paravirt.h 29 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078
  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 (*post_allocator_init)(void);
  48. void (*init_IRQ)(void);
  49. void (*time_init)(void);
  50. /*
  51. * Called before/after init_mm pagetable setup. setup_start
  52. * may reset %cr3, and may pre-install parts of the pagetable;
  53. * pagetable setup is expected to preserve any existing
  54. * mapping.
  55. */
  56. void (*pagetable_setup_start)(pgd_t *pgd_base);
  57. void (*pagetable_setup_done)(pgd_t *pgd_base);
  58. /* Print a banner to identify the environment */
  59. void (*banner)(void);
  60. /* Set and set time of day */
  61. unsigned long (*get_wallclock)(void);
  62. int (*set_wallclock)(unsigned long);
  63. /* cpuid emulation, mostly so that caps bits can be disabled */
  64. void (*cpuid)(unsigned int *eax, unsigned int *ebx,
  65. unsigned int *ecx, unsigned int *edx);
  66. /* hooks for various privileged instructions */
  67. unsigned long (*get_debugreg)(int regno);
  68. void (*set_debugreg)(int regno, unsigned long value);
  69. void (*clts)(void);
  70. unsigned long (*read_cr0)(void);
  71. void (*write_cr0)(unsigned long);
  72. unsigned long (*read_cr2)(void);
  73. void (*write_cr2)(unsigned long);
  74. unsigned long (*read_cr3)(void);
  75. void (*write_cr3)(unsigned long);
  76. unsigned long (*read_cr4_safe)(void);
  77. unsigned long (*read_cr4)(void);
  78. void (*write_cr4)(unsigned long);
  79. /*
  80. * Get/set interrupt state. save_fl and restore_fl are only
  81. * expected to use X86_EFLAGS_IF; all other bits
  82. * returned from save_fl are undefined, and may be ignored by
  83. * restore_fl.
  84. */
  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);
  90. void (*halt)(void);
  91. void (*wbinvd)(void);
  92. /* MSR, PMC and TSR operations.
  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);
  96. u64 (*read_tsc)(void);
  97. u64 (*read_pmc)(void);
  98. u64 (*get_scheduled_cycles)(void);
  99. unsigned long (*get_cpu_khz)(void);
  100. /* Segment descriptor handling */
  101. void (*load_tr_desc)(void);
  102. void (*load_gdt)(const struct Xgt_desc_struct *);
  103. void (*load_idt)(const struct Xgt_desc_struct *);
  104. void (*store_gdt)(struct Xgt_desc_struct *);
  105. void (*store_idt)(struct Xgt_desc_struct *);
  106. void (*set_ldt)(const void *desc, unsigned entries);
  107. unsigned long (*store_tr)(void);
  108. void (*load_tls)(struct thread_struct *t, unsigned int cpu);
  109. void (*write_ldt_entry)(struct desc_struct *,
  110. int entrynum, u32 low, u32 high);
  111. void (*write_gdt_entry)(struct desc_struct *,
  112. int entrynum, u32 low, u32 high);
  113. void (*write_idt_entry)(struct desc_struct *,
  114. int entrynum, u32 low, u32 high);
  115. void (*load_esp0)(struct tss_struct *tss, struct thread_struct *t);
  116. void (*set_iopl_mask)(unsigned mask);
  117. void (*io_delay)(void);
  118. /*
  119. * Hooks for intercepting the creation/use/destruction of an
  120. * mm_struct.
  121. */
  122. void (*activate_mm)(struct mm_struct *prev,
  123. struct mm_struct *next);
  124. void (*dup_mmap)(struct mm_struct *oldmm,
  125. struct mm_struct *mm);
  126. void (*exit_mmap)(struct mm_struct *mm);
  127. #ifdef CONFIG_X86_LOCAL_APIC
  128. /*
  129. * Direct APIC operations, principally for VMI. Ideally
  130. * these shouldn't be in this interface.
