arch/x86/include/asm/paravirt.h

   1 #ifndef _ASM_X86_PARAVIRT_H
   2 #define _ASM_X86_PARAVIRT_H
   3 /* Various instructions on x86 need to be replaced for
   4  * para-virtualization: those hooks are defined here. */
   5
   6 #ifdef CONFIG_PARAVIRT
   7 #include <asm/pgtable_types.h>
   8 #include <asm/asm.h>
   9
  10 /* Bitmask of what can be clobbered: usually at least eax. */
  11 #define CLBR_NONE 0
  12 #define CLBR_EAX  (1 << 0)
  13 #define CLBR_ECX  (1 << 1)
  14 #define CLBR_EDX  (1 << 2)
  15 #define CLBR_EDI  (1 << 3)
  16
  17 #ifdef CONFIG_X86_32
  18 /* CLBR_ANY should match all regs platform has. For i386, that's just it */
  19 #define CLBR_ANY  ((1 << 4) - 1)
  20
  21 #define CLBR_ARG_REGS   (CLBR_EAX | CLBR_EDX | CLBR_ECX)
  22 #define CLBR_RET_REG    (CLBR_EAX | CLBR_EDX)
  23 #define CLBR_SCRATCH    (0)
  24 #else
  25 #define CLBR_RAX  CLBR_EAX
  26 #define CLBR_RCX  CLBR_ECX
  27 #define CLBR_RDX  CLBR_EDX
  28 #define CLBR_RDI  CLBR_EDI
  29 #define CLBR_RSI  (1 << 4)
  30 #define CLBR_R8   (1 << 5)
  31 #define CLBR_R9   (1 << 6)
  32 #define CLBR_R10  (1 << 7)
  33 #define CLBR_R11  (1 << 8)
  34
  35 #define CLBR_ANY  ((1 << 9) - 1)
  36
  37 #define CLBR_ARG_REGS   (CLBR_RDI | CLBR_RSI | CLBR_RDX | \
  38                          CLBR_RCX | CLBR_R8 | CLBR_R9)
  39 #define CLBR_RET_REG    (CLBR_RAX)
  40 #define CLBR_SCRATCH    (CLBR_R10 | CLBR_R11)
  41
  42 #include <asm/desc_defs.h>
  43 #endif /* X86_64 */
  44
  45 #define CLBR_CALLEE_SAVE ((CLBR_ARG_REGS | CLBR_SCRATCH) & ~CLBR_RET_REG)
  46
  47 #ifndef __ASSEMBLY__
  48 #include <linux/types.h>
  49 #include <linux/cpumask.h>
  50 #include <asm/kmap_types.h>
  51 #include <asm/desc_defs.h>
  52
  53 struct page;
  54 struct thread_struct;
  55 struct desc_ptr;
  56 struct tss_struct;
  57 struct mm_struct;
  58 struct desc_struct;
  59
  60 /*
  61  * Wrapper type for pointers to code which uses the non-standard
  62  * calling convention.  See PV_CALL_SAVE_REGS_THUNK below.
  63  */
  64 struct paravirt_callee_save {
  65         void *func;
  66 };
  67
  68 /* general info */
  69 struct pv_info {
  70         unsigned int kernel_rpl;
  71         int shared_kernel_pmd;
  72         int paravirt_enabled;
  73         const char *name;
  74 };
  75
  76 struct pv_init_ops {
  77         /*
  78          * Patch may replace one of the defined code sequences with
  79          * arbitrary code, subject to the same register constraints.
  80          * This generally means the code is not free to clobber any
  81          * registers other than EAX.  The patch function should return
  82          * the number of bytes of code generated, as we nop pad the
  83          * rest in generic code.
  84          */
  85         unsigned (*patch)(u8 type, u16 clobber, void *insnbuf,
  86                           unsigned long addr, unsigned len);
  87
  88         /* Basic arch-specific setup */
  89         void (*arch_setup)(void);
  90         char *(*memory_setup)(void);
  91         void (*post_allocator_init)(void);
  92
  93         /* Print a banner to identify the environment */
  94         void (*banner)(void);
  95 };
  96
  97
  98 struct pv_lazy_ops {
  99         /* Set deferred update mode, used for batching operations. */
 100         void (*enter)(void);
 101         void (*leave)(void);
 102 };
 103
 104 struct pv_time_ops {
 105         void (*time_init)(void);
 106
 107         /* Set and set time of day */
 108         unsigned long (*get_wallclock)(void);
 109         int (*set_wallclock)(unsigned long);
 110
 111         unsigned long long (*sched_clock)(void);
 112         unsigned long (*get_tsc_khz)(void);
 113 };
 114
 115 struct pv_cpu_ops {
 116         /* hooks for various privileged instructions */
 117         unsigned long (*get_debugreg)(int regno);
 118         void (*set_debugreg)(int regno, unsigned long value);
 119
 120         void (*clts)(void);
 121
 122         unsigned long (*read_cr0)(void);
 123         void (*write_cr0)(unsigned long);
 124
 125         unsigned long (*read_cr4_safe)(void);
 126         unsigned long (*read_cr4)(void);
 127         void (*write_cr4)(unsigned long);
 128
 129 #ifdef CONFIG_X86_64
 130         unsigned long (*read_cr8)(void);
 131         void (*write_cr8)(unsigned long);
 132 #endif
 133
 134         /* Segment descriptor handling */
 135         void (*load_tr_desc)(void);
 136         void (*load_gdt)(const struct desc_ptr *);
 137         void (*load_idt)(const struct desc_ptr *);
 138         void (*store_gdt)(struct desc_ptr *);
 139         void (*store_idt)(struct desc_ptr *);
 140         void (*set_ldt)(const void *desc, unsigned entries);
 141         unsigned long (*store_tr)(void);
 142         void (*load_tls)(struct thread_struct *t, unsigned int cpu);
 143 #ifdef CONFIG_X86_64
 144         void (*load_gs_index)(unsigned int idx);
 145 #endif
 146         void (*write_ldt_entry)(struct desc_struct *ldt, int entrynum,
 147                                 const void *desc);
 148         void (*write_gdt_entry)(struct desc_struct *,
 149                                 int entrynum, const void *desc, int size);
 150         void (*write_idt_entry)(gate_desc *,
 151                                 int entrynum, const gate_desc *gate);
 152         void (*alloc_ldt)(struct desc_struct *ldt, unsigned entries);
 153         void (*free_ldt)(struct desc_struct *ldt, unsigned entries);
 154
 155         void (*load_sp0)(struct tss_struct *tss, struct thread_struct *t);
 156
 157         void (*set_iopl_mask)(unsigned mask);
 158
 159         void (*wbinvd)(void);
 160         void (*io_delay)(void);
 161
 162         /* cpuid emulation, mostly so that caps bits can be disabled */
 163         void (*cpuid)(unsigned int *eax, unsigned int *ebx,
 164                       unsigned int *ecx, unsigned int *edx);
 165
 166         /* MSR, PMC and TSR operations.
 167            err = 0/-EFAULT.  wrmsr returns 0/-EFAULT. */
 168         u64 (*read_msr_amd)(unsigned int msr, int *err);
 169         u64 (*read_msr)(unsigned int msr, int *err);
 170         int (*write_msr)(unsigned int msr, unsigned low, unsigned high);
 171
 172         u64 (*read_tsc)(void);
 173         u64 (*read_pmc)(int counter);
 174         unsigned long long (*read_tscp)(unsigned int *aux);
 175
 176         /*
 177          * Atomically enable interrupts and return to userspace.  This
 178          * is only ever used to return to 32-bit processes; in a
 179          * 64-bit kernel, it's used for 32-on-64 compat processes, but
 180          * never native 64-bit processes.  (Jump, not call.)
 181          */
 182         void (*irq_enable_sysexit)(void);
 183
 184         /*
 185          * Switch to usermode gs and return to 64-bit usermode using
 186          * sysret.  Only used in 64-bit kernels to return to 64-bit
 187          * processes.  Usermode register state, including %rsp, must
 188          * already be restored.
 189          */
 190         void (*usergs_sysret64)(void);
 191
 192         /*
 193          * Switch to usermode gs and return to 32-bit usermode using
 194          * sysret.  Used to return to 32-on-64 compat processes.
