enlighten.c 37 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598
  1. /*
  2. * Core of Xen paravirt_ops implementation.
  3. *
  4. * This file contains the xen_paravirt_ops structure itself, and the
  5. * implementations for:
  6. * - privileged instructions
  7. * - interrupt flags
  8. * - segment operations
  9. * - booting and setup
  10. *
  11. * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  12. */
  13. #include <linux/cpu.h>
  14. #include <linux/kernel.h>
  15. #include <linux/init.h>
  16. #include <linux/smp.h>
  17. #include <linux/preempt.h>
  18. #include <linux/hardirq.h>
  19. #include <linux/percpu.h>
  20. #include <linux/delay.h>
  21. #include <linux/start_kernel.h>
  22. #include <linux/sched.h>
  23. #include <linux/kprobes.h>
  24. #include <linux/bootmem.h>
  25. #include <linux/module.h>
  26. #include <linux/mm.h>
  27. #include <linux/page-flags.h>
  28. #include <linux/highmem.h>
  29. #include <linux/console.h>
  30. #include <linux/pci.h>
  31. #include <linux/gfp.h>
  32. #include <linux/memblock.h>
  33. #include <xen/xen.h>
  34. #include <xen/interface/xen.h>
  35. #include <xen/interface/version.h>
  36. #include <xen/interface/physdev.h>
  37. #include <xen/interface/vcpu.h>
  38. #include <xen/interface/memory.h>
  39. #include <xen/interface/xen-mca.h>
  40. #include <xen/features.h>
  41. #include <xen/page.h>
  42. #include <xen/hvm.h>
  43. #include <xen/hvc-console.h>
  44. #include <xen/acpi.h>
  45. #include <asm/paravirt.h>
  46. #include <asm/apic.h>
  47. #include <asm/page.h>
  48. #include <asm/xen/pci.h>
  49. #include <asm/xen/hypercall.h>
  50. #include <asm/xen/hypervisor.h>
  51. #include <asm/fixmap.h>
  52. #include <asm/processor.h>
  53. #include <asm/proto.h>
  54. #include <asm/msr-index.h>
  55. #include <asm/traps.h>
  56. #include <asm/setup.h>
  57. #include <asm/desc.h>
  58. #include <asm/pgalloc.h>
  59. #include <asm/pgtable.h>
  60. #include <asm/tlbflush.h>
  61. #include <asm/reboot.h>
  62. #include <asm/stackprotector.h>
  63. #include <asm/hypervisor.h>
  64. #include <asm/mwait.h>
  65. #include <asm/pci_x86.h>
  66. #ifdef CONFIG_ACPI
  67. #include <linux/acpi.h>
  68. #include <asm/acpi.h>
  69. #include <acpi/pdc_intel.h>
  70. #include <acpi/processor.h>
  71. #include <xen/interface/platform.h>
  72. #endif
  73. #include "xen-ops.h"
  74. #include "mmu.h"
  75. #include "smp.h"
  76. #include "multicalls.h"
  77. EXPORT_SYMBOL_GPL(hypercall_page);
  78. DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  79. DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
  80. enum xen_domain_type xen_domain_type = XEN_NATIVE;
  81. EXPORT_SYMBOL_GPL(xen_domain_type);
  82. unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
  83. EXPORT_SYMBOL(machine_to_phys_mapping);
  84. unsigned long machine_to_phys_nr;
  85. EXPORT_SYMBOL(machine_to_phys_nr);
  86. struct start_info *xen_start_info;
  87. EXPORT_SYMBOL_GPL(xen_start_info);
  88. struct shared_info xen_dummy_shared_info;
  89. void *xen_initial_gdt;
  90. RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
  91. __read_mostly int xen_have_vector_callback;
  92. EXPORT_SYMBOL_GPL(xen_have_vector_callback);
  93. /*
  94. * Point at some empty memory to start with. We map the real shared_info
  95. * page as soon as fixmap is up and running.
  96. */
  97. struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
  98. /*
  99. * Flag to determine whether vcpu info placement is available on all
  100. * VCPUs. We assume it is to start with, and then set it to zero on
  101. * the first failure. This is because it can succeed on some VCPUs
  102. * and not others, since it can involve hypervisor memory allocation,
  103. * or because the guest failed to guarantee all the appropriate
  104. * constraints on all VCPUs (ie buffer can't cross a page boundary).
  105. *
  106. * Note that any particular CPU may be using a placed vcpu structure,
  107. * but we can only optimise if the all are.
  108. *
  109. * 0: not available, 1: available
  110. */
  111. static int have_vcpu_info_placement = 1;
  112. struct tls_descs {
  113. struct desc_struct desc[3];
  114. };
  115. /*
  116. * Updating the 3 TLS descriptors in the GDT on every task switch is
  117. * surprisingly expensive so we avoid updating them if they haven't
  118. * changed. Since Xen writes different descriptors than the one
  119. * passed in the update_descriptor hypercall we keep shadow copies to
  120. * compare against.
  121. */
  122. static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
  123. static void clamp_max_cpus(void)
  124. {
  125. #ifdef CONFIG_SMP
  126. if (setup_max_cpus > MAX_VIRT_CPUS)
  127. setup_max_cpus = MAX_VIRT_CPUS;
  128. #endif
  129. }
  130. static void xen_vcpu_setup(int cpu)
  131. {
  132. struct vcpu_register_vcpu_info info;
  133. int err;
  134. struct vcpu_info *vcpup;
  135. BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
  136. if (cpu < MAX_VIRT_CPUS)
  137. per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
  138. if (!have_vcpu_info_placement) {
  139. if (cpu >= MAX_VIRT_CPUS)
  140. clamp_max_cpus();
  141. return;
  142. }
  143. vcpup = &per_cpu(xen_vcpu_info, cpu);
  144. info.mfn = arbitrary_virt_to_mfn(vcpup);
  145. info.offset = offset_in_page(vcpup);
  146. /* Check to see if the hypervisor will put the vcpu_info
  147. structure where we want it, which allows direct access via
  148. a percpu-variable. */
  149. err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
  150. if (err) {
  151. printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
  152. have_vcpu_info_placement = 0;
  153. clamp_max_cpus();
  154. } else {
  155. /* This cpu is using the registered vcpu info, even if
  156. later ones fail to. */
  157. per_cpu(xen_vcpu, cpu) = vcpup;
  158. }
  159. }
  160. /*
  161. * On restore, set the vcpu placement up again.
