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
14#include <linux/cpu.h>
15#include <linux/kernel.h>
16#include <linux/init.h>
17#include <linux/smp.h>
18#include <linux/preempt.h>
19#include <linux/hardirq.h>
20#include <linux/percpu.h>
21#include <linux/delay.h>
22#include <linux/start_kernel.h>
23#include <linux/sched.h>
24#include <linux/kprobes.h>
25#include <linux/bootmem.h>
26#include <linux/module.h>
27#include <linux/mm.h>
28#include <linux/page-flags.h>
29#include <linux/highmem.h>
30#include <linux/console.h>
31#include <linux/pci.h>
32#include <linux/gfp.h>
33#include <linux/memblock.h>
34#include <linux/edd.h>
35#include <linux/frame.h>
36
37#ifdef CONFIG_KEXEC_CORE
38#include <linux/kexec.h>
39#endif
40
41#include <xen/xen.h>
42#include <xen/events.h>
43#include <xen/interface/xen.h>
44#include <xen/interface/version.h>
45#include <xen/interface/physdev.h>
46#include <xen/interface/vcpu.h>
47#include <xen/interface/memory.h>
48#include <xen/interface/nmi.h>
49#include <xen/interface/xen-mca.h>
50#include <xen/features.h>
51#include <xen/page.h>
52#include <xen/hvm.h>
53#include <xen/hvc-console.h>
54#include <xen/acpi.h>
55
56#include <asm/paravirt.h>
57#include <asm/apic.h>
58#include <asm/page.h>
59#include <asm/xen/pci.h>
60#include <asm/xen/hypercall.h>
61#include <asm/xen/hypervisor.h>
62#include <asm/fixmap.h>
63#include <asm/processor.h>
64#include <asm/proto.h>
65#include <asm/msr-index.h>
66#include <asm/traps.h>
67#include <asm/setup.h>
68#include <asm/desc.h>
69#include <asm/pgalloc.h>
70#include <asm/pgtable.h>
71#include <asm/tlbflush.h>
72#include <asm/reboot.h>
73#include <asm/stackprotector.h>
74#include <asm/hypervisor.h>
75#include <asm/mach_traps.h>
76#include <asm/mwait.h>
77#include <asm/pci_x86.h>
78#include <asm/pat.h>
79#include <asm/cpu.h>
80
81#ifdef CONFIG_ACPI
82#include <linux/acpi.h>
83#include <asm/acpi.h>
84#include <acpi/pdc_intel.h>
85#include <acpi/processor.h>
86#include <xen/interface/platform.h>
87#endif
88
89#include "xen-ops.h"
90#include "mmu.h"
91#include "smp.h"
92#include "multicalls.h"
93#include "pmu.h"
94
95EXPORT_SYMBOL_GPL(hypercall_page);
96
97/*
98 * Pointer to the xen_vcpu_info structure or
99 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
100 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
101 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
102 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
103 * acknowledge pending events.
104 * Also more subtly it is used by the patched version of irq enable/disable
105 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
106 *
107 * The desire to be able to do those mask/unmask operations as a single
108 * instruction by using the per-cpu offset held in %gs is the real reason
109 * vcpu info is in a per-cpu pointer and the original reason for this
110 * hypercall.
111 *
112 */
113DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
114
115/*
116 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
117 * hypercall. This can be used both in PV and PVHVM mode. The structure
118 * overrides the default per_cpu(xen_vcpu, cpu) value.
119 */
120DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
121
122enum xen_domain_type xen_domain_type = XEN_NATIVE;
123EXPORT_SYMBOL_GPL(xen_domain_type);
124
125unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
126EXPORT_SYMBOL(machine_to_phys_mapping);
127unsigned long machine_to_phys_nr;
128EXPORT_SYMBOL(machine_to_phys_nr);
129
130struct start_info *xen_start_info;
131EXPORT_SYMBOL_GPL(xen_start_info);
132
133struct shared_info xen_dummy_shared_info;
134
135void *xen_initial_gdt;
136
137RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
138__read_mostly int xen_have_vector_callback;
139EXPORT_SYMBOL_GPL(xen_have_vector_callback);
140
141/*
142 * Point at some empty memory to start with. We map the real shared_info
143 * page as soon as fixmap is up and running.
144 */
145struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
146
147/*
148 * Flag to determine whether vcpu info placement is available on all
149 * VCPUs. We assume it is to start with, and then set it to zero on
150 * the first failure. This is because it can succeed on some VCPUs
151 * and not others, since it can involve hypervisor memory allocation,
152 * or because the guest failed to guarantee all the appropriate
153 * constraints on all VCPUs (ie buffer can't cross a page boundary).
154 *
155 * Note that any particular CPU may be using a placed vcpu structure,
156 * but we can only optimise if the all are.
157 *
158 * 0: not available, 1: available
159 */
160static int have_vcpu_info_placement = 1;
161
162struct tls_descs {
163 struct desc_struct desc[3];
164};
165
166/*
167 * Updating the 3 TLS descriptors in the GDT on every task switch is
168 * surprisingly expensive so we avoid updating them if they haven't
169 * changed. Since Xen writes different descriptors than the one
170 * passed in the update_descriptor hypercall we keep shadow copies to
171 * compare against.
172 */
173static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
174
175static void clamp_max_cpus(void)
176{
177#ifdef CONFIG_SMP
178 if (setup_max_cpus > MAX_VIRT_CPUS)
179 setup_max_cpus = MAX_VIRT_CPUS;
180#endif
181}
182
183static void xen_vcpu_setup(int cpu)
184{
185 struct vcpu_register_vcpu_info info;
186 int err;
187 struct vcpu_info *vcpup;
188
189 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
190
191 /*
192 * This path is called twice on PVHVM - first during bootup via
193 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
194 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
195 * As we can only do the VCPUOP_register_vcpu_info once lets
196 * not over-write its result.
197 *
198 * For PV it is called during restore (xen_vcpu_restore) and bootup
199 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
200 * use this function.
201 */
202 if (xen_hvm_domain()) {
203 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
204 return;
205 }
206 if (cpu < MAX_VIRT_CPUS)
207 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
208
209 if (!have_vcpu_info_placement) {
210 if (cpu >= MAX_VIRT_CPUS)
211 clamp_max_cpus();
212 return;
213 }
214
215 vcpup = &per_cpu(xen_vcpu_info, cpu);
216 info.mfn = arbitrary_virt_to_mfn(vcpup);
217 info.offset = offset_in_page(vcpup);
218
219 /* Check to see if the hypervisor will put the vcpu_info
220 structure where we want it, which allows direct access via
221 a percpu-variable.
222 N.B. This hypercall can _only_ be called once per CPU. Subsequent
223 calls will error out with -EINVAL. This is due to the fact that
224 hypervisor has no unregister variant and this hypercall does not
225 allow to over-write info.mfn and info.offset.
226 */
227 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
228
229 if (err) {
230 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
231 have_vcpu_info_placement = 0;
232 clamp_max_cpus();
233 } else {
234 /* This cpu is using the registered vcpu info, even if
235 later ones fail to. */
236 per_cpu(xen_vcpu, cpu) = vcpup;
237 }
238}
239
240/*
241 * On restore, set the vcpu placement up again.
