xref: /openbmc/linux/arch/x86/include/asm/paravirt.h (revision 4f3db074)
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_idt(struct desc_ptr *dtr)
266 {
267 	PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
268 }
269 static inline unsigned long paravirt_store_tr(void)
270 {
271 	return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
272 }
273 #define store_tr(tr)	((tr) = paravirt_store_tr())
274 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
275 {
276 	PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
277 }
278 
279 #ifdef CONFIG_X86_64
280 static inline void load_gs_index(unsigned int gs)
281 {
282 	PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
283 }
284 #endif
285 
286 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
287 				   const void *desc)
288 {
289 	PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
290 }
291 
292 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
293 				   void *desc, int type)
294 {
295 	PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
296 }
297 
298 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
299 {
300 	PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
301 }
302 static inline void set_iopl_mask(unsigned mask)
303 {
304 	PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
305 }
306 
307 /* The paravirtualized I/O functions */
308 static inline void slow_down_io(void)
309 {
310 	pv_cpu_ops.io_delay();
311 #ifdef REALLY_SLOW_IO
312 	pv_cpu_ops.io_delay();
313 	pv_cpu_ops.io_delay();
314 	pv_cpu_ops.io_delay();
315 #endif
316 }
317 
318 #ifdef CONFIG_SMP
319 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
320 				    unsigned long start_esp)
321 {
322 	PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
323 		    phys_apicid, start_eip, start_esp);
324 }
325 #endif
326 
327 static inline void paravirt_activate_mm(struct mm_struct *prev,
328 					struct mm_struct *next)
329 {
330 	PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
331 }
332 
333 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
334 					  struct mm_struct *mm)
335 {
336 	PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
337 }
338 
339 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
340 {
341 	PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
342 }
343 
344 static inline void __flush_tlb(void)
345 {
346 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
347 }
348 static inline void __flush_tlb_global(void)
349 {
350 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
351 }
352 static inline void __flush_tlb_single(unsigned long addr)
353 {
354 	PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
355 }
356 
357 static inline void flush_tlb_others(const struct cpumask *cpumask,
358 				    struct mm_struct *mm,
359 				    unsigned long start,
360 				    unsigned long end)
361 {
362 	PVOP_VCALL4(pv_mmu_ops.flush_tlb_others, cpumask, mm, start, end);
363 }
364 
365 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
366 {
367 	return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
368 }
369 
370 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
371 {
372 	PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
373 }
374 
375 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
376 {
377 	PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
378 }
379 static inline void paravirt_release_pte(unsigned long pfn)
380 {
381 	PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
382 }
383 
384 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
385 {
386 	PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
387 }
388 
389 static inline void paravirt_release_pmd(unsigned long pfn)
390 {
391 	PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
392 }
393 
394 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
395 {
396 	PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
397 }
398 static inline void paravirt_release_pud(unsigned long pfn)
399 {
400 	PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
401 }
402 
403 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
404 			      pte_t *ptep)
405 {
406 	PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
407 }
408 static inline void pmd_update(struct mm_struct *mm, unsigned long addr,
409 			      pmd_t *pmdp)
410 {
411 	PVOP_VCALL3(pv_mmu_ops.pmd_update, mm, addr, pmdp);
412 }
413 
414 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
415 				    pte_t *ptep)
416 {
417 	PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
418 }
419 
420 static inline void pmd_update_defer(struct mm_struct *mm, unsigned long addr,
421 				    pmd_t *pmdp)
422 {
423 	PVOP_VCALL3(pv_mmu_ops.pmd_update_defer, mm, addr, pmdp);
424 }
425 
426 static inline pte_t __pte(pteval_t val)
427 {
428 	pteval_t ret;
429 
430 	if (sizeof(pteval_t) > sizeof(long))
431 		ret = PVOP_CALLEE2(pteval_t,
