powerpc: Move spinlock implementation to simple_spinlock
authorNicholas Piggin <npiggin@gmail.com>
Fri, 24 Jul 2020 13:14:19 +0000 (23:14 +1000)
committerMichael Ellerman <mpe@ellerman.id.au>
Sun, 26 Jul 2020 13:34:26 +0000 (23:34 +1000)
To prepare for queued spinlocks. This is a simple rename except to
update preprocessor guard name and a file reference.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Acked-by: Waiman Long <longman@redhat.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200724131423.1362108-3-npiggin@gmail.com
arch/powerpc/include/asm/simple_spinlock.h [new file with mode: 0644]
arch/powerpc/include/asm/simple_spinlock_types.h [new file with mode: 0644]
arch/powerpc/include/asm/spinlock.h
arch/powerpc/include/asm/spinlock_types.h

diff --git a/arch/powerpc/include/asm/simple_spinlock.h b/arch/powerpc/include/asm/simple_spinlock.h
new file mode 100644 (file)
index 0000000..9c3c305
--- /dev/null
@@ -0,0 +1,288 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef _ASM_POWERPC_SIMPLE_SPINLOCK_H
+#define _ASM_POWERPC_SIMPLE_SPINLOCK_H
+
+/*
+ * Simple spin lock operations.
+ *
+ * Copyright (C) 2001-2004 Paul Mackerras <paulus@au.ibm.com>, IBM
+ * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
+ * Copyright (C) 2002 Dave Engebretsen <engebret@us.ibm.com>, IBM
+ *     Rework to support virtual processors
+ *
+ * Type of int is used as a full 64b word is not necessary.
+ *
+ * (the type definitions are in asm/simple_spinlock_types.h)
+ */
+#include <linux/irqflags.h>
+#include <asm/paravirt.h>
+#include <asm/paca.h>
+#include <asm/synch.h>
+#include <asm/ppc-opcode.h>
+
+#ifdef CONFIG_PPC64
+/* use 0x800000yy when locked, where yy == CPU number */
+#ifdef __BIG_ENDIAN__
+#define LOCK_TOKEN     (*(u32 *)(&get_paca()->lock_token))
+#else
+#define LOCK_TOKEN     (*(u32 *)(&get_paca()->paca_index))
+#endif
+#else
+#define LOCK_TOKEN     1
+#endif
+
+static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
+{
+       return lock.slock == 0;
+}
+
+static inline int arch_spin_is_locked(arch_spinlock_t *lock)
+{
+       smp_mb();
+       return !arch_spin_value_unlocked(*lock);
+}
+
+/*
+ * This returns the old value in the lock, so we succeeded
+ * in getting the lock if the return value is 0.
+ */
+static inline unsigned long __arch_spin_trylock(arch_spinlock_t *lock)
+{
+       unsigned long tmp, token;
+
+       token = LOCK_TOKEN;
+       __asm__ __volatile__(
+"1:    " PPC_LWARX(%0,0,%2,1) "\n\
+       cmpwi           0,%0,0\n\
+       bne-            2f\n\
+       stwcx.          %1,0,%2\n\
+       bne-            1b\n"
+       PPC_ACQUIRE_BARRIER
+"2:"
+       : "=&r" (tmp)
+       : "r" (token), "r" (&lock->slock)
+       : "cr0", "memory");
+
+       return tmp;
+}
+
+static inline int arch_spin_trylock(arch_spinlock_t *lock)
+{
+       return __arch_spin_trylock(lock) == 0;
+}
+
+/*
+ * On a system with shared processors (that is, where a physical
+ * processor is multiplexed between several virtual processors),
+ * there is no point spinning on a lock if the holder of the lock
+ * isn't currently scheduled on a physical processor.  Instead
+ * we detect this situation and ask the hypervisor to give the
+ * rest of our timeslice to the lock holder.
+ *
+ * So that we can tell which virtual processor is holding a lock,
+ * we put 0x80000000 | smp_processor_id() in the lock when it is
+ * held.  Conveniently, we have a word in the paca that holds this
+ * value.
