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
36#include <xen/xen.h>
37#include <xen/events.h>
38#include <xen/interface/xen.h>
39#include <xen/interface/version.h>
40#include <xen/interface/physdev.h>
41#include <xen/interface/vcpu.h>
42#include <xen/interface/memory.h>
43#include <xen/interface/nmi.h>
44#include <xen/interface/xen-mca.h>
45#include <xen/features.h>
46#include <xen/page.h>
47#include <xen/hvm.h>
48#include <xen/hvc-console.h>
49#include <xen/acpi.h>
50
51#include <asm/paravirt.h>
52#include <asm/apic.h>
53#include <asm/page.h>
54#include <asm/xen/pci.h>
55#include <asm/xen/hypercall.h>
56#include <asm/xen/hypervisor.h>
57#include <asm/fixmap.h>
58#include <asm/processor.h>
59#include <asm/proto.h>
60#include <asm/msr-index.h>
61#include <asm/traps.h>
62#include <asm/setup.h>
63#include <asm/desc.h>
64#include <asm/pgalloc.h>
65#include <asm/pgtable.h>
66#include <asm/tlbflush.h>
67#include <asm/reboot.h>
68#include <asm/stackprotector.h>
69#include <asm/hypervisor.h>
70#include <asm/mach_traps.h>
71#include <asm/mwait.h>
72#include <asm/pci_x86.h>
73#include <asm/pat.h>
74
75#ifdef CONFIG_ACPI
76#include <linux/acpi.h>
77#include <asm/acpi.h>
78#include <acpi/pdc_intel.h>
79#include <acpi/processor.h>
80#include <xen/interface/platform.h>
81#endif
82
83#include "xen-ops.h"
84#include "mmu.h"
85#include "smp.h"
86#include "multicalls.h"
87
88EXPORT_SYMBOL_GPL(hypercall_page);
89
90/*
91 * Pointer to the xen_vcpu_info structure or
92 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
93 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
94 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
95 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
96 * acknowledge pending events.
97 * Also more subtly it is used by the patched version of irq enable/disable
98 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
99 *
100 * The desire to be able to do those mask/unmask operations as a single
101 * instruction by using the per-cpu offset held in %gs is the real reason
102 * vcpu info is in a per-cpu pointer and the original reason for this
103 * hypercall.
104 *
105 */
106DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
107
108/*
109 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
110 * hypercall. This can be used both in PV and PVHVM mode. The structure
111 * overrides the default per_cpu(xen_vcpu, cpu) value.
112 */
113DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
114
115enum xen_domain_type xen_domain_type = XEN_NATIVE;
116EXPORT_SYMBOL_GPL(xen_domain_type);
117
118unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
119EXPORT_SYMBOL(machine_to_phys_mapping);
120unsigned long machine_to_phys_nr;
121EXPORT_SYMBOL(machine_to_phys_nr);
122
123struct start_info *xen_start_info;
124EXPORT_SYMBOL_GPL(xen_start_info);
125
126struct shared_info xen_dummy_shared_info;
127
128void *xen_initial_gdt;
129
130RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
131__read_mostly int xen_have_vector_callback;
132EXPORT_SYMBOL_GPL(xen_have_vector_callback);
133
134/*
135 * Point at some empty memory to start with. We map the real shared_info
136 * page as soon as fixmap is up and running.
137 */
138struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
139
140/*
141 * Flag to determine whether vcpu info placement is available on all
142 * VCPUs. We assume it is to start with, and then set it to zero on
143 * the first failure. This is because it can succeed on some VCPUs
144 * and not others, since it can involve hypervisor memory allocation,
145 * or because the guest failed to guarantee all the appropriate
146 * constraints on all VCPUs (ie buffer can't cross a page boundary).
147 *
148 * Note that any particular CPU may be using a placed vcpu structure,
149 * but we can only optimise if the all are.
150 *
151 * 0: not available, 1: available
152 */
153static int have_vcpu_info_placement = 1;
154
155struct tls_descs {
156 struct desc_struct desc[3];
157};
158
159/*
160 * Updating the 3 TLS descriptors in the GDT on every task switch is
161 * surprisingly expensive so we avoid updating them if they haven't
162 * changed. Since Xen writes different descriptors than the one
163 * passed in the update_descriptor hypercall we keep shadow copies to
164 * compare against.
165 */
166static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
167
168static void clamp_max_cpus(void)
169{
170#ifdef CONFIG_SMP
171 if (setup_max_cpus > MAX_VIRT_CPUS)
172 setup_max_cpus = MAX_VIRT_CPUS;
173#endif
174}
175
176static void xen_vcpu_setup(int cpu)
177{
178 struct vcpu_register_vcpu_info info;
179 int err;
180 struct vcpu_info *vcpup;
181
182 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
183
184 /*
185 * This path is called twice on PVHVM - first during bootup via
186 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
187 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
188 * As we can only do the VCPUOP_register_vcpu_info once lets
189 * not over-write its result.
190 *
191 * For PV it is called during restore (xen_vcpu_restore) and bootup
192 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
193 * use this function.
194 */
195 if (xen_hvm_domain()) {
196 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
197 return;
198 }
199 if (cpu < MAX_VIRT_CPUS)
200 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
201
202 if (!have_vcpu_info_placement) {
203 if (cpu >= MAX_VIRT_CPUS)
204 clamp_max_cpus();
205 return;
206 }
207
208 vcpup = &per_cpu(xen_vcpu_info, cpu);
209 info.mfn = arbitrary_virt_to_mfn(vcpup);
210 info.offset = offset_in_page(vcpup);
211
212 /* Check to see if the hypervisor will put the vcpu_info
213 structure where we want it, which allows direct access via
214 a percpu-variable.
215 N.B. This hypercall can _only_ be called once per CPU. Subsequent
216 calls will error out with -EINVAL. This is due to the fact that
217 hypervisor has no unregister variant and this hypercall does not
218 allow to over-write info.mfn and info.offset.
219 */
220 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
221
222 if (err) {
223 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
224 have_vcpu_info_placement = 0;
225 clamp_max_cpus();
226 } else {
227 /* This cpu is using the registered vcpu info, even if
228 later ones fail to. */
229 per_cpu(xen_vcpu, cpu) = vcpup;
230 }
231}
232
233/*
234 * On restore, set the vcpu placement up again.
235 * If it fails, then we're in a bad state, since
236 * we can't back out from using it...
237 */
238void xen_vcpu_restore(void)
239{
240 int cpu;
241
242 for_each_possible_cpu(cpu) {
243 bool other_cpu = (cpu != smp_processor_id());
244 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL);
245
246 if (other_cpu && is_up &&
247 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
248 BUG();
249
250 xen_setup_runstate_info(cpu);
251
252 if (have_vcpu_info_placement)
253 xen_vcpu_setup(cpu);
254
255 if (other_cpu && is_up &&
256 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
257 BUG();
258 }
259}
260
261static void __init xen_banner(void)
262{
263 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
264 struct xen_extraversion extra;
265 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
266
267 pr_info("Booting paravirtualized kernel %son %s\n",
268 xen_feature(XENFEAT_auto_translated_physmap) ?
