1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _ASM_X86_PARAVIRT_H 3 #define _ASM_X86_PARAVIRT_H 4 /* Various instructions on x86 need to be replaced for 5 * para-virtualization: those hooks are defined here. */ 6 7 #ifdef CONFIG_PARAVIRT 8 #include <asm/pgtable_types.h> 9 #include <asm/asm.h> 10 #include <asm/nospec-branch.h> 11 12 #include <asm/paravirt_types.h> 13 14 #ifndef __ASSEMBLY__ 15 #include <linux/bug.h> 16 #include <linux/types.h> 17 #include <linux/cpumask.h> 18 #include <asm/frame.h> 19 20 static inline void load_sp0(unsigned long sp0) 21 { 22 PVOP_VCALL1(cpu.load_sp0, sp0); 23 } 24 25 /* The paravirtualized CPUID instruction. */ 26 static inline void __cpuid(unsigned int *eax, unsigned int *ebx, 27 unsigned int *ecx, unsigned int *edx) 28 { 29 PVOP_VCALL4(cpu.cpuid, eax, ebx, ecx, edx); 30 } 31 32 /* 33 * These special macros can be used to get or set a debugging register 34 */ 35 static inline unsigned long paravirt_get_debugreg(int reg) 36 { 37 return PVOP_CALL1(unsigned long, cpu.get_debugreg, reg); 38 } 39 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg) 40 static inline void set_debugreg(unsigned long val, int reg) 41 { 42 PVOP_VCALL2(cpu.set_debugreg, reg, val); 43 } 44 45 static inline unsigned long read_cr0(void) 46 { 47 return PVOP_CALL0(unsigned long, cpu.read_cr0); 48 } 49 50 static inline void write_cr0(unsigned long x) 51 { 52 PVOP_VCALL1(cpu.write_cr0, x); 53 } 54 55 static inline unsigned long read_cr2(void) 56 { 57 return PVOP_CALL0(unsigned long, mmu.read_cr2); 58 } 59 60 static inline void write_cr2(unsigned long x) 61 { 62 PVOP_VCALL1(mmu.write_cr2, x); 63 } 64 65 static inline unsigned long __read_cr3(void) 66 { 67 return PVOP_CALL0(unsigned long, mmu.read_cr3); 68 } 69 70 static inline void write_cr3(unsigned long x) 71 { 72 PVOP_VCALL1(mmu.write_cr3, x); 73 } 74 75 static inline void __write_cr4(unsigned long x) 76 { 77 PVOP_VCALL1(cpu.write_cr4, x); 78 } 79 80 #ifdef CONFIG_X86_64 81 static inline unsigned long read_cr8(void) 82 { 83 return PVOP_CALL0(unsigned long, cpu.read_cr8); 84 } 85 86 static inline void write_cr8(unsigned long x) 87 { 88 PVOP_VCALL1(cpu.write_cr8, x); 89 } 90 #endif 91 92 static inline void arch_safe_halt(void) 93 { 94 PVOP_VCALL0(irq.safe_halt); 95 } 96 97 static inline void halt(void) 98 { 99 PVOP_VCALL0(irq.halt); 100 } 101 102 static inline void wbinvd(void) 103 { 104 PVOP_VCALL0(cpu.wbinvd); 105 } 106 107 #define get_kernel_rpl() (pv_info.kernel_rpl) 108 109 static inline u64 paravirt_read_msr(unsigned msr) 110 { 111 return PVOP_CALL1(u64, cpu.read_msr, msr); 112 } 113 114 static inline void paravirt_write_msr(unsigned msr, 115 unsigned low, unsigned high) 116 { 117 PVOP_VCALL3(cpu.write_msr, msr, low, high); 118 } 119 120 static inline u64 paravirt_read_msr_safe(unsigned msr, int *err) 121 { 122 return PVOP_CALL2(u64, cpu.read_msr_safe, msr, err); 123 } 124 125 static inline int paravirt_write_msr_safe(unsigned msr, 126 unsigned low, unsigned high) 127 { 128 return PVOP_CALL3(int, cpu.write_msr_safe, msr, low, high); 129 } 130 131 #define rdmsr(msr, val1, val2) \ 132 do { \ 133 u64 _l = paravirt_read_msr(msr); \ 134 val1 = (u32)_l; \ 135 val2 = _l >> 32; \ 136 } while (0) 137 138 #define wrmsr(msr, val1, val2) \ 139 do { \ 140 paravirt_write_msr(msr, val1, val2); \ 141 } while (0) 142 143 #define rdmsrl(msr, val) \ 144 do { \ 145 val = paravirt_read_msr(msr); \ 146 } while (0) 147 148 static inline void wrmsrl(unsigned msr, u64 val) 149 { 150 wrmsr(msr, (u32)val, (u32)(val>>32)); 151 } 152 153 #define wrmsr_safe(msr, a, b) paravirt_write_msr_safe(msr, a, b) 154 155 /* rdmsr with exception handling */ 156 #define rdmsr_safe(msr, a, b) \ 157 ({ \ 158 int _err; \ 159 u64 _l = paravirt_read_msr_safe(msr, &_err); \ 160 (*a) = (u32)_l; \ 161 (*b) = _l >> 32; \ 162 _err; \ 163 }) 164 165 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p) 166 { 167 int err; 168 169 *p = paravirt_read_msr_safe(msr, &err); 170 return err; 171 } 172 173 static inline unsigned long long paravirt_sched_clock(void) 174 { 175 return PVOP_CALL0(unsigned long long, time.sched_clock); 176 } 177 178 struct static_key; 179 extern struct static_key paravirt_steal_enabled; 180 extern struct static_key paravirt_steal_rq_enabled; 181 182 static inline u64 paravirt_steal_clock(int cpu) 183 { 184 return PVOP_CALL1(u64, time.steal_clock, cpu); 185 } 186 187 static inline unsigned long long paravirt_read_pmc(int counter) 188 { 189 return PVOP_CALL1(u64, cpu.read_pmc, counter); 190 } 191 192 #define rdpmc(counter, low, high) \ 193 do { \ 194 u64 _l = paravirt_read_pmc(counter); \ 195 low = (u32)_l; \ 196 high = _l >> 32; \ 197 } while (0) 198 199 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter)) 200 201 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries) 202 { 203 PVOP_VCALL2(cpu.alloc_ldt, ldt, entries); 204 } 205 206 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries) 207 { 208 PVOP_VCALL2(cpu.free_ldt, ldt, entries); 209 } 210 211 static inline void load_TR_desc(void) 212 { 213 PVOP_VCALL0(cpu.load_tr_desc); 214 } 215 static inline void load_gdt(const struct desc_ptr *dtr) 216 { 217 PVOP_VCALL1(cpu.load_gdt, dtr); 218 } 219 static inline void load_idt(const struct desc_ptr *dtr) 220 { 221 PVOP_VCALL1(cpu.load_idt, dtr); 222 } 223 static inline void set_ldt(const void *addr, unsigned entries) 224 { 225 PVOP_VCALL2(cpu.set_ldt, addr, entries); 226 } 227 static inline unsigned long paravirt_store_tr(void) 228 { 229 return PVOP_CALL0(unsigned long, cpu.store_tr); 230 } 231 #define store_tr(tr) ((tr) = paravirt_store_tr()) 232 static inline void load_TLS(struct thread_struct *t, unsigned cpu) 233 { 234 PVOP_VCALL2(cpu.load_tls, t, cpu); 235 } 236 237 #ifdef CONFIG_X86_64 238 static inline void load_gs_index(unsigned int gs) 239 { 240 PVOP_VCALL1(cpu.load_gs_index, gs); 241 } 242 #endif 243 244 static inline void write_ldt_entry(struct desc_struct *dt, int entry, 245 const void *desc) 246 { 247 PVOP_VCALL3(cpu.write_ldt_entry, dt, entry, desc); 248 } 249 250 static inline void write_gdt_entry(struct desc_struct *dt, int entry, 251 void *desc, int type) 252 { 253 PVOP_VCALL4(cpu.write_gdt_entry, dt, entry, desc, type); 254 } 255 256 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g) 257 { 258 PVOP_VCALL3(cpu.write_idt_entry, dt, entry, g); 