  131. */
  132. void (*apic_write)(unsigned long reg, unsigned long v);
  133. void (*apic_write_atomic)(unsigned long reg, unsigned long v);
  134. unsigned long (*apic_read)(unsigned long reg);
  135. void (*setup_boot_clock)(void);
  136. void (*setup_secondary_clock)(void);
  137. void (*startup_ipi_hook)(int phys_apicid,
  138. unsigned long start_eip,
  139. unsigned long start_esp);
  140. #endif
  141. /* TLB operations */
  142. void (*flush_tlb_user)(void);
  143. void (*flush_tlb_kernel)(void);
  144. void (*flush_tlb_single)(unsigned long addr);
  145. void (*flush_tlb_others)(const cpumask_t *cpus, struct mm_struct *mm,
  146. unsigned long va);
  147. /* Hooks for allocating/releasing pagetable pages */
  148. void (*alloc_pt)(struct mm_struct *mm, u32 pfn);
  149. void (*alloc_pd)(u32 pfn);
  150. void (*alloc_pd_clone)(u32 pfn, u32 clonepfn, u32 start, u32 count);
  151. void (*release_pt)(u32 pfn);
  152. void (*release_pd)(u32 pfn);
  153. /* Pagetable manipulation functions */
  154. void (*set_pte)(pte_t *ptep, pte_t pteval);
  155. void (*set_pte_at)(struct mm_struct *mm, unsigned long addr,
  156. pte_t *ptep, pte_t pteval);
  157. void (*set_pmd)(pmd_t *pmdp, pmd_t pmdval);
  158. void (*pte_update)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
  159. void (*pte_update_defer)(struct mm_struct *mm,
  160. unsigned long addr, pte_t *ptep);
  161. #ifdef CONFIG_HIGHPTE
  162. void *(*kmap_atomic_pte)(struct page *page, enum km_type type);
  163. #endif
  164. #ifdef CONFIG_X86_PAE
  165. void (*set_pte_atomic)(pte_t *ptep, pte_t pteval);
  166. void (*set_pte_present)(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte);
  167. void (*set_pud)(pud_t *pudp, pud_t pudval);
  168. void (*pte_clear)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
  169. void (*pmd_clear)(pmd_t *pmdp);
  170. unsigned long long (*pte_val)(pte_t);
  171. unsigned long long (*pmd_val)(pmd_t);
  172. unsigned long long (*pgd_val)(pgd_t);
  173. pte_t (*make_pte)(unsigned long long pte);
  174. pmd_t (*make_pmd)(unsigned long long pmd);
  175. pgd_t (*make_pgd)(unsigned long long pgd);
  176. #else
  177. unsigned long (*pte_val)(pte_t);
  178. unsigned long (*pgd_val)(pgd_t);
  179. pte_t (*make_pte)(unsigned long pte);
  180. pgd_t (*make_pgd)(unsigned long pgd);
  181. #endif
  182. /* Set deferred update mode, used for batching operations. */
  183. void (*set_lazy_mode)(enum paravirt_lazy_mode mode);
  184. /* These two are jmp to, not actually called. */
  185. void (*irq_enable_sysexit)(void);
  186. void (*iret)(void);
  187. };
  188. extern struct paravirt_ops paravirt_ops;
  189. #define PARAVIRT_PATCH(x) \
  190. (offsetof(struct paravirt_ops, x) / sizeof(void *))
  191. #define paravirt_type(type) \
  192. [paravirt_typenum] "i" (PARAVIRT_PATCH(type))
  193. #define paravirt_clobber(clobber) \
  194. [paravirt_clobber] "i" (clobber)
  195. /*
  196. * Generate some code, and mark it as patchable by the
  197. * apply_paravirt() alternate instruction patcher.
  198. */
  199. #define _paravirt_alt(insn_string, type, clobber) \
  200. "771:\n\t" insn_string "\n" "772:\n" \
  201. ".pushsection .parainstructions,\"a\"\n" \
  202. " .long 771b\n" \
  203. " .byte " type "\n" \
  204. " .byte 772b-771b\n" \
  205. " .short " clobber "\n" \
  206. ".popsection\n"
  207. /* Generate patchable code, with the default asm parameters. */
  208. #define paravirt_alt(insn_string) \
  209. _paravirt_alt(insn_string, "%c[paravirt_typenum]", "%c[paravirt_clobber]")
  210. unsigned paravirt_patch_nop(void);
  211. unsigned paravirt_patch_ignore(unsigned len);
  212. unsigned paravirt_patch_call(void *target, u16 tgt_clobbers,
  213. void *site, u16 site_clobbers,
  214. unsigned len);
  215. unsigned paravirt_patch_jmp(void *target, void *site, unsigned len);
  216. unsigned paravirt_patch_default(u8 type, u16 clobbers, void *site, unsigned len);
  217. unsigned paravirt_patch_insns(void *site, unsigned len,
  218. const char *start, const char *end);
  219. int paravirt_disable_iospace(void);
  220. /*
  221. * This generates an indirect call based on the operation type number.
  222. * The type number, computed in PARAVIRT_PATCH, is derived from the
  223. * offset into the paravirt_ops structure, and can therefore be freely
  224. * converted back into a structure offset.
  225. */
  226. #define PARAVIRT_CALL "call *(paravirt_ops+%c[paravirt_typenum]*4);"
  227. /*
  228. * These macros are intended to wrap calls into a paravirt_ops
  229. * operation, so that they can be later identified and patched at
  230. * runtime.
  231. *
  232. * Normally, a call to a pv_op function is a simple indirect call:
  233. * (paravirt_ops.operations)(args...).
  234. *
  235. * Unfortunately, this is a relatively slow operation for modern CPUs,
  236. * because it cannot necessarily determine what the destination
  237. * address is. In this case, the address is a runtime constant, so at
  238. * the very least we can patch the call to e a simple direct call, or
  239. * ideally, patch an inline implementation into the callsite. (Direct
  240. * calls are essentially free, because the call and return addresses
  241. * are completely predictable.)
