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