Commit | Line | Data |
---|---|---|
5ead97c8 JF |
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/kernel.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/smp.h> | |
17 | #include <linux/preempt.h> | |
f120f13e | 18 | #include <linux/hardirq.h> |
5ead97c8 JF |
19 | #include <linux/percpu.h> |
20 | #include <linux/delay.h> | |
21 | #include <linux/start_kernel.h> | |
22 | #include <linux/sched.h> | |
6cac5a92 | 23 | #include <linux/kprobes.h> |
5ead97c8 JF |
24 | #include <linux/bootmem.h> |
25 | #include <linux/module.h> | |
f4f97b3e JF |
26 | #include <linux/mm.h> |
27 | #include <linux/page-flags.h> | |
28 | #include <linux/highmem.h> | |
b8c2d3df | 29 | #include <linux/console.h> |
5d990b62 | 30 | #include <linux/pci.h> |
5a0e3ad6 | 31 | #include <linux/gfp.h> |
5ead97c8 | 32 | |
1ccbf534 | 33 | #include <xen/xen.h> |
5ead97c8 | 34 | #include <xen/interface/xen.h> |
ecbf29cd | 35 | #include <xen/interface/version.h> |
5ead97c8 JF |
36 | #include <xen/interface/physdev.h> |
37 | #include <xen/interface/vcpu.h> | |
38 | #include <xen/features.h> | |
39 | #include <xen/page.h> | |
084a2a4e | 40 | #include <xen/hvc-console.h> |
5ead97c8 JF |
41 | |
42 | #include <asm/paravirt.h> | |
7b6aa335 | 43 | #include <asm/apic.h> |
5ead97c8 JF |
44 | #include <asm/page.h> |
45 | #include <asm/xen/hypercall.h> | |
46 | #include <asm/xen/hypervisor.h> | |
47 | #include <asm/fixmap.h> | |
48 | #include <asm/processor.h> | |
707ebbc8 | 49 | #include <asm/proto.h> |
1153968a | 50 | #include <asm/msr-index.h> |
6cac5a92 | 51 | #include <asm/traps.h> |
5ead97c8 JF |
52 | #include <asm/setup.h> |
53 | #include <asm/desc.h> | |
817a824b | 54 | #include <asm/pgalloc.h> |
5ead97c8 | 55 | #include <asm/pgtable.h> |
f87e4cac | 56 | #include <asm/tlbflush.h> |
fefa629a | 57 | #include <asm/reboot.h> |
577eebea | 58 | #include <asm/stackprotector.h> |
5ead97c8 JF |
59 | |
60 | #include "xen-ops.h" | |
3b827c1b | 61 | #include "mmu.h" |
5ead97c8 JF |
62 | #include "multicalls.h" |
63 | ||
64 | EXPORT_SYMBOL_GPL(hypercall_page); | |
65 | ||
5ead97c8 JF |
66 | DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); |
67 | DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); | |
9f79991d | 68 | |
6e833587 JF |
69 | enum xen_domain_type xen_domain_type = XEN_NATIVE; |
70 | EXPORT_SYMBOL_GPL(xen_domain_type); | |
71 | ||
5ead97c8 JF |
72 | struct start_info *xen_start_info; |
73 | EXPORT_SYMBOL_GPL(xen_start_info); | |
74 | ||
a0d695c8 | 75 | struct shared_info xen_dummy_shared_info; |
60223a32 | 76 | |
38341432 JF |
77 | void *xen_initial_gdt; |
78 | ||
60223a32 JF |
79 | /* |
80 | * Point at some empty memory to start with. We map the real shared_info | |
81 | * page as soon as fixmap is up and running. | |
82 | */ | |
a0d695c8 | 83 | struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; |
60223a32 JF |
84 | |
85 | /* | |
86 | * Flag to determine whether vcpu info placement is available on all | |
87 | * VCPUs. We assume it is to start with, and then set it to zero on | |
88 | * the first failure. This is because it can succeed on some VCPUs | |
89 | * and not others, since it can involve hypervisor memory allocation, | |
90 | * or because the guest failed to guarantee all the appropriate | |
91 | * constraints on all VCPUs (ie buffer can't cross a page boundary). | |
92 | * | |
93 | * Note that any particular CPU may be using a placed vcpu structure, | |
94 | * but we can only optimise if the all are. | |
95 | * | |
96 | * 0: not available, 1: available | |
97 | */ | |
e4d04071 | 98 | static int have_vcpu_info_placement = 1; |
60223a32 | 99 | |
c06ee78d MR |
100 | static void clamp_max_cpus(void) |
101 | { | |
102 | #ifdef CONFIG_SMP | |
103 | if (setup_max_cpus > MAX_VIRT_CPUS) | |
104 | setup_max_cpus = MAX_VIRT_CPUS; | |
105 | #endif | |
106 | } | |
107 | ||
9c7a7942 | 108 | static void xen_vcpu_setup(int cpu) |
5ead97c8 | 109 | { |
60223a32 JF |
110 | struct vcpu_register_vcpu_info info; |
111 | int err; | |
112 | struct vcpu_info *vcpup; | |
113 | ||
a0d695c8 | 114 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
60223a32 | 115 | |
c06ee78d MR |
116 | if (cpu < MAX_VIRT_CPUS) |
117 | per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; | |
60223a32 | 118 | |
c06ee78d MR |
119 | if (!have_vcpu_info_placement) { |
120 | if (cpu >= MAX_VIRT_CPUS) | |
121 | clamp_max_cpus(); | |
122 | return; | |
123 | } | |
60223a32 | 124 | |
c06ee78d | 125 | vcpup = &per_cpu(xen_vcpu_info, cpu); |
9976b39b | 126 | info.mfn = arbitrary_virt_to_mfn(vcpup); |
60223a32 JF |
127 | info.offset = offset_in_page(vcpup); |
128 | ||
e3d26976 | 129 | printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", |
60223a32 JF |
130 | cpu, vcpup, info.mfn, info.offset); |
131 | ||
132 | /* Check to see if the hypervisor will put the vcpu_info | |
133 | structure where we want it, which allows direct access via | |
134 | a percpu-variable. */ | |
135 | err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); | |
136 | ||
137 | if (err) { | |
138 | printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); | |
139 | have_vcpu_info_placement = 0; | |
c06ee78d | 140 | clamp_max_cpus(); |
60223a32 JF |
141 | } else { |
142 | /* This cpu is using the registered vcpu info, even if | |
143 | later ones fail to. */ | |
144 | per_cpu(xen_vcpu, cpu) = vcpup; | |
6487673b | 145 | |
60223a32 JF |
146 | printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", |