259 } 260 static inline void set_iopl_mask(unsigned mask) 261 { 262 PVOP_VCALL1(cpu.set_iopl_mask, mask); 263 } 264 265 /* The paravirtualized I/O functions */ 266 static inline void slow_down_io(void) 267 { 268 pv_ops.cpu.io_delay(); 269 #ifdef REALLY_SLOW_IO 270 pv_ops.cpu.io_delay(); 271 pv_ops.cpu.io_delay(); 272 pv_ops.cpu.io_delay(); 273 #endif 274 } 275 276 static inline void paravirt_activate_mm(struct mm_struct *prev, 277 struct mm_struct *next) 278 { 279 PVOP_VCALL2(mmu.activate_mm, prev, next); 280 } 281 282 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm, 283 struct mm_struct *mm) 284 { 285 PVOP_VCALL2(mmu.dup_mmap, oldmm, mm); 286 } 287 288 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm) 289 { 290 PVOP_VCALL1(mmu.exit_mmap, mm); 291 } 292 293 static inline void __flush_tlb(void) 294 { 295 PVOP_VCALL0(mmu.flush_tlb_user); 296 } 297 static inline void __flush_tlb_global(void) 298 { 299 PVOP_VCALL0(mmu.flush_tlb_kernel); 300 } 301 static inline void __flush_tlb_one_user(unsigned long addr) 302 { 303 PVOP_VCALL1(mmu.flush_tlb_one_user, addr); 304 } 305 306 static inline void flush_tlb_others(const struct cpumask *cpumask, 307 const struct flush_tlb_info *info) 308 { 309 PVOP_VCALL2(mmu.flush_tlb_others, cpumask, info); 310 } 311 312 static inline void paravirt_tlb_remove_table(struct mmu_gather *tlb, void *table) 313 { 314 PVOP_VCALL2(mmu.tlb_remove_table, tlb, table); 315 } 316 317 static inline int paravirt_pgd_alloc(struct mm_struct *mm) 318 { 319 return PVOP_CALL1(int, mmu.pgd_alloc, mm); 320 } 321 322 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd) 323 { 324 PVOP_VCALL2(mmu.pgd_free, mm, pgd); 325 } 326 327 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn) 328 { 329 PVOP_VCALL2(mmu.alloc_pte, mm, pfn); 330 } 331 static inline void paravirt_release_pte(unsigned long pfn) 332 { 333 PVOP_VCALL1(mmu.release_pte, pfn); 334 } 335 336 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn) 337 { 338 PVOP_VCALL2(mmu.alloc_pmd, mm, pfn); 339 } 340 341 static inline void paravirt_release_pmd(unsigned long pfn) 342 { 343 PVOP_VCALL1(mmu.release_pmd, pfn); 344 } 345 346 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn) 347 { 348 PVOP_VCALL2(mmu.alloc_pud, mm, pfn); 349 } 350 static inline void paravirt_release_pud(unsigned long pfn) 351 { 352 PVOP_VCALL1(mmu.release_pud, pfn); 353 } 354 355 static inline void paravirt_alloc_p4d(struct mm_struct *mm, unsigned long pfn) 356 { 357 PVOP_VCALL2(mmu.alloc_p4d, mm, pfn); 358 } 359 360 static inline void paravirt_release_p4d(unsigned long pfn) 361 { 362 PVOP_VCALL1(mmu.release_p4d, pfn); 363 } 364 365 static inline pte_t __pte(pteval_t val) 366 { 367 pteval_t ret; 368 369 if (sizeof(pteval_t) > sizeof(long)) 370 ret = PVOP_CALLEE2(pteval_t, mmu.make_pte, val, (u64)val >> 32); 371 else 372 ret = PVOP_CALLEE1(pteval_t, mmu.make_pte, val); 373 374 return (pte_t) { .pte = ret }; 375 } 376 377 static inline pteval_t pte_val(pte_t pte) 378 { 379 pteval_t ret; 380 381 if (sizeof(pteval_t) > sizeof(long)) 382 ret = PVOP_CALLEE2(pteval_t, mmu.pte_val, 383 pte.pte, (u64)pte.pte >> 32); 384 else 385 ret = PVOP_CALLEE1(pteval_t, mmu.pte_val, pte.pte); 386 387 return ret; 388 } 389 390 static inline pgd_t __pgd(pgdval_t val) 391 { 392 pgdval_t ret; 393 394 if (sizeof(pgdval_t) > sizeof(long)) 395 ret = PVOP_CALLEE2(pgdval_t, mmu.make_pgd, val, (u64)val >> 32); 396 else 397 ret = PVOP_CALLEE1(pgdval_t, mmu.make_pgd, val); 398 399 return (pgd_t) { ret }; 400 } 401 402 static inline pgdval_t pgd_val(pgd_t pgd) 403 { 404 pgdval_t ret; 405 406 if (sizeof(pgdval_t) > sizeof(long)) 407 ret = PVOP_CALLEE2(pgdval_t, mmu.pgd_val, 408 pgd.pgd, (u64)pgd.pgd >> 32); 409 else 410 ret = PVOP_CALLEE1(pgdval_t, mmu.pgd_val, pgd.pgd); 411 412 return ret; 413 } 414 415 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION 416 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr, 417 pte_t *ptep) 418 { 419 pteval_t ret; 420 421 ret = PVOP_CALL3(pteval_t, mmu.ptep_modify_prot_start, mm, addr, ptep); 422 423 return (pte_t) { .pte = ret }; 424 } 425 426 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, 427 pte_t *ptep, pte_t pte) 428 { 429 if (sizeof(pteval_t) > sizeof(long)) 430 /* 5 arg words */ 431 pv_ops.mmu.ptep_modify_prot_commit(mm, addr, ptep, pte); 432 else 433 PVOP_VCALL4(mmu.ptep_modify_prot_commit, 434 mm, addr, ptep, pte.pte); 435 } 436 437 static inline void set_pte(pte_t *ptep, pte_t pte) 438 { 439 if (sizeof(pteval_t) > sizeof(long)) 440 PVOP_VCALL3(mmu.set_pte, ptep, pte.pte, (u64)pte.pte >> 32); 441 else 442 PVOP_VCALL2(mmu.set_pte, ptep, pte.pte); 443 } 444 445 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr, 446 pte_t *ptep, pte_t pte) 447 { 448 if (sizeof(pteval_t) > sizeof(long)) 449 /* 5 arg words */ 450 pv_ops.mmu.set_pte_at(mm, addr, ptep, pte); 451 else 452 PVOP_VCALL4(mmu.set_pte_at, mm, addr, ptep, pte.pte); 453 } 454 455 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd) 456 { 457 pmdval_t val = native_pmd_val(pmd); 458 459 if (sizeof(pmdval_t) > sizeof(long)) 460 PVOP_VCALL3(mmu.set_pmd, pmdp, val, (u64)val >> 32); 461 else 462 PVOP_VCALL2(mmu.set_pmd, pmdp, val); 463 } 464 465 #if CONFIG_PGTABLE_LEVELS >= 3 466 static inline pmd_t __pmd(pmdval_t val) 467 { 468 pmdval_t ret; 469 470 if (sizeof(pmdval_t) > sizeof(long)) 471 ret = PVOP_CALLEE2(pmdval_t, mmu.make_pmd, val, (u64)val >> 32); 472 else 473 ret = PVOP_CALLEE1(pmdval_t, mmu.make_pmd, val); 474 475 return (pmd_t) { ret }; 476 } 477 478 static inline pmdval_t pmd_val(pmd_t pmd) 479 { 480 pmdval_t ret; 481 482 if (sizeof(pmdval_t) > sizeof(long)) 483 ret = PVOP_CALLEE2(pmdval_t, mmu.pmd_val, 484 pmd.pmd, (u64)pmd.pmd >> 32); 485 else 486 ret = PVOP_CALLEE1(pmdval_t, mmu.pmd_val, pmd.pmd); 487 488 return ret; 489 } 490 491 static inline void set_pud(pud_t *pudp, pud_t pud) 492 { 493 pudval_t val = native_pud_val(pud); 494 495 if (sizeof(pudval_t) > sizeof(long)) 496 PVOP_VCALL3(mmu.set_pud, pudp, val, (u64)val >> 32); 497 else 498 PVOP_VCALL2(mmu.set_pud, pudp, val); 499 } 500 #if CONFIG_PGTABLE_LEVELS >= 4 501 static inline pud_t __pud(pudval_t val) 502 { 503 pudval_t ret; 504 505 if (sizeof(pudval_t) > sizeof(long)) 506 ret = PVOP_CALLEE2(pudval_t, mmu.make_pud, val, (u64)val >> 32); 507 else 508 ret = PVOP_CALLEE1(pudval_t, mmu.make_pud, val); 