xref: /openbmc/linux/arch/x86/include/asm/paravirt.h (revision a36954f5)
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 #include <asm/frame.h>
17 
18 static inline void load_sp0(struct tss_struct *tss,
19 			     struct thread_struct *thread)
20 {
21 	PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
22 }
23 
24 /* The paravirtualized CPUID instruction. */
25 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
26 			   unsigned int *ecx, unsigned int *edx)
27 {
28 	PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
29 }
30 
31 /*
32  * These special macros can be used to get or set a debugging register
33  */
34 static inline unsigned long paravirt_get_debugreg(int reg)
35 {
36 	return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
37 }
38 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
39 static inline void set_debugreg(unsigned long val, int reg)
40 {
41 	PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
42 }
43 
44 static inline unsigned long read_cr0(void)
45 {
46 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
47 }
48 
49 static inline void write_cr0(unsigned long x)
50 {
51 	PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
52 }
53 
54 static inline unsigned long read_cr2(void)
55 {
56 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
57 }
58 
59 static inline void write_cr2(unsigned long x)
60 {
61 	PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
62 }
63 
64 static inline unsigned long read_cr3(void)
65 {
66 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
67 }
68 
69 static inline void write_cr3(unsigned long x)
70 {
71 	PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
72 }
73 
74 static inline unsigned long __read_cr4(void)
75 {
76 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
77 }
78 
79 static inline void __write_cr4(unsigned long x)
80 {
81 	PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
82 }
83 
84 #ifdef CONFIG_X86_64
85 static inline unsigned long read_cr8(void)
86 {
87 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
88 }
89 
90 static inline void write_cr8(unsigned long x)
91 {
92 	PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
93 }
94 #endif
95 
96 static inline void arch_safe_halt(void)
97 {
98 	PVOP_VCALL0(pv_irq_ops.safe_halt);
99 }
100 
101 static inline void halt(void)
102 {
103 	PVOP_VCALL0(pv_irq_ops.halt);
104 }
105 
106 static inline void wbinvd(void)
107 {
108 	PVOP_VCALL0(pv_cpu_ops.wbinvd);
109 }
110 
111 #define get_kernel_rpl()  (pv_info.kernel_rpl)
112 
113 static inline u64 paravirt_read_msr(unsigned msr)
114 {
115 	return PVOP_CALL1(u64, pv_cpu_ops.read_msr, msr);
116 }
117 
118 static inline void paravirt_write_msr(unsigned msr,
119 				      unsigned low, unsigned high)
120 {
121 	return PVOP_VCALL3(pv_cpu_ops.write_msr, msr, low, high);
122 }
123 
124 static inline u64 paravirt_read_msr_safe(unsigned msr, int *err)
125 {
126 	return PVOP_CALL2(u64, pv_cpu_ops.read_msr_safe, msr, err);
127 }
128 
129 static inline int paravirt_write_msr_safe(unsigned msr,
130 					  unsigned low, unsigned high)
131 {
132 	return PVOP_CALL3(int, pv_cpu_ops.write_msr_safe, msr, low, high);
133 }
134 
135 #define rdmsr(msr, val1, val2)			\
136 do {						\
137 	u64 _l = paravirt_read_msr(msr);	\
138 	val1 = (u32)_l;				\
139 	val2 = _l >> 32;			\
140 } while (0)
141 
142 #define wrmsr(msr, val1, val2)			\
143 do {						\
144 	paravirt_write_msr(msr, val1, val2);	\
145 } while (0)
146 
147 #define rdmsrl(msr, val)			\
148 do {						\
149 	val = paravirt_read_msr(msr);		\
150 } while (0)
151 
152 static inline void wrmsrl(unsigned msr, u64 val)
153 {
154 	wrmsr(msr, (u32)val, (u32)(val>>32));
155 }
156 
157 #define wrmsr_safe(msr, a, b)	paravirt_write_msr_safe(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_safe(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_safe(msr, &err);
174 	return err;
175 }
176 
177 static inline unsigned long long paravirt_sched_clock(void)
178 {
179 	return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
180 }
181 
182 struct static_key;
183 extern struct static_key paravirt_steal_enabled;
