xref: /openbmc/linux/arch/x86/include/asm/paravirt.h (revision a06c488d)
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 int paravirt_has_feature(unsigned int feature)
23 {
24 	WARN_ON_ONCE(!pv_info.paravirt_enabled);
25 	return (pv_info.features & feature);
26 }
27 
28 static inline void load_sp0(struct tss_struct *tss,
29 			     struct thread_struct *thread)
30 {
31 	PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
32 }
33 
34 /* The paravirtualized CPUID instruction. */
35 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
36 			   unsigned int *ecx, unsigned int *edx)
37 {
38 	PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
39 }
40 
41 /*
42  * These special macros can be used to get or set a debugging register
43  */
44 static inline unsigned long paravirt_get_debugreg(int reg)
45 {
46 	return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
47 }
48 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
49 static inline void set_debugreg(unsigned long val, int reg)
50 {
51 	PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
52 }
53 
54 static inline void clts(void)
55 {
56 	PVOP_VCALL0(pv_cpu_ops.clts);
57 }
58 
59 static inline unsigned long read_cr0(void)
60 {
61 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
62 }
63 
64 static inline void write_cr0(unsigned long x)
65 {
66 	PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
67 }
68 
69 static inline unsigned long read_cr2(void)
70 {
71 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
72 }
73 
74 static inline void write_cr2(unsigned long x)
75 {
76 	PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
77 }
78 
79 static inline unsigned long read_cr3(void)
80 {
81 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
82 }
83 
84 static inline void write_cr3(unsigned long x)
85 {
86 	PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
87 }
88 
89 static inline unsigned long __read_cr4(void)
90 {
91 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
92 }
93 static inline unsigned long __read_cr4_safe(void)
94 {
95 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
96 }
97 
98 static inline void __write_cr4(unsigned long x)
99 {
100 	PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
101 }
102 
103 #ifdef CONFIG_X86_64
104 static inline unsigned long read_cr8(void)
105 {
106 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
107 }
108 
109 static inline void write_cr8(unsigned long x)
110 {
111 	PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
112 }
113 #endif
114 
115 static inline void arch_safe_halt(void)
116 {
117 	PVOP_VCALL0(pv_irq_ops.safe_halt);
118 }
119 
120 static inline void halt(void)
121 {
122 	PVOP_VCALL0(pv_irq_ops.halt);
123 }
124 
125 static inline void wbinvd(void)
126 {
127 	PVOP_VCALL0(pv_cpu_ops.wbinvd);
128 }
129 
130 #define get_kernel_rpl()  (pv_info.kernel_rpl)
131 
132 static inline u64 paravirt_read_msr(unsigned msr, int *err)
133 {
134 	return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
135 }
136 
137 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
138 {
139 	return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
140 }
141 
142 /* These should all do BUG_ON(_err), but our headers are too tangled. */
143 #define rdmsr(msr, val1, val2)			\
144 do {						\
145 	int _err;				\
146 	u64 _l = paravirt_read_msr(msr, &_err);	\
147 	val1 = (u32)_l;				\
148 	val2 = _l >> 32;			\
149 } while (0)
150 
151 #define wrmsr(msr, val1, val2)			\
152 do {						\
153 	paravirt_write_msr(msr, val1, val2);	\
154 } while (0)
155 
156 #define rdmsrl(msr, val)			\
157 do {						\
158 	int _err;				\
159 	val = paravirt_read_msr(msr, &_err);	\
160 } while (0)
161 
162 static inline void wrmsrl(unsigned msr, u64 val)
163 {
164 	wrmsr(msr, (u32)val, (u32)(val>>32));
165 }
166 
167 #define wrmsr_safe(msr, a, b)	paravirt_write_msr(msr, a, b)
168 
169 /* rdmsr with exception handling */
170 #define rdmsr_safe(msr, a, b)			\
171 ({						\
172 	int _err;				\
173 	u64 _l = paravirt_read_msr(msr, &_err);	\
174 	(*a) = (u32)_l;				\
175 	(*b) = _l >> 32;			\
176 	_err;					\
177 })
178 
179 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
180 {
181 	int err;
182 
183 	*p = paravirt_read_msr(msr, &err);
184 	return err;
185 }
186 
