| Message ID | 20230515080554.453785148@infradead.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | Introduce cmpxchg128() -- aka. the demise of cmpxchg_double() | expand |
On Mon, May 15, 2023 at 09:57:07AM +0200, Peter Zijlstra wrote: > @@ -3008,6 +3029,22 @@ 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) > +{ > +#ifdef system_has_freelist_aba > + freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; > + freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; > + > + return this_cpu_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, > + old.full, new.full) == old.full; > +#else > + return false; > +#endif > +} > + > /* > * Check the slab->freelist and either transfer the freelist to the > * per cpu freelist or deactivate the slab. > @@ -3359,11 +3396,7 @@ 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; > } > @@ -3736,11 +3769,7 @@ 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; > } This isn't right; the this_cpu_cmpxchg_double() was unconditional and relied on the local_irq_save() fallback when no native cmpxchg128 is present. The below delta makes things boot again when system_has_cmpxchg128 is not defined. I'm going to zap these patches from tip/locking/core for a few days and fold the below back into the series and let it run through the robots again. --- mm/slab.h | 20 +++++++++++--------- mm/slub.c | 6 +----- 2 files changed, 12 insertions(+), 14 deletions(-) diff --git a/mm/slab.h b/mm/slab.h index 5880c70de3d6..b191bf68e6e0 100644 --- a/mm/slab.h +++ b/mm/slab.h @@ -6,36 +6,36 @@ */ void __init kmem_cache_init(void); -#ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE #ifdef CONFIG_64BIT # ifdef system_has_cmpxchg128 # define system_has_freelist_aba() system_has_cmpxchg128() # define try_cmpxchg_freelist try_cmpxchg128 -# define this_cpu_cmpxchg_freelist this_cpu_cmpxchg128 -typedef u128 freelist_full_t; # endif +#define this_cpu_cmpxchg_freelist this_cpu_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 -# define this_cpu_cmpxchg_freelist this_cpu_cmpxchg64 -typedef u64 freelist_full_t; # endif +#define this_cpu_cmpxchg_freelist this_cpu_cmpxchg64 +typedef u64 freelist_full_t; #endif /* CONFIG_64BIT */ -#endif /* CONFIG_HAVE_ALIGNED_STRUCT_PAGE */ + +#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 { -#ifdef system_has_freelist_aba struct { void *freelist; unsigned long counter; }; freelist_full_t full; -#endif } freelist_aba_t; /* Reuses the bits in struct page */ @@ -82,7 +82,9 @@ struct slab { }; }; }; +#ifdef system_has_freelist_aba freelist_aba_t freelist_counter; +#endif }; }; struct rcu_head rcu_head; @@ -110,7 +112,7 @@ SLAB_MATCH(memcg_data, memcg_data); #undef SLAB_MATCH static_assert(sizeof(struct slab) <= sizeof(struct page)); #if defined(system_has_freelist_aba) && defined(CONFIG_SLUB) -static_assert(IS_ALIGNED(offsetof(struct slab, freelist), 2*sizeof(void *))); +static_assert(IS_ALIGNED(offsetof(struct slab, freelist), sizeof(freelist_aba_t))); #endif /** diff --git a/mm/slub.c b/mm/slub.c index 161b091746b7..af92c770606d 100644 --- a/mm/slub.c +++ b/mm/slub.c @@ -3034,15 +3034,11 @@ __update_cpu_freelist_fast(struct kmem_cache *s, void *freelist_old, void *freelist_new, unsigned long tid) { -#ifdef system_has_freelist_aba freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; return this_cpu_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, old.full, new.full) == old.full; -#else - return false; -#endif } /*
On 5/24/23 11:32, Peter Zijlstra wrote: > On Mon, May 15, 2023 at 09:57:07AM +0200, Peter Zijlstra wrote: > >> @@ -3008,6 +3029,22 @@ 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) >> +{ >> +#ifdef system_has_freelist_aba >> + freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; >> + freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; >> + >> + return this_cpu_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, >> + old.full, new.full) == old.full; >> +#else >> + return false; >> +#endif >> +} >> + >> /* >> * Check the slab->freelist and either transfer the freelist to the >> * per cpu freelist or deactivate the slab. >> @@ -3359,11 +3396,7 @@ 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; >> } >> @@ -3736,11 +3769,7 @@ 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; >> } > > This isn't right; the this_cpu_cmpxchg_double() was unconditional and > relied on the local_irq_save() fallback when no native cmpxchg128 is > present. > > The below delta makes things boot again when system_has_cmpxchg128 is > not defined. Right, that should do. > I'm going to zap these patches from tip/locking/core for a few days and > fold the below back into the series and let it run through the robots > again. I noticed some comments in mm/slub.c still mention "cmpxchg_double", dunno how much you want to clean it right now or can be postponed. Also some sysfs stats files for CONFIG_SLUB_STATS (not widely used) which we probably might try renaming without breaking anyone, but it's not guaranteed.
