]>
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> | |
f87e4cac | 28 | #include <linux/smp.h> |
5ead97c8 JF |
29 | |
30 | #include <xen/interface/xen.h> | |
31 | #include <xen/interface/physdev.h> | |
32 | #include <xen/interface/vcpu.h> | |
fefa629a | 33 | #include <xen/interface/sched.h> |
5ead97c8 JF |
34 | #include <xen/features.h> |
35 | #include <xen/page.h> | |
36 | ||
37 | #include <asm/paravirt.h> | |
38 | #include <asm/page.h> | |
39 | #include <asm/xen/hypercall.h> | |
40 | #include <asm/xen/hypervisor.h> | |
41 | #include <asm/fixmap.h> | |
42 | #include <asm/processor.h> | |
43 | #include <asm/setup.h> | |
44 | #include <asm/desc.h> | |
45 | #include <asm/pgtable.h> | |
f87e4cac | 46 | #include <asm/tlbflush.h> |
fefa629a | 47 | #include <asm/reboot.h> |
5ead97c8 JF |
48 | |
49 | #include "xen-ops.h" | |
3b827c1b | 50 | #include "mmu.h" |
5ead97c8 JF |
51 | #include "multicalls.h" |
52 | ||
53 | EXPORT_SYMBOL_GPL(hypercall_page); | |
54 | ||
55 | DEFINE_PER_CPU(enum paravirt_lazy_mode, xen_lazy_mode); | |
56 | ||
57 | DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); | |
58 | DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); | |
59 | DEFINE_PER_CPU(unsigned long, xen_cr3); | |
60 | ||
61 | struct start_info *xen_start_info; | |
62 | EXPORT_SYMBOL_GPL(xen_start_info); | |
63 | ||
60223a32 JF |
64 | static /* __initdata */ struct shared_info dummy_shared_info; |
65 | ||
66 | /* | |
67 | * Point at some empty memory to start with. We map the real shared_info | |
68 | * page as soon as fixmap is up and running. | |
69 | */ | |
70 | struct shared_info *HYPERVISOR_shared_info = (void *)&dummy_shared_info; | |
71 | ||
72 | /* | |
73 | * Flag to determine whether vcpu info placement is available on all | |
74 | * VCPUs. We assume it is to start with, and then set it to zero on | |
75 | * the first failure. This is because it can succeed on some VCPUs | |
76 | * and not others, since it can involve hypervisor memory allocation, | |
77 | * or because the guest failed to guarantee all the appropriate | |
78 | * constraints on all VCPUs (ie buffer can't cross a page boundary). | |
79 | * | |
80 | * Note that any particular CPU may be using a placed vcpu structure, | |
81 | * but we can only optimise if the all are. | |
82 | * | |
83 | * 0: not available, 1: available | |
84 | */ | |
85 | static int have_vcpu_info_placement = 1; | |
86 | ||
87 | static void __init xen_vcpu_setup(int cpu) | |
5ead97c8 | 88 | { |
60223a32 JF |
89 | struct vcpu_register_vcpu_info info; |
90 | int err; | |
91 | struct vcpu_info *vcpup; | |
92 | ||
5ead97c8 | 93 | per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; |
60223a32 JF |
94 | |
95 | if (!have_vcpu_info_placement) | |
96 | return; /* already tested, not available */ | |
97 | ||
98 | vcpup = &per_cpu(xen_vcpu_info, cpu); | |
99 | ||
100 | info.mfn = virt_to_mfn(vcpup); | |
101 | info.offset = offset_in_page(vcpup); | |
102 | ||
103 | printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %x, offset %d\n", | |
104 | cpu, vcpup, info.mfn, info.offset); | |
105 | ||
106 | /* Check to see if the hypervisor will put the vcpu_info | |
107 | structure where we want it, which allows direct access via | |
108 | a percpu-variable. */ | |
109 | err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); | |
110 | ||
111 | if (err) { | |
112 | printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); | |
113 | have_vcpu_info_placement = 0; | |
114 | } else { | |
115 | /* This cpu is using the registered vcpu info, even if | |
116 | later ones fail to. */ | |
117 | per_cpu(xen_vcpu, cpu) = vcpup; | |
118 | printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", | |
119 | cpu, vcpup); | |
120 | } | |
5ead97c8 JF |
121 | } |
122 | ||
123 | static void __init xen_banner(void) | |
124 | { | |
125 | printk(KERN_INFO "Booting paravirtualized kernel on %s\n", | |
126 | paravirt_ops.name); | |
127 | printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic); | |
128 | } | |
129 | ||
130 | static void xen_cpuid(unsigned int *eax, unsigned int *ebx, | |
131 | unsigned int *ecx, unsigned int *edx) | |
132 | { | |
