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/types.h>
14 #include <linux/cpumask.h>
15 
paravirt_enabled(void)16 static inline int paravirt_enabled(void)
17 {
18 	return pv_info.paravirt_enabled;
19 }
20 
load_sp0(struct tss_struct * tss,struct thread_struct * thread)21 static inline void load_sp0(struct tss_struct *tss,
22 			     struct thread_struct *thread)
23 {
24 	PVOP_VCALL2(pv_cpu_ops.load_sp0, tss, thread);
25 }
26 
27 /* The paravirtualized CPUID instruction. */
__cpuid(unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)28 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
29 			   unsigned int *ecx, unsigned int *edx)
30 {
31 	PVOP_VCALL4(pv_cpu_ops.cpuid, eax, ebx, ecx, edx);
32 }
33 
34 /*
35  * These special macros can be used to get or set a debugging register
36  */
paravirt_get_debugreg(int reg)37 static inline unsigned long paravirt_get_debugreg(int reg)
38 {
39 	return PVOP_CALL1(unsigned long, pv_cpu_ops.get_debugreg, reg);
40 }
41 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
set_debugreg(unsigned long val,int reg)42 static inline void set_debugreg(unsigned long val, int reg)
43 {
44 	PVOP_VCALL2(pv_cpu_ops.set_debugreg, reg, val);
45 }
46 
clts(void)47 static inline void clts(void)
48 {
49 	PVOP_VCALL0(pv_cpu_ops.clts);
50 }
51 
read_cr0(void)52 static inline unsigned long read_cr0(void)
53 {
54 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr0);
55 }
56 
write_cr0(unsigned long x)57 static inline void write_cr0(unsigned long x)
58 {
59 	PVOP_VCALL1(pv_cpu_ops.write_cr0, x);
60 }
61 
read_cr2(void)62 static inline unsigned long read_cr2(void)
63 {
64 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr2);
65 }
66 
write_cr2(unsigned long x)67 static inline void write_cr2(unsigned long x)
68 {
69 	PVOP_VCALL1(pv_mmu_ops.write_cr2, x);
70 }
71 
read_cr3(void)72 static inline unsigned long read_cr3(void)
73 {
74 	return PVOP_CALL0(unsigned long, pv_mmu_ops.read_cr3);
75 }
76 
write_cr3(unsigned long x)77 static inline void write_cr3(unsigned long x)
78 {
79 	PVOP_VCALL1(pv_mmu_ops.write_cr3, x);
80 }
81 
read_cr4(void)82 static inline unsigned long read_cr4(void)
83 {
84 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4);
85 }
read_cr4_safe(void)86 static inline unsigned long read_cr4_safe(void)
87 {
88 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr4_safe);
89 }
90 
write_cr4(unsigned long x)91 static inline void write_cr4(unsigned long x)
92 {
93 	PVOP_VCALL1(pv_cpu_ops.write_cr4, x);
94 }
95 
96 #ifdef CONFIG_X86_64
read_cr8(void)97 static inline unsigned long read_cr8(void)
98 {
99 	return PVOP_CALL0(unsigned long, pv_cpu_ops.read_cr8);
100 }
101 
write_cr8(unsigned long x)102 static inline void write_cr8(unsigned long x)
103 {
104 	PVOP_VCALL1(pv_cpu_ops.write_cr8, x);
105 }
106 #endif
107 
arch_safe_halt(void)108 static inline void arch_safe_halt(void)
109 {
110 	PVOP_VCALL0(pv_irq_ops.safe_halt);
111 }
112 
halt(void)113 static inline void halt(void)
114 {
115 	PVOP_VCALL0(pv_irq_ops.halt);
116 }
117 
wbinvd(void)118 static inline void wbinvd(void)
119 {
120 	PVOP_VCALL0(pv_cpu_ops.wbinvd);
121 }
122 
123 #define get_kernel_rpl()  (pv_info.kernel_rpl)
124 
paravirt_read_msr(unsigned msr,int * err)125 static inline u64 paravirt_read_msr(unsigned msr, int *err)
126 {
127 	return PVOP_CALL2(u64, pv_cpu_ops.read_msr, msr, err);
128 }
129 
paravirt_rdmsr_regs(u32 * regs)130 static inline int paravirt_rdmsr_regs(u32 *regs)
131 {
132 	return PVOP_CALL1(int, pv_cpu_ops.rdmsr_regs, regs);
133 }
134 
paravirt_write_msr(unsigned msr,unsigned low,unsigned high)135 static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
136 {
137 	return PVOP_CALL3(int, pv_cpu_ops.write_msr, msr, low, high);
138 }
139 
paravirt_wrmsr_regs(u32 * regs)140 static inline int paravirt_wrmsr_regs(u32 *regs)
141 {
142 	return PVOP_CALL1(int, pv_cpu_ops.wrmsr_regs, regs);
143 }
144 
145 /* These should all do BUG_ON(_err), but our headers are too tangled. */
146 #define rdmsr(msr, val1, val2)			\
147 do {						\
148 	int _err;				\
149 	u64 _l = paravirt_read_msr(msr, &_err);	\
150 	val1 = (u32)_l;				\
151 	val2 = _l >> 32;			\
152 } while (0)
153 
154 #define wrmsr(msr, val1, val2)			\
155 do {						\
156 	paravirt_write_msr(msr, val1, val2);	\
157 } while (0)
158 
