LXR linux/include/linux/percpu.h

   1#ifndef __LINUX_PERCPU_H
   2#define __LINUX_PERCPU_H
   3
   4#include <linux/mmdebug.h>
   5#include <linux/preempt.h>
   6#include <linux/smp.h>
   7#include <linux/cpumask.h>
   8#include <linux/pfn.h>
   9#include <linux/init.h>
  10
  11#include <asm/percpu.h>
  12
  13/* enough to cover all DEFINE_PER_CPUs in modules */
  14#ifdef CONFIG_MODULES
  15#define PERCPU_MODULE_RESERVE           (8 << 10)
  16#else
  17#define PERCPU_MODULE_RESERVE           0
  18#endif
  19
  20#ifndef PERCPU_ENOUGH_ROOM
  21#define PERCPU_ENOUGH_ROOM                                              \
  22        (ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES) +      \
  23         PERCPU_MODULE_RESERVE)
  24#endif
  25
  26/*
  27 * Must be an lvalue. Since @var must be a simple identifier,
  28 * we force a syntax error here if it isn't.
  29 */
  30#define get_cpu_var(var) (*({                           \
  31        preempt_disable();                              \
  32        &__get_cpu_var(var); }))
  33
  34/*
  35 * The weird & is necessary because sparse considers (void)(var) to be
  36 * a direct dereference of percpu variable (var).
  37 */
  38#define put_cpu_var(var) do {                           \
  39        (void)&(var);                                   \
  40        preempt_enable();                               \
  41} while (0)
  42
  43#define get_cpu_ptr(var) ({                             \
  44        preempt_disable();                              \
  45        this_cpu_ptr(var); })
  46
  47#define put_cpu_ptr(var) do {                           \
  48        (void)(var);                                    \
  49        preempt_enable();                               \
  50} while (0)
  51
  52/* minimum unit size, also is the maximum supported allocation size */
  53#define PCPU_MIN_UNIT_SIZE              PFN_ALIGN(32 << 10)
  54
  55/*
  56 * Percpu allocator can serve percpu allocations before slab is
  57 * initialized which allows slab to depend on the percpu allocator.
  58 * The following two parameters decide how much resource to
  59 * preallocate for this.  Keep PERCPU_DYNAMIC_RESERVE equal to or
  60 * larger than PERCPU_DYNAMIC_EARLY_SIZE.
  61 */
  62#define PERCPU_DYNAMIC_EARLY_SLOTS      128
  63#define PERCPU_DYNAMIC_EARLY_SIZE       (12 << 10)
  64
  65/*
  66 * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
  67 * back on the first chunk for dynamic percpu allocation if arch is
  68 * manually allocating and mapping it for faster access (as a part of
  69 * large page mapping for example).
  70 *
  71 * The following values give between one and two pages of free space
  72 * after typical minimal boot (2-way SMP, single disk and NIC) with
  73 * both defconfig and a distro config on x86_64 and 32.  More
  74 * intelligent way to determine this would be nice.
  75 */
  76#if BITS_PER_LONG > 32
  77#define PERCPU_DYNAMIC_RESERVE          (20 << 10)
  78#else
  79#define PERCPU_DYNAMIC_RESERVE          (12 << 10)
  80#endif
  81
  82extern void *pcpu_base_addr;
  83extern const unsigned long *pcpu_unit_offsets;
  84
  85struct pcpu_group_info {
  86        int                     nr_units;       /* aligned # of units */
  87        unsigned long           base_offset;    /* base address offset */
  88        unsigned int            *cpu_map;       /* unit->cpu map, empty
  89                                                 * entries contain NR_CPUS */
  90};
  91
  92struct pcpu_alloc_info {
  93        size_t                  static_size;
  94        size_t                  reserved_size;
  95        size_t                  dyn_size;
  96        size_t                  unit_size;
  97        size_t                  atom_size;
  98        size_t                  alloc_size;
  99        size_t                  __ai_size;      /* internal, don't use */
 100        int                     nr_groups;      /* 0 if grouping unnecessary */
 101        struct pcpu_group_info  groups[];
 102};
 103
 104enum pcpu_fc {
 105        PCPU_FC_AUTO,
 106        PCPU_FC_EMBED,
 107        PCPU_FC_PAGE,
 108
 109        PCPU_FC_NR,
 110};
 111extern const char * const pcpu_fc_names[PCPU_FC_NR];
 112
 113extern enum pcpu_fc pcpu_chosen_fc;
 114
 115typedef void * (*pcpu_fc_alloc_fn_t)(unsigned int cpu, size_t size,
 116                                     size_t align);
 117typedef void (*pcpu_fc_free_fn_t)(void *ptr, size_t size);
 118typedef void (*pcpu_fc_populate_pte_fn_t)(unsigned long addr);
 119typedef int (pcpu_fc_cpu_distance_fn_t)(unsigned int from, unsigned int to);
 120