+ */
+
+#if defined(CONFIG_PPC_SPLPAR)
+/* We only yield to the hypervisor if we are in shared processor mode */
+void splpar_spin_yield(arch_spinlock_t *lock);
+void splpar_rw_yield(arch_rwlock_t *lock);
+#else /* SPLPAR */
+static inline void splpar_spin_yield(arch_spinlock_t *lock) {};
+static inline void splpar_rw_yield(arch_rwlock_t *lock) {};
+#endif
+
+static inline void spin_yield(arch_spinlock_t *lock)
+{
+       if (is_shared_processor())
+               splpar_spin_yield(lock);
+       else
+               barrier();
+}
+
+static inline void rw_yield(arch_rwlock_t *lock)
+{
+       if (is_shared_processor())
+               splpar_rw_yield(lock);
+       else
+               barrier();
+}
+
+static inline void arch_spin_lock(arch_spinlock_t *lock)
+{
+       while (1) {
+               if (likely(__arch_spin_trylock(lock) == 0))
+                       break;
+               do {
+                       HMT_low();
+                       if (is_shared_processor())
+                               splpar_spin_yield(lock);
+               } while (unlikely(lock->slock != 0));
+               HMT_medium();
+       }
+}
+
+static inline
+void arch_spin_lock_flags(arch_spinlock_t *lock, unsigned long flags)
+{
+       unsigned long flags_dis;
+
+       while (1) {
+               if (likely(__arch_spin_trylock(lock) == 0))
+                       break;
+               local_save_flags(flags_dis);
+               local_irq_restore(flags);
+               do {
+                       HMT_low();
+                       if (is_shared_processor())
+                               splpar_spin_yield(lock);
+               } while (unlikely(lock->slock != 0));
+               HMT_medium();
+               local_irq_restore(flags_dis);
+       }
+}
+#define arch_spin_lock_flags arch_spin_lock_flags
+
+static inline void arch_spin_unlock(arch_spinlock_t *lock)
+{
+       __asm__ __volatile__("# arch_spin_unlock\n\t"
+                               PPC_RELEASE_BARRIER: : :"memory");
+       lock->slock = 0;
+}
+
+/*
+ * Read-write spinlocks, allowing multiple readers
+ * but only one writer.
+ *
+ * NOTE! it is quite common to have readers in interrupts
+ * but no interrupt writers. For those circumstances we
+ * can "mix" irq-safe locks - any writer needs to get a
+ * irq-safe write-lock, but readers can get non-irqsafe
+ * read-locks.
+ */
+
+#ifdef CONFIG_PPC64
+#define __DO_SIGN_EXTEND       "extsw  %0,%0\n"
+#define WRLOCK_TOKEN           LOCK_TOKEN      /* it's negative */
+#else
+#define __DO_SIGN_EXTEND
+#define WRLOCK_TOKEN           (-1)
+#endif
+
+/*
+ * This returns the old value in the lock + 1,
+ * so we got a read lock if the return value is > 0.
+ */
+static inline long __arch_read_trylock(arch_rwlock_t *rw)
+{
+       long tmp;
+
+       __asm__ __volatile__(
+"1:    " PPC_LWARX(%0,0,%1,1) "\n"
+       __DO_SIGN_EXTEND
+"      addic.          %0,%0,1\n\
+       ble-            2f\n"
+"      stwcx.          %0,0,%1\n\
+       bne-            1b\n"
+       PPC_ACQUIRE_BARRIER
+"2:"   : "=&r" (tmp)
+       : "r" (&rw->lock)
+       : "cr0", "xer", "memory");
+
+       return tmp;
+}
+
+/*
+ * This returns the old value in the lock,
+ * so we got the write lock if the return value is 0.