509 510 return (pud_t) { ret }; 511 } 512 513 static inline pudval_t pud_val(pud_t pud) 514 { 515 pudval_t ret; 516 517 if (sizeof(pudval_t) > sizeof(long)) 518 ret = PVOP_CALLEE2(pudval_t, mmu.pud_val, 519 pud.pud, (u64)pud.pud >> 32); 520 else 521 ret = PVOP_CALLEE1(pudval_t, mmu.pud_val, pud.pud); 522 523 return ret; 524 } 525 526 static inline void pud_clear(pud_t *pudp) 527 { 528 set_pud(pudp, __pud(0)); 529 } 530 531 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d) 532 { 533 p4dval_t val = native_p4d_val(p4d); 534 535 if (sizeof(p4dval_t) > sizeof(long)) 536 PVOP_VCALL3(mmu.set_p4d, p4dp, val, (u64)val >> 32); 537 else 538 PVOP_VCALL2(mmu.set_p4d, p4dp, val); 539 } 540 541 #if CONFIG_PGTABLE_LEVELS >= 5 542 543 static inline p4d_t __p4d(p4dval_t val) 544 { 545 p4dval_t ret = PVOP_CALLEE1(p4dval_t, mmu.make_p4d, val); 546 547 return (p4d_t) { ret }; 548 } 549 550 static inline p4dval_t p4d_val(p4d_t p4d) 551 { 552 return PVOP_CALLEE1(p4dval_t, mmu.p4d_val, p4d.p4d); 553 } 554 555 static inline void __set_pgd(pgd_t *pgdp, pgd_t pgd) 556 { 557 PVOP_VCALL2(mmu.set_pgd, pgdp, native_pgd_val(pgd)); 558 } 559 560 #define set_pgd(pgdp, pgdval) do { \ 561 if (pgtable_l5_enabled()) \ 562 __set_pgd(pgdp, pgdval); \ 563 else \ 564 set_p4d((p4d_t *)(pgdp), (p4d_t) { (pgdval).pgd }); \ 565 } while (0) 566 567 #define pgd_clear(pgdp) do { \ 568 if (pgtable_l5_enabled()) \ 569 set_pgd(pgdp, __pgd(0)); \ 570 } while (0) 571 572 #endif /* CONFIG_PGTABLE_LEVELS == 5 */ 573 574 static inline void p4d_clear(p4d_t *p4dp) 575 { 576 set_p4d(p4dp, __p4d(0)); 577 } 578 579 #endif /* CONFIG_PGTABLE_LEVELS == 4 */ 580 581 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */ 582 583 #ifdef CONFIG_X86_PAE 584 /* Special-case pte-setting operations for PAE, which can't update a 585 64-bit pte atomically */ 586 static inline void set_pte_atomic(pte_t *ptep, pte_t pte) 587 { 588 PVOP_VCALL3(mmu.set_pte_atomic, ptep, pte.pte, pte.pte >> 32); 589 } 590 591 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, 592 pte_t *ptep) 593 { 594 PVOP_VCALL3(mmu.pte_clear, mm, addr, ptep); 595 } 596 597 static inline void pmd_clear(pmd_t *pmdp) 598 { 599 PVOP_VCALL1(mmu.pmd_clear, pmdp); 600 } 601 #else /* !CONFIG_X86_PAE */ 602 static inline void set_pte_atomic(pte_t *ptep, pte_t pte) 603 { 604 set_pte(ptep, pte); 605 } 606 607 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, 608 pte_t *ptep) 609 { 610 set_pte_at(mm, addr, ptep, __pte(0)); 611 } 612 613 static inline void pmd_clear(pmd_t *pmdp) 614 { 615 set_pmd(pmdp, __pmd(0)); 616 } 617 #endif /* CONFIG_X86_PAE */ 618 619 #define __HAVE_ARCH_START_CONTEXT_SWITCH 620 static inline void arch_start_context_switch(struct task_struct *prev) 621 { 622 PVOP_VCALL1(cpu.start_context_switch, prev); 623 } 624 625 static inline void arch_end_context_switch(struct task_struct *next) 626 { 627 PVOP_VCALL1(cpu.end_context_switch, next); 628 } 629 630 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE 631 static inline void arch_enter_lazy_mmu_mode(void) 632 { 633 PVOP_VCALL0(mmu.lazy_mode.enter); 634 } 635 636 static inline void arch_leave_lazy_mmu_mode(void) 637 { 638 