paravirt.h 29 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079
  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. */
  5. #ifdef CONFIG_PARAVIRT
  6. #include <asm/page.h>
  7. /* Bitmask of what can be clobbered: usually at least eax. */
  8. #define CLBR_NONE 0x0
  9. #define CLBR_EAX 0x1
  10. #define CLBR_ECX 0x2
  11. #define CLBR_EDX 0x4
  12. #define CLBR_ANY 0x7
  13. #ifndef __ASSEMBLY__
  14. #include <linux/types.h>
  15. #include <linux/cpumask.h>
  16. #include <asm/kmap_types.h>
  17. struct page;
  18. struct thread_struct;
  19. struct Xgt_desc_struct;
  20. struct tss_struct;
  21. struct mm_struct;
  22. struct desc_struct;
  23. /* Lazy mode for batching updates / context switch */
  24. enum paravirt_lazy_mode {
  25. PARAVIRT_LAZY_NONE = 0,
  26. PARAVIRT_LAZY_MMU = 1,
  27. PARAVIRT_LAZY_CPU = 2,
  28. PARAVIRT_LAZY_FLUSH = 3,
  29. };
  30. struct paravirt_ops
  31. {
  32. unsigned int kernel_rpl;
  33. int shared_kernel_pmd;
  34. int paravirt_enabled;
  35. const char *name;
  36. /*
  37. * Patch may replace one of the defined code sequences with arbitrary
  38. * code, subject to the same register constraints. This generally
  39. * means the code is not free to clobber any registers other than EAX.
  40. * The patch function should return the number of bytes of code
  41. * generated, as we nop pad the rest in generic code.
  42. */
  43. unsigned (*patch)(u8 type, u16 clobber, void *firstinsn, unsigned len);
  44. /* Basic arch-specific setup */
  45. void (*arch_setup)(void);
  46. char *(*memory_setup)(void);
  47. void (*init_IRQ)(void);
  48. void (*time_init)(void);
  49. /*
  50. * Called before/after init_mm pagetable setup. setup_start
  51. * may reset %cr3, and may pre-install parts of the pagetable;
  52. * pagetable setup is expected to preserve any existing
  53. * mapping.
  54. */
  55. void (*pagetable_setup_start)(pgd_t *pgd_base);
  56. void (*pagetable_setup_done)(pgd_t *pgd_base);
  57. /* Print a banner to identify the environment */
  58. void (*banner)(void);
  59. /* Set and set time of day */
  60. unsigned long (*get_wallclock)(void);
  61. int (*set_wallclock)(unsigned long);
  62. /* cpuid emulation, mostly so that caps bits can be disabled */
  63. void (*cpuid)(unsigned int *eax, unsigned int *ebx,
  64. unsigned int *ecx, unsigned int *edx);
  65. /* hooks for various privileged instructions */
  66. unsigned long (*get_debugreg)(int regno);
  67. void (*set_debugreg)(int regno, unsigned long value);
  68. void (*clts)(void);
  69. unsigned long (*read_cr0)(void);
  70. void (*write_cr0)(unsigned long);
  71. unsigned long (*read_cr2)(void);
  72. void (*write_cr2)(unsigned long);
  73. unsigned long (*read_cr3)(void);
  74. void (*write_cr3)(unsigned long);
  75. unsigned long (*read_cr4_safe)(void);
  76. unsigned long (*read_cr4)(void);
  77. void (*write_cr4)(unsigned long);
  78. /*
  79. * Get/set interrupt state. save_fl and restore_fl are only
  80. * expected to use X86_EFLAGS_IF; all other bits
  81. * returned from save_fl are undefined, and may be ignored by
  82. * restore_fl.
  83. */
  84. unsigned long (*save_fl)(void);
  85. void (*restore_fl)(unsigned long);
  86. void (*irq_disable)(void);
  87. void (*irq_enable)(void);
  88. void (*safe_halt)(void);
  89. void (*halt)(void);
  90. void (*wbinvd)(void);
  91. /* MSR, PMC and TSR operations.
  92. err = 0/-EFAULT. wrmsr returns 0/-EFAULT. */
  93. u64 (*read_msr)(unsigned int msr, int *err);
  94. int (*write_msr)(unsigned int msr, u64 val);
  95. u64 (*read_tsc)(void);
  96. u64 (*read_pmc)(void);
  97. u64 (*get_scheduled_cycles)(void);
  98. unsigned long (*get_cpu_khz)(void);
  99. /* Segment descriptor handling */
  100. void (*load_tr_desc)(void);
  101. void (*load_gdt)(const struct Xgt_desc_struct *);
  102. void (*load_idt)(const struct Xgt_desc_struct *);
  103. void (*store_gdt)(struct Xgt_desc_struct *);
  104. void (*store_idt)(struct Xgt_desc_struct *);
  105. void (*set_ldt)(const void *desc, unsigned entries);
  106. unsigned long (*store_tr)(void);
  107. void (*load_tls)(struct thread_struct *t, unsigned int cpu);
  108. void (*write_ldt_entry)(struct desc_struct *,
  109. int entrynum, u32 low, u32 high);
  110. void (*write_gdt_entry)(struct desc_struct *,
  111. int entrynum, u32 low, u32 high);
  112. void (*write_idt_entry)(struct desc_struct *,
  113. int entrynum, u32 low, u32 high);
  114. void (*load_esp0)(struct tss_struct *tss, struct thread_struct *t);
  115. void (*set_iopl_mask)(unsigned mask);
  116. void (*io_delay)(void);
  117. /*
  118. * Hooks for intercepting the creation/use/destruction of an
  119. * mm_struct.
