From 3032f0c9008088a3effdc2622ce16c3e1bcb13a2 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Thu, 7 Mar 2019 13:29:59 -0800 Subject: [PATCH] ARCv2: spinlock: remove the extra smp_mb before lock, after unlock - ARCv2 LLSC spinlocks have smp_mb() both before and after the LLSC instructions, which is not required per lkmm ACQ/REL semantics. smp_mb() is only needed _after_ lock and _before_ unlock. So remove the extra barriers. The reason they were there was mainly historical. At the time of initial SMP Linux bringup on HS38 cores, I was too conservative, given the fluidity of both hw and sw. The last attempt to ditch the extra barrier showed some hackbench regression which is apparently not the case now (atleast for LLSC case, read on...) - EX based spinlocks (!CONFIG_ARC_HAS_LLSC) still needs the extra smp_mb(), not due to lkmm, but due to some hardware shenanigans. W/o that, hackbench triggers RCU stall splat so extra DMB is retained !LLSC based systems are not realistic Linux sstem anyways so they can afford to be a nit suboptimal ;-) | [ARCLinux]# for i in (seq 1 1 5) ; do hackbench; done | Running with 10 groups 400 process | INFO: task hackbench:158 blocked for more than 10 seconds. | Not tainted 4.20.0-00005-g96b18288a88e-dirty #117 | "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. | hackbench D 0 158 135 0x00000000 | | Stack Trace: | watchdog: BUG: soft lockup - CPU#3 stuck for 59s! [hackbench:469] | Modules linked in: | Path: (null) | CPU: 3 PID: 469 Comm: hackbench Not tainted 4.20.0-00005-g96b18288a88e-dirty | | [ECR ]: 0x00000000 => Check Programmer's Manual | [EFA ]: 0x00000000 | [BLINK ]: do_exit+0x4a6/0x7d0 | [ERET ]: _raw_write_unlock_irq+0x44/0x5c - And while at it, remove the extar smp_mb() from EX based arch_read_trylock() since the spin lock there guarantees a full barrier anyways - For LLSC case, hackbench threads improves with this patch (HAPS @ 50MHz) ---- before ---- | | [ARCLinux]# for i in 1 2 3 4 5; do hackbench 10 thread; done | Running with 10 groups 400 threads | Time: 16.253 | Time: 16.445 | Time: 16.590 | Time: 16.721 | Time: 16.544 ---- after ---- | | [ARCLinux]# for i in 1 2 3 4 5; do hackbench 10 thread; done | Running with 10 groups 400 threads | Time: 15.638 | Time: 15.730 | Time: 15.870 | Time: 15.842 | Time: 15.729 Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/include/asm/spinlock.h | 49 ++++++++++++----------------------------- 1 file changed, 14 insertions(+), 35 deletions(-) diff --git a/arch/arc/include/asm/spinlock.h b/arch/arc/include/asm/spinlock.h index 2ba04a7..daa914d 100644 --- a/arch/arc/include/asm/spinlock.h +++ b/arch/arc/include/asm/spinlock.h @@ -21,8 +21,6 @@ static inline void arch_spin_lock(arch_spinlock_t *lock) { unsigned int val; - smp_mb(); - __asm__ __volatile__( "1: llock %[val], [%[slock]] \n" " breq %[val], %[LOCKED], 1b \n" /* spin while LOCKED */ @@ -34,6 +32,14 @@ static inline void arch_spin_lock(arch_spinlock_t *lock) [LOCKED] "r" (__ARCH_SPIN_LOCK_LOCKED__) : "memory", "cc"); + /* + * ACQUIRE barrier to ensure load/store after taking the lock + * don't "bleed-up" out of the critical section (leak-in is allowed) + * http://www.spinics.net/lists/kernel/msg2010409.html + * + * ARCv2 only has load-load, store-store