269 "with PVH extensions " : "", pv_info.name);
270 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
271 version >> 16, version & 0xffff, extra.extraversion,
272 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
273}
274/* Check if running on Xen version (major, minor) or later */
275bool
276xen_running_on_version_or_later(unsigned int major, unsigned int minor)
277{
278 unsigned int version;
279
280 if (!xen_domain())
281 return false;
282
283 version = HYPERVISOR_xen_version(XENVER_version, NULL);
284 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
285 ((version >> 16) > major))
286 return true;
287 return false;
288}
289
290#define CPUID_THERM_POWER_LEAF 6
291#define APERFMPERF_PRESENT 0
292
293static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
294static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
295
296static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
297static __read_mostly unsigned int cpuid_leaf5_ecx_val;
298static __read_mostly unsigned int cpuid_leaf5_edx_val;
299
300static void xen_cpuid(unsigned int *ax, unsigned int *bx,
301 unsigned int *cx, unsigned int *dx)
302{
303 unsigned maskebx = ~0;
304 unsigned maskecx = ~0;
305 unsigned maskedx = ~0;
306 unsigned setecx = 0;
307 /*
308 * Mask out inconvenient features, to try and disable as many
309 * unsupported kernel subsystems as possible.
310 */
311 switch (*ax) {
312 case 1:
313 maskecx = cpuid_leaf1_ecx_mask;
314 setecx = cpuid_leaf1_ecx_set_mask;
315 maskedx = cpuid_leaf1_edx_mask;
316 break;
317
318 case CPUID_MWAIT_LEAF:
319 /* Synthesize the values.. */
320 *ax = 0;
321 *bx = 0;
322 *cx = cpuid_leaf5_ecx_val;
323 *dx = cpuid_leaf5_edx_val;
324 return;
325
326 case CPUID_THERM_POWER_LEAF:
327 /* Disabling APERFMPERF for kernel usage */
328 maskecx = ~(1 << APERFMPERF_PRESENT);
329 break;
330
331 case 0xb:
332 /* Suppress extended topology stuff */
333 maskebx = 0;
334 break;
335 }
336
337 asm(XEN_EMULATE_PREFIX "cpuid"
338 : "=a" (*ax),
339 "=b" (*bx),
340 "=c" (*cx),
341 "=d" (*dx)
342 : "0" (*ax), "2" (*cx));
343
344 *bx &= maskebx;
345 *cx &= maskecx;
346 *cx |= setecx;
347 *dx &= maskedx;
348
349}
350
351static bool __init xen_check_mwait(void)
352{
353#ifdef CONFIG_ACPI
354 struct xen_platform_op op = {
355 .cmd = XENPF_set_processor_pminfo,
356 .u.set_pminfo.id = -1,
357 .u.set_pminfo.type = XEN_PM_PDC,
358 };
359 uint32_t buf[3];
360 unsigned int ax, bx, cx, dx;
361 unsigned int mwait_mask;
362
363 /* We need to determine whether it is OK to expose the MWAIT
364 * capability to the kernel to harvest deeper than C3 states from ACPI
365 * _CST using the processor_harvest_xen.c module. For this to work, we
366 * need to gather the MWAIT_LEAF values (which the cstate.c code
367 * checks against). The hypervisor won't expose the MWAIT flag because
368 * it would break backwards compatibility; so we will find out directly
369 * from the hardware and hypercall.
370 */
371 if (!xen_initial_domain())
372 return false;
373
374 /*
375 * When running under platform earlier than Xen4.2, do not expose
376 * mwait, to avoid the risk of loading native acpi pad driver
377 */
378 if (!xen_running_on_version_or_later(4, 2))
379 return false;
380
381 ax = 1;
382 cx = 0;
383
384 native_cpuid(&ax, &bx, &cx, &dx);
385
386 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
387 (1 << (X86_FEATURE_MWAIT % 32));
388
389 if ((cx & mwait_mask) != mwait_mask)
390 return false;
391
392 /* We need to emulate the MWAIT_LEAF and for that we need both
393 * ecx and edx. The hypercall provides only partial information.
394 */
395
396 ax = CPUID_MWAIT_LEAF;
397 bx = 0;
398 cx = 0;
399 dx = 0;
400
401 native_cpuid(&ax, &bx, &cx, &dx);
402
403 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
404 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
405 */
406 buf[0] = ACPI_PDC_REVISION_ID;
407 buf[1] = 1;
408 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
409
410 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
411
412 if ((HYPERVISOR_dom0_op(&op) == 0) &&
413 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
414 cpuid_leaf5_ecx_val = cx;
415 cpuid_leaf5_edx_val = dx;
416 }
417 return true;
418#else
419 return false;
420#endif
421}
422static void __init xen_init_cpuid_mask(void)
423{
424 unsigned int ax, bx, cx, dx;
425 unsigned int xsave_mask;
426
427 cpuid_leaf1_edx_mask =
428 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
429 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
430
431 if (!xen_initial_domain())
432 cpuid_leaf1_edx_mask &=
433 ~((1 << X86_FEATURE_ACPI)); /* disable ACPI */
434
435 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
436
437 ax = 1;
438 cx = 0;
439 cpuid(1, &ax, &bx, &cx, &dx);
440
441 xsave_mask =
442 (1 << (X86_FEATURE_XSAVE % 32)) |
443 (1 << (X86_FEATURE_OSXSAVE % 32));
444
445 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
446 if ((cx & xsave_mask) != xsave_mask)
447 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
448 if (xen_check_mwait())
449 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
450}
451
452static void xen_set_debugreg(int reg, unsigned long val)
453{
454 HYPERVISOR_set_debugreg(reg, val);
455}
456
457static unsigned long xen_get_debugreg(int reg)
458{
459 return HYPERVISOR_get_debugreg(reg);
460}
461
462static void xen_end_context_switch(struct task_struct *next)
463{
464 xen_mc_flush();
465 paravirt_end_context_switch(next);
466}
467
468static unsigned long xen_store_tr(void)
469{
470 return 0;
471}
472
473/*
474 * Set the page permissions for a particular virtual address. If the
475 * address is a vmalloc mapping (or other non-linear mapping), then
476 * find the linear mapping of the page and also set its protections to
477 * match.
