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> | |
7b6aa335 | 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 | */ | |
e4d04071 | 90 | static int have_vcpu_info_placement = 1; |
60223a32 | 91 | |
9c7a7942 | 92 | static void xen_vcpu_setup(int cpu) |
5ead97c8 | 93 | { |
60223a32 JF |
94 | struct vcpu_register_vcpu_info info; |
95 | int err; | |
96 | struct vcpu_info *vcpup; | |
97 | ||
a0d695c8 | 98 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
5ead97c8 | 99 | per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; |
60223a32 JF |
100 | |
101 | if (!have_vcpu_info_placement) | |
102 | return; /* already tested, not available */ | |
103 | ||
104 | vcpup = &per_cpu(xen_vcpu_info, cpu); | |
105 | ||
9976b39b | 106 | info.mfn = arbitrary_virt_to_mfn(vcpup); |
60223a32 JF |
107 | info.offset = offset_in_page(vcpup); |
108 | ||
e3d26976 | 109 | printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", |
60223a32 JF |
110 | cpu, vcpup, info.mfn, info.offset); |
111 | ||
112 | /* Check to see if the hypervisor will put the vcpu_info | |
113 | structure where we want it, which allows direct access via | |
114 | a percpu-variable. */ | |
115 | err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); | |
116 | ||
117 | if (err) { | |
118 | printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); | |
119 | have_vcpu_info_placement = 0; | |
120 | } else { | |
121 | /* This cpu is using the registered vcpu info, even if | |
122 | later ones fail to. */ | |
123 | per_cpu(xen_vcpu, cpu) = vcpup; | |
6487673b | 124 | |
60223a32 JF |
125 | printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", |
126 | cpu, vcpup); | |
127 | } | |
5ead97c8 JF |
128 | } |
129 | ||
9c7a7942 JF |
130 | /* |
131 | * On restore, set the vcpu placement up again. | |
132 | * If it fails, then we're in a bad state, since | |
133 | * we can't back out from using it... | |
134 | */ | |
135 | void xen_vcpu_restore(void) | |
136 | { | |
137 | if (have_vcpu_info_placement) { | |
138 | int cpu; | |
139 | ||
140 | for_each_online_cpu(cpu) { | |
141 | bool other_cpu = (cpu != smp_processor_id()); | |
142 | ||
143 | if (other_cpu && | |
144 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) | |
145 | BUG(); | |
146 | ||
147 | xen_vcpu_setup(cpu); | |
148 | ||
149 | if (other_cpu && | |
150 | HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) | |
151 | BUG(); | |
152 | } | |
153 | ||
154 | BUG_ON(!have_vcpu_info_placement); | |
155 | } | |
156 | } | |
157 | ||
5ead97c8 JF |
158 | static void __init xen_banner(void) |
159 | { | |
95c7c23b JF |
160 | unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); |
161 | struct xen_extraversion extra; | |
162 | HYPERVISOR_xen_version(XENVER_extraversion, &extra); | |
163 | ||
5ead97c8 | 164 | printk(KERN_INFO "Booting paravirtualized kernel on %s\n", |
93b1eab3 | 165 | pv_info.name); |
95c7c23b JF |
166 | printk(KERN_INFO "Xen version: %d.%d%s%s\n", |
167 | version >> 16, version & 0xffff, extra.extraversion, | |
e57778a1 | 168 | xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); |
5ead97c8 JF |
169 | } |
170 | ||
65ea5b03 PA |
171 | static void xen_cpuid(unsigned int *ax, unsigned int *bx, |
172 | unsigned int *cx, unsigned int *dx) | |
5ead97c8 JF |
173 | { |
174 | unsigned maskedx = ~0; | |
175 | ||
176 | /* | |
177 | * Mask out inconvenient features, to try and disable as many | |
178 | * unsupported kernel subsystems as possible. | |
179 | */ | |
65ea5b03 | 180 | if (*ax == 1) |
5ead97c8 JF |
181 | maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */ |
182 | (1 << X86_FEATURE_ACPI) | /* disable ACPI */ | |
dbe9e994 JF |
183 | (1 << X86_FEATURE_MCE) | /* disable MCE */ |
184 | (1 << X86_FEATURE_MCA) | /* disable MCA */ | |
5ead97c8 JF |
185 | (1 << X86_FEATURE_ACC)); /* thermal monitoring */ |
186 | ||
187 | asm(XEN_EMULATE_PREFIX "cpuid" | |
65ea5b03 PA |
188 | : "=a" (*ax), |
189 | "=b" (*bx), | |
190 | "=c" (*cx), | |
191 | "=d" (*dx) | |
192 | : "0" (*ax), "2" (*cx)); | |
193 | *dx &= maskedx; | |