PVOP_VCALL0(mmu.lazy_mode.leave); 639 } 640 641 static inline void arch_flush_lazy_mmu_mode(void) 642 { 643 PVOP_VCALL0(mmu.lazy_mode.flush); 644 } 645 646 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx, 647 phys_addr_t phys, pgprot_t flags) 648 { 649 pv_ops.mmu.set_fixmap(idx, phys, flags); 650 } 651 652 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS) 653 654 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock, 655 u32 val) 656 { 657 PVOP_VCALL2(lock.queued_spin_lock_slowpath, lock, val); 658 } 659 660 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock) 661 { 662 PVOP_VCALLEE1(lock.queued_spin_unlock, lock); 663 } 664 665 static __always_inline void pv_wait(u8 *ptr, u8 val) 666 { 667 PVOP_VCALL2(lock.wait, ptr, val); 668 } 669 670 static __always_inline void pv_kick(int cpu) 671 { 672 PVOP_VCALL1(lock.kick, cpu); 673 } 674 675 static __always_inline bool pv_vcpu_is_preempted(long cpu) 676 { 677 return PVOP_CALLEE1(bool, lock.vcpu_is_preempted, cpu); 678 } 679 680 void __raw_callee_save___native_queued_spin_unlock(struct qspinlock *lock); 681 bool __raw_callee_save___native_vcpu_is_preempted(long cpu); 682 683 #endif /* SMP && PARAVIRT_SPINLOCKS */ 684 685 #ifdef CONFIG_X86_32 686 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;" 687 #define PV_RESTORE_REGS "popl %edx; popl %ecx;" 688 689 /* save and restore all caller-save registers, except return value */ 690 #define PV_SAVE_ALL_CALLER_REGS "pushl %ecx;" 691 #define PV_RESTORE_ALL_CALLER_REGS "popl %ecx;" 692 693 #define PV_FLAGS_ARG "0" 694 #define PV_EXTRA_CLOBBERS 695 #define PV_VEXTRA_CLOBBERS 696 #else 697 /* save and restore all caller-save registers, except return value */ 698 #define PV_SAVE_ALL_CALLER_REGS \ 699 "push %rcx;" \ 700 "push %rdx;" \ 701 "push %rsi;" \ 702 "push %rdi;" \ 703 "push %r8;" \ 704 "push %r9;" \ 705 "push %r10;" \ 706 "push %r11;" 707 #define PV_RESTORE_ALL_CALLER_REGS \ 708 "pop %r11;" \ 709 "pop %r10;" \ 710 "pop %r9;" \ 711 "pop %r8;" \ 712 "pop %rdi;" \ 713 "pop %rsi;" \ 714 "pop %rdx;" \ 715 "pop %rcx;" 716 717 /* We save some registers, but all of them, that's too much. We clobber all 718 * caller saved registers but the argument parameter */ 719 #define PV_SAVE_REGS "pushq %%rdi;" 720 #define PV_RESTORE_REGS "popq %%rdi;" 721 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi" 722 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi" 723 #define PV_FLAGS_ARG "D" 724 #endif 725 726 /* 727 * Generate a thunk around a function which saves all caller-save 728 * registers except for the return value. This allows C functions to 729 * be called from assembler code where fewer than normal registers are 730 * available. It may also help code generation around calls from C 731 * code if the common case doesn't use many registers. 732 * 733 * When a callee is wrapped in a thunk, the caller can assume that all 734 * arg regs and all scratch registers are preserved across the 735 * call. The return value in rax/eax will not be saved, even for void 736 * functions. 