Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-next-2.6
[linux-2.6.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/types.h>
14 #include <linux/cpumask.h>
15
16 static inline int paravirt_enabled(void)
17 {
18         return pv_info.paravirt_enabled;
19 }
20
21 static inline void load_sp0(struct tss_struct *tss,
22                              struct thread_struct *thread)
23 {
24         PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
25 }
26
27 /* The paravirtualized CPUID instruction. */
28 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
29                            unsigned int *ecx, unsigned int *edx)
30 {
31         PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
32 }
33
34 /*
35  * These special macros can be used to get or set a debugging register
36  */
37 static inline unsigned long paravirt_get_debugreg(int reg)
38 {
39         return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
40 }
41 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
42 static inline void set_debugreg(unsigned long val, int reg)
43 {
44         PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
45 }
46
47 static inline void clts(void)
48 {
49         PVOP_VCALL0(pv_cpu_ops.clts);
50 }
51
52 static inline unsigned long read_cr0(void)
53 {
54         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
55 }
56
57 static inline void write_cr0(unsigned long x)
58 {
59         PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
60 }
61
62 static inline unsigned long read_cr2(void)
63 {
64         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
65 }
66
67 static inline void write_cr2(unsigned long x)
68 {
69         PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
70 }
71
72 static inline unsigned long read_cr3(void)
73 {
74         return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
75 }
76
77 static inline void write_cr3(unsigned long x)
78 {
79         PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
80 }
81
82 static inline unsigned long read_cr4(void)
83 {
84         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
85 }
86 static inline unsigned long read_cr4_safe(void)
87 {
88         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
89 }
90
91 static inline void write_cr4(unsigned long x)
92 {
93         PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
94 }
95
96 #ifdef CONFIG_X86_64
97 static inline unsigned long read_cr8(void)
98 {
99         return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
100 }
101
102 static inline void write_cr8(unsigned long x)
103 {
104         PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
105 }
106 #endif
107
108 static inline void raw_safe_halt(void)
109 {
110         PVOP_VCALL0(pv_irq_ops.safe_halt);
111 }
112
113 static inline void halt(void)
114 {
115         PVOP_VCALL0(pv_irq_ops.safe_halt);
116 }
117
118 static inline void wbinvd(void)
119 {
120         PVOP_VCALL0(pv_cpu_ops.wbinvd);
121 }
122
123 #define get_kernel_rpl()  (pv_info.kernel_rpl)
124
125 static inline u64 paravirt_read_msr(unsigned msr, int *err)
126 {
127         return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
128 }
129
130 static inline int paravirt_rdmsr_regs(u32 *regs)
131 {
132         return PVOP_CALL1(int, pv_cpu_ops.rdmsr_regs, regs);
133 }
134
135 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
136 {
137         return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
138 }
139
140 static inline int paravirt_wrmsr_regs(u32 *regs)
141 {
142         return PVOP_CALL1(int, pv_cpu_ops.wrmsr_regs, regs);
143 }
144
145 /* These should all do BUG_ON(_err), but our headers are too tangled. */
146 #define rdmsr(msr, val1, val2)                  \
147 do {                                            \
148         int _err;                               \
149         u64 _l = paravirt_read_msr(msr, &_err); \
150         val1 = (u32)_l;                         \
151         val2 = _l >> 32;                        \
152 } while (0)
153
154 #define wrmsr(msr, val1, val2)                  \
155 do {                                            \
156         paravirt_write_msr(msr, val1, val2);    \
157 } while (0)
158
159 #define rdmsrl(msr, val)                        \
160 do {                                            \
161         int _err;                               \
162         val = paravirt_read_msr(msr, &_err);    \
163 } while (0)
164
165 #define wrmsrl(msr, val)        wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
166 #define wrmsr_safe(msr, a, b)   paravirt_write_msr(msr, a, b)
167
168 /* rdmsr with exception handling */
169 #define rdmsr_safe(msr, a, b)                   \
170 ({                                              \
171         int _err;                               \
172         u64 _l = paravirt_read_msr(msr, &_err); \
173         (*a) = (u32)_l;                         \
174         (*b) = _l >> 32;                        \
175         _err;                                   \
176 })
177
178 #define rdmsr_safe_regs(regs)   paravirt_rdmsr_regs(regs)
179 #define wrmsr_safe_regs(regs)   paravirt_wrmsr_regs(regs)
180
181 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
182 {
183         int err;
184
185         *p = paravirt_read_msr(msr, &err);
186         return err;
187 }
