Merge git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core-2.6
[linux/fpc-iii.git] / arch / x86 / include / asm / paravirt.h
blob40d6586af25be1b9e54e8741352fe3ed4f235800
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. */
6 #ifdef CONFIG_PARAVIRT
7 #include <asm/pgtable_types.h>
8 #include <asm/asm.h>
10 #include <asm/paravirt_types.h>
12 #ifndef __ASSEMBLY__
13 #include <linux/types.h>
14 #include <linux/cpumask.h>
16 static inline int paravirt_enabled(void)
18 return pv_info.paravirt_enabled;
21 static inline void load_sp0(struct tss_struct *tss,
22 struct thread_struct *thread)
24 PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
27 #define ARCH_SETUP pv_init_ops.arch_setup();
28 static inline unsigned long get_wallclock(void)
30 return PVOP_CALL0(unsigned long, pv_time_ops.get_wallclock);
33 static inline int set_wallclock(unsigned long nowtime)
35 return PVOP_CALL1(int, pv_time_ops.set_wallclock, nowtime);
38 static inline void (*choose_time_init(void))(void)
40 return pv_time_ops.time_init;
43 /* The paravirtualized CPUID instruction. */
44 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
45 unsigned int *ecx, unsigned int *edx)
47 PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
51 * These special macros can be used to get or set a debugging register
53 static inline unsigned long paravirt_get_debugreg(int reg)
55 return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
57 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
58 static inline void set_debugreg(unsigned long val, int reg)
60 PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
63 static inline void clts(void)
65 PVOP_VCALL0(pv_cpu_ops.clts);
68 static inline unsigned long read_cr0(void)
70 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
73 static inline void write_cr0(unsigned long x)
75 PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
78 static inline unsigned long read_cr2(void)
80 return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
83 static inline void write_cr2(unsigned long x)
85 PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
88 static inline unsigned long read_cr3(void)
90 return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
93 static inline void write_cr3(unsigned long x)
95 PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
98 static inline unsigned long read_cr4(void)
100 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
102 static inline unsigned long read_cr4_safe(void)
104 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
107 static inline void write_cr4(unsigned long x)
109 PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
112 #ifdef CONFIG_X86_64
113 static inline unsigned long read_cr8(void)
115 return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
118 static inline void write_cr8(unsigned long x)
120 PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
122 #endif
124 static inline void raw_safe_halt(void)
126 PVOP_VCALL0(pv_irq_ops.safe_halt);
129 static inline void halt(void)
131 PVOP_VCALL0(pv_irq_ops.safe_halt);
134 static inline void wbinvd(void)
136 PVOP_VCALL0(pv_cpu_ops.wbinvd);
139 #define get_kernel_rpl() (pv_info.kernel_rpl)
141 static inline u64 paravirt_read_msr(unsigned msr, int *err)
143 return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
146 static inline int paravirt_rdmsr_regs(u32 *regs)
148 return PVOP_CALL1(int, pv_cpu_ops.rdmsr_regs, regs);
151 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
153 return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
156 static inline int paravirt_wrmsr_regs(u32 *regs)
158 return PVOP_CALL1(int, pv_cpu_ops.wrmsr_regs, regs);
161 /* These should all do BUG_ON(_err), but our headers are too tangled. */
162 #define rdmsr(msr, val1, val2) \
163 do { \
164 int _err; \
165 u64 _l = paravirt_read_msr(msr, &_err); \
166 val1 = (u32)_l; \
167 val2 = _l >> 32; \
168 } while (0)
170 #define wrmsr(msr, val1, val2) \
171 do { \
172 paravirt_write_msr(msr, val1, val2); \
173 } while (0)
175 #define rdmsrl(msr, val) \
176 do { \
177 int _err; \
178 val = paravirt_read_msr(msr, &_err); \
179 } while (0)
181 #define wrmsrl(msr, val) wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
182 #define wrmsr_safe(msr, a, b) paravirt_write_msr(msr, a, b)
184 /* rdmsr with exception handling */
185 #define rdmsr_safe(msr, a, b) \
186 ({ \
187 int _err; \
188 u64 _l = paravirt_read_msr(msr, &_err); \
189 (*a) = (u32)_l; \
190 (*b) = _l >> 32; \
191 _err; \
194 #define rdmsr_safe_regs(regs) paravirt_rdmsr_regs(regs)
195 #define wrmsr_safe_regs(regs) paravirt_wrmsr_regs(regs)
197 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
199 int err;
201 *p = paravirt_read_msr(msr, &err);
202 return err;
204 static inline int rdmsrl_amd_safe(unsigned msr, unsigned long long *p)
206 u32 gprs[8] = { 0 };
207 int err;
209 gprs[1] = msr;
210 gprs[7] = 0x9c5a203a;
212 err = paravirt_rdmsr_regs(gprs);
214 *p = gprs[0] | ((u64)gprs[2] << 32);
216 return err;
219 static inline int wrmsrl_amd_safe(unsigned msr, unsigned long long val)
221 u32 gprs[8] = { 0 };
223 gprs[0] = (u32)val;
224 gprs[1] = msr;
225 gprs[2] = val >> 32;
226 gprs[7] = 0x9c5a203a;