242 * If it fails, then we're in a bad state, since
243 * we can't back out from using it...
244 */
245void xen_vcpu_restore(void)
246{
247 int cpu;
248
249 for_each_possible_cpu(cpu) {
250 bool other_cpu = (cpu != smp_processor_id());
251 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL);
252
253 if (other_cpu && is_up &&
254 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
255 BUG();
256
257 xen_setup_runstate_info(cpu);
258
259 if (have_vcpu_info_placement)
260 xen_vcpu_setup(cpu);
261
262 if (other_cpu && is_up &&
263 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
264 BUG();
265 }
266}
267
268static void __init xen_banner(void)
269{
270 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
271 struct xen_extraversion extra;
272 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
273
274 pr_info("Booting paravirtualized kernel %son %s\n",
275 xen_feature(XENFEAT_auto_translated_physmap) ?
276 "with PVH extensions " : "", pv_info.name);
277 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
278 version >> 16, version & 0xffff, extra.extraversion,
279 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
280}
281/* Check if running on Xen version (major, minor) or later */
282bool
283xen_running_on_version_or_later(unsigned int major, unsigned int minor)
284{
285 unsigned int version;
286
287 if (!xen_domain())
288 return false;
289
290 version = HYPERVISOR_xen_version(XENVER_version, NULL);
291 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
292 ((version >> 16) > major))
293 return true;
294 return false;
295}
296
297#define CPUID_THERM_POWER_LEAF 6
298#define APERFMPERF_PRESENT 0
299
300static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
301static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
302
303static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
304static __read_mostly unsigned int cpuid_leaf5_ecx_val;
305static __read_mostly unsigned int cpuid_leaf5_edx_val;
306
307static void xen_cpuid(unsigned int *ax, unsigned int *bx,
308 unsigned int *cx, unsigned int *dx)
309{
310 unsigned maskebx = ~0;
311 unsigned maskecx = ~0;
312 unsigned maskedx = ~0;
313 unsigned setecx = 0;
314 /*
315 * Mask out inconvenient features, to try and disable as many
316 * unsupported kernel subsystems as possible.
317 */
318 switch (*ax) {
319 case 1:
320 maskecx = cpuid_leaf1_ecx_mask;
321 setecx = cpuid_leaf1_ecx_set_mask;
322 maskedx = cpuid_leaf1_edx_mask;
323 break;
324
325 case CPUID_MWAIT_LEAF:
326 /* Synthesize the values.. */
327 *ax = 0;
328 *bx = 0;
329 *cx = cpuid_leaf5_ecx_val;
330 *dx = cpuid_leaf5_edx_val;
331 return;
332
333 case CPUID_THERM_POWER_LEAF:
334 /* Disabling APERFMPERF for kernel usage */
335 maskecx = ~(1 << APERFMPERF_PRESENT);
336 break;
337
338 case 0xb:
339 /* Suppress extended topology stuff */
340 maskebx = 0;
341 break;
342 }
343
344 asm(XEN_EMULATE_PREFIX "cpuid"
345 : "=a" (*ax),
346 "=b" (*bx),
347 "=c" (*cx),
348 "=d" (*dx)
349 : "0" (*ax), "2" (*cx));
350
351 *bx &= maskebx;
352 *cx &= maskecx;
353 *cx |= setecx;
354 *dx &= maskedx;
355}
356STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
357
358static bool __init xen_check_mwait(void)
359{
360#ifdef CONFIG_ACPI
361 struct xen_platform_op op = {
362 .cmd = XENPF_set_processor_pminfo,
363 .u.set_pminfo.id = -1,
364 .u.set_pminfo.type = XEN_PM_PDC,
365 };
366 uint32_t buf[3];
367 unsigned int ax, bx, cx, dx;
368 unsigned int mwait_mask;
369
370 /* We need to determine whether it is OK to expose the MWAIT
371 * capability to the kernel to harvest deeper than C3 states from ACPI
372 * _CST using the processor_harvest_xen.c module. For this to work, we
373 * need to gather the MWAIT_LEAF values (which the cstate.c code
374 * checks against). The hypervisor won't expose the MWAIT flag because
375 * it would break backwards compatibility; so we will find out directly
376 * from the hardware and hypercall.
377 */
378 if (!xen_initial_domain())
379 return false;
380
381 /*
382 * When running under platform earlier than Xen4.2, do not expose
383 * mwait, to avoid the risk of loading native acpi pad driver
384 */
385 if (!xen_running_on_version_or_later(4, 2))
386 return false;
387
388 ax = 1;
389 cx = 0;
390
391 native_cpuid(&ax, &bx, &cx, &dx);
392
393 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
394 (1 << (X86_FEATURE_MWAIT % 32));
395
396 if ((cx & mwait_mask) != mwait_mask)
397 return false;
398
399 /* We need to emulate the MWAIT_LEAF and for that we need both
400 * ecx and edx. The hypercall provides only partial information.
401 */
402
403 ax = CPUID_MWAIT_LEAF;
404 bx = 0;
405 cx = 0;
406 dx = 0;
407
408 native_cpuid(&ax, &bx, &cx, &dx);
409
410 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
411 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
412 */
413 buf[0] = ACPI_PDC_REVISION_ID;
414 buf[1] = 1;
415 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
416
417 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
418
419 if ((HYPERVISOR_platform_op(&op) == 0) &&
420 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
421 cpuid_leaf5_ecx_val = cx;
422 cpuid_leaf5_edx_val = dx;
423 }
424 return true;
425#else
426 return false;
427#endif
428}
429static void __init xen_init_cpuid_mask(void)
430{
431 unsigned int ax, bx, cx, dx;
432 unsigned int xsave_mask;
433
434 cpuid_leaf1_edx_mask =
435 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
436 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
437
438 if (!xen_initial_domain())
439 cpuid_leaf1_edx_mask &=
440 ~((1 << X86_FEATURE_ACPI)); /* disable ACPI */
441
442 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
443
444 ax = 1;
445 cx = 0;
446 cpuid(1, &ax, &bx, &cx, &dx);
447
448 xsave_mask =
449 (1 << (X86_FEATURE_XSAVE % 32)) |
450 (1 << (X86_FEATURE_OSXSAVE % 32));
451
452 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
453 if ((cx & xsave_mask) != xsave_mask)
454 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
455 if (xen_check_mwait())
456 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
457}
458
459static void xen_set_debugreg(int reg, unsigned long val)
460{
461 HYPERVISOR_set_debugreg(reg, val);
462}
463
464static unsigned long xen_get_debugreg(int reg)
465{
466 return HYPERVISOR_get_debugreg(reg);
467}
468
469static void xen_end_context_switch(struct task_struct *next)
470{
471 xen_mc_flush();
472 paravirt_end_context_switch(next);
473}
474
475static unsigned long xen_store_tr(void)
476{
477 return 0;
478}
479
480/*
481 * Set the page permissions for a particular virtual address. If the
482 * address is a vmalloc mapping (or other non-linear mapping), then
483 * find the linear mapping of the page and also set its protections to
484 * match.