On Wed, May 24, 2023 at 11:32:47AM +0200, Peter Zijlstra wrote: > On Mon, May 15, 2023 at 09:57:07AM +0200, Peter Zijlstra wrote: > > > @@ -3008,6 +3029,22 @@ 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) > > +{ > > +#ifdef system_has_freelist_aba > > + freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; > > + freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; > > + > > + return this_cpu_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, > > + old.full, new.full) == old.full; > > +#else > > + return false; > > +#endif > > +} > > + > > /* > > * Check the slab->freelist and either transfer the freelist to the > > * per cpu freelist or deactivate the slab. > > @@ -3359,11 +3396,7 @@ 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; > > } > > @@ -3736,11 +3769,7 @@ 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; > > } > > This isn't right; the this_cpu_cmpxchg_double() was unconditional and > relied on the local_irq_save() fallback when no native cmpxchg128 is > present. This then also means I need to look at this_cpu_cmpxchg128 and this_cpu_cmoxchg64 behaviour when we dont have the CPUID feature. Because current verions seem to assume the instruction is present.
On Thu, May 25, 2023, at 12:29, Peter Zijlstra wrote: > On Wed, May 24, 2023 at 11:32:47AM +0200, Peter Zijlstra wrote: >> On Mon, May 15, 2023 at 09:57:07AM +0200, Peter Zijlstra wrote: > > This then also means I need to look at this_cpu_cmpxchg128 and > this_cpu_cmoxchg64 behaviour when we dont have the CPUID feature. > > Because current verions seem to assume the instruction is present. As far as I could tell when reviewing your series, this_cpu_cmpxchg64() is always available on all architectures. Depending on compile-time feature detection this would be either a native instruction that is guaranteed to work, or the irq-disabled version. On x86, this is handled at runtime with alternative_io(). this_cpu_cmpxchg128() clearly needed the system_has_cmpxchg128() check, same as system_has_cmpxchg_double() today. Arnd
On Thu, May 25, 2023 at 12:52:06PM +0200, Arnd Bergmann wrote: > On Thu, May 25, 2023, at 12:29, Peter Zijlstra wrote: > > On Wed, May 24, 2023 at 11:32:47AM +0200, Peter Zijlstra wrote: > >> On Mon, May 15, 2023 at 09:57:07AM +0200, Peter Zijlstra wrote: > > > > This then also means I need to look at this_cpu_cmpxchg128 and > > this_cpu_cmoxchg64 behaviour when we dont have the CPUID feature. > > > > Because current verions seem to assume the instruction is present. > > As far as I could tell when reviewing your series, this_cpu_cmpxchg64() > is always available on all architectures. Depending on compile-time > feature detection this would be either a native instruction that > is guaranteed to work, or the irq-disabled version. On x86, this > is handled at runtime with alternative_io(). > > this_cpu_cmpxchg128() clearly needed the system_has_cmpxchg128() > check, same as system_has_cmpxchg_double() today. So, having just dug through all that, on x86: this_cpu_cmpxchg64() is: X86_CMPXCHG64=n -> fallback, irrespective of CX8 X86_CMPXCHG64=y -> cmpxchg8b X86_64 -> cmpxchg I've changed it to be similar between 32bit and 64bit such that both: cmpxchg#b when CX#, otherwise this_cpu_cmpxchg#b_emu
