arch/ia64/include/asm/spinlock.h

   1 #ifndef _ASM_IA64_SPINLOCK_H
   2 #define _ASM_IA64_SPINLOCK_H
   3
   4 /*
   5  * Copyright (C) 1998-2003 Hewlett-Packard Co
   6  *      David Mosberger-Tang <davidm@hpl.hp.com>
   7  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
   8  *
   9  * This file is used for SMP configurations only.
  10  */
  11
  12 #include <linux/compiler.h>
  13 #include <linux/kernel.h>
  14 #include <linux/bitops.h>
  15
  16 #include <asm/atomic.h>
  17 #include <asm/intrinsics.h>
  18 #include <asm/system.h>
  19
  20 #define arch_spin_lock_init(x)                  ((x)->lock = 0)
  21
  22 /*
  23  * Ticket locks are conceptually two parts, one indicating the current head of
  24  * the queue, and the other indicating the current tail. The lock is acquired
  25  * by atomically noting the tail and incrementing it by one (thus adding
  26  * ourself to the queue and noting our position), then waiting until the head
  27  * becomes equal to the the initial value of the tail.
  28  * The pad bits in the middle are used to prevent the next_ticket number
  29  * overflowing into the now_serving number.
  30  *
  31  *   31             17  16    15  14                    0
  32  *  +----------------------------------------------------+
  33  *  |  now_serving     | padding |   next_ticket         |
  34  *  +----------------------------------------------------+
  35  */
  36
  37 #define TICKET_SHIFT    17
  38 #define TICKET_BITS     15
  39 #define TICKET_MASK     ((1 << TICKET_BITS) - 1)
  40
  41 static __always_inline void __ticket_spin_lock(arch_spinlock_t *lock)
  42 {
  43         int     *p = (int *)&lock->lock, ticket, serve;
  44
  45         ticket = ia64_fetchadd(1, p, acq);
  46
  47         if (!(((ticket >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
  48                 return;
  49
  50         ia64_invala();
  51
  52         for (;;) {
  53                 asm volatile ("ld4.c.nc %0=[%1]" : "=r"(serve) : "r"(p) : "memory");
  54
  55                 if (!(((serve >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
  56                         return;
  57                 cpu_relax();
  58         }
  59 }
  60
  61 static __always_inline int __ticket_spin_trylock(arch_spinlock_t *lock)
  62 {
  63         int tmp = ACCESS_ONCE(lock->lock);
  64
  65         if (!(((tmp >> TICKET_SHIFT) ^ tmp) & TICKET_MASK))
  66                 return ia64_cmpxchg(acq, &lock->lock, tmp, tmp + 1, sizeof (tmp)) == tmp;
  67         return 0;
  68 }
  69
  70 static __always_inline void __ticket_spin_unlock(arch_spinlock_t *lock)
  71 {
  72         unsigned short  *p = (unsigned short *)&lock->lock + 1, tmp;
  73
  74         asm volatile ("ld2.bias %0=[%1]" : "=r"(tmp) : "r"(p));
  75         ACCESS_ONCE(*p) = (tmp + 2) & ~1;
  76 }
  77
  78 static __always_inline void __ticket_spin_unlock_wait(arch_spinlock_t *lock)
  79 {
  80         int     *p = (int *)&lock->lock, ticket;
  81
  82         ia64_invala();
  83
  84         for (;;) {
  85                 asm volatile ("ld4.c.nc %0=[%1]" : "=r"(ticket) : "r"(p) : "memory");
  86                 if (!(((ticket >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
  87                         return;
  88                 cpu_relax();
  89         }
  90 }
  91
  92 static inline int __ticket_spin_is_locked(arch_spinlock_t *lock)
  93 {
  94         long tmp = ACCESS_ONCE(lock->lock);
  95
  96         return !!(((tmp >> TICKET_SHIFT) ^ tmp) & TICKET_MASK);
  97 }
  98
  99 static inline int __ticket_spin_is_contended(arch_spinlock_t *lock)
 100 {
 101         long tmp = ACCESS_ONCE(lock->lock);
 102
 103         return ((tmp - (tmp >> TICKET_SHIFT)) & TICKET_MASK) > 1;
 104 }
 105
 106 static inline int arch_spin_is_locked(arch_spinlock_t *lock)
 107 {
 108         return __ticket_spin_is_locked(lock);
 109 }
 110
 111 static inline int arch_spin_is_contended(arch_spinlock_t *lock)