187 static inline unsigned long long paravirt_sched_clock(void)
188 {
189 	return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
190 }
191 
192 struct static_key;
193 extern struct static_key paravirt_steal_enabled;
194 extern struct static_key paravirt_steal_rq_enabled;
195 
196 static inline u64 paravirt_steal_clock(int cpu)
197 {
198 	return PVOP_CALL1(u64, pv_time_ops.steal_clock, cpu);
199 }
200 
201 static inline unsigned long long paravirt_read_pmc(int counter)
202 {
203 	return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
204 }
205 
206 #define rdpmc(counter, low, high)		\
207 do {						\
208 	u64 _l = paravirt_read_pmc(counter);	\
209 	low = (u32)_l;				\
210 	high = _l >> 32;			\
211 } while (0)
212 
213 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter))
214 
215 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
216 {
217 	PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
218 }
219 
220 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
221 {
222 	PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
223 }
224 
225 static inline void load_TR_desc(void)
226 {
227 	PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
228 }
229 static inline void load_gdt(const struct desc_ptr *dtr)
230 {
231 	PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
232 }
233 static inline void load_idt(const struct desc_ptr *dtr)
234 {
235 	PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
236 }
237 static inline void set_ldt(const void *addr, unsigned entries)
238 {
239 	PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
240 }
241 static inline void store_idt(struct desc_ptr *dtr)
242 {
243 	PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
244 }
245 static inline unsigned long paravirt_store_tr(void)
246 {
247 	return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
248 }
249 #define store_tr(tr)	((tr) = paravirt_store_tr())
250 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
251 {
252 	PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
253 }
254 
255 #ifdef CONFIG_X86_64
256 static inline void load_gs_index(unsigned int gs)
257 {
258 	PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
259 }
260 #endif
261 
262 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
263 				   const void *desc)
264 {
265 	PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
266 }
267 
268 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
269 				   void *desc, int type)
270 {
271 	PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
272 }
273 
274 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
275 {
276 	PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
277 }
278 static inline void set_iopl_mask(unsigned mask)
279 {
280 	PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
281 }
282 
283 /* The paravirtualized I/O functions */
284 static inline void slow_down_io(void)
285 {
286 	pv_cpu_ops.io_delay();
287 #ifdef REALLY_SLOW_IO
288 	pv_cpu_ops.io_delay();
289 	pv_cpu_ops.io_delay();
290 	pv_cpu_ops.io_delay();
291 #endif
292 }
293 
294 static inline void paravirt_activate_mm(struct mm_struct *prev,
295 					struct mm_struct *next)
296 {
297 	PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
298 }
299 
300 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
301 					  struct mm_struct *mm)
302 {
303 	PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
304 }
305 
306 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
307 {
308 	PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
309 }
310 
311 static inline void __flush_tlb(void)
312 {
313 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
314 }
315 static inline void __flush_tlb_global(void)
316 {
317 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
318 }
319 static inline void __flush_tlb_single(unsigned long addr)
320 {
321 	PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
322 }
323 
324 static inline void flush_tlb_others(const struct cpumask *cpumask,
325 				    struct mm_struct *mm,
326 				    unsigned long start,
327 				    unsigned long end)
328 {
329 	PVOP_VCALL4(pv_mmu_ops.flush_tlb_others, cpumask, mm, start, end);
330 }
331 
332 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
333 {
334 	return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