5ead97c8 JF |
194 | } |
195 | ||
196 | static void xen_set_debugreg(int reg, unsigned long val) | |
197 | { | |
198 | HYPERVISOR_set_debugreg(reg, val); | |
199 | } | |
200 | ||
201 | static unsigned long xen_get_debugreg(int reg) | |
202 | { | |
203 | return HYPERVISOR_get_debugreg(reg); | |
204 | } | |
205 | ||
224101ed | 206 | static void xen_end_context_switch(struct task_struct *next) |
5ead97c8 | 207 | { |
5ead97c8 | 208 | xen_mc_flush(); |
224101ed | 209 | paravirt_end_context_switch(next); |
5ead97c8 JF |
210 | } |
211 | ||
212 | static unsigned long xen_store_tr(void) | |
213 | { | |
214 | return 0; | |
215 | } | |
216 | ||
a05d2eba | 217 | /* |
cef43bf6 JF |
218 | * Set the page permissions for a particular virtual address. If the |
219 | * address is a vmalloc mapping (or other non-linear mapping), then | |
220 | * find the linear mapping of the page and also set its protections to | |
221 | * match. | |
a05d2eba JF |
222 | */ |
223 | static void set_aliased_prot(void *v, pgprot_t prot) | |
224 | { | |
225 | int level; | |
226 | pte_t *ptep; | |
227 | pte_t pte; | |
228 | unsigned long pfn; | |
229 | struct page *page; | |
230 | ||
231 | ptep = lookup_address((unsigned long)v, &level); | |
232 | BUG_ON(ptep == NULL); | |
233 | ||
234 | pfn = pte_pfn(*ptep); | |
235 | page = pfn_to_page(pfn); | |
236 | ||
237 | pte = pfn_pte(pfn, prot); | |
238 | ||
239 | if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) | |
240 | BUG(); | |
241 | ||
242 | if (!PageHighMem(page)) { | |
243 | void *av = __va(PFN_PHYS(pfn)); | |
244 | ||
245 | if (av != v) | |
246 | if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) | |
247 | BUG(); | |
248 | } else | |
249 | kmap_flush_unused(); | |
250 | } | |
251 | ||
38ffbe66 JF |
252 | static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) |
253 | { | |
a05d2eba | 254 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
255 | int i; |
256 | ||
a05d2eba JF |
257 | for(i = 0; i < entries; i += entries_per_page) |
258 | set_aliased_prot(ldt + i, PAGE_KERNEL_RO); | |
38ffbe66 JF |
259 | } |
260 | ||
261 | static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) | |
262 | { | |
a05d2eba | 263 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
264 | int i; |
265 | ||
a05d2eba JF |
266 | for(i = 0; i < entries; i += entries_per_page) |
267 | set_aliased_prot(ldt + i, PAGE_KERNEL); | |
38ffbe66 JF |
268 | } |
269 | ||
5ead97c8 JF |
270 | static void xen_set_ldt(const void *addr, unsigned entries) |
271 | { | |
5ead97c8 JF |
272 | struct mmuext_op *op; |
273 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
274 | ||
275 | op = mcs.args; | |
276 | op->cmd = MMUEXT_SET_LDT; | |
4dbf7af6 | 277 | op->arg1.linear_addr = (unsigned long)addr; |
5ead97c8 JF |
278 | op->arg2.nr_ents = entries; |
279 | ||
280 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
281 | ||
282 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
283 | } | |
284 | ||
6b68f01b | 285 | static void xen_load_gdt(const struct desc_ptr *dtr) |
5ead97c8 JF |
286 | { |
287 | unsigned long *frames; | |
288 | unsigned long va = dtr->address; | |
289 | unsigned int size = dtr->size + 1; | |
290 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
291 | int f; | |
292 | struct multicall_space mcs; | |
293 | ||
294 | /* A GDT can be up to 64k in size, which corresponds to 8192 | |
295 | 8-byte entries, or 16 4k pages.. */ | |
296 | ||
297 | BUG_ON(size > 65536); | |
298 | BUG_ON(va & ~PAGE_MASK); | |
299 | ||
300 | mcs = xen_mc_entry(sizeof(*frames) * pages); | |
301 | frames = mcs.args; | |
302 | ||
303 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
6ed6bf42 JF |
304 | int level; |
305 | pte_t *ptep = lookup_address(va, &level); | |
306 | unsigned long pfn, mfn; | |
307 | void *virt; | |
308 | ||
309 | BUG_ON(ptep == NULL); | |
310 | ||
311 | pfn = pte_pfn(*ptep); | |
312 | mfn = pfn_to_mfn(pfn); | |
313 | virt = __va(PFN_PHYS(pfn)); | |
314 | ||
315 | frames[f] = mfn; | |
9976b39b | 316 | |
5ead97c8 | 317 | make_lowmem_page_readonly((void *)va); |