133 | unsigned maskedx = ~0; | |
134 | ||
135 | /* | |
136 | * Mask out inconvenient features, to try and disable as many | |
137 | * unsupported kernel subsystems as possible. | |
138 | */ | |
139 | if (*eax == 1) | |
140 | maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */ | |
141 | (1 << X86_FEATURE_ACPI) | /* disable ACPI */ | |
142 | (1 << X86_FEATURE_ACC)); /* thermal monitoring */ | |
143 | ||
144 | asm(XEN_EMULATE_PREFIX "cpuid" | |
145 | : "=a" (*eax), | |
146 | "=b" (*ebx), | |
147 | "=c" (*ecx), | |
148 | "=d" (*edx) | |
149 | : "0" (*eax), "2" (*ecx)); | |
150 | *edx &= maskedx; | |
151 | } | |
152 | ||
153 | static void xen_set_debugreg(int reg, unsigned long val) | |
154 | { | |
155 | HYPERVISOR_set_debugreg(reg, val); | |
156 | } | |
157 | ||
158 | static unsigned long xen_get_debugreg(int reg) | |
159 | { | |
160 | return HYPERVISOR_get_debugreg(reg); | |
161 | } | |
162 | ||
163 | static unsigned long xen_save_fl(void) | |
164 | { | |
165 | struct vcpu_info *vcpu; | |
166 | unsigned long flags; | |
167 | ||
5ead97c8 | 168 | vcpu = x86_read_percpu(xen_vcpu); |
f120f13e | 169 | |
5ead97c8 JF |
170 | /* flag has opposite sense of mask */ |
171 | flags = !vcpu->evtchn_upcall_mask; | |
5ead97c8 JF |
172 | |
173 | /* convert to IF type flag | |
174 | -0 -> 0x00000000 | |
175 | -1 -> 0xffffffff | |
176 | */ | |
177 | return (-flags) & X86_EFLAGS_IF; | |
178 | } | |
179 | ||
60223a32 JF |
180 | static unsigned long xen_save_fl_direct(void) |
181 | { | |
182 | unsigned long flags; | |
183 | ||
184 | /* flag has opposite sense of mask */ | |
185 | flags = !x86_read_percpu(xen_vcpu_info.evtchn_upcall_mask); | |
186 | ||
187 | /* convert to IF type flag | |
188 | -0 -> 0x00000000 | |
189 | -1 -> 0xffffffff | |
190 | */ | |
191 | return (-flags) & X86_EFLAGS_IF; | |
192 | } | |
193 | ||
5ead97c8 JF |
194 | static void xen_restore_fl(unsigned long flags) |
195 | { | |
196 | struct vcpu_info *vcpu; | |
197 | ||
5ead97c8 JF |
198 | /* convert from IF type flag */ |
199 | flags = !(flags & X86_EFLAGS_IF); | |
f120f13e JF |
200 | |
201 | /* There's a one instruction preempt window here. We need to | |
202 | make sure we're don't switch CPUs between getting the vcpu | |
203 | pointer and updating the mask. */ | |
204 | preempt_disable(); | |
5ead97c8 JF |
205 | vcpu = x86_read_percpu(xen_vcpu); |
206 | vcpu->evtchn_upcall_mask = flags; | |
f120f13e | 207 | preempt_enable_no_resched(); |
5ead97c8 | 208 | |
f120f13e JF |
209 | /* Doesn't matter if we get preempted here, because any |
210 | pending event will get dealt with anyway. */ | |
5ead97c8 | 211 | |
f120f13e JF |
212 | if (flags == 0) { |
213 | preempt_check_resched(); | |
214 | barrier(); /* unmask then check (avoid races) */ | |
5ead97c8 JF |
215 | if (unlikely(vcpu->evtchn_upcall_pending)) |
216 | force_evtchn_callback(); | |
f120f13e | 217 | } |
5ead97c8 JF |
218 | } |
219 | ||
60223a32 JF |
220 | static void xen_restore_fl_direct(unsigned long flags) |
221 | { | |
222 | /* convert from IF type flag */ | |
223 | flags = !(flags & X86_EFLAGS_IF); | |
224 | ||
225 | /* This is an atomic update, so no need to worry about | |
226 | preemption. */ | |
227 | x86_write_percpu(xen_vcpu_info.evtchn_upcall_mask, flags); | |
228 | ||
229 | /* If we get preempted here, then any pending event will be | |
230 | handled anyway. */ | |
231 | ||
232 | if (flags == 0) { | |
233 | barrier(); /* unmask then check (avoid races) */ | |
234 | if (unlikely(x86_read_percpu(xen_vcpu_info.evtchn_upcall_pending))) | |
235 | force_evtchn_callback(); | |
236 | } | |
237 | } | |
238 | ||
5ead97c8 JF |
239 | static void xen_irq_disable(void) |
240 | { | |
f120f13e JF |
241 | /* There's a one instruction preempt window here. We need to |
242 | make sure we're don't switch CPUs between getting the vcpu | |
243 | pointer and updating the mask. */ | |
5ead97c8 | 244 | preempt_disable(); |
f120f13e | 245 | x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1; |
5ead97c8 JF |
246 | preempt_enable_no_resched(); |
247 | } | |
248 | ||
60223a32 JF |
249 | static void xen_irq_disable_direct(void) |
250 | { | |
251 | /* Atomic update, so preemption not a concern. */ | |
