Merge branch 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-block.git] / arch / x86 / include / asm / paravirt.h
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 unsigned long long paravirt_sched_clock(void)
21 {
22         return PVOP_CALL0(unsigned long long, time.sched_clock);
23 }
24
25 struct static_key;
26 extern struct static_key paravirt_steal_enabled;
27 extern struct static_key paravirt_steal_rq_enabled;
28
29 __visible void __native_queued_spin_unlock(struct qspinlock *lock);
30 bool pv_is_native_spin_unlock(void);
31 __visible bool __native_vcpu_is_preempted(long cpu);
32 bool pv_is_native_vcpu_is_preempted(void);
33
34 static inline u64 paravirt_steal_clock(int cpu)
35 {
36         return PVOP_CALL1(u64, time.steal_clock, cpu);
37 }
38
39 /* The paravirtualized I/O functions */
40 static inline void slow_down_io(void)
41 {
42         pv_ops.cpu.io_delay();
43 #ifdef REALLY_SLOW_IO
44         pv_ops.cpu.io_delay();
45         pv_ops.cpu.io_delay();
46         pv_ops.cpu.io_delay();
47 #endif
48 }
49
50 static inline void __flush_tlb(void)
51 {
52         PVOP_VCALL0(mmu.flush_tlb_user);
53 }
54
55 static inline void __flush_tlb_global(void)
56 {
57         PVOP_VCALL0(mmu.flush_tlb_kernel);
58 }
59
60 static inline void __flush_tlb_one_user(unsigned long addr)
61 {
62         PVOP_VCALL1(mmu.flush_tlb_one_user, addr);
63 }
64
65 static inline void flush_tlb_others(const struct cpumask *cpumask,
66                                     const struct flush_tlb_info *info)
67 {
68         PVOP_VCALL2(mmu.flush_tlb_others, cpumask, info);
69 }
70
71 static inline void paravirt_tlb_remove_table(struct mmu_gather *tlb, void *table)
72 {
73         PVOP_VCALL2(mmu.tlb_remove_table, tlb, table);
74 }
75
76 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
77 {
78         PVOP_VCALL1(mmu.exit_mmap, mm);
79 }
80
81 #ifdef CONFIG_PARAVIRT_XXL
82 static inline void load_sp0(unsigned long sp0)
83 {
84         PVOP_VCALL1(cpu.load_sp0, sp0);
85 }
86
87 /* The paravirtualized CPUID instruction. */
88 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
89                            unsigned int *ecx, unsigned int *edx)
90 {
91         PVOP_VCALL4(cpu.cpuid, eax, ebx, ecx, edx);
92 }
93
94 /*
95  * These special macros can be used to get or set a debugging register
96  */
97 static inline unsigned long paravirt_get_debugreg(int reg)
98 {
99         return PVOP_CALL1(unsigned long, cpu.get_debugreg, reg);
100 }
101 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
102 static inline void set_debugreg(unsigned long val, int reg)
103 {
104         PVOP_VCALL2(cpu.set_debugreg, reg, val);
105 }
106
107 static inline unsigned long read_cr0(void)
108 {
109         return PVOP_CALL0(unsigned long, cpu.read_cr0);
110 }
111
112 static inline void write_cr0(unsigned long x)
113 {
114         PVOP_VCALL1(cpu.write_cr0, x);
115 }
116
117 static inline unsigned long read_cr2(void)
118 {
119         return PVOP_CALLEE0(unsigned long, mmu.read_cr2);
120 }
121
122 static inline void write_cr2(unsigned long x)
123 {
124         PVOP_VCALL1(mmu.write_cr2, x);
125 }
126
127 static inline unsigned long __read_cr3(void)
128 {
129         return PVOP_CALL0(unsigned long, mmu.read_cr3);
130 }
131
132 static inline void write_cr3(unsigned long x)
133 {
134         PVOP_VCALL1(mmu.write_cr3, x);
135 }
136
137 static inline void __write_cr4(unsigned long x)
138 {
139         PVOP_VCALL1(cpu.write_cr4, x);
140 }
141
142 static inline void arch_safe_halt(void)
143 {
144         PVOP_VCALL0(irq.safe_halt);
145 }
146
147 static inline void halt(void)
148 {
149         PVOP_VCALL0(irq.halt);
150 }
151
152 static inline void wbinvd(void)
153 {
154         PVOP_VCALL0(cpu.wbinvd);
155 }
156
157 #define get_kernel_rpl()  (pv_info.kernel_rpl)
158
159 static inline u64 paravirt_read_msr(unsigned msr)
160 {
161         return PVOP_CALL1(u64, cpu.read_msr, msr);
162 }
163
164 static inline void paravirt_write_msr(unsigned msr,
165                                       unsigned low, unsigned high)
166 {
167         PVOP_VCALL3(cpu.write_msr, msr, low, high);
168 }
169