On Wed, May 24, 2023 at 11:32:47AM +0200, Peter Zijlstra wrote: > On Mon, May 15, 2023 at 09:57:07AM +0200, Peter Zijlstra wrote: > > > @@ -3008,6 +3029,22 @@ 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) > > +{ > > +#ifdef system_has_freelist_aba > > + freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; > > + freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; > > + > > + return this_cpu_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, > > + old.full, new.full) == old.full; > > +#else > > + return false; > > +#endif > > +} > > + > > /* > > * Check the slab->freelist and either transfer the freelist to the > > * per cpu freelist or deactivate the slab. > > @@ -3359,11 +3396,7 @@ 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; > > } > > @@ -3736,11 +3769,7 @@ 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; > > } > > This isn't right; the this_cpu_cmpxchg_double() was unconditional and > relied on the local_irq_save() fallback when no native cmpxchg128 is > present. This means this_cpu_cmpxchg128 is expected to be present on all 64bit archs, except Mark just found out that HPPA doens't support __int128 until gcc-11. (I've been building using gcc-12.2) And because the cmpxchg128 fallback relies on '==' we can't trivally fudge that with a struct type either :/ Now, afaict it all magically works if I use: #ifdef __SIZEOF_INT128__ typedef __s128 s128 typedef __u128 u128 #else #if defined(CONFIG_PARISC) && defined(CONFIG_64BIT) typedef long double u128; #endif #endif but that is *super* gross. The alternative is raising the minimum GCC for PARISC to gcc-11.. Yet another alternative is using a struct type and an equality function, just for this. Anybody?
On Tue, May 30, 2023 at 04:22:32PM +0200, Peter Zijlstra wrote: > Yet another alternative is using a struct type and an equality function, > just for this. The best I could come up with in the regard is the below. It builds on HPPA64 and x86_64, but I've not ran it yet. (also, the introduction of this_cpu_try_cmpxchg() should probably be split out into its own patch) --- a/include/asm-generic/percpu.h +++ b/include/asm-generic/percpu.h @@ -99,6 +99,15 @@ do { \ __ret; \ }) +#define raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval) \ +({ \ + typeof(pcp) __ret, __old = *(ovalp); \ + __ret = raw_cpu_cmpxchg(pcp, __old, nval); \ + if (!likely(__ret == __old)) \ + *(ovalp) = __ret; \ + likely(__ret == __old); \ +}) + #define __this_cpu_generic_read_nopreempt(pcp) \ ({ \ typeof(pcp) ___ret; \ @@ -167,6 +176,15 @@ do { \ __ret; \ }) +#define this_cpu_generic_try_cmpxchg(pcp, ovalp, nval) \ +({ \ + typeof(pcp) __ret, __old = *(ovalp); \ + __ret = this_cpu_cmpxchg(pcp, __old, nval); \ + if (!likely(__ret == __old)) \ + *(ovalp) = __ret; \ + likely(__ret == __old); \ +}) + #ifndef raw_cpu_read_1 #define raw_cpu_read_1(pcp) raw_cpu_generic_read(pcp) #endif @@ -258,6 +276,36 @@ do { \ #define raw_cpu_xchg_8(pcp, nval) raw_cpu_generic_xchg(pcp, nval) #endif +#ifndef __SIZEOF_INT128__ +#define raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval) \ +({ \ + typeof(pcp) *__p = raw_cpu_ptr(&(pcp)); \ + typeof(pcp) __ret, __old = *(ovalp); \ + bool __s; \ + __ret = *__p; \ + if (!