and all-all barrier + * thus need the full all-all barrier + */ smp_mb(); } @@ -42,8 +48,6 @@ static inline int arch_spin_trylock(arch_spinlock_t *lock) { unsigned int val, got_it = 0; - smp_mb(); - __asm__ __volatile__( "1: llock %[val], [%[slock]] \n" " breq %[val], %[LOCKED], 4f \n" /* already LOCKED, just bail */ @@ -67,9 +71,7 @@ static inline void arch_spin_unlock(arch_spinlock_t *lock) { smp_mb(); - lock->slock = __ARCH_SPIN_LOCK_UNLOCKED__; - - smp_mb(); + WRITE_ONCE(lock->slock, __ARCH_SPIN_LOCK_UNLOCKED__); } /* @@ -81,8 +83,6 @@ static inline void arch_read_lock(arch_rwlock_t *rw) { unsigned int val; - smp_mb(); - /* * zero means writer holds the lock exclusively, deny Reader. * Otherwise grant lock to first/subseq reader @@ -113,8 +113,6 @@ static inline int arch_read_trylock(arch_rwlock_t *rw) { unsigned int val, got_it = 0; - smp_mb(); - __asm__ __volatile__( "1: llock %[val], [%[rwlock]] \n" " brls %[val], %[WR_LOCKED], 4f\n" /* <= 0: already write locked, bail */ @@ -140,8 +138,6 @@ static inline void arch_write_lock(arch_rwlock_t *rw) { unsigned int val; - smp_mb(); - /* * If reader(s) hold lock (lock < __ARCH_RW_LOCK_UNLOCKED__), * deny writer. Otherwise if unlocked grant to writer @@ -175,8 +171,6 @@ static inline int arch_write_trylock(arch_rwlock_t *rw) { unsigned int val, got_it = 0; - smp_mb(); - __asm__ __volatile__( "1: llock %[val], [%[rwlock]] \n" " brne %[val], %[UNLOCKED], 4f \n" /* !UNLOCKED, bail */ @@ -217,17 +211,13 @@ static inline void arch_read_unlock(arch_rwlock_t *rw) : [val] "=&r" (val) : [rwlock] "r" (&(rw->counter)) : "memory", "cc"); - - smp_mb(); } static inline void arch_write_unlock(arch_rwlock_t *rw) { smp_mb(); - rw->counter = __ARCH_RW_LOCK_UNLOCKED__; - - smp_mb(); + WRITE_ONCE(rw->counter, __ARCH_RW_LOCK_UNLOCKED__); } #else /* !CONFIG_ARC_HAS_LLSC */ @@ -237,10 +227,9 @@ static inline void arch_spin_lock(arch_spinlock_t *lock) unsigned int val = __ARCH_SPIN_LOCK_LOCKED__; /* - * This smp_mb() is technically superfluous, we only need the one - * after the lock for providing the ACQUIRE semantics. - * However doing the "right" thing was regressing hackbench - * so keeping this, pending further investigation + * Per lkmm, smp_mb() is only required after _lock (and before_unlock) + * for ACQ and REL semantics respectively. However EX based spinlocks + * need the extra smp_mb to workaround a hardware quirk. */ smp_mb(); @@ -257,14 +246,6 @@ static inline void arch_spin_lock(arch_spinlock_t *lock) #endif : "memory"); - /* - * ACQUIRE barrier to ensure load/store after taking the lock - * don't "bleed-up" out of the critical section (leak-in is allowed) - * http://www.spinics.net/lists/kernel/msg2010409.html - * - * ARCv2 only has load-load, store-store and all-all barrier - * thus need the full all-all barrier - */ smp_mb(); } @@ -309,8 +290,7 @@ static inline void arch_spin_unlock(arch_spinlock_t *lock) : "memory"); /* - * superfluous, but keeping for now - see pairing version in - * arch_spin_lock above + * see pairing version/comment in arch_spin_lock above */ smp_mb(); } @@ -344,7 +324,6 @@ static inline int arch_read_trylock(arch_rwlock_t *rw) arch_spin_unlock(&(rw->lock_mutex)); local_irq_restore(flags); - smp_mb(); return ret; } -- 2.7.4