 195          * Other usermode register state, including %esp, must already
 196          * be restored.
 197          */
 198         void (*usergs_sysret32)(void);
 199
 200         /* Normal iret.  Jump to this with the standard iret stack
 201            frame set up. */
 202         void (*iret)(void);
 203
 204         void (*swapgs)(void);
 205
 206         struct pv_lazy_ops lazy_mode;
 207 };
 208
 209 struct pv_irq_ops {
 210         void (*init_IRQ)(void);
 211
 212         /*
 213          * Get/set interrupt state.  save_fl and restore_fl are only
 214          * expected to use X86_EFLAGS_IF; all other bits
 215          * returned from save_fl are undefined, and may be ignored by
 216          * restore_fl.
 217          *
 218          * NOTE: These functions callers expect the callee to preserve
 219          * more registers than the standard C calling convention.
 220          */
 221         struct paravirt_callee_save save_fl;
 222         struct paravirt_callee_save restore_fl;
 223         struct paravirt_callee_save irq_disable;
 224         struct paravirt_callee_save irq_enable;
 225
 226         void (*safe_halt)(void);
 227         void (*halt)(void);
 228
 229 #ifdef CONFIG_X86_64
 230         void (*adjust_exception_frame)(void);
 231 #endif
 232 };
 233
 234 struct pv_apic_ops {
 235 #ifdef CONFIG_X86_LOCAL_APIC
 236         void (*setup_boot_clock)(void);
 237         void (*setup_secondary_clock)(void);
 238
 239         void (*startup_ipi_hook)(int phys_apicid,
 240                                  unsigned long start_eip,
 241                                  unsigned long start_esp);
 242 #endif
 243 };
 244
 245 struct pv_mmu_ops {
 246         /*
 247          * Called before/after init_mm pagetable setup. setup_start
 248          * may reset %cr3, and may pre-install parts of the pagetable;
 249          * pagetable setup is expected to preserve any existing
 250          * mapping.
 251          */
 252         void (*pagetable_setup_start)(pgd_t *pgd_base);
 253         void (*pagetable_setup_done)(pgd_t *pgd_base);
 254
 255         unsigned long (*read_cr2)(void);
 256         void (*write_cr2)(unsigned long);
 257
 258         unsigned long (*read_cr3)(void);
 259         void (*write_cr3)(unsigned long);
 260
 261         /*
 262          * Hooks for intercepting the creation/use/destruction of an
 263          * mm_struct.
 264          */
 265         void (*activate_mm)(struct mm_struct *prev,
 266                             struct mm_struct *next);
 267         void (*dup_mmap)(struct mm_struct *oldmm,
 268                          struct mm_struct *mm);
 269         void (*exit_mmap)(struct mm_struct *mm);
 270
 271
 272         /* TLB operations */
 273         void (*flush_tlb_user)(void);
 274         void (*flush_tlb_kernel)(void);
 275         void (*flush_tlb_single)(unsigned long addr);
 276         void (*flush_tlb_others)(const struct cpumask *cpus,
 277                                  struct mm_struct *mm,
 278                                  unsigned long va);
 279
 280         /* Hooks for allocating and freeing a pagetable top-level */
 281         int  (*pgd_alloc)(struct mm_struct *mm);
 282         void (*pgd_free)(struct mm_struct *mm, pgd_t *pgd);
 283
 284         /*
 285          * Hooks for allocating/releasing pagetable pages when they're
 286          * attached to a pagetable
 287          */
 288         void (*alloc_pte)(struct mm_struct *mm, unsigned long pfn);
 289         void (*alloc_pmd)(struct mm_struct *mm, unsigned long pfn);
 290         void (*alloc_pmd_clone)(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count);
 291         void (*alloc_pud)(struct mm_struct *mm, unsigned long pfn);
 292         void (*release_pte)(unsigned long pfn);
 293         void (*release_pmd)(unsigned long pfn);
 294         void (*release_pud)(unsigned long pfn);
 295
 296         /* Pagetable manipulation functions */
 297         void (*set_pte)(pte_t *ptep, pte_t pteval);
 298         void (*set_pte_at)(struct mm_struct *mm, unsigned long addr,
 299                            pte_t *ptep, pte_t pteval);
 300         void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval);
 301         void (*pte_update)(struct mm_struct *mm, unsigned long addr,
 302                            pte_t *ptep);
 303         void (*pte_update_defer)(struct mm_struct *mm,
 304                                  unsigned long addr, pte_t *ptep);
 305
 306         pte_t (*ptep_modify_prot_start)(struct mm_struct *mm, unsigned long addr,
 307                                         pte_t *ptep);
 308         void (*ptep_modify_prot_commit)(struct mm_struct *mm, unsigned long addr,
 309                                         pte_t *ptep, pte_t pte);
 310
 311         struct paravirt_callee_save pte_val;
 312         struct paravirt_callee_save make_pte;
 313
 314         struct paravirt_callee_save pgd_val;
 315         struct paravirt_callee_save make_pgd;
 316
 317 #if PAGETABLE_LEVELS >= 3
 318 #ifdef CONFIG_X86_PAE
 319         void (*set_pte_atomic)(pte_t *ptep, pte_t pteval);
 320         void (*pte_clear)(struct mm_struct *mm, unsigned long addr,
 321                           pte_t *ptep);
 322         void (*pmd_clear)(pmd_t *pmdp);
 323
 324 #endif  /* CONFIG_X86_PAE */
 325
 326         void (*set_pud)(pud_t *pudp, pud_t pudval);
 327
 328         struct paravirt_callee_save pmd_val;
 329         struct paravirt_callee_save make_pmd;
 330
 331 #if PAGETABLE_LEVELS == 4
 332         struct paravirt_callee_save pud_val;
 333         struct paravirt_callee_save make_pud;
 334
 335         void (*set_pgd)(pgd_t *pudp, pgd_t pgdval);
 336 #endif  /* PAGETABLE_LEVELS == 4 */
 337 #endif  /* PAGETABLE_LEVELS >= 3 */
 338
 339 #ifdef CONFIG_HIGHPTE
 340         void *(*kmap_atomic_pte)(struct page *page, enum km_type type);
 341 #endif
 342
 343         struct pv_lazy_ops lazy_mode;
 344
 345         /* dom0 ops */
 346
 347         /* Sometimes the physical address is a pfn, and sometimes its
 348            an mfn.  We can tell which is which from the index. */
 349         void (*set_fixmap)(unsigned /* enum fixed_addresses */ idx,
 350                            phys_addr_t phys, pgprot_t flags);
 351 };
 352
 353 struct raw_spinlock;
 354 struct pv_lock_ops {
 355         int (*spin_is_locked)(struct raw_spinlock *lock);
 356         int (*spin_is_contended)(struct raw_spinlock *lock);
 357         void (*spin_lock)(struct raw_spinlock *lock);
 358         void (*spin_lock_flags)(struct raw_spinlock *lock, unsigned long flags);
 359         int (*spin_trylock)(struct raw_spinlock *lock);
 360         void (*spin_unlock)(struct raw_spinlock *lock);
 361 };
 362
 363 /* This contains all the paravirt structures: we get a convenient
 364  * number for each function using the offset which we use to indicate
 365  * what to patch. */
 366 struct paravirt_patch_template {
 367         struct pv_init_ops pv_init_ops;
 368         struct pv_time_ops pv_time_ops;
 369         struct pv_cpu_ops pv_cpu_ops;
 370         struct pv_irq_ops pv_irq_ops;
 371         struct pv_apic_ops pv_apic_ops;
 372         struct pv_mmu_ops pv_mmu_ops;
 373         struct pv_lock_ops pv_lock_ops;
 374 };
 375
 376 extern struct pv_info pv_info;
 377 extern struct pv_init_ops pv_init_ops;
 378 extern struct pv_time_ops pv_time_ops;
 379 extern struct pv_cpu_ops pv_cpu_ops;
 380 extern struct pv_irq_ops pv_irq_ops;
 381 extern struct pv_apic_ops pv_apic_ops;
 382 extern struct pv_mmu_ops pv_mmu_ops;
 383 extern struct pv_lock_ops pv_lock_ops;
 384
 385 #define PARAVIRT_PATCH(x)                                       \
 386         (offsetof(struct paravirt_patch_template, x) / sizeof(void *))
 387
 388 #define paravirt_type(op)                               \
 389         [paravirt_typenum] "i" (PARAVIRT_PATCH(op)),    \
 390         [paravirt_opptr] "i" (&(op))
 391 #define paravirt_clobber(clobber)               \
 392         [paravirt_clobber] "i" (clobber)
 393
 394 /*
 395  * Generate some code, and mark it as patchable by the
 396  * apply_paravirt() alternate instruction patcher.
