include/asm-ia64/processor.h

   1 #ifndef _ASM_IA64_PROCESSOR_H
   2 #define _ASM_IA64_PROCESSOR_H
   3
   4 /*
   5  * Copyright (C) 1998-2004 Hewlett-Packard Co
   6  *      David Mosberger-Tang <davidm@hpl.hp.com>
   7  *      Stephane Eranian <eranian@hpl.hp.com>
   8  * Copyright (C) 1999 Asit Mallick <asit.k.mallick@intel.com>
   9  * Copyright (C) 1999 Don Dugger <don.dugger@intel.com>
  10  *
  11  * 11/24/98     S.Eranian       added ia64_set_iva()
  12  * 12/03/99     D. Mosberger    implement thread_saved_pc() via kernel unwind API
  13  * 06/16/00     A. Mallick      added csd/ssd/tssd for ia32 support
  14  */
  15
  16 #include <linux/config.h>
  17
  18 #include <asm/intrinsics.h>
  19 #include <asm/kregs.h>
  20 #include <asm/ptrace.h>
  21 #include <asm/ustack.h>
  22
  23 #define IA64_NUM_DBG_REGS       8
  24 /*
  25  * Limits for PMC and PMD are set to less than maximum architected values
  26  * but should be sufficient for a while
  27  */
  28 #define IA64_NUM_PMC_REGS       32
  29 #define IA64_NUM_PMD_REGS       32
  30
  31 #define DEFAULT_MAP_BASE        __IA64_UL_CONST(0x2000000000000000)
  32 #define DEFAULT_TASK_SIZE       __IA64_UL_CONST(0xa000000000000000)
  33
  34 /*
  35  * TASK_SIZE really is a mis-named.  It really is the maximum user
  36  * space address (plus one).  On IA-64, there are five regions of 2TB
  37  * each (assuming 8KB page size), for a total of 8TB of user virtual
  38  * address space.
  39  */
  40 #define TASK_SIZE               (current->thread.task_size)
  41
  42 /*
  43  * MM_VM_SIZE(mm) gives the maximum address (plus 1) which may contain a mapping for
  44  * address-space MM.  Note that with 32-bit tasks, this is still DEFAULT_TASK_SIZE,
  45  * because the kernel may have installed helper-mappings above TASK_SIZE.  For example,
  46  * for x86 emulation, the LDT and GDT are mapped above TASK_SIZE.
  47  */
  48 #define MM_VM_SIZE(mm)          DEFAULT_TASK_SIZE
  49
  50 /*
  51  * This decides where the kernel will search for a free chunk of vm
  52  * space during mmap's.
  53  */
  54 #define TASK_UNMAPPED_BASE      (current->thread.map_base)
  55
  56 #define IA64_THREAD_FPH_VALID   (__IA64_UL(1) << 0)     /* floating-point high state valid? */
  57 #define IA64_THREAD_DBG_VALID   (__IA64_UL(1) << 1)     /* debug registers valid? */
  58 #define IA64_THREAD_PM_VALID    (__IA64_UL(1) << 2)     /* performance registers valid? */
  59 #define IA64_THREAD_UAC_NOPRINT (__IA64_UL(1) << 3)     /* don't log unaligned accesses */
  60 #define IA64_THREAD_UAC_SIGBUS  (__IA64_UL(1) << 4)     /* generate SIGBUS on unaligned acc. */
  61                                                         /* bit 5 is currently unused */
  62 #define IA64_THREAD_FPEMU_NOPRINT (__IA64_UL(1) << 6)   /* don't log any fpswa faults */
  63 #define IA64_THREAD_FPEMU_SIGFPE  (__IA64_UL(1) << 7)   /* send a SIGFPE for fpswa faults */
  64
  65 #define IA64_THREAD_UAC_SHIFT   3
  66 #define IA64_THREAD_UAC_MASK    (IA64_THREAD_UAC_NOPRINT | IA64_THREAD_UAC_SIGBUS)
  67 #define IA64_THREAD_FPEMU_SHIFT 6
  68 #define IA64_THREAD_FPEMU_MASK  (IA64_THREAD_FPEMU_NOPRINT | IA64_THREAD_FPEMU_SIGFPE)
  69
  70
  71 /*
  72  * This shift should be large enough to be able to represent 1000000000/itc_freq with good
  73  * accuracy while being small enough to fit 10*1000000000<<IA64_NSEC_PER_CYC_SHIFT in 64 bits
  74  * (this will give enough slack to represent 10 seconds worth of time as a scaled number).
