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/export.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#include <linux/kexec.h>
38
39#include <xen/xen.h>
40#include <xen/events.h>
41#include <xen/interface/xen.h>
42#include <xen/interface/version.h>
43#include <xen/interface/physdev.h>
44#include <xen/interface/vcpu.h>
45#include <xen/interface/memory.h>
46#include <xen/interface/nmi.h>
47#include <xen/interface/xen-mca.h>
48#include <xen/interface/hvm/start_info.h>
49#include <xen/features.h>
50#include <xen/page.h>
51#include <xen/hvm.h>
52#include <xen/hvc-console.h>
53#include <xen/acpi.h>
54
55#include <asm/paravirt.h>
56#include <asm/apic.h>
57#include <asm/page.h>
58#include <asm/xen/pci.h>
59#include <asm/xen/hypercall.h>
60#include <asm/xen/hypervisor.h>
61#include <asm/xen/cpuid.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/cpu.h>
79
80#ifdef CONFIG_ACPI
81#include <linux/acpi.h>
82#include <asm/acpi.h>
83#include <acpi/pdc_intel.h>
84#include <acpi/processor.h>
85#include <xen/interface/platform.h>
86#endif
87
88#include "xen-ops.h"
89#include "mmu.h"
90#include "smp.h"
91#include "multicalls.h"
92#include "pmu.h"
93
94EXPORT_SYMBOL_GPL(hypercall_page);
95
96/*
97 * Pointer to the xen_vcpu_info structure or
98 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
99 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
100 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
101 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
102 * acknowledge pending events.
103 * Also more subtly it is used by the patched version of irq enable/disable
104 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
105 *
106 * The desire to be able to do those mask/unmask operations as a single
107 * instruction by using the per-cpu offset held in %gs is the real reason
108 * vcpu info is in a per-cpu pointer and the original reason for this
109 * hypercall.
110 *
111 */
112DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
113
114/*
115 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
116 * hypercall. This can be used both in PV and PVHVM mode. The structure
117 * overrides the default per_cpu(xen_vcpu, cpu) value.
118 */
119DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
120
121/* Linux <-> Xen vCPU id mapping */
122DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
123EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
124
125enum xen_domain_type xen_domain_type = XEN_NATIVE;
126EXPORT_SYMBOL_GPL(xen_domain_type);
127
128unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
129EXPORT_SYMBOL(machine_to_phys_mapping);
130unsigned long machine_to_phys_nr;
131EXPORT_SYMBOL(machine_to_phys_nr);
132
133struct start_info *xen_start_info;
134EXPORT_SYMBOL_GPL(xen_start_info);
135
136struct shared_info xen_dummy_shared_info;
137
138void *xen_initial_gdt;
139
140RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
141
142static int xen_cpu_up_prepare(unsigned int cpu);
143static int xen_cpu_up_online(unsigned int cpu);
144static int xen_cpu_dead(unsigned int cpu);
145
146/*
147 * Point at some empty memory to start with. We map the real shared_info
148 * page as soon as fixmap is up and running.
149 */
150struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
151
152/*
153 * Flag to determine whether vcpu info placement is available on all
154 * VCPUs. We assume it is to start with, and then set it to zero on
155 * the first failure. This is because it can succeed on some VCPUs
156 * and not others, since it can involve hypervisor memory allocation,
157 * or because the guest failed to guarantee all the appropriate
158 * constraints on all VCPUs (ie buffer can't cross a page boundary).
159 *
160 * Note that any particular CPU may be using a placed vcpu structure,
161 * but we can only optimise if the all are.
162 *
163 * 0: not available, 1: available
164 */
165static int have_vcpu_info_placement = 1;
166
167struct tls_descs {
168 struct desc_struct desc[3];
169};
170
171/*
172 * Updating the 3 TLS descriptors in the GDT on every task switch is
173 * surprisingly expensive so we avoid updating them if they haven't
174 * changed. Since Xen writes different descriptors than the one
175 * passed in the update_descriptor hypercall we keep shadow copies to
176 * compare against.
177 */
178static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
179
180#ifdef CONFIG_XEN_PVH
181/*
182 * PVH variables.
183 *
184 * xen_pvh and pvh_bootparams need to live in data segment since they
185 * are used after startup_{32|64}, which clear .bss, are invoked.
186 */
187bool xen_pvh __attribute__((section(".data"))) = 0;
188struct boot_params pvh_bootparams __attribute__((section(".data")));
189
190struct hvm_start_info pvh_start_info;
191unsigned int pvh_start_info_sz = sizeof(pvh_start_info);
192#endif
193
194static void clamp_max_cpus(void)
195{
196#ifdef CONFIG_SMP
197 if (setup_max_cpus > MAX_VIRT_CPUS)
198 setup_max_cpus = MAX_VIRT_CPUS;
199#endif
200}
201
202void xen_vcpu_setup(int cpu)
203{
204 struct vcpu_register_vcpu_info info;
205 int err;
206 struct vcpu_info *vcpup;
207
208 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
209
210 /*
211 * This path is called twice on PVHVM - first during bootup via
212 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
213 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
214 * As we can only do the VCPUOP_register_vcpu_info once lets
215 * not over-write its result.
216 *
217 * For PV it is called during restore (xen_vcpu_restore) and bootup
218 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
219 * use this function.
220 */
221 if (xen_hvm_domain()) {
222 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
223 return;
224 }
225 if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS)
226 per_cpu(xen_vcpu, cpu) =
227 &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
228
229 if (!have_vcpu_info_placement) {
230 if (cpu >= MAX_VIRT_CPUS)
231 clamp_max_cpus();
232 return;
233 }
234
235 vcpup = &per_cpu(xen_vcpu_info, cpu);
236 info.mfn = arbitrary_virt_to_mfn(vcpup);
237 info.offset = offset_in_page(vcpup);
238
239 /* Check to see if the hypervisor will put the vcpu_info
240 structure where we want it, which allows direct access via
241 a percpu-variable.
242 N.B. This hypercall can _only_ be called once per CPU. Subsequent
243 calls will error out with -EINVAL. This is due to the fact that
244 hypervisor has no unregister variant and this hypercall does not
245 allow to over-write info.mfn and info.offset.
246 */
247 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
248 &info);
249
250 if (err) {
251 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
252 have_vcpu_info_placement = 0;
253 clamp_max_cpus();
254 } else {
255 /* This cpu is using the registered vcpu info, even if
256 later ones fail to. */
257 per_cpu(xen_vcpu, cpu) = vcpup;
258 }
259}
260
261/*
262 * On restore, set the vcpu placement up again.
263 * If it fails, then we're in a bad state, since
264 * we can't back out from using it...
265 */
266void xen_vcpu_restore(void)
267{
268 int cpu;
269
270 for_each_possible_cpu(cpu) {
271 bool other_cpu = (cpu != smp_processor_id());
272 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu),
273 NULL);
274
275 if (other_cpu && is_up &&
276 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
277 BUG();
278
279 xen_setup_runstate_info(cpu);
280
281 if (have_vcpu_info_placement)
282 xen_vcpu_setup(cpu);
283
284 if (other_cpu && is_up &&
285 HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
286 BUG();
287 }
288}
289
290static void __init xen_banner(void)
291{
292 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
293 struct xen_extraversion extra;
294 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
295
296 pr_info("Booting paravirtualized kernel %son %s\n",
297 xen_feature(XENFEAT_auto_translated_physmap) ?
