]>
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> | |
23 | #include <linux/bootmem.h> | |
24 | #include <linux/module.h> | |
f4f97b3e JF |
25 | #include <linux/mm.h> |
26 | #include <linux/page-flags.h> | |
27 | #include <linux/highmem.h> | |
b8c2d3df | 28 | #include <linux/console.h> |
5ead97c8 JF |
29 | |
30 | #include <xen/interface/xen.h> | |
ecbf29cd | 31 | #include <xen/interface/version.h> |
5ead97c8 JF |
32 | #include <xen/interface/physdev.h> |
33 | #include <xen/interface/vcpu.h> | |
34 | #include <xen/features.h> | |
35 | #include <xen/page.h> | |
084a2a4e | 36 | #include <xen/hvc-console.h> |
5ead97c8 JF |
37 | |
38 | #include <asm/paravirt.h> | |
caf43bf7 | 39 | #include <asm/apic.h> |
5ead97c8 JF |
40 | #include <asm/page.h> |
41 | #include <asm/xen/hypercall.h> | |
42 | #include <asm/xen/hypervisor.h> | |
43 | #include <asm/fixmap.h> | |
44 | #include <asm/processor.h> | |
1153968a | 45 | #include <asm/msr-index.h> |
5ead97c8 JF |
46 | #include <asm/setup.h> |
47 | #include <asm/desc.h> | |
48 | #include <asm/pgtable.h> | |
f87e4cac | 49 | #include <asm/tlbflush.h> |
fefa629a | 50 | #include <asm/reboot.h> |
5ead97c8 JF |
51 | |
52 | #include "xen-ops.h" | |
3b827c1b | 53 | #include "mmu.h" |
5ead97c8 JF |
54 | #include "multicalls.h" |
55 | ||
56 | EXPORT_SYMBOL_GPL(hypercall_page); | |
57 | ||
5ead97c8 JF |
58 | DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); |
59 | DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); | |
9f79991d | 60 | |
6e833587 JF |
61 | enum xen_domain_type xen_domain_type = XEN_NATIVE; |
62 | EXPORT_SYMBOL_GPL(xen_domain_type); | |
63 | ||
5ead97c8 JF |
64 | struct start_info *xen_start_info; |
65 | EXPORT_SYMBOL_GPL(xen_start_info); | |
66 | ||
a0d695c8 | 67 | struct shared_info xen_dummy_shared_info; |
60223a32 | 68 | |
38341432 JF |
69 | void *xen_initial_gdt; |
70 | ||
60223a32 JF |
71 | /* |
72 | * Point at some empty memory to start with. We map the real shared_info | |
73 | * page as soon as fixmap is up and running. | |
74 | */ | |
a0d695c8 | 75 | struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; |
60223a32 JF |
76 | |
77 | /* | |
78 | * Flag to determine whether vcpu info placement is available on all | |
79 | * VCPUs. We assume it is to start with, and then set it to zero on | |
80 | * the first failure. This is because it can succeed on some VCPUs | |
81 | * and not others, since it can involve hypervisor memory allocation, | |
82 | * or because the guest failed to guarantee all the appropriate | |
83 | * constraints on all VCPUs (ie buffer can't cross a page boundary). | |
84 | * | |
85 | * Note that any particular CPU may be using a placed vcpu structure, | |
86 | * but we can only optimise if the all are. | |
87 | * | |
88 | * 0: not available, 1: available | |
89 | */ | |
db053b86 JF |
90 | static int have_vcpu_info_placement = |
91 | #ifdef CONFIG_X86_32 | |
92 | 1 | |
93 | #else | |
94 | 0 | |
95 | #endif | |
96 | ; | |
97 | ||
60223a32 | 98 | |
9c7a7942 | 99 | static void xen_vcpu_setup(int cpu) |
5ead97c8 | 100 | { |
60223a32 JF |
101 | struct vcpu_register_vcpu_info info; |
102 | int err; | |
103 | struct vcpu_info *vcpup; | |
104 | ||
a0d695c8 | 105 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
5ead97c8 | 106 | per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; |
60223a32 JF |
107 | |
108 | if (!have_vcpu_info_placement) | |
109 | return; /* already tested, not available */ | |
110 | ||
111 | vcpup = &per_cpu(xen_vcpu_info, cpu); | |
112 | ||
113 | info.mfn = virt_to_mfn(vcpup); | |
114 | info.offset = offset_in_page(vcpup); | |
115 | ||
e3d26976 | 116 | printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", |
60223a32 JF |
117 | cpu, vcpup, info.mfn, info.offset); |
118 | ||
119 | /* Check to see if the hypervisor will put the vcpu_info | |
120 | structure where we want it, which allows direct access via | |
121 | a percpu-variable. */ | |
122 | err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); | |
123 | ||
124 | if (err) { | |
125 | printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); | |
126 | have_vcpu_info_placement = 0; | |
127 | } else { | |
128 | /* This cpu is using the registered vcpu info, even if | |
129 | later ones fail to. */ | |
130 | per_cpu(xen_vcpu, cpu) = vcpup; | |
6487673b | 131 | |