  120. */
  121. void (*activate_mm)(struct mm_struct *prev,
  122. struct mm_struct *next);
  123. void (*dup_mmap)(struct mm_struct *oldmm,
  124. struct mm_struct *mm);
  125. void (*exit_mmap)(struct mm_struct *mm);
  126. #ifdef CONFIG_X86_LOCAL_APIC
  127. /*
  128. * Direct APIC operations, principally for VMI. Ideally
  129. * these shouldn't be in this interface.
  130. */
  131. void (*apic_write)(unsigned long reg, unsigned long v);
  132. void (*apic_write_atomic)(unsigned long reg, unsigned long v);
  133. unsigned long (*apic_read)(unsigned long reg);
  134. void (*setup_boot_clock)(void);
  135. void (*setup_secondary_clock)(void);
  136. void (*startup_ipi_hook)(int phys_apicid,
  137. unsigned long start_eip,
  138. unsigned long start_esp);
  139. #endif
  140. /* TLB operations */
  141. void (*flush_tlb_user)(void);
  142. void (*flush_tlb_kernel)(void);
  143. void (*flush_tlb_single)(unsigned long addr);
  144. void (*flush_tlb_others)(const cpumask_t *cpus, struct mm_struct *mm,
  145. unsigned long va);
  146. /* Hooks for allocating/releasing pagetable pages */
  147. void (*alloc_pt)(u32 pfn);
  148. void (*alloc_pd)(u32 pfn);
  149. void (*alloc_pd_clone)(u32 pfn, u32 clonepfn, u32 start, u32 count);
  150. void (*release_pt)(u32 pfn);
  151. void (*release_pd)(u32 pfn);
  152. /* Pagetable manipulation functions */
  153. void (*set_pte)(pte_t *ptep, pte_t pteval);
  154. void (*set_pte_at)(struct mm_struct *mm, unsigned long addr,
  155. pte_t *ptep, pte_t pteval);
  156. void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval);
  157. void (*pte_update)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
  158. void (*pte_update_defer)(struct mm_struct *mm,
  159. unsigned long addr, pte_t *ptep);
  160. #ifdef CONFIG_HIGHPTE
  161. void *(*kmap_atomic_pte)(struct page *page, enum km_type type);
  162. #endif
  163. #ifdef CONFIG_X86_PAE
  164. void (*set_pte_atomic)(pte_t *ptep, pte_t pteval);
  165. void (*set_pte_present)(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte);
  166. void (*set_pud)(pud_t *pudp, pud_t pudval);
  167. void (*pte_clear)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
  168. void (*pmd_clear)(pmd_t *pmdp);
  169. unsigned long long (*pte_val)(pte_t);
  170. unsigned long long (*pmd_val)(pmd_t);
  171. unsigned long long (*pgd_val)(pgd_t);
  172. pte_t (*make_pte)(unsigned long long pte);
  173. pmd_t (*make_pmd)(unsigned long long pmd);
  174. pgd_t (*make_pgd)(unsigned long long pgd);
  175. #else
  176. unsigned long (*pte_val)(pte_t);
  177. unsigned long (*pgd_val)(pgd_t);
  178. pte_t (*make_pte)(unsigned long pte);
  179. pgd_t (*make_pgd)(unsigned long pgd);
  180. #endif
  181. /* Set deferred update mode, used for batching operations. */
  182. void (*set_lazy_mode)(enum paravirt_lazy_mode mode);
  183. /* These two are jmp to, not actually called. */
  184. void (*irq_enable_sysexit)(void);
  185. void (*iret)(void);
  186. };
  187. /* Mark a paravirt probe function. */
  188. #define paravirt_probe(fn) \
  189. static asmlinkage void (*__paravirtprobe_##fn)(void) __attribute_used__ \
  190. __attribute__((__section__(".paravirtprobe"))) = fn
  191. extern struct paravirt_ops paravirt_ops;
  192. #define PARAVIRT_PATCH(x) \
  193. (offsetof(struct paravirt_ops, x) / sizeof(void *))
  194. #define paravirt_type(type) \
  195. [paravirt_typenum] "i" (PARAVIRT_PATCH(type))
  196. #define paravirt_clobber(clobber) \
  197. [paravirt_clobber] "i" (clobber)
  198. /*
  199. * Generate some code, and mark it as patchable by the
  200. * apply_paravirt() alternate instruction patcher.
