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 #ifdef CONFIG_SMP
 289 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
 290                                     unsigned long start_esp)
 291 {
 292         PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
 293                     phys_apicid, start_eip, start_esp);
 294 }
 295 #endif
 296
 297 static inline void paravirt_activate_mm(struct mm_struct *prev,
 298                                         struct mm_struct *next)
 299 {
 300         PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
 301 }
 302
 303 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
 304                                           struct mm_struct *mm)
 305 {
 306         PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
 307 }
 308
 309 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
 310 {
 311         PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
 312 }
 313
 314 static inline void __flush_tlb(void)
 315 {
 316         PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
 317 }
 318 static inline void __flush_tlb_global(void)
 319 {
 320         PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
 321 }
 322 static inline void __flush_tlb_single(unsigned long addr)
 323 {
 324         PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
 325 }
 326
 327 static inline void flush_tlb_others(const struct cpumask *cpumask,
 328                                     struct mm_struct *mm,
 329                                     unsigned long start,
 330                                     unsigned long end)
 331 {
 332         PVOP_VCALL4(pv_mmu_ops.flush_tlb_others, cpumask, mm, start, end);
 333 }
 334
 335 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
 336 {
 337         return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
 338 }
 339
 340 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
 341 {
 342         PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
 343 }
 344
 345 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
 346 {
 347         PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
 348 }
 349 static inline void paravirt_release_pte(unsigned long pfn)
 350 {
 351         PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
 352 }
 353
 354 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
 355 {
 356         PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
 357 }
 358
 359 static inline void paravirt_release_pmd(unsigned long pfn)
 360 {
 361         PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
 362 }
 363
 364 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
 365 {
 366         PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
 367 }
 368 static inline void paravirt_release_pud(unsigned long pfn)
 369 {
 370         PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
 371 }
 372
 373 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
 374                               pte_t *ptep)
 375 {
 376         PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
 377 }
 378 static inline void pmd_update(struct mm_struct *mm, unsigned long addr,
 379                               pmd_t *pmdp)
 380 {
 381         PVOP_VCALL3(pv_mmu_ops.pmd_update, mm, addr, pmdp);
 382 }
 383
 384 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
 385                                     pte_t *ptep)
 386 {
 387         PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
 388 }
 389
 390 static inline void pmd_update_defer(struct mm_struct *mm, unsigned long addr,
 391                                     pmd_t *pmdp)
 392 {
 393         PVOP_VCALL3(pv_mmu_ops.pmd_update_defer, mm, addr, pmdp);
 394 }
 395
 396 static inline pte_t __pte(pteval_t val)
 397 {
 398         pteval_t ret;
 399
 400         if (sizeof(pteval_t) > sizeof(long))
 401                 ret = PVOP_CALLEE2(pteval_t,
 402                                    pv_mmu_ops.make_pte,
 403                                    val, (u64)val >> 32);
 404         else
 405                 ret = PVOP_CALLEE1(pteval_t,
 406                                    pv_mmu_ops.make_pte,
 407                                    val);
 408
 409         return (pte_t) { .pte = ret };
 410 }
 411
 412 static inline pteval_t pte_val(pte_t pte)
 413 {
 414         pteval_t ret;
 415
 416         if (sizeof(pteval_t) > sizeof(long))
 417                 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
 418                                    pte.pte, (u64)pte.pte >> 32);
 419         else
 420                 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
 421                                    pte.pte);
 422
 423         return ret;
 424 }
 425
 426 static inline pgd_t __pgd(pgdval_t val)
 427 {
 428         pgdval_t ret;
 429
 430         if (sizeof(pgdval_t) > sizeof(long))
 431                 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
 432                                    val, (u64)val >> 32);
 433         else
 434                 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
 435                                    val);
 436
 437         return (pgd_t) { ret };
 438 }
 439
 440 static inline pgdval_t pgd_val(pgd_t pgd)
 441 {
 442         pgdval_t ret;
 443
 444         if (sizeof(pgdval_t) > sizeof(long))
 445                 ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
 446                                     pgd.pgd, (u64)pgd.pgd >> 32);
 447         else
 448                 ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
 449                                     pgd.pgd);
 450
 451         return ret;
 452 }
 453
 454 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 455 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