252 | x86_write_percpu(xen_vcpu_info.evtchn_upcall_mask, 1); | |
253 | } | |
254 | ||
5ead97c8 JF |
255 | static void xen_irq_enable(void) |
256 | { | |
257 | struct vcpu_info *vcpu; | |
258 | ||
f120f13e JF |
259 | /* There's a one instruction preempt window here. We need to |
260 | make sure we're don't switch CPUs between getting the vcpu | |
261 | pointer and updating the mask. */ | |
5ead97c8 JF |
262 | preempt_disable(); |
263 | vcpu = x86_read_percpu(xen_vcpu); | |
264 | vcpu->evtchn_upcall_mask = 0; | |
f120f13e | 265 | preempt_enable_no_resched(); |
5ead97c8 | 266 | |
f120f13e JF |
267 | /* Doesn't matter if we get preempted here, because any |
268 | pending event will get dealt with anyway. */ | |
5ead97c8 | 269 | |
f120f13e | 270 | barrier(); /* unmask then check (avoid races) */ |
5ead97c8 JF |
271 | if (unlikely(vcpu->evtchn_upcall_pending)) |
272 | force_evtchn_callback(); | |
5ead97c8 JF |
273 | } |
274 | ||
60223a32 JF |
275 | static void xen_irq_enable_direct(void) |
276 | { | |
277 | /* Atomic update, so preemption not a concern. */ | |
278 | x86_write_percpu(xen_vcpu_info.evtchn_upcall_mask, 0); | |
279 | ||
280 | /* Doesn't matter if we get preempted here, because any | |
281 | pending event will get dealt with anyway. */ | |
282 | ||
283 | barrier(); /* unmask then check (avoid races) */ | |
284 | if (unlikely(x86_read_percpu(xen_vcpu_info.evtchn_upcall_pending))) | |
285 | force_evtchn_callback(); | |
286 | } | |
287 | ||
5ead97c8 JF |
288 | static void xen_safe_halt(void) |
289 | { | |
290 | /* Blocking includes an implicit local_irq_enable(). */ | |
291 | if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0) | |
292 | BUG(); | |
293 | } | |
294 | ||
295 | static void xen_halt(void) | |
296 | { | |
297 | if (irqs_disabled()) | |
298 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
299 | else | |
300 | xen_safe_halt(); | |
301 | } | |
302 | ||
303 | static void xen_set_lazy_mode(enum paravirt_lazy_mode mode) | |
304 | { | |
f120f13e JF |
305 | BUG_ON(preemptible()); |
306 | ||
5ead97c8 JF |
307 | switch (mode) { |
308 | case PARAVIRT_LAZY_NONE: | |
309 | BUG_ON(x86_read_percpu(xen_lazy_mode) == PARAVIRT_LAZY_NONE); | |
310 | break; | |
311 | ||
312 | case PARAVIRT_LAZY_MMU: | |
313 | case PARAVIRT_LAZY_CPU: | |
314 | BUG_ON(x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE); | |
315 | break; | |
316 | ||
317 | case PARAVIRT_LAZY_FLUSH: | |
318 | /* flush if necessary, but don't change state */ | |
319 | if (x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE) | |
320 | xen_mc_flush(); | |
321 | return; | |
322 | } | |
323 | ||
324 | xen_mc_flush(); | |
325 | x86_write_percpu(xen_lazy_mode, mode); | |
326 | } | |
327 | ||
328 | static unsigned long xen_store_tr(void) | |
329 | { | |
330 | return 0; | |
331 | } | |
332 | ||
333 | static void xen_set_ldt(const void *addr, unsigned entries) | |
334 | { | |
335 | unsigned long linear_addr = (unsigned long)addr; | |
336 | struct mmuext_op *op; | |
337 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
338 | ||
339 | op = mcs.args; | |
340 | op->cmd = MMUEXT_SET_LDT; | |
341 | if (linear_addr) { | |
342 | /* ldt my be vmalloced, use arbitrary_virt_to_machine */ | |
343 | xmaddr_t maddr; | |
344 | maddr = arbitrary_virt_to_machine((unsigned long)addr); | |
345 | linear_addr = (unsigned long)maddr.maddr; | |
346 | } | |
347 | op->arg1.linear_addr = linear_addr; | |
348 | op->arg2.nr_ents = entries; | |
349 | ||
350 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
351 | ||
352 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
353 | } | |
354 | ||
355 | static void xen_load_gdt(const struct Xgt_desc_struct *dtr) | |
356 | { | |
357 | unsigned long *frames; | |
358 | unsigned long va = dtr->address; | |
359 | unsigned int size = dtr->size + 1; | |
360 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
361 | int f; | |
362 | struct multicall_space mcs; | |
363 | ||
364 | /* A GDT can be up to 64k in size, which corresponds to 8192 | |
365 | 8-byte entries, or 16 4k pages.. */ | |
366 | ||
367 | BUG_ON(size > 65536); | |
368 | BUG_ON(va & ~PAGE_MASK); | |