228 return paravirt_wrmsr_regs(gprs);
231 static inline u64 paravirt_read_tsc(void)
233 return PVOP_CALL0(u64, pv_cpu_ops.read_tsc);
236 #define rdtscl(low) \
237 do { \
238 u64 _l = paravirt_read_tsc(); \
239 low = (int)_l; \
240 } while (0)
242 #define rdtscll(val) (val = paravirt_read_tsc())
244 static inline unsigned long long paravirt_sched_clock(void)
246 return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
248 #define calibrate_tsc() (pv_time_ops.get_tsc_khz())
250 static inline unsigned long long paravirt_read_pmc(int counter)
252 return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
255 #define rdpmc(counter, low, high) \
256 do { \
257 u64 _l = paravirt_read_pmc(counter); \
258 low = (u32)_l; \
259 high = _l >> 32; \
260 } while (0)
262 static inline unsigned long long paravirt_rdtscp(unsigned int *aux)
264 return PVOP_CALL1(u64, pv_cpu_ops.read_tscp, aux);
267 #define rdtscp(low, high, aux) \
268 do { \
269 int __aux; \
270 unsigned long __val = paravirt_rdtscp(&__aux); \
271 (low) = (u32)__val; \
272 (high) = (u32)(__val >> 32); \
273 (aux) = __aux; \
274 } while (0)
276 #define rdtscpll(val, aux) \
277 do { \
278 unsigned long __aux; \
279 val = paravirt_rdtscp(&__aux); \
280 (aux) = __aux; \
281 } while (0)
283 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
285 PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
288 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
290 PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
293 static inline void load_TR_desc(void)
295 PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
297 static inline void load_gdt(const struct desc_ptr *dtr)
299 PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
301 static inline void load_idt(const struct desc_ptr *dtr)
303 PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
305 static inline void set_ldt(const void *addr, unsigned entries)
307 PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
309 static inline void store_gdt(struct desc_ptr *dtr)
311 PVOP_VCALL1(pv_cpu_ops.store_gdt, dtr);
313 static inline void store_idt(struct desc_ptr *dtr)
315 PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
317 static inline unsigned long paravirt_store_tr(void)
319 return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
321 #define store_tr(tr) ((tr) = paravirt_store_tr())
322 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
324 PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
327 #ifdef CONFIG_X86_64
328 static inline void load_gs_index(unsigned int gs)
330 PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
332 #endif
334 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
335 const void *desc)
337 PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
340 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
341 void *desc, int type)
343 PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
346 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
348 PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
350 static inline void set_iopl_mask(unsigned mask)
352 PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
355 /* The paravirtualized I/O functions */
356 static inline void slow_down_io(void)
358 pv_cpu_ops.io_delay();
359 #ifdef REALLY_SLOW_IO
360 pv_cpu_ops.io_delay();
361 pv_cpu_ops.io_delay();
362 pv_cpu_ops.io_delay();
363 #endif
366 #ifdef CONFIG_X86_LOCAL_APIC
367 static inline void setup_boot_clock(void)
369 PVOP_VCALL0(pv_apic_ops.setup_boot_clock);
372 static inline void setup_secondary_clock(void)
374 PVOP_VCALL0(pv_apic_ops.setup_secondary_clock);
376 #endif
378 static inline void paravirt_post_allocator_init(void)
380 if (pv_init_ops.post_allocator_init)
381 (*pv_init_ops.post_allocator_init)();
384 static inline void paravirt_pagetable_setup_start(pgd_t *base)
386 (*pv_mmu_ops.pagetable_setup_start)(base);
389 static inline void paravirt_pagetable_setup_done(pgd_t *base)
391 (*pv_mmu_ops.pagetable_setup_done)(base);
394 #ifdef CONFIG_SMP
395 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
396 unsigned long start_esp)
398 PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
399 phys_apicid, start_eip, start_esp);
401 #endif
403 static inline void paravirt_activate_mm(struct mm_struct *prev,
404 struct mm_struct *next)
406 PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
409 static inline void arch_dup_mmap(struct mm_struct *oldmm,
410 struct mm_struct *mm)
412 PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
415 static inline void arch_exit_mmap(struct mm_struct *mm)
417 PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
420 static inline void __flush_tlb(void)
422 PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
424 static inline void __flush_tlb_global(void)
426 PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
428 static inline void __flush_tlb_single(unsigned long addr)
430 PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
433 static inline void flush_tlb_others(const struct cpumask *cpumask,
434 struct mm_struct *mm,
435 unsigned long va)
437 PVOP_VCALL3(pv_mmu_ops.flush_tlb_others, cpumask, mm, va);
440 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
442 return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