  75  */
  76 #define IA64_NSEC_PER_CYC_SHIFT 30
  77
  78 #ifndef __ASSEMBLY__
  79
  80 #include <linux/cache.h>
  81 #include <linux/compiler.h>
  82 #include <linux/threads.h>
  83 #include <linux/types.h>
  84
  85 #include <asm/fpu.h>
  86 #include <asm/page.h>
  87 #include <asm/percpu.h>
  88 #include <asm/rse.h>
  89 #include <asm/unwind.h>
  90 #include <asm/atomic.h>
  91 #ifdef CONFIG_NUMA
  92 #include <asm/nodedata.h>
  93 #endif
  94
  95 /* like above but expressed as bitfields for more efficient access: */
  96 struct ia64_psr {
  97         __u64 reserved0 : 1;
  98         __u64 be : 1;
  99         __u64 up : 1;
 100         __u64 ac : 1;
 101         __u64 mfl : 1;
 102         __u64 mfh : 1;
 103         __u64 reserved1 : 7;
 104         __u64 ic : 1;
 105         __u64 i : 1;
 106         __u64 pk : 1;
 107         __u64 reserved2 : 1;
 108         __u64 dt : 1;
 109         __u64 dfl : 1;
 110         __u64 dfh : 1;
 111         __u64 sp : 1;
 112         __u64 pp : 1;
 113         __u64 di : 1;
 114         __u64 si : 1;
 115         __u64 db : 1;
 116         __u64 lp : 1;
 117         __u64 tb : 1;
 118         __u64 rt : 1;
 119         __u64 reserved3 : 4;
 120         __u64 cpl : 2;
 121         __u64 is : 1;
 122         __u64 mc : 1;
 123         __u64 it : 1;
 124         __u64 id : 1;
 125         __u64 da : 1;
 126         __u64 dd : 1;
 127         __u64 ss : 1;
 128         __u64 ri : 2;
 129         __u64 ed : 1;
 130         __u64 bn : 1;
 131         __u64 reserved4 : 19;
 132 };
 133
 134 /*
 135  * CPU type, hardware bug flags, and per-CPU state.  Frequently used
 136  * state comes earlier:
 137  */
 138 struct cpuinfo_ia64 {
 139         __u32 softirq_pending;
 140         __u64 itm_delta;        /* # of clock cycles between clock ticks */
 141         __u64 itm_next;         /* interval timer mask value to use for next clock tick */
 142         __u64 nsec_per_cyc;     /* (1000000000<<IA64_NSEC_PER_CYC_SHIFT)/itc_freq */
 143         __u64 unimpl_va_mask;   /* mask of unimplemented virtual address bits (from PAL) */
 144         __u64 unimpl_pa_mask;   /* mask of unimplemented physical address bits (from PAL) */
 145         __u64 *pgd_quick;
 146         __u64 *pmd_quick;
 147         __u64 pgtable_cache_sz;
 148         __u64 itc_freq;         /* frequency of ITC counter */
 149         __u64 proc_freq;        /* frequency of processor */
 150         __u64 cyc_per_usec;     /* itc_freq/1000000 */
 151         __u64 ptce_base;
 152         __u32 ptce_count[2];
 153         __u32 ptce_stride[2];
 154         struct task_struct *ksoftirqd;  /* kernel softirq daemon for this CPU */
 155
 156 #ifdef CONFIG_SMP
 157         __u64 loops_per_jiffy;
 158         int cpu;
 159 #endif
 160
 161         /* CPUID-derived information: */
 162         __u64 ppn;
 163         __u64 features;
 164         __u8 number;
 165         __u8 revision;
 166         __u8 model;
 167         __u8 family;
 168         __u8 archrev;
 169         char vendor[16];
 170
 171 #ifdef CONFIG_NUMA
 172         struct ia64_node_data *node_data;
 173 #endif
 174 };
 175
 176 DECLARE_PER_CPU(struct cpuinfo_ia64, cpu_info);
 177
 178 /*
 179  * The "local" data variable.  It refers to the per-CPU data of the currently executing
 180  * CPU, much like "current" points to the per-task data of the currently executing task.