__builtin_memcmp(&__ret, &__old, sizeof(pcp))) { \ + *__p = nval; \ + __s = true; \ + } else { \ + *(ovalp) = __ret; \ + __s = false; \ + } \ + __s; \ +}) + +#define raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval) \ +({ \ + typeof(pcp) __old = (oval); \ + raw_cpu_generic_try_cmpxchg_memcpy(pcp, &__old, nval); \ + __old; \ +}) + +#define raw_cpu_cmpxchg128(pcp, oval, nval) \ + raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval) +#define raw_cpu_try_cmpxchg128(pcp, ovalp, nval) \ + raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval) +#endif + #ifndef raw_cpu_cmpxchg_1 #define raw_cpu_cmpxchg_1(pcp, oval, nval) \ raw_cpu_generic_cmpxchg(pcp, oval, nval) @@ -283,6 +331,31 @@ do { \ raw_cpu_generic_cmpxchg(pcp, oval, nval) #endif +#ifndef raw_cpu_try_cmpxchg_1 +#define raw_cpu_try_cmpxchg_1(pcp, ovalp, nval) \ + raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef raw_cpu_try_cmpxchg_2 +#define raw_cpu_try_cmpxchg_2(pcp, ovalp, nval) \ + raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef raw_cpu_try_cmpxchg_4 +#define raw_cpu_try_cmpxchg_4(pcp, ovalp, nval) \ + raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef raw_cpu_try_cmpxchg_8 +#define raw_cpu_try_cmpxchg_8(pcp, ovalp, nval) \ + raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef raw_cpu_try_cmpxchg64 +#define raw_cpu_try_cmpxchg64(pcp, ovalp, nval) \ + raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef raw_cpu_try_cmpxchg128 +#define raw_cpu_try_cmpxchg128(pcp, ovalp, nval) \ + raw_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif + #ifndef this_cpu_read_1 #define this_cpu_read_1(pcp) this_cpu_generic_read(pcp) #endif @@ -374,6 +447,33 @@ do { \ #define this_cpu_xchg_8(pcp, nval) this_cpu_generic_xchg(pcp, nval) #endif +#ifndef __SIZEOF_INT128__ +#define this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval) \ +({ \ + bool __ret; \ + unsigned long __flags; \ + raw_local_irq_save(__flags); \ + __ret = raw_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval); \ + raw_local_irq_restore(__flags); \ + __ret; \ +}) + +#define this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval) \ +({ \ + typeof(pcp) __ret; \ + unsigned long __flags; \ + raw_local_irq_save(__flags); \ + __ret = raw_cpu_generic_cmpxchg_memcmp(pcp, oval, nval); \ + raw_local_irq_restore(__flags); \ + __ret; \ +}) + +#define this_cpu_cmpxchg128(pcp, oval, nval) \ + this_cpu_generic_cmpxchg_memcmp(pcp, oval, nval) +#define this_cpu_try_cmpxchg128(pcp, ovalp, nval) \ + this_cpu_generic_try_cmpxchg_memcmp(pcp, ovalp, nval) +#endif + #ifndef this_cpu_cmpxchg_1 #define this_cpu_cmpxchg_1(pcp, oval, nval) \ this_cpu_generic_cmpxchg(pcp, oval, nval) @@ -399,4 +499,29 @@ do { \ this_cpu_generic_cmpxchg(pcp, oval, nval) #endif +#ifndef this_cpu_try_cmpxchg_1 +#define this_cpu_try_cmpxchg_1(pcp, ovalp, nval) \ + this_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef this_cpu_try_cmpxchg_2 +#define this_cpu_try_cmpxchg_2(pcp, ovalp, nval) \ + this_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef this_cpu_try_cmpxchg_4 +#define this_cpu_try_cmpxchg_4(pcp, ovalp, nval) \ + this_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef this_cpu_try_cmpxchg_8 +#define this_cpu_try_cmpxchg_8(pcp, ovalp, nval) \ + this_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef this_cpu_try_cmpxchg64 +#define this_cpu_try_cmpxchg64(pcp, ovalp, nval) \ + this_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif +#ifndef this_cpu_try_cmpxchg128 +#define this_cpu_try_cmpxchg128(pcp, ovalp, nval) \ + this_cpu_generic_try_cmpxchg(pcp, ovalp, nval) +#endif + #endif /* _ASM_GENERIC_PERCPU_H_ */ --- a/include/linux/types.h +++ b/include/linux/types.h @@ -13,6 +13,13 @@ #ifdef __SIZEOF_INT128__ typedef __s128 s128; typedef __u128 u128; +#else +#ifdef CONFIG_64BIT +/* hack for this_cpu_cmpxchg128 */ +typedef struct { + u64 a, b; +} u128 __attribute__((aligned(16))); +#endif #endif typedef u32 __kernel_dev_t; --- a/mm/slab.h +++ b/mm/slab.h @@ -11,14 +11,14 @@ void __init kmem_cache_init(void); # define system_has_freelist_aba() system_has_cmpxchg128() # define try_cmpxchg_freelist try_cmpxchg128 # endif -#define this_cpu_cmpxchg_freelist this_cpu_cmpxchg128 +#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_cmpxchg_freelist this_cpu_cmpxchg64 +#define this_cpu_try_cmpxchg_freelist this_cpu_try_cmpxchg64 typedef u64 freelist_full_t; #endif /* CONFIG_64BIT */ --- a/mm/slub.c +++ b/mm/slub.c @@ -3037,8 +3037,8 @@ __update_cpu_freelist_fast(struct kmem_c freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; - return this_cpu_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, - old.full, new.full) == old.full; + return this_cpu_try_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, + &old.full, new.full); } /*
--- 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_HAVE_ALIGNED_STRUCT_PAGE +#ifdef CONFIG_64BIT +# ifdef system_has_cmpxchg128 +# define system_has_freelist_aba() system_has_cmpxchg128() +# define try_cmpxchg_freelist try_cmpxchg128 +# define this_cpu_cmpxchg_freelist this_cpu_cmpxchg128 +typedef u128 freelist_full_t; +# endif +#else /* CONFIG_64BIT */ +# ifdef system_has_cmpxchg64 +# define system_has_freelist_aba() system_has_cmpxchg64() +# define try_cmpxchg_freelist try_cmpxchg64 +# define this_cpu_cmpxchg_freelist this_cpu_cmpxchg64 +typedef u64 freelist_full_t; +# endif +#endif /* CONFIG_64BIT */ +#endif /* CONFIG_HAVE_ALIGNED_STRUCT_PAGE */ + +/* + * Freelist pointer and counter to cmpxchg together, avoids the typical ABA + * problems with cmpxchg of just a pointer. + */ +typedef union { +#ifdef system_has_freelist_aba + struct { + void *freelist; + unsigned long counter; + }; + freelist_full_t full; +#endif +} freelist_aba_t; + /* Reuses the bits in struct page */ struct slab { unsigned long __page_flags; @@ -38,14 +70,19 @@ 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; + }; + }; }; + freelist_aba_t freelist_counter; }; }; struct rcu_head rcu_head; @@ -72,7 +109,7 @@ 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) +#if defined(system_has_freelist_aba) && defined(CONFIG_SLUB) static_assert(IS_ALIGNED(offsetof(struct slab, freelist), 2*sizeof(void *))); #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 +3029,22 @@ 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) +{ +#ifdef system_has_freelist_aba + freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; + freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; + + return this_cpu_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, + old.full, new.full) == old.full; +#else + return false; +#endif +} + /* * Check the slab->freelist and either transfer the freelist to the * per cpu freelist or deactivate the slab. @@ -3034,7 +3071,7 @@ 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 +3396,7 @@ 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 +3664,7 @@ 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 +3769,7 @@ 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 +4534,11 @@ 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 /*