[IA64] implement ticket locks for Itanium

Back in January 2008 Nick Piggin implemented "ticket" spinlocks
for X86 (See commit 314cdbef).

IA64 implementation has a couple of differences because of the
available atomic operations ... e.g. we have no fetchadd2 instruction
that operates on a 16-bit quantity so we make ticket locks use
a 32-bit word for each of the current ticket and now-serving values.

Performance on uncontended locks is about 8% worse than the previous
implementation, but this seems a good trade for determinism in the
contended case. Performance impact on macro-level benchmarks is in
the noise.
Signed-off-by: Tony Luck <tony.luck@intel.com>

arch/ia64/Kconfig

@@ -60,9 +60,7 @@ config IOMMU_HELPER
        bool
 
 config GENERIC_LOCKBREAK
-	bool
-	default y
-	depends on SMP && PREEMPT
+	def_bool n
 
 config RWSEM_XCHGADD_ALGORITHM
 	bool

arch/ia64/include/asm/spinlock.h

@@ -19,103 +19,106 @@
 
 #define __raw_spin_lock_init(x)			((x)->lock = 0)
 
-#ifdef ASM_SUPPORTED
 /*
- * Try to get the lock.  If we fail to get the lock, make a non-standard call to
- * ia64_spinlock_contention().  We do not use a normal call because that would force all
- * callers of __raw_spin_lock() to be non-leaf routines.  Instead, ia64_spinlock_contention() is
- * carefully coded to touch only those registers that __raw_spin_lock() marks "clobbered".
+ * Ticket locks are conceptually two parts, one indicating the current head of
+ * the queue, and the other indicating the current tail. The lock is acquired
+ * by atomically noting the tail and incrementing it by one (thus adding
+ * ourself to the queue and noting our position), then waiting until the head
+ * becomes equal to the the initial value of the tail.
+ *
+ *   63                     32  31                      0
+ *  +----------------------------------------------------+
+ *  |  next_ticket_number      |     now_serving         |
+ *  +----------------------------------------------------+
  */
 
-#define IA64_SPINLOCK_CLOBBERS "ar.ccv", "ar.pfs", "p14", "p15", "r27", "r28", "r29", "r30", "b6", "memory"
+#define TICKET_SHIFT	32
 
-static inline void
-__raw_spin_lock_flags (raw_spinlock_t *lock, unsigned long flags)
+static __always_inline void __ticket_spin_lock(raw_spinlock_t *lock)
 {
-	register volatile unsigned int *ptr asm ("r31") = &lock->lock;
-
-#if (__GNUC__ == 3 && __GNUC_MINOR__ < 3)
-# ifdef CONFIG_ITANIUM
-	/* don't use brl on Itanium... */
-	asm volatile ("{\n\t"
-		      "  mov ar.ccv = r0\n\t"
-		      "  mov r28 = ip\n\t"
-		      "  mov r30 = 1;;\n\t"
-		      "}\n\t"
-		      "cmpxchg4.acq r30 = [%1], r30, ar.ccv\n\t"
-		      "movl r29 = ia64_spinlock_contention_pre3_4;;\n\t"
-		      "cmp4.ne p14, p0 = r30, r0\n\t"
-		      "mov b6 = r29;;\n\t"
-		      "mov r27=%2\n\t"
-		      "(p14) br.cond.spnt.many b6"
-		      : "=r"(ptr) : "r"(ptr), "r" (flags) : IA64_SPINLOCK_CLOBBERS);
-# else
-	asm volatile ("{\n\t"
-		      "  mov ar.ccv = r0\n\t"
-		      "  mov r28 = ip\n\t"
-		      "  mov r30 = 1;;\n\t"
-		      "}\n\t"
-		      "cmpxchg4.acq r30 = [%1], r30, ar.ccv;;\n\t"
-		      "cmp4.ne p14, p0 = r30, r0\n\t"
-		      "mov r27=%2\n\t"
-		      "(p14) brl.cond.spnt.many ia64_spinlock_contention_pre3_4;;"
-		      : "=r"(ptr) : "r"(ptr), "r" (flags) : IA64_SPINLOCK_CLOBBERS);
-# endif /* CONFIG_MCKINLEY */
-#else
-# ifdef CONFIG_ITANIUM
-	/* don't use brl on Itanium... */
-	/* mis-declare, so we get the entry-point, not it's function descriptor: */
-	asm volatile ("mov r30 = 1\n\t"
-		      "mov r27=%2\n\t"
-		      "mov ar.ccv = r0;;\n\t"
-		      "cmpxchg4.acq r30 = [%0], r30, ar.ccv\n\t"
-		      "movl r29 = ia64_spinlock_contention;;\n\t"
-		      "cmp4.ne p14, p0 = r30, r0\n\t"
-		      "mov b6 = r29;;\n\t"
-		      "(p14) br.call.spnt.many b6 = b6"
-		      : "=r"(ptr) : "r"(ptr), "r" (flags) : IA64_SPINLOCK_CLOBBERS);
-# else
-	asm volatile ("mov r30 = 1\n\t"
-		      "mov r27=%2\n\t"
-		      "mov ar.ccv = r0;;\n\t"
-		      "cmpxchg4.acq r30 = [%0], r30, ar.ccv;;\n\t"
-		      "cmp4.ne p14, p0 = r30, r0\n\t"
-		      "(p14) brl.call.spnt.many b6=ia64_spinlock_contention;;"
-		      : "=r"(ptr) : "r"(ptr), "r" (flags) : IA64_SPINLOCK_CLOBBERS);
-# endif /* CONFIG_MCKINLEY */
-#endif
+	int	*p = (int *)&lock->lock, turn, now_serving;
+
+	now_serving = *p;
+	turn = ia64_fetchadd(1, p+1, acq);
+
+	if (turn == now_serving)
+		return;
+
+	do {
+		cpu_relax();
+	} while (ACCESS_ONCE(*p) != turn);
 }
 
