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/types.h>
  14 #include <linux/cpumask.h>
  15
  16 static inline int paravirt_enabled(void)
  17 {
  18         return pv_info.paravirt_enabled;
  19 }
  20
  21 static inline void load_sp0(struct tss_struct *tss,
  22                              struct thread_struct *thread)
  23 {
  24         PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
  25 }
  26
  27 static inline unsigned long get_wallclock(void)
  28 {
  29         return PVOP_CALL0(unsigned long, pv_time_ops.get_wallclock);
  30 }
  31
  32 static inline int set_wallclock(unsigned long nowtime)
  33 {
  34         return PVOP_CALL1(int, pv_time_ops.set_wallclock, nowtime);
  35 }
  36
  37 /* The paravirtualized CPUID instruction. */
  38 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
  39                            unsigned int *ecx, unsigned int *edx)
  40 {
  41         PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
  42 }
  43
  44 /*
  45  * These special macros can be used to get or set a debugging register
  46  */
  47 static inline unsigned long paravirt_get_debugreg(int reg)
  48 {
  49         return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
  50 }
  51 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
  52 static inline void set_debugreg(unsigned long val, int reg)
  53 {
  54         PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
  55 }
  56
  57 static inline void clts(void)
  58 {
  59         PVOP_VCALL0(pv_cpu_ops.clts);
  60 }
  61
  62 static inline unsigned long read_cr0(void)
  63 {
  64         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
  65 }
  66
  67 static inline void write_cr0(unsigned long x)
  68 {
  69         PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
  70 }
  71
  72 static inline unsigned long read_cr2(void)
  73 {
  74         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
  75 }
  76
  77 static inline void write_cr2(unsigned long x)
  78 {
  79         PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
  80 }
  81
  82 static inline unsigned long read_cr3(void)
  83 {
  84         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
  85 }
  86
  87 static inline void write_cr3(unsigned long x)
  88 {
  89         PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
  90 }
  91
  92 static inline unsigned long read_cr4(void)
  93 {
  94         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
  95 }
  96 static inline unsigned long read_cr4_safe(void)
  97 {
  98         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
  99 }
 100
 101 static inline void write_cr4(unsigned long x)
 102 {
 103         PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
 104 }
 105
 106 #ifdef CONFIG_X86_64
 107 static inline unsigned long read_cr8(void)
 108 {
 109         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
 110 }
 111
 112 static inline void write_cr8(unsigned long x)
 113 {
 114         PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
 115 }
 116 #endif
 117
 118 static inline void raw_safe_halt(void)
 119 {
 120         PVOP_VCALL0(pv_irq_ops.safe_halt);
 121 }
 122
 123 static inline void halt(void)
 124 {
 125         PVOP_VCALL0(pv_irq_ops.safe_halt);
 126 }
 127
 128 static inline void wbinvd(void)
 129 {
 130         PVOP_VCALL0(pv_cpu_ops.wbinvd);
 131 }
 132
 133 #define get_kernel_rpl()  (pv_info.kernel_rpl)
 134
 135 static inline u64 paravirt_read_msr(unsigned msr, int *err)
 136 {
 137         return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
 138 }
 139 static inline u64 paravirt_read_msr_amd(unsigned msr, int *err)
 140 {
 141         return PVOP_CALL2(u64, pv_cpu_ops.read_msr_amd, msr, err);
 142 }
 143 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
 144 {
 145         return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
 146 }
 147
 148 /* These should all do BUG_ON(_err), but our headers are too tangled. */
 149 #define rdmsr(msr, val1, val2)                  \
 150 do {                                            \
 151         int _err;                               \
 152         u64 _l = paravirt_read_msr(msr, &_err); \
 153         val1 = (u32)_l;                         \
 154         val2 = _l >> 32;                        \
 155 } while (0)
 156
 157 #define wrmsr(msr, val1, val2)                  \
 158 do {                                            \
 159         paravirt_write_msr(msr, val1, val2);    \
 160 } while (0)
 161
 162 #define rdmsrl(msr, val)                        \
 163 do {                                            \
 164         int _err;                               \
 165         val = paravirt_read_msr(msr, &_err);    \
