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 #include <asm/paravirt_types.h>
  11
  12 #ifndef __ASSEMBLY__
  13 #include <linux/bug.h>
  14 #include <linux/types.h>
  15 #include <linux/cpumask.h>
  16
  17 static inline int paravirt_enabled(void)
  18 {
  19         return pv_info.paravirt_enabled;
  20 }
  21
  22 static inline void load_sp0(struct tss_struct *tss,
  23                              struct thread_struct *thread)
  24 {
  25         PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
  26 }
  27
  28 /* The paravirtualized CPUID instruction. */
  29 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
  30                            unsigned int *ecx, unsigned int *edx)
  31 {
  32         PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
  33 }
  34
  35 /*
  36  * These special macros can be used to get or set a debugging register
  37  */
  38 static inline unsigned long paravirt_get_debugreg(int reg)
  39 {
  40         return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
  41 }
  42 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
  43 static inline void set_debugreg(unsigned long val, int reg)
  44 {
  45         PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
  46 }
  47
  48 static inline void clts(void)
  49 {
  50         PVOP_VCALL0(pv_cpu_ops.clts);
  51 }
  52
  53 static inline unsigned long read_cr0(void)
  54 {
  55         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
  56 }
  57
  58 static inline void write_cr0(unsigned long x)
  59 {
  60         PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
  61 }
  62
  63 static inline unsigned long read_cr2(void)
  64 {
  65         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
  66 }
  67
  68 static inline void write_cr2(unsigned long x)
  69 {
  70         PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
  71 }
  72
  73 static inline unsigned long read_cr3(void)
  74 {
  75         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
  76 }
  77
  78 static inline void write_cr3(unsigned long x)
  79 {
  80         PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
  81 }
  82
  83 static inline unsigned long __read_cr4(void)
  84 {
  85         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
  86 }
  87 static inline unsigned long __read_cr4_safe(void)
  88 {
  89         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
  90 }
  91
  92 static inline void __write_cr4(unsigned long x)
  93 {
  94         PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
  95 }
  96
  97 #ifdef CONFIG_X86_64
  98 static inline unsigned long read_cr8(void)
  99 {
 100         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
 101 }
 102
 103 static inline void write_cr8(unsigned long x)
 104 {
 105         PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
 106 }
 107 #endif
 108
 109 static inline void arch_safe_halt(void)
 110 {
 111         PVOP_VCALL0(pv_irq_ops.safe_halt);
 112 }
 113
 114 static inline void halt(void)
 115 {
 116         PVOP_VCALL0(pv_irq_ops.halt);
 117 }
 118
 119 static inline void wbinvd(void)
 120 {
 121         PVOP_VCALL0(pv_cpu_ops.wbinvd);
 122 }
 123
 124 #define get_kernel_rpl()  (pv_info.kernel_rpl)
 125
 126 static inline u64 paravirt_read_msr(unsigned msr, int *err)
 127 {
 128         return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
 129 }
 130
 131 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
 132 {
 133         return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
 134 }
 135
 136 /* These should all do BUG_ON(_err), but our headers are too tangled. */
 137 #define rdmsr(msr, val1, val2)                  \
 138 do {                                            \
 139         int _err;                               \
 140         u64 _l = paravirt_read_msr(msr, &_err); \
 141         val1 = (u32)_l;                         \
 142         val2 = _l >> 32;                        \
 143 } while (0)
 144
 145 #define wrmsr(msr, val1, val2)                  \
 146 do {                                            \
 147         paravirt_write_msr(msr, val1, val2);    \
 148 } while (0)
 149
 150 #define rdmsrl(msr, val)                        \
 151 do {                                            \
 152         int _err;                               \
 153         val = paravirt_read_msr(msr, &_err);    \
 154 } while (0)
 155
 156 static inline void wrmsrl(unsigned msr, u64 val)
 157 {
 158         wrmsr(msr, (u32)val, (u32)(val>>32));
 159 }
 160
 161 #define wrmsr_safe(msr, a, b)   paravirt_write_msr(msr, a, b)
 162
 163 /* rdmsr with exception handling */
 164 #define rdmsr_safe(msr, a, b)                   \
 165 ({                                              \
 166         int _err;                               \
 167         u64 _l = paravirt_read_msr(msr, &_err); \
 168         (*a) = (u32)_l;                         \
 169         (*b) = _l >> 32;                        \
 170         _err;                                   \
 171 })
 172