485 */
486static void set_aliased_prot(void *v, pgprot_t prot)
487{
488 int level;
489 pte_t *ptep;
490 pte_t pte;
491 unsigned long pfn;
492 struct page *page;
493 unsigned char dummy;
494
495 ptep = lookup_address((unsigned long)v, &level);
496 BUG_ON(ptep == NULL);
497
498 pfn = pte_pfn(*ptep);
499 page = pfn_to_page(pfn);
500
501 pte = pfn_pte(pfn, prot);
502
503 /*
504 * Careful: update_va_mapping() will fail if the virtual address
505 * we're poking isn't populated in the page tables. We don't
506 * need to worry about the direct map (that's always in the page
507 * tables), but we need to be careful about vmap space. In
508 * particular, the top level page table can lazily propagate
509 * entries between processes, so if we've switched mms since we
510 * vmapped the target in the first place, we might not have the
511 * top-level page table entry populated.
512 *
513 * We disable preemption because we want the same mm active when
514 * we probe the target and when we issue the hypercall. We'll
515 * have the same nominal mm, but if we're a kernel thread, lazy
516 * mm dropping could change our pgd.
517 *
518 * Out of an abundance of caution, this uses __get_user() to fault
519 * in the target address just in case there's some obscure case
520 * in which the target address isn't readable.
521 */
522
523 preempt_disable();
524
525 pagefault_disable(); /* Avoid warnings due to being atomic. */
526 __get_user(dummy, (unsigned char __user __force *)v);
527 pagefault_enable();
528
529 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
530 BUG();
531
532 if (!PageHighMem(page)) {
533 void *av = __va(PFN_PHYS(pfn));
534
535 if (av != v)
536 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
537 BUG();
538 } else
539 kmap_flush_unused();
540
541 preempt_enable();
542}
543
544static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
545{
546 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
547 int i;
548
549 /*
550 * We need to mark the all aliases of the LDT pages RO. We
551 * don't need to call vm_flush_aliases(), though, since that's
552 * only responsible for flushing aliases out the TLBs, not the
553 * page tables, and Xen will flush the TLB for us if needed.
554 *
555 * To avoid confusing future readers: none of this is necessary
556 * to load the LDT. The hypervisor only checks this when the
557 * LDT is faulted in due to subsequent descriptor access.
558 */
559
560 for(i = 0; i < entries; i += entries_per_page)
561 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
562}
563
564static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
565{
566 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
567 int i;
568
569 for(i = 0; i < entries; i += entries_per_page)
570 set_aliased_prot(ldt + i, PAGE_KERNEL);
571}
572
573static void xen_set_ldt(const void *addr, unsigned entries)
574{
575 struct mmuext_op *op;
576 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
577
578 trace_xen_cpu_set_ldt(addr, entries);
579
580 op = mcs.args;
581 op->cmd = MMUEXT_SET_LDT;
582 op->arg1.linear_addr = (unsigned long)addr;
583 op->arg2.nr_ents = entries;
584
585 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
586
587 xen_mc_issue(PARAVIRT_LAZY_CPU);
588}
589
590static void xen_load_gdt(const struct desc_ptr *dtr)
591{
592 unsigned long va = dtr->address;
593 unsigned int size = dtr->size + 1;
594 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
595 unsigned long frames[pages];
596 int f;
597
598 /*
599 * A GDT can be up to 64k in size, which corresponds to 8192
600 * 8-byte entries, or 16 4k pages..
601 */
602
603 BUG_ON(size > 65536);
604 BUG_ON(va & ~PAGE_MASK);
605
606 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
607 int level;
608 pte_t *ptep;
609 unsigned long pfn, mfn;
610 void *virt;
611
612 /*
613 * The GDT is per-cpu and is in the percpu data area.
614 * That can be virtually mapped, so we need to do a
615 * page-walk to get the underlying MFN for the
616 * hypercall. The page can also be in the kernel's
617 * linear range, so we need to RO that mapping too.
618 */
619 ptep = lookup_address(va, &level);
620 BUG_ON(ptep == NULL);
621
622 pfn = pte_pfn(*ptep);
623 mfn = pfn_to_mfn(pfn);
624 virt = __va(PFN_PHYS(pfn));
625
626 frames[f] = mfn;
627
628 make_lowmem_page_readonly((void *)va);
629 make_lowmem_page_readonly(virt);
630 }
631
632 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
633 BUG();
634}
635
636/*
637 * load_gdt for early boot, when the gdt is only mapped once
638 */
639static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
640{
641 unsigned long va = dtr->address;
642 unsigned int size = dtr->size + 1;
643 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
644 unsigned long frames[pages];
645 int f;
646
647 /*
648 * A GDT can be up to 64k in size, which corresponds to 8192
649 * 8-byte entries, or 16 4k pages..
650 */
651
652 BUG_ON(size > 65536);
653 BUG_ON(va & ~PAGE_MASK);
654
655 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
656 pte_t pte;
657 unsigned long pfn, mfn;
658
659 pfn = virt_to_pfn(va);
660 mfn = pfn_to_mfn(pfn);
661
662 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
663
664 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
665 BUG();
666
667 frames[f] = mfn;
668 }
669
670 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
671 BUG();
672}
673
674static inline bool desc_equal(const struct desc_struct *d1,
675 const struct desc_struct *d2)
676{
677 return d1->a == d2->a && d1->b == d2->b;
678}
679
680static void load_TLS_descriptor(struct thread_struct *t,
681 unsigned int cpu, unsigned int i)
682{
683 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
684 struct desc_struct *gdt;
685 xmaddr_t maddr;
686 struct multicall_space mc;
687
688 if (desc_equal(shadow, &t->tls_array[i]))
689 return;
690
691 *shadow = t->tls_array[i];
692
693 gdt = get_cpu_gdt_table(cpu);
694 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
695 mc = __xen_mc_entry(0);
696
697 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
698}
699
700static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
701{
702 /*
703 * XXX sleazy hack: If we're being called in a lazy-cpu zone
704 * and lazy gs handling is enabled, it means we're in a
705 * context switch, and %gs has just been saved. This means we
706 * can zero it out to prevent faults on exit from the
707 * hypervisor if the next process has no %gs. Either way, it
708 * has been saved, and the new value will get loaded properly.
709 * This will go away as soon as Xen has been modified to not
710 * save/restore %gs for normal hypercalls.
711 *
712 * On x86_64, this hack is not used for %gs, because gs points
713 * to KERNEL_GS_BASE (and uses it for PDA references), so we
714 * must not zero %gs on x86_64
715 *
716 * For x86_64, we need to zero %fs, otherwise we may get an
717 * exception between the new %fs descriptor being loaded and
718 * %fs being effectively cleared at __switch_to().