369 | ||
370 | mcs = xen_mc_entry(sizeof(*frames) * pages); | |
371 | frames = mcs.args; | |
372 | ||
373 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
374 | frames[f] = virt_to_mfn(va); | |
375 | make_lowmem_page_readonly((void *)va); | |
376 | } | |
377 | ||
378 | MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct)); | |
379 | ||
380 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
381 | } | |
382 | ||
383 | static void load_TLS_descriptor(struct thread_struct *t, | |
384 | unsigned int cpu, unsigned int i) | |
385 | { | |
386 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
387 | xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); | |
388 | struct multicall_space mc = __xen_mc_entry(0); | |
389 | ||
390 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
391 | } | |
392 | ||
393 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
394 | { | |
395 | xen_mc_batch(); | |
396 | ||
397 | load_TLS_descriptor(t, cpu, 0); | |
398 | load_TLS_descriptor(t, cpu, 1); | |
399 | load_TLS_descriptor(t, cpu, 2); | |
400 | ||
401 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
8b84ad94 JF |
402 | |
403 | /* | |
404 | * XXX sleazy hack: If we're being called in a lazy-cpu zone, | |
405 | * it means we're in a context switch, and %gs has just been | |
406 | * saved. This means we can zero it out to prevent faults on | |
407 | * exit from the hypervisor if the next process has no %gs. | |
408 | * Either way, it has been saved, and the new value will get | |
409 | * loaded properly. This will go away as soon as Xen has been | |
410 | * modified to not save/restore %gs for normal hypercalls. | |
411 | */ | |
412 | if (xen_get_lazy_mode() == PARAVIRT_LAZY_CPU) | |
413 | loadsegment(gs, 0); | |
5ead97c8 JF |
414 | } |
415 | ||
416 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
417 | u32 low, u32 high) | |
418 | { | |
419 | unsigned long lp = (unsigned long)&dt[entrynum]; | |
420 | xmaddr_t mach_lp = virt_to_machine(lp); | |
421 | u64 entry = (u64)high << 32 | low; | |
422 | ||
f120f13e JF |
423 | preempt_disable(); |
424 | ||
5ead97c8 JF |
425 | xen_mc_flush(); |
426 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
427 | BUG(); | |
f120f13e JF |
428 | |
429 | preempt_enable(); | |
5ead97c8 JF |
430 | } |
431 | ||
432 | static int cvt_gate_to_trap(int vector, u32 low, u32 high, | |
433 | struct trap_info *info) | |
434 | { | |
435 | u8 type, dpl; | |
436 | ||
437 | type = (high >> 8) & 0x1f; | |
438 | dpl = (high >> 13) & 3; | |
439 | ||
440 | if (type != 0xf && type != 0xe) | |
441 | return 0; | |
442 | ||
443 | info->vector = vector; | |
444 | info->address = (high & 0xffff0000) | (low & 0x0000ffff); | |
445 | info->cs = low >> 16; | |
446 | info->flags = dpl; | |
447 | /* interrupt gates clear IF */ | |
448 | if (type == 0xe) | |
449 | info->flags |= 4; | |
450 | ||
451 | return 1; | |
452 | } | |
453 | ||
454 | /* Locations of each CPU's IDT */ | |
455 | static DEFINE_PER_CPU(struct Xgt_desc_struct, idt_desc); | |
456 | ||
457 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
458 | also update Xen. */ | |
459 | static void xen_write_idt_entry(struct desc_struct *dt, int entrynum, | |
460 | u32 low, u32 high) | |
461 | { | |
5ead97c8 | 462 | unsigned long p = (unsigned long)&dt[entrynum]; |
f120f13e JF |
463 | unsigned long start, end; |
464 | ||
465 | preempt_disable(); | |
466 | ||
467 | start = __get_cpu_var(idt_desc).address; | |
468 | end = start + __get_cpu_var(idt_desc).size + 1; | |
5ead97c8 JF |
469 | |
470 | xen_mc_flush(); | |
471 | ||
472 | write_dt_entry(dt, entrynum, low, high); | |
473 | ||
474 | if (p >= start && (p + 8) <= end) { | |
475 | struct trap_info info[2]; | |
476 | ||
477 | info[1].address = 0; | |
478 | ||
479 | if (cvt_gate_to_trap(entrynum, low, high, &info[0])) | |
480 | if (HYPERVISOR_set_trap_table(info)) | |
481 | BUG(); | |
482 | } | |
f120f13e JF |
483 | |
484 | preempt_enable(); | |
5ead97c8 JF |
485 | } |
486 | ||
f87e4cac JF |
487 | static void xen_convert_trap_info(const struct Xgt_desc_struct *desc, |
488 | struct trap_info *traps) | |
5ead97c8 | 489 | { |
5ead97c8 JF |
490 | unsigned in, out, count; |