478 */
479static void set_aliased_prot(void *v, pgprot_t prot)
480{
481 int level;
482 pte_t *ptep;
483 pte_t pte;
484 unsigned long pfn;
485 struct page *page;
486 unsigned char dummy;
487
488 ptep = lookup_address((unsigned long)v, &level);
489 BUG_ON(ptep == NULL);
490
491 pfn = pte_pfn(*ptep);
492 page = pfn_to_page(pfn);
493
494 pte = pfn_pte(pfn, prot);
495
496 /*
497 * Careful: update_va_mapping() will fail if the virtual address
498 * we're poking isn't populated in the page tables. We don't
499 * need to worry about the direct map (that's always in the page
500 * tables), but we need to be careful about vmap space. In
501 * particular, the top level page table can lazily propagate
502 * entries between processes, so if we've switched mms since we
503 * vmapped the target in the first place, we might not have the
504 * top-level page table entry populated.
505 *
506 * We disable preemption because we want the same mm active when
507 * we probe the target and when we issue the hypercall. We'll
508 * have the same nominal mm, but if we're a kernel thread, lazy
509 * mm dropping could change our pgd.
510 *
511 * Out of an abundance of caution, this uses __get_user() to fault
512 * in the target address just in case there's some obscure case
513 * in which the target address isn't readable.
514 */
515
516 preempt_disable();
517
518 pagefault_disable(); /* Avoid warnings due to being atomic. */
519 __get_user(dummy, (unsigned char __user __force *)v);
520 pagefault_enable();
521
522 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
523 BUG();
524
525 if (!PageHighMem(page)) {
526 void *av = __va(PFN_PHYS(pfn));
527
528 if (av != v)
529 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
530 BUG();
531 } else
532 kmap_flush_unused();
533
534 preempt_enable();
535}
536
537static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
538{
539 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
540 int i;
541
542 /*
543 * We need to mark the all aliases of the LDT pages RO. We
544 * don't need to call vm_flush_aliases(), though, since that's
545 * only responsible for flushing aliases out the TLBs, not the
546 * page tables, and Xen will flush the TLB for us if needed.
547 *
548 * To avoid confusing future readers: none of this is necessary
549 * to load the LDT. The hypervisor only checks this when the
550 * LDT is faulted in due to subsequent descriptor access.
551 */
552
553 for(i = 0; i < entries; i += entries_per_page)
554 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
555}
556
557static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
558{
559 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
560 int i;
561
562 for(i = 0; i < entries; i += entries_per_page)
563 set_aliased_prot(ldt + i, PAGE_KERNEL);
564}
565
566static void xen_set_ldt(const void *addr, unsigned entries)
567{
568 struct mmuext_op *op;
569 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
570
571 trace_xen_cpu_set_ldt(addr, entries);
572
573 op = mcs.args;
574 op->cmd = MMUEXT_SET_LDT;
575 op->arg1.linear_addr = (unsigned long)addr;
576 op->arg2.nr_ents = entries;
577
578 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
579
580 xen_mc_issue(PARAVIRT_LAZY_CPU);
581}
582
583static void xen_load_gdt(const struct desc_ptr *dtr)
584{
585 unsigned long va = dtr->address;
586 unsigned int size = dtr->size + 1;
587 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
588 unsigned long frames[pages];
589 int f;
590
591 /*
592 * A GDT can be up to 64k in size, which corresponds to 8192
593 * 8-byte entries, or 16 4k pages..
594 */
595
596 BUG_ON(size > 65536);
597 BUG_ON(va & ~PAGE_MASK);
598
599 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
600 int level;
601 pte_t *ptep;
602 unsigned long pfn, mfn;
603 void *virt;
604
605 /*
606 * The GDT is per-cpu and is in the percpu data area.
607 * That can be virtually mapped, so we need to do a
608 * page-walk to get the underlying MFN for the
609 * hypercall. The page can also be in the kernel's
610 * linear range, so we need to RO that mapping too.
611 */
612 ptep = lookup_address(va, &level);
613 BUG_ON(ptep == NULL);
614
615 pfn = pte_pfn(*ptep);
616 mfn = pfn_to_mfn(pfn);
617 virt = __va(PFN_PHYS(pfn));
618
619 frames[f] = mfn;
620
621 make_lowmem_page_readonly((void *)va);
622 make_lowmem_page_readonly(virt);
623 }
624
625 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
626 BUG();
627}
628
629/*
630 * load_gdt for early boot, when the gdt is only mapped once
631 */
632static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
633{
634 unsigned long va = dtr->address;
635 unsigned int size = dtr->size + 1;
636 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
637 unsigned long frames[pages];
638 int f;
639
640 /*
641 * A GDT can be up to 64k in size, which corresponds to 8192
642 * 8-byte entries, or 16 4k pages..
643 */
644
645 BUG_ON(size > 65536);
646 BUG_ON(va & ~PAGE_MASK);
647
648 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
649 pte_t pte;
650 unsigned long pfn, mfn;
651
652 pfn = virt_to_pfn(va);
653 mfn = pfn_to_mfn(pfn);
654
655 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
656
657 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
658 BUG();
659
660 frames[f] = mfn;
661 }
662
663 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
664 BUG();
665}
666
667static inline bool desc_equal(const struct desc_struct *d1,
668 const struct desc_struct *d2)
669{
670 return d1->a == d2->a && d1->b == d2->b;
671}
672
673static void load_TLS_descriptor(struct thread_struct *t,
674 unsigned int cpu, unsigned int i)
675{
676 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
677 struct desc_struct *gdt;
678 xmaddr_t maddr;
679 struct multicall_space mc;
680
681 if (desc_equal(shadow, &t->tls_array[i]))
682 return;
683
684 *shadow = t->tls_array[i];
685
686 gdt = get_cpu_gdt_table(cpu);
687 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
688 mc = __xen_mc_entry(0);
689
690 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
691}
692
693static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
694{
695 /*
696 * XXX sleazy hack: If we're being called in a lazy-cpu zone
697 * and lazy gs handling is enabled, it means we're in a
698 * context switch, and %gs has just been saved. This means we
699 * can zero it out to prevent faults on exit from the
700 * hypervisor if the next process has no %gs. Either way, it
701 * has been saved, and the new value will get loaded properly.
702 * This will go away as soon as Xen has been modified to not
703 * save/restore %gs for normal hypercalls.
704 *
705 * On x86_64, this hack is not used for %gs, because gs points
706 * to KERNEL_GS_BASE (and uses it for PDA references), so we
707 * must not zero %gs on x86_64
708 *
709 * For x86_64, we need to zero %fs, otherwise we may get an
710 * exception between the new %fs descriptor being loaded and
711 * %fs being effectively cleared at __switch_to().