170 static inline u64 paravirt_read_msr_safe(unsigned msr, int *err)
171 {
172         return PVOP_CALL2(u64, cpu.read_msr_safe, msr, err);
173 }
174
175 static inline int paravirt_write_msr_safe(unsigned msr,
176                                           unsigned low, unsigned high)
177 {
178         return PVOP_CALL3(int, cpu.write_msr_safe, msr, low, high);
179 }
180
181 #define rdmsr(msr, val1, val2)                  \
182 do {                                            \
183         u64 _l = paravirt_read_msr(msr);        \
184         val1 = (u32)_l;                         \
185         val2 = _l >> 32;                        \
186 } while (0)
187
188 #define wrmsr(msr, val1, val2)                  \
189 do {                                            \
190         paravirt_write_msr(msr, val1, val2);    \
191 } while (0)
192
193 #define rdmsrl(msr, val)                        \
194 do {                                            \
195         val = paravirt_read_msr(msr);           \
196 } while (0)
197
198 static inline void wrmsrl(unsigned msr, u64 val)
199 {
200         wrmsr(msr, (u32)val, (u32)(val>>32));
201 }
202
203 #define wrmsr_safe(msr, a, b)   paravirt_write_msr_safe(msr, a, b)
204
205 /* rdmsr with exception handling */
206 #define rdmsr_safe(msr, a, b)                           \
207 ({                                                      \
208         int _err;                                       \
209         u64 _l = paravirt_read_msr_safe(msr, &_err);    \
210         (*a) = (u32)_l;                                 \
211         (*b) = _l >> 32;                                \
212         _err;                                           \
213 })
214
215 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
216 {
217         int err;
218
219         *p = paravirt_read_msr_safe(msr, &err);
220         return err;
221 }
222
223 static inline unsigned long long paravirt_read_pmc(int counter)
224 {
225         return PVOP_CALL1(u64, cpu.read_pmc, counter);
226 }
227
228 #define rdpmc(counter, low, high)               \
229 do {                                            \
230         u64 _l = paravirt_read_pmc(counter);    \
231         low = (u32)_l;                          \
232         high = _l >> 32;                        \
233 } while (0)
234
235 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter))
236
237 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
238 {
239         PVOP_VCALL2(cpu.alloc_ldt, ldt, entries);
240 }
241
242 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
243 {
244         PVOP_VCALL2(cpu.free_ldt, ldt, entries);
245 }
246
247 static inline void load_TR_desc(void)
248 {
249         PVOP_VCALL0(cpu.load_tr_desc);
250 }
251 static inline void load_gdt(const struct desc_ptr *dtr)
252 {
253         PVOP_VCALL1(cpu.load_gdt, dtr);
254 }
255 static inline void load_idt(const struct desc_ptr *dtr)
256 {
257         PVOP_VCALL1(cpu.load_idt, dtr);
258 }
259 static inline void set_ldt(const void *addr, unsigned entries)
260 {
261         PVOP_VCALL2(cpu.set_ldt, addr, entries);
262 }
263 static inline unsigned long paravirt_store_tr(void)
264 {
265         return PVOP_CALL0(unsigned long, cpu.store_tr);
266 }
267
268 #define store_tr(tr)    ((tr) = paravirt_store_tr())
269 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
270 {
271         PVOP_VCALL2(cpu.load_tls, t, cpu);
272 }
273
274 #ifdef CONFIG_X86_64
275 static inline void load_gs_index(unsigned int gs)
276 {
277         PVOP_VCALL1(cpu.load_gs_index, gs);
278 }
279 #endif
280
281 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
282                                    const void *desc)
283 {
284         PVOP_VCALL3(cpu.write_ldt_entry, dt, entry, desc);
285 }
286
287 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
288                                    void *desc, int type)
289 {
290         PVOP_VCALL4(cpu.write_gdt_entry, dt, entry, desc, type);
291 }
292
293 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
294 {
295         PVOP_VCALL3(cpu.write_idt_entry, dt, entry, g);
296 }
297
298 #ifdef CONFIG_X86_IOPL_IOPERM
299 static inline void tss_update_io_bitmap(void)