  162. * If it fails, then we're in a bad state, since
  163. * we can't back out from using it...
  164. */
  165. void xen_vcpu_restore(void)
  166. {
  167. int cpu;
  168. for_each_online_cpu(cpu) {
  169. bool other_cpu = (cpu != smp_processor_id());
  170. if (other_cpu &&
  171. HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
  172. BUG();
  173. xen_setup_runstate_info(cpu);
  174. if (have_vcpu_info_placement)
  175. xen_vcpu_setup(cpu);
  176. if (other_cpu &&
  177. HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
  178. BUG();
  179. }
  180. }
  181. static void __init xen_banner(void)
  182. {
  183. unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
  184. struct xen_extraversion extra;
  185. HYPERVISOR_xen_version(XENVER_extraversion, &extra);
  186. printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
  187. pv_info.name);
  188. printk(KERN_INFO "Xen version: %d.%d%s%s\n",
  189. version >> 16, version & 0xffff, extra.extraversion,
  190. xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
  191. }
  192. #define CPUID_THERM_POWER_LEAF 6
  193. #define APERFMPERF_PRESENT 0
  194. static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
  195. static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
  196. static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
  197. static __read_mostly unsigned int cpuid_leaf5_ecx_val;
  198. static __read_mostly unsigned int cpuid_leaf5_edx_val;
  199. static void xen_cpuid(unsigned int *ax, unsigned int *bx,
  200. unsigned int *cx, unsigned int *dx)
  201. {
  202. unsigned maskebx = ~0;
  203. unsigned maskecx = ~0;
  204. unsigned maskedx = ~0;
  205. unsigned setecx = 0;
  206. /*
  207. * Mask out inconvenient features, to try and disable as many
  208. * unsupported kernel subsystems as possible.
  209. */
  210. switch (*ax) {
  211. case 1:
  212. maskecx = cpuid_leaf1_ecx_mask;
  213. setecx = cpuid_leaf1_ecx_set_mask;
  214. maskedx = cpuid_leaf1_edx_mask;
  215. break;
  216. case CPUID_MWAIT_LEAF:
  217. /* Synthesize the values.. */
  218. *ax = 0;
  219. *bx = 0;
  220. *cx = cpuid_leaf5_ecx_val;
  221. *dx = cpuid_leaf5_edx_val;
  222. return;
  223. case CPUID_THERM_POWER_LEAF:
  224. /* Disabling APERFMPERF for kernel usage */
  225. maskecx = ~(1 << APERFMPERF_PRESENT);
  226. break;
  227. case 0xb:
  228. /* Suppress extended topology stuff */
  229. maskebx = 0;
  230. break;
  231. }
  232. asm(XEN_EMULATE_PREFIX "cpuid"
  233. : "=a" (*ax),
  234. "=b" (*bx),
  235. "=c" (*cx),
  236. "=d" (*dx)
  237. : "0" (*ax), "2" (*cx));
  238. *bx &= maskebx;
  239. *cx &= maskecx;
  240. *cx |= setecx;
  241. *dx &= maskedx;
  242. }
  243. static bool __init xen_check_mwait(void)
  244. {
  245. #if defined(CONFIG_ACPI) && !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR) && \
  246. !defined(CONFIG_ACPI_PROCESSOR_AGGREGATOR_MODULE)
  247. struct xen_platform_op op = {
  248. .cmd = XENPF_set_processor_pminfo,
  249. .u.set_pminfo.id = -1,
  250. .u.set_pminfo.type = XEN_PM_PDC,
  251. };
  252. uint32_t buf[3];
  253. unsigned int ax, bx, cx, dx;
  254. unsigned int mwait_mask;
  255. /* We need to determine whether it is OK to expose the MWAIT
  256. * capability to the kernel to harvest deeper than C3 states from ACPI
  257. * _CST using the processor_harvest_xen.c module. For this to work, we
  258. * need to gather the MWAIT_LEAF values (which the cstate.c code
  259. * checks against). The hypervisor won't expose the MWAIT flag because
  260. * it would break backwards compatibility; so we will find out directly
  261. * from the hardware and hypercall.
  262. */
  263. if (!xen_initial_domain())
  264. return false;
  265. ax = 1;
  266. cx = 0;
  267. native_cpuid(&ax, &bx, &cx, &dx);
  268. mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
  269. (1 << (X86_FEATURE_MWAIT % 32));
  270. if ((cx & mwait_mask) != mwait_mask)
  271. return false;
  272. /* We need to emulate the MWAIT_LEAF and for that we need both
  273. * ecx and edx. The hypercall provides only partial information.
  274. */
  275. ax = CPUID_MWAIT_LEAF;
  276. bx = 0;
  277. cx = 0;
  278. dx = 0;
  279. native_cpuid(&ax, &bx, &cx, &dx);
  280. /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
  281. * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
  282. */
  283. buf[0] = ACPI_PDC_REVISION_ID;
  284. buf[1] = 1;
  285. buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
  286. set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
  287. if ((HYPERVISOR_dom0_op(&op) == 0) &&
  288. (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
  289. cpuid_leaf5_ecx_val = cx;
  290. cpuid_leaf5_edx_val = dx;
  291. }
  292. return true;
  293. #else
  294. return false;
  295. #endif
  296. }
  297. static void __init xen_init_cpuid_mask(void)
  298. {
  299. unsigned int ax, bx, cx, dx;
  300. unsigned int xsave_mask;
  301. cpuid_leaf1_edx_mask =
  302. ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
  303. (1 << X86_FEATURE_ACC)); /* thermal monitoring */
  304. if (!xen_initial_domain())
  305. cpuid_leaf1_edx_mask &=
  306. ~((1 << X86_FEATURE_APIC) | /* disable local APIC */
  307. (1 << X86_FEATURE_ACPI)); /* disable ACPI */
  308. ax = 1;
  309. cx = 0;
  310. xen_cpuid(&ax, &bx, &cx, &dx);
  311. xsave_mask =
  312. (1 << (X86_FEATURE_XSAVE % 32)) |
  313. (1 << (X86_FEATURE_OSXSAVE % 32));
  314. /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
  315. if ((cx & xsave_mask) != xsave_mask)
  316. cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
  317. if (xen_check_mwait())
  318. cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
  319. }
  320. static void xen_set_debugreg(int reg, unsigned long val)
  321. {
  322. HYPERVISOR_set_debugreg(reg, val);
  323. }
  324. static unsigned long xen_get_debugreg(int reg)
  325. {
  326. return HYPERVISOR_get_debugreg(reg);
  327. }
  328. static void xen_end_context_switch(struct task_struct *next)
  329. {
  330. xen_mc_flush();
  331. paravirt_end_context_switch(next);
  332. }
  333. static unsigned long xen_store_tr(void)
  334. {
  335. return 0;
  336. }
  337. /*
  338. * Set the page permissions for a particular virtual address. If the
  339. * address is a vmalloc mapping (or other non-linear mapping), then
  340. * find the linear mapping of the page and also set its protections to
  341. * match.
