Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
[firefly-linux-kernel-4.4.55.git] / arch / x86 / include / asm / paravirt.h
1 #ifndef _ASM_X86_PARAVIRT_H
2 #define _ASM_X86_PARAVIRT_H
3 /* Various instructions on x86 need to be replaced for
4  * para-virtualization: those hooks are defined here. */
5
6 #ifdef CONFIG_PARAVIRT
7 #include <asm/pgtable_types.h>
8 #include <asm/asm.h>
9
10 #include <asm/paravirt_types.h>
11
12 #ifndef __ASSEMBLY__
13 #include <linux/bug.h>
14 #include <linux/types.h>
15 #include <linux/cpumask.h>
16
17 static inline int paravirt_enabled(void)
18 {
19         return pv_info.paravirt_enabled;
20 }
21
22 static inline void load_sp0(struct tss_struct *tss,
23                              struct thread_struct *thread)
24 {
25         PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
26 }
27
28 /* The paravirtualized CPUID instruction. */
29 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
30                            unsigned int *ecx, unsigned int *edx)
31 {
32         PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
33 }
34
35 /*
36  * These special macros can be used to get or set a debugging register
37  */
38 static inline unsigned long paravirt_get_debugreg(int reg)
39 {
40         return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
41 }
42 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
43 static inline void set_debugreg(unsigned long val, int reg)
44 {
45         PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
46 }
47
48 static inline void clts(void)
49 {
50         PVOP_VCALL0(pv_cpu_ops.clts);
51 }
52
53 static inline unsigned long read_cr0(void)
54 {
55         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
56 }
57
58 static inline void write_cr0(unsigned long x)
59 {
60         PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
61 }
62
63 static inline unsigned long read_cr2(void)
64 {
65         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
66 }
67
68 static inline void write_cr2(unsigned long x)
69 {
70         PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
71 }
72
73 static inline unsigned long read_cr3(void)
74 {
75         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
76 }
77
78 static inline void write_cr3(unsigned long x)
79 {
80         PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
81 }
82
83 static inline unsigned long read_cr4(void)
84 {
85         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
86 }
87 static inline unsigned long read_cr4_safe(void)
88 {
89         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
90 }
91
92 static inline void write_cr4(unsigned long x)
93 {
94         PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
95 }
96
97 #ifdef CONFIG_X86_64
98 static inline unsigned long read_cr8(void)
99 {
100         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
101 }
102
103 static inline void write_cr8(unsigned long x)
104 {
105         PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
106 }
107 #endif
108
109 static inline void arch_safe_halt(void)
110 {
111         PVOP_VCALL0(pv_irq_ops.safe_halt);
112 }
113
114 static inline void halt(void)
115 {
116         PVOP_VCALL0(pv_irq_ops.halt);
117 }
118
119 static inline void wbinvd(void)
120 {
121         PVOP_VCALL0(pv_cpu_ops.wbinvd);
122 }
123
124 #define get_kernel_rpl()  (pv_info.kernel_rpl)
125
126 static inline u64 paravirt_read_msr(unsigned msr, int *err)
127 {
128         return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
129 }
130
131 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
132 {
133         return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
134 }
135
136 /* These should all do BUG_ON(_err), but our headers are too tangled. */
137 #define rdmsr(msr, val1, val2)                  \
138 do {                                            \
139         int _err;                               \
140         u64 _l = paravirt_read_msr(msr, &_err); \
141         val1 = (u32)_l;                         \
142         val2 = _l >> 32;                        \
143 } while (0)
144
145 #define wrmsr(msr, val1, val2)                  \
146 do {                                            \
147         paravirt_write_msr(msr, val1, val2);    \
148 } while (0)
149
150 #define rdmsrl(msr, val)                        \
151 do {                                            \
152         int _err;                               \
153         val = paravirt_read_msr(msr, &_err);    \
154 } while (0)
155
156 #define wrmsrl(msr, val)        wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
157 #define wrmsr_safe(msr, a, b)   paravirt_write_msr(msr, a, b)
158
159 /* rdmsr with exception handling */
160 #define rdmsr_safe(msr, a, b)                   \
161 ({                                              \
162         int _err;                               \
163         u64 _l = paravirt_read_msr(msr, &_err); \
164         (*a) = (u32)_l;                         \
165         (*b) = _l >> 32;                        \
166         _err;                                   \
167 })
168
169 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
170 {
171         int err;
172
173         *p = paravirt_read_msr(msr, &err);
174         return err;
175 }
176
177 static inline u64 paravirt_read_tsc(void)
178 {