 121extern struct pcpu_alloc_info * __init pcpu_alloc_alloc_info(int nr_groups,
 122                                                             int nr_units);
 123extern void __init pcpu_free_alloc_info(struct pcpu_alloc_info *ai);
 124
 125extern int __init pcpu_setup_first_chunk(const struct pcpu_alloc_info *ai,
 126                                         void *base_addr);
 127
 128#ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
 129extern int __init pcpu_embed_first_chunk(size_t reserved_size, size_t dyn_size,
 130                                size_t atom_size,
 131                                pcpu_fc_cpu_distance_fn_t cpu_distance_fn,
 132                                pcpu_fc_alloc_fn_t alloc_fn,
 133                                pcpu_fc_free_fn_t free_fn);
 134#endif
 135
 136#ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
 137extern int __init pcpu_page_first_chunk(size_t reserved_size,
 138                                pcpu_fc_alloc_fn_t alloc_fn,
 139                                pcpu_fc_free_fn_t free_fn,
 140                                pcpu_fc_populate_pte_fn_t populate_pte_fn);
 141#endif
 142
 143/*
 144 * Use this to get to a cpu's version of the per-cpu object
 145 * dynamically allocated. Non-atomic access to the current CPU's
 146 * version should probably be combined with get_cpu()/put_cpu().
 147 */
 148#ifdef CONFIG_SMP
 149#define per_cpu_ptr(ptr, cpu)   SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
 150#else
 151#define per_cpu_ptr(ptr, cpu)   ({ (void)(cpu); VERIFY_PERCPU_PTR((ptr)); })
 152#endif
 153
 154extern void __percpu *__alloc_reserved_percpu(size_t size, size_t align);
 155extern bool is_kernel_percpu_address(unsigned long addr);
 156
 157#if !defined(CONFIG_SMP) || !defined(CONFIG_HAVE_SETUP_PER_CPU_AREA)
 158extern void __init setup_per_cpu_areas(void);
 159#endif
 160extern void __init percpu_init_late(void);
 161
 162extern void __percpu *__alloc_percpu(size_t size, size_t align);
 163extern void free_percpu(void __percpu *__pdata);
 164extern phys_addr_t per_cpu_ptr_to_phys(void *addr);
 165
 166#define alloc_percpu(type)      \
 167        (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type))
 168
 169/*
 170 * Branching function to split up a function into a set of functions that
 171 * are called for different scalar sizes of the objects handled.
 172 */
 173
 174extern void __bad_size_call_parameter(void);
 175
 176#define __pcpu_size_call_return(stem, variable)                         \
 177({      typeof(variable) pscr_ret__;                                    \
 178        __verify_pcpu_ptr(&(variable));                                 \
 179        switch(sizeof(variable)) {                                      \
 180        case 1: pscr_ret__ = stem##1(variable);break;                   \
 181        case 2: pscr_ret__ = stem##2(variable);break;                   \
 182        case 4: pscr_ret__ = stem##4(variable);break;                   \
 183        case 8: pscr_ret__ = stem##8(variable);break;                   \
 184        default:                                                        \
 185                __bad_size_call_parameter();break;                      \
 186        }                                                               \
 187        pscr_ret__;                                                     \
 188})
 189
 190#define __pcpu_size_call_return2(stem, variable, ...)                   \
 191({                                                                      \
 192        typeof(variable) pscr2_ret__;                                   \
 193        __verify_pcpu_ptr(&(variable));                                 \
 194        switch(sizeof(variable)) {                                      \
 195        case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;    \
 196        case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;    \
 197        case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;    \
 198        case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;    \
 199        default:                                                        \
 200                __bad_size_call_parameter(); break;                     \
 201        }                                                               \
 202        pscr2_ret__;                                                    \
 203})
 204
 205/*
 206 * Special handling for cmpxchg_double.  cmpxchg_double is passed two
 207 * percpu variables.  The first has to be aligned to a double word
 208 * boundary and the second has to follow directly thereafter.