298 "with PVH extensions " : "", pv_info.name);
299 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
300 version >> 16, version & 0xffff, extra.extraversion,
301 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
302}
303/* Check if running on Xen version (major, minor) or later */
304bool
305xen_running_on_version_or_later(unsigned int major, unsigned int minor)
306{
307 unsigned int version;
308
309 if (!xen_domain())
310 return false;
311
312 version = HYPERVISOR_xen_version(XENVER_version, NULL);
313 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
314 ((version >> 16) > major))
315 return true;
316 return false;
317}
318
319#define CPUID_THERM_POWER_LEAF 6
320#define APERFMPERF_PRESENT 0
321
322static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
323static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
324
325static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
326static __read_mostly unsigned int cpuid_leaf5_ecx_val;
327static __read_mostly unsigned int cpuid_leaf5_edx_val;
328
329static void xen_cpuid(unsigned int *ax, unsigned int *bx,
330 unsigned int *cx, unsigned int *dx)
331{
332 unsigned maskebx = ~0;
333 unsigned maskecx = ~0;
334 unsigned maskedx = ~0;
335 unsigned setecx = 0;
336 /*
337 * Mask out inconvenient features, to try and disable as many
338 * unsupported kernel subsystems as possible.
339 */
340 switch (*ax) {
341 case 1:
342 maskecx = cpuid_leaf1_ecx_mask;
343 setecx = cpuid_leaf1_ecx_set_mask;
344 maskedx = cpuid_leaf1_edx_mask;
345 break;
346
347 case CPUID_MWAIT_LEAF:
348 /* Synthesize the values.. */
349 *ax = 0;
350 *bx = 0;
351 *cx = cpuid_leaf5_ecx_val;
352 *dx = cpuid_leaf5_edx_val;
353 return;
354
355 case CPUID_THERM_POWER_LEAF:
356 /* Disabling APERFMPERF for kernel usage */
357 maskecx = ~(1 << APERFMPERF_PRESENT);
358 break;
359
360 case 0xb:
361 /* Suppress extended topology stuff */
362 maskebx = 0;
363 break;
364 }
365
366 asm(XEN_EMULATE_PREFIX "cpuid"
367 : "=a" (*ax),
368 "=b" (*bx),
369 "=c" (*cx),
370 "=d" (*dx)
371 : "0" (*ax), "2" (*cx));
372
373 *bx &= maskebx;
374 *cx &= maskecx;
375 *cx |= setecx;
376 *dx &= maskedx;
377}
378STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
379
380static bool __init xen_check_mwait(void)
381{
382#ifdef CONFIG_ACPI
383 struct xen_platform_op op = {
384 .cmd = XENPF_set_processor_pminfo,
385 .u.set_pminfo.id = -1,
386 .u.set_pminfo.type = XEN_PM_PDC,
387 };
388 uint32_t buf[3];
389 unsigned int ax, bx, cx, dx;
390 unsigned int mwait_mask;
391
392 /* We need to determine whether it is OK to expose the MWAIT
393 * capability to the kernel to harvest deeper than C3 states from ACPI
394 * _CST using the processor_harvest_xen.c module. For this to work, we
395 * need to gather the MWAIT_LEAF values (which the cstate.c code
396 * checks against). The hypervisor won't expose the MWAIT flag because
397 * it would break backwards compatibility; so we will find out directly
398 * from the hardware and hypercall.
399 */
400 if (!xen_initial_domain())
401 return false;
402
403 /*
404 * When running under platform earlier than Xen4.2, do not expose
405 * mwait, to avoid the risk of loading native acpi pad driver
406 */
407 if (!xen_running_on_version_or_later(4, 2))
408 return false;
409
410 ax = 1;
411 cx = 0;
412
413 native_cpuid(&ax, &bx, &cx, &dx);
414
415 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
416 (1 << (X86_FEATURE_MWAIT % 32));
417
418 if ((cx & mwait_mask) != mwait_mask)
419 return false;
420
421 /* We need to emulate the MWAIT_LEAF and for that we need both
422 * ecx and edx. The hypercall provides only partial information.
423 */
424
425 ax = CPUID_MWAIT_LEAF;
426 bx = 0;
427 cx = 0;
428 dx = 0;
429
430 native_cpuid(&ax, &bx, &cx, &dx);
431
432 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
433 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
434 */
435 buf[0] = ACPI_PDC_REVISION_ID;
436 buf[1] = 1;
437 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
438
439 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
440
441 if ((HYPERVISOR_platform_op(&op) == 0) &&
442 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
443 cpuid_leaf5_ecx_val = cx;
444 cpuid_leaf5_edx_val = dx;
445 }
446 return true;
447#else
448 return false;
449#endif
450}
451static void __init xen_init_cpuid_mask(void)
452{
453 unsigned int ax, bx, cx, dx;
454 unsigned int xsave_mask;
455
456 cpuid_leaf1_edx_mask =
457 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
458 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
459
460 if (!xen_initial_domain())
461 cpuid_leaf1_edx_mask &=
462 ~((1 << X86_FEATURE_ACPI)); /* disable ACPI */
463
464 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
465
466 ax = 1;
467 cx = 0;
468 cpuid(1, &ax, &bx, &cx, &dx);
469
470 xsave_mask =
471 (1 << (X86_FEATURE_XSAVE % 32)) |
472 (1 << (X86_FEATURE_OSXSAVE % 32));
473
474 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
475 if ((cx & xsave_mask) != xsave_mask)
476 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
477 if (xen_check_mwait())
478 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
479}
480
481static void xen_set_debugreg(int reg, unsigned long val)
482{
483 HYPERVISOR_set_debugreg(reg, val);
484}
485
486static unsigned long xen_get_debugreg(int reg)
487{
488 return HYPERVISOR_get_debugreg(reg);
489}
490
491static void xen_end_context_switch(struct task_struct *next)
492{
493 xen_mc_flush();
494 paravirt_end_context_switch(next);
495}
496
497static unsigned long xen_store_tr(void)
498{
499 return 0;
500}
501
502/*
503 * Set the page permissions for a particular virtual address. If the
504 * address is a vmalloc mapping (or other non-linear mapping), then
505 * find the linear mapping of the page and also set its protections to
506 * match.
507 */
508static void set_aliased_prot(void *v, pgprot_t prot)
509{
510 int level;
511 pte_t *ptep;
512 pte_t pte;
513 unsigned long pfn;
514 struct page *page;
515 unsigned char dummy;
516
517 ptep = lookup_address((unsigned long)v, &level);
518 BUG_ON(ptep == NULL);
519
520 pfn = pte_pfn(*ptep);
521 page = pfn_to_page(pfn);
522
523 pte = pfn_pte(pfn, prot);
524
525 /*
526 * Careful: update_va_mapping() will fail if the virtual address
527 * we're poking isn't populated in the page tables. We don't
528 * need to worry about the direct map (that's always in the page
529 * tables), but we need to be careful about vmap space. In
530 * particular, the top level page table can lazily propagate
531 * entries between processes, so if we've switched mms since we
532 * vmapped the target in the first place, we might not have the
533 * top-level page table entry populated.
534 *
535 * We disable preemption because we want the same mm active when
536 * we probe the target and when we issue the hypercall. We'll
537 * have the same nominal mm, but if we're a kernel thread, lazy
538 * mm dropping could change our pgd.