491 | ||
5ead97c8 JF |
492 | count = (desc->size+1) / 8; |
493 | BUG_ON(count > 256); | |
494 | ||
5ead97c8 JF |
495 | for (in = out = 0; in < count; in++) { |
496 | const u32 *entry = (u32 *)(desc->address + in * 8); | |
497 | ||
498 | if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out])) | |
499 | out++; | |
500 | } | |
501 | traps[out].address = 0; | |
f87e4cac JF |
502 | } |
503 | ||
504 | void xen_copy_trap_info(struct trap_info *traps) | |
505 | { | |
f120f13e | 506 | const struct Xgt_desc_struct *desc = &__get_cpu_var(idt_desc); |
f87e4cac JF |
507 | |
508 | xen_convert_trap_info(desc, traps); | |
f87e4cac JF |
509 | } |
510 | ||
511 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
512 | hold a spinlock to protect the static traps[] array (static because | |
513 | it avoids allocation, and saves stack space). */ | |
514 | static void xen_load_idt(const struct Xgt_desc_struct *desc) | |
515 | { | |
516 | static DEFINE_SPINLOCK(lock); | |
517 | static struct trap_info traps[257]; | |
f87e4cac JF |
518 | |
519 | spin_lock(&lock); | |
520 | ||
f120f13e JF |
521 | __get_cpu_var(idt_desc) = *desc; |
522 | ||
f87e4cac | 523 | xen_convert_trap_info(desc, traps); |
5ead97c8 JF |
524 | |
525 | xen_mc_flush(); | |
526 | if (HYPERVISOR_set_trap_table(traps)) | |
527 | BUG(); | |
528 | ||
529 | spin_unlock(&lock); | |
530 | } | |
531 | ||
532 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
533 | they're handled differently. */ | |
534 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
535 | u32 low, u32 high) | |
536 | { | |
f120f13e JF |
537 | preempt_disable(); |
538 | ||
5ead97c8 JF |
539 | switch ((high >> 8) & 0xff) { |
540 | case DESCTYPE_LDT: | |
541 | case DESCTYPE_TSS: | |
542 | /* ignore */ | |
543 | break; | |
544 | ||
545 | default: { | |
546 | xmaddr_t maddr = virt_to_machine(&dt[entry]); | |
547 | u64 desc = (u64)high << 32 | low; | |
548 | ||
549 | xen_mc_flush(); | |
550 | if (HYPERVISOR_update_descriptor(maddr.maddr, desc)) | |
551 | BUG(); | |
552 | } | |
553 | ||
554 | } | |
f120f13e JF |
555 | |
556 | preempt_enable(); | |
5ead97c8 JF |
557 | } |
558 | ||
559 | static void xen_load_esp0(struct tss_struct *tss, | |
f120f13e | 560 | struct thread_struct *thread) |
5ead97c8 JF |
561 | { |
562 | struct multicall_space mcs = xen_mc_entry(0); | |
563 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->esp0); | |
564 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
565 | } | |
566 | ||
567 | static void xen_set_iopl_mask(unsigned mask) | |
568 | { | |
569 | struct physdev_set_iopl set_iopl; | |
570 | ||
571 | /* Force the change at ring 0. */ | |
572 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
573 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
574 | } | |
575 | ||
576 | static void xen_io_delay(void) | |
577 | { | |
578 | } | |
579 | ||
580 | #ifdef CONFIG_X86_LOCAL_APIC | |
581 | static unsigned long xen_apic_read(unsigned long reg) | |
582 | { | |
583 | return 0; | |
584 | } | |
f87e4cac JF |
585 | |
586 | static void xen_apic_write(unsigned long reg, unsigned long val) | |
587 | { | |
588 | /* Warn to see if there's any stray references */ | |
589 | WARN_ON(1); | |
590 | } | |
5ead97c8 JF |
591 | #endif |
592 | ||
593 | static void xen_flush_tlb(void) | |
594 | { | |
d66bf8fc JF |
595 | struct mmuext_op *op; |
596 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
5ead97c8 | 597 | |
d66bf8fc JF |
598 | op = mcs.args; |
599 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
600 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
601 | ||
602 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
5ead97c8 JF |
603 | } |
604 | ||
605 | static void xen_flush_tlb_single(unsigned long addr) | |
606 | { | |
d66bf8fc JF |
607 | struct mmuext_op *op; |
608 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
5ead97c8 | 609 | |
d66bf8fc JF |
610 | op = mcs.args; |
611 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
612 | op->arg1.linear_addr = addr & PAGE_MASK; | |
613 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
614 | ||
615 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
5ead97c8 JF |