  342. */
  343. static void set_aliased_prot(void *v, pgprot_t prot)
  344. {
  345. int level;
  346. pte_t *ptep;
  347. pte_t pte;
  348. unsigned long pfn;
  349. struct page *page;
  350. ptep = lookup_address((unsigned long)v, &level);
  351. BUG_ON(ptep == NULL);
  352. pfn = pte_pfn(*ptep);
  353. page = pfn_to_page(pfn);
  354. pte = pfn_pte(pfn, prot);
  355. if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
  356. BUG();
  357. if (!PageHighMem(page)) {
  358. void *av = __va(PFN_PHYS(pfn));
  359. if (av != v)
  360. if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
  361. BUG();
  362. } else
  363. kmap_flush_unused();
  364. }
  365. static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
  366. {
  367. const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
  368. int i;
  369. for(i = 0; i < entries; i += entries_per_page)
  370. set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
  371. }
  372. static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
  373. {
  374. const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
  375. int i;
  376. for(i = 0; i < entries; i += entries_per_page)
  377. set_aliased_prot(ldt + i, PAGE_KERNEL);
  378. }
  379. static void xen_set_ldt(const void *addr, unsigned entries)
  380. {
  381. struct mmuext_op *op;
  382. struct multicall_space mcs = xen_mc_entry(sizeof(*op));
  383. trace_xen_cpu_set_ldt(addr, entries);
  384. op = mcs.args;
  385. op->cmd = MMUEXT_SET_LDT;
  386. op->arg1.linear_addr = (unsigned long)addr;
  387. op->arg2.nr_ents = entries;
  388. MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
  389. xen_mc_issue(PARAVIRT_LAZY_CPU);
  390. }
  391. static void xen_load_gdt(const struct desc_ptr *dtr)
  392. {
  393. unsigned long va = dtr->address;
  394. unsigned int size = dtr->size + 1;
  395. unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
  396. unsigned long frames[pages];
  397. int f;
  398. /*
  399. * A GDT can be up to 64k in size, which corresponds to 8192
  400. * 8-byte entries, or 16 4k pages..
  401. */
  402. BUG_ON(size > 65536);
  403. BUG_ON(va & ~PAGE_MASK);
  404. for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
  405. int level;
  406. pte_t *ptep;
  407. unsigned long pfn, mfn;
  408. void *virt;
  409. /*
  410. * The GDT is per-cpu and is in the percpu data area.
  411. * That can be virtually mapped, so we need to do a
  412. * page-walk to get the underlying MFN for the
  413. * hypercall. The page can also be in the kernel's
  414. * linear range, so we need to RO that mapping too.
  415. */
  416. ptep = lookup_address(va, &level);
  417. BUG_ON(ptep == NULL);
  418. pfn = pte_pfn(*ptep);
  419. mfn = pfn_to_mfn(pfn);
  420. virt = __va(PFN_PHYS(pfn));
  421. frames[f] = mfn;
  422. make_lowmem_page_readonly((void *)va);
  423. make_lowmem_page_readonly(virt);
  424. }
  425. if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
  426. BUG();
  427. }
  428. /*
  429. * load_gdt for early boot, when the gdt is only mapped once
  430. */
  431. static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
  432. {
  433. unsigned long va = dtr->address;
  434. unsigned int size = dtr->size + 1;
  435. unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
  436. unsigned long frames[pages];
  437. int f;
  438. /*
  439. * A GDT can be up to 64k in size, which corresponds to 8192
  440. * 8-byte entries, or 16 4k pages..
  441. */
  442. BUG_ON(size > 65536);
  443. BUG_ON(va & ~PAGE_MASK);
  444. for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
  445. pte_t pte;
  446. unsigned long pfn, mfn;
  447. pfn = virt_to_pfn(va);
  448. mfn = pfn_to_mfn(pfn);
  449. pte = pfn_pte(pfn, PAGE_KERNEL_RO);
  450. if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
  451. BUG();
  452. frames[f] = mfn;
  453. }
  454. if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
  455. BUG();
  456. }
  457. static inline bool desc_equal(const struct desc_struct *d1,
  458. const struct desc_struct *d2)
  459. {
  460. return d1->a == d2->a && d1->b == d2->b;
  461. }
  462. static void load_TLS_descriptor(struct thread_struct *t,
  463. unsigned int cpu, unsigned int i)
  464. {
  465. struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
  466. struct desc_struct *gdt;
  467. xmaddr_t maddr;
  468. struct multicall_space mc;
  469. if (desc_equal(shadow, &t->tls_array[i]))
  470. return;
  471. *shadow = t->tls_array[i];
  472. gdt = get_cpu_gdt_table(cpu);
  473. maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
  474. mc = __xen_mc_entry(0);
  475. MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
  476. }
  477. static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
  478. {
  479. /*
  480. * XXX sleazy hack: If we're being called in a lazy-cpu zone
  481. * and lazy gs handling is enabled, it means we're in a
  482. * context switch, and %gs has just been saved. This means we
  483. * can zero it out to prevent faults on exit from the
  484. * hypervisor if the next process has no %gs. Either way, it
  485. * has been saved, and the new value will get loaded properly.
  486. * This will go away as soon as Xen has been modified to not
  487. * save/restore %gs for normal hypercalls.