6ed6bf42 | 318 | make_lowmem_page_readonly(virt); |
5ead97c8 JF |
319 | } |
320 | ||
321 | MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct)); | |
322 | ||
323 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
324 | } | |
325 | ||
326 | static void load_TLS_descriptor(struct thread_struct *t, | |
327 | unsigned int cpu, unsigned int i) | |
328 | { | |
329 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
9976b39b | 330 | xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); |
5ead97c8 JF |
331 | struct multicall_space mc = __xen_mc_entry(0); |
332 | ||
333 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
334 | } | |
335 | ||
336 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
337 | { | |
8b84ad94 | 338 | /* |
ccbeed3a TH |
339 | * XXX sleazy hack: If we're being called in a lazy-cpu zone |
340 | * and lazy gs handling is enabled, it means we're in a | |
341 | * context switch, and %gs has just been saved. This means we | |
342 | * can zero it out to prevent faults on exit from the | |
343 | * hypervisor if the next process has no %gs. Either way, it | |
344 | * has been saved, and the new value will get loaded properly. | |
345 | * This will go away as soon as Xen has been modified to not | |
346 | * save/restore %gs for normal hypercalls. | |
8a95408e EH |
347 | * |
348 | * On x86_64, this hack is not used for %gs, because gs points | |
349 | * to KERNEL_GS_BASE (and uses it for PDA references), so we | |
350 | * must not zero %gs on x86_64 | |
351 | * | |
352 | * For x86_64, we need to zero %fs, otherwise we may get an | |
353 | * exception between the new %fs descriptor being loaded and | |
354 | * %fs being effectively cleared at __switch_to(). | |
8b84ad94 | 355 | */ |
8a95408e EH |
356 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { |
357 | #ifdef CONFIG_X86_32 | |
ccbeed3a | 358 | lazy_load_gs(0); |
8a95408e EH |
359 | #else |
360 | loadsegment(fs, 0); | |
361 | #endif | |
362 | } | |
363 | ||
364 | xen_mc_batch(); | |
365 | ||
366 | load_TLS_descriptor(t, cpu, 0); | |
367 | load_TLS_descriptor(t, cpu, 1); | |
368 | load_TLS_descriptor(t, cpu, 2); | |
369 | ||
370 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
5ead97c8 JF |
371 | } |
372 | ||
a8fc1089 EH |
373 | #ifdef CONFIG_X86_64 |
374 | static void xen_load_gs_index(unsigned int idx) | |
375 | { | |
376 | if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) | |
377 | BUG(); | |
5ead97c8 | 378 | } |
a8fc1089 | 379 | #endif |
5ead97c8 JF |
380 | |
381 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
75b8bb3e | 382 | const void *ptr) |
5ead97c8 | 383 | { |
cef43bf6 | 384 | xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); |
75b8bb3e | 385 | u64 entry = *(u64 *)ptr; |
5ead97c8 | 386 | |
f120f13e JF |
387 | preempt_disable(); |
388 | ||
5ead97c8 JF |
389 | xen_mc_flush(); |
390 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
391 | BUG(); | |
f120f13e JF |
392 | |
393 | preempt_enable(); | |
5ead97c8 JF |
394 | } |
395 | ||
e176d367 | 396 | static int cvt_gate_to_trap(int vector, const gate_desc *val, |
5ead97c8 JF |
397 | struct trap_info *info) |
398 | { | |
e176d367 | 399 | if (val->type != 0xf && val->type != 0xe) |
5ead97c8 JF |
400 | return 0; |
401 | ||
402 | info->vector = vector; | |
e176d367 EH |
403 | info->address = gate_offset(*val); |
404 | info->cs = gate_segment(*val); | |
405 | info->flags = val->dpl; | |
5ead97c8 | 406 | /* interrupt gates clear IF */ |
e176d367 | 407 | if (val->type == 0xe) |
5ead97c8 JF |
408 | info->flags |= 4; |
409 | ||
410 | return 1; | |
411 | } | |
412 | ||
413 | /* Locations of each CPU's IDT */ | |
6b68f01b | 414 | static DEFINE_PER_CPU(struct desc_ptr, idt_desc); |
5ead97c8 JF |
415 | |
416 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
417 | also update Xen. */ | |
8d947344 | 418 | static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) |
5ead97c8 | 419 | { |
5ead97c8 | 420 | unsigned long p = (unsigned long)&dt[entrynum]; |
f120f13e JF |
421 | unsigned long start, end; |