 166 } while (0)
 167
 168 #define wrmsrl(msr, val)        wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
 169 #define wrmsr_safe(msr, a, b)   paravirt_write_msr(msr, a, b)
 170
 171 /* rdmsr with exception handling */
 172 #define rdmsr_safe(msr, a, b)                   \
 173 ({                                              \
 174         int _err;                               \
 175         u64 _l = paravirt_read_msr(msr, &_err); \
 176         (*a) = (u32)_l;                         \
 177         (*b) = _l >> 32;                        \
 178         _err;                                   \
 179 })
 180
 181 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
 182 {
 183         int err;
 184
 185         *p = paravirt_read_msr(msr, &err);
 186         return err;
 187 }
 188 static inline int rdmsrl_amd_safe(unsigned msr, unsigned long long *p)
 189 {
 190         int err;
 191
 192         *p = paravirt_read_msr_amd(msr, &err);
 193         return err;
 194 }
 195
 196 static inline u64 paravirt_read_tsc(void)
 197 {
 198         return PVOP_CALL0(u64, pv_cpu_ops.read_tsc);
 199 }
 200
 201 #define rdtscl(low)                             \
 202 do {                                            \
 203         u64 _l = paravirt_read_tsc();           \
 204         low = (int)_l;                          \
 205 } while (0)
 206
 207 #define rdtscll(val) (val = paravirt_read_tsc())
 208
 209 static inline unsigned long long paravirt_sched_clock(void)
 210 {
 211         return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
 212 }
 213 #define calibrate_tsc() (pv_time_ops.get_tsc_khz())
 214
 215 static inline unsigned long long paravirt_read_pmc(int counter)
 216 {
 217         return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
 218 }
 219
 220 #define rdpmc(counter, low, high)               \
 221 do {                                            \
 222         u64 _l = paravirt_read_pmc(counter);    \
 223         low = (u32)_l;                          \
 224         high = _l >> 32;                        \
 225 } while (0)
 226
 227 static inline unsigned long long paravirt_rdtscp(unsigned int *aux)
 228 {
 229         return PVOP_CALL1(u64, pv_cpu_ops.read_tscp, aux);
 230 }
 231
 232 #define rdtscp(low, high, aux)                          \
 233 do {                                                    \
 234         int __aux;                                      \
 235         unsigned long __val = paravirt_rdtscp(&__aux);  \
 236         (low) = (u32)__val;                             \
 237         (high) = (u32)(__val >> 32);                    \
 238         (aux) = __aux;                                  \
 239 } while (0)
 240
 241 #define rdtscpll(val, aux)                              \
 242 do {                                                    \
 243         unsigned long __aux;                            \
 244         val = paravirt_rdtscp(&__aux);                  \
 245         (aux) = __aux;                                  \
 246 } while (0)
 247
 248 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
 249 {
 250         PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
 251 }
 252
 253 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
 254 {
 255         PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
 256 }
 257
 258 static inline void load_TR_desc(void)
 259 {
 260         PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
 261 }
 262 static inline void load_gdt(const struct desc_ptr *dtr)
 263 {
 264         PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
 265 }
 266 static inline void load_idt(const struct desc_ptr *dtr)
 267 {
 268         PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
 269 }
 270 static inline void set_ldt(const void *addr, unsigned entries)
 271 {
 272         PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
 273 }
 274 static inline void store_gdt(struct desc_ptr *dtr)
 275 {
 276         PVOP_VCALL1(pv_cpu_ops.store_gdt, dtr);
 277 }
 278 static inline void store_idt(struct desc_ptr *dtr)
 279 {
 280         PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
 281 }
 282 static inline unsigned long paravirt_store_tr(void)
 283 {
 284         return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
 285 }
 286 #define store_tr(tr)    ((tr) = paravirt_store_tr())
 287 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
 288 {
 289         PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
 290 }
 291
 292 #ifdef CONFIG_X86_64
 293 static inline void load_gs_index(unsigned int gs)
 294 {
 295         PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
 296 }
 297 #endif
 298