+ */
+static inline long __arch_write_trylock(arch_rwlock_t *rw)
+{
+       long tmp, token;
+
+       token = WRLOCK_TOKEN;
+       __asm__ __volatile__(
+"1:    " PPC_LWARX(%0,0,%2,1) "\n\
+       cmpwi           0,%0,0\n\
+       bne-            2f\n"
+"      stwcx.          %1,0,%2\n\
+       bne-            1b\n"
+       PPC_ACQUIRE_BARRIER
+"2:"   : "=&r" (tmp)
+       : "r" (token), "r" (&rw->lock)
+       : "cr0", "memory");
+
+       return tmp;
+}
+
+static inline void arch_read_lock(arch_rwlock_t *rw)
+{
+       while (1) {
+               if (likely(__arch_read_trylock(rw) > 0))
+                       break;
+               do {
+                       HMT_low();
+                       if (is_shared_processor())
+                               splpar_rw_yield(rw);
+               } while (unlikely(rw->lock < 0));
+               HMT_medium();
+       }
+}
+
+static inline void arch_write_lock(arch_rwlock_t *rw)
+{
+       while (1) {
+               if (likely(__arch_write_trylock(rw) == 0))
+                       break;
+               do {
+                       HMT_low();
+                       if (is_shared_processor())
+                               splpar_rw_yield(rw);
+               } while (unlikely(rw->lock != 0));
+               HMT_medium();
+       }
+}
+
+static inline int arch_read_trylock(arch_rwlock_t *rw)
+{
+       return __arch_read_trylock(rw) > 0;
+}
+
+static inline int arch_write_trylock(arch_rwlock_t *rw)
+{
+       return __arch_write_trylock(rw) == 0;
+}
+
+static inline void arch_read_unlock(arch_rwlock_t *rw)
+{
+       long tmp;
+
+       __asm__ __volatile__(
+       "# read_unlock\n\t"
+       PPC_RELEASE_BARRIER
+"1:    lwarx           %0,0,%1\n\
+       addic           %0,%0,-1\n"
+"      stwcx.          %0,0,%1\n\
+       bne-            1b"
+       : "=&r"(tmp)
+       : "r"(&rw->lock)
+       : "cr0", "xer", "memory");
+}
+
+static inline void arch_write_unlock(arch_rwlock_t *rw)
+{
+       __asm__ __volatile__("# write_unlock\n\t"
+                               PPC_RELEASE_BARRIER: : :"memory");
+       rw->lock = 0;
+}
+
+#define arch_spin_relax(lock)  spin_yield(lock)
+#define arch_read_relax(lock)  rw_yield(lock)
+#define arch_write_relax(lock) rw_yield(lock)
+
+/* See include/linux/spinlock.h */
+#define smp_mb__after_spinlock()   smp_mb()
+
+#endif /* _ASM_POWERPC_SIMPLE_SPINLOCK_H */
diff --git a/arch/powerpc/include/asm/simple_spinlock_types.h b/arch/powerpc/include/asm/simple_spinlock_types.h
new file mode 100644 (file)
index 0000000..0f3cdd8
--- /dev/null
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_POWERPC_SIMPLE_SPINLOCK_TYPES_H
+#define _ASM_POWERPC_SIMPLE_SPINLOCK_TYPES_H
+
+#ifndef __LINUX_SPINLOCK_TYPES_H
+# error "please don't include this file directly"
+#endif
+
+typedef struct {
+       volatile unsigned int slock;
+} arch_spinlock_t;
+
+#define __ARCH_SPIN_LOCK_UNLOCKED      { 0 }
+
+typedef struct {
+       volatile signed int lock;
+} arch_rwlock_t;
+
+#define __ARCH_RW_LOCK_UNLOCKED                { 0 }
+
+#endif /* _ASM_POWERPC_SIMPLE_SPINLOCK_TYPES_H */
index 79be9bb..21357fe 100644 (file)
@@ -3,290 +3,7 @@
 #define __ASM_SPINLOCK_H
 #ifdef __KERNEL__
 
-/*
- * Simple spin lock operations.  
- *
- * Copyright (C) 2001-2004 Paul Mackerras <paulus@au.ibm.com>, IBM
- * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
- * Copyright (C) 2002 Dave Engebretsen <engebret@us.ibm.com>, IBM
- *     Rework to support virtual processors
- *
- * Type of int is used as a full 64b word is not necessary.
- *
- * (the type definitions are in asm/spinlock_types.h)
- */
-#include <linux/irqflags.h>
-#include <asm/paravirt.h>
-#ifdef CONFIG_PPC64
-#include <asm/paca.h>
-#endif
-#include <asm/synch.h>
-#include <asm/ppc-opcode.h>
-
-#ifdef CONFIG_PPC64
-/* use 0x800000yy when locked, where yy == CPU number */
-#ifdef __BIG_ENDIAN__
-#define LOCK_TOKEN     (*(u32 *)(&get_paca()->lock_token))
-#else
-#define LOCK_TOKEN     (*(u32 *)(&get_paca()->paca_index))
-#endif
-#else
-#define LOCK_TOKEN     1
-#endif
-
-static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
-{
-       return lock.slock == 0;
-}
-
-static inline int arch_spin_is_locked(arch_spinlock_t *lock)
-{
-       smp_mb();
-       return !arch_spin_value_unlocked(*lock);
-}
-
-/*
- * This returns the old value in the lock, so we succeeded
- * in getting the lock if the return value is 0.