422 | ||
423 | preempt_disable(); | |
424 | ||
425 | start = __get_cpu_var(idt_desc).address; | |
426 | end = start + __get_cpu_var(idt_desc).size + 1; | |
5ead97c8 JF |
427 | |
428 | xen_mc_flush(); | |
429 | ||
8d947344 | 430 | native_write_idt_entry(dt, entrynum, g); |
5ead97c8 JF |
431 | |
432 | if (p >= start && (p + 8) <= end) { | |
433 | struct trap_info info[2]; | |
434 | ||
435 | info[1].address = 0; | |
436 | ||
e176d367 | 437 | if (cvt_gate_to_trap(entrynum, g, &info[0])) |
5ead97c8 JF |
438 | if (HYPERVISOR_set_trap_table(info)) |
439 | BUG(); | |
440 | } | |
f120f13e JF |
441 | |
442 | preempt_enable(); | |
5ead97c8 JF |
443 | } |
444 | ||
6b68f01b | 445 | static void xen_convert_trap_info(const struct desc_ptr *desc, |
f87e4cac | 446 | struct trap_info *traps) |
5ead97c8 | 447 | { |
5ead97c8 JF |
448 | unsigned in, out, count; |
449 | ||
e176d367 | 450 | count = (desc->size+1) / sizeof(gate_desc); |
5ead97c8 JF |
451 | BUG_ON(count > 256); |
452 | ||
5ead97c8 | 453 | for (in = out = 0; in < count; in++) { |
e176d367 | 454 | gate_desc *entry = (gate_desc*)(desc->address) + in; |
5ead97c8 | 455 | |
e176d367 | 456 | if (cvt_gate_to_trap(in, entry, &traps[out])) |
5ead97c8 JF |
457 | out++; |
458 | } | |
459 | traps[out].address = 0; | |
f87e4cac JF |
460 | } |
461 | ||
462 | void xen_copy_trap_info(struct trap_info *traps) | |
463 | { | |
6b68f01b | 464 | const struct desc_ptr *desc = &__get_cpu_var(idt_desc); |
f87e4cac JF |
465 | |
466 | xen_convert_trap_info(desc, traps); | |
f87e4cac JF |
467 | } |
468 | ||
469 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
470 | hold a spinlock to protect the static traps[] array (static because | |
471 | it avoids allocation, and saves stack space). */ | |
6b68f01b | 472 | static void xen_load_idt(const struct desc_ptr *desc) |
f87e4cac JF |
473 | { |
474 | static DEFINE_SPINLOCK(lock); | |
475 | static struct trap_info traps[257]; | |
f87e4cac JF |
476 | |
477 | spin_lock(&lock); | |
478 | ||
f120f13e JF |
479 | __get_cpu_var(idt_desc) = *desc; |
480 | ||
f87e4cac | 481 | xen_convert_trap_info(desc, traps); |
5ead97c8 JF |
482 | |
483 | xen_mc_flush(); | |
484 | if (HYPERVISOR_set_trap_table(traps)) | |
485 | BUG(); | |
486 | ||
487 | spin_unlock(&lock); | |
488 | } | |
489 | ||
490 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
491 | they're handled differently. */ | |
492 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
014b15be | 493 | const void *desc, int type) |
5ead97c8 | 494 | { |
f120f13e JF |
495 | preempt_disable(); |
496 | ||
014b15be GOC |
497 | switch (type) { |
498 | case DESC_LDT: | |
499 | case DESC_TSS: | |
5ead97c8 JF |
500 | /* ignore */ |
501 | break; | |
502 | ||
503 | default: { | |
9976b39b | 504 | xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); |
5ead97c8 JF |
505 | |
506 | xen_mc_flush(); | |
014b15be | 507 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) |
5ead97c8 JF |
508 | BUG(); |
509 | } | |
510 | ||
511 | } | |
f120f13e JF |
512 | |
513 | preempt_enable(); | |
5ead97c8 JF |
514 | } |
515 | ||
faca6227 | 516 | static void xen_load_sp0(struct tss_struct *tss, |
a05d2eba | 517 | struct thread_struct *thread) |
5ead97c8 JF |
518 | { |
519 | struct multicall_space mcs = xen_mc_entry(0); | |
faca6227 | 520 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); |
5ead97c8 JF |
521 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
522 | } | |
523 | ||
524 | static void xen_set_iopl_mask(unsigned mask) | |
525 | { | |
526 | struct physdev_set_iopl set_iopl; | |
527 | ||
528 | /* Force the change at ring 0. */ | |
529 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
530 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
531 | } | |
532 | ||
533 | static void xen_io_delay(void) | |
534 | { | |
535 | } | |
536 | ||
537 | #ifdef CONFIG_X86_LOCAL_APIC | |
ad66dd34 | 538 | static u32 xen_apic_read(u32 reg) |
5ead97c8 JF |
539 | { |
540 | return 0; | |
541 | } | |
f87e4cac | 542 | |
ad66dd34 | 543 | static void xen_apic_write(u32 reg, u32 val) |
