paravirt.h 29 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074
  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 rdtscl(low) do { \
  483. u64 _l = paravirt_read_tsc(); \
  484. low = (int)_l; \
  485. } while(0)
  486. #define rdtscll(val) (val = paravirt_read_tsc())
  487. #define get_scheduled_cycles(val) (val = paravirt_ops.get_scheduled_cycles())
  488. #define calculate_cpu_khz() (paravirt_ops.get_cpu_khz())
  489. #define write_tsc(val1,val2) wrmsr(0x10, val1, val2)
  490. static inline unsigned long long paravirt_read_pmc(int counter)
  491. {
  492. return PVOP_CALL1(u64, read_pmc, counter);
  493. }
  494. #define rdpmc(counter,low,high) do { \
  495. u64 _l = paravirt_read_pmc(counter); \
  496. low = (u32)_l; \
  497. high = _l >> 32; \
  498. } while(0)
  499. static inline void load_TR_desc(void)
  500. {
  501. PVOP_VCALL0(load_tr_desc);
  502. }
  503. static inline void load_gdt(const struct Xgt_desc_struct *dtr)
  504. {
  505. PVOP_VCALL1(load_gdt, dtr);
  506. }
  507. static inline void load_idt(const struct Xgt_desc_struct *dtr)
  508. {
  509. PVOP_VCALL1(load_idt, dtr);
  510. }
  511. static inline void set_ldt(const void *addr, unsigned entries)
  512. {
  513. PVOP_VCALL2(set_ldt, addr, entries);
  514. }
  515. static inline void store_gdt(struct Xgt_desc_struct *dtr)
  516. {
  517. PVOP_VCALL1(store_gdt, dtr);
  518. }
  519. static inline void store_idt(struct Xgt_desc_struct *dtr)
  520. {
  521. PVOP_VCALL1(store_idt, dtr);
  522. }
  523. static inline unsigned long paravirt_store_tr(void)
  524. {
  525. return PVOP_CALL0(unsigned long, store_tr);
  526. }
  527. #define store_tr(tr) ((tr) = paravirt_store_tr())
  528. static inline void load_TLS(struct thread_struct *t, unsigned cpu)
  529. {
  530. PVOP_VCALL2(load_tls, t, cpu);
  531. }
  532. static inline void write_ldt_entry(void *dt, int entry, u32 low, u32 high)
  533. {
  534. PVOP_VCALL4(write_ldt_entry, dt, entry, low, high);
  535. }
  536. static inline void write_gdt_entry(void *dt, int entry, u32 low, u32 high)
  537. {
  538. PVOP_VCALL4(write_gdt_entry, dt, entry, low, high);
  539. }
  540. static inline void write_idt_entry(void *dt, int entry, u32 low, u32 high)
  541. {
  542. PVOP_VCALL4(write_idt_entry, dt, entry, low, high);
  543. }
  544. static inline void set_iopl_mask(unsigned mask)
  545. {
  546. PVOP_VCALL1(set_iopl_mask, mask);
  547. }
  548. /* The paravirtualized I/O functions */
  549. static inline void slow_down_io(void) {
  550. paravirt_ops.io_delay();
  551. #ifdef REALLY_SLOW_IO
  552. paravirt_ops.io_delay();
  553. paravirt_ops.io_delay();
  554. paravirt_ops.io_delay();
  555. #endif
  556. }
  557. #ifdef CONFIG_X86_LOCAL_APIC
  558. /*
  559. * Basic functions accessing APICs.