300 {
301         PVOP_VCALL0(cpu.update_io_bitmap);
302 }
303 #endif
304
305 static inline void paravirt_activate_mm(struct mm_struct *prev,
306                                         struct mm_struct *next)
307 {
308         PVOP_VCALL2(mmu.activate_mm, prev, next);
309 }
310
311 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
312                                           struct mm_struct *mm)
313 {
314         PVOP_VCALL2(mmu.dup_mmap, oldmm, mm);
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 vm_area_struct *vma, unsigned long addr,
417                                            pte_t *ptep)
418 {
419         pteval_t ret;
420
421         ret = PVOP_CALL3(pteval_t, mmu.ptep_modify_prot_start, vma, addr, ptep);
422
423         return (pte_t) { .pte = ret };
424 }
425
426 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
427                                            pte_t *ptep, pte_t old_pte, pte_t pte)
428 {
429
430         if (sizeof(pteval_t) > sizeof(long))
431                 /* 5 arg words */
432                 pv_ops.mmu.ptep_modify_prot_commit(vma, addr, ptep, pte);
433         else
434                 PVOP_VCALL4(mmu.ptep_modify_prot_commit,
435                             vma, addr, ptep, pte.pte);
436 }
437
438 static inline void set_pte(pte_t *ptep, pte_t pte)
439 {
440         if (sizeof(pteval_t) > sizeof(long))
441                 PVOP_VCALL3(mmu.set_pte, ptep, pte.pte, (u64)pte.pte >> 32);
442         else
443                 PVOP_VCALL2(mmu.set_pte, ptep, pte.pte);
444 }
445
446 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
447                               pte_t *ptep, pte_t pte)
448 {
449         if (sizeof(pteval_t) > sizeof(long))
450                 /* 5 arg words */
451                 pv_ops.mmu.set_pte_at(mm, addr, ptep, pte);
452         else
453                 PVOP_VCALL4(mmu.set_pte_at, mm, addr, ptep, pte.pte);
454 }
455
456 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
457 {
458         pmdval_t val = native_pmd_val(pmd);
459
460         if (sizeof(pmdval_t) > sizeof(long))
461                 PVOP_VCALL3(mmu.set_pmd, pmdp, val, (u64)val >> 32);
462         else
463                 PVOP_VCALL2(mmu.set_pmd, pmdp, val);
464 }
465
466 #if CONFIG_PGTABLE_LEVELS >= 3
467 static inline pmd_t __pmd(pmdval_t val)
468 {
469         pmdval_t ret;
470
471         if (sizeof(pmdval_t) > sizeof(long))
472                 ret = PVOP_CALLEE2(pmdval_t, mmu.make_pmd, val, (u64)val >> 32);
473         else
474                 ret = PVOP_CALLEE1(pmdval_t, mmu.make_pmd, val);
475
476         return (pmd_t) { ret };
477 }
478
479 static inline pmdval_t pmd_val(pmd_t pmd)
480 {
481         pmdval_t ret;
482
483         if (sizeof(pmdval_t) > sizeof(long))
484                 ret =  PVOP_CALLEE2(pmdval_t, mmu.pmd_val,
485                                     pmd.pmd, (u64)pmd.pmd >> 32);
486         else
487                 ret =  PVOP_CALLEE1(pmdval_t, mmu.pmd_val, pmd.pmd);
488
489         return ret;
490 }
491
492 static inline void set_pud(pud_t *pudp, pud_t pud)
493 {
494         pudval_t val = native_pud_val(pud);
495
496         if (sizeof(pudval_t) > sizeof(long))
497                 PVOP_VCALL3(mmu.set_pud, pudp, val, (u64)val >> 32);
498         else
499                 PVOP_VCALL2(mmu.set_pud, pudp, val);
500 }
501 #if CONFIG_PGTABLE_LEVELS >= 4
502 static inline pud_t __pud(pudval_t val)
503 {
504         pudval_t ret;
505
506         ret = PVOP_CALLEE1(pudval_t, mmu.make_pud, val);
507
508         return (pud_t) { ret };
509 }
510
511 static inline pudval_t pud_val(pud_t pud)
512 {
513         return PVOP_CALLEE1(pudval_t, mmu.pud_val, pud.pud);
514 }
515
516 static inline void pud_clear(pud_t *pudp)
517 {
518         set_pud(pudp, __pud(0));
519 }
520
521 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
522 {
523         p4dval_t val = native_p4d_val(p4d);
524
525         PVOP_VCALL2(mmu.set_p4d, p4dp, val);
526 }
527
528 #if CONFIG_PGTABLE_LEVELS >= 5
529
530 static inline p4d_t __p4d(p4dval_t val)
531 {
532         p4dval_t ret = PVOP_CALLEE1(p4dval_t, mmu.make_p4d, val);
533