  488. *
  489. * On x86_64, this hack is not used for %gs, because gs points
  490. * to KERNEL_GS_BASE (and uses it for PDA references), so we
  491. * must not zero %gs on x86_64
  492. *
  493. * For x86_64, we need to zero %fs, otherwise we may get an
  494. * exception between the new %fs descriptor being loaded and
  495. * %fs being effectively cleared at __switch_to().
  496. */
  497. if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
  498. #ifdef CONFIG_X86_32
  499. lazy_load_gs(0);
  500. #else
  501. loadsegment(fs, 0);
  502. #endif
  503. }
  504. xen_mc_batch();
  505. load_TLS_descriptor(t, cpu, 0);
  506. load_TLS_descriptor(t, cpu, 1);
  507. load_TLS_descriptor(t, cpu, 2);
  508. xen_mc_issue(PARAVIRT_LAZY_CPU);
  509. }
  510. #ifdef CONFIG_X86_64
  511. static void xen_load_gs_index(unsigned int idx)
  512. {
  513. if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
  514. BUG();
  515. }
  516. #endif
  517. static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
  518. const void *ptr)
  519. {
  520. xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
  521. u64 entry = *(u64 *)ptr;
  522. trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
  523. preempt_disable();
  524. xen_mc_flush();
  525. if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
  526. BUG();
  527. preempt_enable();
  528. }
  529. static int cvt_gate_to_trap(int vector, const gate_desc *val,
  530. struct trap_info *info)
  531. {
  532. unsigned long addr;
  533. if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
  534. return 0;
  535. info->vector = vector;
  536. addr = gate_offset(*val);
  537. #ifdef CONFIG_X86_64
  538. /*
  539. * Look for known traps using IST, and substitute them
  540. * appropriately. The debugger ones are the only ones we care
  541. * about. Xen will handle faults like double_fault,
  542. * so we should never see them. Warn if
  543. * there's an unexpected IST-using fault handler.
  544. */
  545. if (addr == (unsigned long)debug)
  546. addr = (unsigned long)xen_debug;
  547. else if (addr == (unsigned long)int3)
  548. addr = (unsigned long)xen_int3;
  549. else if (addr == (unsigned long)stack_segment)
  550. addr = (unsigned long)xen_stack_segment;
  551. else if (addr == (unsigned long)double_fault ||
  552. addr == (unsigned long)nmi) {
  553. /* Don't need to handle these */
  554. return 0;
  555. #ifdef CONFIG_X86_MCE
  556. } else if (addr == (unsigned long)machine_check) {
  557. /*
  558. * when xen hypervisor inject vMCE to guest,
  559. * use native mce handler to handle it
  560. */
  561. ;
  562. #endif
  563. } else {
  564. /* Some other trap using IST? */
  565. if (WARN_ON(val->ist != 0))
  566. return 0;
  567. }
  568. #endif /* CONFIG_X86_64 */
  569. info->address = addr;
  570. info->cs = gate_segment(*val);
  571. info->flags = val->dpl;
  572. /* interrupt gates clear IF */
  573. if (val->type == GATE_INTERRUPT)
  574. info->flags |= 1 << 2;
  575. return 1;
  576. }
  577. /* Locations of each CPU's IDT */
  578. static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
  579. /* Set an IDT entry. If the entry is part of the current IDT, then
  580. also update Xen. */
  581. static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
  582. {
  583. unsigned long p = (unsigned long)&dt[entrynum];
  584. unsigned long start, end;
  585. trace_xen_cpu_write_idt_entry(dt, entrynum, g);
  586. preempt_disable();
  587. start = __this_cpu_read(idt_desc.address);
  588. end = start + __this_cpu_read(idt_desc.size) + 1;
  589. xen_mc_flush();
  590. native_write_idt_entry(dt, entrynum, g);
  591. if (p >= start && (p + 8) <= end) {
  592. struct trap_info info[2];
  593. info[1].address = 0;
  594. if (cvt_gate_to_trap(entrynum, g, &info[0]))
  595. if (HYPERVISOR_set_trap_table(info))
  596. BUG();
  597. }
  598. preempt_enable();
  599. }
  600. static void xen_convert_trap_info(const struct desc_ptr *desc,
  601. struct trap_info *traps)
  602. {
  603. unsigned in, out, count;
  604. count = (desc->size+1) / sizeof(gate_desc);
  605. BUG_ON(count > 256);
  606. for (in = out = 0; in < count; in++) {
  607. gate_desc *entry = (gate_desc*)(desc->address) + in;
  608. if (cvt_gate_to_trap(in, entry, &traps[out]))
  609. out++;
  610. }
  611. traps[out].address = 0;
  612. }
  613. void xen_copy_trap_info(struct trap_info *traps)
  614. {
  615. const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
  616. xen_convert_trap_info(desc, traps);
  617. }
  618. /* Load a new IDT into Xen. In principle this can be per-CPU, so we
  619. hold a spinlock to protect the static traps[] array (static because
  620. it avoids allocation, and saves stack space). */
  621. static void xen_load_idt(const struct desc_ptr *desc)
  622. {
  623. static DEFINE_SPINLOCK(lock);
  624. static struct trap_info traps[257];
  625. trace_xen_cpu_load_idt(desc);
  626. spin_lock(&lock);
  627. __get_cpu_var(idt_desc) = *desc;
  628. xen_convert_trap_info(desc, traps);
  629. xen_mc_flush();
  630. if (HYPERVISOR_set_trap_table(traps))
  631. BUG();
  632. spin_unlock(&lock);
  633. }
  634. /* Write a GDT descriptor entry. Ignore LDT descriptors, since
  635. they're handled differently. */
  636. static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
  637. const void *desc, int type)
  638. {
  639. trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
  640. preempt_disable();
  641. switch (type) {
  642. case DESC_LDT:
  643. case DESC_TSS:
  644. /* ignore */
  645. break;
  646. default: {
  647. xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
  648. xen_mc_flush();
  649. if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
  650. BUG();
  651. }
  652. }
  653. preempt_enable();
  654. }
  655. /*
  656. * Version of write_gdt_entry for use at early boot-time needed to
  657. * update an entry as simply as possible.