  560. */
  561. static inline void apic_write(unsigned long reg, unsigned long v)
  562. {
  563. PVOP_VCALL2(apic_write, reg, v);
  564. }
  565. static inline void apic_write_atomic(unsigned long reg, unsigned long v)
  566. {
  567. PVOP_VCALL2(apic_write_atomic, reg, v);
  568. }
  569. static inline unsigned long apic_read(unsigned long reg)
  570. {
  571. return PVOP_CALL1(unsigned long, apic_read, reg);
  572. }
  573. static inline void setup_boot_clock(void)
  574. {
  575. PVOP_VCALL0(setup_boot_clock);
  576. }
  577. static inline void setup_secondary_clock(void)
  578. {
  579. PVOP_VCALL0(setup_secondary_clock);
  580. }
  581. #endif
  582. static inline void paravirt_pagetable_setup_start(pgd_t *base)
  583. {
  584. if (paravirt_ops.pagetable_setup_start)
  585. (*paravirt_ops.pagetable_setup_start)(base);
  586. }
  587. static inline void paravirt_pagetable_setup_done(pgd_t *base)
  588. {
  589. if (paravirt_ops.pagetable_setup_done)
  590. (*paravirt_ops.pagetable_setup_done)(base);
  591. }
  592. #ifdef CONFIG_SMP
  593. static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
  594. unsigned long start_esp)
  595. {
  596. PVOP_VCALL3(startup_ipi_hook, phys_apicid, start_eip, start_esp);
  597. }
  598. #endif
  599. static inline void paravirt_activate_mm(struct mm_struct *prev,
  600. struct mm_struct *next)
  601. {
  602. PVOP_VCALL2(activate_mm, prev, next);
  603. }
  604. static inline void arch_dup_mmap(struct mm_struct *oldmm,
  605. struct mm_struct *mm)
  606. {
  607. PVOP_VCALL2(dup_mmap, oldmm, mm);
  608. }
  609. static inline void arch_exit_mmap(struct mm_struct *mm)
  610. {
  611. PVOP_VCALL1(exit_mmap, mm);
  612. }
  613. static inline void __flush_tlb(void)
  614. {
  615. PVOP_VCALL0(flush_tlb_user);
  616. }
  617. static inline void __flush_tlb_global(void)
  618. {
  619. PVOP_VCALL0(flush_tlb_kernel);
  620. }
  621. static inline void __flush_tlb_single(unsigned long addr)
  622. {
  623. PVOP_VCALL1(flush_tlb_single, addr);
  624. }
  625. static inline void flush_tlb_others(cpumask_t cpumask, struct mm_struct *mm,
  626. unsigned long va)
  627. {
  628. PVOP_VCALL3(flush_tlb_others, &cpumask, mm, va);
  629. }
  630. static inline void paravirt_alloc_pt(unsigned pfn)
  631. {
  632. PVOP_VCALL1(alloc_pt, pfn);
  633. }
  634. static inline void paravirt_release_pt(unsigned pfn)
  635. {
  636. PVOP_VCALL1(release_pt, pfn);
  637. }
  638. static inline void paravirt_alloc_pd(unsigned pfn)
  639. {
  640. PVOP_VCALL1(alloc_pd, pfn);
  641. }
  642. static inline void paravirt_alloc_pd_clone(unsigned pfn, unsigned clonepfn,
  643. unsigned start, unsigned count)
  644. {
  645. PVOP_VCALL4(alloc_pd_clone, pfn, clonepfn, start, count);
  646. }
  647. static inline void paravirt_release_pd(unsigned pfn)
  648. {
  649. PVOP_VCALL1(release_pd, pfn);
  650. }
  651. #ifdef CONFIG_HIGHPTE
  652. static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
  653. {
  654. unsigned long ret;
  655. ret = PVOP_CALL2(unsigned long, kmap_atomic_pte, page, type);
  656. return (void *)ret;
  657. }
  658. #endif
  659. static inline void pte_update(struct mm_struct *mm, unsigned long addr,
  660. pte_t *ptep)
  661. {
  662. PVOP_VCALL3(pte_update, mm, addr, ptep);
  663. }
  664. static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
  665. pte_t *ptep)
  666. {
  667. PVOP_VCALL3(pte_update_defer, mm, addr, ptep);
  668. }
  669. #ifdef CONFIG_X86_PAE
  670. static inline pte_t __pte(unsigned long long val)
  671. {
  672. unsigned long long ret = PVOP_CALL2(unsigned long long, make_pte,
  673. val, val >> 32);
  674. return (pte_t) { ret, ret >> 32 };
  675. }
  676. static inline pmd_t __pmd(unsigned long long val)
  677. {
  678. return (pmd_t) { PVOP_CALL2(unsigned long long, make_pmd, val, val >> 32) };
  679. }
  680. static inline pgd_t __pgd(unsigned long long val)
  681. {
  682. return (pgd_t) { PVOP_CALL2(unsigned long long, make_pgd, val, val >> 32) };
  683. }
  684. static inline unsigned long long pte_val(pte_t x)
  685. {