179         return PVOP_CALL0(u64, pv_cpu_ops.read_tsc);
180 }
181
182 #define rdtscl(low)                             \
183 do {                                            \
184         u64 _l = paravirt_read_tsc();           \
185         low = (int)_l;                          \
186 } while (0)
187
188 #define rdtscll(val) (val = paravirt_read_tsc())
189
190 static inline unsigned long long paravirt_sched_clock(void)
191 {
192         return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
193 }
194
195 struct static_key;
196 extern struct static_key paravirt_steal_enabled;
197 extern struct static_key paravirt_steal_rq_enabled;
198
199 static inline u64 paravirt_steal_clock(int cpu)
200 {
201         return PVOP_CALL1(u64, pv_time_ops.steal_clock, cpu);
202 }
203
204 static inline unsigned long long paravirt_read_pmc(int counter)
205 {
206         return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
207 }
208
209 #define rdpmc(counter, low, high)               \
210 do {                                            \
211         u64 _l = paravirt_read_pmc(counter);    \
212         low = (u32)_l;                          \
213         high = _l >> 32;                        \
214 } while (0)
215
216 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter))
217
218 static inline unsigned long long paravirt_rdtscp(unsigned int *aux)
219 {
220         return PVOP_CALL1(u64, pv_cpu_ops.read_tscp, aux);
221 }
222
223 #define rdtscp(low, high, aux)                          \
224 do {                                                    \
225         int __aux;                                      \
226         unsigned long __val = paravirt_rdtscp(&__aux);  \
227         (low) = (u32)__val;                             \
228         (high) = (u32)(__val >> 32);                    \
229         (aux) = __aux;                                  \
230 } while (0)
231
232 #define rdtscpll(val, aux)                              \
233 do {                                                    \
234         unsigned long __aux;                            \
235         val = paravirt_rdtscp(&__aux);                  \
236         (aux) = __aux;                                  \
237 } while (0)
238
239 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
240 {
241         PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
242 }
243
244 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
245 {
246         PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
247 }
248
249 static inline void load_TR_desc(void)
250 {
251         PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
252 }
253 static inline void load_gdt(const struct desc_ptr *dtr)
254 {
255         PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
256 }
257 static inline void load_idt(const struct desc_ptr *dtr)
258 {
259         PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
260 }
261 static inline void set_ldt(const void *addr, unsigned entries)
262 {
263         PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
264 }
265 static inline void store_gdt(struct desc_ptr *dtr)
266 {
267         PVOP_VCALL1(pv_cpu_ops.store_gdt, dtr);
268 }
269 static inline void store_idt(struct desc_ptr *dtr)
270 {
271         PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
272 }
273 static inline unsigned long paravirt_store_tr(void)
274 {
275         return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
276 }
277 #define store_tr(tr)    ((tr) = paravirt_store_tr())
278 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
279 {
280         PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
281 }
282
283 #ifdef CONFIG_X86_64
284 static inline void load_gs_index(unsigned int gs)
285 {
286         PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
287 }
288 #endif
289
290 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
291                                    const void *desc)
292 {
293         PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
294 }
295
296 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
297                                    void *desc, int type)
298 {
299         PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
300 }
301
302 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
303 {
304         PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
305 }
306 static inline void set_iopl_mask(unsigned mask)
307 {
308         PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
309 }
310
311 /* The paravirtualized I/O functions */
312 static inline void slow_down_io(void)
313 {
314         pv_cpu_ops.io_delay();
315 #ifdef REALLY_SLOW_IO
316         pv_cpu_ops.io_delay();
317         pv_cpu_ops.io_delay();
318         pv_cpu_ops.io_delay();
319 #endif
320 }
321
322 #ifdef CONFIG_SMP
323 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
324                                     unsigned long start_esp)
325 {
326         PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
327                     phys_apicid, start_eip, start_esp);
328 }
329 #endif
330
331 static inline void paravirt_activate_mm(struct mm_struct *prev,
332                                         struct mm_struct *next)
333 {
334         PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
335 }
336
337 static inline void arch_dup_mmap(struct mm_struct *oldmm,