 173 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
 174 {
 175         int err;
 176
 177         *p = paravirt_read_msr(msr, &err);
 178         return err;
 179 }
 180
 181 static inline unsigned long long paravirt_sched_clock(void)
 182 {
 183         return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
 184 }
 185
 186 struct static_key;
 187 extern struct static_key paravirt_steal_enabled;
 188 extern struct static_key paravirt_steal_rq_enabled;
 189
 190 static inline u64 paravirt_steal_clock(int cpu)
 191 {
 192         return PVOP_CALL1(u64, pv_time_ops.steal_clock, cpu);
 193 }
 194
 195 static inline unsigned long long paravirt_read_pmc(int counter)
 196 {
 197         return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
 198 }
 199
 200 #define rdpmc(counter, low, high)               \
 201 do {                                            \
 202         u64 _l = paravirt_read_pmc(counter);    \
 203         low = (u32)_l;                          \
 204         high = _l >> 32;                        \
 205 } while (0)
 206
 207 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter))
 208
 209 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
 210 {
 211         PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
 212 }
 213
 214 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
 215 {
 216         PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
 217 }
 218
 219 static inline void load_TR_desc(void)
 220 {
 221         PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
 222 }
 223 static inline void load_gdt(const struct desc_ptr *dtr)
 224 {
 225         PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
 226 }
 227 static inline void load_idt(const struct desc_ptr *dtr)
 228 {
 229         PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
 230 }
 231 static inline void set_ldt(const void *addr, unsigned entries)
 232 {
 233         PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
 234 }
 235 static inline void store_idt(struct desc_ptr *dtr)
 236 {
 237         PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
 238 }
 239 static inline unsigned long paravirt_store_tr(void)
 240 {
 241         return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
 242 }
 243 #define store_tr(tr)    ((tr) = paravirt_store_tr())
 244 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
 245 {
 246         PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
 247 }
 248
 249 #ifdef CONFIG_X86_64
 250 static inline void load_gs_index(unsigned int gs)
 251 {
 252         PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
 253 }
 254 #endif
 255
 256 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
 257                                    const void *desc)
 258 {
 259         PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
 260 }
 261
 262 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
 263                                    void *desc, int type)
 264 {
 265         PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
 266 }
 267
 268 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
 269 {
 270         PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
 271 }
 272 static inline void set_iopl_mask(unsigned mask)
 273 {
 274         PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
 275 }
 276
 277 /* The paravirtualized I/O functions */
 278 static inline void slow_down_io(void)
 279 {
 280         pv_cpu_ops.io_delay();
 281 #ifdef REALLY_SLOW_IO
 282         pv_cpu_ops.io_delay();
 283         pv_cpu_ops.io_delay();
 284         pv_cpu_ops.io_delay();
 285 #endif
 286 }
 287
 288 static inline void paravirt_activate_mm(struct mm_struct *prev,
 289                                         struct mm_struct *next)
 290 {
 291         PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
 292 }
 293
 294 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
 295                                           struct mm_struct *mm)
 296 {
 297         PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
 298 }
 299
 300 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
 301 {
 302         PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
 303 }
 304
 305 static inline void __flush_tlb(void)
 306 {
 307         PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
 308 }
 309 static inline void __flush_tlb_global(void)
 310 {
 311         PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
 312 }
 313 static inline void __flush_tlb_single(unsigned long addr)
 314 {
 315         PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
 316 }
 317
 318 static inline void flush_tlb_others(const struct cpumask *cpumask,
 319                                     struct mm_struct *mm,
 320                                     unsigned long start,
 321                                     unsigned long end)
 322 {
 323         PVOP_VCALL4(pv_mmu_ops.flush_tlb_others, cpumask, mm, start, end);
 324 }
 325
 326 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
 327 {