534         return (p4d_t) { ret };
535 }
536
537 static inline p4dval_t p4d_val(p4d_t p4d)
538 {
539         return PVOP_CALLEE1(p4dval_t, mmu.p4d_val, p4d.p4d);
540 }
541
542 static inline void __set_pgd(pgd_t *pgdp, pgd_t pgd)
543 {
544         PVOP_VCALL2(mmu.set_pgd, pgdp, native_pgd_val(pgd));
545 }
546
547 #define set_pgd(pgdp, pgdval) do {                                      \
548         if (pgtable_l5_enabled())                                               \
549                 __set_pgd(pgdp, pgdval);                                \
550         else                                                            \
551                 set_p4d((p4d_t *)(pgdp), (p4d_t) { (pgdval).pgd });     \
552 } while (0)
553
554 #define pgd_clear(pgdp) do {                                            \
555         if (pgtable_l5_enabled())                                               \
556                 set_pgd(pgdp, __pgd(0));                                \
557 } while (0)
558
559 #endif  /* CONFIG_PGTABLE_LEVELS == 5 */
560
561 static inline void p4d_clear(p4d_t *p4dp)
562 {
563         set_p4d(p4dp, __p4d(0));
564 }
565
566 #endif  /* CONFIG_PGTABLE_LEVELS == 4 */
567
568 #endif  /* CONFIG_PGTABLE_LEVELS >= 3 */
569
570 #ifdef CONFIG_X86_PAE
571 /* Special-case pte-setting operations for PAE, which can't update a
572    64-bit pte atomically */
573 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
574 {
575         PVOP_VCALL3(mmu.set_pte_atomic, ptep, pte.pte, pte.pte >> 32);
576 }
577
578 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
579                              pte_t *ptep)
580 {
581         PVOP_VCALL3(mmu.pte_clear, mm, addr, ptep);
582 }
583
584 static inline void pmd_clear(pmd_t *pmdp)
585 {
586         PVOP_VCALL1(mmu.pmd_clear, pmdp);
587 }
588 #else  /* !CONFIG_X86_PAE */
589 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
590 {
591         set_pte(ptep, pte);
592 }
593
594 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
595                              pte_t *ptep)
596 {
597         set_pte_at(mm, addr, ptep, __pte(0));
598 }
599
600 static inline void pmd_clear(pmd_t *pmdp)
601 {
602         set_pmd(pmdp, __pmd(0));
603 }
604 #endif  /* CONFIG_X86_PAE */
605
606 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
607 static inline void arch_start_context_switch(struct task_struct *prev)
608 {
609         PVOP_VCALL1(cpu.start_context_switch, prev);
610 }
611
612 static inline void arch_end_context_switch(struct task_struct *next)
613 {
614         PVOP_VCALL1(cpu.end_context_switch, next);
615 }
616
617 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
618 static inline void arch_enter_lazy_mmu_mode(void)
619 {
620         PVOP_VCALL0(mmu.lazy_mode.enter);
621 }
622
623 static inline void arch_leave_lazy_mmu_mode(void)
624 {
625         PVOP_VCALL0(mmu.lazy_mode.leave);
626 }
627
628 static inline void arch_flush_lazy_mmu_mode(void)
629 {
630         PVOP_VCALL0(mmu.lazy_mode.flush);
631 }
632
633 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
634                                 phys_addr_t phys, pgprot_t flags)
635 {
636         pv_ops.mmu.set_fixmap(idx, phys, flags);
637 }
638 #endif
639
640 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
641
642 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock,
643                                                         u32 val)
644 {
645         PVOP_VCALL2(lock.queued_spin_lock_slowpath, lock, val);
646 }
647
648 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock)
649 {
650         PVOP_VCALLEE1(lock.queued_spin_unlock, lock);
651 }
652
653 static __always_inline void pv_wait(u8 *ptr, u8 val)
654 {
655         PVOP_VCALL2(lock.wait, ptr, val);
656 }
657
658 static __always_inline void pv_kick(int cpu)
659 {
660         PVOP_VCALL1(lock.kick, cpu);
661 }
662
663 static __always_inline bool pv_vcpu_is_preempted(long cpu)
664 {
665         return PVOP_CALLEE1(bool, lock.vcpu_is_preempted, cpu);
666 }
667
668 void __raw_callee_save___native_queued_spin_unlock(struct qspinlock *lock);
669 bool __raw_callee_save___native_vcpu_is_preempted(long cpu);
670