 181  * Do not use the address of local_cpu_data, since it will be different from
 182  * cpu_data(smp_processor_id())!
 183  */
 184 #define local_cpu_data          (&__ia64_per_cpu_var(cpu_info))
 185 #define cpu_data(cpu)           (&per_cpu(cpu_info, cpu))
 186
 187 extern void identify_cpu (struct cpuinfo_ia64 *);
 188 extern void print_cpu_info (struct cpuinfo_ia64 *);
 189
 190 typedef struct {
 191         unsigned long seg;
 192 } mm_segment_t;
 193
 194 #define SET_UNALIGN_CTL(task,value)                                                             \
 195 ({                                                                                              \
 196         (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_UAC_MASK)                  \
 197                                 | (((value) << IA64_THREAD_UAC_SHIFT) & IA64_THREAD_UAC_MASK)); \
 198         0;                                                                                      \
 199 })
 200 #define GET_UNALIGN_CTL(task,addr)                                                              \
 201 ({                                                                                              \
 202         put_user(((task)->thread.flags & IA64_THREAD_UAC_MASK) >> IA64_THREAD_UAC_SHIFT,        \
 203                  (int __user *) (addr));                                                        \
 204 })
 205
 206 #define SET_FPEMU_CTL(task,value)                                                               \
 207 ({                                                                                              \
 208         (task)->thread.flags = (((task)->thread.flags & ~IA64_THREAD_FPEMU_MASK)                \
 209                           | (((value) << IA64_THREAD_FPEMU_SHIFT) & IA64_THREAD_FPEMU_MASK));   \
 210         0;                                                                                      \
 211 })
 212 #define GET_FPEMU_CTL(task,addr)                                                                \
 213 ({                                                                                              \
 214         put_user(((task)->thread.flags & IA64_THREAD_FPEMU_MASK) >> IA64_THREAD_FPEMU_SHIFT,    \
 215                  (int __user *) (addr));                                                        \
 216 })
 217
 218 #ifdef CONFIG_IA32_SUPPORT
 219 struct desc_struct {
 220         unsigned int a, b;
 221 };
 222
 223 #define desc_empty(desc)                (!((desc)->a + (desc)->b))
 224 #define desc_equal(desc1, desc2)        (((desc1)->a == (desc2)->a) && ((desc1)->b == (desc2)->b))
 225
 226 #define GDT_ENTRY_TLS_ENTRIES   3
 227 #define GDT_ENTRY_TLS_MIN       6
 228 #define GDT_ENTRY_TLS_MAX       (GDT_ENTRY_TLS_MIN + GDT_ENTRY_TLS_ENTRIES - 1)
 229
 230 #define TLS_SIZE (GDT_ENTRY_TLS_ENTRIES * 8)
 231
 232 struct partial_page_list;
 233 #endif
 234
 235 struct thread_struct {
 236         __u32 flags;                    /* various thread flags (see IA64_THREAD_*) */
 237         /* writing on_ustack is performance-critical, so it's worth spending 8 bits on it... */
 238         __u8 on_ustack;                 /* executing on user-stacks? */
 239         __u8 pad[3];
 240         __u64 ksp;                      /* kernel stack pointer */
 241         __u64 map_base;                 /* base address for get_unmapped_area() */
 242         __u64 task_size;                /* limit for task size */
 243         __u64 rbs_bot;                  /* the base address for the RBS */
 244         int last_fph_cpu;               /* CPU that may hold the contents of f32-f127 */
 245
 246 #ifdef CONFIG_IA32_SUPPORT
 247         __u64 eflag;                    /* IA32 EFLAGS reg */
 248         __u64 fsr;                      /* IA32 floating pt status reg */
 249         __u64 fcr;                      /* IA32 floating pt control reg */
 250         __u64 fir;                      /* IA32 fp except. instr. reg */
 251         __u64 fdr;                      /* IA32 fp except. data reg */
 252         __u64 old_k1;                   /* old value of ar.k1 */
 253         __u64 old_iob;                  /* old IOBase value */
 254         struct partial_page_list *ppl;  /* partial page list for 4K page size issue */
 255         /* cached TLS descriptors. */
 256         struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
 257
 258 # define INIT_THREAD_IA32       .eflag =        0,                      \
 259                                 .fsr =          0,                      \
 260                                 .fcr =          0x17800000037fULL,      \