  201. */
  202. #define _paravirt_alt(insn_string, type, clobber) \
  203. "771:\n\t" insn_string "\n" "772:\n" \
  204. ".pushsection .parainstructions,\"a\"\n" \
  205. " .long 771b\n" \
  206. " .byte " type "\n" \
  207. " .byte 772b-771b\n" \
  208. " .short " clobber "\n" \
  209. ".popsection\n"
  210. /* Generate patchable code, with the default asm parameters. */
  211. #define paravirt_alt(insn_string) \
  212. _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
  213. unsigned paravirt_patch_nop(void);
  214. unsigned paravirt_patch_ignore(unsigned len);
  215. unsigned paravirt_patch_call(void *target, u16 tgt_clobbers,
  216. void *site, u16 site_clobbers,
  217. unsigned len);
  218. unsigned paravirt_patch_jmp(void *target, void *site, unsigned len);
  219. unsigned paravirt_patch_default(u8 type, u16 clobbers, void *site, unsigned len);
  220. unsigned paravirt_patch_insns(void *site, unsigned len,
  221. const char *start, const char *end);
  222. /*
  223. * This generates an indirect call based on the operation type number.
  224. * The type number, computed in PARAVIRT_PATCH, is derived from the
  225. * offset into the paravirt_ops structure, and can therefore be freely
  226. * converted back into a structure offset.
  227. */
  228. #define PARAVIRT_CALL "call *(paravirt_ops+%c[paravirt_typenum]*4);"
  229. /*
  230. * These macros are intended to wrap calls into a paravirt_ops
  231. * operation, so that they can be later identified and patched at
  232. * runtime.
  233. *
  234. * Normally, a call to a pv_op function is a simple indirect call:
  235. * (paravirt_ops.operations)(args...).
  236. *
  237. * Unfortunately, this is a relatively slow operation for modern CPUs,
  238. * because it cannot necessarily determine what the destination
  239. * address is. In this case, the address is a runtime constant, so at
  240. * the very least we can patch the call to e a simple direct call, or
  241. * ideally, patch an inline implementation into the callsite. (Direct
  242. * calls are essentially free, because the call and return addresses
  243. * are completely predictable.)
  244. *
  245. * These macros rely on the standard gcc "regparm(3)" calling
  246. * convention, in which the first three arguments are placed in %eax,
  247. * %edx, %ecx (in that order), and the remaining arguments are placed
  248. * on the stack. All caller-save registers (eax,edx,ecx) are expected
  249. * to be modified (either clobbered or used for return values).
  250. *
  251. * The call instruction itself is marked by placing its start address
  252. * and size into the .parainstructions section, so that
  253. * apply_paravirt() in arch/i386/kernel/alternative.c can do the
  254. * appropriate patching under the control of the backend paravirt_ops
  255. * implementation.
  256. *
  257. * Unfortunately there's no way to get gcc to generate the args setup
  258. * for the call, and then allow the call itself to be generated by an
  259. * inline asm. Because of this, we must do the complete arg setup and
  260. * return value handling from within these macros. This is fairly
  261. * cumbersome.
  262. *
  263. * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
  264. * It could be extended to more arguments, but there would be little
  265. * to be gained from that. For each number of arguments, there are
  266. * the two VCALL and CALL variants for void and non-void functions.
  267. *
  268. * When there is a return value, the invoker of the macro must specify
  269. * the return type. The macro then uses sizeof() on that type to
  270. * determine whether its a 32 or 64 bit value, and places the return
  271. * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
  272. * 64-bit).
  273. *
  274. * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
  275. * in low,high order.
  276. *
  277. * Small structures are passed and returned in registers. The macro
  278. * calling convention can't directly deal with this, so the wrapper
  279. * functions must do this.
  280. *
  281. * These PVOP_* macros are only defined within this header. This
  282. * means that all uses must be wrapped in inline functions. This also
  283. * makes sure the incoming and outgoing types are always correct.