  242. *
  243. * These macros rely on the standard gcc "regparm(3)" calling
  244. * convention, in which the first three arguments are placed in %eax,
  245. * %edx, %ecx (in that order), and the remaining arguments are placed
  246. * on the stack. All caller-save registers (eax,edx,ecx) are expected
  247. * to be modified (either clobbered or used for return values).
  248. *
  249. * The call instruction itself is marked by placing its start address
  250. * and size into the .parainstructions section, so that
  251. * apply_paravirt() in arch/i386/kernel/alternative.c can do the
  252. * appropriate patching under the control of the backend paravirt_ops
  253. * implementation.
  254. *
  255. * Unfortunately there's no way to get gcc to generate the args setup
  256. * for the call, and then allow the call itself to be generated by an
  257. * inline asm. Because of this, we must do the complete arg setup and
  258. * return value handling from within these macros. This is fairly
  259. * cumbersome.
  260. *
  261. * There are 5 sets of PVOP_* macros for dealing with 0-4 arguments.
  262. * It could be extended to more arguments, but there would be little
  263. * to be gained from that. For each number of arguments, there are
  264. * the two VCALL and CALL variants for void and non-void functions.
  265. *
  266. * When there is a return value, the invoker of the macro must specify
  267. * the return type. The macro then uses sizeof() on that type to
  268. * determine whether its a 32 or 64 bit value, and places the return
  269. * in the right register(s) (just %eax for 32-bit, and %edx:%eax for
  270. * 64-bit).
  271. *
  272. * 64-bit arguments are passed as a pair of adjacent 32-bit arguments
  273. * in low,high order.
  274. *
  275. * Small structures are passed and returned in registers. The macro
  276. * calling convention can't directly deal with this, so the wrapper
  277. * functions must do this.
  278. *
  279. * These PVOP_* macros are only defined within this header. This
  280. * means that all uses must be wrapped in inline functions. This also
  281. * makes sure the incoming and outgoing types are always correct.
  282. */
  283. #define __PVOP_CALL(rettype, op, pre, post, ...) \
  284. ({ \
  285. rettype __ret; \
  286. unsigned long __eax, __edx, __ecx; \
  287. if (sizeof(rettype) > sizeof(unsigned long)) { \
  288. asm volatile(pre \
  289. paravirt_alt(PARAVIRT_CALL) \
  290. post \
  291. : "=a" (__eax), "=d" (__edx), \
  292. "=c" (__ecx) \
  293. : paravirt_type(op), \
  294. paravirt_clobber(CLBR_ANY), \
  295. ##__VA_ARGS__ \
  296. : "memory", "cc"); \
  297. __ret = (rettype)((((u64)__edx) << 32) | __eax); \
  298. } else { \
  299. asm volatile(pre \
  300. paravirt_alt(PARAVIRT_CALL) \
  301. post \
  302. : "=a" (__eax), "=d" (__edx), \
  303. "=c" (__ecx) \
  304. : paravirt_type(op), \
  305. paravirt_clobber(CLBR_ANY), \
  306. ##__VA_ARGS__ \
  307. : "memory", "cc"); \
  308. __ret = (rettype)__eax; \
  309. } \
  310. __ret; \
  311. })
  312. #define __PVOP_VCALL(op, pre, post, ...) \
  313. ({ \
  314. unsigned long __eax, __edx, __ecx; \
  315. asm volatile(pre \
  316. paravirt_alt(PARAVIRT_CALL) \
  317. post \
  318. : "=a" (__eax), "=d" (__edx), "=c" (__ecx) \
  319. : paravirt_type(op), \
  320. paravirt_clobber(CLBR_ANY), \
  321. ##__VA_ARGS__ \
  322. : "memory", "cc"); \
  323. })
  324. #define PVOP_CALL0(rettype, op) \
  325. __PVOP_CALL(rettype, op, "", "")
  326. #define PVOP_VCALL0(op) \
  327. __PVOP_VCALL(op, "", "")
  328. #define PVOP_CALL1(rettype, op, arg1) \
  329. __PVOP_CALL(rettype, op, "", "", "0" ((u32)(arg1)))
  330. #define PVOP_VCALL1(op, arg1) \
  331. __PVOP_VCALL(op, "", "", "0" ((u32)(arg1)))
  332. #define PVOP_CALL2(rettype, op, arg1, arg2) \