 299 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
 300                                    const void *desc)
 301 {
 302         PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
 303 }
 304
 305 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
 306                                    void *desc, int type)
 307 {
 308         PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
 309 }
 310
 311 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
 312 {
 313         PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
 314 }
 315 static inline void set_iopl_mask(unsigned mask)
 316 {
 317         PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
 318 }
 319
 320 /* The paravirtualized I/O functions */
 321 static inline void slow_down_io(void)
 322 {
 323         pv_cpu_ops.io_delay();
 324 #ifdef REALLY_SLOW_IO
 325         pv_cpu_ops.io_delay();
 326         pv_cpu_ops.io_delay();
 327         pv_cpu_ops.io_delay();
 328 #endif
 329 }
 330
 331 #ifdef CONFIG_SMP
 332 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
 333                                     unsigned long start_esp)
 334 {
 335         PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
 336                     phys_apicid, start_eip, start_esp);
 337 }
 338 #endif
 339
 340 static inline void paravirt_activate_mm(struct mm_struct *prev,
 341                                         struct mm_struct *next)
 342 {
 343         PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
 344 }
 345
 346 static inline void arch_dup_mmap(struct mm_struct *oldmm,
 347                                  struct mm_struct *mm)
 348 {
 349         PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
 350 }
 351
 352 static inline void arch_exit_mmap(struct mm_struct *mm)
 353 {
 354         PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
 355 }
 356
 357 static inline void __flush_tlb(void)
 358 {
 359         PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
 360 }
 361 static inline void __flush_tlb_global(void)
 362 {
 363         PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
 364 }
 365 static inline void __flush_tlb_single(unsigned long addr)
 366 {
 367         PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
 368 }
 369
 370 static inline void flush_tlb_others(const struct cpumask *cpumask,
 371                                     struct mm_struct *mm,
 372                                     unsigned long va)
 373 {
 374         PVOP_VCALL3(pv_mmu_ops.flush_tlb_others, cpumask, mm, va);
 375 }
 376
 377 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
 378 {
 379         return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
 380 }
 381
 382 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
 383 {
 384         PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
 385 }
 386
 387 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
 388 {
 389         PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
 390 }
 391 static inline void paravirt_release_pte(unsigned long pfn)
 392 {
 393         PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
 394 }
 395
 396 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
 397 {
 398         PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
 399 }
 400
 401 static inline void paravirt_alloc_pmd_clone(unsigned long pfn, unsigned long clonepfn,
 402                                             unsigned long start, unsigned long count)
 403 {
 404         PVOP_VCALL4(pv_mmu_ops.alloc_pmd_clone, pfn, clonepfn, start, count);
 405 }
 406 static inline void paravirt_release_pmd(unsigned long pfn)
 407 {
 408         PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
 409 }
 410
 411 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
 412 {
 413         PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
 414 }
 415 static inline void paravirt_release_pud(unsigned long pfn)
 416 {
 417         PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
 418 }
 419
 420 #ifdef CONFIG_HIGHPTE
 421 static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
 422 {
 423         unsigned long ret;
 424         ret = PVOP_CALL2(unsigned long, pv_mmu_ops.kmap_atomic_pte, page, type);
 425         return (void *)ret;
 426 }
 427 #endif
 428
 429 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
 430                               pte_t *ptep)
 431 {
 432         PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
 433 }
 434
 435 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
 436                                     pte_t *ptep)
 437 {
 438         PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
 439 }
 440
 441 static inline pte_t __pte(pteval_t val)