  658. */
  659. static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
  660. const void *desc, int type)
  661. {
  662. trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
  663. switch (type) {
  664. case DESC_LDT:
  665. case DESC_TSS:
  666. /* ignore */
  667. break;
  668. default: {
  669. xmaddr_t maddr = virt_to_machine(&dt[entry]);
  670. if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
  671. dt[entry] = *(struct desc_struct *)desc;
  672. }
  673. }
  674. }
  675. static void xen_load_sp0(struct tss_struct *tss,
  676. struct thread_struct *thread)
  677. {
  678. struct multicall_space mcs;
  679. mcs = xen_mc_entry(0);
  680. MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
  681. xen_mc_issue(PARAVIRT_LAZY_CPU);
  682. }
  683. static void xen_set_iopl_mask(unsigned mask)
  684. {
  685. struct physdev_set_iopl set_iopl;
  686. /* Force the change at ring 0. */
  687. set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
  688. HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
  689. }
  690. static void xen_io_delay(void)
  691. {
  692. }
  693. #ifdef CONFIG_X86_LOCAL_APIC
  694. static unsigned long xen_set_apic_id(unsigned int x)
  695. {
  696. WARN_ON(1);
  697. return x;
  698. }
  699. static unsigned int xen_get_apic_id(unsigned long x)
  700. {
  701. return ((x)>>24) & 0xFFu;
  702. }
  703. static u32 xen_apic_read(u32 reg)
  704. {
  705. struct xen_platform_op op = {
  706. .cmd = XENPF_get_cpuinfo,
  707. .interface_version = XENPF_INTERFACE_VERSION,
  708. .u.pcpu_info.xen_cpuid = 0,
  709. };
  710. int ret = 0;
  711. /* Shouldn't need this as APIC is turned off for PV, and we only
  712. * get called on the bootup processor. But just in case. */
  713. if (!xen_initial_domain() || smp_processor_id())
  714. return 0;
  715. if (reg == APIC_LVR)
  716. return 0x10;
  717. if (reg != APIC_ID)
  718. return 0;
  719. ret = HYPERVISOR_dom0_op(&op);
  720. if (ret)
  721. return 0;
  722. return op.u.pcpu_info.apic_id << 24;
  723. }
  724. static void xen_apic_write(u32 reg, u32 val)
  725. {
  726. /* Warn to see if there's any stray references */
  727. WARN_ON(1);
  728. }
  729. static u64 xen_apic_icr_read(void)
  730. {
  731. return 0;
  732. }
  733. static void xen_apic_icr_write(u32 low, u32 id)
  734. {
  735. /* Warn to see if there's any stray references */
  736. WARN_ON(1);
  737. }
  738. static void xen_apic_wait_icr_idle(void)
  739. {
  740. return;
  741. }
  742. static u32 xen_safe_apic_wait_icr_idle(void)
  743. {
  744. return 0;
  745. }
  746. static void set_xen_basic_apic_ops(void)
  747. {
  748. apic->read = xen_apic_read;
  749. apic->write = xen_apic_write;
  750. apic->icr_read = xen_apic_icr_read;
  751. apic->icr_write = xen_apic_icr_write;
  752. apic->wait_icr_idle = xen_apic_wait_icr_idle;
  753. apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
  754. apic->set_apic_id = xen_set_apic_id;
  755. apic->get_apic_id = xen_get_apic_id;
  756. #ifdef CONFIG_SMP
  757. apic->send_IPI_allbutself = xen_send_IPI_allbutself;
  758. apic->send_IPI_mask_allbutself = xen_send_IPI_mask_allbutself;
  759. apic->send_IPI_mask = xen_send_IPI_mask;
  760. apic->send_IPI_all = xen_send_IPI_all;
  761. apic->send_IPI_self = xen_send_IPI_self;
  762. #endif
  763. }
  764. #endif
  765. static void xen_clts(void)
  766. {
  767. struct multicall_space mcs;
  768. mcs = xen_mc_entry(0);
  769. MULTI_fpu_taskswitch(mcs.mc, 0);
  770. xen_mc_issue(PARAVIRT_LAZY_CPU);
  771. }
  772. static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
  773. static unsigned long xen_read_cr0(void)
  774. {
  775. unsigned long cr0 = this_cpu_read(xen_cr0_value);
  776. if (unlikely(cr0 == 0)) {
  777. cr0 = native_read_cr0();
  778. this_cpu_write(xen_cr0_value, cr0);
  779. }
  780. return cr0;
  781. }
  782. static void xen_write_cr0(unsigned long cr0)
  783. {
  784. struct multicall_space mcs;
  785. this_cpu_write(xen_cr0_value, cr0);
  786. /* Only pay attention to cr0.TS; everything else is
  787. ignored. */
  788. mcs = xen_mc_entry(0);
  789. MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
  790. xen_mc_issue(PARAVIRT_LAZY_CPU);
  791. }
  792. static void xen_write_cr4(unsigned long cr4)
  793. {
  794. cr4 &= ~X86_CR4_PGE;
  795. cr4 &= ~X86_CR4_PSE;
  796. native_write_cr4(cr4);
  797. }
  798. static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
  799. {
  800. int ret;
  801. ret = 0;
  802. switch (msr) {
  803. #ifdef CONFIG_X86_64
  804. unsigned which;
  805. u64 base;
  806. case MSR_FS_BASE: which = SEGBASE_FS; goto set;
  807. case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
  808. case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
  809. set:
  810. base = ((u64)high << 32) | low;
  811. if (HYPERVISOR_set_segment_base(which, base) != 0)
  812. ret = -EIO;
  813. break;
  814. #endif
  815. case MSR_STAR:
  816. case MSR_CSTAR:
  817. case MSR_LSTAR:
  818. case MSR_SYSCALL_MASK:
  819. case MSR_IA32_SYSENTER_CS:
  820. case MSR_IA32_SYSENTER_ESP:
  821. case MSR_IA32_SYSENTER_EIP:
  822. /* Fast syscall setup is all done in hypercalls, so
  823. these are all ignored. Stub them out here to stop
  824. Xen console noise. */
  825. break;
  826. case MSR_IA32_CR_PAT:
  827. if (smp_processor_id() == 0)
  828. xen_set_pat(((u64)high << 32) | low);
  829. break;
  830. default:
  831. ret = native_write_msr_safe(msr, low, high);
  832. }
  833. return ret;
  834. }
  835. void xen_setup_shared_info(void)
  836. {
  837. if (!xen_feature(XENFEAT_auto_translated_physmap)) {
  838. set_fixmap(FIX_PARAVIRT_BOOTMAP,
  839. xen_start_info->shared_info);
  840. HYPERVISOR_shared_info =
  841. (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
  842. } else
  843. HYPERVISOR_shared_info =
  844. (struct shared_info *)__va(xen_start_info->shared_info);
  845. #ifndef CONFIG_SMP
  846. /* In UP this is as good a place as any to set up shared info */
  847. xen_setup_vcpu_info_placement();