 397  */
 398 #define _paravirt_alt(insn_string, type, clobber)       \
 399         "771:\n\t" insn_string "\n" "772:\n"            \
 400         ".pushsection .parainstructions,\"a\"\n"        \
 401         _ASM_ALIGN "\n"                                 \
 402         _ASM_PTR " 771b\n"                              \
 403         "  .byte " type "\n"                            \
 404         "  .byte 772b-771b\n"                           \
 405         "  .short " clobber "\n"                        \
 406         ".popsection\n"
 407
 408 /* Generate patchable code, with the default asm parameters. */
 409 #define paravirt_alt(insn_string)                                       \
 410         _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
 411
 412 /* Simple instruction patching code. */
 413 #define DEF_NATIVE(ops, name, code)                                     \
 414         extern const char start_##ops##_##name[], end_##ops##_##name[]; \
 415         asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
 416
 417 unsigned paravirt_patch_nop(void);
 418 unsigned paravirt_patch_ident_32(void *insnbuf, unsigned len);
 419 unsigned paravirt_patch_ident_64(void *insnbuf, unsigned len);
 420 unsigned paravirt_patch_ignore(unsigned len);
 421 unsigned paravirt_patch_call(void *insnbuf,
 422                              const void *target, u16 tgt_clobbers,
 423                              unsigned long addr, u16 site_clobbers,
 424                              unsigned len);
 425 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
 426                             unsigned long addr, unsigned len);
 427 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
 428                                 unsigned long addr, unsigned len);
 429
 430 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
 431                               const char *start, const char *end);
 432
 433 unsigned native_patch(u8 type, u16 clobbers, void *ibuf,
 434                       unsigned long addr, unsigned len);
 435
 436 int paravirt_disable_iospace(void);
 437
 438 /*
 439  * This generates an indirect call based on the operation type number.
 440  * The type number, computed in PARAVIRT_PATCH, is derived from the
 441  * offset into the paravirt_patch_template structure, and can therefore be
 442  * freely converted back into a structure offset.
 443  */
 444 #define PARAVIRT_CALL   "call *%c[paravirt_opptr];"
 445
 446 /*
 447  * These macros are intended to wrap calls through one of the paravirt
 448  * ops structs, so that they can be later identified and patched at
 449  * runtime.
 450  *
 451  * Normally, a call to a pv_op function is a simple indirect call:
 452  * (pv_op_struct.operations)(args...).
 453  *
 454  * Unfortunately, this is a relatively slow operation for modern CPUs,
 455  * because it cannot necessarily determine what the destination
 456  * address is.  In this case, the address is a runtime constant, so at
 457  * the very least we can patch the call to e a simple direct call, or
 458  * ideally, patch an inline implementation into the callsite.  (Direct
 459  * calls are essentially free, because the call and return addresses
 460  * are completely predictable.)
 461  *
 462  * For i386, these macros rely on the standard gcc "regparm(3)" calling
 463  * convention, in which the first three arguments are placed in %eax,
 464  * %edx, %ecx (in that order), and the remaining arguments are placed
 465  * on the stack.  All caller-save registers (eax,edx,ecx) are expected
 466  * to be modified (either clobbered or used for return values).
 467  * X86_64, on the other hand, already specifies a register-based calling
 468  * conventions, returning at %rax, with parameteres going on %rdi, %rsi,
 469  * %rdx, and %rcx. Note that for this reason, x86_64 does not need any
 470  * special handling for dealing with 4 arguments, unlike i386.
 471  * However, x86_64 also have to clobber all caller saved registers, which
 472  * unfortunately, are quite a bit (r8 - r11)
 473  *
 474  * The call instruction itself is marked by placing its start address
 475  * and size into the .parainstructions section, so that
 476  * apply_paravirt() in arch/i386/kernel/alternative.c can do the
 477  * appropriate patching under the control of the backend pv_init_ops
 478  * implementation.
 479  *
 480  * Unfortunately there's no way to get gcc to generate the args setup
 481  * for the call, and then allow the call itself to be generated by an
 482  * inline asm.  Because of this, we must do the complete arg setup and
 483  * return value handling from within these macros.  This is fairly
 484  * cumbersome.
 485  *
 486  * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
 487  * It could be extended to more arguments, but there would be little
 488  * to be gained from that.  For each number of arguments, there are
 489  * the two VCALL and CALL variants for void and non-void functions.
 490  *
 491  * When there is a return value, the invoker of the macro must specify
 492  * the return type.  The macro then uses sizeof() on that type to
 493  * determine whether its a 32 or 64 bit value, and places the return
 494  * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
 495  * 64-bit). For x86_64 machines, it just returns at %rax regardless of
 496  * the return value size.
 497  *
 498  * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
 499  * i386 also passes 64-bit arguments as a pair of adjacent 32-bit arguments
 500  * in low,high order
 501  *
 502  * Small structures are passed and returned in registers.  The macro
 503  * calling convention can't directly deal with this, so the wrapper
 504  * functions must do this.
 505  *
 506  * These PVOP_* macros are only defined within this header.  This
 507  * means that all uses must be wrapped in inline functions.  This also
 508  * makes sure the incoming and outgoing types are always correct.