147 | cpu, vcpup); | |
148 | } | |
5ead97c8 JF |
149 | } |
150 | ||
9c7a7942 JF |
151 | /* |
152 | * On restore, set the vcpu placement up again. | |
153 | * If it fails, then we're in a bad state, since | |
154 | * we can't back out from using it... | |
155 | */ | |
156 | void xen_vcpu_restore(void) | |
157 | { | |
3905bb2a | 158 | int cpu; |
9c7a7942 | 159 | |
3905bb2a JF |
160 | for_each_online_cpu(cpu) { |
161 | bool other_cpu = (cpu != smp_processor_id()); | |
9c7a7942 | 162 | |
3905bb2a JF |
163 | if (other_cpu && |
164 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) | |
165 | BUG(); | |
9c7a7942 | 166 | |
3905bb2a | 167 | xen_setup_runstate_info(cpu); |
9c7a7942 | 168 | |
3905bb2a | 169 | if (have_vcpu_info_placement) |
9c7a7942 | 170 | xen_vcpu_setup(cpu); |
9c7a7942 | 171 | |
3905bb2a JF |
172 | if (other_cpu && |
173 | HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) | |
174 | BUG(); | |
9c7a7942 JF |
175 | } |
176 | } | |
177 | ||
5ead97c8 JF |
178 | static void __init xen_banner(void) |
179 | { | |
95c7c23b JF |
180 | unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); |
181 | struct xen_extraversion extra; | |
182 | HYPERVISOR_xen_version(XENVER_extraversion, &extra); | |
183 | ||
5ead97c8 | 184 | printk(KERN_INFO "Booting paravirtualized kernel on %s\n", |
93b1eab3 | 185 | pv_info.name); |
95c7c23b JF |
186 | printk(KERN_INFO "Xen version: %d.%d%s%s\n", |
187 | version >> 16, version & 0xffff, extra.extraversion, | |
e57778a1 | 188 | xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); |
5ead97c8 JF |
189 | } |
190 | ||
e826fe1b JF |
191 | static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0; |
192 | static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0; | |
193 | ||
65ea5b03 PA |
194 | static void xen_cpuid(unsigned int *ax, unsigned int *bx, |
195 | unsigned int *cx, unsigned int *dx) | |
5ead97c8 | 196 | { |
82d64699 | 197 | unsigned maskebx = ~0; |
e826fe1b | 198 | unsigned maskecx = ~0; |
5ead97c8 JF |
199 | unsigned maskedx = ~0; |
200 | ||
201 | /* | |
202 | * Mask out inconvenient features, to try and disable as many | |
203 | * unsupported kernel subsystems as possible. | |
204 | */ | |
82d64699 JF |
205 | switch (*ax) { |
206 | case 1: | |
e826fe1b JF |
207 | maskecx = cpuid_leaf1_ecx_mask; |
208 | maskedx = cpuid_leaf1_edx_mask; | |
82d64699 JF |
209 | break; |
210 | ||
211 | case 0xb: | |
212 | /* Suppress extended topology stuff */ | |
213 | maskebx = 0; | |
214 | break; | |
e826fe1b | 215 | } |
5ead97c8 JF |
216 | |
217 | asm(XEN_EMULATE_PREFIX "cpuid" | |
65ea5b03 PA |
218 | : "=a" (*ax), |
219 | "=b" (*bx), | |
220 | "=c" (*cx), | |
221 | "=d" (*dx) | |
222 | : "0" (*ax), "2" (*cx)); | |
e826fe1b | 223 | |
82d64699 | 224 | *bx &= maskebx; |
e826fe1b | 225 | *cx &= maskecx; |
65ea5b03 | 226 | *dx &= maskedx; |
5ead97c8 JF |
227 | } |
228 | ||
e826fe1b JF |
229 | static __init void xen_init_cpuid_mask(void) |
230 | { | |
231 | unsigned int ax, bx, cx, dx; | |
232 | ||
233 | cpuid_leaf1_edx_mask = | |
234 | ~((1 << X86_FEATURE_MCE) | /* disable MCE */ | |
235 | (1 << X86_FEATURE_MCA) | /* disable MCA */ | |
236 | (1 << X86_FEATURE_ACC)); /* thermal monitoring */ | |
237 | ||
238 | if (!xen_initial_domain()) | |
239 | cpuid_leaf1_edx_mask &= | |
240 | ~((1 << X86_FEATURE_APIC) | /* disable local APIC */ | |
241 | (1 << X86_FEATURE_ACPI)); /* disable ACPI */ | |
242 | ||
243 | ax = 1; | |
7adb4df4 | 244 | cx = 0; |
e826fe1b JF |
245 | xen_cpuid(&ax, &bx, &cx, &dx); |
246 | ||
247 | /* cpuid claims we support xsave; try enabling it to see what happens */ | |
248 | if (cx & (1 << (X86_FEATURE_XSAVE % 32))) { | |
249 | unsigned long cr4; | |
250 | ||
251 | set_in_cr4(X86_CR4_OSXSAVE); | |
252 | ||
253 | cr4 = read_cr4(); | |
254 | ||
255 | if ((cr4 & X86_CR4_OSXSAVE) == 0) | |
256 | cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_XSAVE % 32)); | |
257 | ||
258 | clear_in_cr4(X86_CR4_OSXSAVE); | |
259 | } | |
260 | } | |
261 | ||
5ead97c8 JF |
262 | static void xen_set_debugreg(int reg, unsigned long val) |
263 | { | |
264 | HYPERVISOR_set_debugreg(reg, val); | |
265 | } | |
266 | ||
267 | static unsigned long xen_get_debugreg(int reg) | |
268 | { | |
269 | return HYPERVISOR_get_debugreg(reg); | |
270 | } | |
271 | ||
224101ed | 272 | static void xen_end_context_switch(struct task_struct *next) |
5ead97c8 | 273 | { |
5ead97c8 | 274 | xen_mc_flush(); |
224101ed | 275 | paravirt_end_context_switch(next); |
5ead97c8 JF |
276 | } |
277 | ||
278 | static unsigned long xen_store_tr(void) | |
279 | { | |
280 | return 0; | |
281 | } | |
282 | ||
a05d2eba | 283 | /* |
cef43bf6 JF |
284 | * Set the page permissions for a particular virtual address. If the |
285 | * address is a vmalloc mapping (or other non-linear mapping), then | |
286 | * find the linear mapping of the page and also set its protections to | |
287 | * match. | |
a05d2eba JF |
288 | */ |
289 | static void set_aliased_prot(void *v, pgprot_t prot) | |
290 | { | |
291 | int level; | |
292 | pte_t *ptep; | |
293 | pte_t pte; | |
294 | unsigned long pfn; | |
295 | struct page *page; | |
296 | ||
297 | ptep = lookup_address((unsigned long)v, &level); | |
298 | BUG_ON(ptep == NULL); | |
299 | ||
300 | pfn = pte_pfn(*ptep); | |
301 | page = pfn_to_page(pfn); | |
302 | ||
303 | pte = pfn_pte(pfn, prot); | |
304 | ||
305 | if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) | |