184 extern struct static_key paravirt_steal_rq_enabled;
185 
186 static inline u64 paravirt_steal_clock(int cpu)
187 {
188 	return PVOP_CALL1(u64, pv_time_ops.steal_clock, cpu);
189 }
190 
191 static inline unsigned long long paravirt_read_pmc(int counter)
192 {
193 	return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
194 }
195 
196 #define rdpmc(counter, low, high)		\
197 do {						\
198 	u64 _l = paravirt_read_pmc(counter);	\
199 	low = (u32)_l;				\
200 	high = _l >> 32;			\
201 } while (0)
202 
203 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter))
204 
205 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
206 {
207 	PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
208 }
209 
210 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
211 {
212 	PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
213 }
214 
215 static inline void load_TR_desc(void)
216 {
217 	PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
218 }
219 static inline void load_gdt(const struct desc_ptr *dtr)
220 {
221 	PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
222 }
223 static inline void load_idt(const struct desc_ptr *dtr)
224 {
225 	PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
226 }
227 static inline void set_ldt(const void *addr, unsigned entries)
228 {
229 	PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
230 }
231 static inline void store_idt(struct desc_ptr *dtr)
232 {
233 	PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
234 }
235 static inline unsigned long paravirt_store_tr(void)
236 {
237 	return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
238 }
239 #define store_tr(tr)	((tr) = paravirt_store_tr())
240 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
241 {
242 	PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
243 }
244 
245 #ifdef CONFIG_X86_64
246 static inline void load_gs_index(unsigned int gs)
247 {
248 	PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
249 }
250 #endif
251 
252 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
253 				   const void *desc)
254 {
255 	PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
256 }
257 
258 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
259 				   void *desc, int type)
260 {
261 	PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
262 }
263 
264 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
265 {
266 	PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
267 }
268 static inline void set_iopl_mask(unsigned mask)
269 {
270 	PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
271 }
272 
273 /* The paravirtualized I/O functions */
274 static inline void slow_down_io(void)
275 {
276 	pv_cpu_ops.io_delay();
277 #ifdef REALLY_SLOW_IO
278 	pv_cpu_ops.io_delay();
279 	pv_cpu_ops.io_delay();
280 	pv_cpu_ops.io_delay();
281 #endif
282 }
283 
284 static inline void paravirt_activate_mm(struct mm_struct *prev,
285 					struct mm_struct *next)
286 {
287 	PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
288 }
289 
290 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
291 					  struct mm_struct *mm)
292 {
293 	PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
294 }
295 
296 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
297 {
298 	PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
299 }
300 
301 static inline void __flush_tlb(void)
302 {
303 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
304 }
305 static inline void __flush_tlb_global(void)
306 {
307 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
308 }
309 static inline void __flush_tlb_single(unsigned long addr)
310 {
311 	PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
312 }
313 
314 static inline void flush_tlb_others(const struct cpumask *cpumask,
315 				    struct mm_struct *mm,
316 				    unsigned long start,
317 				    unsigned long end)
318 {
319 	PVOP_VCALL4(pv_mmu_ops.flush_tlb_others, cpumask, mm, start, end);
320 }
321 
322 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
323 {
324 	return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
325 }
326 
327 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
328 {