159 #define rdmsrl(msr, val)			\
160 do {						\
161 	int _err;				\
162 	val = paravirt_read_msr(msr, &_err);	\
163 } while (0)
164 
165 #define wrmsrl(msr, val)	wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
166 #define wrmsr_safe(msr, a, b)	paravirt_write_msr(msr, a, b)
167 
168 /* rdmsr with exception handling */
169 #define rdmsr_safe(msr, a, b)			\
170 ({						\
171 	int _err;				\
172 	u64 _l = paravirt_read_msr(msr, &_err);	\
173 	(*a) = (u32)_l;				\
174 	(*b) = _l >> 32;			\
175 	_err;					\
176 })
177 
178 #define rdmsr_safe_regs(regs)	paravirt_rdmsr_regs(regs)
179 #define wrmsr_safe_regs(regs)	paravirt_wrmsr_regs(regs)
180 
rdmsrl_safe(unsigned msr,unsigned long long * p)181 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
182 {
183 	int err;
184 
185 	*p = paravirt_read_msr(msr, &err);
186 	return err;
187 }
rdmsrl_amd_safe(unsigned msr,unsigned long long * p)188 static inline int rdmsrl_amd_safe(unsigned msr, unsigned long long *p)
189 {
190 	u32 gprs[8] = { 0 };
191 	int err;
192 
193 	gprs[1] = msr;
194 	gprs[7] = 0x9c5a203a;
195 
196 	err = paravirt_rdmsr_regs(gprs);
197 
198 	*p = gprs[0] | ((u64)gprs[2] << 32);
199 
200 	return err;
201 }
202 
wrmsrl_amd_safe(unsigned msr,unsigned long long val)203 static inline int wrmsrl_amd_safe(unsigned msr, unsigned long long val)
204 {
205 	u32 gprs[8] = { 0 };
206 
207 	gprs[0] = (u32)val;
208 	gprs[1] = msr;
209 	gprs[2] = val >> 32;
210 	gprs[7] = 0x9c5a203a;
211 
212 	return paravirt_wrmsr_regs(gprs);
213 }
214 
paravirt_read_tsc(void)215 static inline u64 paravirt_read_tsc(void)
216 {
217 	return PVOP_CALL0(u64, pv_cpu_ops.read_tsc);
218 }
219 
220 #define rdtscl(low)				\
221 do {						\
222 	u64 _l = paravirt_read_tsc();		\
223 	low = (int)_l;				\
224 } while (0)
225 
226 #define rdtscll(val) (val = paravirt_read_tsc())
227 
paravirt_sched_clock(void)228 static inline unsigned long long paravirt_sched_clock(void)
229 {
230 	return PVOP_CALL0(unsigned long long, pv_time_ops.sched_clock);
231 }
232 
paravirt_read_pmc(int counter)233 static inline unsigned long long paravirt_read_pmc(int counter)
234 {
235 	return PVOP_CALL1(u64, pv_cpu_ops.read_pmc, counter);
236 }
237 
238 #define rdpmc(counter, low, high)		\
239 do {						\
240 	u64 _l = paravirt_read_pmc(counter);	\
241 	low = (u32)_l;				\
242 	high = _l >> 32;			\
243 } while (0)
244 
paravirt_rdtscp(unsigned int * aux)245 static inline unsigned long long paravirt_rdtscp(unsigned int *aux)
246 {
247 	return PVOP_CALL1(u64, pv_cpu_ops.read_tscp, aux);
248 }
249 
250 #define rdtscp(low, high, aux)				\
251 do {							\
252 	int __aux;					\
253 	unsigned long __val = paravirt_rdtscp(&__aux);	\
254 	(low) = (u32)__val;				\
255 	(high) = (u32)(__val >> 32);			\
256 	(aux) = __aux;					\
257 } while (0)
258 
259 #define rdtscpll(val, aux)				\
260 do {							\
261 	unsigned long __aux; 				\
262 	val = paravirt_rdtscp(&__aux);			\
263 	(aux) = __aux;					\
264 } while (0)
265 
paravirt_alloc_ldt(struct desc_struct * ldt,unsigned entries)266 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
267 {
268 	PVOP_VCALL2(pv_cpu_ops.alloc_ldt, ldt, entries);
269 }
270 
paravirt_free_ldt(struct desc_struct * ldt,unsigned entries)271 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
272 {
273 	PVOP_VCALL2(pv_cpu_ops.free_ldt, ldt, entries);
274 }
275 
load_TR_desc(void)276 static inline void load_TR_desc(void)
277 {
278 	PVOP_VCALL0(pv_cpu_ops.load_tr_desc);
279 }
load_gdt(const struct desc_ptr * dtr)280 static inline void load_gdt(const struct desc_ptr *dtr)
281 {
282 	PVOP_VCALL1(pv_cpu_ops.load_gdt, dtr);
283 }
load_idt(const struct desc_ptr * dtr)284 static inline void load_idt(const struct desc_ptr *dtr)
285 {
286 	PVOP_VCALL1(pv_cpu_ops.load_idt, dtr);
287 }
set_ldt(const void * addr,unsigned entries)288 static inline void set_ldt(const void *addr, unsigned entries)
289 {
290 	PVOP_VCALL2(pv_cpu_ops.set_ldt, addr, entries);
291 }
store_gdt(struct desc_ptr * dtr)292 static inline void store_gdt(struct desc_ptr *dtr)
293 {
294 	PVOP_VCALL1(pv_cpu_ops.store_gdt, dtr);
295 }
store_idt(struct desc_ptr * dtr)296 static inline void store_idt(struct desc_ptr *dtr)
297 {
298 	PVOP_VCALL1(pv_cpu_ops.store_idt, dtr);
299 }
paravirt_store_tr(void)300 static inline unsigned long paravirt_store_tr(void)
301 {