60223a32 JF |
132 | printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", |
133 | cpu, vcpup); | |
134 | } | |
5ead97c8 JF |
135 | } |
136 | ||
9c7a7942 JF |
137 | /* |
138 | * On restore, set the vcpu placement up again. | |
139 | * If it fails, then we're in a bad state, since | |
140 | * we can't back out from using it... | |
141 | */ | |
142 | void xen_vcpu_restore(void) | |
143 | { | |
144 | if (have_vcpu_info_placement) { | |
145 | int cpu; | |
146 | ||
147 | for_each_online_cpu(cpu) { | |
148 | bool other_cpu = (cpu != smp_processor_id()); | |
149 | ||
150 | if (other_cpu && | |
151 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) | |
152 | BUG(); | |
153 | ||
154 | xen_vcpu_setup(cpu); | |
155 | ||
156 | if (other_cpu && | |
157 | HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) | |
158 | BUG(); | |
159 | } | |
160 | ||
161 | BUG_ON(!have_vcpu_info_placement); | |
162 | } | |
163 | } | |
164 | ||
5ead97c8 JF |
165 | static void __init xen_banner(void) |
166 | { | |
95c7c23b JF |
167 | unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); |
168 | struct xen_extraversion extra; | |
169 | HYPERVISOR_xen_version(XENVER_extraversion, &extra); | |
170 | ||
5ead97c8 | 171 | printk(KERN_INFO "Booting paravirtualized kernel on %s\n", |
93b1eab3 | 172 | pv_info.name); |
95c7c23b JF |
173 | printk(KERN_INFO "Xen version: %d.%d%s%s\n", |
174 | version >> 16, version & 0xffff, extra.extraversion, | |
e57778a1 | 175 | xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); |
5ead97c8 JF |
176 | } |
177 | ||
65ea5b03 PA |
178 | static void xen_cpuid(unsigned int *ax, unsigned int *bx, |
179 | unsigned int *cx, unsigned int *dx) | |
5ead97c8 JF |
180 | { |
181 | unsigned maskedx = ~0; | |
182 | ||
183 | /* | |
184 | * Mask out inconvenient features, to try and disable as many | |
185 | * unsupported kernel subsystems as possible. | |
186 | */ | |
65ea5b03 | 187 | if (*ax == 1) |
5ead97c8 JF |
188 | maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */ |
189 | (1 << X86_FEATURE_ACPI) | /* disable ACPI */ | |
dbe9e994 JF |
190 | (1 << X86_FEATURE_MCE) | /* disable MCE */ |
191 | (1 << X86_FEATURE_MCA) | /* disable MCA */ | |
5ead97c8 JF |
192 | (1 << X86_FEATURE_ACC)); /* thermal monitoring */ |
193 | ||
194 | asm(XEN_EMULATE_PREFIX "cpuid" | |
65ea5b03 PA |
195 | : "=a" (*ax), |
196 | "=b" (*bx), | |
197 | "=c" (*cx), | |
198 | "=d" (*dx) | |
199 | : "0" (*ax), "2" (*cx)); | |
200 | *dx &= maskedx; | |
5ead97c8 JF |
201 | } |
202 | ||
203 | static void xen_set_debugreg(int reg, unsigned long val) | |
204 | { | |
205 | HYPERVISOR_set_debugreg(reg, val); | |
206 | } | |
207 | ||
208 | static unsigned long xen_get_debugreg(int reg) | |
209 | { | |
210 | return HYPERVISOR_get_debugreg(reg); | |
211 | } | |
212 | ||
319f3ba5 | 213 | void xen_leave_lazy(void) |
5ead97c8 | 214 | { |
8965c1c0 | 215 | paravirt_leave_lazy(paravirt_get_lazy_mode()); |
5ead97c8 | 216 | xen_mc_flush(); |
5ead97c8 JF |
217 | } |
218 | ||
219 | static unsigned long xen_store_tr(void) | |
220 | { | |
221 | return 0; | |
222 | } | |
223 | ||
a05d2eba | 224 | /* |
cef43bf6 JF |
225 | * Set the page permissions for a particular virtual address. If the |
226 | * address is a vmalloc mapping (or other non-linear mapping), then | |
227 | * find the linear mapping of the page and also set its protections to | |
228 | * match. | |
a05d2eba JF |
229 | */ |
230 | static void set_aliased_prot(void *v, pgprot_t prot) | |
231 | { | |
232 | int level; | |
233 | pte_t *ptep; | |
234 | pte_t pte; | |
235 | unsigned long pfn; | |
236 | struct page *page; | |
237 | ||
238 | ptep = lookup_address((unsigned long)v, &level); | |
239 | BUG_ON(ptep == NULL); | |
240 | ||
241 | pfn = pte_pfn(*ptep); | |
242 | page = pfn_to_page(pfn); | |
243 | ||
244 | pte = pfn_pte(pfn, prot); | |
245 | ||
246 | if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) | |
247 | BUG(); | |
248 | ||
249 | if (!PageHighMem(page)) { | |
250 | void *av = __va(PFN_PHYS(pfn)); | |
251 | ||
252 | if (av != v) | |
253 | if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) | |
254 | BUG(); | |
255 | } else | |
256 | kmap_flush_unused(); | |
257 | } | |
258 | ||
38ffbe66 JF |
259 | static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) |
260 | { | |
a05d2eba | 261 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