329 	PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
330 }
331 
332 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
333 {
334 	PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
335 }
336 static inline void paravirt_release_pte(unsigned long pfn)
337 {
338 	PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
339 }
340 
341 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
342 {
343 	PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
344 }
345 
346 static inline void paravirt_release_pmd(unsigned long pfn)
347 {
348 	PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
349 }
350 
351 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
352 {
353 	PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
354 }
355 static inline void paravirt_release_pud(unsigned long pfn)
356 {
357 	PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
358 }
359 
360 static inline void paravirt_alloc_p4d(struct mm_struct *mm, unsigned long pfn)
361 {
362 	PVOP_VCALL2(pv_mmu_ops.alloc_p4d, mm, pfn);
363 }
364 
365 static inline void paravirt_release_p4d(unsigned long pfn)
366 {
367 	PVOP_VCALL1(pv_mmu_ops.release_p4d, pfn);
368 }
369 
370 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
371 			      pte_t *ptep)
372 {
373 	PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
374 }
375 
376 static inline pte_t __pte(pteval_t val)
377 {
378 	pteval_t ret;
379 
380 	if (sizeof(pteval_t) > sizeof(long))
381 		ret = PVOP_CALLEE2(pteval_t,
382 				   pv_mmu_ops.make_pte,
383 				   val, (u64)val >> 32);
384 	else
385 		ret = PVOP_CALLEE1(pteval_t,
386 				   pv_mmu_ops.make_pte,
387 				   val);
388 
389 	return (pte_t) { .pte = ret };
390 }
391 
392 static inline pteval_t pte_val(pte_t pte)
393 {
394 	pteval_t ret;
395 
396 	if (sizeof(pteval_t) > sizeof(long))
397 		ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
398 				   pte.pte, (u64)pte.pte >> 32);
399 	else
400 		ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
401 				   pte.pte);
402 
403 	return ret;
404 }
405 
406 static inline pgd_t __pgd(pgdval_t val)
407 {
408 	pgdval_t ret;
409 
410 	if (sizeof(pgdval_t) > sizeof(long))
411 		ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
412 				   val, (u64)val >> 32);
413 	else
414 		ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
415 				   val);
416 
417 	return (pgd_t) { ret };
418 }
419 
420 static inline pgdval_t pgd_val(pgd_t pgd)
421 {
422 	pgdval_t ret;
423 
424 	if (sizeof(pgdval_t) > sizeof(long))
425 		ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
426 				    pgd.pgd, (u64)pgd.pgd >> 32);
427 	else
428 		ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
429 				    pgd.pgd);
430 
431 	return ret;
432 }
433 
434 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
435 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
436 					   pte_t *ptep)
437 {
438 	pteval_t ret;
439 
440 	ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
441 			 mm, addr, ptep);
442 
443 	return (pte_t) { .pte = ret };
444 }
445 
446 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
447 					   pte_t *ptep, pte_t pte)
448 {
449 	if (sizeof(pteval_t) > sizeof(long))
450 		/* 5 arg words */
451 		pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
452 	else
453 		PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
454 			    mm, addr, ptep, pte.pte);
455 }
456 
457 static inline void set_pte(pte_t *ptep, pte_t pte)
458 {
459 	if (sizeof(pteval_t) > sizeof(long))
460 		PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
461 			    pte.pte, (u64)pte.pte >> 32);
462 	else
463 		PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
464 			    pte.pte);
465 }
466 
467 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
468 			      pte_t *ptep, pte_t pte)
469 {
470 	if (sizeof(pteval_t) > sizeof(long))
471 		/* 5 arg words */
472 		pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
473 	else
474 		PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
475 }
476 
477 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
478 			      pmd_t *pmdp, pmd_t pmd)
479 {
480 	if (sizeof(pmdval_t) > sizeof(long))
481 		/* 5 arg words */
