xref: /linux/arch/x86/include/asm/paravirt.h (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 start,
364 				    unsigned long end)
365 {
366 	PVOP_VCALL4(pv_mmu_ops.flush_tlb_others, cpumask, mm, start, end);
367 }
368 
369 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
370 {
371 	return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
372 }
373 
374 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
375 {
376 	PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
377 }
378 
379 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
380 {
381 	PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
382 }
383 static inline void paravirt_release_pte(unsigned long pfn)
384 {
385 	PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
386 }
387 
388 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
389 {
390 	PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
391 }
392 
393 static inline void paravirt_release_pmd(unsigned long pfn)
394 {
395 	PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
396 }
397 
398 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
399 {
400 	PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
401 }
402 static inline void paravirt_release_pud(unsigned long pfn)
403 {
404 	PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
405 }
406 
407 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
408 			      pte_t *ptep)
409 {
410 	PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
411 }
412 static inline void pmd_update(struct mm_struct *mm, unsigned long addr,
413 			      pmd_t *pmdp)
414 {
415 	PVOP_VCALL3(pv_mmu_ops.pmd_update, mm, addr, pmdp);
416 }
417 
418 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
419 				    pte_t *ptep)
420 {
421 	PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
422 }
423 
424 static inline void pmd_update_defer(struct mm_struct *mm, unsigned long addr,
425 				    pmd_t *pmdp)
426 {
427 	PVOP_VCALL3(pv_mmu_ops.pmd_update_defer, mm, addr, pmdp);
428 }
429 
430 static inline pte_t __pte(pteval_t val)
431 {
432 	pteval_t ret;
433 
434 	if (sizeof(pteval_t) > sizeof(long))
435 		ret = PVOP_CALLEE2(pteval_t,
436 				   pv_mmu_ops.make_pte,
437 				   val, (u64)val >> 32);
438 	else
439 		ret = PVOP_CALLEE1(pteval_t,
440 				   pv_mmu_ops.make_pte,
441 				   val);
442 
443 	return (pte_t) { .pte = ret };
444 }
445 
446 static inline pteval_t pte_val(pte_t pte)
447 {
448 	pteval_t ret;
449 
450 	if (sizeof(pteval_t) > sizeof(long))
451 		ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
452 				   pte.pte, (u64)pte.pte >> 32);
453 	else
454 		ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
455 				   pte.pte);
456 
457 	return ret;
458 }
459 
460 static inline pgd_t __pgd(pgdval_t val)
461 {
462 	pgdval_t ret;
463 
464 	if (sizeof(pgdval_t) > sizeof(long))
465 		ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
466 				   val, (u64)val >> 32);
467 	else
468 		ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
469 				   val);
470 
471 	return (pgd_t) { ret };
472 }
473 
474 static inline pgdval_t pgd_val(pgd_t pgd)
475 {
476 	pgdval_t ret;
477 
478 	if (sizeof(pgdval_t) > sizeof(long))
479 		ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
480 				    pgd.pgd, (u64)pgd.pgd >> 32);
481 	else
482 		ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
483 				    pgd.pgd);
484 
485 	return ret;
486 }
487 
488 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
489 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
490 					   pte_t *ptep)
491 {
492 	pteval_t ret;
493 
494 	ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
495 			 mm, addr, ptep);
496 
497 	return (pte_t) { .pte = ret };
498 }
499 
500 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
501 					   pte_t *ptep, pte_t pte)
502 {
503 	if (sizeof(pteval_t) > sizeof(long))
504 		/* 5 arg words */
505 		pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
506 	else
507 		PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
508 			    mm, addr, ptep, pte.pte);
509 }
510 
511 static inline void set_pte(pte_t *ptep, pte_t pte)
512 {
513 	if (sizeof(pteval_t) > sizeof(long))
514 		PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
515 			    pte.pte, (u64)pte.pte >> 32);
516 	else
517 		PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
518 			    pte.pte);
519 }
520 
521 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
522 			      pte_t *ptep, pte_t pte)
523 {
524 	if (sizeof(pteval_t) > sizeof(long))
525 		/* 5 arg words */
526 		pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
527 	else
528 		PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
529 }
530 
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 
542 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
543 {
544 	pmdval_t val = native_pmd_val(pmd);