262 | int i; |
263 | ||
a05d2eba JF |
264 | for(i = 0; i < entries; i += entries_per_page) |
265 | set_aliased_prot(ldt + i, PAGE_KERNEL_RO); | |
38ffbe66 JF |
266 | } |
267 | ||
268 | static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) | |
269 | { | |
a05d2eba | 270 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
271 | int i; |
272 | ||
a05d2eba JF |
273 | for(i = 0; i < entries; i += entries_per_page) |
274 | set_aliased_prot(ldt + i, PAGE_KERNEL); | |
38ffbe66 JF |
275 | } |
276 | ||
5ead97c8 JF |
277 | static void xen_set_ldt(const void *addr, unsigned entries) |
278 | { | |
5ead97c8 JF |
279 | struct mmuext_op *op; |
280 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
281 | ||
282 | op = mcs.args; | |
283 | op->cmd = MMUEXT_SET_LDT; | |
4dbf7af6 | 284 | op->arg1.linear_addr = (unsigned long)addr; |
5ead97c8 JF |
285 | op->arg2.nr_ents = entries; |
286 | ||
287 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
288 | ||
289 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
290 | } | |
291 | ||
6b68f01b | 292 | static void xen_load_gdt(const struct desc_ptr *dtr) |
5ead97c8 JF |
293 | { |
294 | unsigned long *frames; | |
295 | unsigned long va = dtr->address; | |
296 | unsigned int size = dtr->size + 1; | |
297 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
298 | int f; | |
299 | struct multicall_space mcs; | |
300 | ||
301 | /* A GDT can be up to 64k in size, which corresponds to 8192 | |
302 | 8-byte entries, or 16 4k pages.. */ | |
303 | ||
304 | BUG_ON(size > 65536); | |
305 | BUG_ON(va & ~PAGE_MASK); | |
306 | ||
307 | mcs = xen_mc_entry(sizeof(*frames) * pages); | |
308 | frames = mcs.args; | |
309 | ||
310 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
311 | frames[f] = virt_to_mfn(va); | |
312 | make_lowmem_page_readonly((void *)va); | |
313 | } | |
314 | ||
315 | MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct)); | |
316 | ||
317 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
318 | } | |
319 | ||
320 | static void load_TLS_descriptor(struct thread_struct *t, | |
321 | unsigned int cpu, unsigned int i) | |
322 | { | |
323 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
324 | xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); | |
325 | struct multicall_space mc = __xen_mc_entry(0); | |
326 | ||
327 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
328 | } | |
329 | ||
330 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
331 | { | |
8b84ad94 JF |
332 | /* |
333 | * XXX sleazy hack: If we're being called in a lazy-cpu zone, | |
334 | * it means we're in a context switch, and %gs has just been | |
335 | * saved. This means we can zero it out to prevent faults on | |
336 | * exit from the hypervisor if the next process has no %gs. | |
337 | * Either way, it has been saved, and the new value will get | |
338 | * loaded properly. This will go away as soon as Xen has been | |
339 | * modified to not save/restore %gs for normal hypercalls. | |
8a95408e EH |
340 | * |
341 | * On x86_64, this hack is not used for %gs, because gs points | |
342 | * to KERNEL_GS_BASE (and uses it for PDA references), so we | |
343 | * must not zero %gs on x86_64 | |
344 | * | |
345 | * For x86_64, we need to zero %fs, otherwise we may get an | |
346 | * exception between the new %fs descriptor being loaded and | |
347 | * %fs being effectively cleared at __switch_to(). | |
8b84ad94 | 348 | */ |
8a95408e EH |
349 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { |
350 | #ifdef CONFIG_X86_32 | |
8b84ad94 | 351 | loadsegment(gs, 0); |
8a95408e EH |
352 | #else |
353 | loadsegment(fs, 0); | |
354 | #endif | |
355 | } | |
356 | ||
357 | xen_mc_batch(); | |
358 | ||
359 | load_TLS_descriptor(t, cpu, 0); | |
360 | load_TLS_descriptor(t, cpu, 1); | |
361 | load_TLS_descriptor(t, cpu, 2); | |
362 | ||
363 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
5ead97c8 JF |
364 | } |
365 | ||
a8fc1089 EH |
366 | #ifdef CONFIG_X86_64 |
367 | static void xen_load_gs_index(unsigned int idx) | |
368 | { | |
369 | if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) | |
370 | BUG(); | |
5ead97c8 | 371 | } |
a8fc1089 | 372 | #endif |
5ead97c8 JF |
373 | |
374 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
75b8bb3e | 375 | const void *ptr) |
5ead97c8 | 376 | { |