482 		pv_mmu_ops.set_pmd_at(mm, addr, pmdp, pmd);
483 	else
484 		PVOP_VCALL4(pv_mmu_ops.set_pmd_at, mm, addr, pmdp,
485 			    native_pmd_val(pmd));
486 }
487 
488 static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
489 			      pud_t *pudp, pud_t pud)
490 {
491 	if (sizeof(pudval_t) > sizeof(long))
492 		/* 5 arg words */
493 		pv_mmu_ops.set_pud_at(mm, addr, pudp, pud);
494 	else
495 		PVOP_VCALL4(pv_mmu_ops.set_pud_at, mm, addr, pudp,
496 			    native_pud_val(pud));
497 }
498 
499 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
500 {
501 	pmdval_t val = native_pmd_val(pmd);
502 
503 	if (sizeof(pmdval_t) > sizeof(long))
504 		PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
505 	else
506 		PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
507 }
508 
509 #if CONFIG_PGTABLE_LEVELS >= 3
510 static inline pmd_t __pmd(pmdval_t val)
511 {
512 	pmdval_t ret;
513 
514 	if (sizeof(pmdval_t) > sizeof(long))
515 		ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
516 				   val, (u64)val >> 32);
517 	else
518 		ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
519 				   val);
520 
521 	return (pmd_t) { ret };
522 }
523 
524 static inline pmdval_t pmd_val(pmd_t pmd)
525 {
526 	pmdval_t ret;
527 
528 	if (sizeof(pmdval_t) > sizeof(long))
529 		ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
530 				    pmd.pmd, (u64)pmd.pmd >> 32);
531 	else
532 		ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
533 				    pmd.pmd);
534 
535 	return ret;
536 }
537 
538 static inline void set_pud(pud_t *pudp, pud_t pud)
539 {
540 	pudval_t val = native_pud_val(pud);
541 
542 	if (sizeof(pudval_t) > sizeof(long))
543 		PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
544 			    val, (u64)val >> 32);
545 	else
546 		PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
547 			    val);
548 }
549 #if CONFIG_PGTABLE_LEVELS >= 4
550 static inline pud_t __pud(pudval_t val)
551 {
552 	pudval_t ret;
553 
554 	if (sizeof(pudval_t) > sizeof(long))
555 		ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
556 				   val, (u64)val >> 32);
557 	else
558 		ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
559 				   val);
560 
561 	return (pud_t) { ret };
562 }
563 
564 static inline pudval_t pud_val(pud_t pud)
565 {
566 	pudval_t ret;
567 
568 	if (sizeof(pudval_t) > sizeof(long))
569 		ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
570 				    pud.pud, (u64)pud.pud >> 32);
571 	else
572 		ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
573 				    pud.pud);
574 
575 	return ret;
576 }
577 
578 static inline void pud_clear(pud_t *pudp)
579 {
580 	set_pud(pudp, __pud(0));
581 }
582 
583 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
584 {
585 	p4dval_t val = native_p4d_val(p4d);
586 
587 	if (sizeof(p4dval_t) > sizeof(long))
588 		PVOP_VCALL3(pv_mmu_ops.set_p4d, p4dp,
589 			    val, (u64)val >> 32);
590 	else
591 		PVOP_VCALL2(pv_mmu_ops.set_p4d, p4dp,
592 			    val);
593 }
594 
595 #if CONFIG_PGTABLE_LEVELS >= 5
596 
597 static inline p4d_t __p4d(p4dval_t val)
598 {
599 	p4dval_t ret = PVOP_CALLEE1(p4dval_t, pv_mmu_ops.make_p4d, val);
600 
601 	return (p4d_t) { ret };
602 }
603 
604 static inline p4dval_t p4d_val(p4d_t p4d)
605 {
606 	return PVOP_CALLEE1(p4dval_t, pv_mmu_ops.p4d_val, p4d.p4d);
607 }
608 
609 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
610 {
611 	pgdval_t val = native_pgd_val(pgd);
612 
613 	PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp, val);
614 }
615 
616 static inline void pgd_clear(pgd_t *pgdp)
617 {
618 	set_pgd(pgdp, __pgd(0));
619 }
620 
621 #endif  /* CONFIG_PGTABLE_LEVELS == 5 */
622 
623 static inline void p4d_clear(p4d_t *p4dp)
624 {
625 	set_p4d(p4dp, __p4d(0));
626 }
627 
628 #endif	/* CONFIG_PGTABLE_LEVELS == 4 */
629 
630 #endif	/* CONFIG_PGTABLE_LEVELS >= 3 */
631 
632 #ifdef CONFIG_X86_PAE
633 /* Special-case pte-setting operations for PAE, which can't update a
634    64-bit pte atomically */
635 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
636 {
637 	PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
638 		    pte.pte, pte.pte >> 32);
639 }
640 