 509  */
 510 #ifdef CONFIG_X86_32
 511 #define PVOP_VCALL_ARGS                         \
 512         unsigned long __eax = __eax, __edx = __edx, __ecx = __ecx
 513 #define PVOP_CALL_ARGS                  PVOP_VCALL_ARGS
 514
 515 #define PVOP_CALL_ARG1(x)               "a" ((unsigned long)(x))
 516 #define PVOP_CALL_ARG2(x)               "d" ((unsigned long)(x))
 517 #define PVOP_CALL_ARG3(x)               "c" ((unsigned long)(x))
 518
 519 #define PVOP_VCALL_CLOBBERS             "=a" (__eax), "=d" (__edx),     \
 520                                         "=c" (__ecx)
 521 #define PVOP_CALL_CLOBBERS              PVOP_VCALL_CLOBBERS
 522
 523 #define PVOP_VCALLEE_CLOBBERS           "=a" (__eax), "=d" (__edx)
 524 #define PVOP_CALLEE_CLOBBERS            PVOP_VCALLEE_CLOBBERS
 525
 526 #define EXTRA_CLOBBERS
 527 #define VEXTRA_CLOBBERS
 528 #else  /* CONFIG_X86_64 */
 529 #define PVOP_VCALL_ARGS                                 \
 530         unsigned long __edi = __edi, __esi = __esi,     \
 531                 __edx = __edx, __ecx = __ecx
 532 #define PVOP_CALL_ARGS          PVOP_VCALL_ARGS, __eax
 533
 534 #define PVOP_CALL_ARG1(x)               "D" ((unsigned long)(x))
 535 #define PVOP_CALL_ARG2(x)               "S" ((unsigned long)(x))
 536 #define PVOP_CALL_ARG3(x)               "d" ((unsigned long)(x))
 537 #define PVOP_CALL_ARG4(x)               "c" ((unsigned long)(x))
 538
 539 #define PVOP_VCALL_CLOBBERS     "=D" (__edi),                           \
 540                                 "=S" (__esi), "=d" (__edx),             \
 541                                 "=c" (__ecx)
 542 #define PVOP_CALL_CLOBBERS      PVOP_VCALL_CLOBBERS, "=a" (__eax)
 543
 544 #define PVOP_VCALLEE_CLOBBERS   "=a" (__eax)
 545 #define PVOP_CALLEE_CLOBBERS    PVOP_VCALLEE_CLOBBERS
 546
 547 #define EXTRA_CLOBBERS   , "r8", "r9", "r10", "r11"
 548 #define VEXTRA_CLOBBERS  , "rax", "r8", "r9", "r10", "r11"
 549 #endif  /* CONFIG_X86_32 */
 550
 551 #ifdef CONFIG_PARAVIRT_DEBUG
 552 #define PVOP_TEST_NULL(op)      BUG_ON(op == NULL)
 553 #else
 554 #define PVOP_TEST_NULL(op)      ((void)op)
 555 #endif
 556
 557 #define ____PVOP_CALL(rettype, op, clbr, call_clbr, extra_clbr,         \
 558                       pre, post, ...)                                   \
 559         ({                                                              \
 560                 rettype __ret;                                          \
 561                 PVOP_CALL_ARGS;                                         \
 562                 PVOP_TEST_NULL(op);                                     \
 563                 /* This is 32-bit specific, but is okay in 64-bit */    \
 564                 /* since this condition will never hold */              \
 565                 if (sizeof(rettype) > sizeof(unsigned long)) {          \
 566                         asm volatile(pre                                \
 567                                      paravirt_alt(PARAVIRT_CALL)        \
 568                                      post                               \
 569                                      : call_clbr                        \
 570                                      : paravirt_type(op),               \
 571                                        paravirt_clobber(clbr),          \
 572                                        ##__VA_ARGS__                    \
 573                                      : "memory", "cc" extra_clbr);      \
 574                         __ret = (rettype)((((u64)__edx) << 32) | __eax); \
 575                 } else {                                                \
 576                         asm volatile(pre                                \
 577                                      paravirt_alt(PARAVIRT_CALL)        \
 578                                      post                               \
 579                                      : call_clbr                        \
 580                                      : paravirt_type(op),               \
 581                                        paravirt_clobber(clbr),          \
 582                                        ##__VA_ARGS__                    \
 583                                      : "memory", "cc" extra_clbr);      \
 584                         __ret = (rettype)__eax;                         \
 585                 }                                                       \
 586                 __ret;                                                  \
 587         })
 588
 589 #define __PVOP_CALL(rettype, op, pre, post, ...)                        \
 590         ____PVOP_CALL(rettype, op, CLBR_ANY, PVOP_CALL_CLOBBERS,        \
 591                       EXTRA_CLOBBERS, pre, post, ##__VA_ARGS__)
 592
 593 #define __PVOP_CALLEESAVE(rettype, op, pre, post, ...)                  \
 594         ____PVOP_CALL(rettype, op.func, CLBR_RET_REG,                   \
 595                       PVOP_CALLEE_CLOBBERS, ,                           \
 596                       pre, post, ##__VA_ARGS__)
 597
 598
 599 #define ____PVOP_VCALL(op, clbr, call_clbr, extra_clbr, pre, post, ...) \
 600         ({                                                              \
 601                 PVOP_VCALL_ARGS;                                        \
 602                 PVOP_TEST_NULL(op);                                     \
 603                 asm volatile(pre                                        \
 604                              paravirt_alt(PARAVIRT_CALL)                \
 605                              post                                       \
 606                              : call_clbr                                \
 607                              : paravirt_type(op),                       \
 608                                paravirt_clobber(clbr),                  \
 609                                ##__VA_ARGS__                            \
 610                              : "memory", "cc" extra_clbr);              \
 611         })
 612
 613 #define __PVOP_VCALL(op, pre, post, ...)                                \
 614         ____PVOP_VCALL(op, CLBR_ANY, PVOP_VCALL_CLOBBERS,               \
 615                        VEXTRA_CLOBBERS,                                 \
 616                        pre, post, ##__VA_ARGS__)
 617
 618 #define __PVOP_VCALLEESAVE(rettype, op, pre, post, ...)                 \
 619         ____PVOP_CALL(rettype, op.func, CLBR_RET_REG,                   \
 620                       PVOP_VCALLEE_CLOBBERS, ,                          \
 621                       pre, post, ##__VA_ARGS__)
 622
 623
 624
 625 #define PVOP_CALL0(rettype, op)                                         \
 626         __PVOP_CALL(rettype, op, "", "")
 627 #define PVOP_VCALL0(op)                                                 \
 628         __PVOP_VCALL(op, "", "")
 629
 630 #define PVOP_CALLEE0(rettype, op)                                       \
 631         __PVOP_CALLEESAVE(rettype, op, "", "")
 632 #define PVOP_VCALLEE0(op)                                               \
 633         __PVOP_VCALLEESAVE(op, "", "")
 634
 635
 636 #define PVOP_CALL1(rettype, op, arg1)                                   \
 637         __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1))
 638 #define PVOP_VCALL1(op, arg1)                                           \
 639         __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1))
 640
 641 #define PVOP_CALLEE1(rettype, op, arg1)                                 \
 642         __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1))
 643 #define PVOP_VCALLEE1(op, arg1)                                         \
 644         __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1))
 645
 646
 647 #define PVOP_CALL2(rettype, op, arg1, arg2)                             \
 648         __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1),          \
 649                     PVOP_CALL_ARG2(arg2))
 650 #define PVOP_VCALL2(op, arg1, arg2)                                     \
 651         __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1),                  \
 652                      PVOP_CALL_ARG2(arg2))
 653
 654 #define PVOP_CALLEE2(rettype, op, arg1, arg2)                           \
 655         __PVOP_CALLEESAVE(rettype, op, "", "", PVOP_CALL_ARG1(arg1),    \
 656                           PVOP_CALL_ARG2(arg2))
 657 #define PVOP_VCALLEE2(op, arg1, arg2)                                   \
 658         __PVOP_VCALLEESAVE(op, "", "", PVOP_CALL_ARG1(arg1),            \
 659                            PVOP_CALL_ARG2(arg2))
 660
 661
 662 #define PVOP_CALL3(rettype, op, arg1, arg2, arg3)                       \
 663         __PVOP_CALL(rettype, op, "", "", PVOP_CALL_ARG1(arg1),          \
 664                     PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
 665 #define PVOP_VCALL3(op, arg1, arg2, arg3)                               \
 666         __PVOP_VCALL(op, "", "", PVOP_CALL_ARG1(arg1),                  \
 667                      PVOP_CALL_ARG2(arg2), PVOP_CALL_ARG3(arg3))
 668
 669 /* This is the only difference in x86_64. We can make it much simpler */
 670 #ifdef CONFIG_X86_32
 671 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4)                 \
 672         __PVOP_CALL(rettype, op,                                        \
 673                     "push %[_arg4];", "lea 4(%%esp),%%esp;",            \
 674                     PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),         \
 675                     PVOP_CALL_ARG3(arg3), [_arg4] "mr" ((u32)(arg4)))
 676 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4)                         \
 677         __PVOP_VCALL(op,                                                \
 678                     "push %[_arg4];", "lea 4(%%esp),%%esp;",            \
 679                     "0" ((u32)(arg1)), "1" ((u32)(arg2)),               \
 680                     "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