302 	return PVOP_CALL0(unsigned long, pv_cpu_ops.store_tr);
303 }
304 #define store_tr(tr)	((tr) = paravirt_store_tr())
load_TLS(struct thread_struct * t,unsigned cpu)305 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
306 {
307 	PVOP_VCALL2(pv_cpu_ops.load_tls, t, cpu);
308 }
309 
310 #ifdef CONFIG_X86_64
load_gs_index(unsigned int gs)311 static inline void load_gs_index(unsigned int gs)
312 {
313 	PVOP_VCALL1(pv_cpu_ops.load_gs_index, gs);
314 }
315 #endif
316 
write_ldt_entry(struct desc_struct * dt,int entry,const void * desc)317 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
318 				   const void *desc)
319 {
320 	PVOP_VCALL3(pv_cpu_ops.write_ldt_entry, dt, entry, desc);
321 }
322 
write_gdt_entry(struct desc_struct * dt,int entry,void * desc,int type)323 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
324 				   void *desc, int type)
325 {
326 	PVOP_VCALL4(pv_cpu_ops.write_gdt_entry, dt, entry, desc, type);
327 }
328 
write_idt_entry(gate_desc * dt,int entry,const gate_desc * g)329 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
330 {
331 	PVOP_VCALL3(pv_cpu_ops.write_idt_entry, dt, entry, g);
332 }
set_iopl_mask(unsigned mask)333 static inline void set_iopl_mask(unsigned mask)
334 {
335 	PVOP_VCALL1(pv_cpu_ops.set_iopl_mask, mask);
336 }
337 
338 /* The paravirtualized I/O functions */
slow_down_io(void)339 static inline void slow_down_io(void)
340 {
341 	pv_cpu_ops.io_delay();
342 #ifdef REALLY_SLOW_IO
343 	pv_cpu_ops.io_delay();
344 	pv_cpu_ops.io_delay();
345 	pv_cpu_ops.io_delay();
346 #endif
347 }
348 
349 #ifdef CONFIG_SMP
startup_ipi_hook(int phys_apicid,unsigned long start_eip,unsigned long start_esp)350 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
351 				    unsigned long start_esp)
352 {
353 	PVOP_VCALL3(pv_apic_ops.startup_ipi_hook,
354 		    phys_apicid, start_eip, start_esp);
355 }
356 #endif
357 
paravirt_activate_mm(struct mm_struct * prev,struct mm_struct * next)358 static inline void paravirt_activate_mm(struct mm_struct *prev,
359 					struct mm_struct *next)
360 {
361 	PVOP_VCALL2(pv_mmu_ops.activate_mm, prev, next);
362 }
363 
arch_dup_mmap(struct mm_struct * oldmm,struct mm_struct * mm)364 static inline void arch_dup_mmap(struct mm_struct *oldmm,
365 				 struct mm_struct *mm)
366 {
367 	PVOP_VCALL2(pv_mmu_ops.dup_mmap, oldmm, mm);
368 }
369 
arch_exit_mmap(struct mm_struct * mm)370 static inline void arch_exit_mmap(struct mm_struct *mm)
371 {
372 	PVOP_VCALL1(pv_mmu_ops.exit_mmap, mm);
373 }
374 
__flush_tlb(void)375 static inline void __flush_tlb(void)
376 {
377 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_user);
378 }
__flush_tlb_global(void)379 static inline void __flush_tlb_global(void)
380 {
381 	PVOP_VCALL0(pv_mmu_ops.flush_tlb_kernel);
382 }
__flush_tlb_single(unsigned long addr)383 static inline void __flush_tlb_single(unsigned long addr)
384 {
385 	PVOP_VCALL1(pv_mmu_ops.flush_tlb_single, addr);
386 }
387 
flush_tlb_others(const struct cpumask * cpumask,struct mm_struct * mm,unsigned long va)388 static inline void flush_tlb_others(const struct cpumask *cpumask,
389 				    struct mm_struct *mm,
390 				    unsigned long va)
391 {
392 	PVOP_VCALL3(pv_mmu_ops.flush_tlb_others, cpumask, mm, va);
393 }
394 
paravirt_pgd_alloc(struct mm_struct * mm)395 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
396 {
397 	return PVOP_CALL1(int, pv_mmu_ops.pgd_alloc, mm);
398 }
399 
paravirt_pgd_free(struct mm_struct * mm,pgd_t * pgd)400 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
401 {
402 	PVOP_VCALL2(pv_mmu_ops.pgd_free, mm, pgd);
403 }
404 
paravirt_alloc_pte(struct mm_struct * mm,unsigned long pfn)405 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
406 {
407 	PVOP_VCALL2(pv_mmu_ops.alloc_pte, mm, pfn);
408 }
paravirt_release_pte(unsigned long pfn)409 static inline void paravirt_release_pte(unsigned long pfn)
410 {
411 	PVOP_VCALL1(pv_mmu_ops.release_pte, pfn);
412 }
413 
paravirt_alloc_pmd(struct mm_struct * mm,unsigned long pfn)414 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
415 {
416 	PVOP_VCALL2(pv_mmu_ops.alloc_pmd, mm, pfn);
417 }
418 
paravirt_release_pmd(unsigned long pfn)419 static inline void paravirt_release_pmd(unsigned long pfn)
420 {
421 	PVOP_VCALL1(pv_mmu_ops.release_pmd, pfn);
422 }
423 