432 				   pv_mmu_ops.make_pte,
433 				   val, (u64)val >> 32);
434 	else
435 		ret = PVOP_CALLEE1(pteval_t,
436 				   pv_mmu_ops.make_pte,
437 				   val);
438 
439 	return (pte_t) { .pte = ret };
440 }
441 
442 static inline pteval_t pte_val(pte_t pte)
443 {
444 	pteval_t ret;
445 
446 	if (sizeof(pteval_t) > sizeof(long))
447 		ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
448 				   pte.pte, (u64)pte.pte >> 32);
449 	else
450 		ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
451 				   pte.pte);
452 
453 	return ret;
454 }
455 
456 static inline pgd_t __pgd(pgdval_t val)
457 {
458 	pgdval_t ret;
459 
460 	if (sizeof(pgdval_t) > sizeof(long))
461 		ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
462 				   val, (u64)val >> 32);
463 	else
464 		ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
465 				   val);
466 
467 	return (pgd_t) { ret };
468 }
469 
470 static inline pgdval_t pgd_val(pgd_t pgd)
471 {
472 	pgdval_t ret;
473 
474 	if (sizeof(pgdval_t) > sizeof(long))
475 		ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
476 				    pgd.pgd, (u64)pgd.pgd >> 32);
477 	else
478 		ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
479 				    pgd.pgd);
480 
481 	return ret;
482 }
483 
484 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
485 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
486 					   pte_t *ptep)
487 {
488 	pteval_t ret;
489 
490 	ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
491 			 mm, addr, ptep);
492 
493 	return (pte_t) { .pte = ret };
494 }
495 
496 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
497 					   pte_t *ptep, pte_t pte)
498 {
499 	if (sizeof(pteval_t) > sizeof(long))
500 		/* 5 arg words */
501 		pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
502 	else
503 		PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
504 			    mm, addr, ptep, pte.pte);
505 }
506 
507 static inline void set_pte(pte_t *ptep, pte_t pte)
508 {
509 	if (sizeof(pteval_t) > sizeof(long))
510 		PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
511 			    pte.pte, (u64)pte.pte >> 32);
512 	else
513 		PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
514 			    pte.pte);
515 }
516 
517 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
518 			      pte_t *ptep, pte_t pte)
519 {
520 	if (sizeof(pteval_t) > sizeof(long))
521 		/* 5 arg words */
522 		pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
523 	else
524 		PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
525 }
526 
527 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
528 			      pmd_t *pmdp, pmd_t pmd)
529 {
530 	if (sizeof(pmdval_t) > sizeof(long))
531 		/* 5 arg words */
532 		pv_mmu_ops.set_pmd_at(mm, addr, pmdp, pmd);
533 	else
534 		PVOP_VCALL4(pv_mmu_ops.set_pmd_at, mm, addr, pmdp,
535 			    native_pmd_val(pmd));
536 }
537 
538 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
539 {
540 	pmdval_t val = native_pmd_val(pmd);
541 
542 	if (sizeof(pmdval_t) > sizeof(long))
543 		PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
544 	else
545 		PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
546 }
547 
548 #if CONFIG_PGTABLE_LEVELS >= 3
549 static inline pmd_t __pmd(pmdval_t val)
550 {
551 	pmdval_t ret;
552 
553 	if (sizeof(pmdval_t) > sizeof(long))
554 		ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
555 				   val, (u64)val >> 32);
556 	else
557 		ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
558 				   val);
559 
560 	return (pmd_t) { ret };
561 }
562 
563 static inline pmdval_t pmd_val(pmd_t pmd)
564 {
565 	pmdval_t ret;
566 
567 	if (sizeof(pmdval_t) > sizeof(long))
568 		ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
569 				    pmd.pmd, (u64)pmd.pmd >> 32);
570 	else
571 		ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
572 				    pmd.pmd);
573 
574 	return ret;
575 }
576 
577 static inline void set_pud(pud_t *pudp, pud_t pud)
578 {
579 	pudval_t val = native_pud_val(pud);
580 
581 	if (sizeof(pudval_t) > sizeof(long))
582 		PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
583 			    val, (u64)val >> 32);
584 	else
585 		PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
586 			    val);
587 }
588 #if CONFIG_PGTABLE_LEVELS == 4
589 static inline pud_t __pud(pudval_t val)
590 {
591 	pudval_t ret;
592 
593 	if (sizeof(pudval_t) > sizeof(long))
594 		ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
595 				   val, (u64)val >> 32);
596 	else
597 		ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
598 				   val);
599 
600 	return (pud_t) { ret };
601 }
602 
603 static inline pudval_t pud_val(pud_t pud)
604 {
605 	pudval_t ret;
606 
607 	if (sizeof(pudval_t) > sizeof(long))
608 		ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
609 				    pud.pud, (u64)pud.pud >> 32);
610 	else
611 		ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