-#define __raw_spin_lock(lock) __raw_spin_lock_flags(lock, 0)
+static __always_inline int __ticket_spin_trylock(raw_spinlock_t *lock)
+{
+	long tmp = ACCESS_ONCE(lock->lock), try;
 
-/* Unlock by doing an ordered store and releasing the cacheline with nta */
-static inline void __raw_spin_unlock(raw_spinlock_t *x) {
-	barrier();
-	asm volatile ("st4.rel.nta [%0] = r0\n\t" :: "r"(x));
+	if (!(((tmp >> TICKET_SHIFT) ^ tmp) & ((1L << TICKET_SHIFT) - 1))) {
+		try = tmp + (1L << TICKET_SHIFT);
+
+		return ia64_cmpxchg(acq, &lock->lock, tmp, try, sizeof (tmp)) == tmp;
+	}
+	return 0;
 }
 
-#else /* !ASM_SUPPORTED */
-#define __raw_spin_lock_flags(lock, flags) __raw_spin_lock(lock)
-# define __raw_spin_lock(x)								\
-do {											\
-	__u32 *ia64_spinlock_ptr = (__u32 *) (x);					\
-	__u64 ia64_spinlock_val;							\
-	ia64_spinlock_val = ia64_cmpxchg4_acq(ia64_spinlock_ptr, 1, 0);			\
-	if (unlikely(ia64_spinlock_val)) {						\
-		do {									\
-			while (*ia64_spinlock_ptr)					\
-				ia64_barrier();						\
-			ia64_spinlock_val = ia64_cmpxchg4_acq(ia64_spinlock_ptr, 1, 0);	\
-		} while (ia64_spinlock_val);						\
-	}										\
-} while (0)
-#define __raw_spin_unlock(x)	do { barrier(); ((raw_spinlock_t *) x)->lock = 0; } while (0)
-#endif /* !ASM_SUPPORTED */
+static __always_inline void __ticket_spin_unlock(raw_spinlock_t *lock)
+{
+	int	*p = (int *)&lock->lock;
+
+	(void)ia64_fetchadd(1, p, rel);
+}
+
+static inline int __ticket_spin_is_locked(raw_spinlock_t *lock)
+{
+	long tmp = ACCESS_ONCE(lock->lock);
+
+	return !!(((tmp >> TICKET_SHIFT) ^ tmp) & ((1L << TICKET_SHIFT) - 1));
+}
+
+static inline int __ticket_spin_is_contended(raw_spinlock_t *lock)
+{
+	long tmp = ACCESS_ONCE(lock->lock);
 