 209 * We enforce this on all architectures even if they don't support
 210 * a double cmpxchg instruction, since it's a cheap requirement, and it
 211 * avoids breaking the requirement for architectures with the instruction.
 212 */
 213#define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...)           \
 214({                                                                      \
 215        bool pdcrb_ret__;                                               \
 216        __verify_pcpu_ptr(&pcp1);                                       \
 217        BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2));                     \
 218        VM_BUG_ON((unsigned long)(&pcp1) % (2 * sizeof(pcp1)));         \
 219        VM_BUG_ON((unsigned long)(&pcp2) !=                             \
 220                  (unsigned long)(&pcp1) + sizeof(pcp1));               \
 221        switch(sizeof(pcp1)) {                                          \
 222        case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break;  \
 223        case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break;  \
 224        case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break;  \
 225        case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break;  \
 226        default:                                                        \
 227                __bad_size_call_parameter(); break;                     \
 228        }                                                               \
 229        pdcrb_ret__;                                                    \
 230})
 231
 232#define __pcpu_size_call(stem, variable, ...)                           \
 233do {                                                                    \
 234        __verify_pcpu_ptr(&(variable));                                 \
 235        switch(sizeof(variable)) {                                      \
 236                case 1: stem##1(variable, __VA_ARGS__);break;           \
 237                case 2: stem##2(variable, __VA_ARGS__);break;           \
 238                case 4: stem##4(variable, __VA_ARGS__);break;           \
 239                case 8: stem##8(variable, __VA_ARGS__);break;           \
 240                default:                                                \
 241                        __bad_size_call_parameter();break;              \
 242        }                                                               \
 243} while (0)
 244
 245/*
 246 * Optimized manipulation for memory allocated through the per cpu
 247 * allocator or for addresses of per cpu variables.
 248 *
 249 * These operation guarantee exclusivity of access for other operations
 250 * on the *same* processor. The assumption is that per cpu data is only
 251 * accessed by a single processor instance (the current one).
 252 *
 253 * The first group is used for accesses that must be done in a
 254 * preemption safe way since we know that the context is not preempt
 255 * safe. Interrupts may occur. If the interrupt modifies the variable
 256 * too then RMW actions will not be reliable.
 257 *
 258 * The arch code can provide optimized functions in two ways:
 259 *
 260 * 1. Override the function completely. F.e. define this_cpu_add().
 261 *    The arch must then ensure that the various scalar format passed
 262 *    are handled correctly.
 263 *
 264 * 2. Provide functions for certain scalar sizes. F.e. provide
 265 *    this_cpu_add_2() to provide per cpu atomic operations for 2 byte
 266 *    sized RMW actions. If arch code does not provide operations for
 267 *    a scalar size then the fallback in the generic code will be
 268 *    used.