616 | } |
617 | ||
f87e4cac JF |
618 | static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm, |
619 | unsigned long va) | |
620 | { | |
d66bf8fc JF |
621 | struct { |
622 | struct mmuext_op op; | |
623 | cpumask_t mask; | |
624 | } *args; | |
f87e4cac | 625 | cpumask_t cpumask = *cpus; |
d66bf8fc | 626 | struct multicall_space mcs; |
f87e4cac JF |
627 | |
628 | /* | |
629 | * A couple of (to be removed) sanity checks: | |
630 | * | |
631 | * - current CPU must not be in mask | |
632 | * - mask must exist :) | |
633 | */ | |
634 | BUG_ON(cpus_empty(cpumask)); | |
635 | BUG_ON(cpu_isset(smp_processor_id(), cpumask)); | |
636 | BUG_ON(!mm); | |
637 | ||
638 | /* If a CPU which we ran on has gone down, OK. */ | |
639 | cpus_and(cpumask, cpumask, cpu_online_map); | |
640 | if (cpus_empty(cpumask)) | |
641 | return; | |
642 | ||
d66bf8fc JF |
643 | mcs = xen_mc_entry(sizeof(*args)); |
644 | args = mcs.args; | |
645 | args->mask = cpumask; | |
646 | args->op.arg2.vcpumask = &args->mask; | |
647 | ||
f87e4cac | 648 | if (va == TLB_FLUSH_ALL) { |
d66bf8fc | 649 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; |
f87e4cac | 650 | } else { |
d66bf8fc JF |
651 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
652 | args->op.arg1.linear_addr = va; | |
f87e4cac JF |
653 | } |
654 | ||
d66bf8fc JF |
655 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); |
656 | ||
657 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f87e4cac JF |
658 | } |
659 | ||
60223a32 JF |
660 | static void xen_write_cr2(unsigned long cr2) |
661 | { | |
662 | x86_read_percpu(xen_vcpu)->arch.cr2 = cr2; | |
663 | } | |
664 | ||
5ead97c8 JF |
665 | static unsigned long xen_read_cr2(void) |
666 | { | |
667 | return x86_read_percpu(xen_vcpu)->arch.cr2; | |
668 | } | |
669 | ||
60223a32 JF |
670 | static unsigned long xen_read_cr2_direct(void) |
671 | { | |
672 | return x86_read_percpu(xen_vcpu_info.arch.cr2); | |
673 | } | |
674 | ||
5ead97c8 JF |
675 | static void xen_write_cr4(unsigned long cr4) |
676 | { | |
677 | /* never allow TSC to be disabled */ | |
678 | native_write_cr4(cr4 & ~X86_CR4_TSD); | |
679 | } | |
680 | ||
5ead97c8 JF |
681 | static unsigned long xen_read_cr3(void) |
682 | { | |
683 | return x86_read_percpu(xen_cr3); | |
684 | } | |
685 | ||
686 | static void xen_write_cr3(unsigned long cr3) | |
687 | { | |
f120f13e JF |
688 | BUG_ON(preemptible()); |
689 | ||
5ead97c8 JF |
690 | if (cr3 == x86_read_percpu(xen_cr3)) { |
691 | /* just a simple tlb flush */ | |
692 | xen_flush_tlb(); | |
693 | return; | |
694 | } | |
695 | ||
696 | x86_write_percpu(xen_cr3, cr3); | |
697 | ||
698 | ||
699 | { | |
700 | struct mmuext_op *op; | |
701 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
702 | unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
703 | ||
704 | op = mcs.args; | |
705 | op->cmd = MMUEXT_NEW_BASEPTR; | |
706 | op->arg1.mfn = mfn; | |
707 | ||
708 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
709 | ||
710 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
711 | } | |
712 | } | |
713 | ||
f4f97b3e JF |
714 | /* Early in boot, while setting up the initial pagetable, assume |
715 | everything is pinned. */ | |
9a4029fd | 716 | static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn) |
5ead97c8 | 717 | { |
f4f97b3e | 718 | BUG_ON(mem_map); /* should only be used early */ |
5ead97c8 JF |
719 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
720 | } | |
721 | ||
f4f97b3e JF |
722 | /* This needs to make sure the new pte page is pinned iff its being |
723 | attached to a pinned pagetable. */ | |
724 | static void xen_alloc_pt(struct mm_struct *mm, u32 pfn) | |
5ead97c8 | 725 | { |
f4f97b3e | 726 | struct page *page = pfn_to_page(pfn); |
5ead97c8 | 727 | |
f4f97b3e JF |
728 | if (PagePinned(virt_to_page(mm->pgd))) { |
729 | SetPagePinned(page); | |
730 | ||
731 | if (!PageHighMem(page)) | |
732 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
733 | else | |
734 | /* make sure there are no stray mappings of | |
735 | this page */ | |
736 | kmap_flush_unused(); | |
737 | } | |
5ead97c8 JF |
738 | } |
739 | ||
