1 #ifndef __ASM_PARAVIRT_H
2 #define __ASM_PARAVIRT_H
3 /* Various instructions on x86 need to be replaced for
4 * para-virtualization: those hooks are defined here. */
7 #include <linux/stringify.h>
10 /* These are the most performance critical ops, so we want to be able to patch
12 #define PARAVIRT_IRQ_DISABLE 0
13 #define PARAVIRT_IRQ_ENABLE 1
14 #define PARAVIRT_RESTORE_FLAGS 2
15 #define PARAVIRT_SAVE_FLAGS 3
16 #define PARAVIRT_SAVE_FLAGS_IRQ_DISABLE 4
17 #define PARAVIRT_INTERRUPT_RETURN 5
18 #define PARAVIRT_STI_SYSEXIT 6
20 /* Bitmask of what can be clobbered: usually at least eax. */
28 #include <linux/types.h>
31 struct Xgt_desc_struct;
37 unsigned int kernel_rpl;
38 int shared_kernel_pmd;
43 * Patch may replace one of the defined code sequences with arbitrary
44 * code, subject to the same register constraints. This generally
45 * means the code is not free to clobber any registers other than EAX.
46 * The patch function should return the number of bytes of code
47 * generated, as we nop pad the rest in generic code.
49 unsigned (*patch)(u8 type, u16 clobber, void *firstinsn, unsigned len);
51 void (*arch_setup)(void);
52 char *(*memory_setup)(void);
53 void (*init_IRQ)(void);
55 void (*pagetable_setup_start)(pgd_t *pgd_base);
56 void (*pagetable_setup_done)(pgd_t *pgd_base);
60 unsigned long (*get_wallclock)(void);
61 int (*set_wallclock)(unsigned long);
62 void (*time_init)(void);
64 void (*cpuid)(unsigned int *eax, unsigned int *ebx,
65 unsigned int *ecx, unsigned int *edx);
67 unsigned long (*get_debugreg)(int regno);
68 void (*set_debugreg)(int regno, unsigned long value);
72 unsigned long (*read_cr0)(void);
73 void (*write_cr0)(unsigned long);
75 unsigned long (*read_cr2)(void);
76 void (*write_cr2)(unsigned long);
78 unsigned long (*read_cr3)(void);
79 void (*write_cr3)(unsigned long);
81 unsigned long (*read_cr4_safe)(void);
82 unsigned long (*read_cr4)(void);
83 void (*write_cr4)(unsigned long);
85 unsigned long (*save_fl)(void);
86 void (*restore_fl)(unsigned long);
87 void (*irq_disable)(void);
88 void (*irq_enable)(void);
89 void (*safe_halt)(void);
93 /* err = 0/-EFAULT. wrmsr returns 0/-EFAULT. */
94 u64 (*read_msr)(unsigned int msr, int *err);
95 int (*write_msr)(unsigned int msr, u64 val);
97 u64 (*read_tsc)(void);
98 u64 (*read_pmc)(void);
99 u64 (*get_scheduled_cycles)(void);
100 unsigned long (*get_cpu_khz)(void);
102 void (*load_tr_desc)(void);
103 void (*load_gdt)(const struct Xgt_desc_struct *);
104 void (*load_idt)(const struct Xgt_desc_struct *);
105 void (*store_gdt)(struct Xgt_desc_struct *);
106 void (*store_idt)(struct Xgt_desc_struct *);
107 void (*set_ldt)(const void *desc, unsigned entries);
108 unsigned long (*store_tr)(void);
109 void (*load_tls)(struct thread_struct *t, unsigned int cpu);
110 void (*write_ldt_entry)(struct desc_struct *,
111 int entrynum, u32 low, u32 high);
112 void (*write_gdt_entry)(struct desc_struct *,
113 int entrynum, u32 low, u32 high);
114 void (*write_idt_entry)(struct desc_struct *,
115 int entrynum, u32 low, u32 high);
116 void (*load_esp0)(struct tss_struct *tss, struct thread_struct *t);
118 void (*set_iopl_mask)(unsigned mask);
120 void (*io_delay)(void);
122 #ifdef CONFIG_X86_LOCAL_APIC
123 void (*apic_write)(unsigned long reg, unsigned long v);
