]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/commitdiff
locking/qspinlock: Merge 'struct __qspinlock' into 'struct qspinlock'
authorWill Deacon <will.deacon@arm.com>
Tue, 18 Dec 2018 17:13:54 +0000 (18:13 +0100)
committerKleber Sacilotto de Souza <kleber.souza@canonical.com>
Wed, 14 Aug 2019 09:18:49 +0000 (11:18 +0200)
BugLink: https://bugs.launchpad.net/bugs/1837257
commit 625e88be1f41b53cec55827c984e4a89ea8ee9f9 upstream.

'struct __qspinlock' provides a handy union of fields so that
subcomponents of the lockword can be accessed by name, without having to
manage shifts and masks explicitly and take endianness into account.

This is useful in qspinlock.h and also potentially in arch headers, so
move the 'struct __qspinlock' into 'struct qspinlock' and kill the extra
definition.

Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Acked-by: Boqun Feng <boqun.feng@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-3-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Signed-off-by: Khalid Elmously <khalid.elmously@canonical.com>
arch/x86/include/asm/qspinlock.h
arch/x86/include/asm/qspinlock_paravirt.h
include/asm-generic/qspinlock_types.h
kernel/locking/qspinlock.c
kernel/locking/qspinlock_paravirt.h

index 5e16b5d40d32b7667fb49f8c697f2b2443484151..90b0b0ed8161aa28105fa389171489f3695b2f2a 100644 (file)
@@ -16,7 +16,7 @@
  */
 static inline void native_queued_spin_unlock(struct qspinlock *lock)
 {
-       smp_store_release((u8 *)lock, 0);
+       smp_store_release(&lock->locked, 0);
 }
 
 #ifdef CONFIG_PARAVIRT_SPINLOCKS
index 923307ea11c7128ef585d7056e1af6a755e90961..9ef5ee03d2d79268dc627dd8371653a7e2ec6c33 100644 (file)
@@ -22,8 +22,7 @@ PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath);
  *
  * void __pv_queued_spin_unlock(struct qspinlock *lock)
  * {
- *     struct __qspinlock *l = (void *)lock;
- *     u8 lockval = cmpxchg(&l->locked, _Q_LOCKED_VAL, 0);
+ *     u8 lockval = cmpxchg(&lock->locked, _Q_LOCKED_VAL, 0);
  *
  *     if (likely(lockval == _Q_LOCKED_VAL))
  *             return;
index 034acd0c4956b59932f650e85612a9397236dc66..0763f065b975a543fb0e887d4af8d63bf7354f05 100644 (file)
 #endif
 
 typedef struct qspinlock {
-       atomic_t        val;
+       union {
+               atomic_t val;
+
+               /*
+                * By using the whole 2nd least significant byte for the
+                * pending bit, we can allow better optimization of the lock
+                * acquisition for the pending bit holder.
+                */
+#ifdef __LITTLE_ENDIAN
+               struct {
+                       u8      locked;
+                       u8      pending;
+               };
+               struct {
+                       u16     locked_pending;
+                       u16     tail;
+               };
+#else
+               struct {
+                       u16     tail;
+                       u16     locked_pending;
+               };
+               struct {
+                       u8      reserved[2];
+                       u8      pending;
+                       u8      locked;
+               };
+#endif
+       };
 } arch_spinlock_t;
 
 /*
  * Initializier
  */
-#define        __ARCH_SPIN_LOCK_UNLOCKED       { ATOMIC_INIT(0) }
+#define        __ARCH_SPIN_LOCK_UNLOCKED       { .val = ATOMIC_INIT(0) }
 
 /*
  * Bitfields in the atomic value:
index 18161264227a3fadd6c25c58b3361d3094132ef6..e60e618287b4913758e4f055b8a36c0e801a8161 100644 (file)
@@ -126,40 +126,6 @@ static inline __pure struct mcs_spinlock *decode_tail(u32 tail)
 
 #define _Q_LOCKED_PENDING_MASK (_Q_LOCKED_MASK | _Q_PENDING_MASK)
 