719 */
720 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
721#ifdef CONFIG_X86_32
722 lazy_load_gs(0);
723#else
724 loadsegment(fs, 0);
725#endif
726 }
727
728 xen_mc_batch();
729
730 load_TLS_descriptor(t, cpu, 0);
731 load_TLS_descriptor(t, cpu, 1);
732 load_TLS_descriptor(t, cpu, 2);
733
734 xen_mc_issue(PARAVIRT_LAZY_CPU);
735}
736
737#ifdef CONFIG_X86_64
738static void xen_load_gs_index(unsigned int idx)
739{
740 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
741 BUG();
742}
743#endif
744
745static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
746 const void *ptr)
747{
748 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
749 u64 entry = *(u64 *)ptr;
750
751 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
752
753 preempt_disable();
754
755 xen_mc_flush();
756 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
757 BUG();
758
759 preempt_enable();
760}
761
762static int cvt_gate_to_trap(int vector, const gate_desc *val,
763 struct trap_info *info)
764{
765 unsigned long addr;
766
767 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
768 return 0;
769
770 info->vector = vector;
771
772 addr = gate_offset(*val);
773#ifdef CONFIG_X86_64
774 /*
775 * Look for known traps using IST, and substitute them
776 * appropriately. The debugger ones are the only ones we care
777 * about. Xen will handle faults like double_fault,
778 * so we should never see them. Warn if
779 * there's an unexpected IST-using fault handler.
780 */
781 if (addr == (unsigned long)debug)
782 addr = (unsigned long)xen_debug;
783 else if (addr == (unsigned long)int3)
784 addr = (unsigned long)xen_int3;
785 else if (addr == (unsigned long)stack_segment)
786 addr = (unsigned long)xen_stack_segment;
787 else if (addr == (unsigned long)double_fault) {
788 /* Don't need to handle these */
789 return 0;
790#ifdef CONFIG_X86_MCE
791 } else if (addr == (unsigned long)machine_check) {
792 /*
793 * when xen hypervisor inject vMCE to guest,
794 * use native mce handler to handle it
795 */
796 ;
797#endif
798 } else if (addr == (unsigned long)nmi)
799 /*
800 * Use the native version as well.
801 */
802 ;
803 else {
804 /* Some other trap using IST? */
805 if (WARN_ON(val->ist != 0))
806 return 0;
807 }
808#endif /* CONFIG_X86_64 */
809 info->address = addr;
810
811 info->cs = gate_segment(*val);
812 info->flags = val->dpl;
813 /* interrupt gates clear IF */
814 if (val->type == GATE_INTERRUPT)
815 info->flags |= 1 << 2;
816
817 return 1;
818}
819
820/* Locations of each CPU's IDT */
821static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
822
823/* Set an IDT entry. If the entry is part of the current IDT, then
824 also update Xen. */
825static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
826{
827 unsigned long p = (unsigned long)&dt[entrynum];
828 unsigned long start, end;
829
830 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
831
832 preempt_disable();
833
834 start = __this_cpu_read(idt_desc.address);
835 end = start + __this_cpu_read(idt_desc.size) + 1;
836
837 xen_mc_flush();
838
839 native_write_idt_entry(dt, entrynum, g);
840
841 if (p >= start && (p + 8) <= end) {
842 struct trap_info info[2];
843
844 info[1].address = 0;
845
846 if (cvt_gate_to_trap(entrynum, g, &info[0]))
847 if (HYPERVISOR_set_trap_table(info))
848 BUG();
849 }
850
851 preempt_enable();
852}
853
854static void xen_convert_trap_info(const struct desc_ptr *desc,
855 struct trap_info *traps)
856{
857 unsigned in, out, count;
858
859 count = (desc->size+1) / sizeof(gate_desc);
860 BUG_ON(count > 256);
861
862 for (in = out = 0; in < count; in++) {
863 gate_desc *entry = (gate_desc*)(desc->address) + in;
864
865 if (cvt_gate_to_trap(in, entry, &traps[out]))
866 out++;
867 }
868 traps[out].address = 0;
869}
870
871void xen_copy_trap_info(struct trap_info *traps)
872{
873 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
874
875 xen_convert_trap_info(desc, traps);
876}
877
878/* Load a new IDT into Xen. In principle this can be per-CPU, so we
879 hold a spinlock to protect the static traps[] array (static because
880 it avoids allocation, and saves stack space). */
881static void xen_load_idt(const struct desc_ptr *desc)
882{
883 static DEFINE_SPINLOCK(lock);
884 static struct trap_info traps[257];
885
886 trace_xen_cpu_load_idt(desc);
887
888 spin_lock(&lock);
889
890 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
891
892 xen_convert_trap_info(desc, traps);
893
894 xen_mc_flush();
895 if (HYPERVISOR_set_trap_table(traps))
896 BUG();
897
898 spin_unlock(&lock);
899}
900
901/* Write a GDT descriptor entry. Ignore LDT descriptors, since
902 they're handled differently. */
903static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
904 const void *desc, int type)
905{
906 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
907
908 preempt_disable();
909
910 switch (type) {
911 case DESC_LDT:
912 case DESC_TSS:
913 /* ignore */
914 break;
915
916 default: {
917 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
918
919 xen_mc_flush();
920 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
921 BUG();
922 }
923
924 }
925
926 preempt_enable();
927}
928
929/*
930 * Version of write_gdt_entry for use at early boot-time needed to
931 * update an entry as simply as possible.