cef43bf6 | 377 | xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); |
75b8bb3e | 378 | u64 entry = *(u64 *)ptr; |
5ead97c8 | 379 | |
f120f13e JF |
380 | preempt_disable(); |
381 | ||
5ead97c8 JF |
382 | xen_mc_flush(); |
383 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
384 | BUG(); | |
f120f13e JF |
385 | |
386 | preempt_enable(); | |
5ead97c8 JF |
387 | } |
388 | ||
e176d367 | 389 | static int cvt_gate_to_trap(int vector, const gate_desc *val, |
5ead97c8 JF |
390 | struct trap_info *info) |
391 | { | |
e176d367 | 392 | if (val->type != 0xf && val->type != 0xe) |
5ead97c8 JF |
393 | return 0; |
394 | ||
395 | info->vector = vector; | |
e176d367 EH |
396 | info->address = gate_offset(*val); |
397 | info->cs = gate_segment(*val); | |
398 | info->flags = val->dpl; | |
5ead97c8 | 399 | /* interrupt gates clear IF */ |
e176d367 | 400 | if (val->type == 0xe) |
5ead97c8 JF |
401 | info->flags |= 4; |
402 | ||
403 | return 1; | |
404 | } | |
405 | ||
406 | /* Locations of each CPU's IDT */ | |
6b68f01b | 407 | static DEFINE_PER_CPU(struct desc_ptr, idt_desc); |
5ead97c8 JF |
408 | |
409 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
410 | also update Xen. */ | |
8d947344 | 411 | static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) |
5ead97c8 | 412 | { |
5ead97c8 | 413 | unsigned long p = (unsigned long)&dt[entrynum]; |
f120f13e JF |
414 | unsigned long start, end; |
415 | ||
416 | preempt_disable(); | |
417 | ||
418 | start = __get_cpu_var(idt_desc).address; | |
419 | end = start + __get_cpu_var(idt_desc).size + 1; | |
5ead97c8 JF |
420 | |
421 | xen_mc_flush(); | |
422 | ||
8d947344 | 423 | native_write_idt_entry(dt, entrynum, g); |
5ead97c8 JF |
424 | |
425 | if (p >= start && (p + 8) <= end) { | |
426 | struct trap_info info[2]; | |
427 | ||
428 | info[1].address = 0; | |
429 | ||
e176d367 | 430 | if (cvt_gate_to_trap(entrynum, g, &info[0])) |
5ead97c8 JF |
431 | if (HYPERVISOR_set_trap_table(info)) |
432 | BUG(); | |
433 | } | |
f120f13e JF |
434 | |
435 | preempt_enable(); | |
5ead97c8 JF |
436 | } |
437 | ||
6b68f01b | 438 | static void xen_convert_trap_info(const struct desc_ptr *desc, |
f87e4cac | 439 | struct trap_info *traps) |
5ead97c8 | 440 | { |
5ead97c8 JF |
441 | unsigned in, out, count; |
442 | ||
e176d367 | 443 | count = (desc->size+1) / sizeof(gate_desc); |
5ead97c8 JF |
444 | BUG_ON(count > 256); |
445 | ||
5ead97c8 | 446 | for (in = out = 0; in < count; in++) { |
e176d367 | 447 | gate_desc *entry = (gate_desc*)(desc->address) + in; |
5ead97c8 | 448 | |
e176d367 | 449 | if (cvt_gate_to_trap(in, entry, &traps[out])) |
5ead97c8 JF |
450 | out++; |
451 | } | |
452 | traps[out].address = 0; | |
f87e4cac JF |
453 | } |
454 | ||
455 | void xen_copy_trap_info(struct trap_info *traps) | |
456 | { | |
6b68f01b | 457 | const struct desc_ptr *desc = &__get_cpu_var(idt_desc); |
f87e4cac JF |
458 | |
459 | xen_convert_trap_info(desc, traps); | |
f87e4cac JF |
460 | } |
461 | ||
462 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
463 | hold a spinlock to protect the static traps[] array (static because | |
464 | it avoids allocation, and saves stack space). */ | |
6b68f01b | 465 | static void xen_load_idt(const struct desc_ptr *desc) |
f87e4cac JF |
466 | { |
467 | static DEFINE_SPINLOCK(lock); | |
468 | static struct trap_info traps[257]; | |
f87e4cac JF |
469 | |
470 | spin_lock(&lock); | |
471 | ||
f120f13e JF |
472 | __get_cpu_var(idt_desc) = *desc; |
473 | ||
f87e4cac | 474 | xen_convert_trap_info(desc, traps); |
5ead97c8 JF |
475 | |
476 | xen_mc_flush(); | |
477 | if (HYPERVISOR_set_trap_table(traps)) | |
478 | BUG(); | |
479 | ||
480 | spin_unlock(&lock); | |
481 | } | |
482 | ||
483 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
484 | they're handled differently. */ | |
485 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
014b15be | 486 | const void *desc, int type) |
5ead97c8 | 487 | { |
f120f13e JF |
488 | preempt_disable(); |
489 | ||
014b15be GOC |
490 | switch (type) { |
491 | case DESC_LDT: | |
492 | case DESC_TSS: | |
5ead97c8 JF |
493 | /* ignore */ |
494 | break; | |
495 | ||
496 | default: { | |
497 | xmaddr_t maddr = virt_to_machine(&dt[entry]); | |
5ead97c8 JF |
498 | |
499 | xen_mc_flush(); | |
014b15be | 500 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) |