  848. #endif
  849. xen_setup_mfn_list_list();
  850. }
  851. /* This is called once we have the cpu_possible_mask */
  852. void xen_setup_vcpu_info_placement(void)
  853. {
  854. int cpu;
  855. for_each_possible_cpu(cpu)
  856. xen_vcpu_setup(cpu);
  857. /* xen_vcpu_setup managed to place the vcpu_info within the
  858. percpu area for all cpus, so make use of it */
  859. if (have_vcpu_info_placement) {
  860. pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
  861. pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
  862. pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
  863. pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
  864. pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
  865. }
  866. }
  867. static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
  868. unsigned long addr, unsigned len)
  869. {
  870. char *start, *end, *reloc;
  871. unsigned ret;
  872. start = end = reloc = NULL;
  873. #define SITE(op, x) \
  874. case PARAVIRT_PATCH(op.x): \
  875. if (have_vcpu_info_placement) { \
  876. start = (char *)xen_##x##_direct; \
  877. end = xen_##x##_direct_end; \
  878. reloc = xen_##x##_direct_reloc; \
  879. } \
  880. goto patch_site
  881. switch (type) {
  882. SITE(pv_irq_ops, irq_enable);
  883. SITE(pv_irq_ops, irq_disable);
  884. SITE(pv_irq_ops, save_fl);
  885. SITE(pv_irq_ops, restore_fl);
  886. #undef SITE
  887. patch_site:
  888. if (start == NULL || (end-start) > len)
  889. goto default_patch;
  890. ret = paravirt_patch_insns(insnbuf, len, start, end);
  891. /* Note: because reloc is assigned from something that
  892. appears to be an array, gcc assumes it's non-null,
  893. but doesn't know its relationship with start and
  894. end. */
  895. if (reloc > start && reloc < end) {
  896. int reloc_off = reloc - start;
  897. long *relocp = (long *)(insnbuf + reloc_off);
  898. long delta = start - (char *)addr;
  899. *relocp += delta;
  900. }
  901. break;
  902. default_patch:
  903. default:
  904. ret = paravirt_patch_default(type, clobbers, insnbuf,
  905. addr, len);
  906. break;
  907. }
  908. return ret;
  909. }
  910. static const struct pv_info xen_info __initconst = {
  911. .paravirt_enabled = 1,
  912. .shared_kernel_pmd = 0,
  913. #ifdef CONFIG_X86_64
  914. .extra_user_64bit_cs = FLAT_USER_CS64,
  915. #endif
  916. .name = "Xen",
  917. };
  918. static const struct pv_init_ops xen_init_ops __initconst = {
  919. .patch = xen_patch,
  920. };
  921. static const struct pv_cpu_ops xen_cpu_ops __initconst = {
  922. .cpuid = xen_cpuid,
  923. .set_debugreg = xen_set_debugreg,
  924. .get_debugreg = xen_get_debugreg,
  925. .clts = xen_clts,
  926. .read_cr0 = xen_read_cr0,
  927. .write_cr0 = xen_write_cr0,
  928. .read_cr4 = native_read_cr4,
  929. .read_cr4_safe = native_read_cr4_safe,
  930. .write_cr4 = xen_write_cr4,
  931. .wbinvd = native_wbinvd,
  932. .read_msr = native_read_msr_safe,
  933. .write_msr = xen_write_msr_safe,
  934. .read_tsc = native_read_tsc,
  935. .read_pmc = native_read_pmc,
  936. .iret = xen_iret,
  937. .irq_enable_sysexit = xen_sysexit,
  938. #ifdef CONFIG_X86_64
  939. .usergs_sysret32 = xen_sysret32,
  940. .usergs_sysret64 = xen_sysret64,
  941. #endif
  942. .load_tr_desc = paravirt_nop,
  943. .set_ldt = xen_set_ldt,
  944. .load_gdt = xen_load_gdt,
  945. .load_idt = xen_load_idt,
  946. .load_tls = xen_load_tls,
  947. #ifdef CONFIG_X86_64
  948. .load_gs_index = xen_load_gs_index,
  949. #endif
  950. .alloc_ldt = xen_alloc_ldt,
  951. .free_ldt = xen_free_ldt,
  952. .store_gdt = native_store_gdt,
  953. .store_idt = native_store_idt,
  954. .store_tr = xen_store_tr,
  955. .write_ldt_entry = xen_write_ldt_entry,
  956. .write_gdt_entry = xen_write_gdt_entry,
  957. .write_idt_entry = xen_write_idt_entry,
  958. .load_sp0 = xen_load_sp0,
  959. .set_iopl_mask = xen_set_iopl_mask,
  960. .io_delay = xen_io_delay,
  961. /* Xen takes care of %gs when switching to usermode for us */
  962. .swapgs = paravirt_nop,
  963. .start_context_switch = paravirt_start_context_switch,
  964. .end_context_switch = xen_end_context_switch,
  965. };
  966. static const struct pv_apic_ops xen_apic_ops __initconst = {
  967. #ifdef CONFIG_X86_LOCAL_APIC
  968. .startup_ipi_hook = paravirt_nop,
  969. #endif
  970. };
  971. static void xen_reboot(int reason)
  972. {
  973. struct sched_shutdown r = { .reason = reason };
  974. if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
  975. BUG();
  976. }
  977. static void xen_restart(char *msg)
  978. {
  979. xen_reboot(SHUTDOWN_reboot);
  980. }
  981. static void xen_emergency_restart(void)
  982. {
  983. xen_reboot(SHUTDOWN_reboot);
  984. }
  985. static void xen_machine_halt(void)
  986. {
  987. xen_reboot(SHUTDOWN_poweroff);
  988. }
  989. static void xen_machine_power_off(void)
  990. {
  991. if (pm_power_off)
  992. pm_power_off();
  993. xen_reboot(SHUTDOWN_poweroff);
  994. }
  995. static void xen_crash_shutdown(struct pt_regs *regs)
  996. {
  997. xen_reboot(SHUTDOWN_crash);
  998. }
  999. static int
  1000. xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
  1001. {
  1002. xen_reboot(SHUTDOWN_crash);
  1003. return NOTIFY_DONE;
  1004. }
  1005. static struct notifier_block xen_panic_block = {
  1006. .notifier_call= xen_panic_event,
  1007. };
  1008. int xen_panic_handler_init(void)
  1009. {
  1010. atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
  1011. return 0;
  1012. }
  1013. static const struct machine_ops xen_machine_ops __initconst = {
  1014. .restart = xen_restart,
  1015. .halt = xen_machine_halt,
  1016. .power_off = xen_machine_power_off,
  1017. .shutdown = xen_machine_halt,
  1018. .crash_shutdown = xen_crash_shutdown,
  1019. .emergency_restart = xen_emergency_restart,
  1020. };
  1021. /*
  1022. * Set up the GDT and segment registers for -fstack-protector. Until
  1023. * we do this, we have to be careful not to call any stack-protected
  1024. * function, which is most of the kernel.