671 #endif /* SMP && PARAVIRT_SPINLOCKS */
672
673 #ifdef CONFIG_X86_32
674 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
675 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
676
677 /* save and restore all caller-save registers, except return value */
678 #define PV_SAVE_ALL_CALLER_REGS         "pushl %ecx;"
679 #define PV_RESTORE_ALL_CALLER_REGS      "popl  %ecx;"
680
681 #define PV_FLAGS_ARG "0"
682 #define PV_EXTRA_CLOBBERS
683 #define PV_VEXTRA_CLOBBERS
684 #else
685 /* save and restore all caller-save registers, except return value */
686 #define PV_SAVE_ALL_CALLER_REGS                                         \
687         "push %rcx;"                                                    \
688         "push %rdx;"                                                    \
689         "push %rsi;"                                                    \
690         "push %rdi;"                                                    \
691         "push %r8;"                                                     \
692         "push %r9;"                                                     \
693         "push %r10;"                                                    \
694         "push %r11;"
695 #define PV_RESTORE_ALL_CALLER_REGS                                      \
696         "pop %r11;"                                                     \
697         "pop %r10;"                                                     \
698         "pop %r9;"                                                      \
699         "pop %r8;"                                                      \
700         "pop %rdi;"                                                     \
701         "pop %rsi;"                                                     \
702         "pop %rdx;"                                                     \
703         "pop %rcx;"
704
705 /* We save some registers, but all of them, that's too much. We clobber all
706  * caller saved registers but the argument parameter */
707 #define PV_SAVE_REGS "pushq %%rdi;"
708 #define PV_RESTORE_REGS "popq %%rdi;"
709 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
710 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
711 #define PV_FLAGS_ARG "D"
712 #endif
713
714 /*
715  * Generate a thunk around a function which saves all caller-save
716  * registers except for the return value.  This allows C functions to
717  * be called from assembler code where fewer than normal registers are
718  * available.  It may also help code generation around calls from C
719  * code if the common case doesn't use many registers.
720  *
721  * When a callee is wrapped in a thunk, the caller can assume that all
722  * arg regs and all scratch registers are preserved across the
723  * call. The return value in rax/eax will not be saved, even for void
724  * functions.
725  */
726 #define PV_THUNK_NAME(func) "__raw_callee_save_" #func
727 #define PV_CALLEE_SAVE_REGS_THUNK(func)                                 \
728         extern typeof(func) __raw_callee_save_##func;                   \
729                                                                         \
730         asm(".pushsection .text;"                                       \
731             ".globl " PV_THUNK_NAME(func) ";"                           \
732             ".type " PV_THUNK_NAME(func) ", @function;"                 \
733             PV_THUNK_NAME(func) ":"                                     \
734             FRAME_BEGIN                                                 \
735             PV_SAVE_ALL_CALLER_REGS                                     \
736             "call " #func ";"                                           \
737             PV_RESTORE_ALL_CALLER_REGS                                  \
738             FRAME_END                                                   \
739             "ret;"                                                      \
740             ".size " PV_THUNK_NAME(func) ", .-" PV_THUNK_NAME(func) ";" \
741             ".popsection")
742
743 /* Get a reference to a callee-save function */
744 #define PV_CALLEE_SAVE(func)                                            \