  686. return PVOP_CALL2(unsigned long long, pte_val, x.pte_low, x.pte_high);
  687. }
  688. static inline unsigned long long pmd_val(pmd_t x)
  689. {
  690. return PVOP_CALL2(unsigned long long, pmd_val, x.pmd, x.pmd >> 32);
  691. }
  692. static inline unsigned long long pgd_val(pgd_t x)
  693. {
  694. return PVOP_CALL2(unsigned long long, pgd_val, x.pgd, x.pgd >> 32);
  695. }
  696. static inline void set_pte(pte_t *ptep, pte_t pteval)
  697. {
  698. PVOP_VCALL3(set_pte, ptep, pteval.pte_low, pteval.pte_high);
  699. }
  700. static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
  701. pte_t *ptep, pte_t pteval)
  702. {
  703. /* 5 arg words */
  704. paravirt_ops.set_pte_at(mm, addr, ptep, pteval);
  705. }
  706. static inline void set_pte_atomic(pte_t *ptep, pte_t pteval)
  707. {
  708. PVOP_VCALL3(set_pte_atomic, ptep, pteval.pte_low, pteval.pte_high);
  709. }
  710. static inline void set_pte_present(struct mm_struct *mm, unsigned long addr,
  711. pte_t *ptep, pte_t pte)
  712. {
  713. /* 5 arg words */
  714. paravirt_ops.set_pte_present(mm, addr, ptep, pte);
  715. }
  716. static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
  717. {
  718. PVOP_VCALL3(set_pmd, pmdp, pmdval.pmd, pmdval.pmd >> 32);
  719. }
  720. static inline void set_pud(pud_t *pudp, pud_t pudval)
  721. {
  722. PVOP_VCALL3(set_pud, pudp, pudval.pgd.pgd, pudval.pgd.pgd >> 32);
  723. }
  724. static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
  725. {
  726. PVOP_VCALL3(pte_clear, mm, addr, ptep);
  727. }
  728. static inline void pmd_clear(pmd_t *pmdp)
  729. {
  730. PVOP_VCALL1(pmd_clear, pmdp);
  731. }
  732. #else /* !CONFIG_X86_PAE */
  733. static inline pte_t __pte(unsigned long val)
  734. {
  735. return (pte_t) { PVOP_CALL1(unsigned long, make_pte, val) };
  736. }
  737. static inline pgd_t __pgd(unsigned long val)
  738. {
  739. return (pgd_t) { PVOP_CALL1(unsigned long, make_pgd, val) };
  740. }
  741. static inline unsigned long pte_val(pte_t x)
  742. {
  743. return PVOP_CALL1(unsigned long, pte_val, x.pte_low);
  744. }
  745. static inline unsigned long pgd_val(pgd_t x)
  746. {
  747. return PVOP_CALL1(unsigned long, pgd_val, x.pgd);
  748. }
  749. static inline void set_pte(pte_t *ptep, pte_t pteval)
  750. {
  751. PVOP_VCALL2(set_pte, ptep, pteval.pte_low);
  752. }
  753. static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
  754. pte_t *ptep, pte_t pteval)
  755. {
  756. PVOP_VCALL4(set_pte_at, mm, addr, ptep, pteval.pte_low);
  757. }
  758. static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
  759. {
  760. PVOP_VCALL2(set_pmd, pmdp, pmdval.pud.pgd.pgd);
  761. }
  762. #endif /* CONFIG_X86_PAE */
  763. #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
  764. static inline void arch_enter_lazy_cpu_mode(void)
  765. {
  766. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_CPU);
  767. }
  768. static inline void arch_leave_lazy_cpu_mode(void)
  769. {
  770. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_NONE);
  771. }
  772. static inline void arch_flush_lazy_cpu_mode(void)
  773. {
  774. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_FLUSH);
  775. }
  776. #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
  777. static inline void arch_enter_lazy_mmu_mode(void)
  778. {
  779. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_MMU);
  780. }
  781. static inline void arch_leave_lazy_mmu_mode(void)
  782. {
  783. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_NONE);
  784. }
  785. static inline void arch_flush_lazy_mmu_mode(void)
  786. {
  787. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_FLUSH);
  788. }
  789. void _paravirt_nop(void);
  790. #define paravirt_nop ((void *)_paravirt_nop)
  791. /* These all sit in the .parainstructions section to tell us what to patch. */
  792. struct paravirt_patch_site {
  793. u8 *instr; /* original instructions */
  794. u8 instrtype; /* type of this instruction */
  795. u8 len; /* length of original instruction */