932 */
933static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
934 const void *desc, int type)
935{
936 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
937
938 switch (type) {
939 case DESC_LDT:
940 case DESC_TSS:
941 /* ignore */
942 break;
943
944 default: {
945 xmaddr_t maddr = virt_to_machine(&dt[entry]);
946
947 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
948 dt[entry] = *(struct desc_struct *)desc;
949 }
950
951 }
952}
953
954static void xen_load_sp0(struct tss_struct *tss,
955 struct thread_struct *thread)
956{
957 struct multicall_space mcs;
958
959 mcs = xen_mc_entry(0);
960 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
961 xen_mc_issue(PARAVIRT_LAZY_CPU);
962 tss->x86_tss.sp0 = thread->sp0;
963}
964
965void xen_set_iopl_mask(unsigned mask)
966{
967 struct physdev_set_iopl set_iopl;
968
969 /* Force the change at ring 0. */
970 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
971 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
972}
973
974static void xen_io_delay(void)
975{
976}
977
978static void xen_clts(void)
979{
980 struct multicall_space mcs;
981
982 mcs = xen_mc_entry(0);
983
984 MULTI_fpu_taskswitch(mcs.mc, 0);
985
986 xen_mc_issue(PARAVIRT_LAZY_CPU);
987}
988
989static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
990
991static unsigned long xen_read_cr0(void)
992{
993 unsigned long cr0 = this_cpu_read(xen_cr0_value);
994
995 if (unlikely(cr0 == 0)) {
996 cr0 = native_read_cr0();
997 this_cpu_write(xen_cr0_value, cr0);
998 }
999
1000 return cr0;
1001}
1002
1003static void xen_write_cr0(unsigned long cr0)
1004{
1005 struct multicall_space mcs;
1006
1007 this_cpu_write(xen_cr0_value, cr0);
1008
1009 /* Only pay attention to cr0.TS; everything else is
1010 ignored. */
1011 mcs = xen_mc_entry(0);
1012
1013 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1014
1015 xen_mc_issue(PARAVIRT_LAZY_CPU);
1016}
1017
1018static void xen_write_cr4(unsigned long cr4)
1019{
1020 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1021
1022 native_write_cr4(cr4);
1023}
1024#ifdef CONFIG_X86_64
1025static inline unsigned long xen_read_cr8(void)
1026{
1027 return 0;
1028}
1029static inline void xen_write_cr8(unsigned long val)
1030{
1031 BUG_ON(val);
1032}
1033#endif
1034
1035static u64 xen_read_msr_safe(unsigned int msr, int *err)
1036{
1037 u64 val;
1038
1039 if (pmu_msr_read(msr, &val, err))
1040 return val;
1041
1042 val = native_read_msr_safe(msr, err);
1043 switch (msr) {
1044 case MSR_IA32_APICBASE:
1045#ifdef CONFIG_X86_X2APIC
1046 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1047#endif
1048 val &= ~X2APIC_ENABLE;
1049 break;
1050 }
1051 return val;
1052}
1053
1054static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1055{
1056 int ret;
1057
1058 ret = 0;
1059
1060 switch (msr) {
1061#ifdef CONFIG_X86_64
1062 unsigned which;
1063 u64 base;
1064
1065 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
1066 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
1067 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
1068
1069 set:
1070 base = ((u64)high << 32) | low;
1071 if (HYPERVISOR_set_segment_base(which, base) != 0)
1072 ret = -EIO;
1073 break;
1074#endif
1075
1076 case MSR_STAR:
1077 case MSR_CSTAR:
1078 case MSR_LSTAR:
1079 case MSR_SYSCALL_MASK:
1080 case MSR_IA32_SYSENTER_CS:
1081 case MSR_IA32_SYSENTER_ESP:
1082 case MSR_IA32_SYSENTER_EIP:
1083 /* Fast syscall setup is all done in hypercalls, so
1084 these are all ignored. Stub them out here to stop
1085 Xen console noise. */
1086 break;
1087
1088 default:
1089 if (!pmu_msr_write(msr, low, high, &ret))
1090 ret = native_write_msr_safe(msr, low, high);
1091 }
1092
1093 return ret;
1094}
1095
1096void xen_setup_shared_info(void)
1097{
1098 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1099 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1100 xen_start_info->shared_info);
1101
1102 HYPERVISOR_shared_info =
1103 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1104 } else
1105 HYPERVISOR_shared_info =
1106 (struct shared_info *)__va(xen_start_info->shared_info);
1107
1108#ifndef CONFIG_SMP
1109 /* In UP this is as good a place as any to set up shared info */
1110 xen_setup_vcpu_info_placement();
1111#endif
1112
1113 xen_setup_mfn_list_list();
1114}
1115
1116/* This is called once we have the cpu_possible_mask */
1117void xen_setup_vcpu_info_placement(void)
1118{
1119 int cpu;
1120
1121 for_each_possible_cpu(cpu)
1122 xen_vcpu_setup(cpu);
1123
1124 /* xen_vcpu_setup managed to place the vcpu_info within the
1125 * percpu area for all cpus, so make use of it. Note that for
1126 * PVH we want to use native IRQ mechanism. */
1127 if (have_vcpu_info_placement && !xen_pvh_domain()) {
1128 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1129 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1130 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1131 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1132 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1133 }
1134}
1135
1136static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1137 unsigned long addr, unsigned len)
1138{
1139 char *start, *end, *reloc;
1140 unsigned ret;
1141
1142 start = end = reloc = NULL;
1143
1144#define SITE(op, x) \
1145 case PARAVIRT_PATCH(op.x): \
1146 if (have_vcpu_info_placement) { \
1147 start = (char *)xen_##x##_direct; \
1148 end = xen_##x##_direct_end; \
1149 reloc = xen_##x##_direct_reloc; \
1150 } \
1151 goto patch_site
1152
1153 switch (type) {
1154 SITE(pv_irq_ops, irq_enable);
1155 SITE(pv_irq_ops, irq_disable);
1156 SITE(pv_irq_ops, save_fl);
1157 SITE(pv_irq_ops, restore_fl);
1158#undef SITE
1159
1160 patch_site:
1161 if (start == NULL || (end-start) > len)
1162 goto default_patch;
1163
1164 ret = paravirt_patch_insns(insnbuf, len, start, end);
1165
1166 /* Note: because reloc is assigned from something that
1167 appears to be an array, gcc assumes it's non-null,
1168 but doesn't know its relationship with start and
1169 end. */
1170 if (reloc > start && reloc < end) {
1171 int reloc_off = reloc - start;
1172 long *relocp = (long *)(insnbuf + reloc_off);
1173 long delta = start - (char *)addr;
1174
1175 *relocp += delta;
1176 }
1177 break;
1178
1179 default_patch:
1180 default:
1181 ret = paravirt_patch_default(type, clobbers, insnbuf,
1182 addr, len);
1183 break;
1184 }
1185
1186 return ret;
1187}
1188
1189static const struct pv_info xen_info __initconst = {
1190 .paravirt_enabled = 1,
1191 .shared_kernel_pmd = 0,
1192
1193#ifdef CONFIG_X86_64
1194 .extra_user_64bit_cs = FLAT_USER_CS64,
1195#endif
1196 .features = 0,
1197 .name = "Xen",
1198};
1199
1200static const struct pv_init_ops xen_init_ops __initconst = {
1201 .patch = xen_patch,
1202};
1203
1204static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1205 .cpuid = xen_cpuid,