 328         return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
 329 }
 330
 331 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
 332 {
 333         PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
 334 }
 335
 336 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
 337 {
 338         PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
 339 }
 340 static inline void paravirt_release_pte(unsigned long pfn)
 341 {
 342         PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
 343 }
 344
 345 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
 346 {
 347         PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
 348 }
 349
 350 static inline void paravirt_release_pmd(unsigned long pfn)
 351 {
 352         PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
 353 }
 354
 355 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
 356 {
 357         PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
 358 }
 359 static inline void paravirt_release_pud(unsigned long pfn)
 360 {
 361         PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
 362 }
 363
 364 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
 365                               pte_t *ptep)
 366 {
 367         PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
 368 }
 369
 370 static inline pte_t __pte(pteval_t val)
 371 {
 372         pteval_t ret;
 373
 374         if (sizeof(pteval_t) > sizeof(long))
 375                 ret = PVOP_CALLEE2(pteval_t,
 376                                    pv_mmu_ops.make_pte,
 377                                    val, (u64)val >> 32);
 378         else
 379                 ret = PVOP_CALLEE1(pteval_t,
 380                                    pv_mmu_ops.make_pte,
 381                                    val);
 382
 383         return (pte_t) { .pte = ret };
 384 }
 385
 386 static inline pteval_t pte_val(pte_t pte)
 387 {
 388         pteval_t ret;
 389
 390         if (sizeof(pteval_t) > sizeof(long))
 391                 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
 392                                    pte.pte, (u64)pte.pte >> 32);
 393         else
 394                 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
 395                                    pte.pte);
 396
 397         return ret;
 398 }
 399
 400 static inline pgd_t __pgd(pgdval_t val)
 401 {
 402         pgdval_t ret;
 403
 404         if (sizeof(pgdval_t) > sizeof(long))
 405                 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
 406                                    val, (u64)val >> 32);
 407         else
 408                 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
 409                                    val);
 410
 411         return (pgd_t) { ret };
 412 }
 413
 414 static inline pgdval_t pgd_val(pgd_t pgd)
 415 {
 416         pgdval_t ret;
 417
 418         if (sizeof(pgdval_t) > sizeof(long))
 419                 ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
 420                                     pgd.pgd, (u64)pgd.pgd >> 32);
 421         else
 422                 ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
 423                                     pgd.pgd);
 424
 425         return ret;
 426 }
 427
 428 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 429 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
 430                                            pte_t *ptep)
 431 {
 432         pteval_t ret;
 433
 434         ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
 435                          mm, addr, ptep);
 436
 437         return (pte_t) { .pte = ret };
 438 }
 439
 440 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
 441                                            pte_t *ptep, pte_t pte)
 442 {
 443         if (sizeof(pteval_t) > sizeof(long))
 444                 /* 5 arg words */
 445                 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
 446         else
 447                 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
 448                             mm, addr, ptep, pte.pte);
 449 }
 450
 451 static inline void set_pte(pte_t *ptep, pte_t pte)
 452 {
 453         if (sizeof(pteval_t) > sizeof(long))
 454                 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
 455                             pte.pte, (u64)pte.pte >> 32);
 456         else
 457                 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
 458                             pte.pte);
 459 }
 460
 461 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
 462                               pte_t *ptep, pte_t pte)
 463 {
 464         if (sizeof(pteval_t) > sizeof(long))
 465                 /* 5 arg words */
 466                 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
 467         else
 468                 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
 469 }
 470
 471 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
 472                               pmd_t *pmdp, pmd_t pmd)
 473 {
 474         if (sizeof(pmdval_t) > sizeof(long))