338                                  struct mm_struct *mm)
339 {
340         PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
341 }
342
343 static inline void arch_exit_mmap(struct mm_struct *mm)
344 {
345         PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
346 }
347
348 static inline void __flush_tlb(void)
349 {
350         PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
351 }
352 static inline void __flush_tlb_global(void)
353 {
354         PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
355 }
356 static inline void __flush_tlb_single(unsigned long addr)
357 {
358         PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
359 }
360
361 static inline void flush_tlb_others(const struct cpumask *cpumask,
362                                     struct mm_struct *mm,
363                                     unsigned long va)
364 {
365         PVOP_VCALL3(pv_mmu_ops.flush_tlb_others, cpumask, mm, va);
366 }
367
368 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
369 {
370         return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
371 }
372
373 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
374 {
375         PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
376 }
377
378 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
379 {
380         PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
381 }
382 static inline void paravirt_release_pte(unsigned long pfn)
383 {
384         PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
385 }
386
387 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
388 {
389         PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
390 }
391
392 static inline void paravirt_release_pmd(unsigned long pfn)
393 {
394         PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
395 }
396
397 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
398 {
399         PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
400 }
401 static inline void paravirt_release_pud(unsigned long pfn)
402 {
403         PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
404 }
405
406 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
407                               pte_t *ptep)
408 {
409         PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
410 }
411 static inline void pmd_update(struct mm_struct *mm, unsigned long addr,
412                               pmd_t *pmdp)
413 {
414         PVOP_VCALL3(pv_mmu_ops.pmd_update, mm, addr, pmdp);
415 }
416
417 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
418                                     pte_t *ptep)
419 {
420         PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
421 }
422
423 static inline void pmd_update_defer(struct mm_struct *mm, unsigned long addr,
424                                     pmd_t *pmdp)
425 {
426         PVOP_VCALL3(pv_mmu_ops.pmd_update_defer, mm, addr, pmdp);
427 }
428
429 static inline pte_t __pte(pteval_t val)
430 {
431         pteval_t ret;
432
433         if (sizeof(pteval_t) > sizeof(long))
434                 ret = PVOP_CALLEE2(pteval_t,
435                                    pv_mmu_ops.make_pte,
436                                    val, (u64)val >> 32);
437         else
438                 ret = PVOP_CALLEE1(pteval_t,
439                                    pv_mmu_ops.make_pte,
440                                    val);
441
442         return (pte_t) { .pte = ret };
443 }
444
445 static inline pteval_t pte_val(pte_t pte)
446 {
447         pteval_t ret;
448
449         if (sizeof(pteval_t) > sizeof(long))
450                 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
451                                    pte.pte, (u64)pte.pte >> 32);
452         else
453                 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
454                                    pte.pte);
455
456         return ret;
457 }
458
459 static inline pgd_t __pgd(pgdval_t val)
460 {
461         pgdval_t ret;
462
463         if (sizeof(pgdval_t) > sizeof(long))
464                 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
465                                    val, (u64)val >> 32);
466         else
467                 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
468                                    val);
469
470         return (pgd_t) { ret };
471 }
472
473 static inline pgdval_t pgd_val(pgd_t pgd)
474 {
475         pgdval_t ret;
476
477         if (sizeof(pgdval_t) > sizeof(long))
478                 ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
479                                     pgd.pgd, (u64)pgd.pgd >> 32);
480         else
481                 ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
482                                     pgd.pgd);
483
484         return ret;
485 }
486
487 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
488 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
489                                            pte_t *ptep)
490 {
491         pteval_t ret;
492
493         ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
494                          mm, addr, ptep);
495
496         return (pte_t) { .pte = ret };
497 }
498
499 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
500                                            pte_t *ptep, pte_t pte)
501 {