 269 */
 270
 271#define _this_cpu_generic_read(pcp)                                     \
 272({      typeof(pcp) ret__;                                              \
 273        preempt_disable();                                              \
 274        ret__ = *this_cpu_ptr(&(pcp));                                  \
 275        preempt_enable();                                               \
 276        ret__;                                                          \
 277})
 278
 279#ifndef this_cpu_read
 280# ifndef this_cpu_read_1
 281#  define this_cpu_read_1(pcp)  _this_cpu_generic_read(pcp)
 282# endif
 283# ifndef this_cpu_read_2
 284#  define this_cpu_read_2(pcp)  _this_cpu_generic_read(pcp)
 285# endif
 286# ifndef this_cpu_read_4
 287#  define this_cpu_read_4(pcp)  _this_cpu_generic_read(pcp)
 288# endif
 289# ifndef this_cpu_read_8
 290#  define this_cpu_read_8(pcp)  _this_cpu_generic_read(pcp)
 291# endif
 292# define this_cpu_read(pcp)     __pcpu_size_call_return(this_cpu_read_, (pcp))
 293#endif
 294
 295#define _this_cpu_generic_to_op(pcp, val, op)                           \
 296do {                                                                    \
 297        unsigned long flags;                                            \
 298        raw_local_irq_save(flags);                                      \
 299        *__this_cpu_ptr(&(pcp)) op val;                                 \
 300        raw_local_irq_restore(flags);                                   \
 301} while (0)
 302
 303#ifndef this_cpu_write
 304# ifndef this_cpu_write_1
 305#  define this_cpu_write_1(pcp, val)    _this_cpu_generic_to_op((pcp), (val), =)
 306# endif
 307# ifndef this_cpu_write_2
 308#  define this_cpu_write_2(pcp, val)    _this_cpu_generic_to_op((pcp), (val), =)
 309# endif
 310# ifndef this_cpu_write_4
 311#  define this_cpu_write_4(pcp, val)    _this_cpu_generic_to_op((pcp), (val), =)
 312# endif
 313# ifndef this_cpu_write_8
 314#  define this_cpu_write_8(pcp, val)    _this_cpu_generic_to_op((pcp), (val), =)
 315# endif
 316# define this_cpu_write(pcp, val)       __pcpu_size_call(this_cpu_write_, (pcp), (val))
 317#endif
 318
 319#ifndef this_cpu_add
 320# ifndef this_cpu_add_1
 321#  define this_cpu_add_1(pcp, val)      _this_cpu_generic_to_op((pcp), (val), +=)
 322# endif
 323# ifndef this_cpu_add_2
 324#  define this_cpu_add_2(pcp, val)      _this_cpu_generic_to_op((pcp), (val), +=)
 325# endif
 326# ifndef this_cpu_add_4
 327#  define this_cpu_add_4(pcp, val)      _this_cpu_generic_to_op((pcp), (val), +=)
 328# endif
 329# ifndef this_cpu_add_8
 330#  define this_cpu_add_8(pcp, val)      _this_cpu_generic_to_op((pcp), (val), +=)
 331# endif
 332# define this_cpu_add(pcp, val)         __pcpu_size_call(this_cpu_add_, (pcp), (val))
 333#endif
 334
 335#ifndef this_cpu_sub
 336# define this_cpu_sub(pcp, val)         this_cpu_add((pcp), -(typeof(pcp))(val))
 337#endif
 338
 339#ifndef this_cpu_inc
 340# define this_cpu_inc(pcp)              this_cpu_add((pcp), 1)
 341#endif
 342
 343#ifndef this_cpu_dec
 344# define this_cpu_dec(pcp)              this_cpu_sub((pcp), 1)
 345#endif
 346
 347#ifndef this_cpu_and
 348# ifndef this_cpu_and_1
 349#  define this_cpu_and_1(pcp, val)      _this_cpu_generic_to_op((pcp), (val), &=)
 350# endif
 351# ifndef this_cpu_and_2
 352#  define this_cpu_and_2(pcp, val)      _this_cpu_generic_to_op((pcp), (val), &=)
 353# endif
 354# ifndef this_cpu_and_4
 355#  define this_cpu_and_4(pcp, val)      _this_cpu_generic_to_op((pcp), (val), &=)
 356# endif
 357# ifndef this_cpu_and_8
 358#  define this_cpu_and_8(pcp, val)      _this_cpu_generic_to_op((pcp), (val), &=)
 359# endif
 360# define this_cpu_and(pcp, val)         __pcpu_size_call(this_cpu_and_, (pcp), (val))
 361#endif
 362
 363#ifndef this_cpu_or
 364# ifndef this_cpu_or_1
 365#  define this_cpu_or_1(pcp, val)       _this_cpu_generic_to_op((pcp), (val), |=)
 366# endif
 367# ifndef this_cpu_or_2
 368#  define this_cpu_or_2(pcp, val)       _this_cpu_generic_to_op((pcp), (val), |=)
 369# endif
 370# ifndef this_cpu_or_4
 371#  define this_cpu_or_4(pcp, val)       _this_cpu_generic_to_op((pcp), (val), |=)
 372# endif
 373# ifndef this_cpu_or_8
 374#  define this_cpu_or_8(pcp, val)       _this_cpu_generic_to_op((pcp), (val), |=)
 375# endif
 376# define this_cpu_or(pcp, val)          __pcpu_size_call(this_cpu_or_, (pcp), (val))
 377#endif
 378
 379#define _this_cpu_generic_add_return(pcp, val)                          \
 380({                                                                      \
 381        typeof(pcp) ret__;                                              \