712 */
713 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
714#ifdef CONFIG_X86_32
715 lazy_load_gs(0);
716#else
717 loadsegment(fs, 0);
718#endif
719 }
720
721 xen_mc_batch();
722
723 load_TLS_descriptor(t, cpu, 0);
724 load_TLS_descriptor(t, cpu, 1);
725 load_TLS_descriptor(t, cpu, 2);
726
727 xen_mc_issue(PARAVIRT_LAZY_CPU);
728}
729
730#ifdef CONFIG_X86_64
731static void xen_load_gs_index(unsigned int idx)
732{
733 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
734 BUG();
735}
736#endif
737
738static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
739 const void *ptr)
740{
741 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
742 u64 entry = *(u64 *)ptr;
743
744 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
745
746 preempt_disable();
747
748 xen_mc_flush();
749 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
750 BUG();
751
752 preempt_enable();
753}
754
755static int cvt_gate_to_trap(int vector, const gate_desc *val,
756 struct trap_info *info)
757{
758 unsigned long addr;
759
760 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
761 return 0;
762
763 info->vector = vector;
764
765 addr = gate_offset(*val);
766#ifdef CONFIG_X86_64
767 /*
768 * Look for known traps using IST, and substitute them
769 * appropriately. The debugger ones are the only ones we care
770 * about. Xen will handle faults like double_fault,
771 * so we should never see them. Warn if
772 * there's an unexpected IST-using fault handler.
773 */
774 if (addr == (unsigned long)debug)
775 addr = (unsigned long)xen_debug;
776 else if (addr == (unsigned long)int3)
777 addr = (unsigned long)xen_int3;
778 else if (addr == (unsigned long)stack_segment)
779 addr = (unsigned long)xen_stack_segment;
780 else if (addr == (unsigned long)double_fault) {
781 /* Don't need to handle these */
782 return 0;
783#ifdef CONFIG_X86_MCE
784 } else if (addr == (unsigned long)machine_check) {
785 /*
786 * when xen hypervisor inject vMCE to guest,
787 * use native mce handler to handle it
788 */
789 ;
790#endif
791 } else if (addr == (unsigned long)nmi)
792 /*
793 * Use the native version as well.
794 */
795 ;
796 else {
797 /* Some other trap using IST? */
798 if (WARN_ON(val->ist != 0))
799 return 0;
800 }
801#endif /* CONFIG_X86_64 */
802 info->address = addr;
803
804 info->cs = gate_segment(*val);
805 info->flags = val->dpl;
806 /* interrupt gates clear IF */
807 if (val->type == GATE_INTERRUPT)
808 info->flags |= 1 << 2;
809
810 return 1;
811}
812
813/* Locations of each CPU's IDT */
814static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
815
816/* Set an IDT entry. If the entry is part of the current IDT, then
817 also update Xen. */
818static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
819{
820 unsigned long p = (unsigned long)&dt[entrynum];
821 unsigned long start, end;
822
823 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
824
825 preempt_disable();
826
827 start = __this_cpu_read(idt_desc.address);
828 end = start + __this_cpu_read(idt_desc.size) + 1;
829
830 xen_mc_flush();
831
832 native_write_idt_entry(dt, entrynum, g);
833
834 if (p >= start && (p + 8) <= end) {
835 struct trap_info info[2];
836
837 info[1].address = 0;
838
839 if (cvt_gate_to_trap(entrynum, g, &info[0]))
840 if (HYPERVISOR_set_trap_table(info))
841 BUG();
842 }
843
844 preempt_enable();
845}
846
847static void xen_convert_trap_info(const struct desc_ptr *desc,
848 struct trap_info *traps)
849{
850 unsigned in, out, count;
851
852 count = (desc->size+1) / sizeof(gate_desc);
853 BUG_ON(count > 256);
854
855 for (in = out = 0; in < count; in++) {
856 gate_desc *entry = (gate_desc*)(desc->address) + in;
857
858 if (cvt_gate_to_trap(in, entry, &traps[out]))
859 out++;
860 }
861 traps[out].address = 0;
862}
863
864void xen_copy_trap_info(struct trap_info *traps)
865{
866 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
867
868 xen_convert_trap_info(desc, traps);
869}
870
871/* Load a new IDT into Xen. In principle this can be per-CPU, so we
872 hold a spinlock to protect the static traps[] array (static because
873 it avoids allocation, and saves stack space). */
874static void xen_load_idt(const struct desc_ptr *desc)
875{
876 static DEFINE_SPINLOCK(lock);
877 static struct trap_info traps[257];
878
879 trace_xen_cpu_load_idt(desc);
880
881 spin_lock(&lock);
882
883 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
884
885 xen_convert_trap_info(desc, traps);
886
887 xen_mc_flush();
888 if (HYPERVISOR_set_trap_table(traps))
889 BUG();
890
891 spin_unlock(&lock);
892}
893
894/* Write a GDT descriptor entry. Ignore LDT descriptors, since
895 they're handled differently. */
896static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
897 const void *desc, int type)
898{
899 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
900
901 preempt_disable();
902
903 switch (type) {
904 case DESC_LDT:
905 case DESC_TSS:
906 /* ignore */
907 break;
908
909 default: {
910 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
911
912 xen_mc_flush();
913 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
914 BUG();
915 }
916
917 }
918
919 preempt_enable();
920}
921
922/*
923 * Version of write_gdt_entry for use at early boot-time needed to
924 * update an entry as simply as possible.