 456                                            pte_t *ptep)
 457 {
 458         pteval_t ret;
 459
 460         ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
 461                          mm, addr, ptep);
 462
 463         return (pte_t) { .pte = ret };
 464 }
 465
 466 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
 467                                            pte_t *ptep, pte_t pte)
 468 {
 469         if (sizeof(pteval_t) > sizeof(long))
 470                 /* 5 arg words */
 471                 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
 472         else
 473                 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
 474                             mm, addr, ptep, pte.pte);
 475 }
 476
 477 static inline void set_pte(pte_t *ptep, pte_t pte)
 478 {
 479         if (sizeof(pteval_t) > sizeof(long))
 480                 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
 481                             pte.pte, (u64)pte.pte >> 32);
 482         else
 483                 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
 484                             pte.pte);
 485 }
 486
 487 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
 488                               pte_t *ptep, pte_t pte)
 489 {
 490         if (sizeof(pteval_t) > sizeof(long))
 491                 /* 5 arg words */
 492                 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
 493         else
 494                 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
 495 }
 496
 497 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
 498                               pmd_t *pmdp, pmd_t pmd)
 499 {
 500         if (sizeof(pmdval_t) > sizeof(long))
 501                 /* 5 arg words */
 502                 pv_mmu_ops.set_pmd_at(mm, addr, pmdp, pmd);
 503         else
 504                 PVOP_VCALL4(pv_mmu_ops.set_pmd_at, mm, addr, pmdp,
 505                             native_pmd_val(pmd));
 506 }
 507
 508 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
 509 {
 510         pmdval_t val = native_pmd_val(pmd);
 511
 512         if (sizeof(pmdval_t) > sizeof(long))
 513                 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
 514         else
 515                 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
 516 }
 517
 518 #if CONFIG_PGTABLE_LEVELS >= 3
 519 static inline pmd_t __pmd(pmdval_t val)
 520 {
 521         pmdval_t ret;
 522
 523         if (sizeof(pmdval_t) > sizeof(long))
 524                 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
 525                                    val, (u64)val >> 32);
 526         else
 527                 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
 528                                    val);
 529
 530         return (pmd_t) { ret };
 531 }
 532
 533 static inline pmdval_t pmd_val(pmd_t pmd)
 534 {
 535         pmdval_t ret;
 536
 537         if (sizeof(pmdval_t) > sizeof(long))
 538                 ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
 539                                     pmd.pmd, (u64)pmd.pmd >> 32);
 540         else
 541                 ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
 542                                     pmd.pmd);
 543
 544         return ret;
 545 }
 546
 547 static inline void set_pud(pud_t *pudp, pud_t pud)
 548 {
 549         pudval_t val = native_pud_val(pud);
 550
 551         if (sizeof(pudval_t) > sizeof(long))
 552                 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
 553                             val, (u64)val >> 32);
 554         else
 555                 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
 556                             val);
 557 }
 558 #if CONFIG_PGTABLE_LEVELS == 4
 559 static inline pud_t __pud(pudval_t val)
 560 {
 561         pudval_t ret;
 562
 563         if (sizeof(pudval_t) > sizeof(long))
 564                 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
 565                                    val, (u64)val >> 32);
 566         else
 567                 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
 568                                    val);
 569
 570         return (pud_t) { ret };
 571 }
 572
 573 static inline pudval_t pud_val(pud_t pud)
 574 {
 575         pudval_t ret;
 576
 577         if (sizeof(pudval_t) > sizeof(long))
 578                 ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
 579                                     pud.pud, (u64)pud.pud >> 32);
 580         else
 581                 ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
 582                                     pud.pud);
 583
 584         return ret;
 585 }
 586
 587 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
 588 {
 589         pgdval_t val = native_pgd_val(pgd);
 590
 591         if (sizeof(pgdval_t) > sizeof(long))
 592                 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
 593                             val, (u64)val >> 32);
 594         else
 595                 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
 596                             val);
 597 }
 598
 599 static inline void pgd_clear(pgd_t *pgdp)
 600 {
 601         set_pgd(pgdp, __pgd(0));
 602 }
 603
 604 static inline void pud_clear(pud_t *pudp)
 605 {
 606         set_pud(pudp, __pud(0));
 607 }
 608
 609 #endif  /* CONFIG_PGTABLE_LEVELS == 4 */
 610
 611 #endif  /* CONFIG_PGTABLE_LEVELS >= 3 */
 612
 613 #ifdef CONFIG_X86_PAE
 614 /* Special-case pte-setting operations for PAE, which can't update a
 615    64-bit pte atomically */
 616 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
 617 {
 618         PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
 619                     pte.pte, pte.pte >> 32);
 620 }
 621