 681 #else
 682 #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4)                 \
 683         __PVOP_CALL(rettype, op, "", "",                                \
 684                     PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),         \
 685                     PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
 686 #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4)                         \
 687         __PVOP_VCALL(op, "", "",                                        \
 688                      PVOP_CALL_ARG1(arg1), PVOP_CALL_ARG2(arg2),        \
 689                      PVOP_CALL_ARG3(arg3), PVOP_CALL_ARG4(arg4))
 690 #endif
 691
 692 static inline int paravirt_enabled(void)
 693 {
 694         return pv_info.paravirt_enabled;
 695 }
 696
 697 static inline void load_sp0(struct tss_struct *tss,
 698                              struct thread_struct *thread)
 699 {
 700         PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
 701 }
 702
 703 #define ARCH_SETUP                      pv_init_ops.arch_setup();
 704 static inline unsigned long get_wallclock(void)
 705 {
 706         return PVOP_CALL0(unsigned long, pv_time_ops.get_wallclock);
 707 }
 708
 709 static inline int set_wallclock(unsigned long nowtime)
 710 {
 711         return PVOP_CALL1(int, pv_time_ops.set_wallclock, nowtime);
 712 }
 713
 714 static inline void (*choose_time_init(void))(void)
 715 {
 716         return pv_time_ops.time_init;
 717 }
 718
 719 /* The paravirtualized CPUID instruction. */
 720 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
 721                            unsigned int *ecx, unsigned int *edx)
 722 {
 723         PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
 724 }
 725
 726 /*
 727  * These special macros can be used to get or set a debugging register
 728  */
 729 static inline unsigned long paravirt_get_debugreg(int reg)
 730 {
 731         return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
 732 }
 733 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
 734 static inline void set_debugreg(unsigned long val, int reg)
 735 {
 736         PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
 737 }
 738
 739 static inline void clts(void)
 740 {
 741         PVOP_VCALL0(pv_cpu_ops.clts);
 742 }
 743
 744 static inline unsigned long read_cr0(void)
 745 {
 746         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
 747 }
 748
 749 static inline void write_cr0(unsigned long x)
 750 {
 751         PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
 752 }
 753
 754 static inline unsigned long read_cr2(void)
 755 {
 756         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
 757 }
 758
 759 static inline void write_cr2(unsigned long x)
 760 {
 761         PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
 762 }
 763
 764 static inline unsigned long read_cr3(void)
 765 {
 766         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
 767 }
 768
 769 static inline void write_cr3(unsigned long x)
 770 {
 771         PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
 772 }
 773
 774 static inline unsigned long read_cr4(void)
 775 {
 776         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
 777 }
 778 static inline unsigned long read_cr4_safe(void)
 779 {
 780         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
 781 }
 782
 783 static inline void write_cr4(unsigned long x)
 784 {
 785         PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
 786 }
 787
 788 #ifdef CONFIG_X86_64
 789 static inline unsigned long read_cr8(void)
 790 {
 791         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
 792 }
 793
 794 static inline void write_cr8(unsigned long x)
 795 {
 796         PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
 797 }
 798 #endif
 799
 800 static inline void raw_safe_halt(void)
 801 {
 802         PVOP_VCALL0(pv_irq_ops.safe_halt);
 803 }
 804
 805 static inline void halt(void)
 806 {
 807         PVOP_VCALL0(pv_irq_ops.safe_halt);
 808 }
 809
 810 static inline void wbinvd(void)
 811 {
 812         PVOP_VCALL0(pv_cpu_ops.wbinvd);
 813 }
 814
 815 #define get_kernel_rpl()  (pv_info.kernel_rpl)
 816
 817 static inline u64 paravirt_read_msr(unsigned msr, int *err)
 818 {
 819         return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
 820 }
 821 static inline u64 paravirt_read_msr_amd(unsigned msr, int *err)
 822 {
 823         return PVOP_CALL2(u64, pv_cpu_ops.read_msr_amd, msr, err);
 824 }
 825 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
 826 {
 827         return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
 828 }
 829
 830 /* These should all do BUG_ON(_err), but our headers are too tangled. */
 831 #define rdmsr(msr, val1, val2)                  \
 832 do {                                            \
 833         int _err;                               \
 834         u64 _l = paravirt_read_msr(msr, &_err); \
 835         val1 = (u32)_l;                         \
 836         val2 = _l >> 32;                        \
 837 } while (0)
 838
 839 #define wrmsr(msr, val1, val2)                  \
 840 do {                                            \
 841         paravirt_write_msr(msr, val1, val2);    \
 842 } while (0)
 843
 844 #define rdmsrl(msr, val)                        \
 845 do {                                            \
 846         int _err;                               \
 847         val = paravirt_read_msr(msr, &_err);    \
 848 } while (0)
 849
 850 #define wrmsrl(msr, val)        wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
 851 #define wrmsr_safe(msr, a, b)   paravirt_write_msr(msr, a, b)
 852
 853 /* rdmsr with exception handling */
 854 #define rdmsr_safe(msr, a, b)                   \
 855 ({                                              \
 856         int _err;                               \
 857         u64 _l = paravirt_read_msr(msr, &_err); \
 858         (*a) = (u32)_l;                         \
 859         (*b) = _l >> 32;                        \
 860         _err;                                   \
 861 })
 862
 863 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
 864 {
 865         int err;
 866
 867         *p = paravirt_read_msr(msr, &err);
 868         return err;
 869 }
 870 static inline int rdmsrl_amd_safe(unsigned msr, unsigned long long *p)
 871 {
 872         int err;
 873
 874         *p = paravirt_read_msr_amd(msr, &err);
 875         return err;
 876 }
 877
 878 static inline u64 paravirt_read_tsc(void)
 879 {
 880         return PVOP_CALL0(u64, pv_cpu_ops.read_tsc);
 881 }
 882
 883 #define rdtscl(low)                             \
 884 do {                                            \
 885         u64 _l = paravirt_read_tsc();           \
 886         low = (int)_l;                          \
 887 } while (0)
 888
 889 #define rdtscll(val) (val = paravirt_read_tsc())
 890
 891 static inline unsigned long long paravirt_sched_clock(void)
 892 {
 893         return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
 894 }
 895 #define calibrate_tsc() (pv_time_ops.get_tsc_khz())
 896
 897 static inline unsigned long long paravirt_read_pmc(int counter)
 898 {
 899         return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
 900 }
 901
 902 #define rdpmc(counter, low, high)               \
 903 do {                                            \
 904         u64 _l = paravirt_read_pmc(counter);    \
 905         low = (u32)_l;                          \
 906         high = _l >> 32;                        \
 907 } while (0)
 908
 909 static inline unsigned long long paravirt_rdtscp(unsigned int *aux)
 910 {
 911         return PVOP_CALL1(u64, pv_cpu_ops.read_tscp, aux);
 912 }
 913
 914 #define rdtscp(low, high, aux)                          \
 915 do {                                                    \
 916         int __aux;                                      \
 917         unsigned long __val = paravirt_rdtscp(&__aux);  \
 918         (low) = (u32)__val;                             \
 919         (high) = (u32)(__val >> 32);                    \
 920         (aux) = __aux;                                  \
 921 } while (0)
 922
 923 #define rdtscpll(val, aux)                              \
 924 do {                                                    \
 925         unsigned long __aux;                            \
 926         val = paravirt_rdtscp(&__aux);                  \
 927         (aux) = __aux;                                  \
 928 } while (0)
 929
 930 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
 931 {
 932         PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
 933 }
 934
 935 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
 936 {
 937         PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
 938 }
 939
 940 static inline void load_TR_desc(void)
 941 {
 942         PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
 943 }
 944 static inline void load_gdt(const struct desc_ptr *dtr)
 945 {
 946         PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
 947 }
 948 static inline void load_idt(const struct desc_ptr *dtr)
 949 {
 950         PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
 951 }
 952 static inline void set_ldt(const void *addr, unsigned entries)
 953 {
 954         PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
 955 }
 956 static inline void store_gdt(struct desc_ptr *dtr)
 957 {
 958         PVOP_VCALL1(pv_cpu_ops.store_gdt, dtr);
 959 }
 960 static inline void store_idt(struct desc_ptr *dtr)
 961 {
 962         PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
 963 }
 964 static inline unsigned long paravirt_store_tr(void)
 965 {
 966         return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
 967 }
 968 #define store_tr(tr)    ((tr) = paravirt_store_tr())
 969 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
 970 {
 971         PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