737 */ 738 #define PV_THUNK_NAME(func) "__raw_callee_save_" #func 739 #define PV_CALLEE_SAVE_REGS_THUNK(func) \ 740 extern typeof(func) __raw_callee_save_##func; \ 741 \ 742 asm(".pushsection .text;" \ 743 ".globl " PV_THUNK_NAME(func) ";" \ 744 ".type " PV_THUNK_NAME(func) ", @function;" \ 745 PV_THUNK_NAME(func) ":" \ 746 FRAME_BEGIN \ 747 PV_SAVE_ALL_CALLER_REGS \ 748 "call " #func ";" \ 749 PV_RESTORE_ALL_CALLER_REGS \ 750 FRAME_END \ 751 "ret;" \ 752 ".popsection") 753 754 /* Get a reference to a callee-save function */ 755 #define PV_CALLEE_SAVE(func) \ 756 ((struct paravirt_callee_save) { __raw_callee_save_##func }) 757 758 /* Promise that "func" already uses the right calling convention */ 759 #define __PV_IS_CALLEE_SAVE(func) \ 760 ((struct paravirt_callee_save) { func }) 761 762 static inline notrace unsigned long arch_local_save_flags(void) 763 { 764 return PVOP_CALLEE0(unsigned long, irq.save_fl); 765 } 766 767 static inline notrace void arch_local_irq_restore(unsigned long f) 768 { 769 PVOP_VCALLEE1(irq.restore_fl, f); 770 } 771 772 static inline notrace void arch_local_irq_disable(void) 773 { 774 PVOP_VCALLEE0(irq.irq_disable); 775 } 776 777 static inline notrace void arch_local_irq_enable(void) 778 { 779 PVOP_VCALLEE0(irq.irq_enable); 780 } 781 782 static inline notrace unsigned long arch_local_irq_save(void) 783 { 784 unsigned long f; 785 786 f = arch_local_save_flags(); 787 arch_local_irq_disable(); 788 return f; 789 } 790 791 792 /* Make sure as little as possible of this mess escapes. */ 793 #undef PARAVIRT_CALL 794 #undef __PVOP_CALL 795 #undef __PVOP_VCALL 796 #undef PVOP_VCALL0 797 #undef PVOP_CALL0 798 #undef PVOP_VCALL1 799 #undef PVOP_CALL1 800 #undef PVOP_VCALL2 801 #undef PVOP_CALL2 802 #undef PVOP_VCALL3 803 #undef PVOP_CALL3 804 #undef PVOP_VCALL4 805 #undef PVOP_CALL4 806 807 extern void default_banner(void); 808 809 #else /* __ASSEMBLY__ */ 810 811 #define _PVSITE(ptype, ops, word, algn) \ 812 771:; \ 813 ops; \ 814 772:; \ 815 .pushsection .parainstructions,"a"; \ 816 .align algn; \ 817 word 771b; \ 818 .byte ptype; \ 819 .byte 772b-771b; \ 820 .popsection 821 822 823 #define COND_PUSH(set, mask, reg) \ 824 .if ((~(set)) & mask); push %reg; .endif 825 #define COND_POP(set, mask, reg) \ 826 .if ((~(set)) & mask); pop %reg; .endif 827 828 #ifdef CONFIG_X86_64 829 830 #define PV_SAVE_REGS(set) \ 831 COND_PUSH(set, CLBR_RAX, rax); \ 832 COND_PUSH(set, CLBR_RCX, rcx); \ 833 COND_PUSH(set, CLBR_RDX, rdx); \ 834 COND_PUSH(set, CLBR_RSI, rsi); \ 835 COND_PUSH(set, CLBR_RDI, rdi); \ 836 COND_PUSH(set, CLBR_R8, r8); \ 837 COND_PUSH(set, CLBR_R9, r9); \ 838 COND_PUSH(set, CLBR_R10, r10); \ 839 COND_PUSH(set, CLBR_R11, r11) 840 #define PV_RESTORE_REGS(set) \ 841 COND_POP(set, CLBR_R11, r11); \ 842 COND_POP(set, CLBR_R10, r10); \ 843 COND_POP(set, CLBR_R9, r9); \ 844 COND_POP(set, CLBR_R8, r8); \ 845 COND_POP(set, CLBR_RDI, rdi); \ 846 COND_POP(set, CLBR_RSI, rsi); \ 847 COND_POP(set, CLBR_RDX, rdx); \ 848 COND_POP(set, CLBR_RCX, rcx); \ 849 COND_POP(set, CLBR_RAX, rax) 850 851 #define PARA_PATCH(off) ((off) / 8) 852 #define PARA_SITE(ptype, ops) _PVSITE(ptype, ops, .quad, 8) 853 #define PARA_INDIRECT(addr) *addr(%rip) 854 #else 855 #define PV_SAVE_REGS(set) \ 856 COND_PUSH(set, CLBR_EAX, eax); \ 857 