502         if (sizeof(pteval_t) > sizeof(long))
503                 /* 5 arg words */
504                 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
505         else
506                 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
507                             mm, addr, ptep, pte.pte);
508 }
509
510 static inline void set_pte(pte_t *ptep, pte_t pte)
511 {
512         if (sizeof(pteval_t) > sizeof(long))
513                 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
514                             pte.pte, (u64)pte.pte >> 32);
515         else
516                 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
517                             pte.pte);
518 }
519
520 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
521                               pte_t *ptep, pte_t pte)
522 {
523         if (sizeof(pteval_t) > sizeof(long))
524                 /* 5 arg words */
525                 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
526         else
527                 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
528 }
529
530 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
531 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
532                               pmd_t *pmdp, pmd_t pmd)
533 {
534         if (sizeof(pmdval_t) > sizeof(long))
535                 /* 5 arg words */
536                 pv_mmu_ops.set_pmd_at(mm, addr, pmdp, pmd);
537         else
538                 PVOP_VCALL4(pv_mmu_ops.set_pmd_at, mm, addr, pmdp,
539                             native_pmd_val(pmd));
540 }
541 #endif
542
543 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
544 {
545         pmdval_t val = native_pmd_val(pmd);
546
547         if (sizeof(pmdval_t) > sizeof(long))
548                 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
549         else
550                 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
551 }
552
553 #if PAGETABLE_LEVELS >= 3
554 static inline pmd_t __pmd(pmdval_t val)
555 {
556         pmdval_t ret;
557
558         if (sizeof(pmdval_t) > sizeof(long))
559                 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
560                                    val, (u64)val >> 32);
561         else
562                 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
563                                    val);
564
565         return (pmd_t) { ret };
566 }
567
568 static inline pmdval_t pmd_val(pmd_t pmd)
569 {
570         pmdval_t ret;
571
572         if (sizeof(pmdval_t) > sizeof(long))
573                 ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
574                                     pmd.pmd, (u64)pmd.pmd >> 32);
575         else
576                 ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
577                                     pmd.pmd);
578
579         return ret;
580 }
581
582 static inline void set_pud(pud_t *pudp, pud_t pud)
583 {
584         pudval_t val = native_pud_val(pud);
585
586         if (sizeof(pudval_t) > sizeof(long))
587                 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
588                             val, (u64)val >> 32);
589         else
590                 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
591                             val);
592 }
593 #if PAGETABLE_LEVELS == 4
594 static inline pud_t __pud(pudval_t val)
595 {
596         pudval_t ret;
597
598         if (sizeof(pudval_t) > sizeof(long))
599                 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
600                                    val, (u64)val >> 32);
601         else
602                 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
603                                    val);
604
605         return (pud_t) { ret };
606 }
607
608 static inline pudval_t pud_val(pud_t pud)
609 {
610         pudval_t ret;
611
612         if (sizeof(pudval_t) > sizeof(long))
613                 ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
614                                     pud.pud, (u64)pud.pud >> 32);
615         else
616                 ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
617                                     pud.pud);
618
619         return ret;
620 }
621
622 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
623 {
624         pgdval_t val = native_pgd_val(pgd);
625
626         if (sizeof(pgdval_t) > sizeof(long))
627                 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
628                             val, (u64)val >> 32);
629         else
630                 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
631                             val);
632 }
633
634 static inline void pgd_clear(pgd_t *pgdp)
635 {
636         set_pgd(pgdp, __pgd(0));
637 }
638
639 static inline void pud_clear(pud_t *pudp)
640 {
641         set_pud(pudp, __pud(0));
642 }
643
644 #endif  /* PAGETABLE_LEVELS == 4 */
645
646 #endif  /* PAGETABLE_LEVELS >= 3 */
647
648 #ifdef CONFIG_X86_PAE
649 /* Special-case pte-setting operations for PAE, which can't update a
650    64-bit pte atomically */
651 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
652 {
653         PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
654                     pte.pte, pte.pte >> 32);
655 }
656
657 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
658                              pte_t *ptep)