 261                                 .fir =          0,                      \
 262                                 .fdr =          0,                      \
 263                                 .old_k1 =       0,                      \
 264                                 .old_iob =      0,                      \
 265                                 .ppl =          NULL,
 266 #else
 267 # define INIT_THREAD_IA32
 268 #endif /* CONFIG_IA32_SUPPORT */
 269 #ifdef CONFIG_PERFMON
 270         __u64 pmcs[IA64_NUM_PMC_REGS];
 271         __u64 pmds[IA64_NUM_PMD_REGS];
 272         void *pfm_context;                   /* pointer to detailed PMU context */
 273         unsigned long pfm_needs_checking;    /* when >0, pending perfmon work on kernel exit */
 274 # define INIT_THREAD_PM         .pmcs =                 {0UL, },  \
 275                                 .pmds =                 {0UL, },  \
 276                                 .pfm_context =          NULL,     \
 277                                 .pfm_needs_checking =   0UL,
 278 #else
 279 # define INIT_THREAD_PM
 280 #endif
 281         __u64 dbr[IA64_NUM_DBG_REGS];
 282         __u64 ibr[IA64_NUM_DBG_REGS];
 283         struct ia64_fpreg fph[96];      /* saved/loaded on demand */
 284 };
 285
 286 #define INIT_THREAD {                                           \
 287         .flags =        0,                                      \
 288         .on_ustack =    0,                                      \
 289         .ksp =          0,                                      \
 290         .map_base =     DEFAULT_MAP_BASE,                       \
 291         .rbs_bot =      STACK_TOP - DEFAULT_USER_STACK_SIZE,    \
 292         .task_size =    DEFAULT_TASK_SIZE,                      \
 293         .last_fph_cpu =  -1,                                    \
 294         INIT_THREAD_IA32                                        \
 295         INIT_THREAD_PM                                          \
 296         .dbr =          {0, },                                  \
 297         .ibr =          {0, },                                  \
 298         .fph =          {{{{0}}}, }                             \
 299 }
 300
 301 #define start_thread(regs,new_ip,new_sp) do {                                                   \
 302         set_fs(USER_DS);                                                                        \
 303         regs->cr_ipsr = ((regs->cr_ipsr | (IA64_PSR_BITS_TO_SET | IA64_PSR_CPL))                \
 304                          & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_RI | IA64_PSR_IS));              \
 305         regs->cr_iip = new_ip;                                                                  \
 306         regs->ar_rsc = 0xf;             /* eager mode, privilege level 3 */                     \
 307         regs->ar_rnat = 0;                                                                      \
 308         regs->ar_bspstore = current->thread.rbs_bot;                                            \
 309         regs->ar_fpsr = FPSR_DEFAULT;                                                           \
 310         regs->loadrs = 0;                                                                       \
 311         regs->r8 = current->mm->dumpable;       /* set "don't zap registers" flag */            \
 312         regs->r12 = new_sp - 16;        /* allocate 16 byte scratch area */                     \
 313         if (unlikely(!current->mm->dumpable)) {                                                 \
 314                 /*                                                                              \
 315                  * Zap scratch regs to avoid leaking bits between processes with different      \
 316                  * uid/privileges.                                                              \
 317                  */                                                                             \
 318                 regs->ar_pfs = 0; regs->b0 = 0; regs->pr = 0;                                   \
 319                 regs->r1 = 0; regs->r9  = 0; regs->r11 = 0; regs->r13 = 0; regs->r15 = 0;       \
 320         }                                                                                       \
 321 } while (0)
 322
 323 /* Forward declarations, a strange C thing... */
 324 struct mm_struct;
 325 struct task_struct;
 326
 327 /*
 328  * Free all resources held by a thread. This is called after the
 329  * parent of DEAD_TASK has collected the exit status of the task via
 330  * wait().