paravirt_alloc_pud(struct mm_struct * mm,unsigned long pfn)424 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
425 {
426 	PVOP_VCALL2(pv_mmu_ops.alloc_pud, mm, pfn);
427 }
paravirt_release_pud(unsigned long pfn)428 static inline void paravirt_release_pud(unsigned long pfn)
429 {
430 	PVOP_VCALL1(pv_mmu_ops.release_pud, pfn);
431 }
432 
pte_update(struct mm_struct * mm,unsigned long addr,pte_t * ptep)433 static inline void pte_update(struct mm_struct *mm, unsigned long addr,
434 			      pte_t *ptep)
435 {
436 	PVOP_VCALL3(pv_mmu_ops.pte_update, mm, addr, ptep);
437 }
pmd_update(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp)438 static inline void pmd_update(struct mm_struct *mm, unsigned long addr,
439 			      pmd_t *pmdp)
440 {
441 	PVOP_VCALL3(pv_mmu_ops.pmd_update, mm, addr, pmdp);
442 }
443 
pte_update_defer(struct mm_struct * mm,unsigned long addr,pte_t * ptep)444 static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
445 				    pte_t *ptep)
446 {
447 	PVOP_VCALL3(pv_mmu_ops.pte_update_defer, mm, addr, ptep);
448 }
449 
pmd_update_defer(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp)450 static inline void pmd_update_defer(struct mm_struct *mm, unsigned long addr,
451 				    pmd_t *pmdp)
452 {
453 	PVOP_VCALL3(pv_mmu_ops.pmd_update_defer, mm, addr, pmdp);
454 }
455 
__pte(pteval_t val)456 static inline pte_t __pte(pteval_t val)
457 {
458 	pteval_t ret;
459 
460 	if (sizeof(pteval_t) > sizeof(long))
461 		ret = PVOP_CALLEE2(pteval_t,
462 				   pv_mmu_ops.make_pte,
463 				   val, (u64)val >> 32);
464 	else
465 		ret = PVOP_CALLEE1(pteval_t,
466 				   pv_mmu_ops.make_pte,
467 				   val);
468 
469 	return (pte_t) { .pte = ret };
470 }
471 
pte_val(pte_t pte)472 static inline pteval_t pte_val(pte_t pte)
473 {
474 	pteval_t ret;
475 
476 	if (sizeof(pteval_t) > sizeof(long))
477 		ret = PVOP_CALLEE2(pteval_t, pv_mmu_ops.pte_val,
478 				   pte.pte, (u64)pte.pte >> 32);
479 	else
480 		ret = PVOP_CALLEE1(pteval_t, pv_mmu_ops.pte_val,
481 				   pte.pte);
482 
483 	return ret;
484 }
485 
__pgd(pgdval_t val)486 static inline pgd_t __pgd(pgdval_t val)
487 {
488 	pgdval_t ret;
489 
490 	if (sizeof(pgdval_t) > sizeof(long))
491 		ret = PVOP_CALLEE2(pgdval_t, pv_mmu_ops.make_pgd,
492 				   val, (u64)val >> 32);
493 	else
494 		ret = PVOP_CALLEE1(pgdval_t, pv_mmu_ops.make_pgd,
495 				   val);
496 
497 	return (pgd_t) { ret };
498 }
499 
pgd_val(pgd_t pgd)500 static inline pgdval_t pgd_val(pgd_t pgd)
501 {
502 	pgdval_t ret;
503 
504 	if (sizeof(pgdval_t) > sizeof(long))
505 		ret =  PVOP_CALLEE2(pgdval_t, pv_mmu_ops.pgd_val,
506 				    pgd.pgd, (u64)pgd.pgd >> 32);
507 	else
508 		ret =  PVOP_CALLEE1(pgdval_t, pv_mmu_ops.pgd_val,
509 				    pgd.pgd);
510 
511 	return ret;
512 }
513 
514 #define  __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
ptep_modify_prot_start(struct mm_struct * mm,unsigned long addr,pte_t * ptep)515 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr,
516 					   pte_t *ptep)
517 {
518 	pteval_t ret;
519 
520 	ret = PVOP_CALL3(pteval_t, pv_mmu_ops.ptep_modify_prot_start,
521 			 mm, addr, ptep);
522 
523 	return (pte_t) { .pte = ret };
524 }
525 
ptep_modify_prot_commit(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte)526 static inline void ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
527 					   pte_t *ptep, pte_t pte)
528 {
529 	if (sizeof(pteval_t) > sizeof(long))
530 		/* 5 arg words */
531 		pv_mmu_ops.ptep_modify_prot_commit(mm, addr, ptep, pte);
532 	else
533 		PVOP_VCALL4(pv_mmu_ops.ptep_modify_prot_commit,
534 			    mm, addr, ptep, pte.pte);
535 }
536 
set_pte(pte_t * ptep,pte_t pte)537 static inline void set_pte(pte_t *ptep, pte_t pte)
538 {
539 	if (sizeof(pteval_t) > sizeof(long))
540 		PVOP_VCALL3(pv_mmu_ops.set_pte, ptep,
541 			    pte.pte, (u64)pte.pte >> 32);
542 	else
543 		PVOP_VCALL2(pv_mmu_ops.set_pte, ptep,
544 			    pte.pte);
545 }
546 
set_pte_at(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte)547 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
548 			      pte_t *ptep, pte_t pte)
549 {
550 	if (sizeof(pteval_t) > sizeof(long))
551 		/* 5 arg words */
552 		pv_mmu_ops.set_pte_at(mm, addr, ptep, pte);
553 	else
554 		PVOP_VCALL4(pv_mmu_ops.set_pte_at, mm, addr, ptep, pte.pte);
555 }
556 
557 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