5ead97c8 JF |
501 | BUG(); |
502 | } | |
503 | ||
504 | } | |
f120f13e JF |
505 | |
506 | preempt_enable(); | |
5ead97c8 JF |
507 | } |
508 | ||
faca6227 | 509 | static void xen_load_sp0(struct tss_struct *tss, |
a05d2eba | 510 | struct thread_struct *thread) |
5ead97c8 JF |
511 | { |
512 | struct multicall_space mcs = xen_mc_entry(0); | |
faca6227 | 513 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); |
5ead97c8 JF |
514 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
515 | } | |
516 | ||
517 | static void xen_set_iopl_mask(unsigned mask) | |
518 | { | |
519 | struct physdev_set_iopl set_iopl; | |
520 | ||
521 | /* Force the change at ring 0. */ | |
522 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
523 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
524 | } | |
525 | ||
526 | static void xen_io_delay(void) | |
527 | { | |
528 | } | |
529 | ||
530 | #ifdef CONFIG_X86_LOCAL_APIC | |
ad66dd34 | 531 | static u32 xen_apic_read(u32 reg) |
5ead97c8 JF |
532 | { |
533 | return 0; | |
534 | } | |
f87e4cac | 535 | |
ad66dd34 | 536 | static void xen_apic_write(u32 reg, u32 val) |
f87e4cac JF |
537 | { |
538 | /* Warn to see if there's any stray references */ | |
539 | WARN_ON(1); | |
540 | } | |
ad66dd34 | 541 | |
ad66dd34 SS |
542 | static u64 xen_apic_icr_read(void) |
543 | { | |
544 | return 0; | |
545 | } | |
546 | ||
547 | static void xen_apic_icr_write(u32 low, u32 id) | |
548 | { | |
549 | /* Warn to see if there's any stray references */ | |
550 | WARN_ON(1); | |
551 | } | |
552 | ||
553 | static void xen_apic_wait_icr_idle(void) | |
554 | { | |
555 | return; | |
556 | } | |
557 | ||
94a8c3c2 YL |
558 | static u32 xen_safe_apic_wait_icr_idle(void) |
559 | { | |
560 | return 0; | |
561 | } | |
562 | ||
ad66dd34 SS |
563 | static struct apic_ops xen_basic_apic_ops = { |
564 | .read = xen_apic_read, | |
565 | .write = xen_apic_write, | |
ad66dd34 SS |
566 | .icr_read = xen_apic_icr_read, |
567 | .icr_write = xen_apic_icr_write, | |
568 | .wait_icr_idle = xen_apic_wait_icr_idle, | |
94a8c3c2 | 569 | .safe_wait_icr_idle = xen_safe_apic_wait_icr_idle, |
ad66dd34 | 570 | }; |
ad66dd34 | 571 | |
5ead97c8 JF |
572 | #endif |
573 | ||
f87e4cac | 574 | |
7b1333aa JF |
575 | static void xen_clts(void) |
576 | { | |
577 | struct multicall_space mcs; | |
578 | ||
579 | mcs = xen_mc_entry(0); | |
580 | ||
581 | MULTI_fpu_taskswitch(mcs.mc, 0); | |
582 | ||
583 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
584 | } | |
585 | ||
586 | static void xen_write_cr0(unsigned long cr0) | |
587 | { | |
588 | struct multicall_space mcs; | |
589 | ||
590 | /* Only pay attention to cr0.TS; everything else is | |
591 | ignored. */ | |
592 | mcs = xen_mc_entry(0); | |
593 | ||
594 | MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); | |
595 | ||
596 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
597 | } | |
598 | ||
5ead97c8 JF |
599 | static void xen_write_cr4(unsigned long cr4) |
600 | { | |
2956a351 JF |
601 | cr4 &= ~X86_CR4_PGE; |
602 | cr4 &= ~X86_CR4_PSE; | |
603 | ||
604 | native_write_cr4(cr4); | |
5ead97c8 JF |
605 | } |
606 | ||
1153968a JF |
607 | static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) |
608 | { | |
609 | int ret; | |
610 | ||
611 | ret = 0; | |
612 | ||
f63c2f24 | 613 | switch (msr) { |
1153968a JF |
614 | #ifdef CONFIG_X86_64 |
615 | unsigned which; | |
616 | u64 base; | |
617 | ||
618 | case MSR_FS_BASE: which = SEGBASE_FS; goto set; | |
619 | case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; | |
620 | case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; | |
621 | ||
622 | set: | |
623 | base = ((u64)high << 32) | low; | |
624 | if (HYPERVISOR_set_segment_base(which, base) != 0) | |
625 | ret = -EFAULT; | |
626 | break; | |
627 | #endif | |
d89961e2 JF |
628 | |
629 | case MSR_STAR: | |
630 | case MSR_CSTAR: | |
631 | case MSR_LSTAR: | |
632 | case MSR_SYSCALL_MASK: | |
633 | case MSR_IA32_SYSENTER_CS: | |
634 | case MSR_IA32_SYSENTER_ESP: | |
635 | case MSR_IA32_SYSENTER_EIP: | |
636 | /* Fast syscall setup is all done in hypercalls, so | |
637 | these are all ignored. Stub them out here to stop | |
638 | Xen console noise. */ | |
639 | break; | |
640 | ||
1153968a JF |
641 | default: |
642 | ret = native_write_msr_safe(msr, low, high); | |
643 | } | |
644 | ||
645 | return ret; | |