  796. u16 clobbers; /* what registers you may clobber */
  797. };
  798. extern struct paravirt_patch_site __parainstructions[],
  799. __parainstructions_end[];
  800. static inline unsigned long __raw_local_save_flags(void)
  801. {
  802. unsigned long f;
  803. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  804. PARAVIRT_CALL
  805. "popl %%edx; popl %%ecx")
  806. : "=a"(f)
  807. : paravirt_type(save_fl),
  808. paravirt_clobber(CLBR_EAX)
  809. : "memory", "cc");
  810. return f;
  811. }
  812. static inline void raw_local_irq_restore(unsigned long f)
  813. {
  814. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  815. PARAVIRT_CALL
  816. "popl %%edx; popl %%ecx")
  817. : "=a"(f)
  818. : "0"(f),
  819. paravirt_type(restore_fl),
  820. paravirt_clobber(CLBR_EAX)
  821. : "memory", "cc");
  822. }
  823. static inline void raw_local_irq_disable(void)
  824. {
  825. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  826. PARAVIRT_CALL
  827. "popl %%edx; popl %%ecx")
  828. :
  829. : paravirt_type(irq_disable),
  830. paravirt_clobber(CLBR_EAX)
  831. : "memory", "eax", "cc");
  832. }
  833. static inline void raw_local_irq_enable(void)
  834. {
  835. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  836. PARAVIRT_CALL
  837. "popl %%edx; popl %%ecx")
  838. :
  839. : paravirt_type(irq_enable),
  840. paravirt_clobber(CLBR_EAX)
  841. : "memory", "eax", "cc");
  842. }
  843. static inline unsigned long __raw_local_irq_save(void)
  844. {
  845. unsigned long f;
  846. f = __raw_local_save_flags();
  847. raw_local_irq_disable();
  848. return f;
  849. }
  850. #define CLI_STRING \
  851. _paravirt_alt("pushl %%ecx; pushl %%edx;" \
  852. "call *paravirt_ops+%c[paravirt_cli_type]*4;" \
  853. "popl %%edx; popl %%ecx", \
  854. "%c[paravirt_cli_type]", "%c[paravirt_clobber]")
  855. #define STI_STRING \
  856. _paravirt_alt("pushl %%ecx; pushl %%edx;" \
  857. "call *paravirt_ops+%c[paravirt_sti_type]*4;" \
  858. "popl %%edx; popl %%ecx", \
  859. "%c[paravirt_sti_type]", "%c[paravirt_clobber]")
  860. #define CLI_STI_CLOBBERS , "%eax"
  861. #define CLI_STI_INPUT_ARGS \
  862. , \
  863. [paravirt_cli_type] "i" (PARAVIRT_PATCH(irq_disable)), \
  864. [paravirt_sti_type] "i" (PARAVIRT_PATCH(irq_enable)), \
  865. paravirt_clobber(CLBR_EAX)
  866. /* Make sure as little as possible of this mess escapes. */
  867. #undef PARAVIRT_CALL
  868. #undef __PVOP_CALL
  869. #undef __PVOP_VCALL
  870. #undef PVOP_VCALL0
  871. #undef PVOP_CALL0
  872. #undef PVOP_VCALL1
  873. #undef PVOP_CALL1
  874. #undef PVOP_VCALL2
  875. #undef PVOP_CALL2
  876. #undef PVOP_VCALL3
  877. #undef PVOP_CALL3
  878. #undef PVOP_VCALL4
  879. #undef PVOP_CALL4
  880. #else /* __ASSEMBLY__ */
  881. #define PARA_PATCH(off) ((off) / 4)
  882. #define PARA_SITE(ptype, clobbers, ops) \
  883. 771:; \
  884. ops; \
  885. 772:; \
  886. .pushsection .parainstructions,"a"; \
  887. .long 771b; \
  888. .byte ptype; \
  889. .byte 772b-771b; \
  890. .short clobbers; \
  891. .popsection
  892. #define INTERRUPT_RETURN \
  893. PARA_SITE(PARA_PATCH(PARAVIRT_iret), CLBR_NONE, \
  894. jmp *%cs:paravirt_ops+PARAVIRT_iret)
  895. #define DISABLE_INTERRUPTS(clobbers) \
  896. PARA_SITE(PARA_PATCH(PARAVIRT_irq_disable), clobbers, \
  897. pushl %eax; pushl %ecx; pushl %edx; \
  898. call *%cs:paravirt_ops+PARAVIRT_irq_disable; \
  899. popl %edx; popl %ecx; popl %eax) \
  900. #define ENABLE_INTERRUPTS(clobbers) \
  901. PARA_SITE(PARA_PATCH(PARAVIRT_irq_enable), clobbers, \
  902. pushl %eax; pushl %ecx; pushl %edx; \
  903. call *%cs:paravirt_ops+PARAVIRT_irq_enable; \
  904. popl %edx; popl %ecx; popl %eax)
  905. #define ENABLE_INTERRUPTS_SYSEXIT \
  906. PARA_SITE(PARA_PATCH(PARAVIRT_irq_enable_sysexit), CLBR_NONE, \
  907. jmp *%cs:paravirt_ops+PARAVIRT_irq_enable_sysexit)
  908. #define GET_CR0_INTO_EAX \
  909. push %ecx; push %edx; \
  910. call *paravirt_ops+PARAVIRT_read_cr0; \
  911. pop %edx; pop %ecx
  912. #endif /* __ASSEMBLY__ */
  913. #endif /* CONFIG_PARAVIRT */
  914. #endif /* __ASM_PARAVIRT_H */