  1025. */
  1026. static void __init xen_setup_stackprotector(void)
  1027. {
  1028. pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
  1029. pv_cpu_ops.load_gdt = xen_load_gdt_boot;
  1030. setup_stack_canary_segment(0);
  1031. switch_to_new_gdt(0);
  1032. pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
  1033. pv_cpu_ops.load_gdt = xen_load_gdt;
  1034. }
  1035. /* First C function to be called on Xen boot */
  1036. asmlinkage void __init xen_start_kernel(void)
  1037. {
  1038. struct physdev_set_iopl set_iopl;
  1039. int rc;
  1040. pgd_t *pgd;
  1041. if (!xen_start_info)
  1042. return;
  1043. xen_domain_type = XEN_PV_DOMAIN;
  1044. xen_setup_machphys_mapping();
  1045. /* Install Xen paravirt ops */
  1046. pv_info = xen_info;
  1047. pv_init_ops = xen_init_ops;
  1048. pv_cpu_ops = xen_cpu_ops;
  1049. pv_apic_ops = xen_apic_ops;
  1050. x86_init.resources.memory_setup = xen_memory_setup;
  1051. x86_init.oem.arch_setup = xen_arch_setup;
  1052. x86_init.oem.banner = xen_banner;
  1053. xen_init_time_ops();
  1054. /*
  1055. * Set up some pagetable state before starting to set any ptes.
  1056. */
  1057. xen_init_mmu_ops();
  1058. /* Prevent unwanted bits from being set in PTEs. */
  1059. __supported_pte_mask &= ~_PAGE_GLOBAL;
  1060. #if 0
  1061. if (!xen_initial_domain())
  1062. #endif
  1063. __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
  1064. __supported_pte_mask |= _PAGE_IOMAP;
  1065. /*
  1066. * Prevent page tables from being allocated in highmem, even
  1067. * if CONFIG_HIGHPTE is enabled.
  1068. */
  1069. __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
  1070. /* Work out if we support NX */
  1071. x86_configure_nx();
  1072. xen_setup_features();
  1073. /* Get mfn list */
  1074. if (!xen_feature(XENFEAT_auto_translated_physmap))
  1075. xen_build_dynamic_phys_to_machine();
  1076. /*
  1077. * Set up kernel GDT and segment registers, mainly so that
  1078. * -fstack-protector code can be executed.
  1079. */
  1080. xen_setup_stackprotector();
  1081. xen_init_irq_ops();
  1082. xen_init_cpuid_mask();
  1083. #ifdef CONFIG_X86_LOCAL_APIC
  1084. /*
  1085. * set up the basic apic ops.
  1086. */
  1087. set_xen_basic_apic_ops();
  1088. #endif
  1089. if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
  1090. pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
  1091. pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
  1092. }
  1093. machine_ops = xen_machine_ops;
  1094. /*
  1095. * The only reliable way to retain the initial address of the
  1096. * percpu gdt_page is to remember it here, so we can go and
  1097. * mark it RW later, when the initial percpu area is freed.
  1098. */
  1099. xen_initial_gdt = &per_cpu(gdt_page, 0);
  1100. xen_smp_init();
  1101. #ifdef CONFIG_ACPI_NUMA
  1102. /*
  1103. * The pages we from Xen are not related to machine pages, so
  1104. * any NUMA information the kernel tries to get from ACPI will
  1105. * be meaningless. Prevent it from trying.
  1106. */
  1107. acpi_numa = -1;
  1108. #endif
  1109. pgd = (pgd_t *)xen_start_info->pt_base;
  1110. /* Don't do the full vcpu_info placement stuff until we have a
  1111. possible map and a non-dummy shared_info. */
  1112. per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
  1113. local_irq_disable();
  1114. early_boot_irqs_disabled = true;
  1115. xen_raw_console_write("mapping kernel into physical memory\n");
  1116. pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
  1117. /* Allocate and initialize top and mid mfn levels for p2m structure */
  1118. xen_build_mfn_list_list();
  1119. /* keep using Xen gdt for now; no urgent need to change it */
  1120. #ifdef CONFIG_X86_32
  1121. pv_info.kernel_rpl = 1;
  1122. if (xen_feature(XENFEAT_supervisor_mode_kernel))
  1123. pv_info.kernel_rpl = 0;
  1124. #else
  1125. pv_info.kernel_rpl = 0;
  1126. #endif
  1127. /* set the limit of our address space */
  1128. xen_reserve_top();
  1129. /* We used to do this in xen_arch_setup, but that is too late on AMD
  1130. * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
  1131. * which pokes 0xcf8 port.