745         ((struct paravirt_callee_save) { __raw_callee_save_##func })
746
747 /* Promise that "func" already uses the right calling convention */
748 #define __PV_IS_CALLEE_SAVE(func)                       \
749         ((struct paravirt_callee_save) { func })
750
751 #ifdef CONFIG_PARAVIRT_XXL
752 static inline notrace unsigned long arch_local_save_flags(void)
753 {
754         return PVOP_CALLEE0(unsigned long, irq.save_fl);
755 }
756
757 static inline notrace void arch_local_irq_restore(unsigned long f)
758 {
759         PVOP_VCALLEE1(irq.restore_fl, f);
760 }
761
762 static inline notrace void arch_local_irq_disable(void)
763 {
764         PVOP_VCALLEE0(irq.irq_disable);
765 }
766
767 static inline notrace void arch_local_irq_enable(void)
768 {
769         PVOP_VCALLEE0(irq.irq_enable);
770 }
771
772 static inline notrace unsigned long arch_local_irq_save(void)
773 {
774         unsigned long f;
775
776         f = arch_local_save_flags();
777         arch_local_irq_disable();
778         return f;
779 }
780 #endif
781
782
783 /* Make sure as little as possible of this mess escapes. */
784 #undef PARAVIRT_CALL
785 #undef __PVOP_CALL
786 #undef __PVOP_VCALL
787 #undef PVOP_VCALL0
788 #undef PVOP_CALL0
789 #undef PVOP_VCALL1
790 #undef PVOP_CALL1
791 #undef PVOP_VCALL2
792 #undef PVOP_CALL2
793 #undef PVOP_VCALL3
794 #undef PVOP_CALL3
795 #undef PVOP_VCALL4
796 #undef PVOP_CALL4
797
798 extern void default_banner(void);
799
800 #else  /* __ASSEMBLY__ */
801
802 #define _PVSITE(ptype, ops, word, algn)         \
803 771:;                                           \
804         ops;                                    \
805 772:;                                           \
806         .pushsection .parainstructions,"a";     \
807          .align algn;                           \
808          word 771b;                             \
809          .byte ptype;                           \
810          .byte 772b-771b;                       \
811         .popsection
812
813
814 #define COND_PUSH(set, mask, reg)                       \
815         .if ((~(set)) & mask); push %reg; .endif
816 #define COND_POP(set, mask, reg)                        \
817         .if ((~(set)) & mask); pop %reg; .endif
818
819 #ifdef CONFIG_X86_64
820
821 #define PV_SAVE_REGS(set)                       \
822         COND_PUSH(set, CLBR_RAX, rax);          \
823         COND_PUSH(set, CLBR_RCX, rcx);          \
824         COND_PUSH(set, CLBR_RDX, rdx);          \
825         COND_PUSH(set, CLBR_RSI, rsi);          \
826         COND_PUSH(set, CLBR_RDI, rdi);          \
827         COND_PUSH(set, CLBR_R8, r8);            \
828         COND_PUSH(set, CLBR_R9, r9);            \
829         COND_PUSH(set, CLBR_R10, r10);          \
830         COND_PUSH(set, CLBR_R11, r11)
831 #define PV_RESTORE_REGS(set)                    \
832         COND_POP(set, CLBR_R11, r11);           \
833         COND_POP(set, CLBR_R10, r10);           \
834         COND_POP(set, CLBR_R9, r9);             \
835         COND_POP(set, CLBR_R8, r8);             \
836         COND_POP(set, CLBR_RDI, rdi);           \
837         COND_POP(set, CLBR_RSI, rsi);           \
838         COND_POP(set, CLBR_RDX, rdx);           \
839         COND_POP(set, CLBR_RCX, rcx);           \
840         COND_POP(set, CLBR_RAX, rax)
841
842 #define PARA_PATCH(off)         ((off) / 8)
843 #define PARA_SITE(ptype, ops)   _PVSITE(ptype, ops, .quad, 8)
844 #define PARA_INDIRECT(addr)     *addr(%rip)
845 #else
846 #define PV_SAVE_REGS(set)                       \
847         COND_PUSH(set, CLBR_EAX, eax);          \
848         COND_PUSH(set, CLBR_EDI, edi);          \
849         COND_PUSH(set, CLBR_ECX, ecx);          \
850         COND_PUSH(set, CLBR_EDX, edx)
851 #define PV_RESTORE_REGS(set)                    \
852         COND_POP(set, CLBR_EDX, edx);           \
853         COND_POP(set, CLBR_ECX, ecx);           \
854         COND_POP(set, CLBR_EDI, edi);           \
855         COND_POP(set, CLBR_EAX, eax)
856
857 #define PARA_PATCH(off)         ((off) / 4)
858 #define PARA_SITE(ptype, ops)   _PVSITE(ptype, ops, .long, 4)