188 static inline int rdmsrl_amd_safe(unsigned msr, unsigned long long *p)
189 {
190         u32 gprs[8] = { 0 };
191         int err;
192
193         gprs[1] = msr;
194         gprs[7] = 0x9c5a203a;
195
196         err = paravirt_rdmsr_regs(gprs);
197
198         *p = gprs[0] | ((u64)gprs[2] << 32);
199
200         return err;
201 }
202
203 static inline int wrmsrl_amd_safe(unsigned msr, unsigned long long val)
204 {
205         u32 gprs[8] = { 0 };
206
207         gprs[0] = (u32)val;
208         gprs[1] = msr;
209         gprs[2] = val >> 32;
210         gprs[7] = 0x9c5a203a;
211
212         return paravirt_wrmsr_regs(gprs);
213 }
214
215 static inline u64 paravirt_read_tsc(void)
216 {
217         return PVOP_CALL0(u64, pv_cpu_ops.read_tsc);
218 }
219
220 #define rdtscl(low)                             \
221 do {                                            \
222         u64 _l = paravirt_read_tsc();           \
223         low = (int)_l;                          \
224 } while (0)
225
226 #define rdtscll(val) (val = paravirt_read_tsc())
227
228 static inline unsigned long long paravirt_sched_clock(void)
229 {
230         return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
231 }
232
233 static inline unsigned long long paravirt_read_pmc(int counter)
234 {
235         return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
236 }
237
238 #define rdpmc(counter, low, high)               \
239 do {                                            \
240         u64 _l = paravirt_read_pmc(counter);    \
241         low = (u32)_l;                          \
242         high = _l >> 32;                        \
243 } while (0)
244
245 static inline unsigned long long paravirt_rdtscp(unsigned int *aux)
246 {
247         return PVOP_CALL1(u64, pv_cpu_ops.read_tscp, aux);
248 }
249
250 #define rdtscp(low, high, aux)                          \
251 do {                                                    \
252         int __aux;                                      \
253         unsigned long __val = paravirt_rdtscp(&__aux);  \
254         (low) = (u32)__val;                             \
255         (high) = (u32)(__val >> 32);                    \
256         (aux) = __aux;                                  \
257 } while (0)
258
259 #define rdtscpll(val, aux)                              \
260 do {                                                    \
261         unsigned long __aux;                            \
262         val = paravirt_rdtscp(&__aux);                  \
263         (aux) = __aux;                                  \
264 } while (0)
265
266 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
267 {
268         PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
269 }
270
271 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
272 {
273         PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
274 }
275
276 static inline void load_TR_desc(void)
277 {
278         PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
279 }
280 static inline void load_gdt(const struct desc_ptr *dtr)
281 {
282         PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
283 }
284 static inline void load_idt(const struct desc_ptr *dtr)
285 {
286         PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
287 }
288 static inline void set_ldt(const void *addr, unsigned entries)
289 {
290         PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
291 }
292 static inline void store_gdt(struct desc_ptr *dtr)
293 {
294         PVOP_VCALL1(pv_cpu_ops.store_gdt, dtr);
295 }
296 static inline void store_idt(struct desc_ptr *dtr)
297 {
298         PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
299 }
300 static inline unsigned long paravirt_store_tr(void)
301 {
302         return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
303 }
304 #define store_tr(tr)    ((tr) = paravirt_store_tr())
305 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
306 {
307         PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
308 }
309
310 #ifdef CONFIG_X86_64
311 static inline void load_gs_index(unsigned int gs)
312 {
313         PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
314 }
315 #endif
316
317 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
318                                    const void *desc)
319 {
320         PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
321 }
322
323 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
324                                    void *desc, int type)
325 {
326         PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
327 }
328
329 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
330 {
331         PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
332 }
333 static inline void set_iopl_mask(unsigned mask)
334 {
335         PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
336 }
337
338 /* The paravirtualized I/O functions */
339 static inline void slow_down_io(void)
340 {
341         pv_cpu_ops.io_delay();
342 #ifdef REALLY_SLOW_IO
343         pv_cpu_ops.io_delay();
344         pv_cpu_ops.io_delay();
345         pv_cpu_ops.io_delay();
346 #endif
347 }
348
349 #ifdef CONFIG_SMP
350 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
351                                     unsigned long start_esp)
352 {
353         PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
