• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include <asm/nospec-branch.h>
10 
11 #include <asm/paravirt_types.h>
12 
13 #ifndef __ASSEMBLY__
14 #include <linux/bug.h>
15 #include <linux/types.h>
16 #include <linux/cpumask.h>
17 #include <asm/frame.h>
18 
load_sp0(struct tss_struct * tss,struct thread_struct * thread)19 static inline void load_sp0(struct tss_struct *tss,
20 			     struct thread_struct *thread)
21 {
22 	PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
23 }
24 
25 /* The paravirtualized CPUID instruction. */
__cpuid(unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)26 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
27 			   unsigned int *ecx, unsigned int *edx)
28 {
29 	PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
30 }
31 
32 /*
33  * These special macros can be used to get or set a debugging register
34  */
paravirt_get_debugreg(int reg)35 static inline unsigned long paravirt_get_debugreg(int reg)
36 {
37 	return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
38 }
39 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
set_debugreg(unsigned long val,int reg)40 static inline void set_debugreg(unsigned long val, int reg)
41 {
42 	PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
43 }
44 
clts(void)45 static inline void clts(void)
46 {
47 	PVOP_VCALL0(pv_cpu_ops.clts);
48 }
49 
read_cr0(void)50 static inline unsigned long read_cr0(void)
51 {
52 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
53 }
54 
write_cr0(unsigned long x)55 static inline void write_cr0(unsigned long x)
56 {
57 	PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
58 }
59 
read_cr2(void)60 static inline unsigned long read_cr2(void)
61 {
62 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
63 }
64 
write_cr2(unsigned long x)65 static inline void write_cr2(unsigned long x)
66 {
67 	PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
68 }
69 
read_cr3(void)70 static inline unsigned long read_cr3(void)
71 {
72 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
73 }
74 
write_cr3(unsigned long x)75 static inline void write_cr3(unsigned long x)
76 {
77 	PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
78 }
79 
__read_cr4(void)80 static inline unsigned long __read_cr4(void)
81 {
82 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
83 }
84 
__write_cr4(unsigned long x)85 static inline void __write_cr4(unsigned long x)
86 {
87 	PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
88 }
89 
90 #ifdef CONFIG_X86_64
read_cr8(void)91 static inline unsigned long read_cr8(void)
92 {
93 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
94 }
95 
write_cr8(unsigned long x)96 static inline void write_cr8(unsigned long x)
97 {
98 	PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
99 }
100 #endif
101 
arch_safe_halt(void)102 static inline void arch_safe_halt(void)
103 {
104 	PVOP_VCALL0(pv_irq_ops.safe_halt);
105 }
106 
halt(void)107 static inline void halt(void)
108 {
109 	PVOP_VCALL0(pv_irq_ops.halt);
110 }
111 
wbinvd(void)112 static inline void wbinvd(void)
113 {
114 	PVOP_VCALL0(pv_cpu_ops.wbinvd);
115 }
116 
117 #define get_kernel_rpl()  (pv_info.kernel_rpl)
118 
paravirt_read_msr(unsigned msr)119 static inline u64 paravirt_read_msr(unsigned msr)
120 {
121 	return PVOP_CALL1(u64, pv_cpu_ops.read_msr, msr);
122 }
123 
paravirt_write_msr(unsigned msr,unsigned low,unsigned high)124 static inline void paravirt_write_msr(unsigned msr,
125 				      unsigned low, unsigned high)
126 {
127 	return PVOP_VCALL3(pv_cpu_ops.write_msr, msr, low, high);
128 }
129 
paravirt_read_msr_safe(unsigned msr,int * err)130 static inline u64 paravirt_read_msr_safe(unsigned msr, int *err)
131 {
132 	return PVOP_CALL2(u64, pv_cpu_ops.read_msr_safe, msr, err);
133 }
134 
paravirt_write_msr_safe(unsigned msr,unsigned low,unsigned high)135 static inline int paravirt_write_msr_safe(unsigned msr,
136 					  unsigned low, unsigned high)