  284. */
  285. #define __PVOP_CALL(rettype, op, pre, post, ...) \
  286. ({ \
  287. rettype __ret; \
  288. unsigned long __eax, __edx, __ecx; \
  289. if (sizeof(rettype) > sizeof(unsigned long)) { \
  290. asm volatile(pre \
  291. paravirt_alt(PARAVIRT_CALL) \
  292. post \
  293. : "=a" (__eax), "=d" (__edx), \
  294. "=c" (__ecx) \
  295. : paravirt_type(op), \
  296. paravirt_clobber(CLBR_ANY), \
  297. ##__VA_ARGS__ \
  298. : "memory", "cc"); \
  299. __ret = (rettype)((((u64)__edx) << 32) | __eax); \
  300. } else { \
  301. asm volatile(pre \
  302. paravirt_alt(PARAVIRT_CALL) \
  303. post \
  304. : "=a" (__eax), "=d" (__edx), \
  305. "=c" (__ecx) \
  306. : paravirt_type(op), \
  307. paravirt_clobber(CLBR_ANY), \
  308. ##__VA_ARGS__ \
  309. : "memory", "cc"); \
  310. __ret = (rettype)__eax; \
  311. } \
  312. __ret; \
  313. })
  314. #define __PVOP_VCALL(op, pre, post, ...) \
  315. ({ \
  316. unsigned long __eax, __edx, __ecx; \
  317. asm volatile(pre \
  318. paravirt_alt(PARAVIRT_CALL) \
  319. post \
  320. : "=a" (__eax), "=d" (__edx), "=c" (__ecx) \
  321. : paravirt_type(op), \
  322. paravirt_clobber(CLBR_ANY), \
  323. ##__VA_ARGS__ \
  324. : "memory", "cc"); \
  325. })
  326. #define PVOP_CALL0(rettype, op) \
  327. __PVOP_CALL(rettype, op, "", "")
  328. #define PVOP_VCALL0(op) \
  329. __PVOP_VCALL(op, "", "")
  330. #define PVOP_CALL1(rettype, op, arg1) \
  331. __PVOP_CALL(rettype, op, "", "", "0" ((u32)(arg1)))
  332. #define PVOP_VCALL1(op, arg1) \
  333. __PVOP_VCALL(op, "", "", "0" ((u32)(arg1)))
  334. #define PVOP_CALL2(rettype, op, arg1, arg2) \
  335. __PVOP_CALL(rettype, op, "", "", "0" ((u32)(arg1)), "1" ((u32)(arg2)))
  336. #define PVOP_VCALL2(op, arg1, arg2) \
  337. __PVOP_VCALL(op, "", "", "0" ((u32)(arg1)), "1" ((u32)(arg2)))
  338. #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \
  339. __PVOP_CALL(rettype, op, "", "", "0" ((u32)(arg1)), \
  340. "1"((u32)(arg2)), "2"((u32)(arg3)))
  341. #define PVOP_VCALL3(op, arg1, arg2, arg3) \
  342. __PVOP_VCALL(op, "", "", "0" ((u32)(arg1)), "1"((u32)(arg2)), \
  343. "2"((u32)(arg3)))
  344. #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
  345. __PVOP_CALL(rettype, op, \
  346. "push %[_arg4];", "lea 4(%%esp),%%esp;", \
  347. "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
  348. "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
  349. #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
  350. __PVOP_VCALL(op, \
  351. "push %[_arg4];", "lea 4(%%esp),%%esp;", \
  352. "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
  353. "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
  354. static inline int paravirt_enabled(void)
  355. {
  356. return paravirt_ops.paravirt_enabled;
  357. }
  358. static inline void load_esp0(struct tss_struct *tss,
  359. struct thread_struct *thread)
  360. {
  361. PVOP_VCALL2(load_esp0, tss, thread);
  362. }
  363. #define ARCH_SETUP paravirt_ops.arch_setup();
  364. static inline unsigned long get_wallclock(void)
  365. {
  366. return PVOP_CALL0(unsigned long, get_wallclock);
  367. }
  368. static inline int set_wallclock(unsigned long nowtime)
  369. {
  370. return PVOP_CALL1(int, set_wallclock, nowtime);
  371. }
  372. static inline void (*choose_time_init(void))(void)
  373. {
  374. return paravirt_ops.time_init;
  375. }
  376. /* The paravirtualized CPUID instruction. */
  377. static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
  378. unsigned int *ecx, unsigned int *edx)
  379. {
  380. PVOP_VCALL4(cpuid, eax, ebx, ecx, edx);
  381. }
  382. /*
  383. * These special macros can be used to get or set a debugging register
  384. */
  385. static inline unsigned long paravirt_get_debugreg(int reg)
  386. {
  387. return PVOP_CALL1(unsigned long, get_debugreg, reg);
  388. }
  389. #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
  390. static inline void set_debugreg(unsigned long val, int reg)
  391. {
  392. PVOP_VCALL2(set_debugreg, reg, val);
  393. }
  394. static inline void clts(void)
  395. {
  396. PVOP_VCALL0(clts);
  397. }
  398. static inline unsigned long read_cr0(void)
  399. {
  400. return PVOP_CALL0(unsigned long, read_cr0);
  401. }
  402. static inline void write_cr0(unsigned long x)
  403. {
  404. PVOP_VCALL1(write_cr0, x);
  405. }
  406. static inline unsigned long read_cr2(void)
  407. {
  408. return PVOP_CALL0(unsigned long, read_cr2);
  409. }
  410. static inline void write_cr2(unsigned long x)
  411. {
  412. PVOP_VCALL1(write_cr2, x);
  413. }
  414. static inline unsigned long read_cr3(void)
  415. {
  416. return PVOP_CALL0(unsigned long, read_cr3);
  417. }
  418. static inline void write_cr3(unsigned long x)
  419. {