354                     phys_apicid, start_eip, start_esp);
355 }
356 #endif
357
358 static inline void paravirt_activate_mm(struct mm_struct *prev,
359                                         struct mm_struct *next)
360 {
361         PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
362 }
363
364 static inline void arch_dup_mmap(struct mm_struct *oldmm,
365                                  struct mm_struct *mm)
366 {
367         PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
368 }
369
370 static inline void arch_exit_mmap(struct mm_struct *mm)
371 {
372         PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
373 }
374
375 static inline void __flush_tlb(void)
376 {
377         PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
378 }
379 static inline void __flush_tlb_global(void)
380 {
381         PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
382 }
383 static inline void __flush_tlb_single(unsigned long addr)
384 {
385         PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
386 }
387
388 static inline void flush_tlb_others(const struct cpumask *cpumask,
389                                     struct mm_struct *mm,
390                                     unsigned long va)
391 {
392         PVOP_VCALL3(pv_mmu_ops.flush_tlb_others, cpumask, mm, va);
393 }
394
395 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
396 {
397         return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
398 }
399
400 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
401 {
402         PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
403 }
404
405 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
406 {
407         PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
408 }
409 static inline void paravirt_release_pte(unsigned long pfn)
410 {
411         PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
412 }
413
414 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
415 {
416         PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
417 }
418
419 static inline void paravirt_alloc_pmd_clone(unsigned long pfn, unsigned long clonepfn,
420                                             unsigned long start, unsigned long count)
421 {
422         PVOP_VCALL4(pv_mmu_ops.alloc_pmd_clone, pfn, clonepfn, start, count);
423 }
424 static inline void paravirt_release_pmd(unsigned long pfn)
425 {
426         PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
427 }
428
429 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
430 {
431         PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
432 }
433 static inline void paravirt_release_pud(unsigned long pfn)
434 {
435         PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
436 }
437
438 #ifdef CONFIG_HIGHPTE
439 static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
440 {
441         unsigned long ret;
442         ret = PVOP_CALL2(unsigned long, pv_mmu_ops.kmap_atomic_pte, page, type);
443         return (void *)ret;
444 }
445 #endif
446
447 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
448                               pte_t *ptep)
449 {
450         PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
451 }
452
453 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
454                                     pte_t *ptep)
455 {
456         PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
457 }
458
459 static inline pte_t __pte(pteval_t val)
460 {
461         pteval_t ret;
462
463         if (sizeof(pteval_t) > sizeof(long))
464                 ret = PVOP_CALLEE2(pteval_t,
465                                    pv_mmu_ops.make_pte,
466                                    val, (u64)val >> 32);
467         else
468                 ret = PVOP_CALLEE1(pteval_t,
469                                    pv_mmu_ops.make_pte,
470                                    val);
471
472         return (pte_t) { .pte = ret };
473 }
474
475 static inline pteval_t pte_val(pte_t pte)
476 {
477         pteval_t ret;
478
479         if (sizeof(pteval_t) > sizeof(long))
480                 ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
481                                    pte.pte, (u64)pte.pte >> 32);
482         else
483                 ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
484                                    pte.pte);
485
486         return ret;
487 }
488
489 static inline pgd_t __pgd(pgdval_t val)
490 {
491         pgdval_t ret;
492
493         if (sizeof(pgdval_t) > sizeof(long))
494                 ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
495                                    val, (u64)val >> 32);
496         else
497                 ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
498                                    val);
499
500         return (pgd_t) { ret };
501 }
502
503 static inline pgdval_t pgd_val(pgd_t pgd)
504 {
505         pgdval_t ret;
506
507         if (sizeof(pgdval_t) > sizeof(long))
508                 ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
509                                     pgd.pgd, (u64)pgd.pgd >> 32);
510         else
511                 ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
512                                     pgd.pgd);
513
514         return ret;
515 }
516
517 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