-/*
- * By using the whole 2nd least significant byte for the pending bit, we
- * can allow better optimization of the lock acquisition for the pending
- * bit holder.
- *
- * This internal structure is also used by the set_locked function which
- * is not restricted to _Q_PENDING_BITS == 8.
- */
-struct __qspinlock {
-       union {
-               atomic_t val;
-#ifdef __LITTLE_ENDIAN
-               struct {
-                       u8      locked;
-                       u8      pending;
-               };
-               struct {
-                       u16     locked_pending;
-                       u16     tail;
-               };
-#else
-               struct {
-                       u16     tail;
-                       u16     locked_pending;
-               };
-               struct {
-                       u8      reserved[2];
-                       u8      pending;
-                       u8      locked;
-               };
-#endif
-       };
-};
-
 #if _Q_PENDING_BITS == 8
 /**
  * clear_pending_set_locked - take ownership and clear the pending bit.
@@ -171,9 +137,7 @@ struct __qspinlock {
  */
 static __always_inline void clear_pending_set_locked(struct qspinlock *lock)
 {
-       struct __qspinlock *l = (void *)lock;
-
-       WRITE_ONCE(l->locked_pending, _Q_LOCKED_VAL);
+       WRITE_ONCE(lock->locked_pending, _Q_LOCKED_VAL);
 }
 
 /*
@@ -188,13 +152,11 @@ static __always_inline void clear_pending_set_locked(struct qspinlock *lock)
  */
 static __always_inline u32 xchg_tail(struct qspinlock *lock, u32 tail)
 {
-       struct __qspinlock *l = (void *)lock;
-
        /*
         * Use release semantics to make sure that the MCS node is properly
         * initialized before changing the tail code.
         */
-       return (u32)xchg_release(&l->tail,
+       return (u32)xchg_release(&lock->tail,
                                 tail >> _Q_TAIL_OFFSET) << _Q_TAIL_OFFSET;
 }
 
@@ -249,9 +211,7 @@ static __always_inline u32 xchg_tail(struct qspinlock *lock, u32 tail)
  */
 static __always_inline void set_locked(struct qspinlock *lock)
 {
-       struct __qspinlock *l = (void *)lock;
-
-       WRITE_ONCE(l->locked, _Q_LOCKED_VAL);
+       WRITE_ONCE(lock->locked, _Q_LOCKED_VAL);
 }
 
 
index 6ee477765e6c5cf8d0c2ed00dbd9fa20cdb4a1b4..2711940429f5085dc19729e77f828e0afa35378b 100644 (file)
@@ -87,8 +87,6 @@ struct pv_node {
 #define queued_spin_trylock(l) pv_hybrid_queued_unfair_trylock(l)
 static inline bool pv_hybrid_queued_unfair_trylock(struct qspinlock *lock)
 {
-       struct __qspinlock *l = (void *)lock;
-
        /*
         * Stay in unfair lock mode as long as queued mode waiters are
         * present in the MCS wait queue but the pending bit isn't set.
@@ -97,7 +95,7 @@ static inline bool pv_hybrid_queued_unfair_trylock(struct qspinlock *lock)
                int val = atomic_read(&lock->val);
 
                if (!(val & _Q_LOCKED_PENDING_MASK) &&
-                  (cmpxchg_acquire(&l->locked, 0, _Q_LOCKED_VAL) == 0)) {
+                  (cmpxchg_acquire(&lock->locked, 0, _Q_LOCKED_VAL) == 0)) {
                        qstat_inc(qstat_pv_lock_stealing, true);
                        return true;
                }
@@ -117,16 +115,12 @@ static inline bool pv_hybrid_queued_unfair_trylock(struct qspinlock *lock)
 #if _Q_PENDING_BITS == 8
 static __always_inline void set_pending(struct qspinlock *lock)
 {
-       struct __qspinlock *l = (void *)lock;
-
-       WRITE_ONCE(l->pending, 1);
+       WRITE_ONCE(lock->pending, 1);
 }
 
 static __always_inline void clear_pending(struct qspinlock *lock)
 {
-       struct __qspinlock *l = (void *)lock;
-
-       WRITE_ONCE(l->pending, 0);
+       WRITE_ONCE(lock->pending, 0);
 }
 