 475                 /* 5 arg words */
 476                 pv_mmu_ops.set_pmd_at(mm, addr, pmdp, pmd);
 477         else
 478                 PVOP_VCALL4(pv_mmu_ops.set_pmd_at, mm, addr, pmdp,
 479                             native_pmd_val(pmd));
 480 }
 481
 482 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
 483 {
 484         pmdval_t val = native_pmd_val(pmd);
 485
 486         if (sizeof(pmdval_t) > sizeof(long))
 487                 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
 488         else
 489                 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
 490 }
 491
 492 #if CONFIG_PGTABLE_LEVELS >= 3
 493 static inline pmd_t __pmd(pmdval_t val)
 494 {
 495         pmdval_t ret;
 496
 497         if (sizeof(pmdval_t) > sizeof(long))
 498                 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
 499                                    val, (u64)val >> 32);
 500         else
 501                 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
 502                                    val);
 503
 504         return (pmd_t) { ret };
 505 }
 506
 507 static inline pmdval_t pmd_val(pmd_t pmd)
 508 {
 509         pmdval_t ret;
 510
 511         if (sizeof(pmdval_t) > sizeof(long))
 512                 ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
 513                                     pmd.pmd, (u64)pmd.pmd >> 32);
 514         else
 515                 ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
 516                                     pmd.pmd);
 517
 518         return ret;
 519 }
 520
 521 static inline void set_pud(pud_t *pudp, pud_t pud)
 522 {
 523         pudval_t val = native_pud_val(pud);
 524
 525         if (sizeof(pudval_t) > sizeof(long))
 526                 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
 527                             val, (u64)val >> 32);
 528         else
 529                 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
 530                             val);
 531 }
 532 #if CONFIG_PGTABLE_LEVELS == 4
 533 static inline pud_t __pud(pudval_t val)
 534 {
 535         pudval_t ret;
 536
 537         if (sizeof(pudval_t) > sizeof(long))
 538                 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
 539                                    val, (u64)val >> 32);
 540         else
 541                 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
 542                                    val);
 543
 544         return (pud_t) { ret };
 545 }
 546
 547 static inline pudval_t pud_val(pud_t pud)
 548 {
 549         pudval_t ret;
 550
 551         if (sizeof(pudval_t) > sizeof(long))
 552                 ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
 553                                     pud.pud, (u64)pud.pud >> 32);
 554         else
 555                 ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
 556                                     pud.pud);
 557
 558         return ret;
 559 }
 560
 561 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
 562 {
 563         pgdval_t val = native_pgd_val(pgd);
 564
 565         if (sizeof(pgdval_t) > sizeof(long))
 566                 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
 567                             val, (u64)val >> 32);
 568         else
 569                 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
 570                             val);
 571 }
 572
 573 static inline void pgd_clear(pgd_t *pgdp)
 574 {
 575         set_pgd(pgdp, __pgd(0));
 576 }
 577
 578 static inline void pud_clear(pud_t *pudp)
 579 {
 580         set_pud(pudp, __pud(0));
 581 }
 582
 583 #endif  /* CONFIG_PGTABLE_LEVELS == 4 */
 584
 585 #endif  /* CONFIG_PGTABLE_LEVELS >= 3 */
 586
 587 #ifdef CONFIG_X86_PAE
 588 /* Special-case pte-setting operations for PAE, which can't update a
 589    64-bit pte atomically */
 590 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
 591 {
 592         PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
 593                     pte.pte, pte.pte >> 32);
 594 }
 595
 596 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
 597                              pte_t *ptep)
 598 {
 599         PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
 600 }
 601
 602 static inline void pmd_clear(pmd_t *pmdp)
 603 {
 604         PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
 605 }
 606 #else  /* !CONFIG_X86_PAE */
 607 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
 608 {
 609         set_pte(ptep, pte);
 610 }
 611
 612 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
 613                              pte_t *ptep)
 614 {
 615         set_pte_at(mm, addr, ptep, __pte(0));
 616 }
 617
 618 static inline void pmd_clear(pmd_t *pmdp)
 619 {
 620         set_pmd(pmdp, __pmd(0));
 621 }
 622 #endif  /* CONFIG_X86_PAE */
 623
 624 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
 625 static inline void arch_start_context_switch(struct task_struct *prev)
 626 {
 627         PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