 442 {
 443         pteval_t ret;
 444
 445         if (sizeof(pteval_t) > sizeof(long))
 446                 ret = PVOP_CALLEE2(pteval_t,
 447                                    pv_mmu_ops.make_pte,
 448                                    val, (u64)val >> 32);
 449         else
 450                 ret = PVOP_CALLEE1(pteval_t,
 451                                    pv_mmu_ops.make_pte,
 452                                    val);
 453
 454         return (pte_t) { .pte = ret };
 455 }
 456
 457 static inline pteval_t pte_val(pte_t pte)
 458 {
 459         pteval_t ret;
 460
 461         if (sizeof(pteval_t) > sizeof(long))
 462                 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
 463                                    pte.pte, (u64)pte.pte >> 32);
 464         else
 465                 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
 466                                    pte.pte);
 467
 468         return ret;
 469 }
 470
 471 static inline pgd_t __pgd(pgdval_t val)
 472 {
 473         pgdval_t ret;
 474
 475         if (sizeof(pgdval_t) > sizeof(long))
 476                 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
 477                                    val, (u64)val >> 32);
 478         else
 479                 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
 480                                    val);
 481
 482         return (pgd_t) { ret };
 483 }
 484
 485 static inline pgdval_t pgd_val(pgd_t pgd)
 486 {
 487         pgdval_t ret;
 488
 489         if (sizeof(pgdval_t) > sizeof(long))
 490                 ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
 491                                     pgd.pgd, (u64)pgd.pgd >> 32);
 492         else
 493                 ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
 494                                     pgd.pgd);
 495
 496         return ret;
 497 }
 498
 499 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
 500 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
 501                                            pte_t *ptep)
 502 {
 503         pteval_t ret;
 504
 505         ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
 506                          mm, addr, ptep);
 507
 508         return (pte_t) { .pte = ret };
 509 }
 510
 511 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
 512                                            pte_t *ptep, pte_t pte)
 513 {
 514         if (sizeof(pteval_t) > sizeof(long))
 515                 /* 5 arg words */
 516                 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
 517         else
 518                 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
 519                             mm, addr, ptep, pte.pte);
 520 }
 521
 522 static inline void set_pte(pte_t *ptep, pte_t pte)
 523 {
 524         if (sizeof(pteval_t) > sizeof(long))
 525                 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
 526                             pte.pte, (u64)pte.pte >> 32);
 527         else
 528                 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
 529                             pte.pte);
 530 }
 531
 532 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
 533                               pte_t *ptep, pte_t pte)
 534 {
 535         if (sizeof(pteval_t) > sizeof(long))
 536                 /* 5 arg words */
 537                 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
 538         else
 539                 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
 540 }
 541
 542 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
 543 {
 544         pmdval_t val = native_pmd_val(pmd);
 545
 546         if (sizeof(pmdval_t) > sizeof(long))
 547                 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
 548         else
 549                 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
 550 }
 551
 552 #if PAGETABLE_LEVELS >= 3
 553 static inline pmd_t __pmd(pmdval_t val)
 554 {
 555         pmdval_t ret;
 556
 557         if (sizeof(pmdval_t) > sizeof(long))
 558                 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
 559                                    val, (u64)val >> 32);
 560         else
 561                 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
 562                                    val);
 563
 564         return (pmd_t) { ret };
 565 }
 566
 567 static inline pmdval_t pmd_val(pmd_t pmd)
 568 {
 569         pmdval_t ret;
 570
 571         if (sizeof(pmdval_t) > sizeof(long))
 572                 ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
 573                                     pmd.pmd, (u64)pmd.pmd >> 32);
 574         else