646 | } | |
647 | ||
0e91398f | 648 | void xen_setup_shared_info(void) |
5ead97c8 JF |
649 | { |
650 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
15664f96 JF |
651 | set_fixmap(FIX_PARAVIRT_BOOTMAP, |
652 | xen_start_info->shared_info); | |
653 | ||
654 | HYPERVISOR_shared_info = | |
655 | (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); | |
5ead97c8 JF |
656 | } else |
657 | HYPERVISOR_shared_info = | |
658 | (struct shared_info *)__va(xen_start_info->shared_info); | |
659 | ||
2e8fe719 JF |
660 | #ifndef CONFIG_SMP |
661 | /* In UP this is as good a place as any to set up shared info */ | |
662 | xen_setup_vcpu_info_placement(); | |
663 | #endif | |
d5edbc1f JF |
664 | |
665 | xen_setup_mfn_list_list(); | |
2e8fe719 JF |
666 | } |
667 | ||
60223a32 | 668 | /* This is called once we have the cpu_possible_map */ |
0e91398f | 669 | void xen_setup_vcpu_info_placement(void) |
60223a32 JF |
670 | { |
671 | int cpu; | |
672 | ||
673 | for_each_possible_cpu(cpu) | |
674 | xen_vcpu_setup(cpu); | |
675 | ||
676 | /* xen_vcpu_setup managed to place the vcpu_info within the | |
677 | percpu area for all cpus, so make use of it */ | |
678 | if (have_vcpu_info_placement) { | |
679 | printk(KERN_INFO "Xen: using vcpu_info placement\n"); | |
680 | ||
ecb93d1c JF |
681 | pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); |
682 | pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); | |
683 | pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); | |
684 | pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); | |
93b1eab3 | 685 | pv_mmu_ops.read_cr2 = xen_read_cr2_direct; |
60223a32 | 686 | } |
5ead97c8 JF |
687 | } |
688 | ||
ab144f5e AK |
689 | static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, |
690 | unsigned long addr, unsigned len) | |
6487673b JF |
691 | { |
692 | char *start, *end, *reloc; | |
693 | unsigned ret; | |
694 | ||
695 | start = end = reloc = NULL; | |
696 | ||
93b1eab3 JF |
697 | #define SITE(op, x) \ |
698 | case PARAVIRT_PATCH(op.x): \ | |
6487673b JF |
699 | if (have_vcpu_info_placement) { \ |
700 | start = (char *)xen_##x##_direct; \ | |
701 | end = xen_##x##_direct_end; \ | |
702 | reloc = xen_##x##_direct_reloc; \ | |
703 | } \ | |
704 | goto patch_site | |
705 | ||
706 | switch (type) { | |
93b1eab3 JF |
707 | SITE(pv_irq_ops, irq_enable); |
708 | SITE(pv_irq_ops, irq_disable); | |
709 | SITE(pv_irq_ops, save_fl); | |
710 | SITE(pv_irq_ops, restore_fl); | |
6487673b JF |
711 | #undef SITE |
712 | ||
713 | patch_site: | |
714 | if (start == NULL || (end-start) > len) | |
715 | goto default_patch; | |
716 | ||
ab144f5e | 717 | ret = paravirt_patch_insns(insnbuf, len, start, end); |
6487673b JF |
718 | |
719 | /* Note: because reloc is assigned from something that | |
720 | appears to be an array, gcc assumes it's non-null, | |
721 | but doesn't know its relationship with start and | |
722 | end. */ | |
723 | if (reloc > start && reloc < end) { | |
724 | int reloc_off = reloc - start; | |
ab144f5e AK |
725 | long *relocp = (long *)(insnbuf + reloc_off); |
726 | long delta = start - (char *)addr; | |
6487673b JF |
727 | |
728 | *relocp += delta; | |
729 | } | |
730 | break; | |
731 | ||
732 | default_patch: | |
733 | default: | |
ab144f5e AK |
734 | ret = paravirt_patch_default(type, clobbers, insnbuf, |
735 | addr, len); | |
6487673b JF |
736 | break; |
737 | } | |
738 | ||
739 | return ret; | |
740 | } | |
741 | ||
93b1eab3 | 742 | static const struct pv_info xen_info __initdata = { |
5ead97c8 JF |
743 | .paravirt_enabled = 1, |
744 | .shared_kernel_pmd = 0, | |
745 | ||
746 | .name = "Xen", | |
93b1eab3 | 747 | }; |
5ead97c8 | 748 | |
93b1eab3 | 749 | static const struct pv_init_ops xen_init_ops __initdata = { |
6487673b | 750 | .patch = xen_patch, |
5ead97c8 | 751 | |
93b1eab3 | 752 | .banner = xen_banner, |
5ead97c8 JF |
753 | .memory_setup = xen_memory_setup, |
754 | .arch_setup = xen_arch_setup, | |
e2426cf8 | 755 | .post_allocator_init = xen_post_allocator_init, |
93b1eab3 | 756 | }; |
5ead97c8 | 757 | |
93b1eab3 | 758 | static const struct pv_time_ops xen_time_ops __initdata = { |
15c84731 | 759 | .time_init = xen_time_init, |
93b1eab3 | 760 | |
15c84731 JF |
761 | .set_wallclock = xen_set_wallclock, |
762 | .get_wallclock = xen_get_wallclock, | |
e93ef949 | 763 | .get_tsc_khz = xen_tsc_khz, |