124 void (*apic_write_atomic)(unsigned long reg, unsigned long v);
125 unsigned long (*apic_read)(unsigned long reg);
126 void (*setup_boot_clock)(void);
127 void (*setup_secondary_clock)(void);
130 void (*flush_tlb_user)(void);
131 void (*flush_tlb_kernel)(void);
132 void (*flush_tlb_single)(u32 addr);
134 void (*map_pt_hook)(int type, pte_t *va, u32 pfn);
136 void (*alloc_pt)(u32 pfn);
137 void (*alloc_pd)(u32 pfn);
138 void (*alloc_pd_clone)(u32 pfn, u32 clonepfn, u32 start, u32 count);
139 void (*release_pt)(u32 pfn);
140 void (*release_pd)(u32 pfn);
142 void (*set_pte)(pte_t *ptep, pte_t pteval);
143 void (*set_pte_at)(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pteval);
144 void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval);
145 void (*pte_update)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
146 void (*pte_update_defer)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
148 pte_t (*ptep_get_and_clear)(pte_t *ptep);
150 #ifdef CONFIG_X86_PAE
151 void (*set_pte_atomic)(pte_t *ptep, pte_t pteval);
152 void (*set_pte_present)(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte);
153 void (*set_pud)(pud_t *pudp, pud_t pudval);
154 void (*pte_clear)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
155 void (*pmd_clear)(pmd_t *pmdp);
157 unsigned long long (*pte_val)(pte_t);
158 unsigned long long (*pmd_val)(pmd_t);
159 unsigned long long (*pgd_val)(pgd_t);
161 pte_t (*make_pte)(unsigned long long pte);
162 pmd_t (*make_pmd)(unsigned long long pmd);
163 pgd_t (*make_pgd)(unsigned long long pgd);
165 unsigned long (*pte_val)(pte_t);
166 unsigned long (*pgd_val)(pgd_t);
168 pte_t (*make_pte)(unsigned long pte);
169 pgd_t (*make_pgd)(unsigned long pgd);
172 void (*set_lazy_mode)(int mode);
174 /* These two are jmp to, not actually called. */
175 void (*irq_enable_sysexit)(void);
178 void (*startup_ipi_hook)(int phys_apicid, unsigned long start_eip, unsigned long start_esp);
181 /* Mark a paravirt probe function. */
182 #define paravirt_probe(fn) \
183 static asmlinkage void (*__paravirtprobe_##fn)(void) __attribute_used__ \
184 __attribute__((__section__(".paravirtprobe"))) = fn
186 extern struct paravirt_ops paravirt_ops;
188 #define paravirt_enabled() (paravirt_ops.paravirt_enabled)
190 static inline void load_esp0(struct tss_struct *tss,
191 struct thread_struct *thread)
193 paravirt_ops.load_esp0(tss, thread);
196 #define ARCH_SETUP paravirt_ops.arch_setup();
197 static inline unsigned long get_wallclock(void)
199 return paravirt_ops.get_wallclock();
202 static inline int set_wallclock(unsigned long nowtime)
204 return paravirt_ops.set_wallclock(nowtime);
207 static inline void (*choose_time_init(void))(void)
209 return paravirt_ops.time_init;
212 /* The paravirtualized CPUID instruction. */
213 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
214 unsigned int *ecx, unsigned int *edx)
216 paravirt_ops.cpuid(eax, ebx, ecx, edx);
220 * These special macros can be used to get or set a debugging register
222 #define get_debugreg(var, reg) var = paravirt_ops.get_debugreg(reg)
223 #define set_debugreg(val, reg) paravirt_ops.set_debugreg(reg, val)
225 #define clts() paravirt_ops.clts()
227 #define read_cr0() paravirt_ops.read_cr0()
228 #define write_cr0(x) paravirt_ops.write_cr0(x)
230 #define read_cr2() paravirt_ops.read_cr2()