335 }
336 
337 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
338 {
339 	PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
340 }
341 
342 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
343 {
344 	PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
345 }
346 static inline void paravirt_release_pte(unsigned long pfn)
347 {
348 	PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
349 }
350 
351 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
352 {
353 	PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
354 }
355 
356 static inline void paravirt_release_pmd(unsigned long pfn)
357 {
358 	PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
359 }
360 
361 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
362 {
363 	PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
364 }
365 static inline void paravirt_release_pud(unsigned long pfn)
366 {
367 	PVOP_VCALL1(pv_mmu_ops.release_pud, 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_pmd(pmd_t *pmdp, pmd_t pmd)
489 {
490 	pmdval_t val = native_pmd_val(pmd);
491 
492 	if (sizeof(pmdval_t) > sizeof(long))
493 		PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
494 	else
495 		PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
496 }
497 
498 #if CONFIG_PGTABLE_LEVELS >= 3
499 static inline pmd_t __pmd(pmdval_t val)
500 {
501 	pmdval_t ret;
502 
503 	if (sizeof(pmdval_t) > sizeof(long))
504 		ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
505 				   val, (u64)val >> 32);
506 	else
507 		ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
508 				   val);
509 
510 	return (pmd_t) { ret };
511 }
512 
513 static inline pmdval_t pmd_val(pmd_t pmd)
514 {
515 	pmdval_t ret;
516 
517 	if (sizeof(pmdval_t) > sizeof(long))
518 		ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
519 				    pmd.pmd, (u64)pmd.pmd >> 32);
520 	else
521 		ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
522 				    pmd.pmd);
523 
524 	return ret;
525 }
526 
527 static inline void set_pud(pud_t *pudp, pud_t pud)
528 {
529 	pudval_t val = native_pud_val(pud);
530 
531 	if (sizeof(pudval_t) > sizeof(long))
532 		PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
533 			    val, (u64)val >> 32);
534 	else
535 		PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
536 			    val);
537 }
538 #if CONFIG_PGTABLE_LEVELS == 4
539 static inline pud_t __pud(pudval_t val)
540 {
541 	pudval_t ret;
542 
543 	if (sizeof(pudval_t) > sizeof(long))
544 		ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
545 				   val, (u64)val >> 32);
546 	else
547 		ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
548 				   val);
549 
550 	return (pud_t) { ret };
551 }
552 
553 static inline pudval_t pud_val(pud_t pud)
554 {
555 	pudval_t ret;
556 
557 	if (sizeof(pudval_t) > sizeof(long))
558 		ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
559 				    pud.pud, (u64)pud.pud >> 32);
560 	else
561 		ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
562 				    pud.pud);
563 
564 	return ret;
565 }
566 
567 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
568 {
569 	pgdval_t val = native_pgd_val(pgd);
570 
571 	if (sizeof(pgdval_t) > sizeof(long))
572 		PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
573 			    val, (u64)val >> 32);
574 	else
575 		PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
576 			    val);
577 }
578 
579 static inline void pgd_clear(pgd_t *pgdp)
580 {
581 	set_pgd(pgdp, __pgd(0));
582 }
583 
584 static inline void pud_clear(pud_t *pudp)
585 {
586 	set_pud(pudp, __pud(0));
587 }
588 
589 #endif	/* CONFIG_PGTABLE_LEVELS == 4 */
590 
591 #endif	/* CONFIG_PGTABLE_LEVELS >= 3 */
592 
593 #ifdef CONFIG_X86_PAE
594 /* Special-case pte-setting operations for PAE, which can't update a
595    64-bit pte atomically */
596 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
597 {
598 	PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
599 		    pte.pte, pte.pte >> 32);
600 }
601 
602 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
603 			     pte_t *ptep)
604 {
605 	PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
606 }
607 
608 static inline void pmd_clear(pmd_t *pmdp)
609 {
610 	PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