925 */
926static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
927 const void *desc, int type)
928{
929 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
930
931 switch (type) {
932 case DESC_LDT:
933 case DESC_TSS:
934 /* ignore */
935 break;
936
937 default: {
938 xmaddr_t maddr = virt_to_machine(&dt[entry]);
939
940 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
941 dt[entry] = *(struct desc_struct *)desc;
942 }
943
944 }
945}
946
947static void xen_load_sp0(struct tss_struct *tss,
948 struct thread_struct *thread)
949{
950 struct multicall_space mcs;
951
952 mcs = xen_mc_entry(0);
953 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
954 xen_mc_issue(PARAVIRT_LAZY_CPU);
955 tss->x86_tss.sp0 = thread->sp0;
956}
957
958static void xen_set_iopl_mask(unsigned mask)
959{
960 struct physdev_set_iopl set_iopl;
961
962 /* Force the change at ring 0. */
963 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
964 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
965}
966
967static void xen_io_delay(void)
968{
969}
970
971static void xen_clts(void)
972{
973 struct multicall_space mcs;
974
975 mcs = xen_mc_entry(0);
976
977 MULTI_fpu_taskswitch(mcs.mc, 0);
978
979 xen_mc_issue(PARAVIRT_LAZY_CPU);
980}
981
982static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
983
984static unsigned long xen_read_cr0(void)
985{
986 unsigned long cr0 = this_cpu_read(xen_cr0_value);
987
988 if (unlikely(cr0 == 0)) {
989 cr0 = native_read_cr0();
990 this_cpu_write(xen_cr0_value, cr0);
991 }
992
993 return cr0;
994}
995
996static void xen_write_cr0(unsigned long cr0)
997{
998 struct multicall_space mcs;
999
1000 this_cpu_write(xen_cr0_value, cr0);
1001
1002 /* Only pay attention to cr0.TS; everything else is
1003 ignored. */
1004 mcs = xen_mc_entry(0);
1005
1006 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1007
1008 xen_mc_issue(PARAVIRT_LAZY_CPU);
1009}
1010
1011static void xen_write_cr4(unsigned long cr4)
1012{
1013 cr4 &= ~X86_CR4_PGE;
1014 cr4 &= ~X86_CR4_PSE;
1015
1016 native_write_cr4(cr4);
1017}
1018#ifdef CONFIG_X86_64
1019static inline unsigned long xen_read_cr8(void)
1020{
1021 return 0;
1022}
1023static inline void xen_write_cr8(unsigned long val)
1024{
1025 BUG_ON(val);
1026}
1027#endif
1028
1029static u64 xen_read_msr_safe(unsigned int msr, int *err)
1030{
1031 u64 val;
1032
1033 val = native_read_msr_safe(msr, err);
1034 switch (msr) {
1035 case MSR_IA32_APICBASE:
1036#ifdef CONFIG_X86_X2APIC
1037 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1038#endif
1039 val &= ~X2APIC_ENABLE;
1040 break;
1041 }
1042 return val;
1043}
1044
1045static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1046{
1047 int ret;
1048
1049 ret = 0;
1050
1051 switch (msr) {
1052#ifdef CONFIG_X86_64
1053 unsigned which;
1054 u64 base;
1055
1056 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
1057 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
1058 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
1059
1060 set:
1061 base = ((u64)high << 32) | low;
1062 if (HYPERVISOR_set_segment_base(which, base) != 0)
1063 ret = -EIO;
1064 break;
1065#endif
1066
1067 case MSR_STAR:
1068 case MSR_CSTAR:
1069 case MSR_LSTAR:
1070 case MSR_SYSCALL_MASK:
1071 case MSR_IA32_SYSENTER_CS:
1072 case MSR_IA32_SYSENTER_ESP:
1073 case MSR_IA32_SYSENTER_EIP:
1074 /* Fast syscall setup is all done in hypercalls, so
1075 these are all ignored. Stub them out here to stop
1076 Xen console noise. */
1077
1078 default:
1079 ret = native_write_msr_safe(msr, low, high);
1080 }
1081
1082 return ret;
1083}
1084
1085void xen_setup_shared_info(void)
1086{
1087 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1088 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1089 xen_start_info->shared_info);
1090
1091 HYPERVISOR_shared_info =
1092 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1093 } else
1094 HYPERVISOR_shared_info =
1095 (struct shared_info *)__va(xen_start_info->shared_info);
1096
1097#ifndef CONFIG_SMP
1098 /* In UP this is as good a place as any to set up shared info */
1099 xen_setup_vcpu_info_placement();
1100#endif
1101
1102 xen_setup_mfn_list_list();
1103}
1104
1105/* This is called once we have the cpu_possible_mask */
1106void xen_setup_vcpu_info_placement(void)
1107{
1108 int cpu;
1109
1110 for_each_possible_cpu(cpu)
1111 xen_vcpu_setup(cpu);
1112
1113 /* xen_vcpu_setup managed to place the vcpu_info within the
1114 * percpu area for all cpus, so make use of it. Note that for
1115 * PVH we want to use native IRQ mechanism. */
1116 if (have_vcpu_info_placement && !xen_pvh_domain()) {
1117 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1118 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1119 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1120 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1121 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1122 }
1123}
1124
1125static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1126 unsigned long addr, unsigned len)
1127{
1128 char *start, *end, *reloc;
1129 unsigned ret;
1130
1131 start = end = reloc = NULL;
1132
1133#define SITE(op, x) \
1134 case PARAVIRT_PATCH(op.x): \
1135 if (have_vcpu_info_placement) { \
1136 start = (char *)xen_##x##_direct; \
1137 end = xen_##x##_direct_end; \
1138 reloc = xen_##x##_direct_reloc; \
1139 } \
1140 goto patch_site
1141
1142 switch (type) {
1143 SITE(pv_irq_ops, irq_enable);
1144 SITE(pv_irq_ops, irq_disable);
1145 SITE(pv_irq_ops, save_fl);
1146 SITE(pv_irq_ops, restore_fl);
1147#undef SITE
1148
1149 patch_site:
1150 if (start == NULL || (end-start) > len)
1151 goto default_patch;
1152
1153 ret = paravirt_patch_insns(insnbuf, len, start, end);
1154
1155 /* Note: because reloc is assigned from something that
1156 appears to be an array, gcc assumes it's non-null,
1157 but doesn't know its relationship with start and
1158 end. */
1159 if (reloc > start && reloc < end) {
1160 int reloc_off = reloc - start;
1161 long *relocp = (long *)(insnbuf + reloc_off);
1162 long delta = start - (char *)addr;
1163
1164 *relocp += delta;
1165 }
1166 break;
1167
1168 default_patch:
1169 default:
1170 ret = paravirt_patch_default(type, clobbers, insnbuf,
1171 addr, len);
1172 break;
1173 }
1174
1175 return ret;
1176}
1177
1178static const struct pv_info xen_info __initconst = {
1179 .paravirt_enabled = 1,
1180 .shared_kernel_pmd = 0,
1181
1182#ifdef CONFIG_X86_64
1183 .extra_user_64bit_cs = FLAT_USER_CS64,
1184#endif
1185
1186 .name = "Xen",
1187};
1188
1189static const struct pv_init_ops xen_init_ops __initconst = {
