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