306 | BUG(); | |
307 | ||
308 | if (!PageHighMem(page)) { | |
309 | void *av = __va(PFN_PHYS(pfn)); | |
310 | ||
311 | if (av != v) | |
312 | if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) | |
313 | BUG(); | |
314 | } else | |
315 | kmap_flush_unused(); | |
316 | } | |
317 | ||
38ffbe66 JF |
318 | static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) |
319 | { | |
a05d2eba | 320 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
321 | int i; |
322 | ||
a05d2eba JF |
323 | for(i = 0; i < entries; i += entries_per_page) |
324 | set_aliased_prot(ldt + i, PAGE_KERNEL_RO); | |
38ffbe66 JF |
325 | } |
326 | ||
327 | static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) | |
328 | { | |
a05d2eba | 329 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
330 | int i; |
331 | ||
a05d2eba JF |
332 | for(i = 0; i < entries; i += entries_per_page) |
333 | set_aliased_prot(ldt + i, PAGE_KERNEL); | |
38ffbe66 JF |
334 | } |
335 | ||
5ead97c8 JF |
336 | static void xen_set_ldt(const void *addr, unsigned entries) |
337 | { | |
5ead97c8 JF |
338 | struct mmuext_op *op; |
339 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
340 | ||
341 | op = mcs.args; | |
342 | op->cmd = MMUEXT_SET_LDT; | |
4dbf7af6 | 343 | op->arg1.linear_addr = (unsigned long)addr; |
5ead97c8 JF |
344 | op->arg2.nr_ents = entries; |
345 | ||
346 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
347 | ||
348 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
349 | } | |
350 | ||
6b68f01b | 351 | static void xen_load_gdt(const struct desc_ptr *dtr) |
5ead97c8 | 352 | { |
5ead97c8 JF |
353 | unsigned long va = dtr->address; |
354 | unsigned int size = dtr->size + 1; | |
355 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
3ce5fa7e | 356 | unsigned long frames[pages]; |
5ead97c8 | 357 | int f; |
5ead97c8 | 358 | |
577eebea JF |
359 | /* |
360 | * A GDT can be up to 64k in size, which corresponds to 8192 | |
361 | * 8-byte entries, or 16 4k pages.. | |
362 | */ | |
5ead97c8 JF |
363 | |
364 | BUG_ON(size > 65536); | |
365 | BUG_ON(va & ~PAGE_MASK); | |
366 | ||
5ead97c8 | 367 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { |
6ed6bf42 | 368 | int level; |
577eebea | 369 | pte_t *ptep; |
6ed6bf42 JF |
370 | unsigned long pfn, mfn; |
371 | void *virt; | |
372 | ||
577eebea JF |
373 | /* |
374 | * The GDT is per-cpu and is in the percpu data area. | |
375 | * That can be virtually mapped, so we need to do a | |
376 | * page-walk to get the underlying MFN for the | |
377 | * hypercall. The page can also be in the kernel's | |
378 | * linear range, so we need to RO that mapping too. | |
379 | */ | |
380 | ptep = lookup_address(va, &level); | |
6ed6bf42 JF |
381 | BUG_ON(ptep == NULL); |
382 | ||
383 | pfn = pte_pfn(*ptep); | |
384 | mfn = pfn_to_mfn(pfn); | |
385 | virt = __va(PFN_PHYS(pfn)); | |
386 | ||
387 | frames[f] = mfn; | |
9976b39b | 388 | |
5ead97c8 | 389 | make_lowmem_page_readonly((void *)va); |
6ed6bf42 | 390 | make_lowmem_page_readonly(virt); |
5ead97c8 JF |
391 | } |
392 | ||
3ce5fa7e JF |
393 | if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) |
394 | BUG(); | |
5ead97c8 JF |
395 | } |
396 | ||
577eebea JF |
397 | /* |
398 | * load_gdt for early boot, when the gdt is only mapped once | |
399 | */ | |
400 | static __init void xen_load_gdt_boot(const struct desc_ptr *dtr) | |
401 | { | |
402 | unsigned long va = dtr->address; | |
403 | unsigned int size = dtr->size + 1; | |
404 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
405 | unsigned long frames[pages]; | |
406 | int f; | |
407 | ||
408 | /* | |
409 | * A GDT can be up to 64k in size, which corresponds to 8192 | |
410 | * 8-byte entries, or 16 4k pages.. | |
411 | */ | |
412 | ||
413 | BUG_ON(size > 65536); | |
414 | BUG_ON(va & ~PAGE_MASK); | |
415 | ||
416 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
417 | pte_t pte; | |
418 | unsigned long pfn, mfn; | |
419 | ||
420 | pfn = virt_to_pfn(va); | |
421 | mfn = pfn_to_mfn(pfn); | |
422 | ||
423 | pte = pfn_pte(pfn, PAGE_KERNEL_RO); | |
424 | ||
425 | if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) | |
426 | BUG(); | |
427 | ||
428 | frames[f] = mfn; | |
429 | } | |
430 | ||
431 | if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) | |
432 | BUG(); | |
433 | } | |
434 | ||
5ead97c8 JF |
435 | static void load_TLS_descriptor(struct thread_struct *t, |
436 | unsigned int cpu, unsigned int i) | |
437 | { | |
438 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
9976b39b | 439 | xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); |
5ead97c8 JF |
440 | struct multicall_space mc = __xen_mc_entry(0); |
441 | ||
442 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
443 | } | |
444 | ||
445 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
446 | { | |
8b84ad94 | 447 | /* |
ccbeed3a TH |
448 | * XXX sleazy hack: If we're being called in a lazy-cpu zone |
449 | * and lazy gs handling is enabled, it means we're in a | |
450 | * context switch, and %gs has just been saved. This means we | |
451 | * can zero it out to prevent faults on exit from the | |
452 | * hypervisor if the next process has no %gs. Either way, it | |
453 | * has been saved, and the new value will get loaded properly. | |
454 | * This will go away as soon as Xen has been modified to not | |
455 | * save/restore %gs for normal hypercalls. | |
8a95408e EH |