539 *
540 * Out of an abundance of caution, this uses __get_user() to fault
541 * in the target address just in case there's some obscure case
542 * in which the target address isn't readable.
543 */
544
545 preempt_disable();
546
547 probe_kernel_read(&dummy, v, 1);
548
549 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
550 BUG();
551
552 if (!PageHighMem(page)) {
553 void *av = __va(PFN_PHYS(pfn));
554
555 if (av != v)
556 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
557 BUG();
558 } else
559 kmap_flush_unused();
560
561 preempt_enable();
562}
563
564static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
565{
566 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
567 int i;
568
569 /*
570 * We need to mark the all aliases of the LDT pages RO. We
571 * don't need to call vm_flush_aliases(), though, since that's
572 * only responsible for flushing aliases out the TLBs, not the
573 * page tables, and Xen will flush the TLB for us if needed.
574 *
575 * To avoid confusing future readers: none of this is necessary
576 * to load the LDT. The hypervisor only checks this when the
577 * LDT is faulted in due to subsequent descriptor access.
578 */
579
580 for(i = 0; i < entries; i += entries_per_page)
581 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
582}
583
584static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
585{
586 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
587 int i;
588
589 for(i = 0; i < entries; i += entries_per_page)
590 set_aliased_prot(ldt + i, PAGE_KERNEL);
591}
592
593static void xen_set_ldt(const void *addr, unsigned entries)
594{
595 struct mmuext_op *op;
596 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
597
598 trace_xen_cpu_set_ldt(addr, entries);
599
600 op = mcs.args;
601 op->cmd = MMUEXT_SET_LDT;
602 op->arg1.linear_addr = (unsigned long)addr;
603 op->arg2.nr_ents = entries;
604
605 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
606
607 xen_mc_issue(PARAVIRT_LAZY_CPU);
608}
609
610static void xen_load_gdt(const struct desc_ptr *dtr)
611{
612 unsigned long va = dtr->address;
613 unsigned int size = dtr->size + 1;
614 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
615 unsigned long frames[pages];
616 int f;
617
618 /*
619 * A GDT can be up to 64k in size, which corresponds to 8192
620 * 8-byte entries, or 16 4k pages..
621 */
622
623 BUG_ON(size > 65536);
624 BUG_ON(va & ~PAGE_MASK);
625
626 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
627 int level;
628 pte_t *ptep;
629 unsigned long pfn, mfn;
630 void *virt;
631
632 /*
633 * The GDT is per-cpu and is in the percpu data area.
634 * That can be virtually mapped, so we need to do a
635 * page-walk to get the underlying MFN for the
636 * hypercall. The page can also be in the kernel's
637 * linear range, so we need to RO that mapping too.
638 */
639 ptep = lookup_address(va, &level);
640 BUG_ON(ptep == NULL);
641
642 pfn = pte_pfn(*ptep);
643 mfn = pfn_to_mfn(pfn);
644 virt = __va(PFN_PHYS(pfn));
645
646 frames[f] = mfn;
647
648 make_lowmem_page_readonly((void *)va);
649 make_lowmem_page_readonly(virt);
650 }
651
652 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
653 BUG();
654}
655
656/*
657 * load_gdt for early boot, when the gdt is only mapped once
658 */
659static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
660{
661 unsigned long va = dtr->address;
662 unsigned int size = dtr->size + 1;
663 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
664 unsigned long frames[pages];
665 int f;
666
667 /*
668 * A GDT can be up to 64k in size, which corresponds to 8192
669 * 8-byte entries, or 16 4k pages..
670 */
671
672 BUG_ON(size > 65536);
673 BUG_ON(va & ~PAGE_MASK);
674
675 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
676 pte_t pte;
677 unsigned long pfn, mfn;
678
679 pfn = virt_to_pfn(va);
680 mfn = pfn_to_mfn(pfn);
681
682 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
683
684 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
685 BUG();
686
687 frames[f] = mfn;
688 }
689
690 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
691 BUG();
692}
693
694static inline bool desc_equal(const struct desc_struct *d1,
695 const struct desc_struct *d2)
696{
697 return d1->a == d2->a && d1->b == d2->b;
698}
699
700static void load_TLS_descriptor(struct thread_struct *t,
701 unsigned int cpu, unsigned int i)
702{
703 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
704 struct desc_struct *gdt;
705 xmaddr_t maddr;
706 struct multicall_space mc;
707
708 if (desc_equal(shadow, &t->tls_array[i]))
709 return;
710
711 *shadow = t->tls_array[i];
712
713 gdt = get_cpu_gdt_table(cpu);
714 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
715 mc = __xen_mc_entry(0);
716
717 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
718}
719
720static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
721{
722 /*
723 * XXX sleazy hack: If we're being called in a lazy-cpu zone
724 * and lazy gs handling is enabled, it means we're in a
725 * context switch, and %gs has just been saved. This means we
726 * can zero it out to prevent faults on exit from the
727 * hypervisor if the next process has no %gs. Either way, it
728 * has been saved, and the new value will get loaded properly.
729 * This will go away as soon as Xen has been modified to not
730 * save/restore %gs for normal hypercalls.
731 *
732 * On x86_64, this hack is not used for %gs, because gs points
733 * to KERNEL_GS_BASE (and uses it for PDA references), so we
734 * must not zero %gs on x86_64
735 *
736 * For x86_64, we need to zero %fs, otherwise we may get an
737 * exception between the new %fs descriptor being loaded and
738 * %fs being effectively cleared at __switch_to().
739 */
740 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
741#ifdef CONFIG_X86_32
742 lazy_load_gs(0);
743#else
744 loadsegment(fs, 0);
745#endif
746 }
747
748 xen_mc_batch();
749
750 load_TLS_descriptor(t, cpu, 0);
751 load_TLS_descriptor(t, cpu, 1);
752 load_TLS_descriptor(t, cpu, 2);
753
754 xen_mc_issue(PARAVIRT_LAZY_CPU);
755}
756
757#ifdef CONFIG_X86_64
758static void xen_load_gs_index(unsigned int idx)
759{
760 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
761 BUG();
762}
763#endif
764
765static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
766 const void *ptr)
767{
768 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
769 u64 entry = *(u64 *)ptr;
770
771 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
772
773 preempt_disable();
774
775 xen_mc_flush();
776 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
777 BUG();
778
779 preempt_enable();
780}
781
782static int cvt_gate_to_trap(int vector, const gate_desc *val,
783 struct trap_info *info)
784{
785 unsigned long addr;
786
787 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
788 return 0;
789
790 info->vector = vector;
791
792 addr = gate_offset(*val);
793#ifdef CONFIG_X86_64
794 /*
795 * Look for known traps using IST, and substitute them
796 * appropriately. The debugger ones are the only ones we care
797 * about. Xen will handle faults like double_fault,
798 * so we should never see them. Warn if
799 * there's an unexpected IST-using fault handler.
800 */
801 if (addr == (unsigned long)debug)
802 addr = (unsigned long)xen_debug;
803 else if (addr == (unsigned long)int3)
804 addr = (unsigned long)xen_int3;
805 else if (addr == (unsigned long)stack_segment)
806 addr = (unsigned long)xen_stack_segment;
807 else if (addr == (unsigned long)double_fault) {
808 /* Don't need to handle these */
809 return 0;
810#ifdef CONFIG_X86_MCE
811 } else if (addr == (unsigned long)machine_check) {
812 /*
813 * when xen hypervisor inject vMCE to guest,
814 * use native mce handler to handle it
815 */
816 ;
817#endif
818 } else if (addr == (unsigned long)nmi)
819 /*
820 * Use the native version as well.