1190 .patch = xen_patch,
1191};
1192
1193static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1194 .cpuid = xen_cpuid,
1195
1196 .set_debugreg = xen_set_debugreg,
1197 .get_debugreg = xen_get_debugreg,
1198
1199 .clts = xen_clts,
1200
1201 .read_cr0 = xen_read_cr0,
1202 .write_cr0 = xen_write_cr0,
1203
1204 .read_cr4 = native_read_cr4,
1205 .read_cr4_safe = native_read_cr4_safe,
1206 .write_cr4 = xen_write_cr4,
1207
1208#ifdef CONFIG_X86_64
1209 .read_cr8 = xen_read_cr8,
1210 .write_cr8 = xen_write_cr8,
1211#endif
1212
1213 .wbinvd = native_wbinvd,
1214
1215 .read_msr = xen_read_msr_safe,
1216 .write_msr = xen_write_msr_safe,
1217
1218 .read_tsc = native_read_tsc,
1219 .read_pmc = native_read_pmc,
1220
1221 .read_tscp = native_read_tscp,
1222
1223 .iret = xen_iret,
1224#ifdef CONFIG_X86_64
1225 .usergs_sysret32 = xen_sysret32,
1226 .usergs_sysret64 = xen_sysret64,
1227#else
1228 .irq_enable_sysexit = xen_sysexit,
1229#endif
1230
1231 .load_tr_desc = paravirt_nop,
1232 .set_ldt = xen_set_ldt,
1233 .load_gdt = xen_load_gdt,
1234 .load_idt = xen_load_idt,
1235 .load_tls = xen_load_tls,
1236#ifdef CONFIG_X86_64
1237 .load_gs_index = xen_load_gs_index,
1238#endif
1239
1240 .alloc_ldt = xen_alloc_ldt,
1241 .free_ldt = xen_free_ldt,
1242
1243 .store_idt = native_store_idt,
1244 .store_tr = xen_store_tr,
1245
1246 .write_ldt_entry = xen_write_ldt_entry,
1247 .write_gdt_entry = xen_write_gdt_entry,
1248 .write_idt_entry = xen_write_idt_entry,
1249 .load_sp0 = xen_load_sp0,
1250
1251 .set_iopl_mask = xen_set_iopl_mask,
1252 .io_delay = xen_io_delay,
1253
1254 /* Xen takes care of %gs when switching to usermode for us */
1255 .swapgs = paravirt_nop,
1256
1257 .start_context_switch = paravirt_start_context_switch,
1258 .end_context_switch = xen_end_context_switch,
1259};
1260
1261static const struct pv_apic_ops xen_apic_ops __initconst = {
1262#ifdef CONFIG_X86_LOCAL_APIC
1263 .startup_ipi_hook = paravirt_nop,
1264#endif
1265};
1266
1267static void xen_reboot(int reason)
1268{
1269 struct sched_shutdown r = { .reason = reason };
1270
1271 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1272 BUG();
1273}
1274
1275static void xen_restart(char *msg)
1276{
1277 xen_reboot(SHUTDOWN_reboot);
1278}
1279
1280static void xen_emergency_restart(void)
1281{
1282 xen_reboot(SHUTDOWN_reboot);
1283}
1284
1285static void xen_machine_halt(void)
1286{
1287 xen_reboot(SHUTDOWN_poweroff);
1288}
1289
1290static void xen_machine_power_off(void)
1291{
1292 if (pm_power_off)
1293 pm_power_off();
1294 xen_reboot(SHUTDOWN_poweroff);
1295}
1296
1297static void xen_crash_shutdown(struct pt_regs *regs)
1298{
1299 xen_reboot(SHUTDOWN_crash);
1300}
1301
1302static int
1303xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1304{
1305 xen_reboot(SHUTDOWN_crash);
1306 return NOTIFY_DONE;
1307}
1308
1309static struct notifier_block xen_panic_block = {
1310 .notifier_call= xen_panic_event,
1311 .priority = INT_MIN
1312};
1313
1314int xen_panic_handler_init(void)
1315{
1316 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1317 return 0;
1318}
1319
1320static const struct machine_ops xen_machine_ops __initconst = {
1321 .restart = xen_restart,
1322 .halt = xen_machine_halt,
1323 .power_off = xen_machine_power_off,
1324 .shutdown = xen_machine_halt,
1325 .crash_shutdown = xen_crash_shutdown,
1326 .emergency_restart = xen_emergency_restart,
1327};
1328
1329static unsigned char xen_get_nmi_reason(void)
1330{
1331 unsigned char reason = 0;
1332
1333 /* Construct a value which looks like it came from port 0x61. */
1334 if (test_bit(_XEN_NMIREASON_io_error,
1335 &HYPERVISOR_shared_info->arch.nmi_reason))
1336 reason |= NMI_REASON_IOCHK;
1337 if (test_bit(_XEN_NMIREASON_pci_serr,
1338 &HYPERVISOR_shared_info->arch.nmi_reason))
1339 reason |= NMI_REASON_SERR;
1340
1341 return reason;
1342}
1343
1344static void __init xen_boot_params_init_edd(void)
1345{
1346#if IS_ENABLED(CONFIG_EDD)
1347 struct xen_platform_op op;
1348 struct edd_info *edd_info;
1349 u32 *mbr_signature;
1350 unsigned nr;
1351 int ret;
1352
1353 edd_info = boot_params.eddbuf;
1354 mbr_signature = boot_params.edd_mbr_sig_buffer;
1355
1356 op.cmd = XENPF_firmware_info;
1357
1358 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1359 for (nr = 0; nr < EDDMAXNR; nr++) {
1360 struct edd_info *info = edd_info + nr;
1361
1362 op.u.firmware_info.index = nr;
1363 info->params.length = sizeof(info->params);
1364 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1365 &info->params);
1366 ret = HYPERVISOR_dom0_op(&op);
1367 if (ret)
1368 break;
1369
1370#define C(x) info->x = op.u.firmware_info.u.disk_info.x
1371 C(device);
1372 C(version);
1373 C(interface_support);
1374 C(legacy_max_cylinder);
1375 C(legacy_max_head);
1376 C(legacy_sectors_per_track);
1377#undef C
1378 }
1379 boot_params.eddbuf_entries = nr;
1380
1381 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1382 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1383 op.u.firmware_info.index = nr;
1384 ret = HYPERVISOR_dom0_op(&op);
1385 if (ret)
1386 break;
1387 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1388 }
1389 boot_params.edd_mbr_sig_buf_entries = nr;
1390#endif
1391}
1392
1393/*
1394 * Set up the GDT and segment registers for -fstack-protector. Until
1395 * we do this, we have to be careful not to call any stack-protected
1396 * function, which is most of the kernel.
1397 *
1398 * Note, that it is __ref because the only caller of this after init
1399 * is PVH which is not going to use xen_load_gdt_boot or other
1400 * __init functions.
1401 */
1402static void __ref xen_setup_gdt(int cpu)
1403{
1404 if (xen_feature(XENFEAT_auto_translated_physmap)) {
1405#ifdef CONFIG_X86_64
1406 unsigned long dummy;
1407
1408 load_percpu_segment(cpu); /* We need to access per-cpu area */
1409 switch_to_new_gdt(cpu); /* GDT and GS set */
1410
1411 /* We are switching of the Xen provided GDT to our HVM mode
1412 * GDT. The new GDT has __KERNEL_CS with CS.L = 1
1413 * and we are jumping to reload it.
1414 */
1415 asm volatile ("pushq %0\n"
1416 "leaq 1f(%%rip),%0\n"
1417 "pushq %0\n"
1418 "lretq\n"
1419 "1:\n"
1420 : "=&r" (dummy) : "0" (__KERNEL_CS));
1421
1422 /*
1423 * While not needed, we also set the %es, %ds, and %fs
1424 * to zero. We don't care about %ss as it is NULL.
1425 * Strictly speaking this is not needed as Xen zeros those
1426 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1427 *
1428 * Linux zeros them in cpu_init() and in secondary_startup_64
1429 * (for BSP).