611 }
612 #else  /* !CONFIG_X86_PAE */
613 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
614 {
615 	set_pte(ptep, pte);
616 }
617 
618 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
619 			     pte_t *ptep)
620 {
621 	set_pte_at(mm, addr, ptep, __pte(0));
622 }
623 
624 static inline void pmd_clear(pmd_t *pmdp)
625 {
626 	set_pmd(pmdp, __pmd(0));
627 }
628 #endif	/* CONFIG_X86_PAE */
629 
630 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
631 static inline void arch_start_context_switch(struct task_struct *prev)
632 {
633 	PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
634 }
635 
636 static inline void arch_end_context_switch(struct task_struct *next)
637 {
638 	PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
639 }
640 
641 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
642 static inline void arch_enter_lazy_mmu_mode(void)
643 {
644 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
645 }
646 
647 static inline void arch_leave_lazy_mmu_mode(void)
648 {
649 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
650 }
651 
652 static inline void arch_flush_lazy_mmu_mode(void)
653 {
654 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.flush);
655 }
656 
657 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
658 				phys_addr_t phys, pgprot_t flags)
659 {
660 	pv_mmu_ops.set_fixmap(idx, phys, flags);
661 }
662 
663 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
664 
665 #ifdef CONFIG_QUEUED_SPINLOCKS
666 
667 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock,
668 							u32 val)
669 {
670 	PVOP_VCALL2(pv_lock_ops.queued_spin_lock_slowpath, lock, val);
671 }
672 
673 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock)
674 {
675 	PVOP_VCALLEE1(pv_lock_ops.queued_spin_unlock, lock);
676 }
677 
678 static __always_inline void pv_wait(u8 *ptr, u8 val)
679 {
680 	PVOP_VCALL2(pv_lock_ops.wait, ptr, val);
681 }
682 
683 static __always_inline void pv_kick(int cpu)
684 {
685 	PVOP_VCALL1(pv_lock_ops.kick, cpu);
686 }
687 
688 #else /* !CONFIG_QUEUED_SPINLOCKS */
689 
690 static __always_inline void __ticket_lock_spinning(struct arch_spinlock *lock,
691 							__ticket_t ticket)
692 {
693 	PVOP_VCALLEE2(pv_lock_ops.lock_spinning, lock, ticket);
694 }
695 
696 static __always_inline void __ticket_unlock_kick(struct arch_spinlock *lock,
697 							__ticket_t ticket)
698 {
699 	PVOP_VCALL2(pv_lock_ops.unlock_kick, lock, ticket);
700 }
701 
702 #endif /* CONFIG_QUEUED_SPINLOCKS */
703 
704 #endif /* SMP && PARAVIRT_SPINLOCKS */
705 
706 #ifdef CONFIG_X86_32
707 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
708 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
709 
710 /* save and restore all caller-save registers, except return value */
711 #define PV_SAVE_ALL_CALLER_REGS		"pushl %ecx;"
712 #define PV_RESTORE_ALL_CALLER_REGS	"popl  %ecx;"
713 
714 #define PV_FLAGS_ARG "0"
715 #define PV_EXTRA_CLOBBERS
716 #define PV_VEXTRA_CLOBBERS
717 #else
718 /* save and restore all caller-save registers, except return value */
719 #define PV_SAVE_ALL_CALLER_REGS						\
720 	"push %rcx;"							\
721 	"push %rdx;"							\
722 	"push %rsi;"							\
723 	"push %rdi;"							\
724 	"push %r8;"							\
725 	"push %r9;"							\
726 	"push %r10;"							\
727 	"push %r11;"
728 #define PV_RESTORE_ALL_CALLER_REGS					\
729 	"pop %r11;"							\
730 	"pop %r10;"							\
731 	"pop %r9;"							\
732 	"pop %r8;"							\
733 	"pop %rdi;"							\
734 	"pop %rsi;"							\
735 	"pop %rdx;"							\
736 	"pop %rcx;"
737 
738 /* We save some registers, but all of them, that's too much. We clobber all
739  * caller saved registers but the argument parameter */
740 #define PV_SAVE_REGS "pushq %%rdi;"
741 #define PV_RESTORE_REGS "popq %%rdi;"
742 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
743 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
744 #define PV_FLAGS_ARG "D"
745 #endif
746 
747 /*
748  * Generate a thunk around a function which saves all caller-save
749  * registers except for the return value.  This allows C functions to
750  * be called from assembler code where fewer than normal registers are
751  * available.  It may also help code generation around calls from C
752  * code if the common case doesn't use many registers.