  1132. */
  1133. set_iopl.iopl = 1;
  1134. rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
  1135. if (rc != 0)
  1136. xen_raw_printk("physdev_op failed %d\n", rc);
  1137. #ifdef CONFIG_X86_32
  1138. /* set up basic CPUID stuff */
  1139. cpu_detect(&new_cpu_data);
  1140. new_cpu_data.hard_math = 1;
  1141. new_cpu_data.wp_works_ok = 1;
  1142. new_cpu_data.x86_capability[0] = cpuid_edx(1);
  1143. #endif
  1144. /* Poke various useful things into boot_params */
  1145. boot_params.hdr.type_of_loader = (9 << 4) | 0;
  1146. boot_params.hdr.ramdisk_image = xen_start_info->mod_start
  1147. ? __pa(xen_start_info->mod_start) : 0;
  1148. boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
  1149. boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
  1150. if (!xen_initial_domain()) {
  1151. add_preferred_console("xenboot", 0, NULL);
  1152. add_preferred_console("tty", 0, NULL);
  1153. add_preferred_console("hvc", 0, NULL);
  1154. if (pci_xen)
  1155. x86_init.pci.arch_init = pci_xen_init;
  1156. } else {
  1157. const struct dom0_vga_console_info *info =
  1158. (void *)((char *)xen_start_info +
  1159. xen_start_info->console.dom0.info_off);
  1160. xen_init_vga(info, xen_start_info->console.dom0.info_size);
  1161. xen_start_info->console.domU.mfn = 0;
  1162. xen_start_info->console.domU.evtchn = 0;
  1163. xen_init_apic();
  1164. /* Make sure ACS will be enabled */
  1165. pci_request_acs();
  1166. xen_acpi_sleep_register();
  1167. /* Avoid searching for BIOS MP tables */
  1168. x86_init.mpparse.find_smp_config = x86_init_noop;
  1169. x86_init.mpparse.get_smp_config = x86_init_uint_noop;
  1170. }
  1171. #ifdef CONFIG_PCI
  1172. /* PCI BIOS service won't work from a PV guest. */
  1173. pci_probe &= ~PCI_PROBE_BIOS;
  1174. #endif
  1175. xen_raw_console_write("about to get started...\n");
  1176. xen_setup_runstate_info(0);
  1177. /* Start the world */
  1178. #ifdef CONFIG_X86_32
  1179. i386_start_kernel();
  1180. #else
  1181. x86_64_start_reservations((char *)__pa_symbol(&boot_params));
  1182. #endif
  1183. }
  1184. void __ref xen_hvm_init_shared_info(void)
  1185. {
  1186. int cpu;
  1187. struct xen_add_to_physmap xatp;
  1188. static struct shared_info *shared_info_page = 0;
  1189. if (!shared_info_page)
  1190. shared_info_page = (struct shared_info *)
  1191. extend_brk(PAGE_SIZE, PAGE_SIZE);
  1192. xatp.domid = DOMID_SELF;
  1193. xatp.idx = 0;
  1194. xatp.space = XENMAPSPACE_shared_info;
  1195. xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
  1196. if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
  1197. BUG();
  1198. HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
  1199. /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
  1200. * page, we use it in the event channel upcall and in some pvclock
  1201. * related functions. We don't need the vcpu_info placement
  1202. * optimizations because we don't use any pv_mmu or pv_irq op on
  1203. * HVM.
  1204. * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
  1205. * online but xen_hvm_init_shared_info is run at resume time too and
  1206. * in that case multiple vcpus might be online. */
  1207. for_each_online_cpu(cpu) {
  1208. per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
  1209. }
  1210. }
  1211. #ifdef CONFIG_XEN_PVHVM
  1212. static void __init init_hvm_pv_info(void)
  1213. {
  1214. int major, minor;
  1215. uint32_t eax, ebx, ecx, edx, pages, msr, base;
  1216. u64 pfn;
  1217. base = xen_cpuid_base();
  1218. cpuid(base + 1, &eax, &ebx, &ecx, &edx);
  1219. major = eax >> 16;
  1220. minor = eax & 0xffff;
  1221. printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
  1222. cpuid(base + 2, &pages, &msr, &ecx, &edx);
  1223. pfn = __pa(hypercall_page);
  1224. wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
  1225. xen_setup_features();
  1226. pv_info.name = "Xen HVM";
  1227. xen_domain_type = XEN_HVM_DOMAIN;
  1228. }
  1229. static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
  1230. unsigned long action, void *hcpu)
  1231. {
  1232. int cpu = (long)hcpu;
  1233. switch (action) {
  1234. case CPU_UP_PREPARE:
  1235. xen_vcpu_setup(cpu);
  1236. if (xen_have_vector_callback)
  1237. xen_init_lock_cpu(cpu);
  1238. break;
  1239. default:
  1240. break;
  1241. }
  1242. return NOTIFY_OK;
  1243. }
  1244. static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
  1245. .notifier_call = xen_hvm_cpu_notify,
  1246. };
  1247. static void __init xen_hvm_guest_init(void)
  1248. {
  1249. init_hvm_pv_info();
  1250. xen_hvm_init_shared_info();
  1251. if (xen_feature(XENFEAT_hvm_callback_vector))
  1252. xen_have_vector_callback = 1;
  1253. xen_hvm_smp_init();
  1254. register_cpu_notifier(&xen_hvm_cpu_notifier);
  1255. xen_unplug_emulated_devices();
  1256. x86_init.irqs.intr_init = xen_init_IRQ;
  1257. xen_hvm_init_time_ops();
  1258. xen_hvm_init_mmu_ops();
  1259. }
  1260. static bool __init xen_hvm_platform(void)
  1261. {
  1262. if (xen_pv_domain())
  1263. return false;
  1264. if (!xen_cpuid_base())
  1265. return false;
  1266. return true;
  1267. }
  1268. bool xen_hvm_need_lapic(void)
  1269. {
  1270. if (xen_pv_domain())
  1271. return false;
  1272. if (!xen_hvm_domain())
  1273. return false;
  1274. if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
  1275. return false;
  1276. return true;
  1277. }
  1278. EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
  1279. const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
  1280. .name = "Xen HVM",
  1281. .detect = xen_hvm_platform,
  1282. .init_platform = xen_hvm_guest_init,
  1283. };
  1284. EXPORT_SYMBOL(x86_hyper_xen_hvm);
  1285. #endif