 382        unsigned long flags;                                            \
 383        raw_local_irq_save(flags);                                      \
 384        __this_cpu_add(pcp, val);                                       \
 385        ret__ = __this_cpu_read(pcp);                                   \
 386        raw_local_irq_restore(flags);                                   \
 387        ret__;                                                          \
 388})
 389
 390#ifndef this_cpu_add_return
 391# ifndef this_cpu_add_return_1
 392#  define this_cpu_add_return_1(pcp, val)       _this_cpu_generic_add_return(pcp, val)
 393# endif
 394# ifndef this_cpu_add_return_2
 395#  define this_cpu_add_return_2(pcp, val)       _this_cpu_generic_add_return(pcp, val)
 396# endif
 397# ifndef this_cpu_add_return_4
 398#  define this_cpu_add_return_4(pcp, val)       _this_cpu_generic_add_return(pcp, val)
 399# endif
 400# ifndef this_cpu_add_return_8
 401#  define this_cpu_add_return_8(pcp, val)       _this_cpu_generic_add_return(pcp, val)
 402# endif
 403# define this_cpu_add_return(pcp, val)  __pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
 404#endif
 405
 406#define this_cpu_sub_return(pcp, val)   this_cpu_add_return(pcp, -(typeof(pcp))(val))
 407#define this_cpu_inc_return(pcp)        this_cpu_add_return(pcp, 1)
 408#define this_cpu_dec_return(pcp)        this_cpu_add_return(pcp, -1)
 409
 410#define _this_cpu_generic_xchg(pcp, nval)                               \
 411({      typeof(pcp) ret__;                                              \
 412        unsigned long flags;                                            \
 413        raw_local_irq_save(flags);                                      \
 414        ret__ = __this_cpu_read(pcp);                                   \
 415        __this_cpu_write(pcp, nval);                                    \
 416        raw_local_irq_restore(flags);                                   \
 417        ret__;                                                          \
 418})
 419
 420#ifndef this_cpu_xchg
 421# ifndef this_cpu_xchg_1
 422#  define this_cpu_xchg_1(pcp, nval)    _this_cpu_generic_xchg(pcp, nval)
 423# endif
 424# ifndef this_cpu_xchg_2
 425#  define this_cpu_xchg_2(pcp, nval)    _this_cpu_generic_xchg(pcp, nval)
 426# endif
 427# ifndef this_cpu_xchg_4
 428#  define this_cpu_xchg_4(pcp, nval)    _this_cpu_generic_xchg(pcp, nval)
 429# endif
 430# ifndef this_cpu_xchg_8
 431#  define this_cpu_xchg_8(pcp, nval)    _this_cpu_generic_xchg(pcp, nval)
 432# endif
 433# define this_cpu_xchg(pcp, nval)       \
 434        __pcpu_size_call_return2(this_cpu_xchg_, (pcp), nval)
 435#endif
 436
 437#define _this_cpu_generic_cmpxchg(pcp, oval, nval)                      \
 438({                                                                      \
 439        typeof(pcp) ret__;                                              \
 440        unsigned long flags;                                            \
 441        raw_local_irq_save(flags);                                      \
 442        ret__ = __this_cpu_read(pcp);                                   \
 443        if (ret__ == (oval))                                            \
 444                __this_cpu_write(pcp, nval);                            \
 445        raw_local_irq_restore(flags);                                   \
 446        ret__;                                                          \
 447})
 448
 449#ifndef this_cpu_cmpxchg
 450# ifndef this_cpu_cmpxchg_1
 451#  define this_cpu_cmpxchg_1(pcp, oval, nval)   _this_cpu_generic_cmpxchg(pcp, oval, nval)
 452# endif
 453# ifndef this_cpu_cmpxchg_2
 454#  define this_cpu_cmpxchg_2(pcp, oval, nval)   _this_cpu_generic_cmpxchg(pcp, oval, nval)
 455# endif
 456# ifndef this_cpu_cmpxchg_4
 457#  define this_cpu_cmpxchg_4(pcp, oval, nval)   _this_cpu_generic_cmpxchg(pcp, oval, nval)
 458# endif
 459# ifndef this_cpu_cmpxchg_8
 460#  define this_cpu_cmpxchg_8(pcp, oval, nval)   _this_cpu_generic_cmpxchg(pcp, oval, nval)
 461# endif
 462# define this_cpu_cmpxchg(pcp, oval, nval)      \
 463        __pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
 464#endif
 465
 466/*
 467 * cmpxchg_double replaces two adjacent scalars at once.  The first
 468 * two parameters are per cpu variables which have to be of the same
 469 * size.  A truth value is returned to indicate success or failure
 470 * (since a double register result is difficult to handle).  There is
 471 * very limited hardware support for these operations, so only certain
 472 * sizes may work.