821 */
822 ;
823 else {
824 /* Some other trap using IST? */
825 if (WARN_ON(val->ist != 0))
826 return 0;
827 }
828#endif /* CONFIG_X86_64 */
829 info->address = addr;
830
831 info->cs = gate_segment(*val);
832 info->flags = val->dpl;
833 /* interrupt gates clear IF */
834 if (val->type == GATE_INTERRUPT)
835 info->flags |= 1 << 2;
836
837 return 1;
838}
839
840/* Locations of each CPU's IDT */
841static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
842
843/* Set an IDT entry. If the entry is part of the current IDT, then
844 also update Xen. */
845static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
846{
847 unsigned long p = (unsigned long)&dt[entrynum];
848 unsigned long start, end;
849
850 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
851
852 preempt_disable();
853
854 start = __this_cpu_read(idt_desc.address);
855 end = start + __this_cpu_read(idt_desc.size) + 1;
856
857 xen_mc_flush();
858
859 native_write_idt_entry(dt, entrynum, g);
860
861 if (p >= start && (p + 8) <= end) {
862 struct trap_info info[2];
863
864 info[1].address = 0;
865
866 if (cvt_gate_to_trap(entrynum, g, &info[0]))
867 if (HYPERVISOR_set_trap_table(info))
868 BUG();
869 }
870
871 preempt_enable();
872}
873
874static void xen_convert_trap_info(const struct desc_ptr *desc,
875 struct trap_info *traps)
876{
877 unsigned in, out, count;
878
879 count = (desc->size+1) / sizeof(gate_desc);
880 BUG_ON(count > 256);
881
882 for (in = out = 0; in < count; in++) {
883 gate_desc *entry = (gate_desc*)(desc->address) + in;
884
885 if (cvt_gate_to_trap(in, entry, &traps[out]))
886 out++;
887 }
888 traps[out].address = 0;
889}
890
891void xen_copy_trap_info(struct trap_info *traps)
892{
893 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
894
895 xen_convert_trap_info(desc, traps);
896}
897
898/* Load a new IDT into Xen. In principle this can be per-CPU, so we
899 hold a spinlock to protect the static traps[] array (static because
900 it avoids allocation, and saves stack space). */
901static void xen_load_idt(const struct desc_ptr *desc)
902{
903 static DEFINE_SPINLOCK(lock);
904 static struct trap_info traps[257];
905
906 trace_xen_cpu_load_idt(desc);
907
908 spin_lock(&lock);
909
910 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
911
912 xen_convert_trap_info(desc, traps);
913
914 xen_mc_flush();
915 if (HYPERVISOR_set_trap_table(traps))
916 BUG();
917
918 spin_unlock(&lock);
919}
920
921/* Write a GDT descriptor entry. Ignore LDT descriptors, since
922 they're handled differently. */
923static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
924 const void *desc, int type)
925{
926 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
927
928 preempt_disable();
929
930 switch (type) {
931 case DESC_LDT:
932 case DESC_TSS:
933 /* ignore */
934 break;
935
936 default: {
937 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
938
939 xen_mc_flush();
940 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
941 BUG();
942 }
943
944 }
945
946 preempt_enable();
947}
948
949/*
950 * Version of write_gdt_entry for use at early boot-time needed to
951 * update an entry as simply as possible.
952 */
953static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
954 const void *desc, int type)
955{
956 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
957
958 switch (type) {
959 case DESC_LDT:
960 case DESC_TSS:
961 /* ignore */
962 break;
963
964 default: {
965 xmaddr_t maddr = virt_to_machine(&dt[entry]);
966
967 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
968 dt[entry] = *(struct desc_struct *)desc;
969 }
970
971 }
972}
973
974static void xen_load_sp0(struct tss_struct *tss,
975 struct thread_struct *thread)
976{
977 struct multicall_space mcs;
978
979 mcs = xen_mc_entry(0);
980 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
981 xen_mc_issue(PARAVIRT_LAZY_CPU);
982 tss->x86_tss.sp0 = thread->sp0;
983}
984
985void xen_set_iopl_mask(unsigned mask)
986{
987 struct physdev_set_iopl set_iopl;
988
989 /* Force the change at ring 0. */
990 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
991 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
992}
993
994static void xen_io_delay(void)
995{
996}
997
998static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
999
1000static unsigned long xen_read_cr0(void)
1001{
1002 unsigned long cr0 = this_cpu_read(xen_cr0_value);
1003
1004 if (unlikely(cr0 == 0)) {
1005 cr0 = native_read_cr0();
1006 this_cpu_write(xen_cr0_value, cr0);
1007 }
1008
1009 return cr0;
1010}
1011
1012static void xen_write_cr0(unsigned long cr0)
1013{
1014 struct multicall_space mcs;
1015
1016 this_cpu_write(xen_cr0_value, cr0);
1017
1018 /* Only pay attention to cr0.TS; everything else is
1019 ignored. */
1020 mcs = xen_mc_entry(0);
1021
1022 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1023
1024 xen_mc_issue(PARAVIRT_LAZY_CPU);
1025}
1026
1027static void xen_write_cr4(unsigned long cr4)
1028{
1029 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
1030
1031 native_write_cr4(cr4);
1032}
1033#ifdef CONFIG_X86_64
1034static inline unsigned long xen_read_cr8(void)
1035{
1036 return 0;
1037}
1038static inline void xen_write_cr8(unsigned long val)
1039{
1040 BUG_ON(val);
1041}
1042#endif
1043
1044static u64 xen_read_msr_safe(unsigned int msr, int *err)
1045{
1046 u64 val;
1047
1048 if (pmu_msr_read(msr, &val, err))
1049 return val;
1050
1051 val = native_read_msr_safe(msr, err);
1052 switch (msr) {
1053 case MSR_IA32_APICBASE:
1054#ifdef CONFIG_X86_X2APIC
1055 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1056#endif
1057 val &= ~X2APIC_ENABLE;
1058 break;
1059 }
1060 return val;
1061}
1062
1063static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1064{
1065 int ret;
1066
1067 ret = 0;
1068
1069 switch (msr) {
1070#ifdef CONFIG_X86_64
1071 unsigned which;
1072 u64 base;
1073
1074 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
1075 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
1076 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
1077
1078 set:
1079 base = ((u64)high << 32) | low;
1080 if (HYPERVISOR_set_segment_base(which, base) != 0)
1081 ret = -EIO;
1082 break;
1083#endif
1084
1085 case MSR_STAR:
1086 case MSR_CSTAR:
1087 case MSR_LSTAR:
1088 case MSR_SYSCALL_MASK:
1089 case MSR_IA32_SYSENTER_CS:
1090 case MSR_IA32_SYSENTER_ESP:
1091 case MSR_IA32_SYSENTER_EIP:
1092 /* Fast syscall setup is all done in hypercalls, so
1093 these are all ignored. Stub them out here to stop
1094 Xen console noise. */
1095 break;
1096
1097 default:
1098 if (!pmu_msr_write(msr, low, high, &ret))
1099 ret = native_write_msr_safe(msr, low, high);
1100 }
1101
1102 return ret;
1103}
1104
1105static u64 xen_read_msr(unsigned int msr)
1106{
1107 /*
1108 * This will silently swallow a #GP from RDMSR. It may be worth
1109 * changing that.