859 #define PARA_INDIRECT(addr)     *%cs:addr
860 #endif
861
862 #ifdef CONFIG_PARAVIRT_XXL
863 #define INTERRUPT_RETURN                                                \
864         PARA_SITE(PARA_PATCH(PV_CPU_iret),                              \
865                   ANNOTATE_RETPOLINE_SAFE;                              \
866                   jmp PARA_INDIRECT(pv_ops+PV_CPU_iret);)
867
868 #define DISABLE_INTERRUPTS(clobbers)                                    \
869         PARA_SITE(PARA_PATCH(PV_IRQ_irq_disable),                       \
870                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
871                   ANNOTATE_RETPOLINE_SAFE;                              \
872                   call PARA_INDIRECT(pv_ops+PV_IRQ_irq_disable);        \
873                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
874
875 #define ENABLE_INTERRUPTS(clobbers)                                     \
876         PARA_SITE(PARA_PATCH(PV_IRQ_irq_enable),                        \
877                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
878                   ANNOTATE_RETPOLINE_SAFE;                              \
879                   call PARA_INDIRECT(pv_ops+PV_IRQ_irq_enable);         \
880                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
881 #endif
882
883 #ifdef CONFIG_X86_64
884 #ifdef CONFIG_PARAVIRT_XXL
885 /*
886  * If swapgs is used while the userspace stack is still current,
887  * there's no way to call a pvop.  The PV replacement *must* be
888  * inlined, or the swapgs instruction must be trapped and emulated.
889  */
890 #define SWAPGS_UNSAFE_STACK                                             \
891         PARA_SITE(PARA_PATCH(PV_CPU_swapgs), swapgs)
892
893 /*
894  * Note: swapgs is very special, and in practise is either going to be
895  * implemented with a single "swapgs" instruction or something very
896  * special.  Either way, we don't need to save any registers for
897  * it.
898  */
899 #define SWAPGS                                                          \
900         PARA_SITE(PARA_PATCH(PV_CPU_swapgs),                            \
901                   ANNOTATE_RETPOLINE_SAFE;                              \
902                   call PARA_INDIRECT(pv_ops+PV_CPU_swapgs);             \
903                  )
904
905 #define USERGS_SYSRET64                                                 \
906         PARA_SITE(PARA_PATCH(PV_CPU_usergs_sysret64),                   \
907                   ANNOTATE_RETPOLINE_SAFE;                              \
908                   jmp PARA_INDIRECT(pv_ops+PV_CPU_usergs_sysret64);)
909
910 #ifdef CONFIG_DEBUG_ENTRY
911 #define SAVE_FLAGS(clobbers)                                        \
912         PARA_SITE(PARA_PATCH(PV_IRQ_save_fl),                       \
913                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);        \
914                   ANNOTATE_RETPOLINE_SAFE;                          \
915                   call PARA_INDIRECT(pv_ops+PV_IRQ_save_fl);        \
916                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
917 #endif
918 #endif /* CONFIG_PARAVIRT_XXL */
919 #endif  /* CONFIG_X86_64 */
920
921 #ifdef CONFIG_PARAVIRT_XXL
922
923 #define GET_CR2_INTO_AX                                                 \
924         PARA_SITE(PARA_PATCH(PV_MMU_read_cr2),                          \
925                   ANNOTATE_RETPOLINE_SAFE;                              \
926                   call PARA_INDIRECT(pv_ops+PV_MMU_read_cr2);           \
927                  )
928
929 #endif /* CONFIG_PARAVIRT_XXL */
930
931
932 #endif /* __ASSEMBLY__ */
933 #else  /* CONFIG_PARAVIRT */
934 # define default_banner x86_init_noop
935 #endif /* !CONFIG_PARAVIRT */
936
937 #ifndef __ASSEMBLY__
938 #ifndef CONFIG_PARAVIRT_XXL
939 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
940                                           struct mm_struct *mm)
941 {
942 }
943 #endif
944
945 #ifndef CONFIG_PARAVIRT
946 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
947 {
948 }
949 #endif
950 #endif /* __ASSEMBLY__ */
951 #endif /* _ASM_X86_PARAVIRT_H */