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