641 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
642 			     pte_t *ptep)
643 {
644 	PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
645 }
646 
647 static inline void pmd_clear(pmd_t *pmdp)
648 {
649 	PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
650 }
651 #else  /* !CONFIG_X86_PAE */
652 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
653 {
654 	set_pte(ptep, pte);
655 }
656 
657 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
658 			     pte_t *ptep)
659 {
660 	set_pte_at(mm, addr, ptep, __pte(0));
661 }
662 
663 static inline void pmd_clear(pmd_t *pmdp)
664 {
665 	set_pmd(pmdp, __pmd(0));
666 }
667 #endif	/* CONFIG_X86_PAE */
668 
669 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
670 static inline void arch_start_context_switch(struct task_struct *prev)
671 {
672 	PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
673 }
674 
675 static inline void arch_end_context_switch(struct task_struct *next)
676 {
677 	PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
678 }
679 
680 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
681 static inline void arch_enter_lazy_mmu_mode(void)
682 {
683 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
684 }
685 
686 static inline void arch_leave_lazy_mmu_mode(void)
687 {
688 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
689 }
690 
691 static inline void arch_flush_lazy_mmu_mode(void)
692 {
693 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.flush);
694 }
695 
696 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
697 				phys_addr_t phys, pgprot_t flags)
698 {
699 	pv_mmu_ops.set_fixmap(idx, phys, flags);
700 }
701 
702 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
703 
704 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock,
705 							u32 val)
706 {
707 	PVOP_VCALL2(pv_lock_ops.queued_spin_lock_slowpath, lock, val);
708 }
709 
710 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock)
711 {
712 	PVOP_VCALLEE1(pv_lock_ops.queued_spin_unlock, lock);
713 }
714 
715 static __always_inline void pv_wait(u8 *ptr, u8 val)
716 {
717 	PVOP_VCALL2(pv_lock_ops.wait, ptr, val);
718 }
719 
720 static __always_inline void pv_kick(int cpu)
721 {
722 	PVOP_VCALL1(pv_lock_ops.kick, cpu);
723 }
724 
725 static __always_inline bool pv_vcpu_is_preempted(long cpu)
726 {
727 	return PVOP_CALLEE1(bool, pv_lock_ops.vcpu_is_preempted, cpu);
728 }
729 
730 #endif /* SMP && PARAVIRT_SPINLOCKS */
731 
732 #ifdef CONFIG_X86_32
733 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
734 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
735 
736 /* save and restore all caller-save registers, except return value */
737 #define PV_SAVE_ALL_CALLER_REGS		"pushl %ecx;"
738 #define PV_RESTORE_ALL_CALLER_REGS	"popl  %ecx;"
739 
740 #define PV_FLAGS_ARG "0"
741 #define PV_EXTRA_CLOBBERS
742 #define PV_VEXTRA_CLOBBERS
743 #else
744 /* save and restore all caller-save registers, except return value */
745 #define PV_SAVE_ALL_CALLER_REGS						\
746 	"push %rcx;"							\
747 	"push %rdx;"							\
748 	"push %rsi;"							\
749 	"push %rdi;"							\
750 	"push %r8;"							\
751 	"push %r9;"							\
752 	"push %r10;"							\
753 	"push %r11;"
754 #define PV_RESTORE_ALL_CALLER_REGS					\
755 	"pop %r11;"							\
756 	"pop %r10;"							\
757 	"pop %r9;"							\
758 	"pop %r8;"							\
759 	"pop %rdi;"							\
760 	"pop %rsi;"							\
761 	"pop %rdx;"							\
762 	"pop %rcx;"
763 
764 /* We save some registers, but all of them, that's too much. We clobber all
765  * caller saved registers but the argument parameter */
766 #define PV_SAVE_REGS "pushq %%rdi;"
767 #define PV_RESTORE_REGS "popq %%rdi;"
768 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
769 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
770 #define PV_FLAGS_ARG "D"
771 #endif
772 
773 /*
774  * Generate a thunk around a function which saves all caller-save
775  * registers except for the return value.  This allows C functions to
776  * be called from assembler code where fewer than normal registers are
777  * available.  It may also help code generation around calls from C
778  * code if the common case doesn't use many registers.