612 				    pud.pud);
613 
614 	return ret;
615 }
616 
617 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
618 {
619 	pgdval_t val = native_pgd_val(pgd);
620 
621 	if (sizeof(pgdval_t) > sizeof(long))
622 		PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
623 			    val, (u64)val >> 32);
624 	else
625 		PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
626 			    val);
627 }
628 
629 static inline void pgd_clear(pgd_t *pgdp)
630 {
631 	set_pgd(pgdp, __pgd(0));
632 }
633 
634 static inline void pud_clear(pud_t *pudp)
635 {
636 	set_pud(pudp, __pud(0));
637 }
638 
639 #endif	/* CONFIG_PGTABLE_LEVELS == 4 */
640 
641 #endif	/* CONFIG_PGTABLE_LEVELS >= 3 */
642 
643 #ifdef CONFIG_X86_PAE
644 /* Special-case pte-setting operations for PAE, which can't update a
645    64-bit pte atomically */
646 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
647 {
648 	PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
649 		    pte.pte, pte.pte >> 32);
650 }
651 
652 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
653 			     pte_t *ptep)
654 {
655 	PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
656 }
657 
658 static inline void pmd_clear(pmd_t *pmdp)
659 {
660 	PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
661 }
662 #else  /* !CONFIG_X86_PAE */
663 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
664 {
665 	set_pte(ptep, pte);
666 }
667 
668 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
669 			     pte_t *ptep)
670 {
671 	set_pte_at(mm, addr, ptep, __pte(0));
672 }
673 
674 static inline void pmd_clear(pmd_t *pmdp)
675 {
676 	set_pmd(pmdp, __pmd(0));
677 }
678 #endif	/* CONFIG_X86_PAE */
679 
680 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
681 static inline void arch_start_context_switch(struct task_struct *prev)
682 {
683 	PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
684 }
685 
686 static inline void arch_end_context_switch(struct task_struct *next)
687 {
688 	PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
689 }
690 
691 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
692 static inline void arch_enter_lazy_mmu_mode(void)
693 {
694 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
695 }
696 
697 static inline void arch_leave_lazy_mmu_mode(void)
698 {
699 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
700 }
701 
702 static inline void arch_flush_lazy_mmu_mode(void)
703 {
704 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.flush);
705 }
706 
707 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
708 				phys_addr_t phys, pgprot_t flags)
709 {
710 	pv_mmu_ops.set_fixmap(idx, phys, flags);
711 }
712 
713 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
714 
715 static __always_inline void __ticket_lock_spinning(struct arch_spinlock *lock,
716 							__ticket_t ticket)
717 {
718 	PVOP_VCALLEE2(pv_lock_ops.lock_spinning, lock, ticket);
719 }
720 
721 static __always_inline void __ticket_unlock_kick(struct arch_spinlock *lock,
722 							__ticket_t ticket)
723 {
724 	PVOP_VCALL2(pv_lock_ops.unlock_kick, lock, ticket);
725 }
726 
727 #endif
728 
729 #ifdef CONFIG_X86_32
730 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
731 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
732 
733 /* save and restore all caller-save registers, except return value */
734 #define PV_SAVE_ALL_CALLER_REGS		"pushl %ecx;"
735 #define PV_RESTORE_ALL_CALLER_REGS	"popl  %ecx;"
736 
737 #define PV_FLAGS_ARG "0"
738 #define PV_EXTRA_CLOBBERS
739 #define PV_VEXTRA_CLOBBERS
740 #else
741 /* save and restore all caller-save registers, except return value */
742 #define PV_SAVE_ALL_CALLER_REGS						\
743 	"push %rcx;"							\
744 	"push %rdx;"							\
745 	"push %rsi;"							\
746 	"push %rdi;"							\
747 	"push %r8;"							\
748 	"push %r9;"							\
749 	"push %r10;"							\
750 	"push %r11;"
751 #define PV_RESTORE_ALL_CALLER_REGS					\
752 	"pop %r11;"							\
753 	"pop %r10;"							\
754 	"pop %r9;"							\
755 	"pop %r8;"							\
756 	"pop %rdi;"							\
757 	"pop %rsi;"							\
758 	"pop %rdx;"							\
759 	"pop %rcx;"
760 
761 /* We save some registers, but all of them, that's too much. We clobber all
762  * caller saved registers but the argument parameter */
763 #define PV_SAVE_REGS "pushq %%rdi;"
764 #define PV_RESTORE_REGS "popq %%rdi;"
765 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
766 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
767 #define PV_FLAGS_ARG "D"
768 #endif
769 
770 /*
771  * Generate a thunk around a function which saves all caller-save
772  * registers except for the return value.  This allows C functions to
773  * be called from assembler code where fewer than normal registers are
774  * available.  It may also help code generation around calls from C
775  * code if the common case doesn't use many registers.