456 | * |
457 | * On x86_64, this hack is not used for %gs, because gs points | |
458 | * to KERNEL_GS_BASE (and uses it for PDA references), so we | |
459 | * must not zero %gs on x86_64 | |
460 | * | |
461 | * For x86_64, we need to zero %fs, otherwise we may get an | |
462 | * exception between the new %fs descriptor being loaded and | |
463 | * %fs being effectively cleared at __switch_to(). | |
8b84ad94 | 464 | */ |
8a95408e EH |
465 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { |
466 | #ifdef CONFIG_X86_32 | |
ccbeed3a | 467 | lazy_load_gs(0); |
8a95408e EH |
468 | #else |
469 | loadsegment(fs, 0); | |
470 | #endif | |
471 | } | |
472 | ||
473 | xen_mc_batch(); | |
474 | ||
475 | load_TLS_descriptor(t, cpu, 0); | |
476 | load_TLS_descriptor(t, cpu, 1); | |
477 | load_TLS_descriptor(t, cpu, 2); | |
478 | ||
479 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
5ead97c8 JF |
480 | } |
481 | ||
a8fc1089 EH |
482 | #ifdef CONFIG_X86_64 |
483 | static void xen_load_gs_index(unsigned int idx) | |
484 | { | |
485 | if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) | |
486 | BUG(); | |
5ead97c8 | 487 | } |
a8fc1089 | 488 | #endif |
5ead97c8 JF |
489 | |
490 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
75b8bb3e | 491 | const void *ptr) |
5ead97c8 | 492 | { |
cef43bf6 | 493 | xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); |
75b8bb3e | 494 | u64 entry = *(u64 *)ptr; |
5ead97c8 | 495 | |
f120f13e JF |
496 | preempt_disable(); |
497 | ||
5ead97c8 JF |
498 | xen_mc_flush(); |
499 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
500 | BUG(); | |
f120f13e JF |
501 | |
502 | preempt_enable(); | |
5ead97c8 JF |
503 | } |
504 | ||
e176d367 | 505 | static int cvt_gate_to_trap(int vector, const gate_desc *val, |
5ead97c8 JF |
506 | struct trap_info *info) |
507 | { | |
6cac5a92 JF |
508 | unsigned long addr; |
509 | ||
6d02c426 | 510 | if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT) |
5ead97c8 JF |
511 | return 0; |
512 | ||
513 | info->vector = vector; | |
6cac5a92 JF |
514 | |
515 | addr = gate_offset(*val); | |
516 | #ifdef CONFIG_X86_64 | |
b80119bb JF |
517 | /* |
518 | * Look for known traps using IST, and substitute them | |
519 | * appropriately. The debugger ones are the only ones we care | |
520 | * about. Xen will handle faults like double_fault and | |
521 | * machine_check, so we should never see them. Warn if | |
522 | * there's an unexpected IST-using fault handler. | |
523 | */ | |
6cac5a92 JF |
524 | if (addr == (unsigned long)debug) |
525 | addr = (unsigned long)xen_debug; | |
526 | else if (addr == (unsigned long)int3) | |
527 | addr = (unsigned long)xen_int3; | |
528 | else if (addr == (unsigned long)stack_segment) | |
529 | addr = (unsigned long)xen_stack_segment; | |
b80119bb JF |
530 | else if (addr == (unsigned long)double_fault || |
531 | addr == (unsigned long)nmi) { | |
532 | /* Don't need to handle these */ | |
533 | return 0; | |
534 | #ifdef CONFIG_X86_MCE | |
535 | } else if (addr == (unsigned long)machine_check) { | |
536 | return 0; | |
537 | #endif | |
538 | } else { | |
539 | /* Some other trap using IST? */ | |
540 | if (WARN_ON(val->ist != 0)) | |
541 | return 0; | |
542 | } | |
6cac5a92 JF |
543 | #endif /* CONFIG_X86_64 */ |
544 | info->address = addr; | |
545 | ||
e176d367 EH |
546 | info->cs = gate_segment(*val); |
547 | info->flags = val->dpl; | |
5ead97c8 | 548 | /* interrupt gates clear IF */ |
6d02c426 JF |
549 | if (val->type == GATE_INTERRUPT) |
550 | info->flags |= 1 << 2; | |
5ead97c8 JF |
551 | |
552 | return 1; | |
553 | } | |
554 | ||
555 | /* Locations of each CPU's IDT */ | |
6b68f01b | 556 | static DEFINE_PER_CPU(struct desc_ptr, idt_desc); |
5ead97c8 JF |
557 | |
558 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
559 | also update Xen. */ | |
8d947344 | 560 | static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) |
5ead97c8 | 561 | { |
5ead97c8 | 562 | unsigned long p = (unsigned long)&dt[entrynum]; |
f120f13e JF |
563 | unsigned long start, end; |
564 | ||
565 | preempt_disable(); | |
566 | ||
567 | start = __get_cpu_var(idt_desc).address; | |
568 | end = start + __get_cpu_var(idt_desc).size + 1; | |
5ead97c8 JF |
569 | |
570 | xen_mc_flush(); | |
571 | ||
8d947344 | 572 | native_write_idt_entry(dt, entrynum, g); |
5ead97c8 JF |
573 | |
574 | if (p >= start && (p + 8) <= end) { | |
575 | struct trap_info info[2]; | |
576 | ||
577 | info[1].address = 0; | |
578 | ||
e176d367 | 579 | if (cvt_gate_to_trap(entrynum, g, &info[0])) |
5ead97c8 JF |
580 | if (HYPERVISOR_set_trap_table(info)) |
581 | BUG(); | |
582 | } | |
f120f13e JF |
583 | |
584 | preempt_enable(); | |
5ead97c8 JF |
585 | } |
586 | ||
6b68f01b | 587 | static void xen_convert_trap_info(const struct desc_ptr *desc, |
f87e4cac | 588 | struct trap_info *traps) |
5ead97c8 | 589 | { |
5ead97c8 JF |
590 | unsigned in, out, count; |
591 | ||
e176d367 | 592 | count = (desc->size+1) / sizeof(gate_desc); |
5ead97c8 JF |
593 | BUG_ON(count > 256); |
594 | ||
5ead97c8 | 595 | for (in = out = 0; in < count; in++) { |
e176d367 | 596 | gate_desc *entry = (gate_desc*)(desc->address) + in; |
5ead97c8 | 597 | |
e176d367 | 598 | if (cvt_gate_to_trap(in, entry, &traps[out])) |
5ead97c8 JF |
599 | out++; |
600 | } | |
601 | traps[out].address = 0; | |
f87e4cac JF |
602 | } |
603 | ||
604 | void xen_copy_trap_info(struct trap_info *traps) | |
605 | { | |