1430 */
1431 loadsegment(es, 0);
1432 loadsegment(ds, 0);
1433 loadsegment(fs, 0);
1434#else
1435 /* PVH: TODO Implement. */
1436 BUG();
1437#endif
1438 return; /* PVH does not need any PV GDT ops. */
1439 }
1440 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1441 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1442
1443 setup_stack_canary_segment(0);
1444 switch_to_new_gdt(0);
1445
1446 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1447 pv_cpu_ops.load_gdt = xen_load_gdt;
1448}
1449
1450#ifdef CONFIG_XEN_PVH
1451/*
1452 * A PV guest starts with default flags that are not set for PVH, set them
1453 * here asap.
1454 */
1455static void xen_pvh_set_cr_flags(int cpu)
1456{
1457
1458 /* Some of these are setup in 'secondary_startup_64'. The others:
1459 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1460 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1461 write_cr0(read_cr0() | X86_CR0_MP | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM);
1462
1463 if (!cpu)
1464 return;
1465 /*
1466 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
1467 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu__init_cpu().
1468 */
1469 if (cpu_has_pse)
1470 cr4_set_bits_and_update_boot(X86_CR4_PSE);
1471
1472 if (cpu_has_pge)
1473 cr4_set_bits_and_update_boot(X86_CR4_PGE);
1474}
1475
1476/*
1477 * Note, that it is ref - because the only caller of this after init
1478 * is PVH which is not going to use xen_load_gdt_boot or other
1479 * __init functions.
1480 */
1481void __ref xen_pvh_secondary_vcpu_init(int cpu)
1482{
1483 xen_setup_gdt(cpu);
1484 xen_pvh_set_cr_flags(cpu);
1485}
1486
1487static void __init xen_pvh_early_guest_init(void)
1488{
1489 if (!xen_feature(XENFEAT_auto_translated_physmap))
1490 return;
1491
1492 if (!xen_feature(XENFEAT_hvm_callback_vector))
1493 return;
1494
1495 xen_have_vector_callback = 1;
1496
1497 xen_pvh_early_cpu_init(0, false);
1498 xen_pvh_set_cr_flags(0);
1499
1500#ifdef CONFIG_X86_32
1501 BUG(); /* PVH: Implement proper support. */
1502#endif
1503}
1504#endif /* CONFIG_XEN_PVH */
1505
1506/* First C function to be called on Xen boot */
1507asmlinkage __visible void __init xen_start_kernel(void)
1508{
1509 struct physdev_set_iopl set_iopl;
1510 unsigned long initrd_start = 0;
1511 u64 pat;
1512 int rc;
1513
1514 if (!xen_start_info)
1515 return;
1516
1517 xen_domain_type = XEN_PV_DOMAIN;
1518
1519 xen_setup_features();
1520#ifdef CONFIG_XEN_PVH
1521 xen_pvh_early_guest_init();
1522#endif
1523 xen_setup_machphys_mapping();
1524
1525 /* Install Xen paravirt ops */
1526 pv_info = xen_info;
1527 pv_init_ops = xen_init_ops;
1528 pv_apic_ops = xen_apic_ops;
1529 if (!xen_pvh_domain()) {
1530 pv_cpu_ops = xen_cpu_ops;
1531
1532 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1533 }
1534
1535 if (xen_feature(XENFEAT_auto_translated_physmap))
1536 x86_init.resources.memory_setup = xen_auto_xlated_memory_setup;
1537 else
1538 x86_init.resources.memory_setup = xen_memory_setup;
1539 x86_init.oem.arch_setup = xen_arch_setup;
1540 x86_init.oem.banner = xen_banner;
1541
1542 xen_init_time_ops();
1543
1544 /*
1545 * Set up some pagetable state before starting to set any ptes.
1546 */
1547
1548 xen_init_mmu_ops();
1549
1550 /* Prevent unwanted bits from being set in PTEs. */
1551 __supported_pte_mask &= ~_PAGE_GLOBAL;
1552
1553 /*
1554 * Prevent page tables from being allocated in highmem, even
1555 * if CONFIG_HIGHPTE is enabled.
1556 */
1557 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1558
1559 /* Work out if we support NX */
1560 x86_configure_nx();
1561
1562 /* Get mfn list */
1563 xen_build_dynamic_phys_to_machine();
1564
1565 /*
1566 * Set up kernel GDT and segment registers, mainly so that
1567 * -fstack-protector code can be executed.
1568 */
1569 xen_setup_gdt(0);
1570
1571 xen_init_irq_ops();
1572 xen_init_cpuid_mask();
1573
1574#ifdef CONFIG_X86_LOCAL_APIC
1575 /*
1576 * set up the basic apic ops.
1577 */
1578 xen_init_apic();
1579#endif
1580
1581 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1582 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1583 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1584 }
1585
1586 machine_ops = xen_machine_ops;
1587
1588 /*
1589 * The only reliable way to retain the initial address of the
1590 * percpu gdt_page is to remember it here, so we can go and
1591 * mark it RW later, when the initial percpu area is freed.
1592 */
1593 xen_initial_gdt = &per_cpu(gdt_page, 0);
1594
1595 xen_smp_init();
1596
1597#ifdef CONFIG_ACPI_NUMA
1598 /*
1599 * The pages we from Xen are not related to machine pages, so
1600 * any NUMA information the kernel tries to get from ACPI will
1601 * be meaningless. Prevent it from trying.
1602 */
1603 acpi_numa = -1;
1604#endif
1605 /* Don't do the full vcpu_info placement stuff until we have a
1606 possible map and a non-dummy shared_info. */
1607 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1608
1609 local_irq_disable();
1610 early_boot_irqs_disabled = true;
1611
1612 xen_raw_console_write("mapping kernel into physical memory\n");
1613 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base, xen_start_info->nr_pages);
1614
1615 /*
1616 * Modify the cache mode translation tables to match Xen's PAT
1617 * configuration.
1618 */
1619 rdmsrl(MSR_IA32_CR_PAT, pat);
1620 pat_init_cache_modes(pat);
1621
1622 /* keep using Xen gdt for now; no urgent need to change it */
1623
1624#ifdef CONFIG_X86_32
1625 pv_info.kernel_rpl = 1;
1626 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1627 pv_info.kernel_rpl = 0;
1628#else
1629 pv_info.kernel_rpl = 0;
1630#endif
1631 /* set the limit of our address space */
1632 xen_reserve_top();
1633
1634 /* PVH: runs at default kernel iopl of 0 */
1635 if (!xen_pvh_domain()) {
1636 /*
1637 * We used to do this in xen_arch_setup, but that is too late
1638 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1639 * early_amd_init which pokes 0xcf8 port.