set_pmd_at(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp,pmd_t pmd)558 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
559 			      pmd_t *pmdp, pmd_t pmd)
560 {
561 	if (sizeof(pmdval_t) > sizeof(long))
562 		/* 5 arg words */
563 		pv_mmu_ops.set_pmd_at(mm, addr, pmdp, pmd);
564 	else
565 		PVOP_VCALL4(pv_mmu_ops.set_pmd_at, mm, addr, pmdp,
566 			    native_pmd_val(pmd));
567 }
568 #endif
569 
set_pmd(pmd_t * pmdp,pmd_t pmd)570 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
571 {
572 	pmdval_t val = native_pmd_val(pmd);
573 
574 	if (sizeof(pmdval_t) > sizeof(long))
575 		PVOP_VCALL3(pv_mmu_ops.set_pmd, pmdp, val, (u64)val >> 32);
576 	else
577 		PVOP_VCALL2(pv_mmu_ops.set_pmd, pmdp, val);
578 }
579 
580 #if PAGETABLE_LEVELS >= 3
__pmd(pmdval_t val)581 static inline pmd_t __pmd(pmdval_t val)
582 {
583 	pmdval_t ret;
584 
585 	if (sizeof(pmdval_t) > sizeof(long))
586 		ret = PVOP_CALLEE2(pmdval_t, pv_mmu_ops.make_pmd,
587 				   val, (u64)val >> 32);
588 	else
589 		ret = PVOP_CALLEE1(pmdval_t, pv_mmu_ops.make_pmd,
590 				   val);
591 
592 	return (pmd_t) { ret };
593 }
594 
pmd_val(pmd_t pmd)595 static inline pmdval_t pmd_val(pmd_t pmd)
596 {
597 	pmdval_t ret;
598 
599 	if (sizeof(pmdval_t) > sizeof(long))
600 		ret =  PVOP_CALLEE2(pmdval_t, pv_mmu_ops.pmd_val,
601 				    pmd.pmd, (u64)pmd.pmd >> 32);
602 	else
603 		ret =  PVOP_CALLEE1(pmdval_t, pv_mmu_ops.pmd_val,
604 				    pmd.pmd);
605 
606 	return ret;
607 }
608 
set_pud(pud_t * pudp,pud_t pud)609 static inline void set_pud(pud_t *pudp, pud_t pud)
610 {
611 	pudval_t val = native_pud_val(pud);
612 
613 	if (sizeof(pudval_t) > sizeof(long))
614 		PVOP_VCALL3(pv_mmu_ops.set_pud, pudp,
615 			    val, (u64)val >> 32);
616 	else
617 		PVOP_VCALL2(pv_mmu_ops.set_pud, pudp,
618 			    val);
619 }
620 #if PAGETABLE_LEVELS == 4
__pud(pudval_t val)621 static inline pud_t __pud(pudval_t val)
622 {
623 	pudval_t ret;
624 
625 	if (sizeof(pudval_t) > sizeof(long))
626 		ret = PVOP_CALLEE2(pudval_t, pv_mmu_ops.make_pud,
627 				   val, (u64)val >> 32);
628 	else
629 		ret = PVOP_CALLEE1(pudval_t, pv_mmu_ops.make_pud,
630 				   val);
631 
632 	return (pud_t) { ret };
633 }
634 
pud_val(pud_t pud)635 static inline pudval_t pud_val(pud_t pud)
636 {
637 	pudval_t ret;
638 
639 	if (sizeof(pudval_t) > sizeof(long))
640 		ret =  PVOP_CALLEE2(pudval_t, pv_mmu_ops.pud_val,
641 				    pud.pud, (u64)pud.pud >> 32);
642 	else
643 		ret =  PVOP_CALLEE1(pudval_t, pv_mmu_ops.pud_val,
644 				    pud.pud);
645 
646 	return ret;
647 }
648 
set_pgd(pgd_t * pgdp,pgd_t pgd)649 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
650 {
651 	pgdval_t val = native_pgd_val(pgd);
652 
653 	if (sizeof(pgdval_t) > sizeof(long))
654 		PVOP_VCALL3(pv_mmu_ops.set_pgd, pgdp,
655 			    val, (u64)val >> 32);
656 	else
657 		PVOP_VCALL2(pv_mmu_ops.set_pgd, pgdp,
658 			    val);
659 }
660 
pgd_clear(pgd_t * pgdp)661 static inline void pgd_clear(pgd_t *pgdp)
662 {
663 	set_pgd(pgdp, __pgd(0));
664 }
665 
pud_clear(pud_t * pudp)666 static inline void pud_clear(pud_t *pudp)
667 {
668 	set_pud(pudp, __pud(0));
669 }
670 
671 #endif	/* PAGETABLE_LEVELS == 4 */
672 
673 #endif	/* PAGETABLE_LEVELS >= 3 */
674 
675 #ifdef CONFIG_X86_PAE
676 /* Special-case pte-setting operations for PAE, which can't update a
677    64-bit pte atomically */
set_pte_atomic(pte_t * ptep,pte_t pte)678 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
679 {
680 	PVOP_VCALL3(pv_mmu_ops.set_pte_atomic, ptep,
681 		    pte.pte, pte.pte >> 32);
682 }
683 
pte_clear(struct mm_struct * mm,unsigned long addr,pte_t * ptep)684 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
685 			     pte_t *ptep)
686 {
687 	PVOP_VCALL3(pv_mmu_ops.pte_clear, mm, addr, ptep);
688 }
689 
pmd_clear(pmd_t * pmdp)690 static inline void pmd_clear(pmd_t *pmdp)
691 {
692 	PVOP_VCALL1(pv_mmu_ops.pmd_clear, pmdp);
693 }
694 #else  /* !CONFIG_X86_PAE */
set_pte_atomic(pte_t * ptep,pte_t pte)695 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
696 {
697 	set_pte(ptep, pte);
698 }
699 
pte_clear(struct mm_struct * mm,unsigned long addr,pte_t * ptep)700 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
701 			     pte_t *ptep)
702 {
703 	set_pte_at(mm, addr, ptep, __pte(0));
704 }
705 