518 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
519                                            pte_t *ptep)
520 {
521         pteval_t ret;
522
523         ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
524                          mm, addr, ptep);
525
526         return (pte_t) { .pte = ret };
527 }
528
529 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
530                                            pte_t *ptep, pte_t pte)
531 {
532         if (sizeof(pteval_t) > sizeof(long))
533                 /* 5 arg words */
534                 pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
535         else
536                 PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
537                             mm, addr, ptep, pte.pte);
538 }
539
540 static inline void set_pte(pte_t *ptep, pte_t pte)
541 {
542         if (sizeof(pteval_t) > sizeof(long))
543                 PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
544                             pte.pte, (u64)pte.pte >> 32);
545         else
546                 PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
547                             pte.pte);
548 }
549
550 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
551                               pte_t *ptep, pte_t pte)
552 {
553         if (sizeof(pteval_t) > sizeof(long))
554                 /* 5 arg words */
555                 pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
556         else
557                 PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
558 }
559
560 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
561 {
562         pmdval_t val = native_pmd_val(pmd);
563
564         if (sizeof(pmdval_t) > sizeof(long))
565                 PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
566         else
567                 PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
568 }
569
570 #if PAGETABLE_LEVELS >= 3
571 static inline pmd_t __pmd(pmdval_t val)
572 {
573         pmdval_t ret;
574
575         if (sizeof(pmdval_t) > sizeof(long))
576                 ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
577                                    val, (u64)val >> 32);
578         else
579                 ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
580                                    val);
581
582         return (pmd_t) { ret };
583 }
584
585 static inline pmdval_t pmd_val(pmd_t pmd)
586 {
587         pmdval_t ret;
588
589         if (sizeof(pmdval_t) > sizeof(long))
590                 ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
591                                     pmd.pmd, (u64)pmd.pmd >> 32);
592         else
593                 ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
594                                     pmd.pmd);
595
596         return ret;
597 }
598
599 static inline void set_pud(pud_t *pudp, pud_t pud)
600 {
601         pudval_t val = native_pud_val(pud);
602
603         if (sizeof(pudval_t) > sizeof(long))
604                 PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
605                             val, (u64)val >> 32);
606         else
607                 PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
608                             val);
609 }
610 #if PAGETABLE_LEVELS == 4
611 static inline pud_t __pud(pudval_t val)
612 {
613         pudval_t ret;
614
615         if (sizeof(pudval_t) > sizeof(long))
616                 ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
617                                    val, (u64)val >> 32);
618         else
619                 ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
620                                    val);
621
622         return (pud_t) { ret };
623 }
624
625 static inline pudval_t pud_val(pud_t pud)
626 {
627         pudval_t ret;
628
629         if (sizeof(pudval_t) > sizeof(long))
630                 ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
631                                     pud.pud, (u64)pud.pud >> 32);
632         else
633                 ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
634                                     pud.pud);
635
636         return ret;
637 }
638
639 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
640 {
641         pgdval_t val = native_pgd_val(pgd);
642
643         if (sizeof(pgdval_t) > sizeof(long))
644                 PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
645                             val, (u64)val >> 32);
646         else
647                 PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
648                             val);
649 }
650
651 static inline void pgd_clear(pgd_t *pgdp)
652 {
653         set_pgd(pgdp, __pgd(0));
654 }
655
656 static inline void pud_clear(pud_t *pudp)
657 {
658         set_pud(pudp, __pud(0));
659 }
660
661 #endif  /* PAGETABLE_LEVELS == 4 */
662
663 #endif  /* PAGETABLE_LEVELS >= 3 */
664
665 #ifdef CONFIG_X86_PAE
666 /* Special-case pte-setting operations for PAE, which can't update a
667    64-bit pte atomically */
668 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
669 {
670         PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
671                     pte.pte, pte.pte >> 32);
672 }
673
674 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
675                              pte_t *ptep)