  333. __PVOP_CALL(rettype, op, "", "", "0" ((u32)(arg1)), "1" ((u32)(arg2)))
  334. #define PVOP_VCALL2(op, arg1, arg2) \
  335. __PVOP_VCALL(op, "", "", "0" ((u32)(arg1)), "1" ((u32)(arg2)))
  336. #define PVOP_CALL3(rettype, op, arg1, arg2, arg3) \
  337. __PVOP_CALL(rettype, op, "", "", "0" ((u32)(arg1)), \
  338. "1"((u32)(arg2)), "2"((u32)(arg3)))
  339. #define PVOP_VCALL3(op, arg1, arg2, arg3) \
  340. __PVOP_VCALL(op, "", "", "0" ((u32)(arg1)), "1"((u32)(arg2)), \
  341. "2"((u32)(arg3)))
  342. #define PVOP_CALL4(rettype, op, arg1, arg2, arg3, arg4) \
  343. __PVOP_CALL(rettype, op, \
  344. "push %[_arg4];", "lea 4(%%esp),%%esp;", \
  345. "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
  346. "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
  347. #define PVOP_VCALL4(op, arg1, arg2, arg3, arg4) \
  348. __PVOP_VCALL(op, \
  349. "push %[_arg4];", "lea 4(%%esp),%%esp;", \
  350. "0" ((u32)(arg1)), "1" ((u32)(arg2)), \
  351. "2" ((u32)(arg3)), [_arg4] "mr" ((u32)(arg4)))
  352. static inline int paravirt_enabled(void)
  353. {
  354. return paravirt_ops.paravirt_enabled;
  355. }
  356. static inline void load_esp0(struct tss_struct *tss,
  357. struct thread_struct *thread)
  358. {
  359. PVOP_VCALL2(load_esp0, tss, thread);
  360. }
  361. #define ARCH_SETUP paravirt_ops.arch_setup();
  362. static inline unsigned long get_wallclock(void)
  363. {
  364. return PVOP_CALL0(unsigned long, get_wallclock);
  365. }
  366. static inline int set_wallclock(unsigned long nowtime)
  367. {
  368. return PVOP_CALL1(int, set_wallclock, nowtime);
  369. }
  370. static inline void (*choose_time_init(void))(void)
  371. {
  372. return paravirt_ops.time_init;
  373. }
  374. /* The paravirtualized CPUID instruction. */
  375. static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
  376. unsigned int *ecx, unsigned int *edx)
  377. {
  378. PVOP_VCALL4(cpuid, eax, ebx, ecx, edx);
  379. }
  380. /*
  381. * These special macros can be used to get or set a debugging register
  382. */
  383. static inline unsigned long paravirt_get_debugreg(int reg)
  384. {
  385. return PVOP_CALL1(unsigned long, get_debugreg, reg);
  386. }
  387. #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
  388. static inline void set_debugreg(unsigned long val, int reg)
  389. {
  390. PVOP_VCALL2(set_debugreg, reg, val);
  391. }
  392. static inline void clts(void)
  393. {
  394. PVOP_VCALL0(clts);
  395. }
  396. static inline unsigned long read_cr0(void)
  397. {
  398. return PVOP_CALL0(unsigned long, read_cr0);
  399. }
  400. static inline void write_cr0(unsigned long x)
  401. {
  402. PVOP_VCALL1(write_cr0, x);
  403. }
  404. static inline unsigned long read_cr2(void)
  405. {
  406. return PVOP_CALL0(unsigned long, read_cr2);
  407. }
  408. static inline void write_cr2(unsigned long x)
  409. {
  410. PVOP_VCALL1(write_cr2, x);
  411. }
  412. static inline unsigned long read_cr3(void)
  413. {
  414. return PVOP_CALL0(unsigned long, read_cr3);
  415. }
  416. static inline void write_cr3(unsigned long x)
  417. {
  418. PVOP_VCALL1(write_cr3, x);
  419. }
  420. static inline unsigned long read_cr4(void)
  421. {
  422. return PVOP_CALL0(unsigned long, read_cr4);
  423. }
  424. static inline unsigned long read_cr4_safe(void)
  425. {
  426. return PVOP_CALL0(unsigned long, read_cr4_safe);
  427. }
  428. static inline void write_cr4(unsigned long x)
  429. {
  430. PVOP_VCALL1(write_cr4, x);
  431. }
  432. static inline void raw_safe_halt(void)
  433. {
  434. PVOP_VCALL0(safe_halt);
  435. }
  436. static inline void halt(void)
  437. {
  438. PVOP_VCALL0(safe_halt);
  439. }
  440. static inline void wbinvd(void)
  441. {
  442. PVOP_VCALL0(wbinvd);
  443. }
  444. #define get_kernel_rpl() (paravirt_ops.kernel_rpl)
  445. static inline u64 paravirt_read_msr(unsigned msr, int *err)
  446. {