779  *
780  * When a callee is wrapped in a thunk, the caller can assume that all
781  * arg regs and all scratch registers are preserved across the
782  * call. The return value in rax/eax will not be saved, even for void
783  * functions.
784  */
785 #define PV_THUNK_NAME(func) "__raw_callee_save_" #func
786 #define PV_CALLEE_SAVE_REGS_THUNK(func)					\
787 	extern typeof(func) __raw_callee_save_##func;			\
788 									\
789 	asm(".pushsection .text;"					\
790 	    ".globl " PV_THUNK_NAME(func) ";"				\
791 	    ".type " PV_THUNK_NAME(func) ", @function;"			\
792 	    PV_THUNK_NAME(func) ":"					\
793 	    FRAME_BEGIN							\
794 	    PV_SAVE_ALL_CALLER_REGS					\
795 	    "call " #func ";"						\
796 	    PV_RESTORE_ALL_CALLER_REGS					\
797 	    FRAME_END							\
798 	    "ret;"							\
799 	    ".popsection")
800 
801 /* Get a reference to a callee-save function */
802 #define PV_CALLEE_SAVE(func)						\
803 	((struct paravirt_callee_save) { __raw_callee_save_##func })
804 
805 /* Promise that "func" already uses the right calling convention */
806 #define __PV_IS_CALLEE_SAVE(func)			\
807 	((struct paravirt_callee_save) { func })
808 
809 static inline notrace unsigned long arch_local_save_flags(void)
810 {
811 	return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
812 }
813 
814 static inline notrace void arch_local_irq_restore(unsigned long f)
815 {
816 	PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
817 }
818 
819 static inline notrace void arch_local_irq_disable(void)
820 {
821 	PVOP_VCALLEE0(pv_irq_ops.irq_disable);
822 }
823 
824 static inline notrace void arch_local_irq_enable(void)
825 {
826 	PVOP_VCALLEE0(pv_irq_ops.irq_enable);
827 }
828 
829 static inline notrace unsigned long arch_local_irq_save(void)
830 {
831 	unsigned long f;
832 
833 	f = arch_local_save_flags();
834 	arch_local_irq_disable();
835 	return f;
836 }
837 
838 
839 /* Make sure as little as possible of this mess escapes. */
840 #undef PARAVIRT_CALL
841 #undef __PVOP_CALL
842 #undef __PVOP_VCALL
843 #undef PVOP_VCALL0
844 #undef PVOP_CALL0
845 #undef PVOP_VCALL1
846 #undef PVOP_CALL1
847 #undef PVOP_VCALL2
848 #undef PVOP_CALL2
849 #undef PVOP_VCALL3
850 #undef PVOP_CALL3
851 #undef PVOP_VCALL4
852 #undef PVOP_CALL4
853 
854 extern void default_banner(void);
855 
856 #else  /* __ASSEMBLY__ */
857 
858 #define _PVSITE(ptype, clobbers, ops, word, algn)	\
859 771:;						\
860 	ops;					\
861 772:;						\
862 	.pushsection .parainstructions,"a";	\
863 	 .align	algn;				\
864 	 word 771b;				\
865 	 .byte ptype;				\
866 	 .byte 772b-771b;			\
867 	 .short clobbers;			\
868 	.popsection
869 
870 
871 #define COND_PUSH(set, mask, reg)			\
872 	.if ((~(set)) & mask); push %reg; .endif
873 #define COND_POP(set, mask, reg)			\
874 	.if ((~(set)) & mask); pop %reg; .endif
875 
876 #ifdef CONFIG_X86_64
877 
878 #define PV_SAVE_REGS(set)			\
879 	COND_PUSH(set, CLBR_RAX, rax);		\
880 	COND_PUSH(set, CLBR_RCX, rcx);		\
881 	COND_PUSH(set, CLBR_RDX, rdx);		\
882 	COND_PUSH(set, CLBR_RSI, rsi);		\
883 	COND_PUSH(set, CLBR_RDI, rdi);		\
884 	COND_PUSH(set, CLBR_R8, r8);		\
885 	COND_PUSH(set, CLBR_R9, r9);		\
886 	COND_PUSH(set, CLBR_R10, r10);		\
887 	COND_PUSH(set, CLBR_R11, r11)
888 #define PV_RESTORE_REGS(set)			\
889 	COND_POP(set, CLBR_R11, r11);		\
890 	COND_POP(set, CLBR_R10, r10);		\
891 	COND_POP(set, CLBR_R9, r9);		\
892 	COND_POP(set, CLBR_R8, r8);		\
893 	COND_POP(set, CLBR_RDI, rdi);		\
894 	COND_POP(set, CLBR_RSI, rsi);		\