 628 }
 629
 630 static inline void arch_end_context_switch(struct task_struct *next)
 631 {
 632         PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
 633 }
 634
 635 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 636 static inline void arch_enter_lazy_mmu_mode(void)
 637 {
 638         PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
 639 }
 640
 641 static inline void arch_leave_lazy_mmu_mode(void)
 642 {
 643         PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
 644 }
 645
 646 static inline void arch_flush_lazy_mmu_mode(void)
 647 {
 648         PVOP_VCALL0(pv_mmu_ops.lazy_mode.flush);
 649 }
 650
 651 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
 652                                 phys_addr_t phys, pgprot_t flags)
 653 {
 654         pv_mmu_ops.set_fixmap(idx, phys, flags);
 655 }
 656
 657 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
 658
 659 #ifdef CONFIG_QUEUED_SPINLOCKS
 660
 661 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock,
 662                                                         u32 val)
 663 {
 664         PVOP_VCALL2(pv_lock_ops.queued_spin_lock_slowpath, lock, val);
 665 }
 666
 667 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock)
 668 {
 669         PVOP_VCALLEE1(pv_lock_ops.queued_spin_unlock, lock);
 670 }
 671
 672 static __always_inline void pv_wait(u8 *ptr, u8 val)
 673 {
 674         PVOP_VCALL2(pv_lock_ops.wait, ptr, val);
 675 }
 676
 677 static __always_inline void pv_kick(int cpu)
 678 {
 679         PVOP_VCALL1(pv_lock_ops.kick, cpu);
 680 }
 681
 682 #else /* !CONFIG_QUEUED_SPINLOCKS */
 683
 684 static __always_inline void __ticket_lock_spinning(struct arch_spinlock *lock,
 685                                                         __ticket_t ticket)
 686 {
 687         PVOP_VCALLEE2(pv_lock_ops.lock_spinning, lock, ticket);
 688 }
 689
 690 static __always_inline void __ticket_unlock_kick(struct arch_spinlock *lock,
 691                                                         __ticket_t ticket)
 692 {
 693         PVOP_VCALL2(pv_lock_ops.unlock_kick, lock, ticket);
 694 }
 695
 696 #endif /* CONFIG_QUEUED_SPINLOCKS */
 697
 698 #endif /* SMP && PARAVIRT_SPINLOCKS */
 699
 700 #ifdef CONFIG_X86_32
 701 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
 702 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
 703
 704 /* save and restore all caller-save registers, except return value */
 705 #define PV_SAVE_ALL_CALLER_REGS         "pushl %ecx;"
 706 #define PV_RESTORE_ALL_CALLER_REGS      "popl  %ecx;"
 707
 708 #define PV_FLAGS_ARG "0"
 709 #define PV_EXTRA_CLOBBERS
 710 #define PV_VEXTRA_CLOBBERS
 711 #else
 712 /* save and restore all caller-save registers, except return value */
 713 #define PV_SAVE_ALL_CALLER_REGS                                         \
 714         "push %rcx;"                                                    \
 715         "push %rdx;"                                                    \
 716         "push %rsi;"                                                    \
 717         "push %rdi;"                                                    \
 718         "push %r8;"                                                     \
 719         "push %r9;"                                                     \
 720         "push %r10;"                                                    \
 721         "push %r11;"
 722 #define PV_RESTORE_ALL_CALLER_REGS                                      \
 723         "pop %r11;"                                                     \
 724         "pop %r10;"                                                     \
 725         "pop %r9;"                                                      \
 726         "pop %r8;"                                                      \
 727         "pop %rdi;"                                                     \
 728         "pop %rsi;"                                                     \
 729         "pop %rdx;"                                                     \
 730         "pop %rcx;"
 731
 732 /* We save some registers, but all of them, that's too much. We clobber all
 733  * caller saved registers but the argument parameter */
 734 #define PV_SAVE_REGS "pushq %%rdi;"
 735 #define PV_RESTORE_REGS "popq %%rdi;"
 736 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
 737 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
 738 #define PV_FLAGS_ARG "D"
 739 #endif
 740
 741 /*
 742  * Generate a thunk around a function which saves all caller-save
 743  * registers except for the return value.  This allows C functions to
 744  * be called from assembler code where fewer than normal registers are
 745  * available.  It may also help code generation around calls from C
 746  * code if the common case doesn't use many registers.
 747  *
 748  * When a callee is wrapped in a thunk, the caller can assume that all
 749  * arg regs and all scratch registers are preserved across the
 750  * call. The return value in rax/eax will not be saved, even for void
 751  * functions.