f4f97b3e | 740 | /* This should never happen until we're OK to use struct page */ |
5ead97c8 JF |
741 | static void xen_release_pt(u32 pfn) |
742 | { | |
f4f97b3e JF |
743 | struct page *page = pfn_to_page(pfn); |
744 | ||
745 | if (PagePinned(page)) { | |
746 | if (!PageHighMem(page)) | |
747 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
748 | } | |
5ead97c8 JF |
749 | } |
750 | ||
f4f97b3e JF |
751 | #ifdef CONFIG_HIGHPTE |
752 | static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) | |
5ead97c8 | 753 | { |
f4f97b3e JF |
754 | pgprot_t prot = PAGE_KERNEL; |
755 | ||
756 | if (PagePinned(page)) | |
757 | prot = PAGE_KERNEL_RO; | |
758 | ||
759 | if (0 && PageHighMem(page)) | |
760 | printk("mapping highpte %lx type %d prot %s\n", | |
761 | page_to_pfn(page), type, | |
762 | (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); | |
763 | ||
764 | return kmap_atomic_prot(page, type, prot); | |
5ead97c8 | 765 | } |
f4f97b3e | 766 | #endif |
5ead97c8 | 767 | |
9a4029fd JF |
768 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) |
769 | { | |
770 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
771 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
772 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
773 | pte_val_ma(pte)); | |
774 | ||
775 | return pte; | |
776 | } | |
777 | ||
778 | /* Init-time set_pte while constructing initial pagetables, which | |
779 | doesn't allow RO pagetable pages to be remapped RW */ | |
780 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
781 | { | |
782 | pte = mask_rw_pte(ptep, pte); | |
783 | ||
784 | xen_set_pte(ptep, pte); | |
785 | } | |
786 | ||
5ead97c8 JF |
787 | static __init void xen_pagetable_setup_start(pgd_t *base) |
788 | { | |
789 | pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base; | |
790 | ||
9a4029fd JF |
791 | /* special set_pte for pagetable initialization */ |
792 | paravirt_ops.set_pte = xen_set_pte_init; | |
793 | ||
5ead97c8 JF |
794 | init_mm.pgd = base; |
795 | /* | |
796 | * copy top-level of Xen-supplied pagetable into place. For | |
797 | * !PAE we can use this as-is, but for PAE it is a stand-in | |
798 | * while we copy the pmd pages. | |
799 | */ | |
800 | memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t)); | |
801 | ||
802 | if (PTRS_PER_PMD > 1) { | |
803 | int i; | |
804 | /* | |
805 | * For PAE, need to allocate new pmds, rather than | |
806 | * share Xen's, since Xen doesn't like pmd's being | |
807 | * shared between address spaces. | |
808 | */ | |
809 | for (i = 0; i < PTRS_PER_PGD; i++) { | |
810 | if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) { | |
811 | pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE); | |
812 | ||
813 | memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]), | |
814 | PAGE_SIZE); | |
815 | ||
f4f97b3e | 816 | make_lowmem_page_readonly(pmd); |
5ead97c8 JF |
817 | |
818 | set_pgd(&base[i], __pgd(1 + __pa(pmd))); | |
819 | } else | |
820 | pgd_clear(&base[i]); | |
821 | } | |
822 | } | |
823 | ||
824 | /* make sure zero_page is mapped RO so we can use it in pagetables */ | |
825 | make_lowmem_page_readonly(empty_zero_page); | |
826 | make_lowmem_page_readonly(base); | |
827 | /* | |
828 | * Switch to new pagetable. This is done before | |
829 | * pagetable_init has done anything so that the new pages | |
830 | * added to the table can be prepared properly for Xen. | |
831 | */ | |
832 | xen_write_cr3(__pa(base)); | |
833 | } | |
834 | ||
835 | static __init void xen_pagetable_setup_done(pgd_t *base) | |
836 | { | |
f4f97b3e JF |
837 | /* This will work as long as patching hasn't happened yet |
838 | (which it hasn't) */ | |
839 | paravirt_ops.alloc_pt = xen_alloc_pt; | |
9a4029fd | 840 | paravirt_ops.set_pte = xen_set_pte; |
f4f97b3e | 841 | |
5ead97c8 JF |
842 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
843 | /* | |
844 | * Create a mapping for the shared info page. | |
845 | * Should be set_fixmap(), but shared_info is a machine | |
846 | * address with no corresponding pseudo-phys address. | |
847 | */ | |
5ead97c8 JF |
848 | set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP), |
849 | PFN_DOWN(xen_start_info->shared_info), | |