pmd_clear(pmd_t * pmdp)706 static inline void pmd_clear(pmd_t *pmdp)
707 {
708 	set_pmd(pmdp, __pmd(0));
709 }
710 #endif	/* CONFIG_X86_PAE */
711 
712 #define  __HAVE_ARCH_START_CONTEXT_SWITCH
arch_start_context_switch(struct task_struct * prev)713 static inline void arch_start_context_switch(struct task_struct *prev)
714 {
715 	PVOP_VCALL1(pv_cpu_ops.start_context_switch, prev);
716 }
717 
arch_end_context_switch(struct task_struct * next)718 static inline void arch_end_context_switch(struct task_struct *next)
719 {
720 	PVOP_VCALL1(pv_cpu_ops.end_context_switch, next);
721 }
722 
723 #define  __HAVE_ARCH_ENTER_LAZY_MMU_MODE
arch_enter_lazy_mmu_mode(void)724 static inline void arch_enter_lazy_mmu_mode(void)
725 {
726 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.enter);
727 }
728 
arch_leave_lazy_mmu_mode(void)729 static inline void arch_leave_lazy_mmu_mode(void)
730 {
731 	PVOP_VCALL0(pv_mmu_ops.lazy_mode.leave);
732 }
733 
734 void arch_flush_lazy_mmu_mode(void);
735 
__set_fixmap(unsigned idx,phys_addr_t phys,pgprot_t flags)736 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
737 				phys_addr_t phys, pgprot_t flags)
738 {
739 	pv_mmu_ops.set_fixmap(idx, phys, flags);
740 }
741 
742 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
743 
arch_spin_is_locked(struct arch_spinlock * lock)744 static inline int arch_spin_is_locked(struct arch_spinlock *lock)
745 {
746 	return PVOP_CALL1(int, pv_lock_ops.spin_is_locked, lock);
747 }
748 
arch_spin_is_contended(struct arch_spinlock * lock)749 static inline int arch_spin_is_contended(struct arch_spinlock *lock)
750 {
751 	return PVOP_CALL1(int, pv_lock_ops.spin_is_contended, lock);
752 }
753 #define arch_spin_is_contended	arch_spin_is_contended
754 
arch_spin_lock(struct arch_spinlock * lock)755 static __always_inline void arch_spin_lock(struct arch_spinlock *lock)
756 {
757 	PVOP_VCALL1(pv_lock_ops.spin_lock, lock);
758 }
759 
arch_spin_lock_flags(struct arch_spinlock * lock,unsigned long flags)760 static __always_inline void arch_spin_lock_flags(struct arch_spinlock *lock,
761 						  unsigned long flags)
762 {
763 	PVOP_VCALL2(pv_lock_ops.spin_lock_flags, lock, flags);
764 }
765 
arch_spin_trylock(struct arch_spinlock * lock)766 static __always_inline int arch_spin_trylock(struct arch_spinlock *lock)
767 {
768 	return PVOP_CALL1(int, pv_lock_ops.spin_trylock, lock);
769 }
770 
arch_spin_unlock(struct arch_spinlock * lock)771 static __always_inline void arch_spin_unlock(struct arch_spinlock *lock)
772 {
773 	PVOP_VCALL1(pv_lock_ops.spin_unlock, lock);
774 }
775 
776 #endif
777 
778 #ifdef CONFIG_X86_32
779 #define PV_SAVE_REGS "pushl %ecx; pushl %edx;"
780 #define PV_RESTORE_REGS "popl %edx; popl %ecx;"
781 
782 /* save and restore all caller-save registers, except return value */
783 #define PV_SAVE_ALL_CALLER_REGS		"pushl %ecx;"
784 #define PV_RESTORE_ALL_CALLER_REGS	"popl  %ecx;"
785 
786 #define PV_FLAGS_ARG "0"
787 #define PV_EXTRA_CLOBBERS
788 #define PV_VEXTRA_CLOBBERS
789 #else
790 /* save and restore all caller-save registers, except return value */
791 #define PV_SAVE_ALL_CALLER_REGS						\
792 	"push %rcx;"							\
793 	"push %rdx;"							\
794 	"push %rsi;"							\
795 	"push %rdi;"							\
796 	"push %r8;"							\
797 	"push %r9;"							\
798 	"push %r10;"							\
799 	"push %r11;"
800 #define PV_RESTORE_ALL_CALLER_REGS					\
801 	"pop %r11;"							\
802 	"pop %r10;"							\
803 	"pop %r9;"							\
804 	"pop %r8;"							\
805 	"pop %rdi;"							\
806 	"pop %rsi;"							\
807 	"pop %rdx;"							\
808 	"pop %rcx;"
809 
810 /* We save some registers, but all of them, that's too much. We clobber all
811  * caller saved registers but the argument parameter */
812 #define PV_SAVE_REGS "pushq %%rdi;"
813 #define PV_RESTORE_REGS "popq %%rdi;"
814 #define PV_EXTRA_CLOBBERS EXTRA_CLOBBERS, "rcx" , "rdx", "rsi"
815 #define PV_VEXTRA_CLOBBERS EXTRA_CLOBBERS, "rdi", "rcx" , "rdx", "rsi"
816 #define PV_FLAGS_ARG "D"
817 #endif
818 
819 /*
820  * Generate a thunk around a function which saves all caller-save
821  * registers except for the return value.  This allows C functions to
822  * be called from assembler code where fewer than normal registers are
823  * available.  It may also help code generation around calls from C
824  * code if the common case doesn't use many registers.