 752  */
 753 #define PV_CALLEE_SAVE_REGS_THUNK(func)                                 \
 754         extern typeof(func) __raw_callee_save_##func;                   \
 755                                                                         \
 756         asm(".pushsection .text;"                                       \
 757             ".globl __raw_callee_save_" #func " ; "                     \
 758             "__raw_callee_save_" #func ": "                             \
 759             PV_SAVE_ALL_CALLER_REGS                                     \
 760             "call " #func ";"                                           \
 761             PV_RESTORE_ALL_CALLER_REGS                                  \
 762             "ret;"                                                      \
 763             ".popsection")
 764
 765 /* Get a reference to a callee-save function */
 766 #define PV_CALLEE_SAVE(func)                                            \
 767         ((struct paravirt_callee_save) { __raw_callee_save_##func })
 768
 769 /* Promise that "func" already uses the right calling convention */
 770 #define __PV_IS_CALLEE_SAVE(func)                       \
 771         ((struct paravirt_callee_save) { func })
 772
 773 static inline notrace unsigned long arch_local_save_flags(void)
 774 {
 775         return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
 776 }
 777
 778 static inline notrace void arch_local_irq_restore(unsigned long f)
 779 {
 780         PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
 781 }
 782
 783 static inline notrace void arch_local_irq_disable(void)
 784 {
 785         PVOP_VCALLEE0(pv_irq_ops.irq_disable);
 786 }
 787
 788 static inline notrace void arch_local_irq_enable(void)
 789 {
 790         PVOP_VCALLEE0(pv_irq_ops.irq_enable);
 791 }
 792
 793 static inline notrace unsigned long arch_local_irq_save(void)
 794 {
 795         unsigned long f;
 796
 797         f = arch_local_save_flags();
 798         arch_local_irq_disable();
 799         return f;
 800 }
 801
 802
 803 /* Make sure as little as possible of this mess escapes. */
 804 #undef PARAVIRT_CALL
 805 #undef __PVOP_CALL
 806 #undef __PVOP_VCALL
 807 #undef PVOP_VCALL0
 808 #undef PVOP_CALL0
 809 #undef PVOP_VCALL1
 810 #undef PVOP_CALL1
 811 #undef PVOP_VCALL2
 812 #undef PVOP_CALL2
 813 #undef PVOP_VCALL3
 814 #undef PVOP_CALL3
 815 #undef PVOP_VCALL4
 816 #undef PVOP_CALL4
 817
 818 extern void default_banner(void);
 819
 820 #else  /* __ASSEMBLY__ */
 821
 822 #define _PVSITE(ptype, clobbers, ops, word, algn)       \
 823 771:;                                           \
 824         ops;                                    \
 825 772:;                                           \
 826         .pushsection .parainstructions,"a";     \
 827          .align algn;                           \
 828          word 771b;                             \
 829          .byte ptype;                           \
 830          .byte 772b-771b;                       \
 831          .short clobbers;                       \
 832         .popsection
 833
 834
 835 #define COND_PUSH(set, mask, reg)                       \
 836         .if ((~(set)) & mask); push %reg; .endif
 837 #define COND_POP(set, mask, reg)                        \
 838         .if ((~(set)) & mask); pop %reg; .endif
 839
 840 #ifdef CONFIG_X86_64
 841
 842 #define PV_SAVE_REGS(set)                       \
 843         COND_PUSH(set, CLBR_RAX, rax);          \
 844         COND_PUSH(set, CLBR_RCX, rcx);          \
 845         COND_PUSH(set, CLBR_RDX, rdx);          \
 846         COND_PUSH(set, CLBR_RSI, rsi);          \
 847         COND_PUSH(set, CLBR_RDI, rdi);          \
 848         COND_PUSH(set, CLBR_R8, r8);            \
 849         COND_PUSH(set, CLBR_R9, r9);            \
 850         COND_PUSH(set, CLBR_R10, r10);          \
 851         COND_PUSH(set, CLBR_R11, r11)
 852 #define PV_RESTORE_REGS(set)                    \
 853         COND_POP(set, CLBR_R11, r11);           \
 854         COND_POP(set, CLBR_R10, r10);           \
 855         COND_POP(set, CLBR_R9, r9);             \
 856         COND_POP(set, CLBR_R8, r8);             \
 857         COND_POP(set, CLBR_RDI, rdi);           \
 858         COND_POP(set, CLBR_RSI, rsi);           \
 859         COND_POP(set, CLBR_RDX, rdx);           \
 860         COND_POP(set, CLBR_RCX, rcx);           \
 861         COND_POP(set, CLBR_RAX, rax)
 862
 863 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
 864 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
 865 #define PARA_INDIRECT(addr)     *addr(%rip)
 866 #else
 867 #define PV_SAVE_REGS(set)                       \
 868         COND_PUSH(set, CLBR_EAX, eax);          \