 331  */
 332 #define release_thread(dead_task)
 333
 334 /* Prepare to copy thread state - unlazy all lazy status */
 335 #define prepare_to_copy(tsk)    do { } while (0)
 336
 337 #ifdef CONFIG_NUMA
 338 #define SD_NODE_INIT (struct sched_domain) {            \
 339         .span                   = CPU_MASK_NONE,        \
 340         .parent                 = NULL,                 \
 341         .groups                 = NULL,                 \
 342         .min_interval           = 80,                   \
 343         .max_interval           = 320,                  \
 344         .busy_factor            = 320,                  \
 345         .imbalance_pct          = 125,                  \
 346         .cache_hot_time         = (10*1000000),         \
 347         .cache_nice_tries       = 1,                    \
 348         .per_cpu_gain           = 100,                  \
 349         .flags                  = SD_BALANCE_EXEC       \
 350                                 | SD_WAKE_BALANCE,      \
 351         .last_balance           = jiffies,              \
 352         .balance_interval       = 10,                   \
 353         .nr_balance_failed      = 0,                    \
 354 }
 355 #endif
 356
 357 /*
 358  * This is the mechanism for creating a new kernel thread.
 359  *
 360  * NOTE 1: Only a kernel-only process (ie the swapper or direct
 361  * descendants who haven't done an "execve()") should use this: it
 362  * will work within a system call from a "real" process, but the
 363  * process memory space will not be free'd until both the parent and
 364  * the child have exited.
 365  *
 366  * NOTE 2: This MUST NOT be an inlined function.  Otherwise, we get
 367  * into trouble in init/main.c when the child thread returns to
 368  * do_basic_setup() and the timing is such that free_initmem() has
 369  * been called already.
 370  */
 371 extern pid_t kernel_thread (int (*fn)(void *), void *arg, unsigned long flags);
 372
 373 /* Get wait channel for task P.  */
 374 extern unsigned long get_wchan (struct task_struct *p);
 375
 376 /* Return instruction pointer of blocked task TSK.  */
 377 #define KSTK_EIP(tsk)                                   \
 378   ({                                                    \
 379         struct pt_regs *_regs = ia64_task_regs(tsk);    \
 380         _regs->cr_iip + ia64_psr(_regs)->ri;            \
 381   })
 382
 383 /* Return stack pointer of blocked task TSK.  */
 384 #define KSTK_ESP(tsk)  ((tsk)->thread.ksp)
 385
 386 extern void ia64_getreg_unknown_kr (void);
 387 extern void ia64_setreg_unknown_kr (void);
 388
 389 #define ia64_get_kr(regnum)                                     \
 390 ({                                                              \
 391         unsigned long r = 0;                                    \
 392                                                                 \
 393         switch (regnum) {                                       \
 394             case 0: r = ia64_getreg(_IA64_REG_AR_KR0); break;   \
 395             case 1: r = ia64_getreg(_IA64_REG_AR_KR1); break;   \
 396             case 2: r = ia64_getreg(_IA64_REG_AR_KR2); break;   \
 397             case 3: r = ia64_getreg(_IA64_REG_AR_KR3); break;   \
 398             case 4: r = ia64_getreg(_IA64_REG_AR_KR4); break;   \
 399             case 5: r = ia64_getreg(_IA64_REG_AR_KR5); break;   \
 400             case 6: r = ia64_getreg(_IA64_REG_AR_KR6); break;   \
 401             case 7: r = ia64_getreg(_IA64_REG_AR_KR7); break;   \
 402             default: ia64_getreg_unknown_kr(); break;           \
 403         }                                                       \
 404         r;                                                      \
 405 })
 406
 407 #define ia64_set_kr(regnum, r)                                  \