659 {
660         PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
661 }
662
663 static inline void pmd_clear(pmd_t *pmdp)
664 {
665         PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
666 }
667 #else  /* !CONFIG_X86_PAE */
668 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
669 {
670         set_pte(ptep, pte);
671 }
672
673 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
674                              pte_t *ptep)
675 {
676         set_pte_at(mm, addr, ptep, __pte(0));
677 }
678
679 static inline void pmd_clear(pmd_t *pmdp)
680 {
681         set_pmd(pmdp, __pmd(0));
682 }
683 #endif  /* CONFIG_X86_PAE */
684
685 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
686 static inline void arch_start_context_switch(struct task_struct *prev)
687 {
688         PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
689 }
690
691 static inline void arch_end_context_switch(struct task_struct *next)
692 {
693         PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
694 }
695
696 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
697 static inline void arch_enter_lazy_mmu_mode(void)
698 {
699         PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
700 }
701
702 static inline void arch_leave_lazy_mmu_mode(void)
703 {
704         PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
705 }
706
707 void arch_flush_lazy_mmu_mode(void);
708
709 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
710                                 phys_addr_t phys, pgprot_t flags)
711 {
712         pv_mmu_ops.set_fixmap(idx, phys, flags);
713 }
714
715 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
716
717 static inline int arch_spin_is_locked(struct arch_spinlock *lock)
718 {
719         return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
720 }
721
722 static inline int arch_spin_is_contended(struct arch_spinlock *lock)
723 {
724         return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
725 }
726 #define arch_spin_is_contended  arch_spin_is_contended
727
728 static __always_inline void arch_spin_lock(struct arch_spinlock *lock)
729 {
730         PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
731 }
732
733 static __always_inline void arch_spin_lock_flags(struct arch_spinlock *lock,
734                                                   unsigned long flags)
735 {
736         PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
737 }
738
739 static __always_inline int arch_spin_trylock(struct arch_spinlock *lock)
740 {
741         return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
742 }
743
744 static __always_inline void arch_spin_unlock(struct arch_spinlock *lock)
745 {
746         PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
747 }
748
749 #endif
750
751 #ifdef CONFIG_X86_32
752 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
753 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
754
755 /* save and restore all caller-save registers, except return value */
756 #define PV_SAVE_ALL_CALLER_REGS         "pushl %ecx;"
757 #define PV_RESTORE_ALL_CALLER_REGS      "popl  %ecx;"
758
759 #define PV_FLAGS_ARG "0"
760 #define PV_EXTRA_CLOBBERS
761 #define PV_VEXTRA_CLOBBERS
762 #else
763 /* save and restore all caller-save registers, except return value */
764 #define PV_SAVE_ALL_CALLER_REGS                                         \
765         "push %rcx;"                                                    \
766         "push %rdx;"                                                    \
767         "push %rsi;"                                                    \
768         "push %rdi;"                                                    \
769         "push %r8;"                                                     \
770         "push %r9;"                                                     \
771         "push %r10;"                                                    \
772         "push %r11;"
773 #define PV_RESTORE_ALL_CALLER_REGS                                      \
774         "pop %r11;"                                                     \
775         "pop %r10;"                                                     \
776         "pop %r9;"                                                      \
777         "pop %r8;"                                                      \
778         "pop %rdi;"                                                     \
779         "pop %rsi;"                                                     \
780         "pop %rdx;"                                                     \
781         "pop %rcx;"
782
783 /* We save some registers, but all of them, that's too much. We clobber all
784  * caller saved registers but the argument parameter */
785 #define PV_SAVE_REGS "pushq %%rdi;"
786 #define PV_RESTORE_REGS "popq %%rdi;"
787 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
788 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
789 #define PV_FLAGS_ARG "D"
790 #endif
791
792 /*
793  * Generate a thunk around a function which saves all caller-save
794  * registers except for the return value.  This allows C functions to
795  * be called from assembler code where fewer than normal registers are
796  * available.  It may also help code generation around calls from C
797  * code if the common case doesn't use many registers.