445 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
447 PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
450 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
452 PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
454 static inline void paravirt_release_pte(unsigned long pfn)
456 PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
459 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
461 PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
464 static inline void paravirt_alloc_pmd_clone(unsigned long pfn, unsigned long clonepfn,
465 unsigned long start, unsigned long count)
467 PVOP_VCALL4(pv_mmu_ops.alloc_pmd_clone, pfn, clonepfn, start, count);
469 static inline void paravirt_release_pmd(unsigned long pfn)
471 PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
474 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
476 PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
478 static inline void paravirt_release_pud(unsigned long pfn)
480 PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
483 #ifdef CONFIG_HIGHPTE
484 static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
486 unsigned long ret;
487 ret = PVOP_CALL2(unsigned long, pv_mmu_ops.kmap_atomic_pte, page, type);
488 return (void *)ret;
490 #endif
492 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
493 pte_t *ptep)
495 PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
498 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
499 pte_t *ptep)
501 PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
504 static inline pte_t __pte(pteval_t val)
506 pteval_t ret;
508 if (sizeof(pteval_t) > sizeof(long))
509 ret = PVOP_CALLEE2(pteval_t,
510 pv_mmu_ops.make_pte,
511 val, (u64)val >> 32);
512 else
513 ret = PVOP_CALLEE1(pteval_t,
514 pv_mmu_ops.make_pte,
515 val);
517 return (pte_t) { .pte = ret };
520 static inline pteval_t pte_val(pte_t pte)
522 pteval_t ret;
524 if (sizeof(pteval_t) > sizeof(long))
525 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
526 pte.pte, (u64)pte.pte >> 32);
527 else
528 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
529 pte.pte);
531 return ret;
534 static inline pgd_t __pgd(pgdval_t val)
536 pgdval_t ret;
538 if (sizeof(pgdval_t) > sizeof(long))
539 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
540 val, (u64)val >> 32);
541 else
542 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
543 val);
545 return (pgd_t) { ret };
548 static inline pgdval_t pgd_val(pgd_t pgd)
550 pgdval_t ret;
552 if (sizeof(pgdval_t) > sizeof(long))
553 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
554 pgd.pgd, (u64)pgd.pgd >> 32);
555 else
556 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
557 pgd.pgd);
559 return ret;
562 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
563 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
564 pte_t *ptep)
566 pteval_t ret;
568 ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
569 mm, addr, ptep);
571 return (pte_t) { .pte = ret };
574 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
575 pte_t *ptep, pte_t pte)
577 if (sizeof(pteval_t) > sizeof(long))
578 /* 5 arg words */
579 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
580 else
581 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
582 mm, addr, ptep, pte.pte);
585 static inline void set_pte(pte_t *ptep, pte_t pte)
587 if (sizeof(pteval_t) > sizeof(long))
588 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
589 pte.pte, (u64)pte.pte >> 32);
590 else
591 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
592 pte.pte);
595 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
596 pte_t *ptep, pte_t pte)
598 if (sizeof(pteval_t) > sizeof(long))
599 /* 5 arg words */
600 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
601 else
602 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
605 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
607 pmdval_t val = native_pmd_val(pmd);
609 if (sizeof(pmdval_t) > sizeof(long))
610 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
611 else
612 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
615 #if PAGETABLE_LEVELS >= 3
616 static inline pmd_t __pmd(pmdval_t val)
618 pmdval_t ret;
620 if (sizeof(pmdval_t) > sizeof(long))
621 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
622 val, (u64)val >> 32);
623 else
624 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
625 val);
627 return (pmd_t) { ret };
630 static inline pmdval_t pmd_val(pmd_t pmd)
632 pmdval_t ret;
634 if (sizeof(pmdval_t) > sizeof(long))
635 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
636 pmd.pmd, (u64)pmd.pmd >> 32);
637 else
638 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
639 pmd.pmd);
641 return ret;
644 static inline void set_pud(pud_t *pudp, pud_t pud)
646 pudval_t val = native_pud_val(pud);
648 if (sizeof(pudval_t) > sizeof(long))
649 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
650 val, (u64)val >> 32);
651 else
652 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
653 val);
655 #if PAGETABLE_LEVELS == 4
656 static inline pud_t __pud(pudval_t val)
658 pudval_t ret;
660 if (sizeof(pudval_t) > sizeof(long))