  420. PVOP_VCALL1(write_cr3, x);
  421. }
  422. static inline unsigned long read_cr4(void)
  423. {
  424. return PVOP_CALL0(unsigned long, read_cr4);
  425. }
  426. static inline unsigned long read_cr4_safe(void)
  427. {
  428. return PVOP_CALL0(unsigned long, read_cr4_safe);
  429. }
  430. static inline void write_cr4(unsigned long x)
  431. {
  432. PVOP_VCALL1(write_cr4, x);
  433. }
  434. static inline void raw_safe_halt(void)
  435. {
  436. PVOP_VCALL0(safe_halt);
  437. }
  438. static inline void halt(void)
  439. {
  440. PVOP_VCALL0(safe_halt);
  441. }
  442. static inline void wbinvd(void)
  443. {
  444. PVOP_VCALL0(wbinvd);
  445. }
  446. #define get_kernel_rpl() (paravirt_ops.kernel_rpl)
  447. static inline u64 paravirt_read_msr(unsigned msr, int *err)
  448. {
  449. return PVOP_CALL2(u64, read_msr, msr, err);
  450. }
  451. static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
  452. {
  453. return PVOP_CALL3(int, write_msr, msr, low, high);
  454. }
  455. /* These should all do BUG_ON(_err), but our headers are too tangled. */
  456. #define rdmsr(msr,val1,val2) do { \
  457. int _err; \
  458. u64 _l = paravirt_read_msr(msr, &_err); \
  459. val1 = (u32)_l; \
  460. val2 = _l >> 32; \
  461. } while(0)
  462. #define wrmsr(msr,val1,val2) do { \
  463. paravirt_write_msr(msr, val1, val2); \
  464. } while(0)
  465. #define rdmsrl(msr,val) do { \
  466. int _err; \
  467. val = paravirt_read_msr(msr, &_err); \
  468. } while(0)
  469. #define wrmsrl(msr,val) ((void)paravirt_write_msr(msr, val, 0))
  470. #define wrmsr_safe(msr,a,b) paravirt_write_msr(msr, a, b)
  471. /* rdmsr with exception handling */
  472. #define rdmsr_safe(msr,a,b) ({ \
  473. int _err; \
  474. u64 _l = paravirt_read_msr(msr, &_err); \
  475. (*a) = (u32)_l; \
  476. (*b) = _l >> 32; \
  477. _err; })
  478. static inline u64 paravirt_read_tsc(void)
  479. {
  480. return PVOP_CALL0(u64, read_tsc);
  481. }
  482. #define rdtsc(low,high) do { \
  483. u64 _l = paravirt_read_tsc(); \
  484. low = (u32)_l; \
  485. high = _l >> 32; \
  486. } while(0)
  487. #define rdtscl(low) do { \
  488. u64 _l = paravirt_read_tsc(); \
  489. low = (int)_l; \
  490. } while(0)
  491. #define rdtscll(val) (val = paravirt_read_tsc())
  492. #define get_scheduled_cycles(val) (val = paravirt_ops.get_scheduled_cycles())
  493. #define calculate_cpu_khz() (paravirt_ops.get_cpu_khz())
  494. #define write_tsc(val1,val2) wrmsr(0x10, val1, val2)
  495. static inline unsigned long long paravirt_read_pmc(int counter)
  496. {
  497. return PVOP_CALL1(u64, read_pmc, counter);
  498. }
  499. #define rdpmc(counter,low,high) do { \
  500. u64 _l = paravirt_read_pmc(counter); \
  501. low = (u32)_l; \
  502. high = _l >> 32; \
  503. } while(0)
  504. static inline void load_TR_desc(void)
  505. {
  506. PVOP_VCALL0(load_tr_desc);
  507. }
  508. static inline void load_gdt(const struct Xgt_desc_struct *dtr)
  509. {
  510. PVOP_VCALL1(load_gdt, dtr);
  511. }
  512. static inline void load_idt(const struct Xgt_desc_struct *dtr)
  513. {
  514. PVOP_VCALL1(load_idt, dtr);
  515. }
  516. static inline void set_ldt(const void *addr, unsigned entries)
  517. {
  518. PVOP_VCALL2(set_ldt, addr, entries);
  519. }
  520. static inline void store_gdt(struct Xgt_desc_struct *dtr)
  521. {
  522. PVOP_VCALL1(store_gdt, dtr);
  523. }
  524. static inline void store_idt(struct Xgt_desc_struct *dtr)
  525. {
  526. PVOP_VCALL1(store_idt, dtr);
  527. }
  528. static inline unsigned long paravirt_store_tr(void)
  529. {
  530. return PVOP_CALL0(unsigned long, store_tr);
  531. }
  532. #define store_tr(tr) ((tr) = paravirt_store_tr())
  533. static inline void load_TLS(struct thread_struct *t, unsigned cpu)
  534. {
  535. PVOP_VCALL2(load_tls, t, cpu);
  536. }
  537. static inline void write_ldt_entry(void *dt, int entry, u32 low, u32 high)
  538. {
  539. PVOP_VCALL4(write_ldt_entry, dt, entry, low, high);
  540. }
  541. static inline void write_gdt_entry(void *dt, int entry, u32 low, u32 high)
  542. {
  543. PVOP_VCALL4(write_gdt_entry, dt, entry, low, high);
  544. }
  545. static inline void write_idt_entry(void *dt, int entry, u32 low, u32 high)
  546. {
  547. PVOP_VCALL4(write_idt_entry, dt, entry, low, high);
  548. }
  549. static inline void set_iopl_mask(unsigned mask)
  550. {
  551. PVOP_VCALL1(set_iopl_mask, mask);
  552. }
  553. /* The paravirtualized I/O functions */
  554. static inline void slow_down_io(void) {
  555. paravirt_ops.io_delay();
  556. #ifdef REALLY_SLOW_IO
  557. paravirt_ops.io_delay();
  558. paravirt_ops.io_delay();
  559. paravirt_ops.io_delay();
  560. #endif
  561. }
  562. #ifdef CONFIG_X86_LOCAL_APIC
  563. /*
  564. * Basic functions accessing APICs.