 /*
@@ -136,10 +130,8 @@ static __always_inline void clear_pending(struct qspinlock *lock)
  */
 static __always_inline int trylock_clear_pending(struct qspinlock *lock)
 {
-       struct __qspinlock *l = (void *)lock;
-
-       return !READ_ONCE(l->locked) &&
-              (cmpxchg_acquire(&l->locked_pending, _Q_PENDING_VAL,
+       return !READ_ONCE(lock->locked) &&
+              (cmpxchg_acquire(&lock->locked_pending, _Q_PENDING_VAL,
                                _Q_LOCKED_VAL) == _Q_PENDING_VAL);
 }
 #else /* _Q_PENDING_BITS == 8 */
@@ -384,7 +376,6 @@ static void pv_wait_node(struct mcs_spinlock *node, struct mcs_spinlock *prev)
 static void pv_kick_node(struct qspinlock *lock, struct mcs_spinlock *node)
 {
        struct pv_node *pn = (struct pv_node *)node;
-       struct __qspinlock *l = (void *)lock;
 
        /*
         * If the vCPU is indeed halted, advance its state to match that of
@@ -413,7 +404,7 @@ static void pv_kick_node(struct qspinlock *lock, struct mcs_spinlock *node)
         * the hash table later on at unlock time, no atomic instruction is
         * needed.
         */
-       WRITE_ONCE(l->locked, _Q_SLOW_VAL);
+       WRITE_ONCE(lock->locked, _Q_SLOW_VAL);
        (void)pv_hash(lock, pn);
 }
 
@@ -428,7 +419,6 @@ static u32
 pv_wait_head_or_lock(struct qspinlock *lock, struct mcs_spinlock *node)
 {
        struct pv_node *pn = (struct pv_node *)node;
-       struct __qspinlock *l = (void *)lock;
        struct qspinlock **lp = NULL;
        int waitcnt = 0;
        int loop;
@@ -479,13 +469,13 @@ pv_wait_head_or_lock(struct qspinlock *lock, struct mcs_spinlock *node)
                         *
                         * Matches the smp_rmb() in __pv_queued_spin_unlock().
                         */
-                       if (xchg(&l->locked, _Q_SLOW_VAL) == 0) {
+                       if (xchg(&lock->locked, _Q_SLOW_VAL) == 0) {
                                /*
                                 * The lock was free and now we own the lock.
                                 * Change the lock value back to _Q_LOCKED_VAL
                                 * and unhash the table.
                                 */
-                               WRITE_ONCE(l->locked, _Q_LOCKED_VAL);
+                               WRITE_ONCE(lock->locked, _Q_LOCKED_VAL);
                                WRITE_ONCE(*lp, NULL);
                                goto gotlock;
                        }
@@ -493,7 +483,7 @@ pv_wait_head_or_lock(struct qspinlock *lock, struct mcs_spinlock *node)
                WRITE_ONCE(pn->state, vcpu_hashed);
                qstat_inc(qstat_pv_wait_head, true);
                qstat_inc(qstat_pv_wait_again, waitcnt);
-               pv_wait(&l->locked, _Q_SLOW_VAL);
+               pv_wait(&lock->locked, _Q_SLOW_VAL);
 
                /*
                 * Because of lock stealing, the queue head vCPU may not be
@@ -518,7 +508,6 @@ gotlock:
 __visible void
 __pv_queued_spin_unlock_slowpath(struct qspinlock *lock, u8 locked)
 {
-       struct __qspinlock *l = (void *)lock;
        struct pv_node *node;
 
        if (unlikely(locked != _Q_SLOW_VAL)) {
@@ -547,7 +536,7 @@ __pv_queued_spin_unlock_slowpath(struct qspinlock *lock, u8 locked)
         * Now that we have a reference to the (likely) blocked pv_node,
         * release the lock.
         */
-       smp_store_release(&l->locked, 0);
+       smp_store_release(&lock->locked, 0);
 
        /*
         * At this point the memory pointed at by lock can be freed/reused,
@@ -573,7 +562,6 @@ __pv_queued_spin_unlock_slowpath(struct qspinlock *lock, u8 locked)
 #ifndef __pv_queued_spin_unlock
 __visible void __pv_queued_spin_unlock(struct qspinlock *lock)
 {
-       struct __qspinlock *l = (void *)lock;
        u8 locked;
 
        /*
@@ -581,7 +569,7 @@ __visible void __pv_queued_spin_unlock(struct qspinlock *lock)
         * unhash. Otherwise it would be possible to have multiple @lock
         * entries, which would be BAD.
         */
-       locked = cmpxchg_release(&l->locked, _Q_LOCKED_VAL, 0);
+       locked = cmpxchg_release(&lock->locked, _Q_LOCKED_VAL, 0);
        if (likely(locked == _Q_LOCKED_VAL))
                return;