f87e4cac JF |
544 | { |
545 | /* Warn to see if there's any stray references */ | |
546 | WARN_ON(1); | |
547 | } | |
ad66dd34 | 548 | |
ad66dd34 SS |
549 | static u64 xen_apic_icr_read(void) |
550 | { | |
551 | return 0; | |
552 | } | |
553 | ||
554 | static void xen_apic_icr_write(u32 low, u32 id) | |
555 | { | |
556 | /* Warn to see if there's any stray references */ | |
557 | WARN_ON(1); | |
558 | } | |
559 | ||
560 | static void xen_apic_wait_icr_idle(void) | |
561 | { | |
562 | return; | |
563 | } | |
564 | ||
94a8c3c2 YL |
565 | static u32 xen_safe_apic_wait_icr_idle(void) |
566 | { | |
567 | return 0; | |
568 | } | |
569 | ||
c1eeb2de YL |
570 | static void set_xen_basic_apic_ops(void) |
571 | { | |
572 | apic->read = xen_apic_read; | |
573 | apic->write = xen_apic_write; | |
574 | apic->icr_read = xen_apic_icr_read; | |
575 | apic->icr_write = xen_apic_icr_write; | |
576 | apic->wait_icr_idle = xen_apic_wait_icr_idle; | |
577 | apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle; | |
578 | } | |
ad66dd34 | 579 | |
5ead97c8 JF |
580 | #endif |
581 | ||
f87e4cac | 582 | |
7b1333aa JF |
583 | static void xen_clts(void) |
584 | { | |
585 | struct multicall_space mcs; | |
586 | ||
587 | mcs = xen_mc_entry(0); | |
588 | ||
589 | MULTI_fpu_taskswitch(mcs.mc, 0); | |
590 | ||
591 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
592 | } | |
593 | ||
594 | static void xen_write_cr0(unsigned long cr0) | |
595 | { | |
596 | struct multicall_space mcs; | |
597 | ||
598 | /* Only pay attention to cr0.TS; everything else is | |
599 | ignored. */ | |
600 | mcs = xen_mc_entry(0); | |
601 | ||
602 | MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); | |
603 | ||
604 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
605 | } | |
606 | ||
5ead97c8 JF |
607 | static void xen_write_cr4(unsigned long cr4) |
608 | { | |
2956a351 JF |
609 | cr4 &= ~X86_CR4_PGE; |
610 | cr4 &= ~X86_CR4_PSE; | |
611 | ||
612 | native_write_cr4(cr4); | |
5ead97c8 JF |
613 | } |
614 | ||
1153968a JF |
615 | static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) |
616 | { | |
617 | int ret; | |
618 | ||
619 | ret = 0; | |
620 | ||
f63c2f24 | 621 | switch (msr) { |
1153968a JF |
622 | #ifdef CONFIG_X86_64 |
623 | unsigned which; | |
624 | u64 base; | |
625 | ||
626 | case MSR_FS_BASE: which = SEGBASE_FS; goto set; | |
627 | case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; | |
628 | case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; | |
629 | ||
630 | set: | |
631 | base = ((u64)high << 32) | low; | |
632 | if (HYPERVISOR_set_segment_base(which, base) != 0) | |
633 | ret = -EFAULT; | |
634 | break; | |
635 | #endif | |
d89961e2 JF |
636 | |
637 | case MSR_STAR: | |
638 | case MSR_CSTAR: | |
639 | case MSR_LSTAR: | |
640 | case MSR_SYSCALL_MASK: | |
641 | case MSR_IA32_SYSENTER_CS: | |
642 | case MSR_IA32_SYSENTER_ESP: | |
643 | case MSR_IA32_SYSENTER_EIP: | |
644 | /* Fast syscall setup is all done in hypercalls, so | |
645 | these are all ignored. Stub them out here to stop | |
646 | Xen console noise. */ | |
647 | break; | |
648 | ||
1153968a JF |
649 | default: |
650 | ret = native_write_msr_safe(msr, low, high); | |
651 | } | |
652 | ||
653 | return ret; | |
654 | } | |
655 | ||
0e91398f | 656 | void xen_setup_shared_info(void) |
5ead97c8 JF |
657 | { |
658 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
15664f96 JF |
659 | set_fixmap(FIX_PARAVIRT_BOOTMAP, |
660 | xen_start_info->shared_info); | |
661 | ||
662 | HYPERVISOR_shared_info = | |
663 | (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); | |
5ead97c8 JF |
664 | } else |
665 | HYPERVISOR_shared_info = | |
666 | (struct shared_info *)__va(xen_start_info->shared_info); | |
667 | ||
2e8fe719 JF |
668 | #ifndef CONFIG_SMP |
669 | /* In UP this is as good a place as any to set up shared info */ | |
670 | xen_setup_vcpu_info_placement(); | |
671 | #endif | |
d5edbc1f JF |
672 | |
673 | xen_setup_mfn_list_list(); | |
2e8fe719 JF |
674 | } |
675 | ||