  565. */
  566. static inline void apic_write(unsigned long reg, unsigned long v)
  567. {
  568. PVOP_VCALL2(apic_write, reg, v);
  569. }
  570. static inline void apic_write_atomic(unsigned long reg, unsigned long v)
  571. {
  572. PVOP_VCALL2(apic_write_atomic, reg, v);
  573. }
  574. static inline unsigned long apic_read(unsigned long reg)
  575. {
  576. return PVOP_CALL1(unsigned long, apic_read, reg);
  577. }
  578. static inline void setup_boot_clock(void)
  579. {
  580. PVOP_VCALL0(setup_boot_clock);
  581. }
  582. static inline void setup_secondary_clock(void)
  583. {
  584. PVOP_VCALL0(setup_secondary_clock);
  585. }
  586. #endif
  587. static inline void paravirt_pagetable_setup_start(pgd_t *base)
  588. {
  589. if (paravirt_ops.pagetable_setup_start)
  590. (*paravirt_ops.pagetable_setup_start)(base);
  591. }
  592. static inline void paravirt_pagetable_setup_done(pgd_t *base)
  593. {
  594. if (paravirt_ops.pagetable_setup_done)
  595. (*paravirt_ops.pagetable_setup_done)(base);
  596. }
  597. #ifdef CONFIG_SMP
  598. static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
  599. unsigned long start_esp)
  600. {
  601. PVOP_VCALL3(startup_ipi_hook, phys_apicid, start_eip, start_esp);
  602. }
  603. #endif
  604. static inline void paravirt_activate_mm(struct mm_struct *prev,
  605. struct mm_struct *next)
  606. {
  607. PVOP_VCALL2(activate_mm, prev, next);
  608. }
  609. static inline void arch_dup_mmap(struct mm_struct *oldmm,
  610. struct mm_struct *mm)
  611. {
  612. PVOP_VCALL2(dup_mmap, oldmm, mm);
  613. }
  614. static inline void arch_exit_mmap(struct mm_struct *mm)
  615. {
  616. PVOP_VCALL1(exit_mmap, mm);
  617. }
  618. static inline void __flush_tlb(void)
  619. {
  620. PVOP_VCALL0(flush_tlb_user);
  621. }
  622. static inline void __flush_tlb_global(void)
  623. {
  624. PVOP_VCALL0(flush_tlb_kernel);
  625. }
  626. static inline void __flush_tlb_single(unsigned long addr)
  627. {
  628. PVOP_VCALL1(flush_tlb_single, addr);
  629. }
  630. static inline void flush_tlb_others(cpumask_t cpumask, struct mm_struct *mm,
  631. unsigned long va)
  632. {
  633. PVOP_VCALL3(flush_tlb_others, &cpumask, mm, va);
  634. }
  635. static inline void paravirt_alloc_pt(unsigned pfn)
  636. {
  637. PVOP_VCALL1(alloc_pt, pfn);
  638. }
  639. static inline void paravirt_release_pt(unsigned pfn)
  640. {
  641. PVOP_VCALL1(release_pt, pfn);
  642. }
  643. static inline void paravirt_alloc_pd(unsigned pfn)
  644. {
  645. PVOP_VCALL1(alloc_pd, pfn);
  646. }
  647. static inline void paravirt_alloc_pd_clone(unsigned pfn, unsigned clonepfn,
  648. unsigned start, unsigned count)
  649. {
  650. PVOP_VCALL4(alloc_pd_clone, pfn, clonepfn, start, count);
  651. }
  652. static inline void paravirt_release_pd(unsigned pfn)
  653. {
  654. PVOP_VCALL1(release_pd, pfn);
  655. }
  656. #ifdef CONFIG_HIGHPTE
  657. static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
  658. {
  659. unsigned long ret;
  660. ret = PVOP_CALL2(unsigned long, kmap_atomic_pte, page, type);
  661. return (void *)ret;
  662. }
  663. #endif
  664. static inline void pte_update(struct mm_struct *mm, unsigned long addr,
  665. pte_t *ptep)
  666. {
  667. PVOP_VCALL3(pte_update, mm, addr, ptep);
  668. }
  669. static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
  670. pte_t *ptep)
  671. {
  672. PVOP_VCALL3(pte_update_defer, mm, addr, ptep);
  673. }
  674. #ifdef CONFIG_X86_PAE
  675. static inline pte_t __pte(unsigned long long val)
  676. {
  677. unsigned long long ret = PVOP_CALL2(unsigned long long, make_pte,
  678. val, val >> 32);
  679. return (pte_t) { ret, ret >> 32 };
  680. }
  681. static inline pmd_t __pmd(unsigned long long val)
  682. {
  683. return (pmd_t) { PVOP_CALL2(unsigned long long, make_pmd, val, val >> 32) };
  684. }
  685. static inline pgd_t __pgd(unsigned long long val)
  686. {
  687. return (pgd_t) { PVOP_CALL2(unsigned long long, make_pgd, val, val >> 32) };
  688. }
  689. static inline unsigned long long pte_val(pte_t x)
  690. {