 575                 ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
 576                                     pmd.pmd);
 577
 578         return ret;
 579 }
 580
 581 static inline void set_pud(pud_t *pudp, pud_t pud)
 582 {
 583         pudval_t val = native_pud_val(pud);
 584
 585         if (sizeof(pudval_t) > sizeof(long))
 586                 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
 587                             val, (u64)val >> 32);
 588         else
 589                 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
 590                             val);
 591 }
 592 #if PAGETABLE_LEVELS == 4
 593 static inline pud_t __pud(pudval_t val)
 594 {
 595         pudval_t ret;
 596
 597         if (sizeof(pudval_t) > sizeof(long))
 598                 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
 599                                    val, (u64)val >> 32);
 600         else
 601                 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
 602                                    val);
 603
 604         return (pud_t) { ret };
 605 }
 606
 607 static inline pudval_t pud_val(pud_t pud)
 608 {
 609         pudval_t ret;
 610
 611         if (sizeof(pudval_t) > sizeof(long))
 612                 ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
 613                                     pud.pud, (u64)pud.pud >> 32);
 614         else
 615                 ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
 616                                     pud.pud);
 617
 618         return ret;
 619 }
 620
 621 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
 622 {
 623         pgdval_t val = native_pgd_val(pgd);
 624
 625         if (sizeof(pgdval_t) > sizeof(long))
 626                 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
 627                             val, (u64)val >> 32);
 628         else
 629                 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
 630                             val);
 631 }
 632
 633 static inline void pgd_clear(pgd_t *pgdp)
 634 {
 635         set_pgd(pgdp, __pgd(0));
 636 }
 637
 638 static inline void pud_clear(pud_t *pudp)
 639 {
 640         set_pud(pudp, __pud(0));
 641 }
 642
 643 #endif  /* PAGETABLE_LEVELS == 4 */
 644
 645 #endif  /* PAGETABLE_LEVELS >= 3 */
 646
 647 #ifdef CONFIG_X86_PAE
 648 /* Special-case pte-setting operations for PAE, which can't update a
 649    64-bit pte atomically */
 650 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
 651 {
 652         PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
 653                     pte.pte, pte.pte >> 32);
 654 }
 655
 656 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
 657                              pte_t *ptep)
 658 {
 659         PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
 660 }
 661
 662 static inline void pmd_clear(pmd_t *pmdp)
 663 {
 664         PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
 665 }
 666 #else  /* !CONFIG_X86_PAE */
 667 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
 668 {
 669         set_pte(ptep, pte);
 670 }
 671
 672 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
 673                              pte_t *ptep)
 674 {
 675         set_pte_at(mm, addr, ptep, __pte(0));
 676 }
 677
 678 static inline void pmd_clear(pmd_t *pmdp)
 679 {
 680         set_pmd(pmdp, __pmd(0));
 681 }
 682 #endif  /* CONFIG_X86_PAE */
 683
 684 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
 685 static inline void arch_start_context_switch(struct task_struct *prev)
 686 {
 687         PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
 688 }
 689
 690 static inline void arch_end_context_switch(struct task_struct *next)
 691 {
 692         PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
 693 }
 694
 695 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
 696 static inline void arch_enter_lazy_mmu_mode(void)
 697 {
 698         PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
 699 }
 700
 701 static inline void arch_leave_lazy_mmu_mode(void)
 702 {
 703         PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
 704 }
 705
 706 void arch_flush_lazy_mmu_mode(void);
 707
 708 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
 709                                 phys_addr_t phys, pgprot_t flags)
 710 {
 711         pv_mmu_ops.set_fixmap(idx, phys, flags);
 712 }
 713
 714 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
 715
 716 static inline int __raw_spin_is_locked(struct raw_spinlock *lock)
 717 {
 718         return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
 719 }
 720
 721 static inline int __raw_spin_is_contended(struct raw_spinlock *lock)
 722 {