60223a32 | 676 | /* This is called once we have the cpu_possible_map */ |
0e91398f | 677 | void xen_setup_vcpu_info_placement(void) |
60223a32 JF |
678 | { |
679 | int cpu; | |
680 | ||
681 | for_each_possible_cpu(cpu) | |
682 | xen_vcpu_setup(cpu); | |
683 | ||
684 | /* xen_vcpu_setup managed to place the vcpu_info within the | |
685 | percpu area for all cpus, so make use of it */ | |
686 | if (have_vcpu_info_placement) { | |
687 | printk(KERN_INFO "Xen: using vcpu_info placement\n"); | |
688 | ||
ecb93d1c JF |
689 | pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); |
690 | pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); | |
691 | pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); | |
692 | pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); | |
93b1eab3 | 693 | pv_mmu_ops.read_cr2 = xen_read_cr2_direct; |
60223a32 | 694 | } |
5ead97c8 JF |
695 | } |
696 | ||
ab144f5e AK |
697 | static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, |
698 | unsigned long addr, unsigned len) | |
6487673b JF |
699 | { |
700 | char *start, *end, *reloc; | |
701 | unsigned ret; | |
702 | ||
703 | start = end = reloc = NULL; | |
704 | ||
93b1eab3 JF |
705 | #define SITE(op, x) \ |
706 | case PARAVIRT_PATCH(op.x): \ | |
6487673b JF |
707 | if (have_vcpu_info_placement) { \ |
708 | start = (char *)xen_##x##_direct; \ | |
709 | end = xen_##x##_direct_end; \ | |
710 | reloc = xen_##x##_direct_reloc; \ | |
711 | } \ | |
712 | goto patch_site | |
713 | ||
714 | switch (type) { | |
93b1eab3 JF |
715 | SITE(pv_irq_ops, irq_enable); |
716 | SITE(pv_irq_ops, irq_disable); | |
717 | SITE(pv_irq_ops, save_fl); | |
718 | SITE(pv_irq_ops, restore_fl); | |
6487673b JF |
719 | #undef SITE |
720 | ||
721 | patch_site: | |
722 | if (start == NULL || (end-start) > len) | |
723 | goto default_patch; | |
724 | ||
ab144f5e | 725 | ret = paravirt_patch_insns(insnbuf, len, start, end); |
6487673b JF |
726 | |
727 | /* Note: because reloc is assigned from something that | |
728 | appears to be an array, gcc assumes it's non-null, | |
729 | but doesn't know its relationship with start and | |
730 | end. */ | |
731 | if (reloc > start && reloc < end) { | |
732 | int reloc_off = reloc - start; | |
ab144f5e AK |
733 | long *relocp = (long *)(insnbuf + reloc_off); |
734 | long delta = start - (char *)addr; | |
6487673b JF |
735 | |
736 | *relocp += delta; | |
737 | } | |
738 | break; | |
739 | ||
740 | default_patch: | |
741 | default: | |
ab144f5e AK |
742 | ret = paravirt_patch_default(type, clobbers, insnbuf, |
743 | addr, len); | |
6487673b JF |
744 | break; |
745 | } | |
746 | ||
747 | return ret; | |
748 | } | |
749 | ||
93b1eab3 | 750 | static const struct pv_info xen_info __initdata = { |
5ead97c8 JF |
751 | .paravirt_enabled = 1, |
752 | .shared_kernel_pmd = 0, | |
753 | ||
754 | .name = "Xen", | |
93b1eab3 | 755 | }; |
5ead97c8 | 756 | |
93b1eab3 | 757 | static const struct pv_init_ops xen_init_ops __initdata = { |
6487673b | 758 | .patch = xen_patch, |
5ead97c8 | 759 | |
93b1eab3 | 760 | .banner = xen_banner, |
5ead97c8 JF |
761 | .memory_setup = xen_memory_setup, |
762 | .arch_setup = xen_arch_setup, | |
e2426cf8 | 763 | .post_allocator_init = xen_post_allocator_init, |
93b1eab3 | 764 | }; |
5ead97c8 | 765 | |
93b1eab3 | 766 | static const struct pv_time_ops xen_time_ops __initdata = { |
15c84731 | 767 | .time_init = xen_time_init, |
93b1eab3 | 768 | |
15c84731 JF |
769 | .set_wallclock = xen_set_wallclock, |
770 | .get_wallclock = xen_get_wallclock, | |
e93ef949 | 771 | .get_tsc_khz = xen_tsc_khz, |
ab550288 | 772 | .sched_clock = xen_sched_clock, |
93b1eab3 | 773 | }; |
15c84731 | 774 | |
93b1eab3 | 775 | static const struct pv_cpu_ops xen_cpu_ops __initdata = { |
5ead97c8 JF |
776 | .cpuid = xen_cpuid, |
777 | ||
778 | .set_debugreg = xen_set_debugreg, | |
779 | .get_debugreg = xen_get_debugreg, | |
780 | ||
7b1333aa | 781 | .clts = xen_clts, |
5ead97c8 JF |
782 | |
783 | .read_cr0 = native_read_cr0, | |
7b1333aa | 784 | .write_cr0 = xen_write_cr0, |
5ead97c8 | 785 | |