 473 */
 474#define _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)        \
 475({                                                                      \
 476        int ret__;                                                      \
 477        unsigned long flags;                                            \
 478        raw_local_irq_save(flags);                                      \
 479        ret__ = __this_cpu_generic_cmpxchg_double(pcp1, pcp2,           \
 480                        oval1, oval2, nval1, nval2);                    \
 481        raw_local_irq_restore(flags);                                   \
 482        ret__;                                                          \
 483})
 484
 485#ifndef this_cpu_cmpxchg_double
 486# ifndef this_cpu_cmpxchg_double_1
 487#  define this_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)     \
 488        _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 489# endif
 490# ifndef this_cpu_cmpxchg_double_2
 491#  define this_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)     \
 492        _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 493# endif
 494# ifndef this_cpu_cmpxchg_double_4
 495#  define this_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)     \
 496        _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 497# endif
 498# ifndef this_cpu_cmpxchg_double_8
 499#  define this_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)     \
 500        _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 501# endif
 502# define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)        \
 503        __pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
 504#endif
 505
 506/*
 507 * Generic percpu operations for context that are safe from preemption/interrupts.
 508 * Either we do not care about races or the caller has the
 509 * responsibility of handling preemption/interrupt issues. Arch code can still
 510 * override these instructions since the arch per cpu code may be more
 511 * efficient and may actually get race freeness for free (that is the
 512 * case for x86 for example).
 513 *
 514 * If there is no other protection through preempt disable and/or
 515 * disabling interupts then one of these RMW operations can show unexpected
 516 * behavior because the execution thread was rescheduled on another processor
 517 * or an interrupt occurred and the same percpu variable was modified from
 518 * the interrupt context.
 519 */
 520#ifndef __this_cpu_read
 521# ifndef __this_cpu_read_1
 522#  define __this_cpu_read_1(pcp)        (*__this_cpu_ptr(&(pcp)))
 523# endif
 524# ifndef __this_cpu_read_2
 525#  define __this_cpu_read_2(pcp)        (*__this_cpu_ptr(&(pcp)))
 526# endif
 527# ifndef __this_cpu_read_4
 528#  define __this_cpu_read_4(pcp)        (*__this_cpu_ptr(&(pcp)))
 529# endif
 530# ifndef __this_cpu_read_8
 531#  define __this_cpu_read_8(pcp)        (*__this_cpu_ptr(&(pcp)))
 532# endif
 533# define __this_cpu_read(pcp)   __pcpu_size_call_return(__this_cpu_read_, (pcp))
 534#endif
 535
 536#define __this_cpu_generic_to_op(pcp, val, op)                          \
 537do {                                                                    \
 538        *__this_cpu_ptr(&(pcp)) op val;                                 \
 539} while (0)
 540
 541#ifndef __this_cpu_write
 542# ifndef __this_cpu_write_1
 543#  define __this_cpu_write_1(pcp, val)  __this_cpu_generic_to_op((pcp), (val), =)
 544# endif
 545# ifndef __this_cpu_write_2
 546#  define __this_cpu_write_2(pcp, val)  __this_cpu_generic_to_op((pcp), (val), =)
 547# endif
 548# ifndef __this_cpu_write_4
 549#  define __this_cpu_write_4(pcp, val)  __this_cpu_generic_to_op((pcp), (val), =)
 550# endif
 551# ifndef __this_cpu_write_8
 552#  define __this_cpu_write_8(pcp, val)  __this_cpu_generic_to_op((pcp), (val), =)
 553# endif
 554# define __this_cpu_write(pcp, val)     __pcpu_size_call(__this_cpu_write_, (pcp), (val))