1110 */
1111 int err;
1112
1113 return xen_read_msr_safe(msr, &err);
1114}
1115
1116static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1117{
1118 /*
1119 * This will silently swallow a #GP from WRMSR. It may be worth
1120 * changing that.
1121 */
1122 xen_write_msr_safe(msr, low, high);
1123}
1124
1125void xen_setup_shared_info(void)
1126{
1127 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1128 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1129 xen_start_info->shared_info);
1130
1131 HYPERVISOR_shared_info =
1132 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
1133 } else
1134 HYPERVISOR_shared_info =
1135 (struct shared_info *)__va(xen_start_info->shared_info);
1136
1137#ifndef CONFIG_SMP
1138 /* In UP this is as good a place as any to set up shared info */
1139 xen_setup_vcpu_info_placement();
1140#endif
1141
1142 xen_setup_mfn_list_list();
1143}
1144
1145/* This is called once we have the cpu_possible_mask */
1146void xen_setup_vcpu_info_placement(void)
1147{
1148 int cpu;
1149
1150 for_each_possible_cpu(cpu) {
1151 /* Set up direct vCPU id mapping for PV guests. */
1152 per_cpu(xen_vcpu_id, cpu) = cpu;
1153 xen_vcpu_setup(cpu);
1154 }
1155
1156 /*
1157 * xen_vcpu_setup managed to place the vcpu_info within the
1158 * percpu area for all cpus, so make use of it.
1159 */
1160 if (have_vcpu_info_placement) {
1161 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1162 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1163 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1164 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
1165 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
1166 }
1167}
1168
1169static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1170 unsigned long addr, unsigned len)
1171{
1172 char *start, *end, *reloc;
1173 unsigned ret;
1174
1175 start = end = reloc = NULL;
1176
1177#define SITE(op, x) \
1178 case PARAVIRT_PATCH(op.x): \
1179 if (have_vcpu_info_placement) { \
1180 start = (char *)xen_##x##_direct; \
1181 end = xen_##x##_direct_end; \
1182 reloc = xen_##x##_direct_reloc; \
1183 } \
1184 goto patch_site
1185
1186 switch (type) {
1187 SITE(pv_irq_ops, irq_enable);
1188 SITE(pv_irq_ops, irq_disable);
1189 SITE(pv_irq_ops, save_fl);
1190 SITE(pv_irq_ops, restore_fl);
1191#undef SITE
1192
1193 patch_site:
1194 if (start == NULL || (end-start) > len)
1195 goto default_patch;
1196
1197 ret = paravirt_patch_insns(insnbuf, len, start, end);
1198
1199 /* Note: because reloc is assigned from something that
1200 appears to be an array, gcc assumes it's non-null,
1201 but doesn't know its relationship with start and
1202 end. */
1203 if (reloc > start && reloc < end) {
1204 int reloc_off = reloc - start;
1205 long *relocp = (long *)(insnbuf + reloc_off);
1206 long delta = start - (char *)addr;
1207
1208 *relocp += delta;
1209 }
1210 break;
1211
1212 default_patch:
1213 default:
1214 ret = paravirt_patch_default(type, clobbers, insnbuf,
1215 addr, len);
1216 break;
1217 }
1218
1219 return ret;
1220}
1221
1222static const struct pv_info xen_info __initconst = {
1223 .shared_kernel_pmd = 0,
1224
1225#ifdef CONFIG_X86_64
1226 .extra_user_64bit_cs = FLAT_USER_CS64,
1227#endif
1228 .name = "Xen",
1229};
1230
1231static const struct pv_init_ops xen_init_ops __initconst = {
1232 .patch = xen_patch,
1233};
1234
1235static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1236 .cpuid = xen_cpuid,
1237
1238 .set_debugreg = xen_set_debugreg,
1239 .get_debugreg = xen_get_debugreg,
1240
1241 .read_cr0 = xen_read_cr0,
1242 .write_cr0 = xen_write_cr0,
1243
1244 .read_cr4 = native_read_cr4,
1245 .write_cr4 = xen_write_cr4,
1246
1247#ifdef CONFIG_X86_64
1248 .read_cr8 = xen_read_cr8,
1249 .write_cr8 = xen_write_cr8,
1250#endif
1251
1252 .wbinvd = native_wbinvd,
1253
1254 .read_msr = xen_read_msr,
1255 .write_msr = xen_write_msr,
1256
1257 .read_msr_safe = xen_read_msr_safe,
1258 .write_msr_safe = xen_write_msr_safe,
1259
1260 .read_pmc = xen_read_pmc,
1261
1262 .iret = xen_iret,
1263#ifdef CONFIG_X86_64
1264 .usergs_sysret64 = xen_sysret64,
1265#endif
1266
1267 .load_tr_desc = paravirt_nop,
1268 .set_ldt = xen_set_ldt,
1269 .load_gdt = xen_load_gdt,
1270 .load_idt = xen_load_idt,
1271 .load_tls = xen_load_tls,
1272#ifdef CONFIG_X86_64
1273 .load_gs_index = xen_load_gs_index,
1274#endif
1275
1276 .alloc_ldt = xen_alloc_ldt,
1277 .free_ldt = xen_free_ldt,
1278
1279 .store_idt = native_store_idt,
1280 .store_tr = xen_store_tr,
1281
1282 .write_ldt_entry = xen_write_ldt_entry,
1283 .write_gdt_entry = xen_write_gdt_entry,
1284 .write_idt_entry = xen_write_idt_entry,
1285 .load_sp0 = xen_load_sp0,
1286
1287 .set_iopl_mask = xen_set_iopl_mask,
1288 .io_delay = xen_io_delay,
1289
1290 /* Xen takes care of %gs when switching to usermode for us */
1291 .swapgs = paravirt_nop,
1292
1293 .start_context_switch = paravirt_start_context_switch,
1294 .end_context_switch = xen_end_context_switch,
1295};
1296
1297static void xen_reboot(int reason)
1298{
1299 struct sched_shutdown r = { .reason = reason };
1300 int cpu;
1301
1302 for_each_online_cpu(cpu)
1303 xen_pmu_finish(cpu);
1304
1305 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1306 BUG();
1307}
1308
1309static void xen_restart(char *msg)
1310{
1311 xen_reboot(SHUTDOWN_reboot);
1312}
1313
1314static void xen_emergency_restart(void)
1315{
1316 xen_reboot(SHUTDOWN_reboot);
1317}
1318
1319static void xen_machine_halt(void)
1320{
1321 xen_reboot(SHUTDOWN_poweroff);
1322}
1323
1324static void xen_machine_power_off(void)
1325{
1326 if (pm_power_off)
1327 pm_power_off();
1328 xen_reboot(SHUTDOWN_poweroff);
1329}
1330
1331static void xen_crash_shutdown(struct pt_regs *regs)
1332{
1333 xen_reboot(SHUTDOWN_crash);
1334}
1335
1336static int
1337xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1338{
1339 if (!kexec_crash_loaded())
1340 xen_reboot(SHUTDOWN_crash);
1341 return NOTIFY_DONE;
1342}
1343
1344static struct notifier_block xen_panic_block = {
1345 .notifier_call= xen_panic_event,
1346 .priority = INT_MIN
1347};
1348
1349int xen_panic_handler_init(void)
1350{
1351 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1352 return 0;
1353}
1354
1355static const struct machine_ops xen_machine_ops __initconst = {
1356 .restart = xen_restart,
1357 .halt = xen_machine_halt,
1358 .power_off = xen_machine_power_off,
1359 .shutdown = xen_machine_halt,
1360 .crash_shutdown = xen_crash_shutdown,