 408 ({                                                              \
 409         switch (regnum) {                                       \
 410             case 0: ia64_setreg(_IA64_REG_AR_KR0, r); break;    \
 411             case 1: ia64_setreg(_IA64_REG_AR_KR1, r); break;    \
 412             case 2: ia64_setreg(_IA64_REG_AR_KR2, r); break;    \
 413             case 3: ia64_setreg(_IA64_REG_AR_KR3, r); break;    \
 414             case 4: ia64_setreg(_IA64_REG_AR_KR4, r); break;    \
 415             case 5: ia64_setreg(_IA64_REG_AR_KR5, r); break;    \
 416             case 6: ia64_setreg(_IA64_REG_AR_KR6, r); break;    \
 417             case 7: ia64_setreg(_IA64_REG_AR_KR7, r); break;    \
 418             default: ia64_setreg_unknown_kr(); break;           \
 419         }                                                       \
 420 })
 421
 422 /*
 423  * The following three macros can't be inline functions because we don't have struct
 424  * task_struct at this point.
 425  */
 426
 427 /* Return TRUE if task T owns the fph partition of the CPU we're running on. */
 428 #define ia64_is_local_fpu_owner(t)                                                              \
 429 ({                                                                                              \
 430         struct task_struct *__ia64_islfo_task = (t);                                            \
 431         (__ia64_islfo_task->thread.last_fph_cpu == smp_processor_id()                           \
 432          && __ia64_islfo_task == (struct task_struct *) ia64_get_kr(IA64_KR_FPU_OWNER));        \
 433 })
 434
 435 /* Mark task T as owning the fph partition of the CPU we're running on. */
 436 #define ia64_set_local_fpu_owner(t) do {                                                \
 437         struct task_struct *__ia64_slfo_task = (t);                                     \
 438         __ia64_slfo_task->thread.last_fph_cpu = smp_processor_id();                     \
 439         ia64_set_kr(IA64_KR_FPU_OWNER, (unsigned long) __ia64_slfo_task);               \
 440 } while (0)
 441
 442 /* Mark the fph partition of task T as being invalid on all CPUs.  */
 443 #define ia64_drop_fpu(t)        ((t)->thread.last_fph_cpu = -1)
 444
 445 extern void __ia64_init_fpu (void);
 446 extern void __ia64_save_fpu (struct ia64_fpreg *fph);
 447 extern void __ia64_load_fpu (struct ia64_fpreg *fph);
 448 extern void ia64_save_debug_regs (unsigned long *save_area);
 449 extern void ia64_load_debug_regs (unsigned long *save_area);
 450
 451 #ifdef CONFIG_IA32_SUPPORT
 452 extern void ia32_save_state (struct task_struct *task);
 453 extern void ia32_load_state (struct task_struct *task);
 454 #endif
 455
 456 #define ia64_fph_enable()       do { ia64_rsm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
 457 #define ia64_fph_disable()      do { ia64_ssm(IA64_PSR_DFH); ia64_srlz_d(); } while (0)
 458
 459 /* load fp 0.0 into fph */
 460 static inline void
 461 ia64_init_fpu (void) {
 462         ia64_fph_enable();
 463         __ia64_init_fpu();
 464         ia64_fph_disable();
 465 }
 466
 467 /* save f32-f127 at FPH */
 468 static inline void
 469 ia64_save_fpu (struct ia64_fpreg *fph) {
 470         ia64_fph_enable();
 471         __ia64_save_fpu(fph);
 472         ia64_fph_disable();
 473 }
 474
 475 /* load f32-f127 from FPH */
 476 static inline void
 477 ia64_load_fpu (struct ia64_fpreg *fph) {
 478         ia64_fph_enable();
 479         __ia64_load_fpu(fph);
 480         ia64_fph_disable();
 481 }
 482
 483 static inline __u64
 484 ia64_clear_ic (void)
 485 {
 486         __u64 psr;
 487         psr = ia64_getreg(_IA64_REG_PSR);
 488         ia64_stop();
 489         ia64_rsm(IA64_PSR_I | IA64_PSR_IC);
 490         ia64_srlz_i();
 491         return psr;
 492 }
 493
 494 /*
 495  * Restore the psr.