 555#endif
 556
 557#ifndef __this_cpu_add
 558# ifndef __this_cpu_add_1
 559#  define __this_cpu_add_1(pcp, val)    __this_cpu_generic_to_op((pcp), (val), +=)
 560# endif
 561# ifndef __this_cpu_add_2
 562#  define __this_cpu_add_2(pcp, val)    __this_cpu_generic_to_op((pcp), (val), +=)
 563# endif
 564# ifndef __this_cpu_add_4
 565#  define __this_cpu_add_4(pcp, val)    __this_cpu_generic_to_op((pcp), (val), +=)
 566# endif
 567# ifndef __this_cpu_add_8
 568#  define __this_cpu_add_8(pcp, val)    __this_cpu_generic_to_op((pcp), (val), +=)
 569# endif
 570# define __this_cpu_add(pcp, val)       __pcpu_size_call(__this_cpu_add_, (pcp), (val))
 571#endif
 572
 573#ifndef __this_cpu_sub
 574# define __this_cpu_sub(pcp, val)       __this_cpu_add((pcp), -(typeof(pcp))(val))
 575#endif
 576
 577#ifndef __this_cpu_inc
 578# define __this_cpu_inc(pcp)            __this_cpu_add((pcp), 1)
 579#endif
 580
 581#ifndef __this_cpu_dec
 582# define __this_cpu_dec(pcp)            __this_cpu_sub((pcp), 1)
 583#endif
 584
 585#ifndef __this_cpu_and
 586# ifndef __this_cpu_and_1
 587#  define __this_cpu_and_1(pcp, val)    __this_cpu_generic_to_op((pcp), (val), &=)
 588# endif
 589# ifndef __this_cpu_and_2
 590#  define __this_cpu_and_2(pcp, val)    __this_cpu_generic_to_op((pcp), (val), &=)
 591# endif
 592# ifndef __this_cpu_and_4
 593#  define __this_cpu_and_4(pcp, val)    __this_cpu_generic_to_op((pcp), (val), &=)
 594# endif
 595# ifndef __this_cpu_and_8
 596#  define __this_cpu_and_8(pcp, val)    __this_cpu_generic_to_op((pcp), (val), &=)
 597# endif
 598# define __this_cpu_and(pcp, val)       __pcpu_size_call(__this_cpu_and_, (pcp), (val))
 599#endif
 600
 601#ifndef __this_cpu_or
 602# ifndef __this_cpu_or_1
 603#  define __this_cpu_or_1(pcp, val)     __this_cpu_generic_to_op((pcp), (val), |=)
 604# endif
 605# ifndef __this_cpu_or_2
 606#  define __this_cpu_or_2(pcp, val)     __this_cpu_generic_to_op((pcp), (val), |=)
 607# endif
 608# ifndef __this_cpu_or_4
 609#  define __this_cpu_or_4(pcp, val)     __this_cpu_generic_to_op((pcp), (val), |=)
 610# endif
 611# ifndef __this_cpu_or_8
 612#  define __this_cpu_or_8(pcp, val)     __this_cpu_generic_to_op((pcp), (val), |=)
 613# endif
 614# define __this_cpu_or(pcp, val)        __pcpu_size_call(__this_cpu_or_, (pcp), (val))
 615#endif
 616
 617#define __this_cpu_generic_add_return(pcp, val)                         \
 618({                                                                      \
 619        __this_cpu_add(pcp, val);                                       \
 620        __this_cpu_read(pcp);                                           \
 621})
 622
 623#ifndef __this_cpu_add_return
 624# ifndef __this_cpu_add_return_1
 625#  define __this_cpu_add_return_1(pcp, val)     __this_cpu_generic_add_return(pcp, val)
 626# endif
 627# ifndef __this_cpu_add_return_2
 628#  define __this_cpu_add_return_2(pcp, val)     __this_cpu_generic_add_return(pcp, val)
 629# endif
 630# ifndef __this_cpu_add_return_4
 631#  define __this_cpu_add_return_4(pcp, val)     __this_cpu_generic_add_return(pcp, val)
 632# endif
 633# ifndef __this_cpu_add_return_8
 634#  define __this_cpu_add_return_8(pcp, val)     __this_cpu_generic_add_return(pcp, val)
 635# endif
 636# define __this_cpu_add_return(pcp, val)        \
 637        __pcpu_size_call_return2(__this_cpu_add_return_, pcp, val)
 638#endif
 639
 640#define __this_cpu_sub_return(pcp, val) __this_cpu_add_return(pcp, -(typeof(pcp))(val))
 641#define __this_cpu_inc_return(pcp)      __this_cpu_add_return(pcp, 1)
 642#define __this_cpu_dec_return(pcp)      __this_cpu_add_return(pcp, -1)
 643
 644#define __this_cpu_generic_xchg(pcp, nval)                              \
 645({      typeof(pcp) ret__;                                              \
 646        ret__ = __this_cpu_read(pcp);                                   \
 647        __this_cpu_write(pcp, nval);                                    \