661 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
662 val, (u64)val >> 32);
663 else
664 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
665 val);
667 return (pud_t) { ret };
670 static inline pudval_t pud_val(pud_t pud)
672 pudval_t ret;
674 if (sizeof(pudval_t) > sizeof(long))
675 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
676 pud.pud, (u64)pud.pud >> 32);
677 else
678 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
679 pud.pud);
681 return ret;
684 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
686 pgdval_t val = native_pgd_val(pgd);
688 if (sizeof(pgdval_t) > sizeof(long))
689 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
690 val, (u64)val >> 32);
691 else
692 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
693 val);
696 static inline void pgd_clear(pgd_t *pgdp)
698 set_pgd(pgdp, __pgd(0));
701 static inline void pud_clear(pud_t *pudp)
703 set_pud(pudp, __pud(0));
706 #endif /* PAGETABLE_LEVELS == 4 */
708 #endif /* PAGETABLE_LEVELS >= 3 */
710 #ifdef CONFIG_X86_PAE
711 /* Special-case pte-setting operations for PAE, which can't update a
712 64-bit pte atomically */
713 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
715 PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
716 pte.pte, pte.pte >> 32);
719 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
720 pte_t *ptep)
722 PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
725 static inline void pmd_clear(pmd_t *pmdp)
727 PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
729 #else /* !CONFIG_X86_PAE */
730 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
732 set_pte(ptep, pte);
735 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
736 pte_t *ptep)
738 set_pte_at(mm, addr, ptep, __pte(0));
741 static inline void pmd_clear(pmd_t *pmdp)
743 set_pmd(pmdp, __pmd(0));
745 #endif /* CONFIG_X86_PAE */
747 #define __HAVE_ARCH_START_CONTEXT_SWITCH
748 static inline void arch_start_context_switch(struct task_struct *prev)
750 PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
753 static inline void arch_end_context_switch(struct task_struct *next)
755 PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
758 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
759 static inline void arch_enter_lazy_mmu_mode(void)
761 PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
764 static inline void arch_leave_lazy_mmu_mode(void)
766 PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
769 void arch_flush_lazy_mmu_mode(void);
771 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
772 phys_addr_t phys, pgprot_t flags)
774 pv_mmu_ops.set_fixmap(idx, phys, flags);
777 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
779 static inline int __raw_spin_is_locked(struct raw_spinlock *lock)
781 return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
784 static inline int __raw_spin_is_contended(struct raw_spinlock *lock)
786 return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
788 #define __raw_spin_is_contended __raw_spin_is_contended
790 static __always_inline void __raw_spin_lock(struct raw_spinlock *lock)
792 PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
795 static __always_inline void __raw_spin_lock_flags(struct raw_spinlock *lock,
796 unsigned long flags)
798 PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
801 static __always_inline int __raw_spin_trylock(struct raw_spinlock *lock)
803 return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
806 static __always_inline void __raw_spin_unlock(struct raw_spinlock *lock)
808 PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
811 #endif
813 #ifdef CONFIG_X86_32
814 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
815 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
817 /* save and restore all caller-save registers, except return value */
818 #define PV_SAVE_ALL_CALLER_REGS "pushl %ecx;"
819 #define PV_RESTORE_ALL_CALLER_REGS "popl %ecx;"
821 #define PV_FLAGS_ARG "0"
822 #define PV_EXTRA_CLOBBERS
823 #define PV_VEXTRA_CLOBBERS
824 #else
825 /* save and restore all caller-save registers, except return value */
826 #define PV_SAVE_ALL_CALLER_REGS \
827 "push %rcx;" \
828 "push %rdx;" \
829 "push %rsi;" \
830 "push %rdi;" \
831 "push %r8;" \
832 "push %r9;" \
833 "push %r10;" \
834 "push %r11;"
835 #define PV_RESTORE_ALL_CALLER_REGS \
836 "pop %r11;" \
837 "pop %r10;" \
838 "pop %r9;" \
839 "pop %r8;" \
840 "pop %rdi;" \
841 "pop %rsi;" \
842 "pop %rdx;" \
843 "pop %rcx;"
845 /* We save some registers, but all of them, that's too much. We clobber all
846 * caller saved registers but the argument parameter */
847 #define PV_SAVE_REGS "pushq %%rdi;"
848 #define PV_RESTORE_REGS "popq %%rdi;"
849 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
850 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
851 #define PV_FLAGS_ARG "D"
852 #endif
855 * Generate a thunk around a function which saves all caller-save
856 * registers except for the return value. This allows C functions to
857 * be called from assembler code where fewer than normal registers are
858 * available. It may also help code generation around calls from C
859 * code if the common case doesn't use many registers.