776  *
777  * When a callee is wrapped in a thunk, the caller can assume that all
778  * arg regs and all scratch registers are preserved across the
779  * call. The return value in rax/eax will not be saved, even for void
780  * functions.
781  */
782 #define PV_CALLEE_SAVE_REGS_THUNK(func)					\
783 	extern typeof(func) __raw_callee_save_##func;			\
784 									\
785 	asm(".pushsection .text;"					\
786 	    ".globl __raw_callee_save_" #func " ; "			\
787 	    "__raw_callee_save_" #func ": "				\
788 	    PV_SAVE_ALL_CALLER_REGS					\
789 	    "call " #func ";"						\
790 	    PV_RESTORE_ALL_CALLER_REGS					\
791 	    "ret;"							\
792 	    ".popsection")
793 
794 /* Get a reference to a callee-save function */
795 #define PV_CALLEE_SAVE(func)						\
796 	((struct paravirt_callee_save) { __raw_callee_save_##func })
797 
798 /* Promise that "func" already uses the right calling convention */
799 #define __PV_IS_CALLEE_SAVE(func)			\
800 	((struct paravirt_callee_save) { func })
801 
802 static inline notrace unsigned long arch_local_save_flags(void)
803 {
804 	return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
805 }
806 
807 static inline notrace void arch_local_irq_restore(unsigned long f)
808 {
809 	PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
810 }
811 
812 static inline notrace void arch_local_irq_disable(void)
813 {
814 	PVOP_VCALLEE0(pv_irq_ops.irq_disable);
815 }
816 
817 static inline notrace void arch_local_irq_enable(void)
818 {
819 	PVOP_VCALLEE0(pv_irq_ops.irq_enable);
820 }
821 
822 static inline notrace unsigned long arch_local_irq_save(void)
823 {
824 	unsigned long f;
825 
826 	f = arch_local_save_flags();
827 	arch_local_irq_disable();
828 	return f;
829 }
830 
831 
832 /* Make sure as little as possible of this mess escapes. */
833 #undef PARAVIRT_CALL
834 #undef __PVOP_CALL
835 #undef __PVOP_VCALL
836 #undef PVOP_VCALL0
837 #undef PVOP_CALL0
838 #undef PVOP_VCALL1
839 #undef PVOP_CALL1
840 #undef PVOP_VCALL2
841 #undef PVOP_CALL2
842 #undef PVOP_VCALL3
843 #undef PVOP_CALL3
844 #undef PVOP_VCALL4
845 #undef PVOP_CALL4
846 
847 extern void default_banner(void);
848 
849 #else  /* __ASSEMBLY__ */
850 
851 #define _PVSITE(ptype, clobbers, ops, word, algn)	\
852 771:;						\
853 	ops;					\
854 772:;						\
855 	.pushsection .parainstructions,"a";	\
856 	 .align	algn;				\
857 	 word 771b;				\
858 	 .byte ptype;				\
859 	 .byte 772b-771b;			\
860 	 .short clobbers;			\
861 	.popsection
862 
863 
864 #define COND_PUSH(set, mask, reg)			\
865 	.if ((~(set)) & mask); push %reg; .endif
866 #define COND_POP(set, mask, reg)			\
867 	.if ((~(set)) & mask); pop %reg; .endif
868 
869 #ifdef CONFIG_X86_64
870 
871 #define PV_SAVE_REGS(set)			\
872 	COND_PUSH(set, CLBR_RAX, rax);		\
873 	COND_PUSH(set, CLBR_RCX, rcx);		\
874 	COND_PUSH(set, CLBR_RDX, rdx);		\
875 	COND_PUSH(set, CLBR_RSI, rsi);		\
876 	COND_PUSH(set, CLBR_RDI, rdi);		\
877 	COND_PUSH(set, CLBR_R8, r8);		\
878 	COND_PUSH(set, CLBR_R9, r9);		\
879 	COND_PUSH(set, CLBR_R10, r10);		\
880 	COND_PUSH(set, CLBR_R11, r11)
881 #define PV_RESTORE_REGS(set)			\
882 	COND_POP(set, CLBR_R11, r11);		\
883 	COND_POP(set, CLBR_R10, r10);		\
884 	COND_POP(set, CLBR_R9, r9);		\
885 	COND_POP(set, CLBR_R8, r8);		\
886 	COND_POP(set, CLBR_RDI, rdi);		\
887 	COND_POP(set, CLBR_RSI, rsi);		\
888 	COND_POP(set, CLBR_RDX, rdx);		\
889 	COND_POP(set, CLBR_RCX, rcx);		\
890 	COND_POP(set, CLBR_RAX, rax)
891 
892 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
893 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
894 #define PARA_INDIRECT(addr)	*addr(%rip)