 723         return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
 724 }
 725 #define __raw_spin_is_contended __raw_spin_is_contended
 726
 727 static __always_inline void __raw_spin_lock(struct raw_spinlock *lock)
 728 {
 729         PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
 730 }
 731
 732 static __always_inline void __raw_spin_lock_flags(struct raw_spinlock *lock,
 733                                                   unsigned long flags)
 734 {
 735         PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
 736 }
 737
 738 static __always_inline int __raw_spin_trylock(struct raw_spinlock *lock)
 739 {
 740         return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
 741 }
 742
 743 static __always_inline void __raw_spin_unlock(struct raw_spinlock *lock)
 744 {
 745         PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
 746 }
 747
 748 #endif
 749
 750 #ifdef CONFIG_X86_32
 751 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
 752 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
 753
 754 /* save and restore all caller-save registers, except return value */
 755 #define PV_SAVE_ALL_CALLER_REGS         "pushl %ecx;"
 756 #define PV_RESTORE_ALL_CALLER_REGS      "popl  %ecx;"
 757
 758 #define PV_FLAGS_ARG "0"
 759 #define PV_EXTRA_CLOBBERS
 760 #define PV_VEXTRA_CLOBBERS
 761 #else
 762 /* save and restore all caller-save registers, except return value */
 763 #define PV_SAVE_ALL_CALLER_REGS                                         \
 764         "push %rcx;"                                                    \
 765         "push %rdx;"                                                    \
 766         "push %rsi;"                                                    \
 767         "push %rdi;"                                                    \
 768         "push %r8;"                                                     \
 769         "push %r9;"                                                     \
 770         "push %r10;"                                                    \
 771         "push %r11;"
 772 #define PV_RESTORE_ALL_CALLER_REGS                                      \
 773         "pop %r11;"                                                     \
 774         "pop %r10;"                                                     \
 775         "pop %r9;"                                                      \
 776         "pop %r8;"                                                      \
 777         "pop %rdi;"                                                     \
 778         "pop %rsi;"                                                     \
 779         "pop %rdx;"                                                     \
 780         "pop %rcx;"
 781
 782 /* We save some registers, but all of them, that's too much. We clobber all
 783  * caller saved registers but the argument parameter */
 784 #define PV_SAVE_REGS "pushq %%rdi;"
 785 #define PV_RESTORE_REGS "popq %%rdi;"
 786 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
 787 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
 788 #define PV_FLAGS_ARG "D"
 789 #endif
 790
 791 /*
 792  * Generate a thunk around a function which saves all caller-save
 793  * registers except for the return value.  This allows C functions to
 794  * be called from assembler code where fewer than normal registers are
 795  * available.  It may also help code generation around calls from C
 796  * code if the common case doesn't use many registers.
 797  *
 798  * When a callee is wrapped in a thunk, the caller can assume that all
 799  * arg regs and all scratch registers are preserved across the
 800  * call. The return value in rax/eax will not be saved, even for void
 801  * functions.
 802  */
 803 #define PV_CALLEE_SAVE_REGS_THUNK(func)                                 \
 804         extern typeof(func) __raw_callee_save_##func;                   \
 805         static void *__##func##__ __used = func;                        \
 806                                                                         \
 807         asm(".pushsection .text;"                                       \
 808             "__raw_callee_save_" #func ": "                             \
 809             PV_SAVE_ALL_CALLER_REGS                                     \
 810             "call " #func ";"                                           \
 811             PV_RESTORE_ALL_CALLER_REGS                                  \
 812             "ret;"                                                      \
 813             ".popsection")
 814
 815 /* Get a reference to a callee-save function */
 816 #define PV_CALLEE_SAVE(func)                                            \