861 * When a callee is wrapped in a thunk, the caller can assume that all
862 * arg regs and all scratch registers are preserved across the
863 * call. The return value in rax/eax will not be saved, even for void
864 * functions.
866 #define PV_CALLEE_SAVE_REGS_THUNK(func) \
867 extern typeof(func) __raw_callee_save_##func; \
868 static void *__##func##__ __used = func; \
870 asm(".pushsection .text;" \
871 "__raw_callee_save_" #func ": " \
872 PV_SAVE_ALL_CALLER_REGS \
873 "call " #func ";" \
874 PV_RESTORE_ALL_CALLER_REGS \
875 "ret;" \
876 ".popsection")
878 /* Get a reference to a callee-save function */
879 #define PV_CALLEE_SAVE(func) \
880 ((struct paravirt_callee_save) { __raw_callee_save_##func })
882 /* Promise that "func" already uses the right calling convention */
883 #define __PV_IS_CALLEE_SAVE(func) \
884 ((struct paravirt_callee_save) { func })
886 static inline unsigned long __raw_local_save_flags(void)
888 unsigned long f;
890 asm volatile(paravirt_alt(PARAVIRT_CALL)
891 : "=a"(f)
892 : paravirt_type(pv_irq_ops.save_fl),
893 paravirt_clobber(CLBR_EAX)
894 : "memory", "cc");
895 return f;
898 static inline void raw_local_irq_restore(unsigned long f)
900 asm volatile(paravirt_alt(PARAVIRT_CALL)
901 : "=a"(f)
902 : PV_FLAGS_ARG(f),
903 paravirt_type(pv_irq_ops.restore_fl),
904 paravirt_clobber(CLBR_EAX)
905 : "memory", "cc");
908 static inline void raw_local_irq_disable(void)
910 asm volatile(paravirt_alt(PARAVIRT_CALL)
912 : paravirt_type(pv_irq_ops.irq_disable),
913 paravirt_clobber(CLBR_EAX)
914 : "memory", "eax", "cc");
917 static inline void raw_local_irq_enable(void)
919 asm volatile(paravirt_alt(PARAVIRT_CALL)
921 : paravirt_type(pv_irq_ops.irq_enable),
922 paravirt_clobber(CLBR_EAX)
923 : "memory", "eax", "cc");
926 static inline unsigned long __raw_local_irq_save(void)
928 unsigned long f;
930 f = __raw_local_save_flags();
931 raw_local_irq_disable();
932 return f;
936 /* Make sure as little as possible of this mess escapes. */
937 #undef PARAVIRT_CALL
938 #undef __PVOP_CALL
939 #undef __PVOP_VCALL
940 #undef PVOP_VCALL0
941 #undef PVOP_CALL0
942 #undef PVOP_VCALL1
943 #undef PVOP_CALL1
944 #undef PVOP_VCALL2
945 #undef PVOP_CALL2
946 #undef PVOP_VCALL3
947 #undef PVOP_CALL3
948 #undef PVOP_VCALL4
949 #undef PVOP_CALL4
951 #else /* __ASSEMBLY__ */
953 #define _PVSITE(ptype, clobbers, ops, word, algn) \
954 771:; \
955 ops; \
956 772:; \
957 .pushsection .parainstructions,"a"; \
958 .align algn; \
959 word 771b; \
960 .byte ptype; \
961 .byte 772b-771b; \
962 .short clobbers; \
963 .popsection
966 #define COND_PUSH(set, mask, reg) \
967 .if ((~(set)) & mask); push %reg; .endif
968 #define COND_POP(set, mask, reg) \
969 .if ((~(set)) & mask); pop %reg; .endif
971 #ifdef CONFIG_X86_64
973 #define PV_SAVE_REGS(set) \
974 COND_PUSH(set, CLBR_RAX, rax); \
975 COND_PUSH(set, CLBR_RCX, rcx); \
976 COND_PUSH(set, CLBR_RDX, rdx); \
977 COND_PUSH(set, CLBR_RSI, rsi); \
978 COND_PUSH(set, CLBR_RDI, rdi); \
979 COND_PUSH(set, CLBR_R8, r8); \
980 COND_PUSH(set, CLBR_R9, r9); \
981 COND_PUSH(set, CLBR_R10, r10); \
982 COND_PUSH(set, CLBR_R11, r11)
983 #define PV_RESTORE_REGS(set) \
984 COND_POP(set, CLBR_R11, r11); \
985 COND_POP(set, CLBR_R10, r10); \
986 COND_POP(set, CLBR_R9, r9); \
987 COND_POP(set, CLBR_R8, r8); \
988 COND_POP(set, CLBR_RDI, rdi); \
989 COND_POP(set, CLBR_RSI, rsi); \
990 COND_POP(set, CLBR_RDX, rdx); \
991 COND_POP(set, CLBR_RCX, rcx); \
992 COND_POP(set, CLBR_RAX, rax)
994 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 8)