676 {
677         PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
678 }
679
680 static inline void pmd_clear(pmd_t *pmdp)
681 {
682         PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
683 }
684 #else  /* !CONFIG_X86_PAE */
685 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
686 {
687         set_pte(ptep, pte);
688 }
689
690 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
691                              pte_t *ptep)
692 {
693         set_pte_at(mm, addr, ptep, __pte(0));
694 }
695
696 static inline void pmd_clear(pmd_t *pmdp)
697 {
698         set_pmd(pmdp, __pmd(0));
699 }
700 #endif  /* CONFIG_X86_PAE */
701
702 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
703 static inline void arch_start_context_switch(struct task_struct *prev)
704 {
705         PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
706 }
707
708 static inline void arch_end_context_switch(struct task_struct *next)
709 {
710         PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
711 }
712
713 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
714 static inline void arch_enter_lazy_mmu_mode(void)
715 {
716         PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
717 }
718
719 static inline void arch_leave_lazy_mmu_mode(void)
720 {
721         PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
722 }
723
724 void arch_flush_lazy_mmu_mode(void);
725
726 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
727                                 phys_addr_t phys, pgprot_t flags)
728 {
729         pv_mmu_ops.set_fixmap(idx, phys, flags);
730 }
731
732 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
733
734 static inline int __raw_spin_is_locked(struct raw_spinlock *lock)
735 {
736         return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
737 }
738
739 static inline int __raw_spin_is_contended(struct raw_spinlock *lock)
740 {
741         return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
742 }
743 #define __raw_spin_is_contended __raw_spin_is_contended
744
745 static __always_inline void __raw_spin_lock(struct raw_spinlock *lock)
746 {
747         PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
748 }
749
750 static __always_inline void __raw_spin_lock_flags(struct raw_spinlock *lock,
751                                                   unsigned long flags)
752 {
753         PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
754 }
755
756 static __always_inline int __raw_spin_trylock(struct raw_spinlock *lock)
757 {
758         return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
759 }
760
761 static __always_inline void __raw_spin_unlock(struct raw_spinlock *lock)
762 {
763         PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
764 }
765
766 #endif
767
768 #ifdef CONFIG_X86_32
769 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
770 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
771
772 /* save and restore all caller-save registers, except return value */
773 #define PV_SAVE_ALL_CALLER_REGS         "pushl %ecx;"
774 #define PV_RESTORE_ALL_CALLER_REGS      "popl  %ecx;"
775
776 #define PV_FLAGS_ARG "0"
777 #define PV_EXTRA_CLOBBERS
778 #define PV_VEXTRA_CLOBBERS
779 #else
780 /* save and restore all caller-save registers, except return value */
781 #define PV_SAVE_ALL_CALLER_REGS                                         \
782         "push %rcx;"                                                    \
783         "push %rdx;"                                                    \
784         "push %rsi;"                                                    \
785         "push %rdi;"                                                    \
786         "push %r8;"                                                     \
787         "push %r9;"                                                     \
788         "push %r10;"                                                    \
789         "push %r11;"
790 #define PV_RESTORE_ALL_CALLER_REGS                                      \
791         "pop %r11;"                                                     \
792         "pop %r10;"                                                     \
793         "pop %r9;"                                                      \
794         "pop %r8;"                                                      \
795         "pop %rdi;"                                                     \
796         "pop %rsi;"                                                     \
797         "pop %rdx;"                                                     \
798         "pop %rcx;"
799
800 /* We save some registers, but all of them, that's too much. We clobber all
801  * caller saved registers but the argument parameter */
802 #define PV_SAVE_REGS "pushq %%rdi;"
803 #define PV_RESTORE_REGS "popq %%rdi;"
804 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
805 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
806 #define PV_FLAGS_ARG "D"
807 #endif
808
809 /*
810  * Generate a thunk around a function which saves all caller-save
811  * registers except for the return value.  This allows C functions to
812  * be called from assembler code where fewer than normal registers are
813  * available.  It may also help code generation around calls from C
814  * code if the common case doesn't use many registers.