6b68f01b | 606 | const struct desc_ptr *desc = &__get_cpu_var(idt_desc); |
f87e4cac JF |
607 | |
608 | xen_convert_trap_info(desc, traps); | |
f87e4cac JF |
609 | } |
610 | ||
611 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
612 | hold a spinlock to protect the static traps[] array (static because | |
613 | it avoids allocation, and saves stack space). */ | |
6b68f01b | 614 | static void xen_load_idt(const struct desc_ptr *desc) |
f87e4cac JF |
615 | { |
616 | static DEFINE_SPINLOCK(lock); | |
617 | static struct trap_info traps[257]; | |
f87e4cac JF |
618 | |
619 | spin_lock(&lock); | |
620 | ||
f120f13e JF |
621 | __get_cpu_var(idt_desc) = *desc; |
622 | ||
f87e4cac | 623 | xen_convert_trap_info(desc, traps); |
5ead97c8 JF |
624 | |
625 | xen_mc_flush(); | |
626 | if (HYPERVISOR_set_trap_table(traps)) | |
627 | BUG(); | |
628 | ||
629 | spin_unlock(&lock); | |
630 | } | |
631 | ||
632 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
633 | they're handled differently. */ | |
634 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
014b15be | 635 | const void *desc, int type) |
5ead97c8 | 636 | { |
f120f13e JF |
637 | preempt_disable(); |
638 | ||
014b15be GOC |
639 | switch (type) { |
640 | case DESC_LDT: | |
641 | case DESC_TSS: | |
5ead97c8 JF |
642 | /* ignore */ |
643 | break; | |
644 | ||
645 | default: { | |
9976b39b | 646 | xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); |
5ead97c8 JF |
647 | |
648 | xen_mc_flush(); | |
014b15be | 649 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) |
5ead97c8 JF |
650 | BUG(); |
651 | } | |
652 | ||
653 | } | |
f120f13e JF |
654 | |
655 | preempt_enable(); | |
5ead97c8 JF |
656 | } |
657 | ||
577eebea JF |
658 | /* |
659 | * Version of write_gdt_entry for use at early boot-time needed to | |
660 | * update an entry as simply as possible. | |
661 | */ | |
662 | static __init void xen_write_gdt_entry_boot(struct desc_struct *dt, int entry, | |
663 | const void *desc, int type) | |
664 | { | |
665 | switch (type) { | |
666 | case DESC_LDT: | |
667 | case DESC_TSS: | |
668 | /* ignore */ | |
669 | break; | |
670 | ||
671 | default: { | |
672 | xmaddr_t maddr = virt_to_machine(&dt[entry]); | |
673 | ||
674 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) | |
675 | dt[entry] = *(struct desc_struct *)desc; | |
676 | } | |
677 | ||
678 | } | |
679 | } | |
680 | ||
faca6227 | 681 | static void xen_load_sp0(struct tss_struct *tss, |
a05d2eba | 682 | struct thread_struct *thread) |
5ead97c8 JF |
683 | { |
684 | struct multicall_space mcs = xen_mc_entry(0); | |
faca6227 | 685 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); |
5ead97c8 JF |
686 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
687 | } | |
688 | ||
689 | static void xen_set_iopl_mask(unsigned mask) | |
690 | { | |
691 | struct physdev_set_iopl set_iopl; | |
692 | ||
693 | /* Force the change at ring 0. */ | |
694 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
695 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
696 | } | |
697 | ||
698 | static void xen_io_delay(void) | |
699 | { | |
700 | } | |
701 | ||
702 | #ifdef CONFIG_X86_LOCAL_APIC | |
ad66dd34 | 703 | static u32 xen_apic_read(u32 reg) |
5ead97c8 JF |
704 | { |
705 | return 0; | |
706 | } | |
f87e4cac | 707 | |
ad66dd34 | 708 | static void xen_apic_write(u32 reg, u32 val) |
f87e4cac JF |
709 | { |
710 | /* Warn to see if there's any stray references */ | |
711 | WARN_ON(1); | |
712 | } | |
ad66dd34 | 713 | |
ad66dd34 SS |
714 | static u64 xen_apic_icr_read(void) |
715 | { | |
716 | return 0; | |
717 | } | |
718 | ||
719 | static void xen_apic_icr_write(u32 low, u32 id) | |
720 | { | |
721 | /* Warn to see if there's any stray references */ | |
722 | WARN_ON(1); | |
723 | } | |
724 | ||
725 | static void xen_apic_wait_icr_idle(void) | |
726 | { | |
727 | return; | |
728 | } | |
729 | ||
94a8c3c2 YL |
730 | static u32 xen_safe_apic_wait_icr_idle(void) |
731 | { | |
732 | return 0; | |
733 | } | |
734 | ||
c1eeb2de YL |
735 | static void set_xen_basic_apic_ops(void) |
736 | { | |
737 | apic->read = xen_apic_read; | |
738 | apic->write = xen_apic_write; | |
739 | apic->icr_read = xen_apic_icr_read; | |
740 | apic->icr_write = xen_apic_icr_write; | |
741 | apic->wait_icr_idle = xen_apic_wait_icr_idle; | |
742 | apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle; | |
743 | } | |
ad66dd34 | 744 | |
5ead97c8 JF |
745 | #endif |
746 | ||
7b1333aa JF |
747 | static void xen_clts(void) |
748 | { | |
749 | struct multicall_space mcs; | |
750 | ||
751 | mcs = xen_mc_entry(0); | |
752 | ||
753 | MULTI_fpu_taskswitch(mcs.mc, 0); | |
754 | ||
755 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
756 | } | |
757 | ||
a789ed5f JF |
758 | static DEFINE_PER_CPU(unsigned long, xen_cr0_value); |
759 | ||
760 | static unsigned long xen_read_cr0(void) | |
761 | { | |
762 | unsigned long cr0 = percpu_read(xen_cr0_value); | |
763 | ||
764 | if (unlikely(cr0 == 0)) { | |
765 | cr0 = native_read_cr0(); | |
766 | percpu_write(xen_cr0_value, cr0); | |
767 | } | |
768 | ||
769 | return cr0; | |
770 | } | |
771 | ||
7b1333aa JF |
772 | static void xen_write_cr0(unsigned long cr0) |
773 | { | |
774 | struct multicall_space mcs; | |
775 | ||
a789ed5f JF |
776 | percpu_write(xen_cr0_value, cr0); |
777 | ||
7b1333aa JF |
778 | /* Only pay attention to cr0.TS; everything else is |
779 | ignored. */ | |
780 | mcs = xen_mc_entry(0); | |