- */
-static inline unsigned long __arch_spin_trylock(arch_spinlock_t *lock)
-{
-       unsigned long tmp, token;
-
-       token = LOCK_TOKEN;
-       __asm__ __volatile__(
-"1:    " PPC_LWARX(%0,0,%2,1) "\n\
-       cmpwi           0,%0,0\n\
-       bne-            2f\n\
-       stwcx.          %1,0,%2\n\
-       bne-            1b\n"
-       PPC_ACQUIRE_BARRIER
-"2:"
-       : "=&r" (tmp)
-       : "r" (token), "r" (&lock->slock)
-       : "cr0", "memory");
-
-       return tmp;
-}
-
-static inline int arch_spin_trylock(arch_spinlock_t *lock)
-{
-       return __arch_spin_trylock(lock) == 0;
-}
-
-/*
- * On a system with shared processors (that is, where a physical
- * processor is multiplexed between several virtual processors),
- * there is no point spinning on a lock if the holder of the lock
- * isn't currently scheduled on a physical processor.  Instead
- * we detect this situation and ask the hypervisor to give the
- * rest of our timeslice to the lock holder.
- *
- * So that we can tell which virtual processor is holding a lock,
- * we put 0x80000000 | smp_processor_id() in the lock when it is
- * held.  Conveniently, we have a word in the paca that holds this
- * value.
- */
-
-#if defined(CONFIG_PPC_SPLPAR)
-/* We only yield to the hypervisor if we are in shared processor mode */
-void splpar_spin_yield(arch_spinlock_t *lock);
-void splpar_rw_yield(arch_rwlock_t *lock);
-#else /* SPLPAR */
-static inline void splpar_spin_yield(arch_spinlock_t *lock) {};
-static inline void splpar_rw_yield(arch_rwlock_t *lock) {};
-#endif
-
-static inline void spin_yield(arch_spinlock_t *lock)
-{
-       if (is_shared_processor())
-               splpar_spin_yield(lock);
-       else
-               barrier();
-}
-
-static inline void rw_yield(arch_rwlock_t *lock)
-{
-       if (is_shared_processor())
-               splpar_rw_yield(lock);
-       else
-               barrier();
-}
-
-static inline void arch_spin_lock(arch_spinlock_t *lock)
-{
-       while (1) {
-               if (likely(__arch_spin_trylock(lock) == 0))
-                       break;
-               do {
-                       HMT_low();
-                       if (is_shared_processor())
-                               splpar_spin_yield(lock);
-               } while (unlikely(lock->slock != 0));
-               HMT_medium();
-       }
-}
-
-static inline
-void arch_spin_lock_flags(arch_spinlock_t *lock, unsigned long flags)
-{
-       unsigned long flags_dis;
-
-       while (1) {
-               if (likely(__arch_spin_trylock(lock) == 0))
-                       break;
-               local_save_flags(flags_dis);
-               local_irq_restore(flags);
-               do {
-                       HMT_low();
-                       if (is_shared_processor())
-                               splpar_spin_yield(lock);
-               } while (unlikely(lock->slock != 0));
-               HMT_medium();
-               local_irq_restore(flags_dis);
-       }
-}
-#define arch_spin_lock_flags arch_spin_lock_flags
-
-static inline void arch_spin_unlock(arch_spinlock_t *lock)
-{
-       __asm__ __volatile__("# arch_spin_unlock\n\t"
-                               PPC_RELEASE_BARRIER: : :"memory");
-       lock->slock = 0;
-}
-
-/*
- * Read-write spinlocks, allowing multiple readers
- * but only one writer.
- *
- * NOTE! it is quite common to have readers in interrupts
- * but no interrupt writers. For those circumstances we
- * can "mix" irq-safe locks - any writer needs to get a
- * irq-safe write-lock, but readers can get non-irqsafe
- * read-locks.
- */
-
-#ifdef CONFIG_PPC64
-#define __DO_SIGN_EXTEND       "extsw  %0,%0\n"
-#define WRLOCK_TOKEN           LOCK_TOKEN      /* it's negative */
-#else
-#define __DO_SIGN_EXTEND
-#define WRLOCK_TOKEN           (-1)
-#endif
-
-/*
- * This returns the old value in the lock + 1,
- * so we got a read lock if the return value is > 0.