 112 {
 113         return __ticket_spin_is_contended(lock);
 114 }
 115 #define arch_spin_is_contended  arch_spin_is_contended
 116
 117 static __always_inline void arch_spin_lock(arch_spinlock_t *lock)
 118 {
 119         __ticket_spin_lock(lock);
 120 }
 121
 122 static __always_inline int arch_spin_trylock(arch_spinlock_t *lock)
 123 {
 124         return __ticket_spin_trylock(lock);
 125 }
 126
 127 static __always_inline void arch_spin_unlock(arch_spinlock_t *lock)
 128 {
 129         __ticket_spin_unlock(lock);
 130 }
 131
 132 static __always_inline void arch_spin_lock_flags(arch_spinlock_t *lock,
 133                                                   unsigned long flags)
 134 {
 135         arch_spin_lock(lock);
 136 }
 137
 138 static inline void arch_spin_unlock_wait(arch_spinlock_t *lock)
 139 {
 140         __ticket_spin_unlock_wait(lock);
 141 }
 142
 143 #define arch_read_can_lock(rw)          (*(volatile int *)(rw) >= 0)
 144 #define arch_write_can_lock(rw) (*(volatile int *)(rw) == 0)
 145
 146 #ifdef ASM_SUPPORTED
 147
 148 static __always_inline void
 149 arch_read_lock_flags(arch_rwlock_t *lock, unsigned long flags)
 150 {
 151         __asm__ __volatile__ (
 152                 "tbit.nz p6, p0 = %1,%2\n"
 153                 "br.few 3f\n"
 154                 "1:\n"
 155                 "fetchadd4.rel r2 = [%0], -1;;\n"
 156                 "(p6) ssm psr.i\n"
 157                 "2:\n"
 158                 "hint @pause\n"
 159                 "ld4 r2 = [%0];;\n"
 160                 "cmp4.lt p7,p0 = r2, r0\n"
 161                 "(p7) br.cond.spnt.few 2b\n"
 162                 "(p6) rsm psr.i\n"
 163                 ";;\n"
 164                 "3:\n"
 165                 "fetchadd4.acq r2 = [%0], 1;;\n"
 166                 "cmp4.lt p7,p0 = r2, r0\n"
 167                 "(p7) br.cond.spnt.few 1b\n"
 168                 : : "r"(lock), "r"(flags), "i"(IA64_PSR_I_BIT)
 169                 : "p6", "p7", "r2", "memory");
 170 }
 171
 172 #define arch_read_lock(lock) arch_read_lock_flags(lock, 0)
 173
 174 #else /* !ASM_SUPPORTED */
 175
 176 #define arch_read_lock_flags(rw, flags) arch_read_lock(rw)
 177
 178 #define arch_read_lock(rw)                                                              \
 179 do {                                                                                    \
 180         arch_rwlock_t *__read_lock_ptr = (rw);                                          \
 181                                                                                         \
 182         while (unlikely(ia64_fetchadd(1, (int *) __read_lock_ptr, acq) < 0)) {          \
 183                 ia64_fetchadd(-1, (int *) __read_lock_ptr, rel);                        \
 184                 while (*(volatile int *)__read_lock_ptr < 0)                            \
 185                         cpu_relax();                                                    \
 186         }                                                                               \
 187 } while (0)
 188
 189 #endif /* !ASM_SUPPORTED */
 190
 191 #define arch_read_unlock(rw)                                    \
 192 do {                                                            \
 193         arch_rwlock_t *__read_lock_ptr = (rw);                  \
 194         ia64_fetchadd(-1, (int *) __read_lock_ptr, rel);        \
 195 } while (0)
 196
 197 #ifdef ASM_SUPPORTED
 198
 199 static __always_inline void
 200 arch_write_lock_flags(arch_rwlock_t *lock, unsigned long flags)
 201 {
 202         __asm__ __volatile__ (
 203                 "tbit.nz p6, p0 = %1, %2\n"
 204                 "mov ar.ccv = r0\n"
 205                 "dep r29 = -1, r0, 31, 1\n"
 206                 "br.few 3f;;\n"
 207                 "1:\n"
 208                 "(p6) ssm psr.i\n"
 209                 "2:\n"
 210                 "hint @pause\n"
 211                 "ld4 r2 = [%0];;\n"