815  *
816  * When a callee is wrapped in a thunk, the caller can assume that all
817  * arg regs and all scratch registers are preserved across the
818  * call. The return value in rax/eax will not be saved, even for void
819  * functions.
820  */
821 #define PV_CALLEE_SAVE_REGS_THUNK(func)                                 \
822         extern typeof(func) __raw_callee_save_##func;                   \
823         static void *__##func##__ __used = func;                        \
824                                                                         \
825         asm(".pushsection .text;"                                       \
826             "__raw_callee_save_" #func ": "                             \
827             PV_SAVE_ALL_CALLER_REGS                                     \
828             "call " #func ";"                                           \
829             PV_RESTORE_ALL_CALLER_REGS                                  \
830             "ret;"                                                      \
831             ".popsection")
832
833 /* Get a reference to a callee-save function */
834 #define PV_CALLEE_SAVE(func)                                            \
835         ((struct paravirt_callee_save) { __raw_callee_save_##func })
836
837 /* Promise that "func" already uses the right calling convention */
838 #define __PV_IS_CALLEE_SAVE(func)                       \
839         ((struct paravirt_callee_save) { func })
840
841 static inline unsigned long __raw_local_save_flags(void)
842 {
843         return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
844 }
845
846 static inline void raw_local_irq_restore(unsigned long f)
847 {
848         PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
849 }
850
851 static inline void raw_local_irq_disable(void)
852 {
853         PVOP_VCALLEE0(pv_irq_ops.irq_disable);
854 }
855
856 static inline void raw_local_irq_enable(void)
857 {
858         PVOP_VCALLEE0(pv_irq_ops.irq_enable);
859 }
860
861 static inline unsigned long __raw_local_irq_save(void)
862 {
863         unsigned long f;
864
865         f = __raw_local_save_flags();
866         raw_local_irq_disable();
867         return f;
868 }
869
870
871 /* Make sure as little as possible of this mess escapes. */
872 #undef PARAVIRT_CALL
873 #undef __PVOP_CALL
874 #undef __PVOP_VCALL
875 #undef PVOP_VCALL0
876 #undef PVOP_CALL0
877 #undef PVOP_VCALL1
878 #undef PVOP_CALL1
879 #undef PVOP_VCALL2
880 #undef PVOP_CALL2
881 #undef PVOP_VCALL3
882 #undef PVOP_CALL3
883 #undef PVOP_VCALL4
884 #undef PVOP_CALL4
885
886 extern void default_banner(void);
887
888 #else  /* __ASSEMBLY__ */
889
890 #define _PVSITE(ptype, clobbers, ops, word, algn)       \
891 771:;                                           \
892         ops;                                    \
893 772:;                                           \
894         .pushsection .parainstructions,"a";     \
895          .align algn;                           \
896          word 771b;                             \
897          .byte ptype;                           \
898          .byte 772b-771b;                       \
899          .short clobbers;                       \
900         .popsection
901
902
903 #define COND_PUSH(set, mask, reg)                       \
904         .if ((~(set)) & mask); push %reg; .endif
905 #define COND_POP(set, mask, reg)                        \
906         .if ((~(set)) & mask); pop %reg; .endif
907
908 #ifdef CONFIG_X86_64
909
910 #define PV_SAVE_REGS(set)                       \
911         COND_PUSH(set, CLBR_RAX, rax);          \
912         COND_PUSH(set, CLBR_RCX, rcx);          \
913         COND_PUSH(set, CLBR_RDX, rdx);          \
914         COND_PUSH(set, CLBR_RSI, rsi);          \
915         COND_PUSH(set, CLBR_RDI, rdi);          \
916         COND_PUSH(set, CLBR_R8, r8);            \
917         COND_PUSH(set, CLBR_R9, r9);            \
918         COND_PUSH(set, CLBR_R10, r10);          \
919         COND_PUSH(set, CLBR_R11, r11)
920 #define PV_RESTORE_REGS(set)                    \
921         COND_POP(set, CLBR_R11, r11);           \
922         COND_POP(set, CLBR_R10, r10);           \
923         COND_POP(set, CLBR_R9, r9);             \
924         COND_POP(set, CLBR_R8, r8);             \
925         COND_POP(set, CLBR_RDI, rdi);           \
926         COND_POP(set, CLBR_RSI, rsi);           \
927         COND_POP(set, CLBR_RDX, rdx);           \
928         COND_POP(set, CLBR_RCX, rcx);           \
929         COND_POP(set, CLBR_RAX, rax)
930
931 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
932 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
933 #define PARA_INDIRECT(addr)     *addr(%rip)
934 #else
935 #define PV_SAVE_REGS(set)                       \
936         COND_PUSH(set, CLBR_EAX, eax);          \