1206
1207 .set_debugreg = xen_set_debugreg,
1208 .get_debugreg = xen_get_debugreg,
1209
1210 .clts = xen_clts,
1211
1212 .read_cr0 = xen_read_cr0,
1213 .write_cr0 = xen_write_cr0,
1214
1215 .read_cr4 = native_read_cr4,
1216 .read_cr4_safe = native_read_cr4_safe,
1217 .write_cr4 = xen_write_cr4,
1218
1219#ifdef CONFIG_X86_64
1220 .read_cr8 = xen_read_cr8,
1221 .write_cr8 = xen_write_cr8,
1222#endif
1223
1224 .wbinvd = native_wbinvd,
1225
1226 .read_msr = xen_read_msr_safe,
1227 .write_msr = xen_write_msr_safe,
1228
1229 .read_pmc = xen_read_pmc,
1230
1231 .iret = xen_iret,
1232#ifdef CONFIG_X86_64
1233 .usergs_sysret64 = xen_sysret64,
1234#endif
1235
1236 .load_tr_desc = paravirt_nop,
1237 .set_ldt = xen_set_ldt,
1238 .load_gdt = xen_load_gdt,
1239 .load_idt = xen_load_idt,
1240 .load_tls = xen_load_tls,
1241#ifdef CONFIG_X86_64
1242 .load_gs_index = xen_load_gs_index,
1243#endif
1244
1245 .alloc_ldt = xen_alloc_ldt,
1246 .free_ldt = xen_free_ldt,
1247
1248 .store_idt = native_store_idt,
1249 .store_tr = xen_store_tr,
1250
1251 .write_ldt_entry = xen_write_ldt_entry,
1252 .write_gdt_entry = xen_write_gdt_entry,
1253 .write_idt_entry = xen_write_idt_entry,
1254 .load_sp0 = xen_load_sp0,
1255
1256 .set_iopl_mask = xen_set_iopl_mask,
1257 .io_delay = xen_io_delay,
1258
1259 /* Xen takes care of %gs when switching to usermode for us */
1260 .swapgs = paravirt_nop,
1261
1262 .start_context_switch = paravirt_start_context_switch,
1263 .end_context_switch = xen_end_context_switch,
1264};
1265
1266static void xen_reboot(int reason)
1267{
1268 struct sched_shutdown r = { .reason = reason };
1269 int cpu;
1270
1271 for_each_online_cpu(cpu)
1272 xen_pmu_finish(cpu);
1273
1274 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1275 BUG();
1276}
1277
1278static void xen_restart(char *msg)
1279{
1280 xen_reboot(SHUTDOWN_reboot);
1281}
1282
1283static void xen_emergency_restart(void)
1284{
1285 xen_reboot(SHUTDOWN_reboot);
1286}
1287
1288static void xen_machine_halt(void)
1289{
1290 xen_reboot(SHUTDOWN_poweroff);
1291}
1292
1293static void xen_machine_power_off(void)
1294{
1295 if (pm_power_off)
1296 pm_power_off();
1297 xen_reboot(SHUTDOWN_poweroff);
1298}
1299
1300static void xen_crash_shutdown(struct pt_regs *regs)
1301{
1302 xen_reboot(SHUTDOWN_crash);
1303}
1304
1305static int
1306xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1307{
1308 xen_reboot(SHUTDOWN_crash);
1309 return NOTIFY_DONE;
1310}
1311
1312static struct notifier_block xen_panic_block = {
1313 .notifier_call= xen_panic_event,
1314 .priority = INT_MIN
1315};
1316
1317int xen_panic_handler_init(void)
1318{
1319 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1320 return 0;
1321}
1322
1323static const struct machine_ops xen_machine_ops __initconst = {
1324 .restart = xen_restart,
1325 .halt = xen_machine_halt,
1326 .power_off = xen_machine_power_off,
1327 .shutdown = xen_machine_halt,
1328 .crash_shutdown = xen_crash_shutdown,
1329 .emergency_restart = xen_emergency_restart,
1330};
1331
1332static unsigned char xen_get_nmi_reason(void)
1333{
1334 unsigned char reason = 0;
1335
1336 /* Construct a value which looks like it came from port 0x61. */
1337 if (test_bit(_XEN_NMIREASON_io_error,
1338 &HYPERVISOR_shared_info->arch.nmi_reason))
1339 reason |= NMI_REASON_IOCHK;
1340 if (test_bit(_XEN_NMIREASON_pci_serr,
1341 &HYPERVISOR_shared_info->arch.nmi_reason))
1342 reason |= NMI_REASON_SERR;
1343
1344 return reason;
1345}
1346
1347static void __init xen_boot_params_init_edd(void)
1348{
1349#if IS_ENABLED(CONFIG_EDD)
1350 struct xen_platform_op op;
1351 struct edd_info *edd_info;
1352 u32 *mbr_signature;
1353 unsigned nr;
1354 int ret;
1355
1356 edd_info = boot_params.eddbuf;
1357 mbr_signature = boot_params.edd_mbr_sig_buffer;
1358
1359 op.cmd = XENPF_firmware_info;
1360
1361 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1362 for (nr = 0; nr < EDDMAXNR; nr++) {
1363 struct edd_info *info = edd_info + nr;
1364
1365 op.u.firmware_info.index = nr;
1366 info->params.length = sizeof(info->params);
1367 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1368 &info->params);
1369 ret = HYPERVISOR_platform_op(&op);
1370 if (ret)
1371 break;
1372
1373#define C(x) info->x = op.u.firmware_info.u.disk_info.x
1374 C(device);
1375 C(version);
1376 C(interface_support);
1377 C(legacy_max_cylinder);
1378 C(legacy_max_head);
1379 C(legacy_sectors_per_track);
1380#undef C
1381 }
1382 boot_params.eddbuf_entries = nr;
1383
1384 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1385 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1386 op.u.firmware_info.index = nr;
1387 ret = HYPERVISOR_platform_op(&op);
1388 if (ret)
1389 break;
1390 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1391 }
1392 boot_params.edd_mbr_sig_buf_entries = nr;
1393#endif
1394}
1395
1396/*
1397 * Set up the GDT and segment registers for -fstack-protector. Until
1398 * we do this, we have to be careful not to call any stack-protected
1399 * function, which is most of the kernel.
1400 *
1401 * Note, that it is __ref because the only caller of this after init
1402 * is PVH which is not going to use xen_load_gdt_boot or other
1403 * __init functions.
1404 */
1405static void __ref xen_setup_gdt(int cpu)
1406{
1407 if (xen_feature(XENFEAT_auto_translated_physmap)) {
1408#ifdef CONFIG_X86_64
1409 unsigned long dummy;
1410
1411 load_percpu_segment(cpu); /* We need to access per-cpu area */
1412 switch_to_new_gdt(cpu); /* GDT and GS set */
1413
1414 /* We are switching of the Xen provided GDT to our HVM mode
1415 * GDT. The new GDT has __KERNEL_CS with CS.L = 1
1416 * and we are jumping to reload it.
1417 */
1418 asm volatile ("pushq %0\n"
1419 "leaq 1f(%%rip),%0\n"
1420 "pushq %0\n"
1421 "lretq\n"
1422 "1:\n"
1423 : "=&r" (dummy) : "0" (__KERNEL_CS));
1424
1425 /*
1426 * While not needed, we also set the %es, %ds, and %fs
1427 * to zero. We don't care about %ss as it is NULL.
1428 * Strictly speaking this is not needed as Xen zeros those
1429 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1430 *
1431 * Linux zeros them in cpu_init() and in secondary_startup_64
1432 * (for BSP).
1433 */
1434 loadsegment(es, 0);
1435 loadsegment(ds, 0);
1436 loadsegment(fs, 0);
1437#else
1438 /* PVH: TODO Implement. */
1439 BUG();
1440#endif
1441 return; /* PVH does not need any PV GDT ops. */
1442 }
1443 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1444 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1445
1446 setup_stack_canary_segment(0);
1447 switch_to_new_gdt(0);
1448
1449 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1450 pv_cpu_ops.load_gdt = xen_load_gdt;
1451}
1452
1453#ifdef CONFIG_XEN_PVH
1454/*
1455 * A PV guest starts with default flags that are not set for PVH, set them
1456 * here asap.