  447. return PVOP_CALL2(u64, read_msr, msr, err);
  448. }
  449. static inline int paravirt_write_msr(unsigned msr, unsigned low, unsigned high)
  450. {
  451. return PVOP_CALL3(int, write_msr, msr, low, high);
  452. }
  453. /* These should all do BUG_ON(_err), but our headers are too tangled. */
  454. #define rdmsr(msr,val1,val2) do { \
  455. int _err; \
  456. u64 _l = paravirt_read_msr(msr, &_err); \
  457. val1 = (u32)_l; \
  458. val2 = _l >> 32; \
  459. } while(0)
  460. #define wrmsr(msr,val1,val2) do { \
  461. paravirt_write_msr(msr, val1, val2); \
  462. } while(0)
  463. #define rdmsrl(msr,val) do { \
  464. int _err; \
  465. val = paravirt_read_msr(msr, &_err); \
  466. } while(0)
  467. #define wrmsrl(msr,val) wrmsr(msr, (u32)((u64)(val)), ((u64)(val))>>32)
  468. #define wrmsr_safe(msr,a,b) paravirt_write_msr(msr, a, b)
  469. /* rdmsr with exception handling */
  470. #define rdmsr_safe(msr,a,b) ({ \
  471. int _err; \
  472. u64 _l = paravirt_read_msr(msr, &_err); \
  473. (*a) = (u32)_l; \
  474. (*b) = _l >> 32; \
  475. _err; })
  476. static inline u64 paravirt_read_tsc(void)
  477. {
  478. return PVOP_CALL0(u64, read_tsc);
  479. }
  480. #define rdtscl(low) do { \
  481. u64 _l = paravirt_read_tsc(); \
  482. low = (int)_l; \
  483. } while(0)
  484. #define rdtscll(val) (val = paravirt_read_tsc())
  485. #define get_scheduled_cycles(val) (val = paravirt_ops.get_scheduled_cycles())
  486. #define calculate_cpu_khz() (paravirt_ops.get_cpu_khz())
  487. #define write_tsc(val1,val2) wrmsr(0x10, val1, val2)
  488. static inline unsigned long long paravirt_read_pmc(int counter)
  489. {
  490. return PVOP_CALL1(u64, read_pmc, counter);
  491. }
  492. #define rdpmc(counter,low,high) do { \
  493. u64 _l = paravirt_read_pmc(counter); \
  494. low = (u32)_l; \
  495. high = _l >> 32; \
  496. } while(0)
  497. static inline void load_TR_desc(void)
  498. {
  499. PVOP_VCALL0(load_tr_desc);
  500. }
  501. static inline void load_gdt(const struct Xgt_desc_struct *dtr)
  502. {
  503. PVOP_VCALL1(load_gdt, dtr);
  504. }
  505. static inline void load_idt(const struct Xgt_desc_struct *dtr)
  506. {
  507. PVOP_VCALL1(load_idt, dtr);
  508. }
  509. static inline void set_ldt(const void *addr, unsigned entries)
  510. {
  511. PVOP_VCALL2(set_ldt, addr, entries);
  512. }
  513. static inline void store_gdt(struct Xgt_desc_struct *dtr)
  514. {
  515. PVOP_VCALL1(store_gdt, dtr);
  516. }
  517. static inline void store_idt(struct Xgt_desc_struct *dtr)
  518. {
  519. PVOP_VCALL1(store_idt, dtr);
  520. }
  521. static inline unsigned long paravirt_store_tr(void)
  522. {
  523. return PVOP_CALL0(unsigned long, store_tr);
  524. }
  525. #define store_tr(tr) ((tr) = paravirt_store_tr())
  526. static inline void load_TLS(struct thread_struct *t, unsigned cpu)
  527. {
  528. PVOP_VCALL2(load_tls, t, cpu);
  529. }
  530. static inline void write_ldt_entry(void *dt, int entry, u32 low, u32 high)
  531. {
  532. PVOP_VCALL4(write_ldt_entry, dt, entry, low, high);
  533. }
  534. static inline void write_gdt_entry(void *dt, int entry, u32 low, u32 high)
  535. {
  536. PVOP_VCALL4(write_gdt_entry, dt, entry, low, high);
  537. }
  538. static inline void write_idt_entry(void *dt, int entry, u32 low, u32 high)
  539. {
  540. PVOP_VCALL4(write_idt_entry, dt, entry, low, high);
  541. }
  542. static inline void set_iopl_mask(unsigned mask)
  543. {
  544. PVOP_VCALL1(set_iopl_mask, mask);
  545. }
  546. /* The paravirtualized I/O functions */
  547. static inline void slow_down_io(void) {
  548. paravirt_ops.io_delay();
  549. #ifdef REALLY_SLOW_IO
  550. paravirt_ops.io_delay();
  551. paravirt_ops.io_delay();
  552. paravirt_ops.io_delay();
  553. #endif
  554. }
  555. #ifdef CONFIG_X86_LOCAL_APIC
  556. /*
  557. * Basic functions accessing APICs.