937         COND_PUSH(set, CLBR_EDI, edi);          \
938         COND_PUSH(set, CLBR_ECX, ecx);          \
939         COND_PUSH(set, CLBR_EDX, edx)
940 #define PV_RESTORE_REGS(set)                    \
941         COND_POP(set, CLBR_EDX, edx);           \
942         COND_POP(set, CLBR_ECX, ecx);           \
943         COND_POP(set, CLBR_EDI, edi);           \
944         COND_POP(set, CLBR_EAX, eax)
945
946 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
947 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
948 #define PARA_INDIRECT(addr)     *%cs:addr
949 #endif
950
951 #define INTERRUPT_RETURN                                                \
952         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,       \
953                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
954
955 #define DISABLE_INTERRUPTS(clobbers)                                    \
956         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
957                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
958                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);    \
959                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
960
961 #define ENABLE_INTERRUPTS(clobbers)                                     \
962         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,  \
963                   PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);            \
964                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);     \
965                   PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
966
967 #define USERGS_SYSRET32                                                 \
968         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),       \
969                   CLBR_NONE,                                            \
970                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
971
972 #ifdef CONFIG_X86_32
973 #define GET_CR0_INTO_EAX                                \
974         push %ecx; push %edx;                           \
975         call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0); \
976         pop %edx; pop %ecx
977
978 #define ENABLE_INTERRUPTS_SYSEXIT                                       \
979         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
980                   CLBR_NONE,                                            \
981                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
982
983
984 #else   /* !CONFIG_X86_32 */
985
986 /*
987  * If swapgs is used while the userspace stack is still current,
988  * there's no way to call a pvop.  The PV replacement *must* be
989  * inlined, or the swapgs instruction must be trapped and emulated.
990  */
991 #define SWAPGS_UNSAFE_STACK                                             \
992         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
993                   swapgs)
994
995 /*
996  * Note: swapgs is very special, and in practise is either going to be
997  * implemented with a single "swapgs" instruction or something very
998  * special.  Either way, we don't need to save any registers for
999  * it.
1000  */
1001 #define SWAPGS                                                          \
1002         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,     \
1003                   call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)          \
1004                  )
1005
1006 #define GET_CR2_INTO_RCX                                \
1007         call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2); \
1008         movq %rax, %rcx;                                \
1009         xorq %rax, %rax;
1010
1011 #define PARAVIRT_ADJUST_EXCEPTION_FRAME                                 \
1012         PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
1013                   CLBR_NONE,                                            \
1014                   call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
1015
1016 #define USERGS_SYSRET64                                                 \
1017         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),       \
1018                   CLBR_NONE,                                            \
1019                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1020
1021 #define ENABLE_INTERRUPTS_SYSEXIT32                                     \
1022         PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),    \
1023                   CLBR_NONE,                                            \
1024                   jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1025 #endif  /* CONFIG_X86_32 */
1026
1027 #endif /* __ASSEMBLY__ */
1028 #else  /* CONFIG_PARAVIRT */
1029 # define default_banner x86_init_noop
1030 #endif /* !CONFIG_PARAVIRT */
1031 #endif /* _ASM_X86_PARAVIRT_H */