  691. return PVOP_CALL2(unsigned long long, pte_val, x.pte_low, x.pte_high);
  692. }
  693. static inline unsigned long long pmd_val(pmd_t x)
  694. {
  695. return PVOP_CALL2(unsigned long long, pmd_val, x.pmd, x.pmd >> 32);
  696. }
  697. static inline unsigned long long pgd_val(pgd_t x)
  698. {
  699. return PVOP_CALL2(unsigned long long, pgd_val, x.pgd, x.pgd >> 32);
  700. }
  701. static inline void set_pte(pte_t *ptep, pte_t pteval)
  702. {
  703. PVOP_VCALL3(set_pte, ptep, pteval.pte_low, pteval.pte_high);
  704. }
  705. static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
  706. pte_t *ptep, pte_t pteval)
  707. {
  708. /* 5 arg words */
  709. paravirt_ops.set_pte_at(mm, addr, ptep, pteval);
  710. }
  711. static inline void set_pte_atomic(pte_t *ptep, pte_t pteval)
  712. {
  713. PVOP_VCALL3(set_pte_atomic, ptep, pteval.pte_low, pteval.pte_high);
  714. }
  715. static inline void set_pte_present(struct mm_struct *mm, unsigned long addr,
  716. pte_t *ptep, pte_t pte)
  717. {
  718. /* 5 arg words */
  719. paravirt_ops.set_pte_present(mm, addr, ptep, pte);
  720. }
  721. static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
  722. {
  723. PVOP_VCALL3(set_pmd, pmdp, pmdval.pmd, pmdval.pmd >> 32);
  724. }
  725. static inline void set_pud(pud_t *pudp, pud_t pudval)
  726. {
  727. PVOP_VCALL3(set_pud, pudp, pudval.pgd.pgd, pudval.pgd.pgd >> 32);
  728. }
  729. static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
  730. {
  731. PVOP_VCALL3(pte_clear, mm, addr, ptep);
  732. }
  733. static inline void pmd_clear(pmd_t *pmdp)
  734. {
  735. PVOP_VCALL1(pmd_clear, pmdp);
  736. }
  737. #else /* !CONFIG_X86_PAE */
  738. static inline pte_t __pte(unsigned long val)
  739. {
  740. return (pte_t) { PVOP_CALL1(unsigned long, make_pte, val) };
  741. }
  742. static inline pgd_t __pgd(unsigned long val)
  743. {
  744. return (pgd_t) { PVOP_CALL1(unsigned long, make_pgd, val) };
  745. }
  746. static inline unsigned long pte_val(pte_t x)
  747. {
  748. return PVOP_CALL1(unsigned long, pte_val, x.pte_low);
  749. }
  750. static inline unsigned long pgd_val(pgd_t x)
  751. {
  752. return PVOP_CALL1(unsigned long, pgd_val, x.pgd);
  753. }
  754. static inline void set_pte(pte_t *ptep, pte_t pteval)
  755. {
  756. PVOP_VCALL2(set_pte, ptep, pteval.pte_low);
  757. }
  758. static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
  759. pte_t *ptep, pte_t pteval)
  760. {
  761. PVOP_VCALL4(set_pte_at, mm, addr, ptep, pteval.pte_low);
  762. }
  763. static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
  764. {
  765. PVOP_VCALL2(set_pmd, pmdp, pmdval.pud.pgd.pgd);
  766. }
  767. #endif /* CONFIG_X86_PAE */
  768. #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
  769. static inline void arch_enter_lazy_cpu_mode(void)
  770. {
  771. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_CPU);
  772. }
  773. static inline void arch_leave_lazy_cpu_mode(void)
  774. {
  775. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_NONE);
  776. }
  777. static inline void arch_flush_lazy_cpu_mode(void)
  778. {
  779. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_FLUSH);
  780. }
  781. #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
  782. static inline void arch_enter_lazy_mmu_mode(void)
  783. {
  784. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_MMU);
  785. }
  786. static inline void arch_leave_lazy_mmu_mode(void)
  787. {
  788. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_NONE);
  789. }
  790. static inline void arch_flush_lazy_mmu_mode(void)
  791. {
  792. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_FLUSH);
  793. }
  794. void _paravirt_nop(void);
  795. #define paravirt_nop ((void *)_paravirt_nop)
  796. /* These all sit in the .parainstructions section to tell us what to patch. */
  797. struct paravirt_patch_site {
  798. u8 *instr; /* original instructions */
  799. u8 instrtype; /* type of this instruction */
  800. u8 len; /* length of original instruction */