825  *
826  * When a callee is wrapped in a thunk, the caller can assume that all
827  * arg regs and all scratch registers are preserved across the
828  * call. The return value in rax/eax will not be saved, even for void
829  * functions.
830  */
831 #define PV_CALLEE_SAVE_REGS_THUNK(func)					\
832 	extern typeof(func) __raw_callee_save_##func;			\
833 	static void *__##func##__ __used = func;			\
834 									\
835 	asm(".pushsection .text;"					\
836 	    "__raw_callee_save_" #func ": "				\
837 	    PV_SAVE_ALL_CALLER_REGS					\
838 	    "call " #func ";"						\
839 	    PV_RESTORE_ALL_CALLER_REGS					\
840 	    "ret;"							\
841 	    ".popsection")
842 
843 /* Get a reference to a callee-save function */
844 #define PV_CALLEE_SAVE(func)						\
845 	((struct paravirt_callee_save) { __raw_callee_save_##func })
846 
847 /* Promise that "func" already uses the right calling convention */
848 #define __PV_IS_CALLEE_SAVE(func)			\
849 	((struct paravirt_callee_save) { func })
850 
arch_local_save_flags(void)851 static inline notrace unsigned long arch_local_save_flags(void)
852 {
853 	return PVOP_CALLEE0(unsigned long, pv_irq_ops.save_fl);
854 }
855 
arch_local_irq_restore(unsigned long f)856 static inline notrace void arch_local_irq_restore(unsigned long f)
857 {
858 	PVOP_VCALLEE1(pv_irq_ops.restore_fl, f);
859 }
860 
arch_local_irq_disable(void)861 static inline notrace void arch_local_irq_disable(void)
862 {
863 	PVOP_VCALLEE0(pv_irq_ops.irq_disable);
864 }
865 
arch_local_irq_enable(void)866 static inline notrace void arch_local_irq_enable(void)
867 {
868 	PVOP_VCALLEE0(pv_irq_ops.irq_enable);
869 }
870 
arch_local_irq_save(void)871 static inline notrace unsigned long arch_local_irq_save(void)
872 {
873 	unsigned long f;
874 
875 	f = arch_local_save_flags();
876 	arch_local_irq_disable();
877 	return f;
878 }
879 
880 
881 /* Make sure as little as possible of this mess escapes. */
882 #undef PARAVIRT_CALL
883 #undef __PVOP_CALL
884 #undef __PVOP_VCALL
885 #undef PVOP_VCALL0
886 #undef PVOP_CALL0
887 #undef PVOP_VCALL1
888 #undef PVOP_CALL1
889 #undef PVOP_VCALL2
890 #undef PVOP_CALL2
891 #undef PVOP_VCALL3
892 #undef PVOP_CALL3
893 #undef PVOP_VCALL4
894 #undef PVOP_CALL4
895 
896 extern void default_banner(void);
897 
898 #else  /* __ASSEMBLY__ */
899 
900 #define _PVSITE(ptype, clobbers, ops, word, algn)	\
901 771:;						\
902 	ops;					\
903 772:;						\
904 	.pushsection .parainstructions,"a";	\
905 	 .align	algn;				\
906 	 word 771b;				\
907 	 .byte ptype;				\
908 	 .byte 772b-771b;			\
909 	 .short clobbers;			\
910 	.popsection
911 
912 
913 #define COND_PUSH(set, mask, reg)			\
914 	.if ((~(set)) & mask); push %reg; .endif
915 #define COND_POP(set, mask, reg)			\
916 	.if ((~(set)) & mask); pop %reg; .endif
917 
918 #ifdef CONFIG_X86_64
919 
920 #define PV_SAVE_REGS(set)			\
921 	COND_PUSH(set, CLBR_RAX, rax);		\
922 	COND_PUSH(set, CLBR_RCX, rcx);		\
923 	COND_PUSH(set, CLBR_RDX, rdx);		\
924 	COND_PUSH(set, CLBR_RSI, rsi);		\
925 	COND_PUSH(set, CLBR_RDI, rdi);		\
926 	COND_PUSH(set, CLBR_R8, r8);		\
927 	COND_PUSH(set, CLBR_R9, r9);		\
928 	COND_PUSH(set, CLBR_R10, r10);		\
929 	COND_PUSH(set, CLBR_R11, r11)
930 #define PV_RESTORE_REGS(set)			\
931 	COND_POP(set, CLBR_R11, r11);		\
932 	COND_POP(set, CLBR_R10, r10);		\
933 	COND_POP(set, CLBR_R9, r9);		\
934 	COND_POP(set, CLBR_R8, r8);		\
935 	COND_POP(set, CLBR_RDI, rdi);		\
936 	COND_POP(set, CLBR_RSI, rsi);		\
937 	COND_POP(set, CLBR_RDX, rdx);		\
938 	COND_POP(set, CLBR_RCX, rcx);		\
939 	COND_POP(set, CLBR_RAX, rax)
940 
941 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 8)
942 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .quad, 8)
943 #define PARA_INDIRECT(addr)	*addr(%rip)
944 #else
945 #define PV_SAVE_REGS(set)			\