1640 */
1641 set_iopl.iopl = 1;
1642 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1643 if (rc != 0)
1644 xen_raw_printk("physdev_op failed %d\n", rc);
1645 }
1646
1647#ifdef CONFIG_X86_32
1648 /* set up basic CPUID stuff */
1649 cpu_detect(&new_cpu_data);
1650 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1651 new_cpu_data.wp_works_ok = 1;
1652 new_cpu_data.x86_capability[0] = cpuid_edx(1);
1653#endif
1654
1655 if (xen_start_info->mod_start) {
1656 if (xen_start_info->flags & SIF_MOD_START_PFN)
1657 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1658 else
1659 initrd_start = __pa(xen_start_info->mod_start);
1660 }
1661
1662 /* Poke various useful things into boot_params */
1663 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1664 boot_params.hdr.ramdisk_image = initrd_start;
1665 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1666 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1667
1668 if (!xen_initial_domain()) {
1669 add_preferred_console("xenboot", 0, NULL);
1670 add_preferred_console("tty", 0, NULL);
1671 add_preferred_console("hvc", 0, NULL);
1672 if (pci_xen)
1673 x86_init.pci.arch_init = pci_xen_init;
1674 } else {
1675 const struct dom0_vga_console_info *info =
1676 (void *)((char *)xen_start_info +
1677 xen_start_info->console.dom0.info_off);
1678 struct xen_platform_op op = {
1679 .cmd = XENPF_firmware_info,
1680 .interface_version = XENPF_INTERFACE_VERSION,
1681 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1682 };
1683
1684 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1685 xen_start_info->console.domU.mfn = 0;
1686 xen_start_info->console.domU.evtchn = 0;
1687
1688 if (HYPERVISOR_dom0_op(&op) == 0)
1689 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1690
1691 /* Make sure ACS will be enabled */
1692 pci_request_acs();
1693
1694 xen_acpi_sleep_register();
1695
1696 /* Avoid searching for BIOS MP tables */
1697 x86_init.mpparse.find_smp_config = x86_init_noop;
1698 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1699
1700 xen_boot_params_init_edd();
1701 }
1702#ifdef CONFIG_PCI
1703 /* PCI BIOS service won't work from a PV guest. */
1704 pci_probe &= ~PCI_PROBE_BIOS;
1705#endif
1706 xen_raw_console_write("about to get started...\n");
1707
1708 xen_setup_runstate_info(0);
1709
1710 xen_efi_init();
1711
1712 /* Start the world */
1713#ifdef CONFIG_X86_32
1714 i386_start_kernel();
1715#else
1716 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1717 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1718#endif
1719}
1720
1721void __ref xen_hvm_init_shared_info(void)
1722{
1723 int cpu;
1724 struct xen_add_to_physmap xatp;
1725 static struct shared_info *shared_info_page = 0;
1726
1727 if (!shared_info_page)
1728 shared_info_page = (struct shared_info *)
1729 extend_brk(PAGE_SIZE, PAGE_SIZE);
1730 xatp.domid = DOMID_SELF;
1731 xatp.idx = 0;
1732 xatp.space = XENMAPSPACE_shared_info;
1733 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1734 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1735 BUG();
1736
1737 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1738
1739 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1740 * page, we use it in the event channel upcall and in some pvclock
1741 * related functions. We don't need the vcpu_info placement
1742 * optimizations because we don't use any pv_mmu or pv_irq op on
1743 * HVM.
1744 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1745 * online but xen_hvm_init_shared_info is run at resume time too and
1746 * in that case multiple vcpus might be online. */
1747 for_each_online_cpu(cpu) {
1748 /* Leave it to be NULL. */
1749 if (cpu >= MAX_VIRT_CPUS)
1750 continue;
1751 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1752 }
1753}
1754
1755#ifdef CONFIG_XEN_PVHVM
1756static void __init init_hvm_pv_info(void)
1757{
1758 int major, minor;
1759 uint32_t eax, ebx, ecx, edx, pages, msr, base;
1760 u64 pfn;
1761
1762 base = xen_cpuid_base();
1763 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1764
1765 major = eax >> 16;
1766 minor = eax & 0xffff;
1767 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1768
1769 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1770
1771 pfn = __pa(hypercall_page);
1772 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1773
1774 xen_setup_features();
1775
1776 pv_info.name = "Xen HVM";
1777
1778 xen_domain_type = XEN_HVM_DOMAIN;
1779}
1780
1781static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
1782 void *hcpu)
1783{
1784 int cpu = (long)hcpu;
1785 switch (action) {
1786 case CPU_UP_PREPARE:
1787 xen_vcpu_setup(cpu);
1788 if (xen_have_vector_callback) {
1789 if (xen_feature(XENFEAT_hvm_safe_pvclock))
1790 xen_setup_timer(cpu);
1791 }
1792 break;
1793 default:
1794 break;
1795 }
1796 return NOTIFY_OK;
1797}
1798
1799static struct notifier_block xen_hvm_cpu_notifier = {
1800 .notifier_call = xen_hvm_cpu_notify,
1801};
1802
1803static void __init xen_hvm_guest_init(void)
1804{
1805 if (xen_pv_domain())
1806 return;
1807
1808 init_hvm_pv_info();
1809
1810 xen_hvm_init_shared_info();
1811
1812 xen_panic_handler_init();
1813
1814 if (xen_feature(XENFEAT_hvm_callback_vector))
1815 xen_have_vector_callback = 1;
1816 xen_hvm_smp_init();
1817 register_cpu_notifier(&xen_hvm_cpu_notifier);
1818 xen_unplug_emulated_devices();
1819 x86_init.irqs.intr_init = xen_init_IRQ;
1820 xen_hvm_init_time_ops();
1821 xen_hvm_init_mmu_ops();
1822}
1823#endif
1824
1825static bool xen_nopv = false;
1826static __init int xen_parse_nopv(char *arg)
1827{
1828 xen_nopv = true;
1829 return 0;
1830}
1831early_param("xen_nopv", xen_parse_nopv);
1832
1833static uint32_t __init xen_platform(void)
1834{
1835 if (xen_nopv)
1836 return 0;
1837
1838 return xen_cpuid_base();
1839}
1840
1841bool xen_hvm_need_lapic(void)
1842{
1843 if (xen_nopv)
1844 return false;
1845 if (xen_pv_domain())
1846 return false;
1847 if (!xen_hvm_domain())
1848 return false;
1849 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1850 return false;
1851 return true;
1852}
1853EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1854
1855static void xen_set_cpu_features(struct cpuinfo_x86 *c)
1856{
1857 if (xen_pv_domain())
1858 clear_cpu_bug(c, X86_BUG_SYSRET_SS_ATTRS);
1859}
1860
1861const struct hypervisor_x86 x86_hyper_xen = {
1862 .name = "Xen",
1863 .detect = xen_platform,
1864#ifdef CONFIG_XEN_PVHVM
1865 .init_platform = xen_hvm_guest_init,
1866#endif
1867 .x2apic_available = xen_x2apic_para_available,
1868 .set_cpu_features = xen_set_cpu_features,
1869};
1870EXPORT_SYMBOL(x86_hyper_xen);
1871