ab550288 | 764 | .sched_clock = xen_sched_clock, |
93b1eab3 | 765 | }; |
15c84731 | 766 | |
93b1eab3 | 767 | static const struct pv_cpu_ops xen_cpu_ops __initdata = { |
5ead97c8 JF |
768 | .cpuid = xen_cpuid, |
769 | ||
770 | .set_debugreg = xen_set_debugreg, | |
771 | .get_debugreg = xen_get_debugreg, | |
772 | ||
7b1333aa | 773 | .clts = xen_clts, |
5ead97c8 JF |
774 | |
775 | .read_cr0 = native_read_cr0, | |
7b1333aa | 776 | .write_cr0 = xen_write_cr0, |
5ead97c8 | 777 | |
5ead97c8 JF |
778 | .read_cr4 = native_read_cr4, |
779 | .read_cr4_safe = native_read_cr4_safe, | |
780 | .write_cr4 = xen_write_cr4, | |
781 | ||
5ead97c8 JF |
782 | .wbinvd = native_wbinvd, |
783 | ||
784 | .read_msr = native_read_msr_safe, | |
1153968a | 785 | .write_msr = xen_write_msr_safe, |
5ead97c8 JF |
786 | .read_tsc = native_read_tsc, |
787 | .read_pmc = native_read_pmc, | |
788 | ||
81e103f1 | 789 | .iret = xen_iret, |
d75cd22f | 790 | .irq_enable_sysexit = xen_sysexit, |
6fcac6d3 JF |
791 | #ifdef CONFIG_X86_64 |
792 | .usergs_sysret32 = xen_sysret32, | |
793 | .usergs_sysret64 = xen_sysret64, | |
794 | #endif | |
5ead97c8 JF |
795 | |
796 | .load_tr_desc = paravirt_nop, | |
797 | .set_ldt = xen_set_ldt, | |
798 | .load_gdt = xen_load_gdt, | |
799 | .load_idt = xen_load_idt, | |
800 | .load_tls = xen_load_tls, | |
a8fc1089 EH |
801 | #ifdef CONFIG_X86_64 |
802 | .load_gs_index = xen_load_gs_index, | |
803 | #endif | |
5ead97c8 | 804 | |
38ffbe66 JF |
805 | .alloc_ldt = xen_alloc_ldt, |
806 | .free_ldt = xen_free_ldt, | |
807 | ||
5ead97c8 JF |
808 | .store_gdt = native_store_gdt, |
809 | .store_idt = native_store_idt, | |
810 | .store_tr = xen_store_tr, | |
811 | ||
812 | .write_ldt_entry = xen_write_ldt_entry, | |
813 | .write_gdt_entry = xen_write_gdt_entry, | |
814 | .write_idt_entry = xen_write_idt_entry, | |
faca6227 | 815 | .load_sp0 = xen_load_sp0, |
5ead97c8 JF |
816 | |
817 | .set_iopl_mask = xen_set_iopl_mask, | |
818 | .io_delay = xen_io_delay, | |
819 | ||
952d1d70 JF |
820 | /* Xen takes care of %gs when switching to usermode for us */ |
821 | .swapgs = paravirt_nop, | |
822 | ||
8965c1c0 JF |
823 | .lazy_mode = { |
824 | .enter = paravirt_enter_lazy_cpu, | |
825 | .leave = xen_leave_lazy, | |
826 | }, | |
93b1eab3 JF |
827 | }; |
828 | ||
93b1eab3 | 829 | static const struct pv_apic_ops xen_apic_ops __initdata = { |
5ead97c8 | 830 | #ifdef CONFIG_X86_LOCAL_APIC |
5ead97c8 JF |
831 | .setup_boot_clock = paravirt_nop, |
832 | .setup_secondary_clock = paravirt_nop, | |
833 | .startup_ipi_hook = paravirt_nop, | |
834 | #endif | |
93b1eab3 JF |
835 | }; |
836 | ||
fefa629a JF |
837 | static void xen_reboot(int reason) |
838 | { | |
349c709f JF |
839 | struct sched_shutdown r = { .reason = reason }; |
840 | ||
fefa629a JF |
841 | #ifdef CONFIG_SMP |
842 | smp_send_stop(); | |
843 | #endif | |
844 | ||
349c709f | 845 | if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) |
fefa629a JF |
846 | BUG(); |
847 | } | |
848 | ||
849 | static void xen_restart(char *msg) | |
850 | { | |
851 | xen_reboot(SHUTDOWN_reboot); | |
852 | } | |
853 | ||
854 | static void xen_emergency_restart(void) | |
855 | { | |
856 | xen_reboot(SHUTDOWN_reboot); | |
857 | } | |
858 | ||
859 | static void xen_machine_halt(void) | |
860 | { | |
861 | xen_reboot(SHUTDOWN_poweroff); | |
862 | } | |
863 | ||
864 | static void xen_crash_shutdown(struct pt_regs *regs) | |
865 | { | |
866 | xen_reboot(SHUTDOWN_crash); | |
867 | } | |
868 | ||
869 | static const struct machine_ops __initdata xen_machine_ops = { | |
870 | .restart = xen_restart, | |
871 | .halt = xen_machine_halt, | |
872 | .power_off = xen_machine_halt, | |
873 | .shutdown = xen_machine_halt, | |
874 | .crash_shutdown = xen_crash_shutdown, | |
875 | .emergency_restart = xen_emergency_restart, | |
876 | }; | |
877 | ||
6487673b | 878 | |
5ead97c8 JF |
879 | /* First C function to be called on Xen boot */ |
880 | asmlinkage void __init xen_start_kernel(void) | |
881 | { | |
882 | pgd_t *pgd; | |
883 | ||
884 | if (!xen_start_info) | |
885 | return; | |
886 | ||
6e833587 JF |
887 | xen_domain_type = XEN_PV_DOMAIN; |
888 | ||
7999f4b4 | 889 | BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0); |
5ead97c8 | 890 | |
e57778a1 JF |
891 | xen_setup_features(); |
892 | ||