231 #define write_cr2(x) paravirt_ops.write_cr2(x)
233 #define read_cr3() paravirt_ops.read_cr3()
234 #define write_cr3(x) paravirt_ops.write_cr3(x)
236 #define read_cr4() paravirt_ops.read_cr4()
237 #define read_cr4_safe(x) paravirt_ops.read_cr4_safe()
238 #define write_cr4(x) paravirt_ops.write_cr4(x)
240 #define raw_ptep_get_and_clear(xp) (paravirt_ops.ptep_get_and_clear(xp))
242 static inline void raw_safe_halt(void)
244 paravirt_ops.safe_halt();
247 static inline void halt(void)
249 paravirt_ops.safe_halt();
251 #define wbinvd() paravirt_ops.wbinvd()
253 #define get_kernel_rpl() (paravirt_ops.kernel_rpl)
255 /* These should all do BUG_ON(_err), but our headers are too tangled. */
256 #define rdmsr(msr,val1,val2) do { \
258 u64 _l = paravirt_ops.read_msr(msr,&_err); \
263 #define wrmsr(msr,val1,val2) do { \
264 u64 _l = ((u64)(val2) << 32) | (val1); \
265 paravirt_ops.write_msr((msr), _l); \
268 #define rdmsrl(msr,val) do { \
270 val = paravirt_ops.read_msr((msr),&_err); \
273 #define wrmsrl(msr,val) (paravirt_ops.write_msr((msr),(val)))
274 #define wrmsr_safe(msr,a,b) ({ \
275 u64 _l = ((u64)(b) << 32) | (a); \
276 paravirt_ops.write_msr((msr),_l); \
279 /* rdmsr with exception handling */
280 #define rdmsr_safe(msr,a,b) ({ \
282 u64 _l = paravirt_ops.read_msr(msr,&_err); \
287 #define rdtsc(low,high) do { \
288 u64 _l = paravirt_ops.read_tsc(); \
293 #define rdtscl(low) do { \
294 u64 _l = paravirt_ops.read_tsc(); \
298 #define rdtscll(val) (val = paravirt_ops.read_tsc())
300 #define get_scheduled_cycles(val) (val = paravirt_ops.get_scheduled_cycles())
301 #define calculate_cpu_khz() (paravirt_ops.get_cpu_khz())
303 #define write_tsc(val1,val2) wrmsr(0x10, val1, val2)
305 #define rdpmc(counter,low,high) do { \
306 u64 _l = paravirt_ops.read_pmc(); \
311 #define load_TR_desc() (paravirt_ops.load_tr_desc())
312 #define load_gdt(dtr) (paravirt_ops.load_gdt(dtr))
313 #define load_idt(dtr) (paravirt_ops.load_idt(dtr))
314 #define set_ldt(addr, entries) (paravirt_ops.set_ldt((addr), (entries)))
315 #define store_gdt(dtr) (paravirt_ops.store_gdt(dtr))
316 #define store_idt(dtr) (paravirt_ops.store_idt(dtr))
317 #define store_tr(tr) ((tr) = paravirt_ops.store_tr())
318 #define load_TLS(t,cpu) (paravirt_ops.load_tls((t),(cpu)))
319 #define write_ldt_entry(dt, entry, low, high) \
320 (paravirt_ops.write_ldt_entry((dt), (entry), (low), (high)))
321 #define write_gdt_entry(dt, entry, low, high) \
322 (paravirt_ops.write_gdt_entry((dt), (entry), (low), (high)))
323 #define write_idt_entry(dt, entry, low, high) \
324 (paravirt_ops.write_idt_entry((dt), (entry), (low), (high)))
325 #define set_iopl_mask(mask) (paravirt_ops.set_iopl_mask(mask))
327 #define __pte(x) paravirt_ops.make_pte(x)
328 #define __pgd(x) paravirt_ops.make_pgd(x)
330 #define pte_val(x) paravirt_ops.pte_val(x)
331 #define pgd_val(x) paravirt_ops.pgd_val(x)
333 #ifdef CONFIG_X86_PAE
334 #define __pmd(x) paravirt_ops.make_pmd(x)
335 #define pmd_val(x) paravirt_ops.pmd_val(x)
338 /* The paravirtualized I/O functions */
339 static inline void slow_down_io(void) {
340 paravirt_ops.io_delay();
341 #ifdef REALLY_SLOW_IO
342 paravirt_ops.io_delay();
343 paravirt_ops.io_delay();
344 paravirt_ops.io_delay();
348 #ifdef CONFIG_X86_LOCAL_APIC
350 * Basic functions accessing APICs.