 496  */
 497 static inline void
 498 ia64_set_psr (__u64 psr)
 499 {
 500         ia64_stop();
 501         ia64_setreg(_IA64_REG_PSR_L, psr);
 502         ia64_srlz_d();
 503 }
 504
 505 /*
 506  * Insert a translation into an instruction and/or data translation
 507  * register.
 508  */
 509 static inline void
 510 ia64_itr (__u64 target_mask, __u64 tr_num,
 511           __u64 vmaddr, __u64 pte,
 512           __u64 log_page_size)
 513 {
 514         ia64_setreg(_IA64_REG_CR_ITIR, (log_page_size << 2));
 515         ia64_setreg(_IA64_REG_CR_IFA, vmaddr);
 516         ia64_stop();
 517         if (target_mask & 0x1)
 518                 ia64_itri(tr_num, pte);
 519         if (target_mask & 0x2)
 520                 ia64_itrd(tr_num, pte);
 521 }
 522
 523 /*
 524  * Insert a translation into the instruction and/or data translation
 525  * cache.
 526  */
 527 static inline void
 528 ia64_itc (__u64 target_mask, __u64 vmaddr, __u64 pte,
 529           __u64 log_page_size)
 530 {
 531         ia64_setreg(_IA64_REG_CR_ITIR, (log_page_size << 2));
 532         ia64_setreg(_IA64_REG_CR_IFA, vmaddr);
 533         ia64_stop();
 534         /* as per EAS2.6, itc must be the last instruction in an instruction group */
 535         if (target_mask & 0x1)
 536                 ia64_itci(pte);
 537         if (target_mask & 0x2)
 538                 ia64_itcd(pte);
 539 }
 540
 541 /*
 542  * Purge a range of addresses from instruction and/or data translation
 543  * register(s).
 544  */
 545 static inline void
 546 ia64_ptr (__u64 target_mask, __u64 vmaddr, __u64 log_size)
 547 {
 548         if (target_mask & 0x1)
 549                 ia64_ptri(vmaddr, (log_size << 2));
 550         if (target_mask & 0x2)
 551                 ia64_ptrd(vmaddr, (log_size << 2));
 552 }
 553
 554 /* Set the interrupt vector address.  The address must be suitably aligned (32KB).  */
 555 static inline void
 556 ia64_set_iva (void *ivt_addr)
 557 {
 558         ia64_setreg(_IA64_REG_CR_IVA, (__u64) ivt_addr);
 559         ia64_srlz_i();
 560 }
 561
 562 /* Set the page table address and control bits.  */
 563 static inline void
 564 ia64_set_pta (__u64 pta)
 565 {
 566         /* Note: srlz.i implies srlz.d */
 567         ia64_setreg(_IA64_REG_CR_PTA, pta);
 568         ia64_srlz_i();
 569 }
 570
 571 static inline void
 572 ia64_eoi (void)
 573 {
 574         ia64_setreg(_IA64_REG_CR_EOI, 0);
 575         ia64_srlz_d();
 576 }
 577
 578 #define cpu_relax()     ia64_hint(ia64_hint_pause)
 579
 580 static inline void
 581 ia64_set_lrr0 (unsigned long val)
 582 {
 583         ia64_setreg(_IA64_REG_CR_LRR0, val);
 584         ia64_srlz_d();
 585 }
 586
 587 static inline void
 588 ia64_set_lrr1 (unsigned long val)
 589 {
 590         ia64_setreg(_IA64_REG_CR_LRR1, val);
 591         ia64_srlz_d();
 592 }
 593
 594
 595 /*
 596  * Given the address to which a spill occurred, return the unat bit
 597  * number that corresponds to this address.