995 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
996 #define PARA_INDIRECT(addr) *addr(%rip)
997 #else
998 #define PV_SAVE_REGS(set) \
999 COND_PUSH(set, CLBR_EAX, eax); \
1000 COND_PUSH(set, CLBR_EDI, edi); \
1001 COND_PUSH(set, CLBR_ECX, ecx); \
1002 COND_PUSH(set, CLBR_EDX, edx)
1003 #define PV_RESTORE_REGS(set) \
1004 COND_POP(set, CLBR_EDX, edx); \
1005 COND_POP(set, CLBR_ECX, ecx); \
1006 COND_POP(set, CLBR_EDI, edi); \
1007 COND_POP(set, CLBR_EAX, eax)
1009 #define PARA_PATCH(struct, off) ((PARAVIRT_PATCH_##struct + (off)) / 4)
1010 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
1011 #define PARA_INDIRECT(addr) *%cs:addr
1012 #endif
1014 #define INTERRUPT_RETURN \
1015 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE, \
1016 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
1018 #define DISABLE_INTERRUPTS(clobbers) \
1019 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
1020 PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \
1021 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable); \
1022 PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
1024 #define ENABLE_INTERRUPTS(clobbers) \
1025 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers, \
1026 PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE); \
1027 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable); \
1028 PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
1030 #define USERGS_SYSRET32 \
1031 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32), \
1032 CLBR_NONE, \
1033 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
1035 #ifdef CONFIG_X86_32
1036 #define GET_CR0_INTO_EAX \
1037 push %ecx; push %edx; \
1038 call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
1039 pop %edx; pop %ecx
1041 #define ENABLE_INTERRUPTS_SYSEXIT \
1042 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
1043 CLBR_NONE, \
1044 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1047 #else /* !CONFIG_X86_32 */
1050 * If swapgs is used while the userspace stack is still current,
1051 * there's no way to call a pvop. The PV replacement *must* be
1052 * inlined, or the swapgs instruction must be trapped and emulated.
1054 #define SWAPGS_UNSAFE_STACK \
1055 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1056 swapgs)
1059 * Note: swapgs is very special, and in practise is either going to be
1060 * implemented with a single "swapgs" instruction or something very
1061 * special. Either way, we don't need to save any registers for
1062 * it.
1064 #define SWAPGS \
1065 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE, \
1066 call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs) \
1069 #define GET_CR2_INTO_RCX \
1070 call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2); \
1071 movq %rax, %rcx; \
1072 xorq %rax, %rax;
1074 #define PARAVIRT_ADJUST_EXCEPTION_FRAME \
1075 PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
1076 CLBR_NONE, \
1077 call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
1079 #define USERGS_SYSRET64 \
1080 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64), \
1081 CLBR_NONE, \
1082 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1084 #define ENABLE_INTERRUPTS_SYSEXIT32 \
1085 PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit), \
1086 CLBR_NONE, \
1087 jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1088 #endif /* CONFIG_X86_32 */
1090 #endif /* __ASSEMBLY__ */
1091 #endif /* CONFIG_PARAVIRT */
1092 #endif /* _ASM_X86_PARAVIRT_H */