- */
-static inline long __arch_read_trylock(arch_rwlock_t *rw)
-{
-       long tmp;
-
-       __asm__ __volatile__(
-"1:    " PPC_LWARX(%0,0,%1,1) "\n"
-       __DO_SIGN_EXTEND
-"      addic.          %0,%0,1\n\
-       ble-            2f\n"
-"      stwcx.          %0,0,%1\n\
-       bne-            1b\n"
-       PPC_ACQUIRE_BARRIER
-"2:"   : "=&r" (tmp)
-       : "r" (&rw->lock)
-       : "cr0", "xer", "memory");
-
-       return tmp;
-}
-
-/*
- * This returns the old value in the lock,
- * so we got the write lock if the return value is 0.
- */
-static inline long __arch_write_trylock(arch_rwlock_t *rw)
-{
-       long tmp, token;
-
-       token = WRLOCK_TOKEN;
-       __asm__ __volatile__(
-"1:    " PPC_LWARX(%0,0,%2,1) "\n\
-       cmpwi           0,%0,0\n\
-       bne-            2f\n"
-"      stwcx.          %1,0,%2\n\
-       bne-            1b\n"
-       PPC_ACQUIRE_BARRIER
-"2:"   : "=&r" (tmp)
-       : "r" (token), "r" (&rw->lock)
-       : "cr0", "memory");
-
-       return tmp;
-}
-
-static inline void arch_read_lock(arch_rwlock_t *rw)
-{
-       while (1) {
-               if (likely(__arch_read_trylock(rw) > 0))
-                       break;
-               do {
-                       HMT_low();
-                       if (is_shared_processor())
-                               splpar_rw_yield(rw);
-               } while (unlikely(rw->lock < 0));
-               HMT_medium();
-       }
-}
-
-static inline void arch_write_lock(arch_rwlock_t *rw)
-{
-       while (1) {
-               if (likely(__arch_write_trylock(rw) == 0))
-                       break;
-               do {
-                       HMT_low();
-                       if (is_shared_processor())
-                               splpar_rw_yield(rw);
-               } while (unlikely(rw->lock != 0));
-               HMT_medium();
-       }
-}
-
-static inline int arch_read_trylock(arch_rwlock_t *rw)
-{
-       return __arch_read_trylock(rw) > 0;
-}
-
-static inline int arch_write_trylock(arch_rwlock_t *rw)
-{
-       return __arch_write_trylock(rw) == 0;
-}
-
-static inline void arch_read_unlock(arch_rwlock_t *rw)
-{
-       long tmp;
-
-       __asm__ __volatile__(
-       "# read_unlock\n\t"
-       PPC_RELEASE_BARRIER
-"1:    lwarx           %0,0,%1\n\
-       addic           %0,%0,-1\n"
-"      stwcx.          %0,0,%1\n\
-       bne-            1b"
-       : "=&r"(tmp)
-       : "r"(&rw->lock)
-       : "cr0", "xer", "memory");
-}
-
-static inline void arch_write_unlock(arch_rwlock_t *rw)
-{
-       __asm__ __volatile__("# write_unlock\n\t"
-                               PPC_RELEASE_BARRIER: : :"memory");
-       rw->lock = 0;
-}
-
-#define arch_spin_relax(lock)  spin_yield(lock)
-#define arch_read_relax(lock)  rw_yield(lock)
-#define arch_write_relax(lock) rw_yield(lock)
-
-/* See include/linux/spinlock.h */
-#define smp_mb__after_spinlock()   smp_mb()
+#include <asm/simple_spinlock.h>
 
 #endif /* __KERNEL__ */
 #endif /* __ASM_SPINLOCK_H */
index 87adaf1..3906f52 100644 (file)
@@ -6,16 +6,6 @@
 # error "please don't include this file directly"
 #endif
 
-typedef struct {
-       volatile unsigned int slock;
-} arch_spinlock_t;
-
-#define __ARCH_SPIN_LOCK_UNLOCKED      { 0 }
-
-typedef struct {
-       volatile signed int lock;
-} arch_rwlock_t;
-
-#define __ARCH_RW_LOCK_UNLOCKED                { 0 }
+#include <asm/simple_spinlock_types.h>
 
 #endif