 648        ret__;                                                          \
 649})
 650
 651#ifndef __this_cpu_xchg
 652# ifndef __this_cpu_xchg_1
 653#  define __this_cpu_xchg_1(pcp, nval)  __this_cpu_generic_xchg(pcp, nval)
 654# endif
 655# ifndef __this_cpu_xchg_2
 656#  define __this_cpu_xchg_2(pcp, nval)  __this_cpu_generic_xchg(pcp, nval)
 657# endif
 658# ifndef __this_cpu_xchg_4
 659#  define __this_cpu_xchg_4(pcp, nval)  __this_cpu_generic_xchg(pcp, nval)
 660# endif
 661# ifndef __this_cpu_xchg_8
 662#  define __this_cpu_xchg_8(pcp, nval)  __this_cpu_generic_xchg(pcp, nval)
 663# endif
 664# define __this_cpu_xchg(pcp, nval)     \
 665        __pcpu_size_call_return2(__this_cpu_xchg_, (pcp), nval)
 666#endif
 667
 668#define __this_cpu_generic_cmpxchg(pcp, oval, nval)                     \
 669({                                                                      \
 670        typeof(pcp) ret__;                                              \
 671        ret__ = __this_cpu_read(pcp);                                   \
 672        if (ret__ == (oval))                                            \
 673                __this_cpu_write(pcp, nval);                            \
 674        ret__;                                                          \
 675})
 676
 677#ifndef __this_cpu_cmpxchg
 678# ifndef __this_cpu_cmpxchg_1
 679#  define __this_cpu_cmpxchg_1(pcp, oval, nval) __this_cpu_generic_cmpxchg(pcp, oval, nval)
 680# endif
 681# ifndef __this_cpu_cmpxchg_2
 682#  define __this_cpu_cmpxchg_2(pcp, oval, nval) __this_cpu_generic_cmpxchg(pcp, oval, nval)
 683# endif
 684# ifndef __this_cpu_cmpxchg_4
 685#  define __this_cpu_cmpxchg_4(pcp, oval, nval) __this_cpu_generic_cmpxchg(pcp, oval, nval)
 686# endif
 687# ifndef __this_cpu_cmpxchg_8
 688#  define __this_cpu_cmpxchg_8(pcp, oval, nval) __this_cpu_generic_cmpxchg(pcp, oval, nval)
 689# endif
 690# define __this_cpu_cmpxchg(pcp, oval, nval)    \
 691        __pcpu_size_call_return2(__this_cpu_cmpxchg_, pcp, oval, nval)
 692#endif
 693
 694#define __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)       \
 695({                                                                      \
 696        int __ret = 0;                                                  \
 697        if (__this_cpu_read(pcp1) == (oval1) &&                         \
 698                         __this_cpu_read(pcp2)  == (oval2)) {           \
 699                __this_cpu_write(pcp1, (nval1));                        \
 700                __this_cpu_write(pcp2, (nval2));                        \
 701                __ret = 1;                                              \
 702        }                                                               \
 703        (__ret);                                                        \
 704})
 705
 706#ifndef __this_cpu_cmpxchg_double
 707# ifndef __this_cpu_cmpxchg_double_1
 708#  define __this_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)   \
 709        __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 710# endif
 711# ifndef __this_cpu_cmpxchg_double_2
 712#  define __this_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)   \
 713        __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 714# endif
 715# ifndef __this_cpu_cmpxchg_double_4
 716#  define __this_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)   \
 717        __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 718# endif
 719# ifndef __this_cpu_cmpxchg_double_8
 720#  define __this_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)   \
 721        __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 722# endif
 723# define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)      \
 724        __pcpu_double_call_return_bool(__this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
 725#endif
 726
 727#endif /* __LINUX_PERCPU_H */
 728