137 {
138 	return PVOP_CALL3(int, pv_cpu_ops.write_msr_safe, msr, low, high);
139 }
140 
141 #define rdmsr(msr, val1, val2)			\
142 do {						\
143 	u64 _l = paravirt_read_msr(msr);	\
144 	val1 = (u32)_l;				\
145 	val2 = _l >> 32;			\
146 } while (0)
147 
148 #define wrmsr(msr, val1, val2)			\
149 do {						\
150 	paravirt_write_msr(msr, val1, val2);	\
151 } while (0)
152 
153 #define rdmsrl(msr, val)			\
154 do {						\
155 	val = paravirt_read_msr(msr);		\
156 } while (0)
157 
wrmsrl(unsigned msr,u64 val)158 static inline void wrmsrl(unsigned msr, u64 val)
159 {
160 	wrmsr(msr, (u32)val, (u32)(val>>32));
161 }
162 
163 #define wrmsr_safe(msr, a, b)	paravirt_write_msr_safe(msr, a, b)
164 
165 /* rdmsr with exception handling */
166 #define rdmsr_safe(msr, a, b)				\
167 ({							\
168 	int _err;					\
169 	u64 _l = paravirt_read_msr_safe(msr, &_err);	\
170 	(*a) = (u32)_l;					\
171 	(*b) = _l >> 32;				\
172 	_err;						\
173 })
174 
rdmsrl_safe(unsigned msr,unsigned long long * p)175 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
176 {
177 	int err;
178 
179 	*p = paravirt_read_msr_safe(msr, &err);
180 	return err;
181 }
182 
paravirt_sched_clock(void)183 static inline unsigned long long paravirt_sched_clock(void)
184 {
185 	return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
186 }
187 
188 struct static_key;
189 extern struct static_key paravirt_steal_enabled;
190 extern struct static_key paravirt_steal_rq_enabled;
191 
paravirt_steal_clock(int cpu)192 static inline u64 paravirt_steal_clock(int cpu)
193 {
194 	return PVOP_CALL1(u64, pv_time_ops.steal_clock, cpu);
195 }
196 
paravirt_read_pmc(int counter)197 static inline unsigned long long paravirt_read_pmc(int counter)
198 {
199 	return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
200 }
201 
202 #define rdpmc(counter, low, high)		\
203 do {						\
204 	u64 _l = paravirt_read_pmc(counter);	\
205 	low = (u32)_l;				\
206 	high = _l >> 32;			\
207 } while (0)
208 
209 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter))
210 
paravirt_alloc_ldt(struct desc_struct * ldt,unsigned entries)211 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
212 {
213 	PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
214 }
215 
paravirt_free_ldt(struct desc_struct * ldt,unsigned entries)216 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
217 {
218 	PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
219 }
220 
load_TR_desc(void)221 static inline void load_TR_desc(void)
222 {
223 	PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
224 }
load_gdt(const struct desc_ptr * dtr)225 static inline void load_gdt(const struct desc_ptr *dtr)
226 {
227 	PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
228 }
load_idt(const struct desc_ptr * dtr)229 static inline void load_idt(const struct desc_ptr *dtr)
230 {
231 	PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
232 }
set_ldt(const void * addr,unsigned entries)233 static inline void set_ldt(const void *addr, unsigned entries)
234 {
235 	PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
236 }
store_idt(struct desc_ptr * dtr)237 static inline void store_idt(struct desc_ptr *dtr)
238 {
239 	PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
240 }
paravirt_store_tr(void)241 static inline unsigned long paravirt_store_tr(void)
242 {
243 	return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
244 }
245 #define store_tr(tr)	((tr) = paravirt_store_tr())
load_TLS(struct thread_struct * t,unsigned cpu)246 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
247 {
248 	PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
249 }
250 
251 #ifdef CONFIG_X86_64
load_gs_index(unsigned int gs)252 static inline void load_gs_index(unsigned int gs)
253 {
254 	PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
255 }
256 #endif
257 
write_ldt_entry(struct desc_struct * dt,int entry,const void * desc)258 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