 972 }
 973
 974 #ifdef CONFIG_X86_64
 975 static inline void load_gs_index(unsigned int gs)
 976 {
 977         PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
 978 }
 979 #endif
 980
 981 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
 982                                    const void *desc)
 983 {
 984         PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
 985 }
 986
 987 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
 988                                    void *desc, int type)
 989 {
 990         PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
 991 }
 992
 993 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
 994 {
 995         PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
 996 }
 997 static inline void set_iopl_mask(unsigned mask)
 998 {
 999         PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
1000 }
1001
1002 /* The paravirtualized I/O functions */
1003 static inline void slow_down_io(void)
1004 {
1005         pv_cpu_ops.io_delay();
1006 #ifdef REALLY_SLOW_IO
1007         pv_cpu_ops.io_delay();
1008         pv_cpu_ops.io_delay();
1009         pv_cpu_ops.io_delay();
1010 #endif
1011 }
1012
1013 #ifdef CONFIG_X86_LOCAL_APIC
1014 static inline void setup_boot_clock(void)
1015 {
1016         PVOP_VCALL0(pv_apic_ops.setup_boot_clock);
1017 }
1018
1019 static inline void setup_secondary_clock(void)
1020 {
1021         PVOP_VCALL0(pv_apic_ops.setup_secondary_clock);
1022 }
1023 #endif
1024
1025 static inline void paravirt_post_allocator_init(void)
1026 {
1027         if (pv_init_ops.post_allocator_init)
1028                 (*pv_init_ops.post_allocator_init)();
1029 }
1030
1031 static inline void paravirt_pagetable_setup_start(pgd_t *base)
1032 {
1033         (*pv_mmu_ops.pagetable_setup_start)(base);
1034 }
1035
1036 static inline void paravirt_pagetable_setup_done(pgd_t *base)
1037 {
1038         (*pv_mmu_ops.pagetable_setup_done)(base);
1039 }
1040
1041 #ifdef CONFIG_SMP
1042 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
1043                                     unsigned long start_esp)
1044 {
1045         PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
1046                     phys_apicid, start_eip, start_esp);
1047 }
1048 #endif
1049
1050 static inline void paravirt_activate_mm(struct mm_struct *prev,
1051                                         struct mm_struct *next)
1052 {
1053         PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
1054 }
1055
1056 static inline void arch_dup_mmap(struct mm_struct *oldmm,
1057                                  struct mm_struct *mm)
1058 {
1059         PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
1060 }
1061
1062 static inline void arch_exit_mmap(struct mm_struct *mm)
1063 {
1064         PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
1065 }
1066
1067 static inline void __flush_tlb(void)
1068 {
1069         PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
1070 }
1071 static inline void __flush_tlb_global(void)
1072 {
1073         PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
1074 }
1075 static inline void __flush_tlb_single(unsigned long addr)
1076 {
1077         PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
1078 }
1079
1080 static inline void flush_tlb_others(const struct cpumask *cpumask,
1081                                     struct mm_struct *mm,
1082                                     unsigned long va)
1083 {
1084         PVOP_VCALL3(pv_mmu_ops.flush_tlb_others, cpumask, mm, va);
1085 }
1086
1087 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
1088 {
1089         return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
1090 }
1091
1092 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1093 {
1094         PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
1095 }
1096
1097 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1098 {
1099         PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
1100 }
1101 static inline void paravirt_release_pte(unsigned long pfn)
1102 {
1103         PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
1104 }
1105
1106 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1107 {
1108         PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
1109 }
1110
1111 static inline void paravirt_alloc_pmd_clone(unsigned long pfn, unsigned long clonepfn,
1112                                             unsigned long start, unsigned long count)
1113 {
1114         PVOP_VCALL4(pv_mmu_ops.alloc_pmd_clone, pfn, clonepfn, start, count);
1115 }
1116 static inline void paravirt_release_pmd(unsigned long pfn)
1117 {
1118         PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
1119 }
1120
1121 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1122 {
1123         PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
1124 }
1125 static inline void paravirt_release_pud(unsigned long pfn)
1126 {
1127         PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
1128 }
1129
1130 #ifdef CONFIG_HIGHPTE
1131 static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
1132 {
1133         unsigned long ret;
1134         ret = PVOP_CALL2(unsigned long, pv_mmu_ops.kmap_atomic_pte, page, type);
1135         return (void *)ret;
1136 }
1137 #endif
1138
1139 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
1140                               pte_t *ptep)
1141 {
1142         PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
1143 }
1144
1145 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
1146                                     pte_t *ptep)
1147 {
1148         PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
1149 }
1150
1151 static inline pte_t __pte(pteval_t val)
1152 {
1153         pteval_t ret;
1154
1155         if (sizeof(pteval_t) > sizeof(long))
1156                 ret = PVOP_CALLEE2(pteval_t,
1157                                    pv_mmu_ops.make_pte,
1158                                    val, (u64)val >> 32);
1159         else
1160                 ret = PVOP_CALLEE1(pteval_t,
1161                                    pv_mmu_ops.make_pte,
1162                                    val);
1163
1164         return (pte_t) { .pte = ret };
1165 }
1166
1167 static inline pteval_t pte_val(pte_t pte)
1168 {
1169         pteval_t ret;
1170
1171         if (sizeof(pteval_t) > sizeof(long))
1172                 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
1173                                    pte.pte, (u64)pte.pte >> 32);
1174         else
1175                 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
1176                                    pte.pte);
1177
1178         return ret;
1179 }
1180
1181 static inline pgd_t __pgd(pgdval_t val)
1182 {
1183         pgdval_t ret;
1184
1185         if (sizeof(pgdval_t) > sizeof(long))
1186                 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
1187                                    val, (u64)val >> 32);
1188         else
1189                 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
1190                                    val);
1191
1192         return (pgd_t) { ret };
1193 }
1194
1195 static inline pgdval_t pgd_val(pgd_t pgd)
1196 {
1197         pgdval_t ret;
1198
1199         if (sizeof(pgdval_t) > sizeof(long))
1200                 ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
1201                                     pgd.pgd, (u64)pgd.pgd >> 32);
1202         else
1203                 ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
1204                                     pgd.pgd);
1205
1206         return ret;
1207 }
1208
1209 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1210 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
1211                                            pte_t *ptep)
1212 {
1213         pteval_t ret;
1214
1215         ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
1216                          mm, addr, ptep);
1217
1218         return (pte_t) { .pte = ret };
1219 }
1220
1221 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
1222                                            pte_t *ptep, pte_t pte)
1223 {
1224         if (sizeof(pteval_t) > sizeof(long))
1225                 /* 5 arg words */
1226                 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
1227         else
1228                 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
1229                             mm, addr, ptep, pte.pte);
1230 }
1231
1232 static inline void set_pte(pte_t *ptep, pte_t pte)
1233 {
1234         if (sizeof(pteval_t) > sizeof(long))
1235                 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
1236                             pte.pte, (u64)pte.pte >> 32);
1237         else
1238                 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
1239                             pte.pte);
1240 }
1241
1242 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
1243                               pte_t *ptep, pte_t pte)
1244 {
1245         if (sizeof(pteval_t) > sizeof(long))
1246                 /* 5 arg words */
1247                 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
1248         else
1249                 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
1250 }
1251
1252 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
1253 {
1254         pmdval_t val = native_pmd_val(pmd);
1255
1256         if (sizeof(pmdval_t) > sizeof(long))
1257                 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
1258         else
1259                 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
1260 }
1261
1262 #if PAGETABLE_LEVELS >= 3
1263 static inline pmd_t __pmd(pmdval_t val)
1264 {
1265         pmdval_t ret;
1266
1267         if (sizeof(pmdval_t) > sizeof(long))
1268                 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
1269                                    val, (u64)val >> 32);
1270         else
1271                 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