895 #else
896 #define PV_SAVE_REGS(set)			\
897 	COND_PUSH(set, CLBR_EAX, eax);		\
898 	COND_PUSH(set, CLBR_EDI, edi);		\
899 	COND_PUSH(set, CLBR_ECX, ecx);		\
900 	COND_PUSH(set, CLBR_EDX, edx)
901 #define PV_RESTORE_REGS(set)			\
902 	COND_POP(set, CLBR_EDX, edx);		\
903 	COND_POP(set, CLBR_ECX, ecx);		\
904 	COND_POP(set, CLBR_EDI, edi);		\
905 	COND_POP(set, CLBR_EAX, eax)
906 
907 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
908 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
909 #define PARA_INDIRECT(addr)	*%cs:addr
910 #endif
911 
912 #define INTERRUPT_RETURN						\
913 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,	\
914 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
915 
916 #define DISABLE_INTERRUPTS(clobbers)					\
917 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
918 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
919 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);	\
920 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
921 
922 #define ENABLE_INTERRUPTS(clobbers)					\
923 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,	\
924 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
925 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);	\
926 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
927 
928 #define USERGS_SYSRET32							\
929 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),	\
930 		  CLBR_NONE,						\
931 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
932 
933 #ifdef CONFIG_X86_32
934 #define GET_CR0_INTO_EAX				\
935 	push %ecx; push %edx;				\
936 	call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0);	\
937 	pop %edx; pop %ecx
938 
939 #define ENABLE_INTERRUPTS_SYSEXIT					\
940 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),	\
941 		  CLBR_NONE,						\
942 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
943 
944 
945 #else	/* !CONFIG_X86_32 */
946 
947 /*
948  * If swapgs is used while the userspace stack is still current,
949  * there's no way to call a pvop.  The PV replacement *must* be
950  * inlined, or the swapgs instruction must be trapped and emulated.
951  */
952 #define SWAPGS_UNSAFE_STACK						\
953 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
954 		  swapgs)
955 
956 /*
957  * Note: swapgs is very special, and in practise is either going to be
958  * implemented with a single "swapgs" instruction or something very
959  * special.  Either way, we don't need to save any registers for
960  * it.
961  */
962 #define SWAPGS								\
963 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
964 		  call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)		\
965 		 )
966 
967 #define GET_CR2_INTO_RAX				\
968 	call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2)
969 
970 #define PARAVIRT_ADJUST_EXCEPTION_FRAME					\
971 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
972 		  CLBR_NONE,						\
973 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
974 
975 #define USERGS_SYSRET64							\
976 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),	\
977 		  CLBR_NONE,						\
978 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
979 #endif	/* CONFIG_X86_32 */
980 
981 #endif /* __ASSEMBLY__ */
982 #else  /* CONFIG_PARAVIRT */
983 # define default_banner x86_init_noop
984 #ifndef __ASSEMBLY__
985 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
986 					  struct mm_struct *mm)
987 {
988 }
989 
990 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
991 {
992 }
993 #endif /* __ASSEMBLY__ */
994 #endif /* !CONFIG_PARAVIRT */
995 #endif /* _ASM_X86_PARAVIRT_H */
996