545 
546 	if (sizeof(pmdval_t) > sizeof(long))
547 		PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
548 	else
549 		PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
550 }
551 
552 #if PAGETABLE_LEVELS >= 3
553 static inline pmd_t __pmd(pmdval_t val)
554 {
555 	pmdval_t ret;
556 
557 	if (sizeof(pmdval_t) > sizeof(long))
558 		ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
559 				   val, (u64)val >> 32);
560 	else
561 		ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
562 				   val);
563 
564 	return (pmd_t) { ret };
565 }
566 
567 static inline pmdval_t pmd_val(pmd_t pmd)
568 {
569 	pmdval_t ret;
570 
571 	if (sizeof(pmdval_t) > sizeof(long))
572 		ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
573 				    pmd.pmd, (u64)pmd.pmd >> 32);
574 	else
575 		ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
576 				    pmd.pmd);
577 
578 	return ret;
579 }
580 
581 static inline void set_pud(pud_t *pudp, pud_t pud)
582 {
583 	pudval_t val = native_pud_val(pud);
584 
585 	if (sizeof(pudval_t) > sizeof(long))
586 		PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
587 			    val, (u64)val >> 32);
588 	else
589 		PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
590 			    val);
591 }
592 #if PAGETABLE_LEVELS == 4
593 static inline pud_t __pud(pudval_t val)
594 {
595 	pudval_t ret;
596 
597 	if (sizeof(pudval_t) > sizeof(long))
598 		ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
599 				   val, (u64)val >> 32);
600 	else
601 		ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
602 				   val);
603 
604 	return (pud_t) { ret };
605 }
606 
607 static inline pudval_t pud_val(pud_t pud)
608 {
609 	pudval_t ret;
610 
611 	if (sizeof(pudval_t) > sizeof(long))
612 		ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
613 				    pud.pud, (u64)pud.pud >> 32);
614 	else
615 		ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
616 				    pud.pud);
617 
618 	return ret;
619 }
620 
621 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
622 {
623 	pgdval_t val = native_pgd_val(pgd);
624 
625 	if (sizeof(pgdval_t) > sizeof(long))
626 		PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
627 			    val, (u64)val >> 32);
628 	else
629 		PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
630 			    val);
631 }
632 
633 static inline void pgd_clear(pgd_t *pgdp)
634 {
635 	set_pgd(pgdp, __pgd(0));
636 }
637 
638 static inline void pud_clear(pud_t *pudp)
639 {
640 	set_pud(pudp, __pud(0));
641 }
642 
643 #endif	/* PAGETABLE_LEVELS == 4 */
644 
645 #endif	/* PAGETABLE_LEVELS >= 3 */
646 
647 #ifdef CONFIG_X86_PAE
648 /* Special-case pte-setting operations for PAE, which can't update a
649    64-bit pte atomically */
650 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
651 {
652 	PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
653 		    pte.pte, pte.pte >> 32);
654 }
655 
656 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
657 			     pte_t *ptep)
658 {
659 	PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
660 }
661 
662 static inline void pmd_clear(pmd_t *pmdp)
663 {
664 	PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
665 }
666 #else  /* !CONFIG_X86_PAE */
667 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
668 {
669 	set_pte(ptep, pte);
670 }
671 
672 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
673 			     pte_t *ptep)
674 {
675 	set_pte_at(mm, addr, ptep, __pte(0));
676 }
677 
678 static inline void pmd_clear(pmd_t *pmdp)
679 {
680 	set_pmd(pmdp, __pmd(0));
681 }
682 #endif	/* CONFIG_X86_PAE */
683 
684 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
685 static inline void arch_start_context_switch(struct task_struct *prev)
686 {
687 	PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
688 }
689 
690 static inline void arch_end_context_switch(struct task_struct *next)
691 {
692 	PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
693 }
694 
695 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
696 static inline void arch_enter_lazy_mmu_mode(void)
697 {
698 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
699 }
700 
701 static inline void arch_leave_lazy_mmu_mode(void)
702 {
703 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
704 }
705 
706 void arch_flush_lazy_mmu_mode(void);
707 
708 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
709 				phys_addr_t phys, pgprot_t flags)
710 {
711 	pv_mmu_ops.set_fixmap(idx, phys, flags);
712 }
713 
714 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
715 
716 static inline int arch_spin_is_locked(struct arch_spinlock *lock)
717 {
718 	return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
719 }
720 
721 static inline int arch_spin_is_contended(struct arch_spinlock *lock)
722 {
723 	return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
724 }
