead.org> (raw)
In-Reply-To: 20230531130833.635651916@infradead.org
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>
---
include/linux/slub_def.h | 12 +++-
mm/slab.h | 53 +++++++++++++++--
mm/slub.c | 139 +++++++++++++++++++++++++++--------------------
3 files changed, 137 insertions(+), 67 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
@@ -6,6 +6,38 @@
*/
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;
@@ -38,14 +70,21 @@ 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;
@@ -72,8 +111,8 @@ SLAB_MATCH(memcg_data, memcg_data);
#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
/**
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -292,7 +292,12 @@ static inline bool kmem_cache_has_cpu_pa
/* 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.
@@ -512,6 +517,40 @@ 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)
+{
+#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
@@ -519,33 +558,25 @@ 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 +588,26 @@ 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);
@@ -2228,7 +2249,7 @@ 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"))
@@ -2554,7 +2575,7 @@ 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")) {
@@ -2611,7 +2632,7 @@ 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"));
@@ -3008,6 +3029,18 @@ 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_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.
@@ -3034,7 +3067,7 @@ 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"));
@@ -3359,11 +3392,7 @@ 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;
}
@@ -3631,7 +3660,7 @@ 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"));
@@ -3736,11 +3765,7 @@ 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 +4530,11 @@ static int kmem_cache_open(struct kmem_c
}
}
-#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
/*
next prev parent reply other threads:[~2023-05-31 13:28 UTC|newest]
Thread overview: 34+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-05-31 13:08 [PATCH 00/12] Introduce cmpxchg128() -- aka. the demise of cmpxchg_double() Peter Zijlstra
2023-05-31 13:08 ` [PATCH 01/12] cyrpto/b128ops: Remove struct u128 Peter Zijlstra
2023-05-31 13:08 ` [PATCH 02/12] types: Introduce [us]128 Peter Zijlstra
2023-05-31 13:08 ` [PATCH 03/12] arch: Introduce arch_{,try_}_cmpxchg128{,_local}() Peter Zijlstra
2023-05-31 13:08 ` [PATCH 04/12] instrumentation: Wire up cmpxchg128() Peter Zijlstra
2023-05-31 13:08 ` [PATCH 05/12] percpu: Add {raw,this}_cpu_try_cmpxchg() Peter Zijlstra
2023-06-09 16:10 ` [PATCH v2 5/12] " Konrad Dybcio
2023-06-09 16:13 ` Nathan Chancellor
2023-06-09 16:20 ` Konrad Dybcio
2023-05-31 13:08 ` [PATCH 06/12] percpu: Wire up cmpxchg128 Peter Zijlstra
2023-05-31 13:08 ` [PATCH 07/12] percpu: #ifndef __SIZEOF_INT128__ Peter Zijlstra
2023-05-31 14:21 ` Arnd Bergmann
2023-05-31 15:50 ` Peter Zijlstra
2023-06-01 10:14 ` [PATCH v2 07/12] parisc/percpu: Work around the lack of __SIZEOF_INT128__ Peter Zijlstra
2023-06-01 10:32 ` Helge Deller
2023-06-01 10:50 ` Peter Zijlstra
2023-06-01 21:08 ` Sam James
2023-06-01 13:29 ` Linus Torvalds
2023-06-01 17:38 ` Masahiro Yamada
2023-06-02 14:39 ` Peter Zijlstra
2023-06-02 14:50 ` Mark Rutland
2023-06-02 16:09 ` Linus Torvalds
2023-06-02 17:00 ` H. Peter Anvin
2023-06-02 19:10 ` Peter Zijlstra
2023-06-02 19:20 ` H. Peter Anvin
2023-06-02 19:40 ` Peter Zijlstra
2023-06-02 19:57 ` Linus Torvalds
2023-06-02 20:42 ` Helge Deller
2023-05-31 13:08 ` [PATCH 08/12] x86,amd_iommu: Replace cmpxchg_double() Peter Zijlstra
2023-05-31 13:08 ` [PATCH 09/12] x86,intel_iommu: " Peter Zijlstra
2023-05-31 13:08 ` Peter Zijlstra [this message]
2023-05-31 13:08 ` [PATCH 11/12] arch: Remove cmpxchg_double Peter Zijlstra
2023-05-31 13:08 ` [PATCH 12/12] s390/cpum_sf: Convert to cmpxchg128() Peter Zijlstra
2023-05-31 14:47 ` [PATCH 00/12] Introduce cmpxchg128() -- aka. the demise of cmpxchg_double() Mark Rutland
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230531132323.924677086@infradead.org \
--to=peterz@infradead.org \
--cc=42.hyeyoo@gmail.com \
--cc=James.Bottomley@hansenpartnership.com \
--cc=agordeev@linux.ibm.com \
--cc=akpm@linux-foundation.org \
--cc=arnd@arndb.de \
--cc=baolu.lu@linux.intel.com \
--cc=boqun.feng@gmail.com \
--cc=borntraeger@linux.ibm.com \
--cc=bp@alien8.de \
--cc=catalin.marinas@arm.com \
--cc=cl@linux.com \
--cc=corbet@lwn.net \
--cc=dave.hansen@linux.intel.com \
--cc=davem@davemloft.net \
--cc=deller@gmx.de \
--cc=dennis@kernel.org \
--cc=dwmw2@infradead.org \
--cc=gor@linux.ibm.com \
--cc=hca@linux.ibm.com \
--cc=herbert@gondor.apana.org.au \
--cc=hpa@zytor.com \
--cc=iamjoonsoo.kim@lge.com \
--cc=iommu@lists.linux.dev \
--cc=joro@8bytes.org \
--cc=linux-arch@vger.kernel.org \
--cc=linux-crypto@vger.kernel.org \
--cc=linux-doc@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=linux-parisc@vger.kernel.org \
--cc=linux-s390@vger.kernel.org \
--cc=mark.rutland@arm.com \
--cc=mingo@redhat.com \
--cc=mpe@ellerman.id.au \
--cc=penberg@kernel.org \
--cc=rientjes@google.com \
--cc=robin.murphy@arm.com \
--cc=roman.gushchin@linux.dev \
--cc=sfr@canb.auug.org.au \
--cc=suravee.suthikulpanit@amd.com \
--cc=svens@linux.ibm.com \
--cc=tglx@linutronix.de \
--cc=tj@kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=vbabka@suse.cz \
--cc=will@kernel.org \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).