COND_PUSH(set, CLBR_EDI, edi); \ 858 COND_PUSH(set, CLBR_ECX, ecx); \ 859 COND_PUSH(set, CLBR_EDX, edx) 860 #define PV_RESTORE_REGS(set) \ 861 COND_POP(set, CLBR_EDX, edx); \ 862 COND_POP(set, CLBR_ECX, ecx); \ 863 COND_POP(set, CLBR_EDI, edi); \ 864 COND_POP(set, CLBR_EAX, eax) 865 866 #define PARA_PATCH(off) ((off) / 4) 867 #define PARA_SITE(ptype, ops) _PVSITE(ptype, ops, .long, 4) 868 #define PARA_INDIRECT(addr) *%cs:addr 869 #endif 870 871 #define INTERRUPT_RETURN \ 872 PARA_SITE(PARA_PATCH(PV_CPU_iret), \ 873 ANNOTATE_RETPOLINE_SAFE; \ 874 jmp PARA_INDIRECT(pv_ops+PV_CPU_iret);) 875 876 #define DISABLE_INTERRUPTS(clobbers) \ 877 PARA_SITE(PARA_PATCH(PV_IRQ_irq_disable), \ 878 PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \ 879 ANNOTATE_RETPOLINE_SAFE; \ 880 call PARA_INDIRECT(pv_ops+PV_IRQ_irq_disable); \ 881 PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);) 882 883 #define ENABLE_INTERRUPTS(clobbers) \ 884 PARA_SITE(PARA_PATCH(PV_IRQ_irq_enable), \ 885 PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \ 886 ANNOTATE_RETPOLINE_SAFE; \ 887 call PARA_INDIRECT(pv_ops+PV_IRQ_irq_enable); \ 888 PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);) 889 890 #ifdef CONFIG_X86_32 891 #define GET_CR0_INTO_EAX \ 892 push %ecx; push %edx; \ 893 ANNOTATE_RETPOLINE_SAFE; \ 894 call PARA_INDIRECT(pv_ops+PV_CPU_read_cr0); \ 895 pop %edx; pop %ecx 896 #else /* !CONFIG_X86_32 */ 897 898 /* 899 * If swapgs is used while the userspace stack is still current, 900 * there's no way to call a pvop. The PV replacement *must* be 901 * inlined, or the swapgs instruction must be trapped and emulated. 902 */ 903 #define SWAPGS_UNSAFE_STACK \ 904 PARA_SITE(PARA_PATCH(PV_CPU_swapgs), swapgs) 905 906 /* 907 * Note: swapgs is very special, and in practise is either going to be 908 * implemented with a single "swapgs" instruction or something very 909 * special. Either way, we don't need to save any registers for 910 * it. 911 */ 912 #define SWAPGS \ 913 PARA_SITE(PARA_PATCH(PV_CPU_swapgs), \ 914 ANNOTATE_RETPOLINE_SAFE; \ 915 call PARA_INDIRECT(pv_ops+PV_CPU_swapgs); \ 916 ) 917 918 #define GET_CR2_INTO_RAX \ 919 ANNOTATE_RETPOLINE_SAFE; \ 920 call PARA_INDIRECT(pv_ops+PV_MMU_read_cr2); 921 922 #define USERGS_SYSRET64 \ 923 PARA_SITE(PARA_PATCH(PV_CPU_usergs_sysret64), \ 924 ANNOTATE_RETPOLINE_SAFE; \ 925 jmp PARA_INDIRECT(pv_ops+PV_CPU_usergs_sysret64);) 926 927 #ifdef CONFIG_DEBUG_ENTRY 928 #define SAVE_FLAGS(clobbers) \ 929 PARA_SITE(PARA_PATCH(PV_IRQ_save_fl), \ 930 PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \ 931 ANNOTATE_RETPOLINE_SAFE; \ 932 call PARA_INDIRECT(pv_ops+PV_IRQ_save_fl); \ 933 PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);) 934 #endif 935 936 #endif /* CONFIG_X86_32 */ 937 938 #endif /* __ASSEMBLY__ */ 939 #else /* CONFIG_PARAVIRT */ 940 # define default_banner x86_init_noop 941 #ifndef __ASSEMBLY__ 942 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm, 943 struct mm_struct *mm) 944 { 945 } 946 947 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm) 948 { 949 } 950 #endif /* __ASSEMBLY__ */ 951 #endif /* !CONFIG_PARAVIRT */ 952 #endif /* _ASM_X86_PARAVIRT_H */ 953