1361 .emergency_restart = xen_emergency_restart,
1362};
1363
1364static unsigned char xen_get_nmi_reason(void)
1365{
1366 unsigned char reason = 0;
1367
1368 /* Construct a value which looks like it came from port 0x61. */
1369 if (test_bit(_XEN_NMIREASON_io_error,
1370 &HYPERVISOR_shared_info->arch.nmi_reason))
1371 reason |= NMI_REASON_IOCHK;
1372 if (test_bit(_XEN_NMIREASON_pci_serr,
1373 &HYPERVISOR_shared_info->arch.nmi_reason))
1374 reason |= NMI_REASON_SERR;
1375
1376 return reason;
1377}
1378
1379static void __init xen_boot_params_init_edd(void)
1380{
1381#if IS_ENABLED(CONFIG_EDD)
1382 struct xen_platform_op op;
1383 struct edd_info *edd_info;
1384 u32 *mbr_signature;
1385 unsigned nr;
1386 int ret;
1387
1388 edd_info = boot_params.eddbuf;
1389 mbr_signature = boot_params.edd_mbr_sig_buffer;
1390
1391 op.cmd = XENPF_firmware_info;
1392
1393 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1394 for (nr = 0; nr < EDDMAXNR; nr++) {
1395 struct edd_info *info = edd_info + nr;
1396
1397 op.u.firmware_info.index = nr;
1398 info->params.length = sizeof(info->params);
1399 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1400 &info->params);
1401 ret = HYPERVISOR_platform_op(&op);
1402 if (ret)
1403 break;
1404
1405#define C(x) info->x = op.u.firmware_info.u.disk_info.x
1406 C(device);
1407 C(version);
1408 C(interface_support);
1409 C(legacy_max_cylinder);
1410 C(legacy_max_head);
1411 C(legacy_sectors_per_track);
1412#undef C
1413 }
1414 boot_params.eddbuf_entries = nr;
1415
1416 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1417 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1418 op.u.firmware_info.index = nr;
1419 ret = HYPERVISOR_platform_op(&op);
1420 if (ret)
1421 break;
1422 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1423 }
1424 boot_params.edd_mbr_sig_buf_entries = nr;
1425#endif
1426}
1427
1428/*
1429 * Set up the GDT and segment registers for -fstack-protector. Until
1430 * we do this, we have to be careful not to call any stack-protected
1431 * function, which is most of the kernel.
1432 */
1433static void xen_setup_gdt(int cpu)
1434{
1435 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1436 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1437
1438 setup_stack_canary_segment(0);
1439 switch_to_new_gdt(0);
1440
1441 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1442 pv_cpu_ops.load_gdt = xen_load_gdt;
1443}
1444
1445static void __init xen_dom0_set_legacy_features(void)
1446{
1447 x86_platform.legacy.rtc = 1;
1448}
1449
1450static int xen_cpuhp_setup(void)
1451{
1452 int rc;
1453
1454 rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
1455 "x86/xen/hvm_guest:prepare",
1456 xen_cpu_up_prepare, xen_cpu_dead);
1457 if (rc >= 0) {
1458 rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
1459 "x86/xen/hvm_guest:online",
1460 xen_cpu_up_online, NULL);
1461 if (rc < 0)
1462 cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
1463 }
1464
1465 return rc >= 0 ? 0 : rc;
1466}
1467
1468/* First C function to be called on Xen boot */
1469asmlinkage __visible void __init xen_start_kernel(void)
1470{
1471 struct physdev_set_iopl set_iopl;
1472 unsigned long initrd_start = 0;
1473 int rc;
1474
1475 if (!xen_start_info)
1476 return;
1477
1478 xen_domain_type = XEN_PV_DOMAIN;
1479
1480 xen_setup_features();
1481
1482 xen_setup_machphys_mapping();
1483
1484 /* Install Xen paravirt ops */
1485 pv_info = xen_info;
1486 pv_init_ops = xen_init_ops;
1487 pv_cpu_ops = xen_cpu_ops;
1488
1489 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1490
1491 x86_init.resources.memory_setup = xen_memory_setup;
1492 x86_init.oem.arch_setup = xen_arch_setup;
1493 x86_init.oem.banner = xen_banner;
1494
1495 xen_init_time_ops();
1496
1497 /*
1498 * Set up some pagetable state before starting to set any ptes.
1499 */
1500
1501 xen_init_mmu_ops();
1502
1503 /* Prevent unwanted bits from being set in PTEs. */
1504 __supported_pte_mask &= ~_PAGE_GLOBAL;
1505
1506 /*
1507 * Prevent page tables from being allocated in highmem, even
1508 * if CONFIG_HIGHPTE is enabled.
1509 */
1510 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1511
1512 /* Work out if we support NX */
1513 x86_configure_nx();
1514
1515 /* Get mfn list */
1516 xen_build_dynamic_phys_to_machine();
1517
1518 /*
1519 * Set up kernel GDT and segment registers, mainly so that
1520 * -fstack-protector code can be executed.
1521 */
1522 xen_setup_gdt(0);
1523
1524 xen_init_irq_ops();
1525 xen_init_cpuid_mask();
1526
1527#ifdef CONFIG_X86_LOCAL_APIC
1528 /*
1529 * set up the basic apic ops.
1530 */
1531 xen_init_apic();
1532#endif
1533
1534 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1535 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1536 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1537 }
1538
1539 machine_ops = xen_machine_ops;
1540
1541 /*
1542 * The only reliable way to retain the initial address of the
1543 * percpu gdt_page is to remember it here, so we can go and
1544 * mark it RW later, when the initial percpu area is freed.
1545 */
1546 xen_initial_gdt = &per_cpu(gdt_page, 0);
1547
1548 xen_smp_init();
1549
1550#ifdef CONFIG_ACPI_NUMA
1551 /*
1552 * The pages we from Xen are not related to machine pages, so
1553 * any NUMA information the kernel tries to get from ACPI will
1554 * be meaningless. Prevent it from trying.
1555 */
1556 acpi_numa = -1;
1557#endif
1558 /* Don't do the full vcpu_info placement stuff until we have a
1559 possible map and a non-dummy shared_info. */
1560 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1561
1562 WARN_ON(xen_cpuhp_setup());
1563
1564 local_irq_disable();
1565 early_boot_irqs_disabled = true;
1566
1567 xen_raw_console_write("mapping kernel into physical memory\n");
1568 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1569 xen_start_info->nr_pages);
1570 xen_reserve_special_pages();
1571
1572 /* keep using Xen gdt for now; no urgent need to change it */
1573
1574#ifdef CONFIG_X86_32
1575 pv_info.kernel_rpl = 1;
1576 if (xen_feature(XENFEAT_supervisor_mode_kernel))
1577 pv_info.kernel_rpl = 0;
1578#else
1579 pv_info.kernel_rpl = 0;
1580#endif
1581 /* set the limit of our address space */
1582 xen_reserve_top();
1583
1584 /*
1585 * We used to do this in xen_arch_setup, but that is too late
1586 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1587 * early_amd_init which pokes 0xcf8 port.