-#define __raw_spin_is_locked(x)		((x)->lock != 0)
-#define __raw_spin_trylock(x)		(cmpxchg_acq(&(x)->lock, 0, 1) == 0)
-#define __raw_spin_unlock_wait(lock) \
-	do { while (__raw_spin_is_locked(lock)) cpu_relax(); } while (0)
+	return (((tmp >> TICKET_SHIFT) - tmp) & ((1L << TICKET_SHIFT) - 1)) > 1;
+}
+
+static inline int __raw_spin_is_locked(raw_spinlock_t *lock)
+{
+	return __ticket_spin_is_locked(lock);
+}
+
+static inline int __raw_spin_is_contended(raw_spinlock_t *lock)
+{
+	return __ticket_spin_is_contended(lock);
+}
+#define __raw_spin_is_contended	__raw_spin_is_contended
+
+static __always_inline void __raw_spin_lock(raw_spinlock_t *lock)
+{
+	__ticket_spin_lock(lock);
+}
+
+static __always_inline int __raw_spin_trylock(raw_spinlock_t *lock)
+{
+	return __ticket_spin_trylock(lock);
+}
+
+static __always_inline void __raw_spin_unlock(raw_spinlock_t *lock)
+{
+	__ticket_spin_unlock(lock);
+}
+
+static __always_inline void __raw_spin_lock_flags(raw_spinlock_t *lock,
+						  unsigned long flags)
+{
+	__raw_spin_lock(lock);
+}
+
+static inline void __raw_spin_unlock_wait(raw_spinlock_t *lock)
+{
+	while (__raw_spin_is_locked(lock))
+		cpu_relax();
+}
 
 #define __raw_read_can_lock(rw)		(*(volatile int *)(rw) >= 0)
 #define __raw_write_can_lock(rw)	(*(volatile int *)(rw) == 0)

arch/ia64/include/asm/spinlock_types.h

@@ -6,7 +6,7 @@
 #endif
 
 typedef struct {
-	volatile unsigned int lock;
+	volatile unsigned long lock;
 } raw_spinlock_t;
 
 #define __RAW_SPIN_LOCK_UNLOCKED	{ 0 }

arch/ia64/kernel/head.S

@@ -1130,95 +1130,6 @@ SET_REG(b5);
 #endif /* CONFIG_IA64_BRL_EMU */
 
 #ifdef CONFIG_SMP
-	/*
-	 * This routine handles spinlock contention.  It uses a non-standard calling
-	 * convention to avoid converting leaf routines into interior routines.  Because
-	 * of this special convention, there are several restrictions:
-	 *
-	 * - do not use gp relative variables, this code is called from the kernel
-	 *   and from modules, r1 is undefined.
-	 * - do not use stacked registers, the caller owns them.
-	 * - do not use the scratch stack space, the caller owns it.
-	 * - do not use any registers other than the ones listed below
-	 *
-	 * Inputs:
-	 *   ar.pfs - saved CFM of caller
-	 *   ar.ccv - 0 (and available for use)
-	 *   r27    - flags from spin_lock_irqsave or 0.  Must be preserved.
-	 *   r28    - available for use.
-	 *   r29    - available for use.
-	 *   r30    - available for use.
-	 *   r31    - address of lock, available for use.
-	 *   b6     - return address
-	 *   p14    - available for use.
-	 *   p15    - used to track flag status.
-	 *
-	 * If you patch this code to use more registers, do not forget to update
-	 * the clobber lists for spin_lock() in arch/ia64/include/asm/spinlock.h.
-	 */
-
-#if (__GNUC__ == 3 && __GNUC_MINOR__ < 3)
-
-GLOBAL_ENTRY(ia64_spinlock_contention_pre3_4)
-	.prologue
-	.save ar.pfs, r0	// this code effectively has a zero frame size
-	.save rp, r28
-	.body
-	nop 0
-	tbit.nz p15,p0=r27,IA64_PSR_I_BIT
-	.restore sp		// pop existing prologue after next insn
-	mov b6 = r28
-	.prologue
-	.save ar.pfs, r0
-	.altrp b6
-	.body
-	;;
-(p15)	ssm psr.i		// reenable interrupts if they were on
-				// DavidM says that srlz.d is slow and is not required in this case
-.wait:
-	// exponential backoff, kdb, lockmeter etc. go in here
-	hint @pause
-	ld4 r30=[r31]		// don't use ld4.bias; if it's contended, we won't write the word
-	nop 0
-	;;
-	cmp4.ne p14,p0=r30,r0
-(p14)	br.cond.sptk.few .wait
-(p15)	rsm psr.i		// disable interrupts if we reenabled them
-	br.cond.sptk.few b6	// lock is now free, try to acquire
-	.global ia64_spinlock_contention_pre3_4_end	// for kernprof
-ia64_spinlock_contention_pre3_4_end:
-END(ia64_spinlock_contention_pre3_4)
-
-#else
-
-GLOBAL_ENTRY(ia64_spinlock_contention)
-	.prologue
-	.altrp b6
-	.body
-	tbit.nz p15,p0=r27,IA64_PSR_I_BIT
-	;;
-.wait:
-(p15)	ssm psr.i		// reenable interrupts if they were on
-				// DavidM says that srlz.d is slow and is not required in this case
-.wait2:
-	// exponential backoff, kdb, lockmeter etc. go in here
-	hint @pause
-	ld4 r30=[r31]		// don't use ld4.bias; if it's contended, we won't write the word
-	;;
-	cmp4.ne p14,p0=r30,r0
-	mov r30 = 1
-(p14)	br.cond.sptk.few .wait2
-(p15)	rsm psr.i		// disable interrupts if we reenabled them
-	;;
-	cmpxchg4.acq r30=[r31], r30, ar.ccv
-	;;
-	cmp4.ne p14,p0=r0,r30
-(p14)	br.cond.sptk.few .wait
-
-	br.ret.sptk.many b6	// lock is now taken
-END(ia64_spinlock_contention)
-
-#endif
 