725 #define arch_spin_is_contended	arch_spin_is_contended
726 
727 static __always_inline void arch_spin_lock(struct arch_spinlock *lock)
728 {
729 	PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
730 }
731 
732 static __always_inline void arch_spin_lock_flags(struct arch_spinlock *lock,
733 						  unsigned long flags)
734 {
735 	PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
736 }
737 
738 static __always_inline int arch_spin_trylock(struct arch_spinlock *lock)
739 {
740 	return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
741 }
742 
743 static __always_inline void arch_spin_unlock(struct arch_spinlock *lock)
744 {
745 	PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
746 }
747 
748 #endif
749 
750 #ifdef CONFIG_X86_32
751 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
752 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
753 
754 /* save and restore all caller-save registers, except return value */
755 #define PV_SAVE_ALL_CALLER_REGS		"pushl %ecx;"
756 #define PV_RESTORE_ALL_CALLER_REGS	"popl  %ecx;"
757 
758 #define PV_FLAGS_ARG "0"
759 #define PV_EXTRA_CLOBBERS
760 #define PV_VEXTRA_CLOBBERS
761 #else
762 /* save and restore all caller-save registers, except return value */
763 #define PV_SAVE_ALL_CALLER_REGS						\
764 	"push %rcx;"							\
765 	"push %rdx;"							\
766 	"push %rsi;"							\
767 	"push %rdi;"							\
768 	"push %r8;"							\
769 	"push %r9;"							\
770 	"push %r10;"							\
771 	"push %r11;"
772 #define PV_RESTORE_ALL_CALLER_REGS					\
773 	"pop %r11;"							\
774 	"pop %r10;"							\
775 	"pop %r9;"							\
776 	"pop %r8;"							\
777 	"pop %rdi;"							\
778 	"pop %rsi;"							\
779 	"pop %rdx;"							\
780 	"pop %rcx;"
781 
782 /* We save some registers, but all of them, that's too much. We clobber all
783  * caller saved registers but the argument parameter */
784 #define PV_SAVE_REGS "pushq %%rdi;"
785 #define PV_RESTORE_REGS "popq %%rdi;"
786 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
787 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
788 #define PV_FLAGS_ARG "D"
789 #endif
790 
791 /*
792  * Generate a thunk around a function which saves all caller-save
793  * registers except for the return value.  This allows C functions to
794  * be called from assembler code where fewer than normal registers are
795  * available.  It may also help code generation around calls from C
796  * code if the common case doesn't use many registers.
797  *
798  * When a callee is wrapped in a thunk, the caller can assume that all
799  * arg regs and all scratch registers are preserved across the
800  * call. The return value in rax/eax will not be saved, even for void
801  * functions.
802  */
803 #define PV_CALLEE_SAVE_REGS_THUNK(func)					\
804 	extern typeof(func) __raw_callee_save_##func;			\
805 	static void *__##func##__ __used = func;			\
806 									\
807 	asm(".pushsection .text;"					\
808 	    "__raw_callee_save_" #func ": "				\
809 	    PV_SAVE_ALL_CALLER_REGS					\
810 	    "call " #func ";"						\
811 	    PV_RESTORE_ALL_CALLER_REGS					\
812 	    "ret;"							\
813 	    ".popsection")
814 
815 /* Get a reference to a callee-save function */
816 #define PV_CALLEE_SAVE(func)						\
817 	((struct paravirt_callee_save) { __raw_callee_save_##func })
818 
819 /* Promise that "func" already uses the right calling convention */
820 #define __PV_IS_CALLEE_SAVE(func)			\
821 	((struct paravirt_callee_save) { func })
822 
823 static inline notrace unsigned long arch_local_save_flags(void)
824 {
825 	return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
826 }
827 
828 static inline notrace void arch_local_irq_restore(unsigned long f)
829 {
830 	PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
831 }
832 
833 static inline notrace void arch_local_irq_disable(void)
834 {
835 	PVOP_VCALLEE0(pv_irq_ops.irq_disable);
836 }
837 
838 static inline notrace void arch_local_irq_enable(void)
839 {
840 	PVOP_VCALLEE0(pv_irq_ops.irq_enable);
841 }
842 
843 static inline notrace unsigned long arch_local_irq_save(void)
844 {
845 	unsigned long f;
846 
847 	f = arch_local_save_flags();
848 	arch_local_irq_disable();
849 	return f;
850 }
851 
852 
853 /* Make sure as little as possible of this mess escapes. */
854 #undef PARAVIRT_CALL
855 #undef __PVOP_CALL
856 #undef __PVOP_VCALL
857 #undef PVOP_VCALL0
858 #undef PVOP_CALL0
859 #undef PVOP_VCALL1
860 #undef PVOP_CALL1
861 #undef PVOP_VCALL2
862 #undef PVOP_CALL2
863 #undef PVOP_VCALL3
864 #undef PVOP_CALL3
865 #undef PVOP_VCALL4
866 #undef PVOP_CALL4
867 
868 extern void default_banner(void);
869 
870 #else  /* __ASSEMBLY__ */
871 
872 #define _PVSITE(ptype, clobbers, ops, word, algn)	\
873 771:;						\
874 	ops;					\
875 772:;						\
876 	.pushsection .parainstructions,"a";	\
877 	 .align	algn;				\