 212                 "cmp4.eq p0,p7 = r0, r2\n"
 213                 "(p7) br.cond.spnt.few 2b\n"
 214                 "(p6) rsm psr.i\n"
 215                 ";;\n"
 216                 "3:\n"
 217                 "cmpxchg4.acq r2 = [%0], r29, ar.ccv;;\n"
 218                 "cmp4.eq p0,p7 = r0, r2\n"
 219                 "(p7) br.cond.spnt.few 1b;;\n"
 220                 : : "r"(lock), "r"(flags), "i"(IA64_PSR_I_BIT)
 221                 : "ar.ccv", "p6", "p7", "r2", "r29", "memory");
 222 }
 223
 224 #define arch_write_lock(rw) arch_write_lock_flags(rw, 0)
 225
 226 #define arch_write_trylock(rw)                                                  \
 227 ({                                                                              \
 228         register long result;                                                   \
 229                                                                                 \
 230         __asm__ __volatile__ (                                                  \
 231                 "mov ar.ccv = r0\n"                                             \
 232                 "dep r29 = -1, r0, 31, 1;;\n"                                   \
 233                 "cmpxchg4.acq %0 = [%1], r29, ar.ccv\n"                         \
 234                 : "=r"(result) : "r"(rw) : "ar.ccv", "r29", "memory");          \
 235         (result == 0);                                                          \
 236 })
 237
 238 static inline void arch_write_unlock(arch_rwlock_t *x)
 239 {
 240         u8 *y = (u8 *)x;
 241         barrier();
 242         asm volatile ("st1.rel.nta [%0] = r0\n\t" :: "r"(y+3) : "memory" );
 243 }
 244
 245 #else /* !ASM_SUPPORTED */
 246
 247 #define arch_write_lock_flags(l, flags) arch_write_lock(l)
 248
 249 #define arch_write_lock(l)                                                              \
 250 ({                                                                                      \
 251         __u64 ia64_val, ia64_set_val = ia64_dep_mi(-1, 0, 31, 1);                       \
 252         __u32 *ia64_write_lock_ptr = (__u32 *) (l);                                     \
 253         do {                                                                            \
 254                 while (*ia64_write_lock_ptr)                                            \
 255                         ia64_barrier();                                                 \
 256                 ia64_val = ia64_cmpxchg4_acq(ia64_write_lock_ptr, ia64_set_val, 0);     \
 257         } while (ia64_val);                                                             \
 258 })
 259
 260 #define arch_write_trylock(rw)                                          \
 261 ({                                                                      \
 262         __u64 ia64_val;                                                 \
 263         __u64 ia64_set_val = ia64_dep_mi(-1, 0, 31,1);                  \
 264         ia64_val = ia64_cmpxchg4_acq((__u32 *)(rw), ia64_set_val, 0);   \
 265         (ia64_val == 0);                                                \
 266 })
 267
 268 static inline void arch_write_unlock(arch_rwlock_t *x)
 269 {
 270         barrier();
 271         x->write_lock = 0;
 272 }
 273
 274 #endif /* !ASM_SUPPORTED */
 275
 276 static inline int arch_read_trylock(arch_rwlock_t *x)
 277 {
 278         union {
 279                 arch_rwlock_t lock;
 280                 __u32 word;
 281         } old, new;
 282         old.lock = new.lock = *x;
 283         old.lock.write_lock = new.lock.write_lock = 0;
 284         ++new.lock.read_counter;
 285         return (u32)ia64_cmpxchg4_acq((__u32 *)(x), new.word, old.word) == old.word;
 286 }
 287
 288 #define arch_spin_relax(lock)   cpu_relax()
 289 #define arch_read_relax(lock)   cpu_relax()
 290 #define arch_write_relax(lock)  cpu_relax()
 291
 292 #endif /*  _ASM_IA64_SPINLOCK_H */