5ead97c8 | 893 | /* Install Xen paravirt ops */ |
93b1eab3 JF |
894 | pv_info = xen_info; |
895 | pv_init_ops = xen_init_ops; | |
896 | pv_time_ops = xen_time_ops; | |
897 | pv_cpu_ops = xen_cpu_ops; | |
93b1eab3 JF |
898 | pv_apic_ops = xen_apic_ops; |
899 | pv_mmu_ops = xen_mmu_ops; | |
93b1eab3 | 900 | |
0d1edf46 JF |
901 | xen_init_irq_ops(); |
902 | ||
94a8c3c2 | 903 | #ifdef CONFIG_X86_LOCAL_APIC |
ad66dd34 | 904 | /* |
94a8c3c2 | 905 | * set up the basic apic ops. |
ad66dd34 SS |
906 | */ |
907 | apic_ops = &xen_basic_apic_ops; | |
908 | #endif | |
93b1eab3 | 909 | |
e57778a1 JF |
910 | if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { |
911 | pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; | |
912 | pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; | |
913 | } | |
914 | ||
fefa629a JF |
915 | machine_ops = xen_machine_ops; |
916 | ||
f5d36de0 JF |
917 | #ifdef CONFIG_X86_64 |
918 | /* Disable until direct per-cpu data access. */ | |
919 | have_vcpu_info_placement = 0; | |
f87e4cac | 920 | #endif |
5ead97c8 | 921 | |
38341432 JF |
922 | #ifdef CONFIG_X86_64 |
923 | /* | |
924 | * Setup percpu state. We only need to do this for 64-bit | |
925 | * because 32-bit already has %fs set properly. | |
926 | */ | |
795f99b6 | 927 | load_percpu_segment(0); |
38341432 JF |
928 | #endif |
929 | /* | |
930 | * The only reliable way to retain the initial address of the | |
931 | * percpu gdt_page is to remember it here, so we can go and | |
932 | * mark it RW later, when the initial percpu area is freed. | |
933 | */ | |
934 | xen_initial_gdt = &per_cpu(gdt_page, 0); | |
795f99b6 | 935 | |
a9e7062d | 936 | xen_smp_init(); |
5ead97c8 JF |
937 | |
938 | /* Get mfn list */ | |
939 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
d451bb7a | 940 | xen_build_dynamic_phys_to_machine(); |
5ead97c8 JF |
941 | |
942 | pgd = (pgd_t *)xen_start_info->pt_base; | |
943 | ||
084a2a4e JF |
944 | /* Prevent unwanted bits from being set in PTEs. */ |
945 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
6e833587 | 946 | if (!xen_initial_domain()) |
084a2a4e | 947 | __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); |
60223a32 | 948 | |
60223a32 | 949 | /* Don't do the full vcpu_info placement stuff until we have a |
2e8fe719 | 950 | possible map and a non-dummy shared_info. */ |
60223a32 | 951 | per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; |
5ead97c8 | 952 | |
084a2a4e | 953 | xen_raw_console_write("mapping kernel into physical memory\n"); |
d114e198 | 954 | pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages); |
5ead97c8 | 955 | |
084a2a4e | 956 | init_mm.pgd = pgd; |
5ead97c8 JF |
957 | |
958 | /* keep using Xen gdt for now; no urgent need to change it */ | |
959 | ||
93b1eab3 | 960 | pv_info.kernel_rpl = 1; |
5ead97c8 | 961 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) |
93b1eab3 | 962 | pv_info.kernel_rpl = 0; |
5ead97c8 JF |
963 | |
964 | /* set the limit of our address space */ | |
fb1d8404 | 965 | xen_reserve_top(); |
5ead97c8 | 966 | |
7d087b68 | 967 | #ifdef CONFIG_X86_32 |
5ead97c8 JF |
968 | /* set up basic CPUID stuff */ |
969 | cpu_detect(&new_cpu_data); | |
970 | new_cpu_data.hard_math = 1; | |
971 | new_cpu_data.x86_capability[0] = cpuid_edx(1); | |
7d087b68 | 972 | #endif |
5ead97c8 JF |
973 | |
974 | /* Poke various useful things into boot_params */ | |
30c82645 PA |
975 | boot_params.hdr.type_of_loader = (9 << 4) | 0; |
976 | boot_params.hdr.ramdisk_image = xen_start_info->mod_start | |
977 | ? __pa(xen_start_info->mod_start) : 0; | |
978 | boot_params.hdr.ramdisk_size = xen_start_info->mod_len; | |
b7c3c5c1 | 979 | boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); |
5ead97c8 | 980 | |
6e833587 | 981 | if (!xen_initial_domain()) { |
83abc70a | 982 | add_preferred_console("xenboot", 0, NULL); |
9e124fe1 | 983 | add_preferred_console("tty", 0, NULL); |
b8c2d3df | 984 | add_preferred_console("hvc", 0, NULL); |
9e124fe1 | 985 | } |
b8c2d3df | 986 | |
084a2a4e JF |
987 | xen_raw_console_write("about to get started...\n"); |
988 | ||
5ead97c8 | 989 | /* Start the world */ |
f5d36de0 | 990 | #ifdef CONFIG_X86_32 |
f0d43100 | 991 | i386_start_kernel(); |
f5d36de0 | 992 | #else |
084a2a4e | 993 | x86_64_start_reservations((char *)__pa_symbol(&boot_params)); |
f5d36de0 | 994 | #endif |
5ead97c8 | 995 | } |