 622 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
 623                              pte_t *ptep)
 624 {
 625         PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
 626 }
 627
 628 static inline void pmd_clear(pmd_t *pmdp)
 629 {
 630         PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
 631 }
 632 #else  /* !CONFIG_X86_PAE */
 633 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
 634 {
 635         set_pte(ptep, pte);
 636 }
 637
 638 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
 639                              pte_t *ptep)
 640 {
 641         set_pte_at(mm, addr, ptep, __pte(0));
 642 }
 643
 644 static inline void pmd_clear(pmd_t *pmdp)
 645 {
 646         set_pmd(pmdp, __pmd(0));
 647 }
 648 #endif  /* CONFIG_X86_PAE */
 649
 650 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
 651 static inline void arch_start_context_switch(struct task_struct *prev)
 652 {
 653         PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
 654 }
 655
 656 static inline void arch_end_context_switch(struct task_struct *next)
 657 {
 658         PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
 659 }
 660
 661 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 662 static inline void arch_enter_lazy_mmu_mode(void)
 663 {
 664         PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
 665 }
 666
 667 static inline void arch_leave_lazy_mmu_mode(void)
 668 {
 669         PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
 670 }
 671
 672 static inline void arch_flush_lazy_mmu_mode(void)
 673 {
 674         PVOP_VCALL0(pv_mmu_ops.lazy_mode.flush);
 675 }
 676
 677 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
 678                                 phys_addr_t phys, pgprot_t flags)
 679 {
 680         pv_mmu_ops.set_fixmap(idx, phys, flags);
 681 }
 682
 683 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
 684
 685 #ifdef CONFIG_QUEUED_SPINLOCKS
 686
 687 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock,
 688                                                         u32 val)
 689 {
 690         PVOP_VCALL2(pv_lock_ops.queued_spin_lock_slowpath, lock, val);
 691 }
 692
 693 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock)
 694 {
 695         PVOP_VCALLEE1(pv_lock_ops.queued_spin_unlock, lock);
 696 }
 697
 698 static __always_inline void pv_wait(u8 *ptr, u8 val)
 699 {
 700         PVOP_VCALL2(pv_lock_ops.wait, ptr, val);
 701 }
 702
 703 static __always_inline void pv_kick(int cpu)
 704 {
 705         PVOP_VCALL1(pv_lock_ops.kick, cpu);
 706 }
 707
 708 #else /* !CONFIG_QUEUED_SPINLOCKS */
 709
 710 static __always_inline void __ticket_lock_spinning(struct arch_spinlock *lock,
 711                                                         __ticket_t ticket)
 712 {
 713         PVOP_VCALLEE2(pv_lock_ops.lock_spinning, lock, ticket);
 714 }
 715
 716 static __always_inline void __ticket_unlock_kick(struct arch_spinlock *lock,
 717                                                         __ticket_t ticket)
 718 {
 719         PVOP_VCALL2(pv_lock_ops.unlock_kick, lock, ticket);
 720 }
 721
 722 #endif /* CONFIG_QUEUED_SPINLOCKS */
 723
 724 #endif /* SMP && PARAVIRT_SPINLOCKS */
 725
 726 #ifdef CONFIG_X86_32
 727 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
 728 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
 729
 730 /* save and restore all caller-save registers, except return value */
 731 #define PV_SAVE_ALL_CALLER_REGS         "pushl %ecx;"
 732 #define PV_RESTORE_ALL_CALLER_REGS      "popl  %ecx;"
 733
 734 #define PV_FLAGS_ARG "0"
 735 #define PV_EXTRA_CLOBBERS
 736 #define PV_VEXTRA_CLOBBERS
 737 #else
 738 /* save and restore all caller-save registers, except return value */
 739 #define PV_SAVE_ALL_CALLER_REGS                                         \
 740         "push %rcx;"                                                    \
 741         "push %rdx;"                                                    \
 742         "push %rsi;"                                                    \
 743         "push %rdi;"                                                    \
 744         "push %r8;"                                                     \
 745         "push %r9;"                                                     \
 746         "push %r10;"                                                    \
 747         "push %r11;"
 748 #define PV_RESTORE_ALL_CALLER_REGS                                      \
 749         "pop %r11;"                                                     \
 750         "pop %r10;"                                                     \
 751         "pop %r9;"                                                      \
 752         "pop %r8;"                                                      \
 753         "pop %rdi;"                                                     \
 754         "pop %rsi;"                                                     \
 755         "pop %rdx;"                                                     \
 756         "pop %rcx;"
 757
 758 /* We save some registers, but all of them, that's too much. We clobber all
 759  * caller saved registers but the argument parameter */
 760 #define PV_SAVE_REGS "pushq %%rdi;"
 761 #define PV_RESTORE_REGS "popq %%rdi;"
 762 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
 763 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
 764 #define PV_FLAGS_ARG "D"
 765 #endif
 766
 767 /*
 768  * Generate a thunk around a function which saves all caller-save
 769  * registers except for the return value.  This allows C functions to
 770  * be called from assembler code where fewer than normal registers are
 771  * available.  It may also help code generation around calls from C
 772  * code if the common case doesn't use many registers.