259 				   const void *desc)
260 {
261 	PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
262 }
263 
write_gdt_entry(struct desc_struct * dt,int entry,void * desc,int type)264 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
265 				   void *desc, int type)
266 {
267 	PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
268 }
269 
write_idt_entry(gate_desc * dt,int entry,const gate_desc * g)270 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
271 {
272 	PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
273 }
set_iopl_mask(unsigned mask)274 static inline void set_iopl_mask(unsigned mask)
275 {
276 	PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
277 }
278 
279 /* The paravirtualized I/O functions */
slow_down_io(void)280 static inline void slow_down_io(void)
281 {
282 	pv_cpu_ops.io_delay();
283 #ifdef REALLY_SLOW_IO
284 	pv_cpu_ops.io_delay();
285 	pv_cpu_ops.io_delay();
286 	pv_cpu_ops.io_delay();
287 #endif
288 }
289 
paravirt_activate_mm(struct mm_struct * prev,struct mm_struct * next)290 static inline void paravirt_activate_mm(struct mm_struct *prev,
291 					struct mm_struct *next)
292 {
293 	PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
294 }
295 
paravirt_arch_dup_mmap(struct mm_struct * oldmm,struct mm_struct * mm)296 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
297 					  struct mm_struct *mm)
298 {
299 	PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
300 }
301 
paravirt_arch_exit_mmap(struct mm_struct * mm)302 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
303 {
304 	PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
305 }
306 
__flush_tlb(void)307 static inline void __flush_tlb(void)
308 {
309 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
310 }
__flush_tlb_global(void)311 static inline void __flush_tlb_global(void)
312 {
313 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
314 }
__flush_tlb_single(unsigned long addr)315 static inline void __flush_tlb_single(unsigned long addr)
316 {
317 	PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
318 }
319 
flush_tlb_others(const struct cpumask * cpumask,struct mm_struct * mm,unsigned long start,unsigned long end)320 static inline void flush_tlb_others(const struct cpumask *cpumask,
321 				    struct mm_struct *mm,
322 				    unsigned long start,
323 				    unsigned long end)
324 {
325 	PVOP_VCALL4(pv_mmu_ops.flush_tlb_others, cpumask, mm, start, end);
326 }
327 
paravirt_pgd_alloc(struct mm_struct * mm)328 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
329 {
330 	return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
331 }
332 
paravirt_pgd_free(struct mm_struct * mm,pgd_t * pgd)333 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
334 {
335 	PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
336 }
337 
paravirt_alloc_pte(struct mm_struct * mm,unsigned long pfn)338 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
339 {
340 	PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
341 }
paravirt_release_pte(unsigned long pfn)342 static inline void paravirt_release_pte(unsigned long pfn)
343 {
344 	PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
345 }
346 
paravirt_alloc_pmd(struct mm_struct * mm,unsigned long pfn)347 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
348 {
349 	PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
350 }
351 
paravirt_release_pmd(unsigned long pfn)352 static inline void paravirt_release_pmd(unsigned long pfn)
353 {
354 	PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
355 }
356 
paravirt_alloc_pud(struct mm_struct * mm,unsigned long pfn)357 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
358 {
359 	PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
360 }
paravirt_release_pud(unsigned long pfn)361 static inline void paravirt_release_pud(unsigned long pfn)
362 {
363 	PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