781 | ||
782 | MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); | |
783 | ||
784 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
785 | } | |
786 | ||
5ead97c8 JF |
787 | static void xen_write_cr4(unsigned long cr4) |
788 | { | |
2956a351 JF |
789 | cr4 &= ~X86_CR4_PGE; |
790 | cr4 &= ~X86_CR4_PSE; | |
791 | ||
792 | native_write_cr4(cr4); | |
5ead97c8 JF |
793 | } |
794 | ||
1153968a JF |
795 | static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) |
796 | { | |
797 | int ret; | |
798 | ||
799 | ret = 0; | |
800 | ||
f63c2f24 | 801 | switch (msr) { |
1153968a JF |
802 | #ifdef CONFIG_X86_64 |
803 | unsigned which; | |
804 | u64 base; | |
805 | ||
806 | case MSR_FS_BASE: which = SEGBASE_FS; goto set; | |
807 | case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; | |
808 | case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; | |
809 | ||
810 | set: | |
811 | base = ((u64)high << 32) | low; | |
812 | if (HYPERVISOR_set_segment_base(which, base) != 0) | |
0cc0213e | 813 | ret = -EIO; |
1153968a JF |
814 | break; |
815 | #endif | |
d89961e2 JF |
816 | |
817 | case MSR_STAR: | |
818 | case MSR_CSTAR: | |
819 | case MSR_LSTAR: | |
820 | case MSR_SYSCALL_MASK: | |
821 | case MSR_IA32_SYSENTER_CS: | |
822 | case MSR_IA32_SYSENTER_ESP: | |
823 | case MSR_IA32_SYSENTER_EIP: | |
824 | /* Fast syscall setup is all done in hypercalls, so | |
825 | these are all ignored. Stub them out here to stop | |
826 | Xen console noise. */ | |
827 | break; | |
828 | ||
1153968a JF |
829 | default: |
830 | ret = native_write_msr_safe(msr, low, high); | |
831 | } | |
832 | ||
833 | return ret; | |
834 | } | |
835 | ||
0e91398f | 836 | void xen_setup_shared_info(void) |
5ead97c8 JF |
837 | { |
838 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
15664f96 JF |
839 | set_fixmap(FIX_PARAVIRT_BOOTMAP, |
840 | xen_start_info->shared_info); | |
841 | ||
842 | HYPERVISOR_shared_info = | |
843 | (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); | |
5ead97c8 JF |
844 | } else |
845 | HYPERVISOR_shared_info = | |
846 | (struct shared_info *)__va(xen_start_info->shared_info); | |
847 | ||
2e8fe719 JF |
848 | #ifndef CONFIG_SMP |
849 | /* In UP this is as good a place as any to set up shared info */ | |
850 | xen_setup_vcpu_info_placement(); | |
851 | #endif | |
d5edbc1f JF |
852 | |
853 | xen_setup_mfn_list_list(); | |
2e8fe719 JF |
854 | } |
855 | ||
60223a32 | 856 | /* This is called once we have the cpu_possible_map */ |
0e91398f | 857 | void xen_setup_vcpu_info_placement(void) |
60223a32 JF |
858 | { |
859 | int cpu; | |
860 | ||
861 | for_each_possible_cpu(cpu) | |
862 | xen_vcpu_setup(cpu); | |
863 | ||
864 | /* xen_vcpu_setup managed to place the vcpu_info within the | |
865 | percpu area for all cpus, so make use of it */ | |
866 | if (have_vcpu_info_placement) { | |
867 | printk(KERN_INFO "Xen: using vcpu_info placement\n"); | |
868 | ||
ecb93d1c JF |
869 | pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); |
870 | pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); | |
871 | pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); | |
872 | pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); | |
93b1eab3 | 873 | pv_mmu_ops.read_cr2 = xen_read_cr2_direct; |
60223a32 | 874 | } |
5ead97c8 JF |
875 | } |
876 | ||
ab144f5e AK |
877 | static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, |
878 | unsigned long addr, unsigned len) | |
6487673b JF |
879 | { |
880 | char *start, *end, *reloc; | |
881 | unsigned ret; | |
882 | ||
883 | start = end = reloc = NULL; | |
884 | ||
93b1eab3 JF |
885 | #define SITE(op, x) \ |
886 | case PARAVIRT_PATCH(op.x): \ | |
6487673b JF |
887 | if (have_vcpu_info_placement) { \ |
888 | start = (char *)xen_##x##_direct; \ | |
889 | end = xen_##x##_direct_end; \ | |
890 | reloc = xen_##x##_direct_reloc; \ | |
891 | } \ | |
892 | goto patch_site | |
893 | ||
894 | switch (type) { | |
93b1eab3 JF |
895 | SITE(pv_irq_ops, irq_enable); |
896 | SITE(pv_irq_ops, irq_disable); | |
897 | SITE(pv_irq_ops, save_fl); | |
898 | SITE(pv_irq_ops, restore_fl); | |
6487673b JF |
899 | #undef SITE |
900 | ||
901 | patch_site: | |
902 | if (start == NULL || (end-start) > len) | |
903 | goto default_patch; | |
904 | ||
ab144f5e | 905 | ret = paravirt_patch_insns(insnbuf, len, start, end); |
6487673b JF |
906 | |
907 | /* Note: because reloc is assigned from something that | |
908 | appears to be an array, gcc assumes it's non-null, | |
909 | but doesn't know its relationship with start and | |
910 | end. */ | |
911 | if (reloc > start && reloc < end) { | |
912 | int reloc_off = reloc - start; | |
ab144f5e AK |
913 | long *relocp = (long *)(insnbuf + reloc_off); |
914 | long delta = start - (char *)addr; | |
6487673b JF |
915 | |
916 | *relocp += delta; | |
917 | } | |
918 | break; | |
919 | ||
920 | default_patch: | |
921 | default: | |
ab144f5e AK |
922 | ret = paravirt_patch_default(type, clobbers, insnbuf, |
923 | addr, len); | |
6487673b JF |
924 | break; |
925 | } | |
926 | ||
927 | return ret; | |
928 | } | |
929 | ||
93b1eab3 | 930 | static const struct pv_info xen_info __initdata = { |
5ead97c8 JF |
931 | .paravirt_enabled = 1, |
932 | .shared_kernel_pmd = 0, | |
933 | ||
934 | .name = "Xen", | |
93b1eab3 | 935 | }; |
5ead97c8 | 936 | |
93b1eab3 | 937 | static const struct pv_init_ops xen_init_ops __initdata = { |
6487673b | 938 | .patch = xen_patch, |