 #ifdef CONFIG_HOTPLUG_CPU
 GLOBAL_ENTRY(ia64_jump_to_sal)

arch/ia64/kernel/ia64_ksyms.c

@@ -84,26 +84,6 @@ EXPORT_SYMBOL(ia64_save_scratch_fpregs);
 #include <asm/unwind.h>
 EXPORT_SYMBOL(unw_init_running);
 
-#ifdef ASM_SUPPORTED
-# ifdef CONFIG_SMP
-#  if (__GNUC__ == 3 && __GNUC_MINOR__ < 3)
-/*
- * This is not a normal routine and we don't want a function descriptor for it, so we use
- * a fake declaration here.
- */
-extern char ia64_spinlock_contention_pre3_4;
-EXPORT_SYMBOL(ia64_spinlock_contention_pre3_4);
-#  else
-/*
- * This is not a normal routine and we don't want a function descriptor for it, so we use
- * a fake declaration here.
- */
-extern char ia64_spinlock_contention;
-EXPORT_SYMBOL(ia64_spinlock_contention);
-#  endif
-# endif
-#endif
-
 #if defined(CONFIG_IA64_ESI) || defined(CONFIG_IA64_ESI_MODULE)
 extern void esi_call_phys (void);
 EXPORT_SYMBOL_GPL(esi_call_phys);

arch/ia64/oprofile/backtrace.c

@@ -32,24 +32,6 @@ typedef struct
 	u64 *prev_pfs_loc;	/* state for WAR for old spinlock ool code */
 } ia64_backtrace_t;
 
-#if (__GNUC__ == 3 && __GNUC_MINOR__ < 3)
-/*
- * Returns non-zero if the PC is in the spinlock contention out-of-line code
- * with non-standard calling sequence (on older compilers).
- */
-static __inline__ int in_old_ool_spinlock_code(unsigned long pc)
-{
-	extern const char ia64_spinlock_contention_pre3_4[] __attribute__ ((weak));
-	extern const char ia64_spinlock_contention_pre3_4_end[] __attribute__ ((weak));
-	unsigned long sc_start = (unsigned long)ia64_spinlock_contention_pre3_4;
-	unsigned long sc_end = (unsigned long)ia64_spinlock_contention_pre3_4_end;
-	return (sc_start && sc_end && pc >= sc_start && pc < sc_end);
-}
-#else
-/* Newer spinlock code does a proper br.call and works fine with the unwinder */
-#define in_old_ool_spinlock_code(pc)	0
-#endif
-
 /* Returns non-zero if the PC is in the Interrupt Vector Table */
 static __inline__ int in_ivt_code(unsigned long pc)
 {
@@ -80,7 +62,7 @@ static __inline__ int next_frame(ia64_backtrace_t *bt)
 	 */
 	if (bt->prev_pfs_loc && bt->regs && bt->frame.pfs_loc == bt->prev_pfs_loc)
 		bt->frame.pfs_loc = &bt->regs->ar_pfs;
-	bt->prev_pfs_loc = (in_old_ool_spinlock_code(bt->frame.ip) ? bt->frame.pfs_loc : NULL);
+	bt->prev_pfs_loc = NULL;
 
 	return unw_unwind(&bt->frame) == 0;
 }