364 }
365 
pte_update(struct mm_struct * mm,unsigned long addr,pte_t * ptep)366 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
367 			      pte_t *ptep)
368 {
369 	PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
370 }
371 
__pte(pteval_t val)372 static inline pte_t __pte(pteval_t val)
373 {
374 	pteval_t ret;
375 
376 	if (sizeof(pteval_t) > sizeof(long))
377 		ret = PVOP_CALLEE2(pteval_t,
378 				   pv_mmu_ops.make_pte,
379 				   val, (u64)val >> 32);
380 	else
381 		ret = PVOP_CALLEE1(pteval_t,
382 				   pv_mmu_ops.make_pte,
383 				   val);
384 
385 	return (pte_t) { .pte = ret };
386 }
387 
pte_val(pte_t pte)388 static inline pteval_t pte_val(pte_t pte)
389 {
390 	pteval_t ret;
391 
392 	if (sizeof(pteval_t) > sizeof(long))
393 		ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
394 				   pte.pte, (u64)pte.pte >> 32);
395 	else
396 		ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
397 				   pte.pte);
398 
399 	return ret;
400 }
401 
__pgd(pgdval_t val)402 static inline pgd_t __pgd(pgdval_t val)
403 {
404 	pgdval_t ret;
405 
406 	if (sizeof(pgdval_t) > sizeof(long))
407 		ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
408 				   val, (u64)val >> 32);
409 	else
410 		ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
411 				   val);
412 
413 	return (pgd_t) { ret };
414 }
415 
pgd_val(pgd_t pgd)416 static inline pgdval_t pgd_val(pgd_t pgd)
417 {
418 	pgdval_t ret;
419 
420 	if (sizeof(pgdval_t) > sizeof(long))
421 		ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
422 				    pgd.pgd, (u64)pgd.pgd >> 32);
423 	else
424 		ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
425 				    pgd.pgd);
426 
427 	return ret;
428 }
429 
430 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
ptep_modify_prot_start(struct mm_struct * mm,unsigned long addr,pte_t * ptep)431 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
432 					   pte_t *ptep)
433 {
434 	pteval_t ret;
435 
436 	ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
437 			 mm, addr, ptep);
438 
439 	return (pte_t) { .pte = ret };
440 }
441 
ptep_modify_prot_commit(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte)442 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
443 					   pte_t *ptep, pte_t pte)
444 {
445 	if (sizeof(pteval_t) > sizeof(long))
446 		/* 5 arg words */
447 		pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
448 	else
449 		PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
450 			    mm, addr, ptep, pte.pte);
451 }
452 
set_pte(pte_t * ptep,pte_t pte)453 static inline void set_pte(pte_t *ptep, pte_t pte)
454 {
455 	if (sizeof(pteval_t) > sizeof(long))
456 		PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
457 			    pte.pte, (u64)pte.pte >> 32);
458 	else
459 		PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
460 			    pte.pte);
461 }
462 
set_pte_at(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte)463 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
464 			      pte_t *ptep, pte_t pte)
465 {
466 	if (sizeof(pteval_t) > sizeof(long))
467 		/* 5 arg words */
468 		pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
469 	else
470 		PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
471 }
472 
set_pmd_at(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp,pmd_t pmd)473 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
474 			      pmd_t *pmdp, pmd_t pmd)
475 {
476 	if (sizeof(pmdval_t) > sizeof(long))
477 		/* 5 arg words */
478 		pv_mmu_ops.set_pmd_at(mm, addr, pmdp, pmd);
479 	else
480 		PVOP_VCALL4(pv_mmu_ops.set_pmd_at, mm, addr, pmdp,
481 			    native_pmd_val(pmd));
482 }
483 
set_pmd(pmd_t * pmdp,pmd_t pmd)484 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
485 {
486 	pmdval_t val = native_pmd_val(pmd);
487 
488 	if (sizeof(pmdval_t) > sizeof(long))
489 		PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
490 	else