878 	 word 771b;				\
879 	 .byte ptype;				\
880 	 .byte 772b-771b;			\
881 	 .short clobbers;			\
882 	.popsection
883 
884 
885 #define COND_PUSH(set, mask, reg)			\
886 	.if ((~(set)) & mask); push %reg; .endif
887 #define COND_POP(set, mask, reg)			\
888 	.if ((~(set)) & mask); pop %reg; .endif
889 
890 #ifdef CONFIG_X86_64
891 
892 #define PV_SAVE_REGS(set)			\
893 	COND_PUSH(set, CLBR_RAX, rax);		\
894 	COND_PUSH(set, CLBR_RCX, rcx);		\
895 	COND_PUSH(set, CLBR_RDX, rdx);		\
896 	COND_PUSH(set, CLBR_RSI, rsi);		\
897 	COND_PUSH(set, CLBR_RDI, rdi);		\
898 	COND_PUSH(set, CLBR_R8, r8);		\
899 	COND_PUSH(set, CLBR_R9, r9);		\
900 	COND_PUSH(set, CLBR_R10, r10);		\
901 	COND_PUSH(set, CLBR_R11, r11)
902 #define PV_RESTORE_REGS(set)			\
903 	COND_POP(set, CLBR_R11, r11);		\
904 	COND_POP(set, CLBR_R10, r10);		\
905 	COND_POP(set, CLBR_R9, r9);		\
906 	COND_POP(set, CLBR_R8, r8);		\
907 	COND_POP(set, CLBR_RDI, rdi);		\
908 	COND_POP(set, CLBR_RSI, rsi);		\
909 	COND_POP(set, CLBR_RDX, rdx);		\
910 	COND_POP(set, CLBR_RCX, rcx);		\
911 	COND_POP(set, CLBR_RAX, rax)
912 
913 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
914 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
915 #define PARA_INDIRECT(addr)	*addr(%rip)
916 #else
917 #define PV_SAVE_REGS(set)			\
918 	COND_PUSH(set, CLBR_EAX, eax);		\
919 	COND_PUSH(set, CLBR_EDI, edi);		\
920 	COND_PUSH(set, CLBR_ECX, ecx);		\
921 	COND_PUSH(set, CLBR_EDX, edx)
922 #define PV_RESTORE_REGS(set)			\
923 	COND_POP(set, CLBR_EDX, edx);		\
924 	COND_POP(set, CLBR_ECX, ecx);		\
925 	COND_POP(set, CLBR_EDI, edi);		\
926 	COND_POP(set, CLBR_EAX, eax)
927 
928 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
929 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
930 #define PARA_INDIRECT(addr)	*%cs:addr
931 #endif
932 
933 #define INTERRUPT_RETURN						\
934 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,	\
935 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
936 
937 #define DISABLE_INTERRUPTS(clobbers)					\
938 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
939 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
940 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);	\
941 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
942 
943 #define ENABLE_INTERRUPTS(clobbers)					\
944 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,	\
945 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
946 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);	\
947 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
948 
949 #define USERGS_SYSRET32							\
950 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),	\
951 		  CLBR_NONE,						\
952 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
953 
954 #ifdef CONFIG_X86_32
955 #define GET_CR0_INTO_EAX				\
956 	push %ecx; push %edx;				\
957 	call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0);	\
958 	pop %edx; pop %ecx
959 
960 #define ENABLE_INTERRUPTS_SYSEXIT					\
961 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),	\
962 		  CLBR_NONE,						\
963 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
964 
965 
966 #else	/* !CONFIG_X86_32 */
967 
968 /*
969  * If swapgs is used while the userspace stack is still current,
970  * there's no way to call a pvop.  The PV replacement *must* be
971  * inlined, or the swapgs instruction must be trapped and emulated.
972  */
973 #define SWAPGS_UNSAFE_STACK						\
974 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
975 		  swapgs)
976 
977 /*
978  * Note: swapgs is very special, and in practise is either going to be
979  * implemented with a single "swapgs" instruction or something very
980  * special.  Either way, we don't need to save any registers for
981  * it.
982  */
983 #define SWAPGS								\
984 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
985 		  call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)		\
986 		 )
987 
988 #define GET_CR2_INTO_RAX				\
989 	call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2)
990 
991 #define PARAVIRT_ADJUST_EXCEPTION_FRAME					\
992 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
993 		  CLBR_NONE,						\
994 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
995 
996 #define USERGS_SYSRET64							\
997 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),	\
998 		  CLBR_NONE,						\
999 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1000 
1001 #define ENABLE_INTERRUPTS_SYSEXIT32					\
1002 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),	\
1003 		  CLBR_NONE,						\
1004 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1005 #endif	/* CONFIG_X86_32 */
1006 
1007 #endif /* __ASSEMBLY__ */
1008 #else  /* CONFIG_PARAVIRT */
1009 # define default_banner x86_init_noop
1010 #endif /* !CONFIG_PARAVIRT */
1011 #endif /* _ASM_X86_PARAVIRT_H */
1012