  558. */
  559. static inline void apic_write(unsigned long reg, unsigned long v)
  560. {
  561. PVOP_VCALL2(apic_write, reg, v);
  562. }
  563. static inline void apic_write_atomic(unsigned long reg, unsigned long v)
  564. {
  565. PVOP_VCALL2(apic_write_atomic, reg, v);
  566. }
  567. static inline unsigned long apic_read(unsigned long reg)
  568. {
  569. return PVOP_CALL1(unsigned long, apic_read, reg);
  570. }
  571. static inline void setup_boot_clock(void)
  572. {
  573. PVOP_VCALL0(setup_boot_clock);
  574. }
  575. static inline void setup_secondary_clock(void)
  576. {
  577. PVOP_VCALL0(setup_secondary_clock);
  578. }
  579. #endif
  580. static inline void paravirt_post_allocator_init(void)
  581. {
  582. if (paravirt_ops.post_allocator_init)
  583. (*paravirt_ops.post_allocator_init)();
  584. }
  585. static inline void paravirt_pagetable_setup_start(pgd_t *base)
  586. {
  587. if (paravirt_ops.pagetable_setup_start)
  588. (*paravirt_ops.pagetable_setup_start)(base);
  589. }
  590. static inline void paravirt_pagetable_setup_done(pgd_t *base)
  591. {
  592. if (paravirt_ops.pagetable_setup_done)
  593. (*paravirt_ops.pagetable_setup_done)(base);
  594. }
  595. #ifdef CONFIG_SMP
  596. static inline void startup_ipi_hook(int phys_apicid, unsigned long start_eip,
  597. unsigned long start_esp)
  598. {
  599. PVOP_VCALL3(startup_ipi_hook, phys_apicid, start_eip, start_esp);
  600. }
  601. #endif
  602. static inline void paravirt_activate_mm(struct mm_struct *prev,
  603. struct mm_struct *next)
  604. {
  605. PVOP_VCALL2(activate_mm, prev, next);
  606. }
  607. static inline void arch_dup_mmap(struct mm_struct *oldmm,
  608. struct mm_struct *mm)
  609. {
  610. PVOP_VCALL2(dup_mmap, oldmm, mm);
  611. }
  612. static inline void arch_exit_mmap(struct mm_struct *mm)
  613. {
  614. PVOP_VCALL1(exit_mmap, mm);
  615. }
  616. static inline void __flush_tlb(void)
  617. {
  618. PVOP_VCALL0(flush_tlb_user);
  619. }
  620. static inline void __flush_tlb_global(void)
  621. {
  622. PVOP_VCALL0(flush_tlb_kernel);
  623. }
  624. static inline void __flush_tlb_single(unsigned long addr)
  625. {
  626. PVOP_VCALL1(flush_tlb_single, addr);
  627. }
  628. static inline void flush_tlb_others(cpumask_t cpumask, struct mm_struct *mm,
  629. unsigned long va)
  630. {
  631. PVOP_VCALL3(flush_tlb_others, &cpumask, mm, va);
  632. }
  633. static inline void paravirt_alloc_pt(struct mm_struct *mm, unsigned pfn)
  634. {
  635. PVOP_VCALL2(alloc_pt, mm, pfn);
  636. }
  637. static inline void paravirt_release_pt(unsigned pfn)
  638. {
  639. PVOP_VCALL1(release_pt, pfn);
  640. }
  641. static inline void paravirt_alloc_pd(unsigned pfn)
  642. {
  643. PVOP_VCALL1(alloc_pd, pfn);
  644. }
  645. static inline void paravirt_alloc_pd_clone(unsigned pfn, unsigned clonepfn,
  646. unsigned start, unsigned count)
  647. {
  648. PVOP_VCALL4(alloc_pd_clone, pfn, clonepfn, start, count);
  649. }
  650. static inline void paravirt_release_pd(unsigned pfn)
  651. {
  652. PVOP_VCALL1(release_pd, pfn);
  653. }
  654. #ifdef CONFIG_HIGHPTE
  655. static inline void *kmap_atomic_pte(struct page *page, enum km_type type)
  656. {
  657. unsigned long ret;
  658. ret = PVOP_CALL2(unsigned long, kmap_atomic_pte, page, type);
  659. return (void *)ret;
  660. }
  661. #endif
  662. static inline void pte_update(struct mm_struct *mm, unsigned long addr,
  663. pte_t *ptep)
  664. {
  665. PVOP_VCALL3(pte_update, mm, addr, ptep);
  666. }
  667. static inline void pte_update_defer(struct mm_struct *mm, unsigned long addr,
  668. pte_t *ptep)
  669. {
  670. PVOP_VCALL3(pte_update_defer, mm, addr, ptep);
  671. }
  672. #ifdef CONFIG_X86_PAE
  673. static inline pte_t __pte(unsigned long long val)
  674. {
  675. unsigned long long ret = PVOP_CALL2(unsigned long long, make_pte,
  676. val, val >> 32);
  677. return (pte_t) { ret, ret >> 32 };
  678. }
  679. static inline pmd_t __pmd(unsigned long long val)
  680. {
  681. return (pmd_t) { PVOP_CALL2(unsigned long long, make_pmd, val, val >> 32) };
  682. }
  683. static inline pgd_t __pgd(unsigned long long val)