93b1eab3 | 939 | }; |
5ead97c8 | 940 | |
93b1eab3 | 941 | static const struct pv_time_ops xen_time_ops __initdata = { |
8a22b999 | 942 | .sched_clock = xen_clocksource_read, |
93b1eab3 | 943 | }; |
15c84731 | 944 | |
93b1eab3 | 945 | static const struct pv_cpu_ops xen_cpu_ops __initdata = { |
5ead97c8 JF |
946 | .cpuid = xen_cpuid, |
947 | ||
948 | .set_debugreg = xen_set_debugreg, | |
949 | .get_debugreg = xen_get_debugreg, | |
950 | ||
7b1333aa | 951 | .clts = xen_clts, |
5ead97c8 | 952 | |
a789ed5f | 953 | .read_cr0 = xen_read_cr0, |
7b1333aa | 954 | .write_cr0 = xen_write_cr0, |
5ead97c8 | 955 | |
5ead97c8 JF |
956 | .read_cr4 = native_read_cr4, |
957 | .read_cr4_safe = native_read_cr4_safe, | |
958 | .write_cr4 = xen_write_cr4, | |
959 | ||
5ead97c8 JF |
960 | .wbinvd = native_wbinvd, |
961 | ||
962 | .read_msr = native_read_msr_safe, | |
1153968a | 963 | .write_msr = xen_write_msr_safe, |
5ead97c8 JF |
964 | .read_tsc = native_read_tsc, |
965 | .read_pmc = native_read_pmc, | |
966 | ||
81e103f1 | 967 | .iret = xen_iret, |
d75cd22f | 968 | .irq_enable_sysexit = xen_sysexit, |
6fcac6d3 JF |
969 | #ifdef CONFIG_X86_64 |
970 | .usergs_sysret32 = xen_sysret32, | |
971 | .usergs_sysret64 = xen_sysret64, | |
972 | #endif | |
5ead97c8 JF |
973 | |
974 | .load_tr_desc = paravirt_nop, | |
975 | .set_ldt = xen_set_ldt, | |
976 | .load_gdt = xen_load_gdt, | |
977 | .load_idt = xen_load_idt, | |
978 | .load_tls = xen_load_tls, | |
a8fc1089 EH |
979 | #ifdef CONFIG_X86_64 |
980 | .load_gs_index = xen_load_gs_index, | |
981 | #endif | |
5ead97c8 | 982 | |
38ffbe66 JF |
983 | .alloc_ldt = xen_alloc_ldt, |
984 | .free_ldt = xen_free_ldt, | |
985 | ||
5ead97c8 JF |
986 | .store_gdt = native_store_gdt, |
987 | .store_idt = native_store_idt, | |
988 | .store_tr = xen_store_tr, | |
989 | ||
990 | .write_ldt_entry = xen_write_ldt_entry, | |
991 | .write_gdt_entry = xen_write_gdt_entry, | |
992 | .write_idt_entry = xen_write_idt_entry, | |
faca6227 | 993 | .load_sp0 = xen_load_sp0, |
5ead97c8 JF |
994 | |
995 | .set_iopl_mask = xen_set_iopl_mask, | |
996 | .io_delay = xen_io_delay, | |
997 | ||
952d1d70 JF |
998 | /* Xen takes care of %gs when switching to usermode for us */ |
999 | .swapgs = paravirt_nop, | |
1000 | ||
224101ed JF |
1001 | .start_context_switch = paravirt_start_context_switch, |
1002 | .end_context_switch = xen_end_context_switch, | |
93b1eab3 JF |
1003 | }; |
1004 | ||
93b1eab3 | 1005 | static const struct pv_apic_ops xen_apic_ops __initdata = { |
5ead97c8 | 1006 | #ifdef CONFIG_X86_LOCAL_APIC |
5ead97c8 JF |
1007 | .startup_ipi_hook = paravirt_nop, |
1008 | #endif | |
93b1eab3 JF |
1009 | }; |
1010 | ||
fefa629a JF |
1011 | static void xen_reboot(int reason) |
1012 | { | |
349c709f JF |
1013 | struct sched_shutdown r = { .reason = reason }; |
1014 | ||
fefa629a JF |
1015 | #ifdef CONFIG_SMP |
1016 | smp_send_stop(); | |
1017 | #endif | |
1018 | ||
349c709f | 1019 | if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) |
fefa629a JF |
1020 | BUG(); |
1021 | } | |
1022 | ||
1023 | static void xen_restart(char *msg) | |
1024 | { | |
1025 | xen_reboot(SHUTDOWN_reboot); | |
1026 | } | |
1027 | ||
1028 | static void xen_emergency_restart(void) | |
1029 | { | |
1030 | xen_reboot(SHUTDOWN_reboot); | |
1031 | } | |
1032 | ||
1033 | static void xen_machine_halt(void) | |
1034 | { | |
1035 | xen_reboot(SHUTDOWN_poweroff); | |
1036 | } | |
1037 | ||
1038 | static void xen_crash_shutdown(struct pt_regs *regs) | |
1039 | { | |
1040 | xen_reboot(SHUTDOWN_crash); | |
1041 | } | |
1042 | ||
1043 | static const struct machine_ops __initdata xen_machine_ops = { | |
1044 | .restart = xen_restart, | |
1045 | .halt = xen_machine_halt, | |
1046 | .power_off = xen_machine_halt, | |
1047 | .shutdown = xen_machine_halt, | |
1048 | .crash_shutdown = xen_crash_shutdown, | |
1049 | .emergency_restart = xen_emergency_restart, | |
1050 | }; | |
1051 | ||
577eebea JF |
1052 | /* |
1053 | * Set up the GDT and segment registers for -fstack-protector. Until | |
1054 | * we do this, we have to be careful not to call any stack-protected | |
1055 | * function, which is most of the kernel. | |
1056 | */ | |
1057 | static void __init xen_setup_stackprotector(void) | |
1058 | { | |
1059 | pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot; | |
1060 | pv_cpu_ops.load_gdt = xen_load_gdt_boot; | |
1061 | ||
1062 | setup_stack_canary_segment(0); | |
1063 | switch_to_new_gdt(0); | |
1064 | ||
1065 | pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry; | |
1066 | pv_cpu_ops.load_gdt = xen_load_gdt; | |
1067 | } | |
1068 | ||
5ead97c8 JF |
1069 | /* First C function to be called on Xen boot */ |
1070 | asmlinkage void __init xen_start_kernel(void) | |
1071 | { | |
1072 | pgd_t *pgd; | |
1073 | ||
1074 | if (!xen_start_info) | |
1075 | return; | |
1076 | ||
6e833587 JF |
1077 | xen_domain_type = XEN_PV_DOMAIN; |
1078 | ||
5ead97c8 | 1079 | /* Install Xen paravirt ops */ |
93b1eab3 JF |
1080 | pv_info = xen_info; |
1081 | pv_init_ops = xen_init_ops; | |
1082 | pv_time_ops = xen_time_ops; | |
1083 | pv_cpu_ops = xen_cpu_ops; | |
93b1eab3 | 1084 | pv_apic_ops = xen_apic_ops; |
93b1eab3 | 1085 | |
6b18ae3e | 1086 | x86_init.resources.memory_setup = xen_memory_setup; |
42bbdb43 | 1087 | x86_init.oem.arch_setup = xen_arch_setup; |
6f30c1ac | 1088 | x86_init.oem.banner = xen_banner; |
845b3944 TG |
1089 | |