 817         ((struct paravirt_callee_save) { __raw_callee_save_##func })
 818
 819 /* Promise that "func" already uses the right calling convention */
 820 #define __PV_IS_CALLEE_SAVE(func)                       \
 821         ((struct paravirt_callee_save) { func })
 822
 823 static inline unsigned long __raw_local_save_flags(void)
 824 {
 825         unsigned long f;
 826
 827         asm volatile(paravirt_alt(PARAVIRT_CALL)
 828                      : "=a"(f)
 829                      : paravirt_type(pv_irq_ops.save_fl),
 830                        paravirt_clobber(CLBR_EAX)
 831                      : "memory", "cc");
 832         return f;
 833 }
 834
 835 static inline void raw_local_irq_restore(unsigned long f)
 836 {
 837         asm volatile(paravirt_alt(PARAVIRT_CALL)
 838                      : "=a"(f)
 839                      : PV_FLAGS_ARG(f),
 840                        paravirt_type(pv_irq_ops.restore_fl),
 841                        paravirt_clobber(CLBR_EAX)
 842                      : "memory", "cc");
 843 }
 844
 845 static inline void raw_local_irq_disable(void)
 846 {
 847         asm volatile(paravirt_alt(PARAVIRT_CALL)
 848                      :
 849                      : paravirt_type(pv_irq_ops.irq_disable),
 850                        paravirt_clobber(CLBR_EAX)
 851                      : "memory", "eax", "cc");
 852 }
 853
 854 static inline void raw_local_irq_enable(void)
 855 {
 856         asm volatile(paravirt_alt(PARAVIRT_CALL)
 857                      :
 858                      : paravirt_type(pv_irq_ops.irq_enable),
 859                        paravirt_clobber(CLBR_EAX)
 860                      : "memory", "eax", "cc");
 861 }
 862
 863 static inline unsigned long __raw_local_irq_save(void)
 864 {
 865         unsigned long f;
 866
 867         f = __raw_local_save_flags();
 868         raw_local_irq_disable();
 869         return f;
 870 }
 871
 872
 873 /* Make sure as little as possible of this mess escapes. */
 874 #undef PARAVIRT_CALL
 875 #undef __PVOP_CALL
 876 #undef __PVOP_VCALL
 877 #undef PVOP_VCALL0
 878 #undef PVOP_CALL0
 879 #undef PVOP_VCALL1
 880 #undef PVOP_CALL1
 881 #undef PVOP_VCALL2
 882 #undef PVOP_CALL2
 883 #undef PVOP_VCALL3
 884 #undef PVOP_CALL3
 885 #undef PVOP_VCALL4
 886 #undef PVOP_CALL4
 887
 888 extern void default_banner(void);
 889
 890 #else  /* __ASSEMBLY__ */
 891
 892 #define _PVSITE(ptype, clobbers, ops, word, algn)       \
 893 771:;                                           \
 894         ops;                                    \
 895 772:;                                           \
 896         .pushsection .parainstructions,"a";     \
 897          .align algn;                           \
 898          word 771b;                             \
 899          .byte ptype;                           \
 900          .byte 772b-771b;                       \
 901          .short clobbers;                       \
 902         .popsection
 903
 904
 905 #define COND_PUSH(set, mask, reg)                       \
 906         .if ((~(set)) & mask); push %reg; .endif
 907 #define COND_POP(set, mask, reg)                        \
 908         .if ((~(set)) & mask); pop %reg; .endif
 909
 910 #ifdef CONFIG_X86_64
 911
 912 #define PV_SAVE_REGS(set)                       \
 913         COND_PUSH(set, CLBR_RAX, rax);          \
 914         COND_PUSH(set, CLBR_RCX, rcx);          \
 915         COND_PUSH(set, CLBR_RDX, rdx);          \
 916         COND_PUSH(set, CLBR_RSI, rsi);          \
 917         COND_PUSH(set, CLBR_RDI, rdi);          \
 918         COND_PUSH(set, CLBR_R8, r8);            \
 919         COND_PUSH(set, CLBR_R9, r9);            \
 920         COND_PUSH(set, CLBR_R10, r10);          \
 921         COND_PUSH(set, CLBR_R11, r11)
 922 #define PV_RESTORE_REGS(set)                    \
 923         COND_POP(set, CLBR_R11, r11);           \
 924         COND_POP(set, CLBR_R10, r10);           \
 925         COND_POP(set, CLBR_R9, r9);             \
 926         COND_POP(set, CLBR_R8, r8);             \
 927         COND_POP(set, CLBR_RDI, rdi);           \
 928         COND_POP(set, CLBR_RSI, rsi);           \
 929         COND_POP(set, CLBR_RDX, rdx);           \
 930         COND_POP(set, CLBR_RCX, rcx);           \
 931         COND_POP(set, CLBR_RAX, rax)
 932
 933 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
 934 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
 935 #define PARA_INDIRECT(addr)     *addr(%rip)
 936 #else
 937 #define PV_SAVE_REGS(set)                       \
 938         COND_PUSH(set, CLBR_EAX, eax);          \
 939         COND_PUSH(set, CLBR_EDI, edi);          \
 940         COND_PUSH(set, CLBR_ECX, ecx);          \