798  *
799  * When a callee is wrapped in a thunk, the caller can assume that all
800  * arg regs and all scratch registers are preserved across the
801  * call. The return value in rax/eax will not be saved, even for void
802  * functions.
803  */
804 #define PV_CALLEE_SAVE_REGS_THUNK(func)                                 \
805         extern typeof(func) __raw_callee_save_##func;                   \
806         static void *__##func##__ __used = func;                        \
807                                                                         \
808         asm(".pushsection .text;"                                       \
809             "__raw_callee_save_" #func ": "                             \
810             PV_SAVE_ALL_CALLER_REGS                                     \
811             "call " #func ";"                                           \
812             PV_RESTORE_ALL_CALLER_REGS                                  \
813             "ret;"                                                      \
814             ".popsection")
815
816 /* Get a reference to a callee-save function */
817 #define PV_CALLEE_SAVE(func)                                            \
818         ((struct paravirt_callee_save) { __raw_callee_save_##func })
819
820 /* Promise that "func" already uses the right calling convention */
821 #define __PV_IS_CALLEE_SAVE(func)                       \
822         ((struct paravirt_callee_save) { func })
823
824 static inline notrace unsigned long arch_local_save_flags(void)
825 {
826         return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
827 }
828
829 static inline notrace void arch_local_irq_restore(unsigned long f)
830 {
831         PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
832 }
833
834 static inline notrace void arch_local_irq_disable(void)
835 {
836         PVOP_VCALLEE0(pv_irq_ops.irq_disable);
837 }
838
839 static inline notrace void arch_local_irq_enable(void)
840 {
841         PVOP_VCALLEE0(pv_irq_ops.irq_enable);
842 }
843
844 static inline notrace unsigned long arch_local_irq_save(void)
845 {
846         unsigned long f;
847
848         f = arch_local_save_flags();
849         arch_local_irq_disable();
850         return f;
851 }
852
853
854 /* Make sure as little as possible of this mess escapes. */
855 #undef PARAVIRT_CALL
856 #undef __PVOP_CALL
857 #undef __PVOP_VCALL
858 #undef PVOP_VCALL0
859 #undef PVOP_CALL0
860 #undef PVOP_VCALL1
861 #undef PVOP_CALL1
862 #undef PVOP_VCALL2
863 #undef PVOP_CALL2
864 #undef PVOP_VCALL3
865 #undef PVOP_CALL3
866 #undef PVOP_VCALL4
867 #undef PVOP_CALL4
868
869 extern void default_banner(void);
870
871 #else  /* __ASSEMBLY__ */
872
873 #define _PVSITE(ptype, clobbers, ops, word, algn)       \
874 771:;                                           \
875         ops;                                    \
876 772:;                                           \
877         .pushsection .parainstructions,"a";     \
878          .align algn;                           \
879          word 771b;                             \
880          .byte ptype;                           \
881          .byte 772b-771b;                       \
882          .short clobbers;                       \
883         .popsection
884
885
886 #define COND_PUSH(set, mask, reg)                       \
887         .if ((~(set)) & mask); push %reg; .endif
888 #define COND_POP(set, mask, reg)                        \
889         .if ((~(set)) & mask); pop %reg; .endif
890
891 #ifdef CONFIG_X86_64
892
893 #define PV_SAVE_REGS(set)                       \
894         COND_PUSH(set, CLBR_RAX, rax);          \
895         COND_PUSH(set, CLBR_RCX, rcx);          \
896         COND_PUSH(set, CLBR_RDX, rdx);          \
897         COND_PUSH(set, CLBR_RSI, rsi);          \
898         COND_PUSH(set, CLBR_RDI, rdi);          \
899         COND_PUSH(set, CLBR_R8, r8);            \
900         COND_PUSH(set, CLBR_R9, r9);            \
901         COND_PUSH(set, CLBR_R10, r10);          \
902         COND_PUSH(set, CLBR_R11, r11)
903 #define PV_RESTORE_REGS(set)                    \
904         COND_POP(set, CLBR_R11, r11);           \
905         COND_POP(set, CLBR_R10, r10);           \
906         COND_POP(set, CLBR_R9, r9);             \
907         COND_POP(set, CLBR_R8, r8);             \
908         COND_POP(set, CLBR_RDI, rdi);           \
909         COND_POP(set, CLBR_RSI, rsi);           \
910         COND_POP(set, CLBR_RDX, rdx);           \
911         COND_POP(set, CLBR_RCX, rcx);           \
912         COND_POP(set, CLBR_RAX, rax)
913
914 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
915 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
916 #define PARA_INDIRECT(addr)     *addr(%rip)
917 #else