491 		PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
492 }
493 
494 #if CONFIG_PGTABLE_LEVELS >= 3
__pmd(pmdval_t val)495 static inline pmd_t __pmd(pmdval_t val)
496 {
497 	pmdval_t ret;
498 
499 	if (sizeof(pmdval_t) > sizeof(long))
500 		ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
501 				   val, (u64)val >> 32);
502 	else
503 		ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
504 				   val);
505 
506 	return (pmd_t) { ret };
507 }
508 
pmd_val(pmd_t pmd)509 static inline pmdval_t pmd_val(pmd_t pmd)
510 {
511 	pmdval_t ret;
512 
513 	if (sizeof(pmdval_t) > sizeof(long))
514 		ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
515 				    pmd.pmd, (u64)pmd.pmd >> 32);
516 	else
517 		ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
518 				    pmd.pmd);
519 
520 	return ret;
521 }
522 
set_pud(pud_t * pudp,pud_t pud)523 static inline void set_pud(pud_t *pudp, pud_t pud)
524 {
525 	pudval_t val = native_pud_val(pud);
526 
527 	if (sizeof(pudval_t) > sizeof(long))
528 		PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
529 			    val, (u64)val >> 32);
530 	else
531 		PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
532 			    val);
533 }
534 #if CONFIG_PGTABLE_LEVELS == 4
__pud(pudval_t val)535 static inline pud_t __pud(pudval_t val)
536 {
537 	pudval_t ret;
538 
539 	if (sizeof(pudval_t) > sizeof(long))
540 		ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
541 				   val, (u64)val >> 32);
542 	else
543 		ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
544 				   val);
545 
546 	return (pud_t) { ret };
547 }
548 
pud_val(pud_t pud)549 static inline pudval_t pud_val(pud_t pud)
550 {
551 	pudval_t ret;
552 
553 	if (sizeof(pudval_t) > sizeof(long))
554 		ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
555 				    pud.pud, (u64)pud.pud >> 32);
556 	else
557 		ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
558 				    pud.pud);
559 
560 	return ret;
561 }
562 
set_pgd(pgd_t * pgdp,pgd_t pgd)563 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
564 {
565 	pgdval_t val = native_pgd_val(pgd);
566 
567 	if (sizeof(pgdval_t) > sizeof(long))
568 		PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
569 			    val, (u64)val >> 32);
570 	else
571 		PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
572 			    val);
573 }
574 
pgd_clear(pgd_t * pgdp)575 static inline void pgd_clear(pgd_t *pgdp)
576 {
577 	set_pgd(pgdp, __pgd(0));
578 }
579 
pud_clear(pud_t * pudp)580 static inline void pud_clear(pud_t *pudp)
581 {
582 	set_pud(pudp, __pud(0));
583 }
584 
585 #endif	/* CONFIG_PGTABLE_LEVELS == 4 */
586 
587 #endif	/* CONFIG_PGTABLE_LEVELS >= 3 */
588 
589 #ifdef CONFIG_X86_PAE
590 /* Special-case pte-setting operations for PAE, which can't update a
591    64-bit pte atomically */
set_pte_atomic(pte_t * ptep,pte_t pte)592 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
593 {
594 	PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
595 		    pte.pte, pte.pte >> 32);
596 }
597 
pte_clear(struct mm_struct * mm,unsigned long addr,pte_t * ptep)598 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
599 			     pte_t *ptep)
600 {
601 	PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
602 }
603 
pmd_clear(pmd_t * pmdp)604 static inline void pmd_clear(pmd_t *pmdp)
605 {
606 	PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
607 }
608 #else  /* !CONFIG_X86_PAE */
set_pte_atomic(pte_t * ptep,pte_t pte)609 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
610 {
611 	set_pte(ptep, pte);
612 }
613 
pte_clear(struct mm_struct * mm,unsigned long addr,pte_t * ptep)614 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
615 			     pte_t *ptep)
616 {
617 	set_pte_at(mm, addr, ptep, __pte(0));
618 }
619 
pmd_clear(pmd_t * pmdp)620 static inline void pmd_clear(pmd_t *pmdp)
621 {
622 	set_pmd(pmdp, __pmd(0));
623 }
624 #endif	/* CONFIG_X86_PAE */
625 
626 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