 869         COND_PUSH(set, CLBR_EDI, edi);          \
 870         COND_PUSH(set, CLBR_ECX, ecx);          \
 871         COND_PUSH(set, CLBR_EDX, edx)
 872 #define PV_RESTORE_REGS(set)                    \
 873         COND_POP(set, CLBR_EDX, edx);           \
 874         COND_POP(set, CLBR_ECX, ecx);           \
 875         COND_POP(set, CLBR_EDI, edi);           \
 876         COND_POP(set, CLBR_EAX, eax)
 877
 878 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
 879 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
 880 #define PARA_INDIRECT(addr)     *%cs:addr
 881 #endif
 882
 883 #define INTERRUPT_RETURN                                                \
 884         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,       \
 885                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
 886
 887 #define DISABLE_INTERRUPTS(clobbers)                                    \
 888         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
 889                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
 890                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);    \
 891                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
 892
 893 #define ENABLE_INTERRUPTS(clobbers)                                     \
 894         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,  \
 895                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
 896                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);     \
 897                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
 898
 899 #define USERGS_SYSRET32                                                 \
 900         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),       \
 901                   CLBR_NONE,                                            \
 902                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
 903
 904 #ifdef CONFIG_X86_32
 905 #define GET_CR0_INTO_EAX                                \
 906         push %ecx; push %edx;                           \
 907         call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
 908         pop %edx; pop %ecx
 909
 910 #define ENABLE_INTERRUPTS_SYSEXIT                                       \
 911         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
 912                   CLBR_NONE,                                            \
 913                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
 914
 915
 916 #else   /* !CONFIG_X86_32 */
 917
 918 /*
 919  * If swapgs is used while the userspace stack is still current,
 920  * there's no way to call a pvop.  The PV replacement *must* be
 921  * inlined, or the swapgs instruction must be trapped and emulated.
 922  */
 923 #define SWAPGS_UNSAFE_STACK                                             \
 924         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
 925                   swapgs)
 926
 927 /*
 928  * Note: swapgs is very special, and in practise is either going to be
 929  * implemented with a single "swapgs" instruction or something very
 930  * special.  Either way, we don't need to save any registers for
 931  * it.
 932  */
 933 #define SWAPGS                                                          \
 934         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
 935                   call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)          \
 936                  )
 937
 938 #define GET_CR2_INTO_RAX                                \
 939         call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2)
 940
 941 #define PARAVIRT_ADJUST_EXCEPTION_FRAME                                 \
 942         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
 943                   CLBR_NONE,                                            \
 944                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
 945
 946 #define USERGS_SYSRET64                                                 \
 947         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),       \
 948                   CLBR_NONE,                                            \
 949                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
 950 #endif  /* CONFIG_X86_32 */
 951
 952 #endif /* __ASSEMBLY__ */
 953 #else  /* CONFIG_PARAVIRT */
 954 # define default_banner x86_init_noop
 955 #ifndef __ASSEMBLY__
 956 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
 957                                           struct mm_struct *mm)
 958 {
 959 }
 960
 961 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
 962 {
 963 }
 964 #endif /* __ASSEMBLY__ */
 965 #endif /* !CONFIG_PARAVIRT */
 966 #endif /* _ASM_X86_PARAVIRT_H */