753  *
754  * When a callee is wrapped in a thunk, the caller can assume that all
755  * arg regs and all scratch registers are preserved across the
756  * call. The return value in rax/eax will not be saved, even for void
757  * functions.
758  */
759 #define PV_CALLEE_SAVE_REGS_THUNK(func)					\
760 	extern typeof(func) __raw_callee_save_##func;			\
761 									\
762 	asm(".pushsection .text;"					\
763 	    ".globl __raw_callee_save_" #func " ; "			\
764 	    "__raw_callee_save_" #func ": "				\
765 	    PV_SAVE_ALL_CALLER_REGS					\
766 	    "call " #func ";"						\
767 	    PV_RESTORE_ALL_CALLER_REGS					\
768 	    "ret;"							\
769 	    ".popsection")
770 
771 /* Get a reference to a callee-save function */
772 #define PV_CALLEE_SAVE(func)						\
773 	((struct paravirt_callee_save) { __raw_callee_save_##func })
774 
775 /* Promise that "func" already uses the right calling convention */
776 #define __PV_IS_CALLEE_SAVE(func)			\
777 	((struct paravirt_callee_save) { func })
778 
779 static inline notrace unsigned long arch_local_save_flags(void)
780 {
781 	return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
782 }
783 
784 static inline notrace void arch_local_irq_restore(unsigned long f)
785 {
786 	PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
787 }
788 
789 static inline notrace void arch_local_irq_disable(void)
790 {
791 	PVOP_VCALLEE0(pv_irq_ops.irq_disable);
792 }
793 
794 static inline notrace void arch_local_irq_enable(void)
795 {
796 	PVOP_VCALLEE0(pv_irq_ops.irq_enable);
797 }
798 
799 static inline notrace unsigned long arch_local_irq_save(void)
800 {
801 	unsigned long f;
802 
803 	f = arch_local_save_flags();
804 	arch_local_irq_disable();
805 	return f;
806 }
807 
808 
809 /* Make sure as little as possible of this mess escapes. */
810 #undef PARAVIRT_CALL
811 #undef __PVOP_CALL
812 #undef __PVOP_VCALL
813 #undef PVOP_VCALL0
814 #undef PVOP_CALL0
815 #undef PVOP_VCALL1
816 #undef PVOP_CALL1
817 #undef PVOP_VCALL2
818 #undef PVOP_CALL2
819 #undef PVOP_VCALL3
820 #undef PVOP_CALL3
821 #undef PVOP_VCALL4
822 #undef PVOP_CALL4
823 
824 extern void default_banner(void);
825 
826 #else  /* __ASSEMBLY__ */
827 
828 #define _PVSITE(ptype, clobbers, ops, word, algn)	\
829 771:;						\
830 	ops;					\
831 772:;						\
832 	.pushsection .parainstructions,"a";	\
833 	 .align	algn;				\
834 	 word 771b;				\
835 	 .byte ptype;				\
836 	 .byte 772b-771b;			\
837 	 .short clobbers;			\
838 	.popsection
839 
840 
841 #define COND_PUSH(set, mask, reg)			\
842 	.if ((~(set)) & mask); push %reg; .endif
843 #define COND_POP(set, mask, reg)			\
844 	.if ((~(set)) & mask); pop %reg; .endif
845 
846 #ifdef CONFIG_X86_64
847 
848 #define PV_SAVE_REGS(set)			\
849 	COND_PUSH(set, CLBR_RAX, rax);		\
850 	COND_PUSH(set, CLBR_RCX, rcx);		\
851 	COND_PUSH(set, CLBR_RDX, rdx);		\
852 	COND_PUSH(set, CLBR_RSI, rsi);		\
853 	COND_PUSH(set, CLBR_RDI, rdi);		\
854 	COND_PUSH(set, CLBR_R8, r8);		\
855 	COND_PUSH(set, CLBR_R9, r9);		\
856 	COND_PUSH(set, CLBR_R10, r10);		\
857 	COND_PUSH(set, CLBR_R11, r11)
858 #define PV_RESTORE_REGS(set)			\
859 	COND_POP(set, CLBR_R11, r11);		\
860 	COND_POP(set, CLBR_R10, r10);		\
861 	COND_POP(set, CLBR_R9, r9);		\
862 	COND_POP(set, CLBR_R8, r8);		\
863 	COND_POP(set, CLBR_RDI, rdi);		\
864 	COND_POP(set, CLBR_RSI, rsi);		\
865 	COND_POP(set, CLBR_RDX, rdx);		\
866 	COND_POP(set, CLBR_RCX, rcx);		\