arch_start_context_switch(struct task_struct * prev)627 static inline void arch_start_context_switch(struct task_struct *prev)
628 {
629 	PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
630 }
631 
arch_end_context_switch(struct task_struct * next)632 static inline void arch_end_context_switch(struct task_struct *next)
633 {
634 	PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
635 }
636 
637 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
arch_enter_lazy_mmu_mode(void)638 static inline void arch_enter_lazy_mmu_mode(void)
639 {
640 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
641 }
642 
arch_leave_lazy_mmu_mode(void)643 static inline void arch_leave_lazy_mmu_mode(void)
644 {
645 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
646 }
647 
arch_flush_lazy_mmu_mode(void)648 static inline void arch_flush_lazy_mmu_mode(void)
649 {
650 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.flush);
651 }
652 
__set_fixmap(unsigned idx,phys_addr_t phys,pgprot_t flags)653 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
654 				phys_addr_t phys, pgprot_t flags)
655 {
656 	pv_mmu_ops.set_fixmap(idx, phys, flags);
657 }
658 
659 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
660 
pv_queued_spin_lock_slowpath(struct qspinlock * lock,u32 val)661 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock,
662 							u32 val)
663 {
664 	PVOP_VCALL2(pv_lock_ops.queued_spin_lock_slowpath, lock, val);
665 }
666 
pv_queued_spin_unlock(struct qspinlock * lock)667 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock)
668 {
669 	PVOP_VCALLEE1(pv_lock_ops.queued_spin_unlock, lock);
670 }
671 
pv_wait(u8 * ptr,u8 val)672 static __always_inline void pv_wait(u8 *ptr, u8 val)
673 {
674 	PVOP_VCALL2(pv_lock_ops.wait, ptr, val);
675 }
676 
pv_kick(int cpu)677 static __always_inline void pv_kick(int cpu)
678 {
679 	PVOP_VCALL1(pv_lock_ops.kick, cpu);
680 }
681 
682 #endif /* SMP && PARAVIRT_SPINLOCKS */
683 
684 #ifdef CONFIG_X86_32
685 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
686 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
687 
688 /* save and restore all caller-save registers, except return value */
689 #define PV_SAVE_ALL_CALLER_REGS		"pushl %ecx;"
690 #define PV_RESTORE_ALL_CALLER_REGS	"popl  %ecx;"
691 
692 #define PV_FLAGS_ARG "0"
693 #define PV_EXTRA_CLOBBERS
694 #define PV_VEXTRA_CLOBBERS
695 #else
696 /* save and restore all caller-save registers, except return value */
697 #define PV_SAVE_ALL_CALLER_REGS						\
698 	"push %rcx;"							\
699 	"push %rdx;"							\
700 	"push %rsi;"							\
701 	"push %rdi;"							\
702 	"push %r8;"							\
703 	"push %r9;"							\
704 	"push %r10;"							\
705 	"push %r11;"
706 #define PV_RESTORE_ALL_CALLER_REGS					\
707 	"pop %r11;"							\
708 	"pop %r10;"							\
709 	"pop %r9;"							\
710 	"pop %r8;"							\
711 	"pop %rdi;"							\
712 	"pop %rsi;"							\
713 	"pop %rdx;"							\
714 	"pop %rcx;"
715 
716 /* We save some registers, but all of them, that's too much. We clobber all
717  * caller saved registers but the argument parameter */
718 #define PV_SAVE_REGS "pushq %%rdi;"
719 #define PV_RESTORE_REGS "popq %%rdi;"
720 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
721 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
722 #define PV_FLAGS_ARG "D"
723 #endif
724 
725 /*
726  * Generate a thunk around a function which saves all caller-save
727  * registers except for the return value.  This allows C functions to
728  * be called from assembler code where fewer than normal registers are
729  * available.  It may also help code generation around calls from C
730  * code if the common case doesn't use many registers.
731  *
732  * When a callee is wrapped in a thunk, the caller can assume that all
733  * arg regs and all scratch registers are preserved across the
734  * call. The return value in rax/eax will not be saved, even for void
735  * functions.