 773  *
 774  * When a callee is wrapped in a thunk, the caller can assume that all
 775  * arg regs and all scratch registers are preserved across the
 776  * call. The return value in rax/eax will not be saved, even for void
 777  * functions.
 778  */
 779 #define PV_CALLEE_SAVE_REGS_THUNK(func)                                 \
 780         extern typeof(func) __raw_callee_save_##func;                   \
 781                                                                         \
 782         asm(".pushsection .text;"                                       \
 783             ".globl __raw_callee_save_" #func " ; "                     \
 784             "__raw_callee_save_" #func ": "                             \
 785             PV_SAVE_ALL_CALLER_REGS                                     \
 786             "call " #func ";"                                           \
 787             PV_RESTORE_ALL_CALLER_REGS                                  \
 788             "ret;"                                                      \
 789             ".popsection")
 790
 791 /* Get a reference to a callee-save function */
 792 #define PV_CALLEE_SAVE(func)                                            \
 793         ((struct paravirt_callee_save) { __raw_callee_save_##func })
 794
 795 /* Promise that "func" already uses the right calling convention */
 796 #define __PV_IS_CALLEE_SAVE(func)                       \
 797         ((struct paravirt_callee_save) { func })
 798
 799 static inline notrace unsigned long arch_local_save_flags(void)
 800 {
 801         return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
 802 }
 803
 804 static inline notrace void arch_local_irq_restore(unsigned long f)
 805 {
 806         PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
 807 }
 808
 809 static inline notrace void arch_local_irq_disable(void)
 810 {
 811         PVOP_VCALLEE0(pv_irq_ops.irq_disable);
 812 }
 813
 814 static inline notrace void arch_local_irq_enable(void)
 815 {
 816         PVOP_VCALLEE0(pv_irq_ops.irq_enable);
 817 }
 818
 819 static inline notrace unsigned long arch_local_irq_save(void)
 820 {
 821         unsigned long f;
 822
 823         f = arch_local_save_flags();
 824         arch_local_irq_disable();
 825         return f;
 826 }
 827
 828
 829 /* Make sure as little as possible of this mess escapes. */
 830 #undef PARAVIRT_CALL
 831 #undef __PVOP_CALL
 832 #undef __PVOP_VCALL
 833 #undef PVOP_VCALL0
 834 #undef PVOP_CALL0
 835 #undef PVOP_VCALL1
 836 #undef PVOP_CALL1
 837 #undef PVOP_VCALL2
 838 #undef PVOP_CALL2
 839 #undef PVOP_VCALL3
 840 #undef PVOP_CALL3
 841 #undef PVOP_VCALL4
 842 #undef PVOP_CALL4
 843
 844 extern void default_banner(void);
 845
 846 #else  /* __ASSEMBLY__ */
 847
 848 #define _PVSITE(ptype, clobbers, ops, word, algn)       \
 849 771:;                                           \
 850         ops;                                    \
 851 772:;                                           \
 852         .pushsection .parainstructions,"a";     \
 853          .align algn;                           \
 854          word 771b;                             \
 855          .byte ptype;                           \
 856          .byte 772b-771b;                       \
 857          .short clobbers;                       \
 858         .popsection
 859
 860
 861 #define COND_PUSH(set, mask, reg)                       \
 862         .if ((~(set)) & mask); push %reg; .endif
 863 #define COND_POP(set, mask, reg)                        \
 864         .if ((~(set)) & mask); pop %reg; .endif
 865
 866 #ifdef CONFIG_X86_64
 867
 868 #define PV_SAVE_REGS(set)                       \
 869         COND_PUSH(set, CLBR_RAX, rax);          \
 870         COND_PUSH(set, CLBR_RCX, rcx);          \
 871         COND_PUSH(set, CLBR_RDX, rdx);          \
 872         COND_PUSH(set, CLBR_RSI, rsi);          \
 873         COND_PUSH(set, CLBR_RDI, rdi);          \
 874         COND_PUSH(set, CLBR_R8, r8);            \
 875         COND_PUSH(set, CLBR_R9, r9);            \
 876         COND_PUSH(set, CLBR_R10, r10);          \
 877         COND_PUSH(set, CLBR_R11, r11)
 878 #define PV_RESTORE_REGS(set)                    \
 879         COND_POP(set, CLBR_R11, r11);           \
 880         COND_POP(set, CLBR_R10, r10);           \
 881         COND_POP(set, CLBR_R9, r9);             \
 882         COND_POP(set, CLBR_R8, r8);             \
 883         COND_POP(set, CLBR_RDI, rdi);           \
 884         COND_POP(set, CLBR_RSI, rsi);           \
 885         COND_POP(set, CLBR_RDX, rdx);           \
 886         COND_POP(set, CLBR_RCX, rcx);           \
 887         COND_POP(set, CLBR_RAX, rax)
 888
 889 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
 890 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