1272                                    val);
1273
1274         return (pmd_t) { ret };
1275 }
1276
1277 static inline pmdval_t pmd_val(pmd_t pmd)
1278 {
1279         pmdval_t ret;
1280
1281         if (sizeof(pmdval_t) > sizeof(long))
1282                 ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
1283                                     pmd.pmd, (u64)pmd.pmd >> 32);
1284         else
1285                 ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
1286                                     pmd.pmd);
1287
1288         return ret;
1289 }
1290
1291 static inline void set_pud(pud_t *pudp, pud_t pud)
1292 {
1293         pudval_t val = native_pud_val(pud);
1294
1295         if (sizeof(pudval_t) > sizeof(long))
1296                 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
1297                             val, (u64)val >> 32);
1298         else
1299                 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
1300                             val);
1301 }
1302 #if PAGETABLE_LEVELS == 4
1303 static inline pud_t __pud(pudval_t val)
1304 {
1305         pudval_t ret;
1306
1307         if (sizeof(pudval_t) > sizeof(long))
1308                 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
1309                                    val, (u64)val >> 32);
1310         else
1311                 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
1312                                    val);
1313
1314         return (pud_t) { ret };
1315 }
1316
1317 static inline pudval_t pud_val(pud_t pud)
1318 {
1319         pudval_t ret;
1320
1321         if (sizeof(pudval_t) > sizeof(long))
1322                 ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
1323                                     pud.pud, (u64)pud.pud >> 32);
1324         else
1325                 ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
1326                                     pud.pud);
1327
1328         return ret;
1329 }
1330
1331 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
1332 {
1333         pgdval_t val = native_pgd_val(pgd);
1334
1335         if (sizeof(pgdval_t) > sizeof(long))
1336                 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
1337                             val, (u64)val >> 32);
1338         else
1339                 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
1340                             val);
1341 }
1342
1343 static inline void pgd_clear(pgd_t *pgdp)
1344 {
1345         set_pgd(pgdp, __pgd(0));
1346 }
1347
1348 static inline void pud_clear(pud_t *pudp)
1349 {
1350         set_pud(pudp, __pud(0));
1351 }
1352
1353 #endif  /* PAGETABLE_LEVELS == 4 */
1354
1355 #endif  /* PAGETABLE_LEVELS >= 3 */
1356
1357 #ifdef CONFIG_X86_PAE
1358 /* Special-case pte-setting operations for PAE, which can't update a
1359    64-bit pte atomically */
1360 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
1361 {
1362         PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
1363                     pte.pte, pte.pte >> 32);
1364 }
1365
1366 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
1367                              pte_t *ptep)
1368 {
1369         PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
1370 }
1371
1372 static inline void pmd_clear(pmd_t *pmdp)
1373 {
1374         PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
1375 }
1376 #else  /* !CONFIG_X86_PAE */
1377 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
1378 {
1379         set_pte(ptep, pte);
1380 }
1381
1382 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
1383                              pte_t *ptep)
1384 {
1385         set_pte_at(mm, addr, ptep, __pte(0));
1386 }
1387
1388 static inline void pmd_clear(pmd_t *pmdp)
1389 {
1390         set_pmd(pmdp, __pmd(0));
1391 }
1392 #endif  /* CONFIG_X86_PAE */
1393
1394 /* Lazy mode for batching updates / context switch */
1395 enum paravirt_lazy_mode {
1396         PARAVIRT_LAZY_NONE,
1397         PARAVIRT_LAZY_MMU,
1398         PARAVIRT_LAZY_CPU,
1399 };
1400
1401 enum paravirt_lazy_mode paravirt_get_lazy_mode(void);
1402 void paravirt_enter_lazy_cpu(void);
1403 void paravirt_leave_lazy_cpu(void);
1404 void paravirt_enter_lazy_mmu(void);
1405 void paravirt_leave_lazy_mmu(void);
1406 void paravirt_leave_lazy(enum paravirt_lazy_mode mode);
1407
1408 #define  __HAVE_ARCH_ENTER_LAZY_CPU_MODE
1409 static inline void arch_enter_lazy_cpu_mode(void)
1410 {
1411         PVOP_VCALL0(pv_cpu_ops.lazy_mode.enter);
1412 }
1413
1414 static inline void arch_leave_lazy_cpu_mode(void)
1415 {
1416         PVOP_VCALL0(pv_cpu_ops.lazy_mode.leave);
1417 }
1418
1419 void arch_flush_lazy_cpu_mode(void);
1420
1421 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
1422 static inline void arch_enter_lazy_mmu_mode(void)
1423 {
1424         PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
1425 }
1426
1427 static inline void arch_leave_lazy_mmu_mode(void)
1428 {
1429         PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
1430 }
1431
1432 void arch_flush_lazy_mmu_mode(void);
1433
1434 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
1435                                 phys_addr_t phys, pgprot_t flags)
1436 {
1437         pv_mmu_ops.set_fixmap(idx, phys, flags);
1438 }
1439
1440 void _paravirt_nop(void);
1441 u32 _paravirt_ident_32(u32);
1442 u64 _paravirt_ident_64(u64);
1443
1444 #define paravirt_nop    ((void *)_paravirt_nop)
1445
1446 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
1447
1448 static inline int __raw_spin_is_locked(struct raw_spinlock *lock)
1449 {
1450         return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
1451 }
1452
1453 static inline int __raw_spin_is_contended(struct raw_spinlock *lock)
1454 {
1455         return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
1456 }
1457 #define __raw_spin_is_contended __raw_spin_is_contended
1458
1459 static __always_inline void __raw_spin_lock(struct raw_spinlock *lock)
1460 {
1461         PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
1462 }
1463
1464 static __always_inline void __raw_spin_lock_flags(struct raw_spinlock *lock,
1465                                                   unsigned long flags)
1466 {
1467         PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
1468 }
1469
1470 static __always_inline int __raw_spin_trylock(struct raw_spinlock *lock)
1471 {
1472         return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
1473 }
1474
1475 static __always_inline void __raw_spin_unlock(struct raw_spinlock *lock)
1476 {
1477         PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
1478 }
1479
1480 #endif
1481
1482 /* These all sit in the .parainstructions section to tell us what to patch. */
1483 struct paravirt_patch_site {
1484         u8 *instr;              /* original instructions */
1485         u8 instrtype;           /* type of this instruction */
1486         u8 len;                 /* length of original instruction */
1487         u16 clobbers;           /* what registers you may clobber */
1488 };
1489
1490 extern struct paravirt_patch_site __parainstructions[],
1491         __parainstructions_end[];
1492
1493 #ifdef CONFIG_X86_32
1494 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
1495 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
1496
1497 /* save and restore all caller-save registers, except return value */
1498 #define PV_SAVE_ALL_CALLER_REGS         "pushl %ecx;"
1499 #define PV_RESTORE_ALL_CALLER_REGS      "popl  %ecx;"
1500
1501 #define PV_FLAGS_ARG "0"
1502 #define PV_EXTRA_CLOBBERS
1503 #define PV_VEXTRA_CLOBBERS
1504 #else
1505 /* save and restore all caller-save registers, except return value */
1506 #define PV_SAVE_ALL_CALLER_REGS                                         \
1507         "push %rcx;"                                                    \
1508         "push %rdx;"                                                    \
1509         "push %rsi;"                                                    \
1510         "push %rdi;"                                                    \
1511         "push %r8;"                                                     \
1512         "push %r9;"                                                     \
1513         "push %r10;"                                                    \
1514         "push %r11;"
1515 #define PV_RESTORE_ALL_CALLER_REGS                                      \
1516         "pop %r11;"                                                     \
1517         "pop %r10;"                                                     \
1518         "pop %r9;"                                                      \
1519         "pop %r8;"                                                      \
1520         "pop %rdi;"                                                     \
1521         "pop %rsi;"                                                     \
1522         "pop %rdx;"                                                     \
1523         "pop %rcx;"
1524
1525 /* We save some registers, but all of them, that's too much. We clobber all
1526  * caller saved registers but the argument parameter */
1527 #define PV_SAVE_REGS "pushq %%rdi;"
1528 #define PV_RESTORE_REGS "popq %%rdi;"
1529 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
1530 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
1531 #define PV_FLAGS_ARG "D"
1532 #endif
1533
1534 /*
1535  * Generate a thunk around a function which saves all caller-save
1536  * registers except for the return value.  This allows C functions to
1537  * be called from assembler code where fewer than normal registers are
1538  * available.  It may also help code generation around calls from C
1539  * code if the common case doesn't use many registers.
1540  *
1541  * When a callee is wrapped in a thunk, the caller can assume that all
1542  * arg regs and all scratch registers are preserved across the
1543  * call. The return value in rax/eax will not be saved, even for void
1544  * functions.