1588 */
1589 set_iopl.iopl = 1;
1590 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1591 if (rc != 0)
1592 xen_raw_printk("physdev_op failed %d\n", rc);
1593
1594#ifdef CONFIG_X86_32
1595 /* set up basic CPUID stuff */
1596 cpu_detect(&new_cpu_data);
1597 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
1598 new_cpu_data.wp_works_ok = 1;
1599 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
1600#endif
1601
1602 if (xen_start_info->mod_start) {
1603 if (xen_start_info->flags & SIF_MOD_START_PFN)
1604 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1605 else
1606 initrd_start = __pa(xen_start_info->mod_start);
1607 }
1608
1609 /* Poke various useful things into boot_params */
1610 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1611 boot_params.hdr.ramdisk_image = initrd_start;
1612 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1613 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1614 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
1615
1616 if (!xen_initial_domain()) {
1617 add_preferred_console("xenboot", 0, NULL);
1618 add_preferred_console("tty", 0, NULL);
1619 add_preferred_console("hvc", 0, NULL);
1620 if (pci_xen)
1621 x86_init.pci.arch_init = pci_xen_init;
1622 } else {
1623 const struct dom0_vga_console_info *info =
1624 (void *)((char *)xen_start_info +
1625 xen_start_info->console.dom0.info_off);
1626 struct xen_platform_op op = {
1627 .cmd = XENPF_firmware_info,
1628 .interface_version = XENPF_INTERFACE_VERSION,
1629 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1630 };
1631
1632 x86_platform.set_legacy_features =
1633 xen_dom0_set_legacy_features;
1634 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1635 xen_start_info->console.domU.mfn = 0;
1636 xen_start_info->console.domU.evtchn = 0;
1637
1638 if (HYPERVISOR_platform_op(&op) == 0)
1639 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1640
1641 /* Make sure ACS will be enabled */
1642 pci_request_acs();
1643
1644 xen_acpi_sleep_register();
1645
1646 /* Avoid searching for BIOS MP tables */
1647 x86_init.mpparse.find_smp_config = x86_init_noop;
1648 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
1649
1650 xen_boot_params_init_edd();
1651 }
1652#ifdef CONFIG_PCI
1653 /* PCI BIOS service won't work from a PV guest. */
1654 pci_probe &= ~PCI_PROBE_BIOS;
1655#endif
1656 xen_raw_console_write("about to get started...\n");
1657
1658 /* Let's presume PV guests always boot on vCPU with id 0. */
1659 per_cpu(xen_vcpu_id, 0) = 0;
1660
1661 xen_setup_runstate_info(0);
1662
1663 xen_efi_init();
1664
1665 /* Start the world */
1666#ifdef CONFIG_X86_32
1667 i386_start_kernel();
1668#else
1669 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
1670 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1671#endif
1672}
1673
1674#ifdef CONFIG_XEN_PVH
1675
1676static void xen_pvh_arch_setup(void)
1677{
1678#ifdef CONFIG_ACPI
1679 /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
1680 if (nr_ioapics == 0)
1681 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
1682#endif
1683}
1684
1685static void __init init_pvh_bootparams(void)
1686{
1687 struct xen_memory_map memmap;
1688 unsigned int i;
1689 int rc;
1690
1691 memset(&pvh_bootparams, 0, sizeof(pvh_bootparams));
1692
1693 memmap.nr_entries = ARRAY_SIZE(pvh_bootparams.e820_map);
1694 set_xen_guest_handle(memmap.buffer, pvh_bootparams.e820_map);
1695 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
1696 if (rc) {
1697 xen_raw_printk("XENMEM_memory_map failed (%d)\n", rc);
1698 BUG();
1699 }
1700
1701 if (memmap.nr_entries < E820MAX - 1) {
1702 pvh_bootparams.e820_map[memmap.nr_entries].addr =
1703 ISA_START_ADDRESS;
1704 pvh_bootparams.e820_map[memmap.nr_entries].size =
1705 ISA_END_ADDRESS - ISA_START_ADDRESS;
1706 pvh_bootparams.e820_map[memmap.nr_entries].type =
1707 E820_RESERVED;
1708 memmap.nr_entries++;
1709 } else
1710 xen_raw_printk("Warning: Can fit ISA range into e820\n");
1711
1712 sanitize_e820_map(pvh_bootparams.e820_map,
1713 ARRAY_SIZE(pvh_bootparams.e820_map),
1714 &memmap.nr_entries);
1715
1716 pvh_bootparams.e820_entries = memmap.nr_entries;
1717 for (i = 0; i < pvh_bootparams.e820_entries; i++)
1718 e820_add_region(pvh_bootparams.e820_map[i].addr,
1719 pvh_bootparams.e820_map[i].size,
1720 pvh_bootparams.e820_map[i].type);
1721
1722 pvh_bootparams.hdr.cmd_line_ptr =
1723 pvh_start_info.cmdline_paddr;
1724
1725 /* The first module is always ramdisk. */
1726 if (pvh_start_info.nr_modules) {
1727 struct hvm_modlist_entry *modaddr =
1728 __va(pvh_start_info.modlist_paddr);
1729 pvh_bootparams.hdr.ramdisk_image = modaddr->paddr;
1730 pvh_bootparams.hdr.ramdisk_size = modaddr->size;
1731 }
1732
1733 /*
1734 * See Documentation/x86/boot.txt.
1735 *
1736 * Version 2.12 supports Xen entry point but we will use default x86/PC
1737 * environment (i.e. hardware_subarch 0).
1738 */
1739 pvh_bootparams.hdr.version = 0x212;
1740 pvh_bootparams.hdr.type_of_loader = (9 << 4) | 0; /* Xen loader */
1741}
1742
1743/*
1744 * This routine (and those that it might call) should not use
1745 * anything that lives in .bss since that segment will be cleared later.
1746 */
1747void __init xen_prepare_pvh(void)
1748{
1749 u32 msr;
1750 u64 pfn;
1751
1752 if (pvh_start_info.magic != XEN_HVM_START_MAGIC_VALUE) {
1753 xen_raw_printk("Error: Unexpected magic value (0x%08x)\n",
1754 pvh_start_info.magic);
1755 BUG();
1756 }
1757
1758 xen_pvh = 1;
1759
1760 msr = cpuid_ebx(xen_cpuid_base() + 2);
1761 pfn = __pa(hypercall_page);
1762 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1763
1764 init_pvh_bootparams();
1765
1766 x86_init.oem.arch_setup = xen_pvh_arch_setup;
1767}
1768#endif
1769
1770void __ref xen_hvm_init_shared_info(void)
1771{
1772 int cpu;
1773 struct xen_add_to_physmap xatp;
1774 static struct shared_info *shared_info_page = 0;
1775
1776 if (!shared_info_page)
1777 shared_info_page = (struct shared_info *)
1778 extend_brk(PAGE_SIZE, PAGE_SIZE);
1779 xatp.domid = DOMID_SELF;
1780 xatp.idx = 0;
1781 xatp.space = XENMAPSPACE_shared_info;
1782 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1783 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1784 BUG();
1785
1786 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1787
1788 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1789 * page, we use it in the event channel upcall and in some pvclock
1790 * related functions. We don't need the vcpu_info placement
1791 * optimizations because we don't use any pv_mmu or pv_irq op on
1792 * HVM.