  684. {
  685. return (pgd_t) { PVOP_CALL2(unsigned long long, make_pgd, val, val >> 32) };
  686. }
  687. static inline unsigned long long pte_val(pte_t x)
  688. {
  689. return PVOP_CALL2(unsigned long long, pte_val, x.pte_low, x.pte_high);
  690. }
  691. static inline unsigned long long pmd_val(pmd_t x)
  692. {
  693. return PVOP_CALL2(unsigned long long, pmd_val, x.pmd, x.pmd >> 32);
  694. }
  695. static inline unsigned long long pgd_val(pgd_t x)
  696. {
  697. return PVOP_CALL2(unsigned long long, pgd_val, x.pgd, x.pgd >> 32);
  698. }
  699. static inline void set_pte(pte_t *ptep, pte_t pteval)
  700. {
  701. PVOP_VCALL3(set_pte, ptep, pteval.pte_low, pteval.pte_high);
  702. }
  703. static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
  704. pte_t *ptep, pte_t pteval)
  705. {
  706. /* 5 arg words */
  707. paravirt_ops.set_pte_at(mm, addr, ptep, pteval);
  708. }
  709. static inline void set_pte_atomic(pte_t *ptep, pte_t pteval)
  710. {
  711. PVOP_VCALL3(set_pte_atomic, ptep, pteval.pte_low, pteval.pte_high);
  712. }
  713. static inline void set_pte_present(struct mm_struct *mm, unsigned long addr,
  714. pte_t *ptep, pte_t pte)
  715. {
  716. /* 5 arg words */
  717. paravirt_ops.set_pte_present(mm, addr, ptep, pte);
  718. }
  719. static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
  720. {
  721. PVOP_VCALL3(set_pmd, pmdp, pmdval.pmd, pmdval.pmd >> 32);
  722. }
  723. static inline void set_pud(pud_t *pudp, pud_t pudval)
  724. {
  725. PVOP_VCALL3(set_pud, pudp, pudval.pgd.pgd, pudval.pgd.pgd >> 32);
  726. }
  727. static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
  728. {
  729. PVOP_VCALL3(pte_clear, mm, addr, ptep);
  730. }
  731. static inline void pmd_clear(pmd_t *pmdp)
  732. {
  733. PVOP_VCALL1(pmd_clear, pmdp);
  734. }
  735. #else /* !CONFIG_X86_PAE */
  736. static inline pte_t __pte(unsigned long val)
  737. {
  738. return (pte_t) { PVOP_CALL1(unsigned long, make_pte, val) };
  739. }
  740. static inline pgd_t __pgd(unsigned long val)
  741. {
  742. return (pgd_t) { PVOP_CALL1(unsigned long, make_pgd, val) };
  743. }
  744. static inline unsigned long pte_val(pte_t x)
  745. {
  746. return PVOP_CALL1(unsigned long, pte_val, x.pte_low);
  747. }
  748. static inline unsigned long pgd_val(pgd_t x)
  749. {
  750. return PVOP_CALL1(unsigned long, pgd_val, x.pgd);
  751. }
  752. static inline void set_pte(pte_t *ptep, pte_t pteval)
  753. {
  754. PVOP_VCALL2(set_pte, ptep, pteval.pte_low);
  755. }
  756. static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
  757. pte_t *ptep, pte_t pteval)
  758. {
  759. PVOP_VCALL4(set_pte_at, mm, addr, ptep, pteval.pte_low);
  760. }
  761. static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
  762. {
  763. PVOP_VCALL2(set_pmd, pmdp, pmdval.pud.pgd.pgd);
  764. }
  765. #endif /* CONFIG_X86_PAE */
  766. #define __HAVE_ARCH_ENTER_LAZY_CPU_MODE
  767. static inline void arch_enter_lazy_cpu_mode(void)
  768. {
  769. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_CPU);
  770. }
  771. static inline void arch_leave_lazy_cpu_mode(void)
  772. {
  773. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_NONE);
  774. }
  775. static inline void arch_flush_lazy_cpu_mode(void)
  776. {
  777. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_FLUSH);
  778. }
  779. #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
  780. static inline void arch_enter_lazy_mmu_mode(void)
  781. {
  782. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_MMU);
  783. }
  784. static inline void arch_leave_lazy_mmu_mode(void)
  785. {
  786. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_NONE);
  787. }
  788. static inline void arch_flush_lazy_mmu_mode(void)
  789. {
  790. PVOP_VCALL1(set_lazy_mode, PARAVIRT_LAZY_FLUSH);
  791. }
  792. void _paravirt_nop(void);
  793. #define paravirt_nop ((void *)_paravirt_nop)
  794. /* These all sit in the .parainstructions section to tell us what to patch. */
  795. struct paravirt_patch_site {
  796. u8 *instr; /* original instructions */
  797. u8 instrtype; /* type of this instruction */