1545  */
1546 #define PV_CALLEE_SAVE_REGS_THUNK(func)                                 \
1547         extern typeof(func) __raw_callee_save_##func;                   \
1548         static void *__##func##__ __used = func;                        \
1549                                                                         \
1550         asm(".pushsection .text;"                                       \
1551             "__raw_callee_save_" #func ": "                             \
1552             PV_SAVE_ALL_CALLER_REGS                                     \
1553             "call " #func ";"                                           \
1554             PV_RESTORE_ALL_CALLER_REGS                                  \
1555             "ret;"                                                      \
1556             ".popsection")
1557
1558 /* Get a reference to a callee-save function */
1559 #define PV_CALLEE_SAVE(func)                                            \
1560         ((struct paravirt_callee_save) { __raw_callee_save_##func })
1561
1562 /* Promise that "func" already uses the right calling convention */
1563 #define __PV_IS_CALLEE_SAVE(func)                       \
1564         ((struct paravirt_callee_save) { func })
1565
1566 static inline unsigned long __raw_local_save_flags(void)
1567 {
1568         unsigned long f;
1569
1570         asm volatile(paravirt_alt(PARAVIRT_CALL)
1571                      : "=a"(f)
1572                      : paravirt_type(pv_irq_ops.save_fl),
1573                        paravirt_clobber(CLBR_EAX)
1574                      : "memory", "cc");
1575         return f;
1576 }
1577
1578 static inline void raw_local_irq_restore(unsigned long f)
1579 {
1580         asm volatile(paravirt_alt(PARAVIRT_CALL)
1581                      : "=a"(f)
1582                      : PV_FLAGS_ARG(f),
1583                        paravirt_type(pv_irq_ops.restore_fl),
1584                        paravirt_clobber(CLBR_EAX)
1585                      : "memory", "cc");
1586 }
1587
1588 static inline void raw_local_irq_disable(void)
1589 {
1590         asm volatile(paravirt_alt(PARAVIRT_CALL)
1591                      :
1592                      : paravirt_type(pv_irq_ops.irq_disable),
1593                        paravirt_clobber(CLBR_EAX)
1594                      : "memory", "eax", "cc");
1595 }
1596
1597 static inline void raw_local_irq_enable(void)
1598 {
1599         asm volatile(paravirt_alt(PARAVIRT_CALL)
1600                      :
1601                      : paravirt_type(pv_irq_ops.irq_enable),
1602                        paravirt_clobber(CLBR_EAX)
1603                      : "memory", "eax", "cc");
1604 }
1605
1606 static inline unsigned long __raw_local_irq_save(void)
1607 {
1608         unsigned long f;
1609
1610         f = __raw_local_save_flags();
1611         raw_local_irq_disable();
1612         return f;
1613 }
1614
1615
1616 /* Make sure as little as possible of this mess escapes. */
1617 #undef PARAVIRT_CALL
1618 #undef __PVOP_CALL
1619 #undef __PVOP_VCALL
1620 #undef PVOP_VCALL0
1621 #undef PVOP_CALL0
1622 #undef PVOP_VCALL1
1623 #undef PVOP_CALL1
1624 #undef PVOP_VCALL2
1625 #undef PVOP_CALL2
1626 #undef PVOP_VCALL3
1627 #undef PVOP_CALL3
1628 #undef PVOP_VCALL4
1629 #undef PVOP_CALL4
1630
1631 #else  /* __ASSEMBLY__ */
1632
1633 #define _PVSITE(ptype, clobbers, ops, word, algn)       \
1634 771:;                                           \
1635         ops;                                    \
1636 772:;                                           \
1637         .pushsection .parainstructions,"a";     \
1638          .align algn;                           \
1639          word 771b;                             \
1640          .byte ptype;                           \
1641          .byte 772b-771b;                       \
1642          .short clobbers;                       \
1643         .popsection
1644
1645
1646 #define COND_PUSH(set, mask, reg)                       \
1647         .if ((~(set)) & mask); push %reg; .endif
1648 #define COND_POP(set, mask, reg)                        \
1649         .if ((~(set)) & mask); pop %reg; .endif
1650
1651 #ifdef CONFIG_X86_64
1652
1653 #define PV_SAVE_REGS(set)                       \
1654         COND_PUSH(set, CLBR_RAX, rax);          \
1655         COND_PUSH(set, CLBR_RCX, rcx);          \
1656         COND_PUSH(set, CLBR_RDX, rdx);          \
1657         COND_PUSH(set, CLBR_RSI, rsi);          \
1658         COND_PUSH(set, CLBR_RDI, rdi);          \
1659         COND_PUSH(set, CLBR_R8, r8);            \
1660         COND_PUSH(set, CLBR_R9, r9);            \
1661         COND_PUSH(set, CLBR_R10, r10);          \
1662         COND_PUSH(set, CLBR_R11, r11)
1663 #define PV_RESTORE_REGS(set)                    \
1664         COND_POP(set, CLBR_R11, r11);           \
1665         COND_POP(set, CLBR_R10, r10);           \
1666         COND_POP(set, CLBR_R9, r9);             \
1667         COND_POP(set, CLBR_R8, r8);             \
1668         COND_POP(set, CLBR_RDI, rdi);           \
1669         COND_POP(set, CLBR_RSI, rsi);           \
1670         COND_POP(set, CLBR_RDX, rdx);           \
1671         COND_POP(set, CLBR_RCX, rcx);           \
1672         COND_POP(set, CLBR_RAX, rax)
1673
1674 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
1675 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
1676 #define PARA_INDIRECT(addr)     *addr(%rip)
1677 #else
1678 #define PV_SAVE_REGS(set)                       \
1679         COND_PUSH(set, CLBR_EAX, eax);          \
1680         COND_PUSH(set, CLBR_EDI, edi);          \
1681         COND_PUSH(set, CLBR_ECX, ecx);          \
1682         COND_PUSH(set, CLBR_EDX, edx)
1683 #define PV_RESTORE_REGS(set)                    \
1684         COND_POP(set, CLBR_EDX, edx);           \
1685         COND_POP(set, CLBR_ECX, ecx);           \
1686         COND_POP(set, CLBR_EDI, edi);           \
1687         COND_POP(set, CLBR_EAX, eax)
1688
1689 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
1690 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
1691 #define PARA_INDIRECT(addr)     *%cs:addr
1692 #endif
1693
1694 #define INTERRUPT_RETURN                                                \
1695         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,       \
1696                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
1697
1698 #define DISABLE_INTERRUPTS(clobbers)                                    \
1699         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
1700                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
1701                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);    \
1702                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
1703
1704 #define ENABLE_INTERRUPTS(clobbers)                                     \
1705         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,  \
1706                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
1707                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);     \
1708                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
1709
1710 #define USERGS_SYSRET32                                                 \
1711         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),       \
1712                   CLBR_NONE,                                            \
1713                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
1714
1715 #ifdef CONFIG_X86_32
1716 #define GET_CR0_INTO_EAX                                \
1717         push %ecx; push %edx;                           \
1718         call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
1719         pop %edx; pop %ecx
1720
1721 #define ENABLE_INTERRUPTS_SYSEXIT                                       \
1722         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
1723                   CLBR_NONE,                                            \
1724                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1725
1726
1727 #else   /* !CONFIG_X86_32 */
1728
1729 /*
1730  * If swapgs is used while the userspace stack is still current,
1731  * there's no way to call a pvop.  The PV replacement *must* be
1732  * inlined, or the swapgs instruction must be trapped and emulated.
1733  */
1734 #define SWAPGS_UNSAFE_STACK                                             \
1735         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
1736                   swapgs)
1737
1738 /*
1739  * Note: swapgs is very special, and in practise is either going to be
1740  * implemented with a single "swapgs" instruction or something very
1741  * special.  Either way, we don't need to save any registers for
1742  * it.
1743  */
1744 #define SWAPGS                                                          \
1745         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
1746                   call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)          \
1747                  )
1748
1749 #define GET_CR2_INTO_RCX                                \
1750         call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2); \
1751         movq %rax, %rcx;                                \
1752         xorq %rax, %rax;
1753
1754 #define PARAVIRT_ADJUST_EXCEPTION_FRAME                                 \
1755         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
1756                   CLBR_NONE,                                            \
1757                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
1758
1759 #define USERGS_SYSRET64                                                 \
1760         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),       \
1761                   CLBR_NONE,                                            \
1762                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1763
1764 #define ENABLE_INTERRUPTS_SYSEXIT32                                     \
1765         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
1766                   CLBR_NONE,                                            \
1767                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1768 #endif  /* CONFIG_X86_32 */
1769
1770 #endif /* __ASSEMBLY__ */
1771 #endif /* CONFIG_PARAVIRT */
1772 #endif /* _ASM_X86_PARAVIRT_H */