895 	COND_POP(set, CLBR_RDX, rdx);		\
896 	COND_POP(set, CLBR_RCX, rcx);		\
897 	COND_POP(set, CLBR_RAX, rax)
898 
899 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
900 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
901 #define PARA_INDIRECT(addr)	*addr(%rip)
902 #else
903 #define PV_SAVE_REGS(set)			\
904 	COND_PUSH(set, CLBR_EAX, eax);		\
905 	COND_PUSH(set, CLBR_EDI, edi);		\
906 	COND_PUSH(set, CLBR_ECX, ecx);		\
907 	COND_PUSH(set, CLBR_EDX, edx)
908 #define PV_RESTORE_REGS(set)			\
909 	COND_POP(set, CLBR_EDX, edx);		\
910 	COND_POP(set, CLBR_ECX, ecx);		\
911 	COND_POP(set, CLBR_EDI, edi);		\
912 	COND_POP(set, CLBR_EAX, eax)
913 
914 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
915 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
916 #define PARA_INDIRECT(addr)	*%cs:addr
917 #endif
918 
919 #define INTERRUPT_RETURN						\
920 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,	\
921 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
922 
923 #define DISABLE_INTERRUPTS(clobbers)					\
924 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
925 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
926 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);	\
927 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
928 
929 #define ENABLE_INTERRUPTS(clobbers)					\
930 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,	\
931 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
932 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);	\
933 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
934 
935 #ifdef CONFIG_X86_32
936 #define GET_CR0_INTO_EAX				\
937 	push %ecx; push %edx;				\
938 	call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0);	\
939 	pop %edx; pop %ecx
940 #else	/* !CONFIG_X86_32 */
941 
942 /*
943  * If swapgs is used while the userspace stack is still current,
944  * there's no way to call a pvop.  The PV replacement *must* be
945  * inlined, or the swapgs instruction must be trapped and emulated.
946  */
947 #define SWAPGS_UNSAFE_STACK						\
948 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
949 		  swapgs)
950 
951 /*
952  * Note: swapgs is very special, and in practise is either going to be
953  * implemented with a single "swapgs" instruction or something very
954  * special.  Either way, we don't need to save any registers for
955  * it.
956  */
957 #define SWAPGS								\
958 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
959 		  call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)		\
960 		 )
961 
962 #define GET_CR2_INTO_RAX				\
963 	call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2)
964 
965 #define PARAVIRT_ADJUST_EXCEPTION_FRAME					\
966 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
967 		  CLBR_NONE,						\
968 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
969 
970 #define USERGS_SYSRET64							\
971 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),	\
972 		  CLBR_NONE,						\
973 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
974 #endif	/* CONFIG_X86_32 */
975 
976 #endif /* __ASSEMBLY__ */
977 #else  /* CONFIG_PARAVIRT */
978 # define default_banner x86_init_noop
979 #ifndef __ASSEMBLY__
980 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
981 					  struct mm_struct *mm)
982 {
983 }
984 
985 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
986 {
987 }
988 #endif /* __ASSEMBLY__ */
989 #endif /* !CONFIG_PARAVIRT */
990 #endif /* _ASM_X86_PARAVIRT_H */
991