946 	COND_PUSH(set, CLBR_EAX, eax);		\
947 	COND_PUSH(set, CLBR_EDI, edi);		\
948 	COND_PUSH(set, CLBR_ECX, ecx);		\
949 	COND_PUSH(set, CLBR_EDX, edx)
950 #define PV_RESTORE_REGS(set)			\
951 	COND_POP(set, CLBR_EDX, edx);		\
952 	COND_POP(set, CLBR_ECX, ecx);		\
953 	COND_POP(set, CLBR_EDI, edi);		\
954 	COND_POP(set, CLBR_EAX, eax)
955 
956 #define PARA_PATCH(struct, off)        ((PARAVIRT_PATCH_##struct + (off)) / 4)
957 #define PARA_SITE(ptype, clobbers, ops) _PVSITE(ptype, clobbers, ops, .long, 4)
958 #define PARA_INDIRECT(addr)	*%cs:addr
959 #endif
960 
961 #define INTERRUPT_RETURN						\
962 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_iret), CLBR_NONE,	\
963 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_iret))
964 
965 #define DISABLE_INTERRUPTS(clobbers)					\
966 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_disable), clobbers, \
967 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
968 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_disable);	\
969 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
970 
971 #define ENABLE_INTERRUPTS(clobbers)					\
972 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_irq_enable), clobbers,	\
973 		  PV_SAVE_REGS(clobbers | CLBR_CALLEE_SAVE);		\
974 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_irq_enable);	\
975 		  PV_RESTORE_REGS(clobbers | CLBR_CALLEE_SAVE);)
976 
977 #define USERGS_SYSRET32							\
978 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret32),	\
979 		  CLBR_NONE,						\
980 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret32))
981 
982 #ifdef CONFIG_X86_32
983 #define GET_CR0_INTO_EAX				\
984 	push %ecx; push %edx;				\
985 	call PARA_INDIRECT(pv_cpu_ops+PV_CPU_read_cr0);	\
986 	pop %edx; pop %ecx
987 
988 #define ENABLE_INTERRUPTS_SYSEXIT					\
989 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),	\
990 		  CLBR_NONE,						\
991 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
992 
993 
994 #else	/* !CONFIG_X86_32 */
995 
996 /*
997  * If swapgs is used while the userspace stack is still current,
998  * there's no way to call a pvop.  The PV replacement *must* be
999  * inlined, or the swapgs instruction must be trapped and emulated.
1000  */
1001 #define SWAPGS_UNSAFE_STACK						\
1002 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
1003 		  swapgs)
1004 
1005 /*
1006  * Note: swapgs is very special, and in practise is either going to be
1007  * implemented with a single "swapgs" instruction or something very
1008  * special.  Either way, we don't need to save any registers for
1009  * it.
1010  */
1011 #define SWAPGS								\
1012 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_swapgs), CLBR_NONE,	\
1013 		  call PARA_INDIRECT(pv_cpu_ops+PV_CPU_swapgs)		\
1014 		 )
1015 
1016 #define GET_CR2_INTO_RCX				\
1017 	call PARA_INDIRECT(pv_mmu_ops+PV_MMU_read_cr2);	\
1018 	movq %rax, %rcx;				\
1019 	xorq %rax, %rax;
1020 
1021 #define PARAVIRT_ADJUST_EXCEPTION_FRAME					\
1022 	PARA_SITE(PARA_PATCH(pv_irq_ops, PV_IRQ_adjust_exception_frame), \
1023 		  CLBR_NONE,						\
1024 		  call PARA_INDIRECT(pv_irq_ops+PV_IRQ_adjust_exception_frame))
1025 
1026 #define USERGS_SYSRET64							\
1027 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_usergs_sysret64),	\
1028 		  CLBR_NONE,						\
1029 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_usergs_sysret64))
1030 
1031 #define ENABLE_INTERRUPTS_SYSEXIT32					\
1032 	PARA_SITE(PARA_PATCH(pv_cpu_ops, PV_CPU_irq_enable_sysexit),	\
1033 		  CLBR_NONE,						\
1034 		  jmp PARA_INDIRECT(pv_cpu_ops+PV_CPU_irq_enable_sysexit))
1035 #endif	/* CONFIG_X86_32 */
1036 
1037 #endif /* __ASSEMBLY__ */
1038 #else  /* CONFIG_PARAVIRT */
1039 # define default_banner x86_init_noop
1040 #endif /* !CONFIG_PARAVIRT */
1041 #endif /* _ASM_X86_PARAVIRT_H */
1042