5ead97c8 JF |
786 | .read_cr4 = native_read_cr4, |
787 | .read_cr4_safe = native_read_cr4_safe, | |
788 | .write_cr4 = xen_write_cr4, | |
789 | ||
5ead97c8 JF |
790 | .wbinvd = native_wbinvd, |
791 | ||
792 | .read_msr = native_read_msr_safe, | |
1153968a | 793 | .write_msr = xen_write_msr_safe, |
5ead97c8 JF |
794 | .read_tsc = native_read_tsc, |
795 | .read_pmc = native_read_pmc, | |
796 | ||
81e103f1 | 797 | .iret = xen_iret, |
d75cd22f | 798 | .irq_enable_sysexit = xen_sysexit, |
6fcac6d3 JF |
799 | #ifdef CONFIG_X86_64 |
800 | .usergs_sysret32 = xen_sysret32, | |
801 | .usergs_sysret64 = xen_sysret64, | |
802 | #endif | |
5ead97c8 JF |
803 | |
804 | .load_tr_desc = paravirt_nop, | |
805 | .set_ldt = xen_set_ldt, | |
806 | .load_gdt = xen_load_gdt, | |
807 | .load_idt = xen_load_idt, | |
808 | .load_tls = xen_load_tls, | |
a8fc1089 EH |
809 | #ifdef CONFIG_X86_64 |
810 | .load_gs_index = xen_load_gs_index, | |
811 | #endif | |
5ead97c8 | 812 | |
38ffbe66 JF |
813 | .alloc_ldt = xen_alloc_ldt, |
814 | .free_ldt = xen_free_ldt, | |
815 | ||
5ead97c8 JF |
816 | .store_gdt = native_store_gdt, |
817 | .store_idt = native_store_idt, | |
818 | .store_tr = xen_store_tr, | |
819 | ||
820 | .write_ldt_entry = xen_write_ldt_entry, | |
821 | .write_gdt_entry = xen_write_gdt_entry, | |
822 | .write_idt_entry = xen_write_idt_entry, | |
faca6227 | 823 | .load_sp0 = xen_load_sp0, |
5ead97c8 JF |
824 | |
825 | .set_iopl_mask = xen_set_iopl_mask, | |
826 | .io_delay = xen_io_delay, | |
827 | ||
952d1d70 JF |
828 | /* Xen takes care of %gs when switching to usermode for us */ |
829 | .swapgs = paravirt_nop, | |
830 | ||
224101ed JF |
831 | .start_context_switch = paravirt_start_context_switch, |
832 | .end_context_switch = xen_end_context_switch, | |
93b1eab3 JF |
833 | }; |
834 | ||
93b1eab3 | 835 | static const struct pv_apic_ops xen_apic_ops __initdata = { |
5ead97c8 | 836 | #ifdef CONFIG_X86_LOCAL_APIC |
5ead97c8 JF |
837 | .setup_boot_clock = paravirt_nop, |
838 | .setup_secondary_clock = paravirt_nop, | |
839 | .startup_ipi_hook = paravirt_nop, | |
840 | #endif | |
93b1eab3 JF |
841 | }; |
842 | ||
fefa629a JF |
843 | static void xen_reboot(int reason) |
844 | { | |
349c709f JF |
845 | struct sched_shutdown r = { .reason = reason }; |
846 | ||
fefa629a JF |
847 | #ifdef CONFIG_SMP |
848 | smp_send_stop(); | |
849 | #endif | |
850 | ||
349c709f | 851 | if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) |
fefa629a JF |
852 | BUG(); |
853 | } | |
854 | ||
855 | static void xen_restart(char *msg) | |
856 | { | |
857 | xen_reboot(SHUTDOWN_reboot); | |
858 | } | |
859 | ||
860 | static void xen_emergency_restart(void) | |
861 | { | |
862 | xen_reboot(SHUTDOWN_reboot); | |
863 | } | |
864 | ||
865 | static void xen_machine_halt(void) | |
866 | { | |
867 | xen_reboot(SHUTDOWN_poweroff); | |
868 | } | |
869 | ||
870 | static void xen_crash_shutdown(struct pt_regs *regs) | |
871 | { | |
872 | xen_reboot(SHUTDOWN_crash); | |
873 | } | |
874 | ||
875 | static const struct machine_ops __initdata xen_machine_ops = { | |
876 | .restart = xen_restart, | |
877 | .halt = xen_machine_halt, | |
878 | .power_off = xen_machine_halt, | |
879 | .shutdown = xen_machine_halt, | |
880 | .crash_shutdown = xen_crash_shutdown, | |
881 | .emergency_restart = xen_emergency_restart, | |
882 | }; | |
883 | ||
6487673b | 884 | |
5ead97c8 JF |
885 | /* First C function to be called on Xen boot */ |
886 | asmlinkage void __init xen_start_kernel(void) | |
887 | { | |
888 | pgd_t *pgd; | |
889 | ||
890 | if (!xen_start_info) | |
891 | return; | |
892 | ||
6e833587 JF |
893 | xen_domain_type = XEN_PV_DOMAIN; |
894 | ||
7999f4b4 | 895 | BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0); |
5ead97c8 | 896 | |
e57778a1 JF |
897 | xen_setup_features(); |
898 | ||
5ead97c8 | 899 | /* Install Xen paravirt ops */ |
93b1eab3 JF |
900 | pv_info = xen_info; |
901 | pv_init_ops = xen_init_ops; | |
902 | pv_time_ops = xen_time_ops; | |
903 | pv_cpu_ops = xen_cpu_ops; | |
93b1eab3 JF |