1793 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1794 * online but xen_hvm_init_shared_info is run at resume time too and
1795 * in that case multiple vcpus might be online. */
1796 for_each_online_cpu(cpu) {
1797 /* Leave it to be NULL. */
1798 if (xen_vcpu_nr(cpu) >= MAX_VIRT_CPUS)
1799 continue;
1800 per_cpu(xen_vcpu, cpu) =
1801 &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
1802 }
1803}
1804
1805#ifdef CONFIG_XEN_PVHVM
1806static void __init init_hvm_pv_info(void)
1807{
1808 int major, minor;
1809 uint32_t eax, ebx, ecx, edx, base;
1810
1811 base = xen_cpuid_base();
1812 eax = cpuid_eax(base + 1);
1813
1814 major = eax >> 16;
1815 minor = eax & 0xffff;
1816 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1817
1818 xen_domain_type = XEN_HVM_DOMAIN;
1819
1820 /* PVH set up hypercall page in xen_prepare_pvh(). */
1821 if (xen_pvh_domain())
1822 pv_info.name = "Xen PVH";
1823 else {
1824 u64 pfn;
1825 uint32_t msr;
1826
1827 pv_info.name = "Xen HVM";
1828 msr = cpuid_ebx(base + 2);
1829 pfn = __pa(hypercall_page);
1830 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1831 }
1832
1833 xen_setup_features();
1834
1835 cpuid(base + 4, &eax, &ebx, &ecx, &edx);
1836 if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
1837 this_cpu_write(xen_vcpu_id, ebx);
1838 else
1839 this_cpu_write(xen_vcpu_id, smp_processor_id());
1840}
1841#endif
1842
1843static int xen_cpu_up_prepare(unsigned int cpu)
1844{
1845 int rc;
1846
1847 if (xen_hvm_domain()) {
1848 /*
1849 * This can happen if CPU was offlined earlier and
1850 * offlining timed out in common_cpu_die().
1851 */
1852 if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
1853 xen_smp_intr_free(cpu);
1854 xen_uninit_lock_cpu(cpu);
1855 }
1856
1857 if (cpu_acpi_id(cpu) != U32_MAX)
1858 per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
1859 else
1860 per_cpu(xen_vcpu_id, cpu) = cpu;
1861 xen_vcpu_setup(cpu);
1862 }
1863
1864 if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
1865 xen_setup_timer(cpu);
1866
1867 rc = xen_smp_intr_init(cpu);
1868 if (rc) {
1869 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1870 cpu, rc);
1871 return rc;
1872 }
1873 return 0;
1874}
1875
1876static int xen_cpu_dead(unsigned int cpu)
1877{
1878 xen_smp_intr_free(cpu);
1879
1880 if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
1881 xen_teardown_timer(cpu);
1882
1883 return 0;
1884}
1885
1886static int xen_cpu_up_online(unsigned int cpu)
1887{
1888 xen_init_lock_cpu(cpu);
1889 return 0;
1890}
1891
1892#ifdef CONFIG_XEN_PVHVM
1893#ifdef CONFIG_KEXEC_CORE
1894static void xen_hvm_shutdown(void)
1895{
1896 native_machine_shutdown();
1897 if (kexec_in_progress)
1898 xen_reboot(SHUTDOWN_soft_reset);
1899}
1900
1901static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1902{
1903 native_machine_crash_shutdown(regs);
1904 xen_reboot(SHUTDOWN_soft_reset);
1905}
1906#endif
1907
1908static void __init xen_hvm_guest_init(void)
1909{
1910 if (xen_pv_domain())
1911 return;
1912
1913 init_hvm_pv_info();
1914
1915 xen_hvm_init_shared_info();
1916
1917 xen_panic_handler_init();
1918
1919 BUG_ON(!xen_feature(XENFEAT_hvm_callback_vector));
1920
1921 xen_hvm_smp_init();
1922 WARN_ON(xen_cpuhp_setup());
1923 xen_unplug_emulated_devices();
1924 x86_init.irqs.intr_init = xen_init_IRQ;
1925 xen_hvm_init_time_ops();
1926 xen_hvm_init_mmu_ops();
1927
1928 if (xen_pvh_domain())
1929 machine_ops.emergency_restart = xen_emergency_restart;
1930#ifdef CONFIG_KEXEC_CORE
1931 machine_ops.shutdown = xen_hvm_shutdown;
1932 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1933#endif
1934}
1935#endif
1936
1937static bool xen_nopv = false;
1938static __init int xen_parse_nopv(char *arg)
1939{
1940 xen_nopv = true;
1941 return 0;
1942}
1943early_param("xen_nopv", xen_parse_nopv);
1944
1945static uint32_t __init xen_platform(void)
1946{
1947 if (xen_nopv)
1948 return 0;
1949
1950 return xen_cpuid_base();
1951}
1952
1953bool xen_hvm_need_lapic(void)
1954{
1955 if (xen_nopv)
1956 return false;
1957 if (xen_pv_domain())
1958 return false;
1959 if (!xen_hvm_domain())
1960 return false;
1961 if (xen_feature(XENFEAT_hvm_pirqs))
1962 return false;
1963 return true;
1964}
1965EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1966
1967static void xen_set_cpu_features(struct cpuinfo_x86 *c)
1968{
1969 if (xen_pv_domain()) {
1970 clear_cpu_bug(c, X86_BUG_SYSRET_SS_ATTRS);
1971 set_cpu_cap(c, X86_FEATURE_XENPV);
1972 }
1973}
1974
1975static void xen_pin_vcpu(int cpu)
1976{
1977 static bool disable_pinning;
1978 struct sched_pin_override pin_override;
1979 int ret;
1980
1981 if (disable_pinning)
1982 return;
1983
1984 pin_override.pcpu = cpu;
1985 ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
1986
1987 /* Ignore errors when removing override. */
1988 if (cpu < 0)
1989 return;
1990
1991 switch (ret) {
1992 case -ENOSYS:
1993 pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
1994 cpu);
1995 disable_pinning = true;
1996 break;
1997 case -EPERM:
1998 WARN(1, "Trying to pin vcpu without having privilege to do so\n");
1999 disable_pinning = true;
2000 break;
2001 case -EINVAL:
2002 case -EBUSY:
2003 pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
2004 cpu);
2005 break;
2006 case 0:
2007 break;
2008 default:
2009 WARN(1, "rc %d while trying to pin vcpu\n", ret);
2010 disable_pinning = true;
2011 }
2012}
2013
2014const struct hypervisor_x86 x86_hyper_xen = {
2015 .name = "Xen",
2016 .detect = xen_platform,
2017#ifdef CONFIG_XEN_PVHVM
2018 .init_platform = xen_hvm_guest_init,
2019#endif
2020 .x2apic_available = xen_x2apic_para_available,
2021 .set_cpu_features = xen_set_cpu_features,
2022 .pin_vcpu = xen_pin_vcpu,
2023};
2024EXPORT_SYMBOL(x86_hyper_xen);
2025
2026#ifdef CONFIG_HOTPLUG_CPU
2027void xen_arch_register_cpu(int num)
2028{
2029 arch_register_cpu(num);
2030}
2031EXPORT_SYMBOL(xen_arch_register_cpu);
2032
2033void xen_arch_unregister_cpu(int num)
2034{
2035 arch_unregister_cpu(num);
2036}
2037EXPORT_SYMBOL(xen_arch_unregister_cpu);
2038#endif
2039