1457 */
1458static void xen_pvh_set_cr_flags(int cpu)
1459{
1460
1461 /* Some of these are setup in 'secondary_startup_64'. The others:
1462 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1463 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1464 write_cr0(read_cr0() | X86_CR0_MP | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM);
1465
1466 if (!cpu)
1467 return;
1468 /*
1469 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
1470 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu__init_cpu().
1471 */
1472 if (cpu_has_pse)
1473 cr4_set_bits_and_update_boot(X86_CR4_PSE);
1474
1475 if (cpu_has_pge)
1476 cr4_set_bits_and_update_boot(X86_CR4_PGE);
1477}
1478
1479/*
1480 * Note, that it is ref - because the only caller of this after init
1481 * is PVH which is not going to use xen_load_gdt_boot or other
1482 * __init functions.
1483 */
1484void __ref xen_pvh_secondary_vcpu_init(int cpu)
1485{
1486 xen_setup_gdt(cpu);
1487 xen_pvh_set_cr_flags(cpu);
1488}
1489
1490static void __init xen_pvh_early_guest_init(void)
1491{
1492 if (!xen_feature(XENFEAT_auto_translated_physmap))
1493 return;
1494
1495 if (!xen_feature(XENFEAT_hvm_callback_vector))
1496 return;
1497
1498 xen_have_vector_callback = 1;
1499
1500 xen_pvh_early_cpu_init(0, false);
1501 xen_pvh_set_cr_flags(0);
1502
1503#ifdef CONFIG_X86_32
1504 BUG(); /* PVH: Implement proper support. */
1505#endif
1506}
1507#endif /* CONFIG_XEN_PVH */
1508
1509/* First C function to be called on Xen boot */
1510asmlinkage __visible void __init xen_start_kernel(void)
1511{
1512 struct physdev_set_iopl set_iopl;
1513 unsigned long initrd_start = 0;
1514 u64 pat;
1515 int rc;
1516
1517 if (!xen_start_info)
1518 return;
1519
1520 xen_domain_type = XEN_PV_DOMAIN;
1521
1522 xen_setup_features();
1523#ifdef CONFIG_XEN_PVH
1524 xen_pvh_early_guest_init();
1525#endif
1526 xen_setup_machphys_mapping();
1527
1528 /* Install Xen paravirt ops */
1529 pv_info = xen_info;
1530 if (xen_initial_domain())
1531 pv_info.features |= PV_SUPPORTED_RTC;
1532 pv_init_ops = xen_init_ops;
1533 if (!xen_pvh_domain()) {
1534 pv_cpu_ops = xen_cpu_ops;
1535
1536 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1537 }
1538
1539 if (xen_feature(XENFEAT_auto_translated_physmap))
1540 x86_init.resources.memory_setup = xen_auto_xlated_memory_setup;
1541 else
1542 x86_init.resources.memory_setup = xen_memory_setup;
1543 x86_init.oem.arch_setup = xen_arch_setup;
1544 x86_init.oem.banner = xen_banner;
1545
1546 xen_init_time_ops();
1547
1548 /*
1549 * Set up some pagetable state before starting to set any ptes.
1550 */
1551
1552 xen_init_mmu_ops();
1553
1554 /* Prevent unwanted bits from being set in PTEs. */
1555 __supported_pte_mask &= ~_PAGE_GLOBAL;
1556
1557 /*
1558 * Prevent page tables from being allocated in highmem, even
1559 * if CONFIG_HIGHPTE is enabled.
1560 */
1561 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1562
1563 /* Work out if we support NX */
1564 x86_configure_nx();
1565
1566 /* Get mfn list */
1567 xen_build_dynamic_phys_to_machine();
1568
1569 /*
1570 * Set up kernel GDT and segment registers, mainly so that
1571 * -fstack-protector code can be executed.
1572 */
1573 xen_setup_gdt(0);
1574
1575 xen_init_irq_ops();
1576 xen_init_cpuid_mask();
1577
1578#ifdef CONFIG_X86_LOCAL_APIC
1579 /*
1580 * set up the basic apic ops.
1581 */
1582 xen_init_apic();
1583#endif
1584
1585 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1586 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1587 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1588 }
1589
1590 machine_ops = xen_machine_ops;
1591
1592 /*
1593 * The only reliable way to retain the initial address of the
1594 * percpu gdt_page is to remember it here, so we can go and
1595 * mark it RW later, when the initial percpu area is freed.
1596 */
1597 xen_initial_gdt = &per_cpu(gdt_page, 0);
1598
1599 xen_smp_init();
1600
1601#ifdef CONFIG_ACPI_NUMA
1602 /*
1603 * The pages we from Xen are not related to machine pages, so
1604 * any NUMA information the kernel tries to get from ACPI will
1605 * be meaningless. Prevent it from trying.
1606 */
1607 acpi_numa = -1;
1608#endif
1609 /* Don't do the full vcpu_info placement stuff until we have a
1610 possible map and a non-dummy shared_info. */
1611 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1612
1613 local_irq_disable();
1614 early_boot_irqs_disabled = true;
1615
1616 xen_raw_console_write("mapping kernel into physical memory\n");
1617 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1618 xen_start_info->nr_pages);
1619 xen_reserve_special_pages();
1620
1621 /*
1622 * Modify the cache mode translation tables to match Xen's PAT
1623 * configuration.
1624 */
1625 rdmsrl(MSR_IA32_CR_PAT, pat);
1626 pat_init_cache_modes(pat);
1627
1628 /* keep using Xen gdt for now; no urgent need to change it */
1629
1630#ifdef CONFIG_X86_32
1631 pv_info.kernel_rpl = 1;
1632 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1633 pv_info.kernel_rpl = 0;
1634#else
1635 pv_info.kernel_rpl = 0;
1636#endif
1637 /* set the limit of our address space */
1638 xen_reserve_top();
1639
1640 /* PVH: runs at default kernel iopl of 0 */
1641 if (!xen_pvh_domain()) {
1642 /*
1643 * We used to do this in xen_arch_setup, but that is too late
1644 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1645 * early_amd_init which pokes 0xcf8 port.