904 | pv_apic_ops = xen_apic_ops; |
905 | pv_mmu_ops = xen_mmu_ops; | |
93b1eab3 | 906 | |
0d1edf46 JF |
907 | xen_init_irq_ops(); |
908 | ||
94a8c3c2 | 909 | #ifdef CONFIG_X86_LOCAL_APIC |
ad66dd34 | 910 | /* |
94a8c3c2 | 911 | * set up the basic apic ops. |
ad66dd34 | 912 | */ |
c1eeb2de | 913 | set_xen_basic_apic_ops(); |
ad66dd34 | 914 | #endif |
93b1eab3 | 915 | |
e57778a1 JF |
916 | if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { |
917 | pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; | |
918 | pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; | |
919 | } | |
920 | ||
fefa629a JF |
921 | machine_ops = xen_machine_ops; |
922 | ||
38341432 JF |
923 | #ifdef CONFIG_X86_64 |
924 | /* | |
925 | * Setup percpu state. We only need to do this for 64-bit | |
926 | * because 32-bit already has %fs set properly. | |
927 | */ | |
795f99b6 | 928 | load_percpu_segment(0); |
38341432 JF |
929 | #endif |
930 | /* | |
931 | * The only reliable way to retain the initial address of the | |
932 | * percpu gdt_page is to remember it here, so we can go and | |
933 | * mark it RW later, when the initial percpu area is freed. | |
934 | */ | |
935 | xen_initial_gdt = &per_cpu(gdt_page, 0); | |
795f99b6 | 936 | |
a9e7062d | 937 | xen_smp_init(); |
5ead97c8 JF |
938 | |
939 | /* Get mfn list */ | |
940 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
d451bb7a | 941 | xen_build_dynamic_phys_to_machine(); |
5ead97c8 JF |
942 | |
943 | pgd = (pgd_t *)xen_start_info->pt_base; | |
944 | ||
084a2a4e JF |
945 | /* Prevent unwanted bits from being set in PTEs. */ |
946 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
6e833587 | 947 | if (!xen_initial_domain()) |
084a2a4e | 948 | __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); |
60223a32 | 949 | |
60223a32 | 950 | /* Don't do the full vcpu_info placement stuff until we have a |
2e8fe719 | 951 | possible map and a non-dummy shared_info. */ |
60223a32 | 952 | per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; |
5ead97c8 | 953 | |
55d80856 JF |
954 | local_irq_disable(); |
955 | early_boot_irqs_off(); | |
956 | ||
084a2a4e | 957 | xen_raw_console_write("mapping kernel into physical memory\n"); |
d114e198 | 958 | pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages); |
5ead97c8 | 959 | |
084a2a4e | 960 | init_mm.pgd = pgd; |
5ead97c8 JF |
961 | |
962 | /* keep using Xen gdt for now; no urgent need to change it */ | |
963 | ||
93b1eab3 | 964 | pv_info.kernel_rpl = 1; |
5ead97c8 | 965 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) |
93b1eab3 | 966 | pv_info.kernel_rpl = 0; |
5ead97c8 JF |
967 | |
968 | /* set the limit of our address space */ | |
fb1d8404 | 969 | xen_reserve_top(); |
5ead97c8 | 970 | |
7d087b68 | 971 | #ifdef CONFIG_X86_32 |
5ead97c8 JF |
972 | /* set up basic CPUID stuff */ |
973 | cpu_detect(&new_cpu_data); | |
974 | new_cpu_data.hard_math = 1; | |
975 | new_cpu_data.x86_capability[0] = cpuid_edx(1); | |
7d087b68 | 976 | #endif |
5ead97c8 JF |
977 | |
978 | /* Poke various useful things into boot_params */ | |
30c82645 PA |
979 | boot_params.hdr.type_of_loader = (9 << 4) | 0; |
980 | boot_params.hdr.ramdisk_image = xen_start_info->mod_start | |
981 | ? __pa(xen_start_info->mod_start) : 0; | |
982 | boot_params.hdr.ramdisk_size = xen_start_info->mod_len; | |
b7c3c5c1 | 983 | boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); |
5ead97c8 | 984 | |
6e833587 | 985 | if (!xen_initial_domain()) { |
83abc70a | 986 | add_preferred_console("xenboot", 0, NULL); |
9e124fe1 | 987 | add_preferred_console("tty", 0, NULL); |
b8c2d3df | 988 | add_preferred_console("hvc", 0, NULL); |
9e124fe1 | 989 | } |
b8c2d3df | 990 | |
084a2a4e JF |
991 | xen_raw_console_write("about to get started...\n"); |
992 | ||
5ead97c8 | 993 | /* Start the world */ |
f5d36de0 | 994 | #ifdef CONFIG_X86_32 |
f0d43100 | 995 | i386_start_kernel(); |
f5d36de0 | 996 | #else |
084a2a4e | 997 | x86_64_start_reservations((char *)__pa_symbol(&boot_params)); |
f5d36de0 | 998 | #endif |
5ead97c8 | 999 | } |