867 	COND_POP(set, CLBR_RAX, rax)
868 
869 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
870 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
871 #define PARA_INDIRECT(addr)	*addr(%rip)
872 #else
873 #define PV_SAVE_REGS(set)			\
874 	COND_PUSH(set, CLBR_EAX, eax);		\
875 	COND_PUSH(set, CLBR_EDI, edi);		\
876 	COND_PUSH(set, CLBR_ECX, ecx);		\
877 	COND_PUSH(set, CLBR_EDX, edx)
878 #define PV_RESTORE_REGS(set)			\
879 	COND_POP(set, CLBR_EDX, edx);		\
880 	COND_POP(set, CLBR_ECX, ecx);		\
881 	COND_POP(set, CLBR_EDI, edi);		\
882 	COND_POP(set, CLBR_EAX, eax)
883 
884 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
885 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
886 #define PARA_INDIRECT(addr)	*%cs:addr
887 #endif
888 
889 #define INTERRUPT_RETURN						\
890 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,	\
891 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
892 
893 #define DISABLE_INTERRUPTS(clobbers)					\
894 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
895 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
896 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);	\
897 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
898 
899 #define ENABLE_INTERRUPTS(clobbers)					\
900 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,	\
901 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
902 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);	\
903 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
904 
905 #ifdef CONFIG_X86_32
906 #define GET_CR0_INTO_EAX				\
907 	push %ecx; push %edx;				\
908 	call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0);	\
909 	pop %edx; pop %ecx
910 #else	/* !CONFIG_X86_32 */
911 
912 /*
913  * If swapgs is used while the userspace stack is still current,
914  * there's no way to call a pvop.  The PV replacement *must* be
915  * inlined, or the swapgs instruction must be trapped and emulated.
916  */
917 #define SWAPGS_UNSAFE_STACK						\
918 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
919 		  swapgs)
920 
921 /*
922  * Note: swapgs is very special, and in practise is either going to be
923  * implemented with a single "swapgs" instruction or something very
924  * special.  Either way, we don't need to save any registers for
925  * it.
926  */
927 #define SWAPGS								\
928 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
929 		  call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)		\
930 		 )
931 
932 #define GET_CR2_INTO_RAX				\
933 	call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2)
934 
935 #define PARAVIRT_ADJUST_EXCEPTION_FRAME					\
936 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
937 		  CLBR_NONE,						\
938 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
939 
940 #define USERGS_SYSRET64							\
941 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),	\
942 		  CLBR_NONE,						\
943 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
944 #endif	/* CONFIG_X86_32 */
945 
946 #endif /* __ASSEMBLY__ */
947 #else  /* CONFIG_PARAVIRT */
948 # define default_banner x86_init_noop
949 #ifndef __ASSEMBLY__
950 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
951 					  struct mm_struct *mm)
952 {
953 }
954 
955 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
956 {
957 }
958 #endif /* __ASSEMBLY__ */
959 #endif /* !CONFIG_PARAVIRT */
960 #endif /* _ASM_X86_PARAVIRT_H */
961