352 static inline void apic_write(unsigned long reg, unsigned long v)
354 paravirt_ops.apic_write(reg,v);
357 static inline void apic_write_atomic(unsigned long reg, unsigned long v)
359 paravirt_ops.apic_write_atomic(reg,v);
362 static inline unsigned long apic_read(unsigned long reg)
364 return paravirt_ops.apic_read(reg);
367 static inline void setup_boot_clock(void)
369 paravirt_ops.setup_boot_clock();
372 static inline void setup_secondary_clock(void)
374 paravirt_ops.setup_secondary_clock();
378 static inline void paravirt_pagetable_setup_start(pgd_t *base)
380 if (paravirt_ops.pagetable_setup_start)
381 (*paravirt_ops.pagetable_setup_start)(base);
384 static inline void paravirt_pagetable_setup_done(pgd_t *base)
386 if (paravirt_ops.pagetable_setup_done)
387 (*paravirt_ops.pagetable_setup_done)(base);
391 static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
392 unsigned long start_esp)
394 return paravirt_ops.startup_ipi_hook(phys_apicid, start_eip, start_esp);
398 #define __flush_tlb() paravirt_ops.flush_tlb_user()
399 #define __flush_tlb_global() paravirt_ops.flush_tlb_kernel()
400 #define __flush_tlb_single(addr) paravirt_ops.flush_tlb_single(addr)
402 #define paravirt_map_pt_hook(type, va, pfn) paravirt_ops.map_pt_hook(type, va, pfn)
404 #define paravirt_alloc_pt(pfn) paravirt_ops.alloc_pt(pfn)
405 #define paravirt_release_pt(pfn) paravirt_ops.release_pt(pfn)
407 #define paravirt_alloc_pd(pfn) paravirt_ops.alloc_pd(pfn)
408 #define paravirt_alloc_pd_clone(pfn, clonepfn, start, count) \
409 paravirt_ops.alloc_pd_clone(pfn, clonepfn, start, count)
410 #define paravirt_release_pd(pfn) paravirt_ops.release_pd(pfn)
412 static inline void set_pte(pte_t *ptep, pte_t pteval)
414 paravirt_ops.set_pte(ptep, pteval);
417 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
418 pte_t *ptep, pte_t pteval)
420 paravirt_ops.set_pte_at(mm, addr, ptep, pteval);
423 static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
425 paravirt_ops.set_pmd(pmdp, pmdval);
428 static inline void pte_update(struct mm_struct *mm, u32 addr, pte_t *ptep)
430 paravirt_ops.pte_update(mm, addr, ptep);
433 static inline void pte_update_defer(struct mm_struct *mm, u32 addr, pte_t *ptep)
435 paravirt_ops.pte_update_defer(mm, addr, ptep);
438 #ifdef CONFIG_X86_PAE
439 static inline void set_pte_atomic(pte_t *ptep, pte_t pteval)
441 paravirt_ops.set_pte_atomic(ptep, pteval);
444 static inline void set_pte_present(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
446 paravirt_ops.set_pte_present(mm, addr, ptep, pte);
449 static inline void set_pud(pud_t *pudp, pud_t pudval)
451 paravirt_ops.set_pud(pudp, pudval);
454 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
456 paravirt_ops.pte_clear(mm, addr, ptep);
459 static inline void pmd_clear(pmd_t *pmdp)
461 paravirt_ops.pmd_clear(pmdp);
465 /* Lazy mode for batching updates / context switch */
466 #define PARAVIRT_LAZY_NONE 0
467 #define PARAVIRT_LAZY_MMU 1
468 #define PARAVIRT_LAZY_CPU 2
469 #define PARAVIRT_LAZY_FLUSH 3
471 #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
472 #define arch_enter_lazy_cpu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_CPU)
473 #define arch_leave_lazy_cpu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_NONE)
474 #define arch_flush_lazy_cpu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_FLUSH)
476 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
477 #define arch_enter_lazy_mmu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_MMU)
478 #define arch_leave_lazy_mmu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_NONE)