736  */
737 #define PV_THUNK_NAME(func) "__raw_callee_save_" #func
738 #define PV_CALLEE_SAVE_REGS_THUNK(func)					\
739 	extern typeof(func) __raw_callee_save_##func;			\
740 									\
741 	asm(".pushsection .text;"					\
742 	    ".globl " PV_THUNK_NAME(func) ";"				\
743 	    ".type " PV_THUNK_NAME(func) ", @function;"			\
744 	    PV_THUNK_NAME(func) ":"					\
745 	    FRAME_BEGIN							\
746 	    PV_SAVE_ALL_CALLER_REGS					\
747 	    "call " #func ";"						\
748 	    PV_RESTORE_ALL_CALLER_REGS					\
749 	    FRAME_END							\
750 	    "ret;"							\
751 	    ".popsection")
752 
753 /* Get a reference to a callee-save function */
754 #define PV_CALLEE_SAVE(func)						\
755 	((struct paravirt_callee_save) { __raw_callee_save_##func })
756 
757 /* Promise that "func" already uses the right calling convention */
758 #define __PV_IS_CALLEE_SAVE(func)			\
759 	((struct paravirt_callee_save) { func })
760 
arch_local_save_flags(void)761 static inline notrace unsigned long arch_local_save_flags(void)
762 {
763 	return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
764 }
765 
arch_local_irq_restore(unsigned long f)766 static inline notrace void arch_local_irq_restore(unsigned long f)
767 {
768 	PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
769 }
770 
arch_local_irq_disable(void)771 static inline notrace void arch_local_irq_disable(void)
772 {
773 	PVOP_VCALLEE0(pv_irq_ops.irq_disable);
774 }
775 
arch_local_irq_enable(void)776 static inline notrace void arch_local_irq_enable(void)
777 {
778 	PVOP_VCALLEE0(pv_irq_ops.irq_enable);
779 }
780 
arch_local_irq_save(void)781 static inline notrace unsigned long arch_local_irq_save(void)
782 {
783 	unsigned long f;
784 
785 	f = arch_local_save_flags();
786 	arch_local_irq_disable();
787 	return f;
788 }
789 
790 
791 /* Make sure as little as possible of this mess escapes. */
792 #undef PARAVIRT_CALL
793 #undef __PVOP_CALL
794 #undef __PVOP_VCALL
795 #undef PVOP_VCALL0
796 #undef PVOP_CALL0
797 #undef PVOP_VCALL1
798 #undef PVOP_CALL1
799 #undef PVOP_VCALL2
800 #undef PVOP_CALL2
801 #undef PVOP_VCALL3
802 #undef PVOP_CALL3
803 #undef PVOP_VCALL4
804 #undef PVOP_CALL4
805 
806 extern void default_banner(void);
807 
808 #else  /* __ASSEMBLY__ */
809 
810 #define _PVSITE(ptype, clobbers, ops, word, algn)	\
811 771:;						\
812 	ops;					\
813 772:;						\
814 	.pushsection .parainstructions,"a";	\
815 	 .align	algn;				\
816 	 word 771b;				\
817 	 .byte ptype;				\
818 	 .byte 772b-771b;			\
819 	 .short clobbers;			\
820 	.popsection
821 
822 
823 #define COND_PUSH(set, mask, reg)			\
824 	.if ((~(set)) & mask); push %reg; .endif
825 #define COND_POP(set, mask, reg)			\
826 	.if ((~(set)) & mask); pop %reg; .endif
827 
828 #ifdef CONFIG_X86_64
829 
830 #define PV_SAVE_REGS(set)			\
831 	COND_PUSH(set, CLBR_RAX, rax);		\
832 	COND_PUSH(set, CLBR_RCX, rcx);		\
833 	COND_PUSH(set, CLBR_RDX, rdx);		\
834 	COND_PUSH(set, CLBR_RSI, rsi);		\
835 	COND_PUSH(set, CLBR_RDI, rdi);		\
836 	COND_PUSH(set, CLBR_R8, r8);		\
837 	COND_PUSH(set, CLBR_R9, r9);		\
838 	COND_PUSH(set, CLBR_R10, r10);		\
839 	COND_PUSH(set, CLBR_R11, r11)
840 #define PV_RESTORE_REGS(set)			\
841 	COND_POP(set, CLBR_R11, r11);		\
842 	COND_POP(set, CLBR_R10, r10);		\
843 	COND_POP(set, CLBR_R9, r9);		\
844 	COND_POP(set, CLBR_R8, r8);		\
845 	COND_POP(set, CLBR_RDI, rdi);		\
846 	COND_POP(set, CLBR_RSI, rsi);		\
847 	COND_POP(set, CLBR_RDX, rdx);		\
848 	COND_POP(set, CLBR_RCX, rcx);		\
849 	COND_POP(set, CLBR_RAX, rax)
850 
851 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
852 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