1090 | x86_init.timers.timer_init = xen_time_init; | |
736decac TG |
1091 | x86_init.timers.setup_percpu_clockev = x86_init_noop; |
1092 | x86_cpuinit.setup_percpu_clockev = x86_init_noop; | |
6b18ae3e | 1093 | |
2d826404 | 1094 | x86_platform.calibrate_tsc = xen_tsc_khz; |
7bd867df FT |
1095 | x86_platform.get_wallclock = xen_get_wallclock; |
1096 | x86_platform.set_wallclock = xen_set_wallclock; | |
93b1eab3 | 1097 | |
ce2eef33 | 1098 | /* |
577eebea | 1099 | * Set up some pagetable state before starting to set any ptes. |
ce2eef33 | 1100 | */ |
577eebea | 1101 | |
973df35e JF |
1102 | xen_init_mmu_ops(); |
1103 | ||
577eebea JF |
1104 | /* Prevent unwanted bits from being set in PTEs. */ |
1105 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
1106 | if (!xen_initial_domain()) | |
1107 | __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); | |
1108 | ||
1109 | __supported_pte_mask |= _PAGE_IOMAP; | |
1110 | ||
817a824b IC |
1111 | /* |
1112 | * Prevent page tables from being allocated in highmem, even | |
1113 | * if CONFIG_HIGHPTE is enabled. | |
1114 | */ | |
1115 | __userpte_alloc_gfp &= ~__GFP_HIGHMEM; | |
1116 | ||
b75fe4e5 | 1117 | /* Work out if we support NX */ |
4763ed4d | 1118 | x86_configure_nx(); |
b75fe4e5 | 1119 | |
577eebea JF |
1120 | xen_setup_features(); |
1121 | ||
1122 | /* Get mfn list */ | |
1123 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
1124 | xen_build_dynamic_phys_to_machine(); | |
1125 | ||
1126 | /* | |
1127 | * Set up kernel GDT and segment registers, mainly so that | |
1128 | * -fstack-protector code can be executed. | |
1129 | */ | |
1130 | xen_setup_stackprotector(); | |
0d1edf46 | 1131 | |
ce2eef33 | 1132 | xen_init_irq_ops(); |
e826fe1b JF |
1133 | xen_init_cpuid_mask(); |
1134 | ||
94a8c3c2 | 1135 | #ifdef CONFIG_X86_LOCAL_APIC |
ad66dd34 | 1136 | /* |
94a8c3c2 | 1137 | * set up the basic apic ops. |
ad66dd34 | 1138 | */ |
c1eeb2de | 1139 | set_xen_basic_apic_ops(); |
ad66dd34 | 1140 | #endif |
93b1eab3 | 1141 | |
e57778a1 JF |
1142 | if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { |
1143 | pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; | |
1144 | pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; | |
1145 | } | |
1146 | ||
fefa629a JF |
1147 | machine_ops = xen_machine_ops; |
1148 | ||
38341432 JF |
1149 | /* |
1150 | * The only reliable way to retain the initial address of the | |
1151 | * percpu gdt_page is to remember it here, so we can go and | |
1152 | * mark it RW later, when the initial percpu area is freed. | |
1153 | */ | |
1154 | xen_initial_gdt = &per_cpu(gdt_page, 0); | |
795f99b6 | 1155 | |
a9e7062d | 1156 | xen_smp_init(); |
5ead97c8 | 1157 | |
5ead97c8 JF |
1158 | pgd = (pgd_t *)xen_start_info->pt_base; |
1159 | ||
60223a32 | 1160 | /* Don't do the full vcpu_info placement stuff until we have a |
2e8fe719 | 1161 | possible map and a non-dummy shared_info. */ |
60223a32 | 1162 | per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; |
5ead97c8 | 1163 | |
55d80856 JF |
1164 | local_irq_disable(); |
1165 | early_boot_irqs_off(); | |
1166 | ||
084a2a4e | 1167 | xen_raw_console_write("mapping kernel into physical memory\n"); |
d114e198 | 1168 | pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages); |
5ead97c8 | 1169 | |
084a2a4e | 1170 | init_mm.pgd = pgd; |
5ead97c8 JF |
1171 | |
1172 | /* keep using Xen gdt for now; no urgent need to change it */ | |
1173 | ||
e68266b7 | 1174 | #ifdef CONFIG_X86_32 |
93b1eab3 | 1175 | pv_info.kernel_rpl = 1; |
5ead97c8 | 1176 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) |
93b1eab3 | 1177 | pv_info.kernel_rpl = 0; |
e68266b7 IC |
1178 | #else |
1179 | pv_info.kernel_rpl = 0; | |
1180 | #endif | |
5ead97c8 JF |
1181 | |
1182 | /* set the limit of our address space */ | |
fb1d8404 | 1183 | xen_reserve_top(); |
5ead97c8 | 1184 | |
7d087b68 | 1185 | #ifdef CONFIG_X86_32 |
5ead97c8 JF |
1186 | /* set up basic CPUID stuff */ |
1187 | cpu_detect(&new_cpu_data); | |
1188 | new_cpu_data.hard_math = 1; | |
d560bc61 | 1189 | new_cpu_data.wp_works_ok = 1; |
5ead97c8 | 1190 | new_cpu_data.x86_capability[0] = cpuid_edx(1); |
7d087b68 | 1191 | #endif |
5ead97c8 JF |
1192 | |
1193 | /* Poke various useful things into boot_params */ | |
30c82645 PA |
1194 | boot_params.hdr.type_of_loader = (9 << 4) | 0; |
1195 | boot_params.hdr.ramdisk_image = xen_start_info->mod_start | |
1196 | ? __pa(xen_start_info->mod_start) : 0; | |
1197 | boot_params.hdr.ramdisk_size = xen_start_info->mod_len; | |
b7c3c5c1 | 1198 | boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); |
5ead97c8 | 1199 | |
6e833587 | 1200 | if (!xen_initial_domain()) { |
83abc70a | 1201 | add_preferred_console("xenboot", 0, NULL); |
9e124fe1 | 1202 | add_preferred_console("tty", 0, NULL); |
b8c2d3df | 1203 | add_preferred_console("hvc", 0, NULL); |
5d990b62 CW |
1204 | } else { |
1205 | /* Make sure ACS will be enabled */ | |
1206 | pci_request_acs(); | |
9e124fe1 | 1207 | } |
5d990b62 | 1208 | |
b8c2d3df | 1209 | |
084a2a4e JF |
1210 | xen_raw_console_write("about to get started...\n"); |
1211 | ||
499d19b8 JF |
1212 | xen_setup_runstate_info(0); |
1213 | ||
5ead97c8 | 1214 | /* Start the world */ |
f5d36de0 | 1215 | #ifdef CONFIG_X86_32 |
f0d43100 | 1216 | i386_start_kernel(); |
f5d36de0 | 1217 | #else |
084a2a4e | 1218 | x86_64_start_reservations((char *)__pa_symbol(&boot_params)); |
f5d36de0 | 1219 | #endif |
5ead97c8 | 1220 | } |