479 #define arch_flush_lazy_mmu_mode() paravirt_ops.set_lazy_mode(PARAVIRT_LAZY_FLUSH)
481 void _paravirt_nop(void);
482 #define paravirt_nop ((void *)_paravirt_nop)
484 /* These all sit in the .parainstructions section to tell us what to patch. */
485 struct paravirt_patch {
486 u8 *instr; /* original instructions */
487 u8 instrtype; /* type of this instruction */
488 u8 len; /* length of original instruction */
489 u16 clobbers; /* what registers you may clobber */
492 #define paravirt_alt(insn_string, typenum, clobber) \
493 "771:\n\t" insn_string "\n" "772:\n" \
494 ".pushsection .parainstructions,\"a\"\n" \
496 " .byte " __stringify(typenum) "\n" \
497 " .byte 772b-771b\n" \
498 " .short " __stringify(clobber) "\n" \
501 static inline unsigned long __raw_local_save_flags(void)
505 __asm__ __volatile__(paravirt_alt( "pushl %%ecx; pushl %%edx;"
507 "popl %%edx; popl %%ecx",
508 PARAVIRT_SAVE_FLAGS, CLBR_NONE)
509 : "=a"(f): "m"(paravirt_ops.save_fl)
514 static inline void raw_local_irq_restore(unsigned long f)
516 __asm__ __volatile__(paravirt_alt( "pushl %%ecx; pushl %%edx;"
518 "popl %%edx; popl %%ecx",
519 PARAVIRT_RESTORE_FLAGS, CLBR_EAX)
520 : "=a"(f) : "m" (paravirt_ops.restore_fl), "0"(f)
524 static inline void raw_local_irq_disable(void)
526 __asm__ __volatile__(paravirt_alt( "pushl %%ecx; pushl %%edx;"
528 "popl %%edx; popl %%ecx",
529 PARAVIRT_IRQ_DISABLE, CLBR_EAX)
530 : : "m" (paravirt_ops.irq_disable)
531 : "memory", "eax", "cc");
534 static inline void raw_local_irq_enable(void)
536 __asm__ __volatile__(paravirt_alt( "pushl %%ecx; pushl %%edx;"
538 "popl %%edx; popl %%ecx",
539 PARAVIRT_IRQ_ENABLE, CLBR_EAX)
540 : : "m" (paravirt_ops.irq_enable)
541 : "memory", "eax", "cc");
544 static inline unsigned long __raw_local_irq_save(void)
548 __asm__ __volatile__(paravirt_alt( "pushl %%ecx; pushl %%edx;"
549 "call *%1; pushl %%eax;"
550 "call *%2; popl %%eax;"
551 "popl %%edx; popl %%ecx",
552 PARAVIRT_SAVE_FLAGS_IRQ_DISABLE,
555 : "m" (paravirt_ops.save_fl),
556 "m" (paravirt_ops.irq_disable)
561 #define CLI_STRING paravirt_alt("pushl %%ecx; pushl %%edx;" \
562 "call *paravirt_ops+%c[irq_disable];" \
563 "popl %%edx; popl %%ecx", \
564 PARAVIRT_IRQ_DISABLE, CLBR_EAX)
566 #define STI_STRING paravirt_alt("pushl %%ecx; pushl %%edx;" \
567 "call *paravirt_ops+%c[irq_enable];" \
568 "popl %%edx; popl %%ecx", \
569 PARAVIRT_IRQ_ENABLE, CLBR_EAX)
570 #define CLI_STI_CLOBBERS , "%eax"
571 #define CLI_STI_INPUT_ARGS \
573 [irq_disable] "i" (offsetof(struct paravirt_ops, irq_disable)), \
574 [irq_enable] "i" (offsetof(struct paravirt_ops, irq_enable))
576 #else /* __ASSEMBLY__ */
578 #define PARA_PATCH(ptype, clobbers, ops) \
582 .pushsection .parainstructions,"a"; \
589 #define INTERRUPT_RETURN \
590 PARA_PATCH(PARAVIRT_INTERRUPT_RETURN, CLBR_ANY, \
591 jmp *%cs:paravirt_ops+PARAVIRT_iret)
593 #define DISABLE_INTERRUPTS(clobbers) \
594 PARA_PATCH(PARAVIRT_IRQ_DISABLE, clobbers, \
595 pushl %ecx; pushl %edx; \
596 call *paravirt_ops+PARAVIRT_irq_disable; \
597 popl %edx; popl %ecx) \
599 #define ENABLE_INTERRUPTS(clobbers) \
600 PARA_PATCH(PARAVIRT_IRQ_ENABLE, clobbers, \
601 pushl %ecx; pushl %edx; \
602 call *%cs:paravirt_ops+PARAVIRT_irq_enable; \
603 popl %edx; popl %ecx)
605 #define ENABLE_INTERRUPTS_SYSEXIT \
606 PARA_PATCH(PARAVIRT_STI_SYSEXIT, CLBR_ANY, \
607 jmp *%cs:paravirt_ops+PARAVIRT_irq_enable_sysexit)
609 #define GET_CR0_INTO_EAX \
610 call *paravirt_ops+PARAVIRT_read_cr0
612 #endif /* __ASSEMBLY__ */
613 #endif /* CONFIG_PARAVIRT */
614 #endif /* __ASM_PARAVIRT_H */