918 #define PV_SAVE_REGS(set)                       \
919         COND_PUSH(set, CLBR_EAX, eax);          \
920         COND_PUSH(set, CLBR_EDI, edi);          \
921         COND_PUSH(set, CLBR_ECX, ecx);          \
922         COND_PUSH(set, CLBR_EDX, edx)
923 #define PV_RESTORE_REGS(set)                    \
924         COND_POP(set, CLBR_EDX, edx);           \
925         COND_POP(set, CLBR_ECX, ecx);           \
926         COND_POP(set, CLBR_EDI, edi);           \
927         COND_POP(set, CLBR_EAX, eax)
928
929 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
930 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
931 #define PARA_INDIRECT(addr)     *%cs:addr
932 #endif
933
934 #define INTERRUPT_RETURN                                                \
935         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,       \
936                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
937
938 #define DISABLE_INTERRUPTS(clobbers)                                    \
939         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
940                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
941                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);    \
942                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
943
944 #define ENABLE_INTERRUPTS(clobbers)                                     \
945         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,  \
946                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
947                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);     \
948                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
949
950 #define USERGS_SYSRET32                                                 \
951         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),       \
952                   CLBR_NONE,                                            \
953                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
954
955 #ifdef CONFIG_X86_32
956 #define GET_CR0_INTO_EAX                                \
957         push %ecx; push %edx;                           \
958         call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
959         pop %edx; pop %ecx
960
961 #define ENABLE_INTERRUPTS_SYSEXIT                                       \
962         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
963                   CLBR_NONE,                                            \
964                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
965
966
967 #else   /* !CONFIG_X86_32 */
968
969 /*
970  * If swapgs is used while the userspace stack is still current,
971  * there's no way to call a pvop.  The PV replacement *must* be
972  * inlined, or the swapgs instruction must be trapped and emulated.
973  */
974 #define SWAPGS_UNSAFE_STACK                                             \
975         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
976                   swapgs)
977
978 /*
979  * Note: swapgs is very special, and in practise is either going to be
980  * implemented with a single "swapgs" instruction or something very
981  * special.  Either way, we don't need to save any registers for
982  * it.
983  */
984 #define SWAPGS                                                          \
985         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
986                   call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)          \
987                  )
988
989 #define GET_CR2_INTO_RAX                                \
990         call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2)
991
992 #define PARAVIRT_ADJUST_EXCEPTION_FRAME                                 \
993         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
994                   CLBR_NONE,                                            \
995                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
996
997 #define USERGS_SYSRET64                                                 \
998         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),       \
999                   CLBR_NONE,                                            \
1000                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1001
1002 #define ENABLE_INTERRUPTS_SYSEXIT32                                     \
1003         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
1004                   CLBR_NONE,                                            \
1005                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1006 #endif  /* CONFIG_X86_32 */
1007
1008 #endif /* __ASSEMBLY__ */
1009 #else  /* CONFIG_PARAVIRT */
1010 # define default_banner x86_init_noop
1011 #endif /* !CONFIG_PARAVIRT */
1012 #endif /* _ASM_X86_PARAVIRT_H */