 941         COND_PUSH(set, CLBR_EDX, edx)
 942 #define PV_RESTORE_REGS(set)                    \
 943         COND_POP(set, CLBR_EDX, edx);           \
 944         COND_POP(set, CLBR_ECX, ecx);           \
 945         COND_POP(set, CLBR_EDI, edi);           \
 946         COND_POP(set, CLBR_EAX, eax)
 947
 948 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
 949 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
 950 #define PARA_INDIRECT(addr)     *%cs:addr
 951 #endif
 952
 953 #define INTERRUPT_RETURN                                                \
 954         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,       \
 955                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
 956
 957 #define DISABLE_INTERRUPTS(clobbers)                                    \
 958         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
 959                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
 960                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);    \
 961                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
 962
 963 #define ENABLE_INTERRUPTS(clobbers)                                     \
 964         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,  \
 965                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
 966                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);     \
 967                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
 968
 969 #define USERGS_SYSRET32                                                 \
 970         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),       \
 971                   CLBR_NONE,                                            \
 972                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
 973
 974 #ifdef CONFIG_X86_32
 975 #define GET_CR0_INTO_EAX                                \
 976         push %ecx; push %edx;                           \
 977         call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
 978         pop %edx; pop %ecx
 979
 980 #define ENABLE_INTERRUPTS_SYSEXIT                                       \
 981         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
 982                   CLBR_NONE,                                            \
 983                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
 984
 985
 986 #else   /* !CONFIG_X86_32 */
 987
 988 /*
 989  * If swapgs is used while the userspace stack is still current,
 990  * there's no way to call a pvop.  The PV replacement *must* be
 991  * inlined, or the swapgs instruction must be trapped and emulated.
 992  */
 993 #define SWAPGS_UNSAFE_STACK                                             \
 994         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
 995                   swapgs)
 996
 997 /*
 998  * Note: swapgs is very special, and in practise is either going to be
 999  * implemented with a single "swapgs" instruction or something very
1000  * special.  Either way, we don't need to save any registers for
1001  * it.
1002  */
1003 #define SWAPGS                                                          \
1004         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
1005                   call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)          \
1006                  )
1007
1008 #define GET_CR2_INTO_RCX                                \
1009         call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2); \
1010         movq %rax, %rcx;                                \
1011         xorq %rax, %rax;
1012
1013 #define PARAVIRT_ADJUST_EXCEPTION_FRAME                                 \
1014         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
1015                   CLBR_NONE,                                            \
1016                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
1017
1018 #define USERGS_SYSRET64                                                 \
1019         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),       \
1020                   CLBR_NONE,                                            \
1021                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1022
1023 #define ENABLE_INTERRUPTS_SYSEXIT32                                     \
1024         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
1025                   CLBR_NONE,                                            \
1026                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1027 #endif  /* CONFIG_X86_32 */
1028
1029 #endif /* __ASSEMBLY__ */
1030 #else  /* CONFIG_PARAVIRT */
1031 # define default_banner x86_init_noop
1032 #endif /* !CONFIG_PARAVIRT */
1033 #endif /* _ASM_X86_PARAVIRT_H */