  798. u8 len; /* length of original instruction */
  799. u16 clobbers; /* what registers you may clobber */
  800. };
  801. extern struct paravirt_patch_site __parainstructions[],
  802. __parainstructions_end[];
  803. static inline unsigned long __raw_local_save_flags(void)
  804. {
  805. unsigned long f;
  806. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  807. PARAVIRT_CALL
  808. "popl %%edx; popl %%ecx")
  809. : "=a"(f)
  810. : paravirt_type(save_fl),
  811. paravirt_clobber(CLBR_EAX)
  812. : "memory", "cc");
  813. return f;
  814. }
  815. static inline void raw_local_irq_restore(unsigned long f)
  816. {
  817. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  818. PARAVIRT_CALL
  819. "popl %%edx; popl %%ecx")
  820. : "=a"(f)
  821. : "0"(f),
  822. paravirt_type(restore_fl),
  823. paravirt_clobber(CLBR_EAX)
  824. : "memory", "cc");
  825. }
  826. static inline void raw_local_irq_disable(void)
  827. {
  828. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  829. PARAVIRT_CALL
  830. "popl %%edx; popl %%ecx")
  831. :
  832. : paravirt_type(irq_disable),
  833. paravirt_clobber(CLBR_EAX)
  834. : "memory", "eax", "cc");
  835. }
  836. static inline void raw_local_irq_enable(void)
  837. {
  838. asm volatile(paravirt_alt("pushl %%ecx; pushl %%edx;"
  839. PARAVIRT_CALL
  840. "popl %%edx; popl %%ecx")
  841. :
  842. : paravirt_type(irq_enable),
  843. paravirt_clobber(CLBR_EAX)
  844. : "memory", "eax", "cc");
  845. }
  846. static inline unsigned long __raw_local_irq_save(void)
  847. {
  848. unsigned long f;
  849. f = __raw_local_save_flags();
  850. raw_local_irq_disable();
  851. return f;
  852. }
  853. #define CLI_STRING \
  854. _paravirt_alt("pushl %%ecx; pushl %%edx;" \
  855. "call *paravirt_ops+%c[paravirt_cli_type]*4;" \
  856. "popl %%edx; popl %%ecx", \
  857. "%c[paravirt_cli_type]", "%c[paravirt_clobber]")
  858. #define STI_STRING \
  859. _paravirt_alt("pushl %%ecx; pushl %%edx;" \
  860. "call *paravirt_ops+%c[paravirt_sti_type]*4;" \
  861. "popl %%edx; popl %%ecx", \
  862. "%c[paravirt_sti_type]", "%c[paravirt_clobber]")
  863. #define CLI_STI_CLOBBERS , "%eax"
  864. #define CLI_STI_INPUT_ARGS \
  865. , \
  866. [paravirt_cli_type] "i" (PARAVIRT_PATCH(irq_disable)), \
  867. [paravirt_sti_type] "i" (PARAVIRT_PATCH(irq_enable)), \
  868. paravirt_clobber(CLBR_EAX)
  869. /* Make sure as little as possible of this mess escapes. */
  870. #undef PARAVIRT_CALL
  871. #undef __PVOP_CALL
  872. #undef __PVOP_VCALL
  873. #undef PVOP_VCALL0
  874. #undef PVOP_CALL0
  875. #undef PVOP_VCALL1
  876. #undef PVOP_CALL1
  877. #undef PVOP_VCALL2
  878. #undef PVOP_CALL2
  879. #undef PVOP_VCALL3
  880. #undef PVOP_CALL3
  881. #undef PVOP_VCALL4
  882. #undef PVOP_CALL4
  883. #else /* __ASSEMBLY__ */
  884. #define PARA_PATCH(off) ((off) / 4)
  885. #define PARA_SITE(ptype, clobbers, ops) \
  886. 771:; \
  887. ops; \
  888. 772:; \
  889. .pushsection .parainstructions,"a"; \
  890. .long 771b; \
  891. .byte ptype; \
  892. .byte 772b-771b; \
  893. .short clobbers; \
  894. .popsection
  895. #define INTERRUPT_RETURN \
  896. PARA_SITE(PARA_PATCH(PARAVIRT_iret), CLBR_NONE, \
  897. jmp *%cs:paravirt_ops+PARAVIRT_iret)
  898. #define DISABLE_INTERRUPTS(clobbers) \
  899. PARA_SITE(PARA_PATCH(PARAVIRT_irq_disable), clobbers, \
  900. pushl %eax; pushl %ecx; pushl %edx; \
  901. call *%cs:paravirt_ops+PARAVIRT_irq_disable; \
  902. popl %edx; popl %ecx; popl %eax) \
  903. #define ENABLE_INTERRUPTS(clobbers) \
  904. PARA_SITE(PARA_PATCH(PARAVIRT_irq_enable), clobbers, \
  905. pushl %eax; pushl %ecx; pushl %edx; \
  906. call *%cs:paravirt_ops+PARAVIRT_irq_enable; \
  907. popl %edx; popl %ecx; popl %eax)
  908. #define ENABLE_INTERRUPTS_SYSEXIT \
  909. PARA_SITE(PARA_PATCH(PARAVIRT_irq_enable_sysexit), CLBR_NONE, \
  910. jmp *%cs:paravirt_ops+PARAVIRT_irq_enable_sysexit)
  911. #define GET_CR0_INTO_EAX \
  912. push %ecx; push %edx; \
  913. call *paravirt_ops+PARAVIRT_read_cr0; \
  914. pop %edx; pop %ecx
  915. #endif /* __ASSEMBLY__ */
  916. #endif /* CONFIG_PARAVIRT */
  917. #endif /* __ASM_PARAVIRT_H */