 598  */
 599 static inline __u64
 600 ia64_unat_pos (void *spill_addr)
 601 {
 602         return ((__u64) spill_addr >> 3) & 0x3f;
 603 }
 604
 605 /*
 606  * Set the NaT bit of an integer register which was spilled at address
 607  * SPILL_ADDR.  UNAT is the mask to be updated.
 608  */
 609 static inline void
 610 ia64_set_unat (__u64 *unat, void *spill_addr, unsigned long nat)
 611 {
 612         __u64 bit = ia64_unat_pos(spill_addr);
 613         __u64 mask = 1UL << bit;
 614
 615         *unat = (*unat & ~mask) | (nat << bit);
 616 }
 617
 618 /*
 619  * Return saved PC of a blocked thread.
 620  * Note that the only way T can block is through a call to schedule() -> switch_to().
 621  */
 622 static inline unsigned long
 623 thread_saved_pc (struct task_struct *t)
 624 {
 625         struct unw_frame_info info;
 626         unsigned long ip;
 627
 628         unw_init_from_blocked_task(&info, t);
 629         if (unw_unwind(&info) < 0)
 630                 return 0;
 631         unw_get_ip(&info, &ip);
 632         return ip;
 633 }
 634
 635 /*
 636  * Get the current instruction/program counter value.
 637  */
 638 #define current_text_addr() \
 639         ({ void *_pc; _pc = (void *)ia64_getreg(_IA64_REG_IP); _pc; })
 640
 641 static inline __u64
 642 ia64_get_ivr (void)
 643 {
 644         __u64 r;
 645         ia64_srlz_d();
 646         r = ia64_getreg(_IA64_REG_CR_IVR);
 647         ia64_srlz_d();
 648         return r;
 649 }
 650
 651 static inline void
 652 ia64_set_dbr (__u64 regnum, __u64 value)
 653 {
 654         __ia64_set_dbr(regnum, value);
 655 #ifdef CONFIG_ITANIUM
 656         ia64_srlz_d();
 657 #endif
 658 }
 659
 660 static inline __u64
 661 ia64_get_dbr (__u64 regnum)
 662 {
 663         __u64 retval;
 664
 665         retval = __ia64_get_dbr(regnum);
 666 #ifdef CONFIG_ITANIUM
 667         ia64_srlz_d();
 668 #endif
 669         return retval;
 670 }
 671
 672 static inline __u64
 673 ia64_rotr (__u64 w, __u64 n)
 674 {
 675         return (w >> n) | (w << (64 - n));
 676 }
 677
 678 #define ia64_rotl(w,n)  ia64_rotr((w), (64) - (n))
 679
 680 /*
 681  * Take a mapped kernel address and return the equivalent address
 682  * in the region 7 identity mapped virtual area.
 683  */
 684 static inline void *
 685 ia64_imva (void *addr)
 686 {
 687         void *result;
 688         result = (void *) ia64_tpa(addr);
 689         return __va(result);
 690 }
 691
 692 #define ARCH_HAS_PREFETCH
 693 #define ARCH_HAS_PREFETCHW
 694 #define ARCH_HAS_SPINLOCK_PREFETCH
 695 #define PREFETCH_STRIDE                 L1_CACHE_BYTES
 696
 697 static inline void
 698 prefetch (const void *x)
 699 {
 700          ia64_lfetch(ia64_lfhint_none, x);
 701 }
 702
 703 static inline void
 704 prefetchw (const void *x)
 705 {
 706         ia64_lfetch_excl(ia64_lfhint_none, x);
 707 }
 708
 709 #define spin_lock_prefetch(x)   prefetchw(x)
 710
 711 #endif /* !__ASSEMBLY__ */
 712
 713 #endif /* _ASM_IA64_PROCESSOR_H */