853 #define PARA_INDIRECT(addr)	*addr(%rip)
854 #else
855 #define PV_SAVE_REGS(set)			\
856 	COND_PUSH(set, CLBR_EAX, eax);		\
857 	COND_PUSH(set, CLBR_EDI, edi);		\
858 	COND_PUSH(set, CLBR_ECX, ecx);		\
859 	COND_PUSH(set, CLBR_EDX, edx)
860 #define PV_RESTORE_REGS(set)			\
861 	COND_POP(set, CLBR_EDX, edx);		\
862 	COND_POP(set, CLBR_ECX, ecx);		\
863 	COND_POP(set, CLBR_EDI, edi);		\
864 	COND_POP(set, CLBR_EAX, eax)
865 
866 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
867 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
868 #define PARA_INDIRECT(addr)	*%cs:addr
869 #endif
870 
871 #define INTERRUPT_RETURN						\
872 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,	\
873 		  ANNOTATE_RETPOLINE_SAFE;					\
874 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret);)
875 
876 #define DISABLE_INTERRUPTS(clobbers)					\
877 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
878 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
879 		  ANNOTATE_RETPOLINE_SAFE;					\
880 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);	\
881 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
882 
883 #define ENABLE_INTERRUPTS(clobbers)					\
884 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,	\
885 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
886 		  ANNOTATE_RETPOLINE_SAFE;					\
887 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);	\
888 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
889 
890 #ifdef CONFIG_X86_32
891 #define GET_CR0_INTO_EAX				\
892 	push %ecx; push %edx;				\
893 	ANNOTATE_RETPOLINE_SAFE;				\
894 	call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0);	\
895 	pop %edx; pop %ecx
896 #else	/* !CONFIG_X86_32 */
897 
898 /*
899  * If swapgs is used while the userspace stack is still current,
900  * there's no way to call a pvop.  The PV replacement *must* be
901  * inlined, or the swapgs instruction must be trapped and emulated.
902  */
903 #define SWAPGS_UNSAFE_STACK						\
904 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
905 		  swapgs)
906 
907 /*
908  * Note: swapgs is very special, and in practise is either going to be
909  * implemented with a single "swapgs" instruction or something very
910  * special.  Either way, we don't need to save any registers for
911  * it.
912  */
913 #define SWAPGS								\
914 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
915 		  ANNOTATE_RETPOLINE_SAFE;					\
916 		  call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs);		\
917 		 )
918 
919 #define GET_CR2_INTO_RAX				\
920 	ANNOTATE_RETPOLINE_SAFE;				\
921 	call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2);
922 
923 #define PARAVIRT_ADJUST_EXCEPTION_FRAME					\
924 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
925 		  CLBR_NONE,						\
926 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
927 
928 #define USERGS_SYSRET64							\
929 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),	\
930 		  CLBR_NONE,						\
931 		  ANNOTATE_RETPOLINE_SAFE;					\
932 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64);)
933 #endif	/* CONFIG_X86_32 */
934 
935 #endif /* __ASSEMBLY__ */
936 #else  /* CONFIG_PARAVIRT */
937 # define default_banner x86_init_noop
938 #ifndef __ASSEMBLY__
paravirt_arch_dup_mmap(struct mm_struct * oldmm,struct mm_struct * mm)939 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
940 					  struct mm_struct *mm)
941 {
942 }
943 
paravirt_arch_exit_mmap(struct mm_struct * mm)944 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
945 {
946 }
947 #endif /* __ASSEMBLY__ */
948 #endif /* !CONFIG_PARAVIRT */
949 #endif /* _ASM_X86_PARAVIRT_H */
950