 891 #define PARA_INDIRECT(addr)     *addr(%rip)
 892 #else
 893 #define PV_SAVE_REGS(set)                       \
 894         COND_PUSH(set, CLBR_EAX, eax);          \
 895         COND_PUSH(set, CLBR_EDI, edi);          \
 896         COND_PUSH(set, CLBR_ECX, ecx);          \
 897         COND_PUSH(set, CLBR_EDX, edx)
 898 #define PV_RESTORE_REGS(set)                    \
 899         COND_POP(set, CLBR_EDX, edx);           \
 900         COND_POP(set, CLBR_ECX, ecx);           \
 901         COND_POP(set, CLBR_EDI, edi);           \
 902         COND_POP(set, CLBR_EAX, eax)
 903
 904 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
 905 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
 906 #define PARA_INDIRECT(addr)     *%cs:addr
 907 #endif
 908
 909 #define INTERRUPT_RETURN                                                \
 910         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,       \
 911                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
 912
 913 #define DISABLE_INTERRUPTS(clobbers)                                    \
 914         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
 915                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
 916                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);    \
 917                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
 918
 919 #define ENABLE_INTERRUPTS(clobbers)                                     \
 920         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,  \
 921                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
 922                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);     \
 923                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
 924
 925 #define USERGS_SYSRET32                                                 \
 926         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),       \
 927                   CLBR_NONE,                                            \
 928                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
 929
 930 #ifdef CONFIG_X86_32
 931 #define GET_CR0_INTO_EAX                                \
 932         push %ecx; push %edx;                           \
 933         call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
 934         pop %edx; pop %ecx
 935
 936 #define ENABLE_INTERRUPTS_SYSEXIT                                       \
 937         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
 938                   CLBR_NONE,                                            \
 939                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
 940
 941
 942 #else   /* !CONFIG_X86_32 */
 943
 944 /*
 945  * If swapgs is used while the userspace stack is still current,
 946  * there's no way to call a pvop.  The PV replacement *must* be
 947  * inlined, or the swapgs instruction must be trapped and emulated.
 948  */
 949 #define SWAPGS_UNSAFE_STACK                                             \
 950         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
 951                   swapgs)
 952
 953 /*
 954  * Note: swapgs is very special, and in practise is either going to be
 955  * implemented with a single "swapgs" instruction or something very
 956  * special.  Either way, we don't need to save any registers for
 957  * it.
 958  */
 959 #define SWAPGS                                                          \
 960         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
 961                   call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)          \
 962                  )
 963
 964 #define GET_CR2_INTO_RAX                                \
 965         call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2)
 966
 967 #define PARAVIRT_ADJUST_EXCEPTION_FRAME                                 \
 968         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
 969                   CLBR_NONE,                                            \
 970                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
 971
 972 #define USERGS_SYSRET64                                                 \
 973         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),       \
 974                   CLBR_NONE,                                            \
 975                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
 976 #endif  /* CONFIG_X86_32 */
 977
 978 #endif /* __ASSEMBLY__ */
 979 #else  /* CONFIG_PARAVIRT */
 980 # define default_banner x86_init_noop
 981 #ifndef __ASSEMBLY__
 982 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
 983                                           struct mm_struct *mm)
 984 {
 985 }
 986
 987 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
 988 {
 989 }
 990 #endif /* __ASSEMBLY__ */
 991 #endif /* !CONFIG_PARAVIRT */
 992 #endif /* _ASM_X86_PARAVIRT_H */