  801. u16 clobbers; /* what registers you may clobber */
  802. };
  803. extern struct paravirt_patch_site __parainstructions[],
  804. __parainstructions_end[];
  805. static inline unsigned long __raw_local_save_flags(void)
  806. {
  807. unsigned long f;
  808. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  809. PARAVIRT_CALL
  810. "popl %%edx; popl %%ecx")
  811. : "=a"(f)
  812. : paravirt_type(save_fl),
  813. paravirt_clobber(CLBR_EAX)
  814. : "memory", "cc");
  815. return f;
  816. }
  817. static inline void raw_local_irq_restore(unsigned long f)
  818. {
  819. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  820. PARAVIRT_CALL
  821. "popl %%edx; popl %%ecx")
  822. : "=a"(f)
  823. : "0"(f),
  824. paravirt_type(restore_fl),
  825. paravirt_clobber(CLBR_EAX)
  826. : "memory", "cc");
  827. }
  828. static inline void raw_local_irq_disable(void)
  829. {
  830. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  831. PARAVIRT_CALL
  832. "popl %%edx; popl %%ecx")
  833. :
  834. : paravirt_type(irq_disable),
  835. paravirt_clobber(CLBR_EAX)
  836. : "memory", "eax", "cc");
  837. }
  838. static inline void raw_local_irq_enable(void)
  839. {
  840. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  841. PARAVIRT_CALL
  842. "popl %%edx; popl %%ecx")
  843. :
  844. : paravirt_type(irq_enable),
  845. paravirt_clobber(CLBR_EAX)
  846. : "memory", "eax", "cc");
  847. }
  848. static inline unsigned long __raw_local_irq_save(void)
  849. {
  850. unsigned long f;
  851. f = __raw_local_save_flags();
  852. raw_local_irq_disable();
  853. return f;
  854. }
  855. #define CLI_STRING \
  856. _paravirt_alt("pushl %%ecx; pushl %%edx;" \
  857. "call *paravirt_ops+%c[paravirt_cli_type]*4;" \
  858. "popl %%edx; popl %%ecx", \
  859. "%c[paravirt_cli_type]", "%c[paravirt_clobber]")
  860. #define STI_STRING \
  861. _paravirt_alt("pushl %%ecx; pushl %%edx;" \
  862. "call *paravirt_ops+%c[paravirt_sti_type]*4;" \
  863. "popl %%edx; popl %%ecx", \
  864. "%c[paravirt_sti_type]", "%c[paravirt_clobber]")
  865. #define CLI_STI_CLOBBERS , "%eax"
  866. #define CLI_STI_INPUT_ARGS \
  867. , \
  868. [paravirt_cli_type] "i" (PARAVIRT_PATCH(irq_disable)), \
  869. [paravirt_sti_type] "i" (PARAVIRT_PATCH(irq_enable)), \
  870. paravirt_clobber(CLBR_EAX)
  871. /* Make sure as little as possible of this mess escapes. */
  872. #undef PARAVIRT_CALL
  873. #undef __PVOP_CALL
  874. #undef __PVOP_VCALL
  875. #undef PVOP_VCALL0
  876. #undef PVOP_CALL0
  877. #undef PVOP_VCALL1
  878. #undef PVOP_CALL1
  879. #undef PVOP_VCALL2
  880. #undef PVOP_CALL2
  881. #undef PVOP_VCALL3
  882. #undef PVOP_CALL3
  883. #undef PVOP_VCALL4
  884. #undef PVOP_CALL4
  885. #else /* __ASSEMBLY__ */
  886. #define PARA_PATCH(off) ((off) / 4)
  887. #define PARA_SITE(ptype, clobbers, ops) \
  888. 771:; \
  889. ops; \
  890. 772:; \
  891. .pushsection .parainstructions,"a"; \
  892. .long 771b; \
  893. .byte ptype; \
  894. .byte 772b-771b; \
  895. .short clobbers; \
  896. .popsection
  897. #define INTERRUPT_RETURN \
  898. PARA_SITE(PARA_PATCH(PARAVIRT_iret), CLBR_NONE, \
  899. jmp *%cs:paravirt_ops+PARAVIRT_iret)
  900. #define DISABLE_INTERRUPTS(clobbers) \
  901. PARA_SITE(PARA_PATCH(PARAVIRT_irq_disable), clobbers, \
  902. pushl %eax; pushl %ecx; pushl %edx; \
  903. call *%cs:paravirt_ops+PARAVIRT_irq_disable; \
  904. popl %edx; popl %ecx; popl %eax) \
  905. #define ENABLE_INTERRUPTS(clobbers) \
  906. PARA_SITE(PARA_PATCH(PARAVIRT_irq_enable), clobbers, \
  907. pushl %eax; pushl %ecx; pushl %edx; \
  908. call *%cs:paravirt_ops+PARAVIRT_irq_enable; \
  909. popl %edx; popl %ecx; popl %eax)
  910. #define ENABLE_INTERRUPTS_SYSEXIT \
  911. PARA_SITE(PARA_PATCH(PARAVIRT_irq_enable_sysexit), CLBR_NONE, \
  912. jmp *%cs:paravirt_ops+PARAVIRT_irq_enable_sysexit)
  913. #define GET_CR0_INTO_EAX \
  914. push %ecx; push %edx; \
  915. call *paravirt_ops+PARAVIRT_read_cr0; \
  916. pop %edx; pop %ecx
  917. #endif /* __ASSEMBLY__ */
  918. #endif /* CONFIG_PARAVIRT */
  919. #endif /* __ASM_PARAVIRT_H */