1646 */
1647 set_iopl.iopl = 1;
1648 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1649 if (rc != 0)
1650 xen_raw_printk("physdev_op failed %d\n", rc);
1651 }
1652
1653#ifdef CONFIG_X86_32
1654 /* set up basic CPUID stuff */
1655 cpu_detect(&new_cpu_data);
1656 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1657 new_cpu_data.wp_works_ok = 1;
1658 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1659#endif
1660
1661 if (xen_start_info->mod_start) {
1662 if (xen_start_info->flags & SIF_MOD_START_PFN)
1663 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1664 else
1665 initrd_start = __pa(xen_start_info->mod_start);
1666 }
1667
1668 /* Poke various useful things into boot_params */
1669 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1670 boot_params.hdr.ramdisk_image = initrd_start;
1671 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1672 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1673
1674 if (!xen_initial_domain()) {
1675 add_preferred_console("xenboot", 0, NULL);
1676 add_preferred_console("tty", 0, NULL);
1677 add_preferred_console("hvc", 0, NULL);
1678 if (pci_xen)
1679 x86_init.pci.arch_init = pci_xen_init;
1680 } else {
1681 const struct dom0_vga_console_info *info =
1682 (void *)((char *)xen_start_info +
1683 xen_start_info->console.dom0.info_off);
1684 struct xen_platform_op op = {
1685 .cmd = XENPF_firmware_info,
1686 .interface_version = XENPF_INTERFACE_VERSION,
1687 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1688 };
1689
1690 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1691 xen_start_info->console.domU.mfn = 0;
1692 xen_start_info->console.domU.evtchn = 0;
1693
1694 if (HYPERVISOR_platform_op(&op) == 0)
1695 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1696
1697 /* Make sure ACS will be enabled */
1698 pci_request_acs();
1699
1700 xen_acpi_sleep_register();
1701
1702 /* Avoid searching for BIOS MP tables */
1703 x86_init.mpparse.find_smp_config = x86_init_noop;
1704 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1705
1706 xen_boot_params_init_edd();
1707 }
1708#ifdef CONFIG_PCI
1709 /* PCI BIOS service won't work from a PV guest. */
1710 pci_probe &= ~PCI_PROBE_BIOS;
1711#endif
1712 xen_raw_console_write("about to get started...\n");
1713
1714 xen_setup_runstate_info(0);
1715
1716 xen_efi_init();
1717
1718 /* Start the world */
1719#ifdef CONFIG_X86_32
1720 i386_start_kernel();
1721#else
1722 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1723 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1724#endif
1725}
1726
1727void __ref xen_hvm_init_shared_info(void)
1728{
1729 int cpu;
1730 struct xen_add_to_physmap xatp;
1731 static struct shared_info *shared_info_page = 0;
1732
1733 if (!shared_info_page)
1734 shared_info_page = (struct shared_info *)
1735 extend_brk(PAGE_SIZE, PAGE_SIZE);
1736 xatp.domid = DOMID_SELF;
1737 xatp.idx = 0;
1738 xatp.space = XENMAPSPACE_shared_info;
1739 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1740 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1741 BUG();
1742
1743 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1744
1745 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1746 * page, we use it in the event channel upcall and in some pvclock
1747 * related functions. We don't need the vcpu_info placement
1748 * optimizations because we don't use any pv_mmu or pv_irq op on
1749 * HVM.
1750 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1751 * online but xen_hvm_init_shared_info is run at resume time too and
1752 * in that case multiple vcpus might be online. */
1753 for_each_online_cpu(cpu) {
1754 /* Leave it to be NULL. */
1755 if (cpu >= MAX_VIRT_CPUS)
1756 continue;
1757 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1758 }
1759}
1760
1761#ifdef CONFIG_XEN_PVHVM
1762static void __init init_hvm_pv_info(void)
1763{
1764 int major, minor;
1765 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1766 u64 pfn;
1767
1768 base = xen_cpuid_base();
1769 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1770
1771 major = eax >> 16;
1772 minor = eax & 0xffff;
1773 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1774
1775 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1776
1777 pfn = __pa(hypercall_page);
1778 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1779
1780 xen_setup_features();
1781
1782 pv_info.name = "Xen HVM";
1783
1784 xen_domain_type = XEN_HVM_DOMAIN;
1785}
1786
1787static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
1788 void *hcpu)
1789{
1790 int cpu = (long)hcpu;
1791 switch (action) {
1792 case CPU_UP_PREPARE:
1793 xen_vcpu_setup(cpu);
1794 if (xen_have_vector_callback) {
1795 if (xen_feature(XENFEAT_hvm_safe_pvclock))
1796 xen_setup_timer(cpu);
1797 }
1798 break;
1799 default:
1800 break;
1801 }
1802 return NOTIFY_OK;
1803}
1804
1805static struct notifier_block xen_hvm_cpu_notifier = {
1806 .notifier_call = xen_hvm_cpu_notify,
1807};
1808
1809#ifdef CONFIG_KEXEC_CORE
1810static void xen_hvm_shutdown(void)
1811{
1812 native_machine_shutdown();
1813 if (kexec_in_progress)
1814 xen_reboot(SHUTDOWN_soft_reset);
1815}
1816
1817static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1818{
1819 native_machine_crash_shutdown(regs);
1820 xen_reboot(SHUTDOWN_soft_reset);
1821}
1822#endif
1823
1824static void __init xen_hvm_guest_init(void)
1825{
1826 if (xen_pv_domain())
1827 return;
1828
1829 init_hvm_pv_info();
1830
1831 xen_hvm_init_shared_info();
1832
1833 xen_panic_handler_init();
1834
1835 if (xen_feature(XENFEAT_hvm_callback_vector))
1836 xen_have_vector_callback = 1;
1837 xen_hvm_smp_init();
1838 register_cpu_notifier(&xen_hvm_cpu_notifier);
1839 xen_unplug_emulated_devices();
1840 x86_init.irqs.intr_init = xen_init_IRQ;
1841 xen_hvm_init_time_ops();
1842 xen_hvm_init_mmu_ops();
1843#ifdef CONFIG_KEXEC_CORE
1844 machine_ops.shutdown = xen_hvm_shutdown;
1845 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1846#endif
1847}
1848#endif
1849
1850static bool xen_nopv = false;
1851static __init int xen_parse_nopv(char *arg)
1852{
1853 xen_nopv = true;
1854 return 0;
1855}
1856early_param("xen_nopv", xen_parse_nopv);
1857
1858static uint32_t __init xen_platform(void)
1859{
1860 if (xen_nopv)
1861 return 0;
1862
1863 return xen_cpuid_base();
1864}
1865
1866bool xen_hvm_need_lapic(void)
1867{
1868 if (xen_nopv)
1869 return false;
1870 if (xen_pv_domain())
1871 return false;
1872 if (!xen_hvm_domain())
1873 return false;
1874 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1875 return false;
1876 return true;
1877}
1878EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1879
1880static void xen_set_cpu_features(struct cpuinfo_x86 *c)
1881{
1882 if (xen_pv_domain()) {
1883 clear_cpu_bug(c, X86_BUG_SYSRET_SS_ATTRS);
1884 set_cpu_cap(c, X86_FEATURE_XENPV);
1885 }
1886}
1887
1888const struct hypervisor_x86 x86_hyper_xen = {
1889 .name = "Xen",
1890 .detect = xen_platform,
1891#ifdef CONFIG_XEN_PVHVM
1892 .init_platform = xen_hvm_guest_init,
1893#endif
1894 .x2apic_available = xen_x2apic_para_available,
1895 .set_cpu_features = xen_set_cpu_features,
1896};
1897EXPORT_SYMBOL(x86_hyper_xen);
1898
1899#ifdef CONFIG_HOTPLUG_CPU
1900void xen_arch_register_cpu(int num)
1901{
1902 arch_register_cpu(num);
1903}
1904EXPORT_SYMBOL(xen_arch_register_cpu);
1905
1906void xen_arch_unregister_cpu(int num)
1907{
1908 arch_unregister_cpu(num);
1909}
1910EXPORT_SYMBOL(xen_arch_unregister_cpu);
1911#endif
1912