850 | PAGE_KERNEL); | |
5ead97c8 JF |
851 | |
852 | HYPERVISOR_shared_info = | |
853 | (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); | |
854 | ||
855 | } else | |
856 | HYPERVISOR_shared_info = | |
857 | (struct shared_info *)__va(xen_start_info->shared_info); | |
858 | ||
f4f97b3e JF |
859 | /* Actually pin the pagetable down, but we can't set PG_pinned |
860 | yet because the page structures don't exist yet. */ | |
861 | { | |
862 | struct mmuext_op op; | |
863 | #ifdef CONFIG_X86_PAE | |
864 | op.cmd = MMUEXT_PIN_L3_TABLE; | |
865 | #else | |
866 | op.cmd = MMUEXT_PIN_L3_TABLE; | |
867 | #endif | |
868 | op.arg1.mfn = pfn_to_mfn(PFN_DOWN(__pa(base))); | |
869 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
870 | BUG(); | |
871 | } | |
60223a32 | 872 | } |
5ead97c8 | 873 | |
60223a32 JF |
874 | /* This is called once we have the cpu_possible_map */ |
875 | void __init xen_setup_vcpu_info_placement(void) | |
876 | { | |
877 | int cpu; | |
878 | ||
879 | for_each_possible_cpu(cpu) | |
880 | xen_vcpu_setup(cpu); | |
881 | ||
882 | /* xen_vcpu_setup managed to place the vcpu_info within the | |
883 | percpu area for all cpus, so make use of it */ | |
884 | if (have_vcpu_info_placement) { | |
885 | printk(KERN_INFO "Xen: using vcpu_info placement\n"); | |
886 | ||
887 | paravirt_ops.save_fl = xen_save_fl_direct; | |
888 | paravirt_ops.restore_fl = xen_restore_fl_direct; | |
889 | paravirt_ops.irq_disable = xen_irq_disable_direct; | |
890 | paravirt_ops.irq_enable = xen_irq_enable_direct; | |
891 | paravirt_ops.read_cr2 = xen_read_cr2_direct; | |
892 | } | |
5ead97c8 JF |
893 | } |
894 | ||
895 | static const struct paravirt_ops xen_paravirt_ops __initdata = { | |
896 | .paravirt_enabled = 1, | |
897 | .shared_kernel_pmd = 0, | |
898 | ||
899 | .name = "Xen", | |
900 | .banner = xen_banner, | |
901 | ||
902 | .patch = paravirt_patch_default, | |
903 | ||
904 | .memory_setup = xen_memory_setup, | |
905 | .arch_setup = xen_arch_setup, | |
e46cdb66 | 906 | .init_IRQ = xen_init_IRQ, |
f4f97b3e | 907 | .post_allocator_init = xen_mark_init_mm_pinned, |
5ead97c8 | 908 | |
15c84731 JF |
909 | .time_init = xen_time_init, |
910 | .set_wallclock = xen_set_wallclock, | |
911 | .get_wallclock = xen_get_wallclock, | |
912 | .get_cpu_khz = xen_cpu_khz, | |
ab550288 | 913 | .sched_clock = xen_sched_clock, |
15c84731 | 914 | |
5ead97c8 JF |
915 | .cpuid = xen_cpuid, |
916 | ||
917 | .set_debugreg = xen_set_debugreg, | |
918 | .get_debugreg = xen_get_debugreg, | |
919 | ||
920 | .clts = native_clts, | |
921 | ||
922 | .read_cr0 = native_read_cr0, | |
923 | .write_cr0 = native_write_cr0, | |
924 | ||
925 | .read_cr2 = xen_read_cr2, | |
60223a32 | 926 | .write_cr2 = xen_write_cr2, |
5ead97c8 JF |
927 | |
928 | .read_cr3 = xen_read_cr3, | |
929 | .write_cr3 = xen_write_cr3, | |
930 | ||
931 | .read_cr4 = native_read_cr4, | |
932 | .read_cr4_safe = native_read_cr4_safe, | |
933 | .write_cr4 = xen_write_cr4, | |
934 | ||
935 | .save_fl = xen_save_fl, | |
936 | .restore_fl = xen_restore_fl, | |
937 | .irq_disable = xen_irq_disable, | |
938 | .irq_enable = xen_irq_enable, | |
939 | .safe_halt = xen_safe_halt, | |
940 | .halt = xen_halt, | |
941 | .wbinvd = native_wbinvd, | |
942 | ||
943 | .read_msr = native_read_msr_safe, | |
944 | .write_msr = native_write_msr_safe, | |
945 | .read_tsc = native_read_tsc, | |
946 | .read_pmc = native_read_pmc, | |
947 | ||
948 | .iret = (void *)&hypercall_page[__HYPERVISOR_iret], | |
949 | .irq_enable_sysexit = NULL, /* never called */ | |
950 | ||
951 | .load_tr_desc = paravirt_nop, | |
952 | .set_ldt = xen_set_ldt, | |
953 | .load_gdt = xen_load_gdt, | |
954 | .load_idt = xen_load_idt, | |
955 | .load_tls = xen_load_tls, | |
956 | ||
957 | .store_gdt = native_store_gdt, | |
958 | .store_idt = native_store_idt, | |
959 | .store_tr = xen_store_tr, | |
960 | ||
961 | .write_ldt_entry = xen_write_ldt_entry, | |
962 | .write_gdt_entry = xen_write_gdt_entry, | |
963 | .write_idt_entry = xen_write_idt_entry, | |
964 | .load_esp0 = xen_load_esp0, | |
965 | ||
966 | .set_iopl_mask = xen_set_iopl_mask, | |
967 | .io_delay = xen_io_delay, | |
968 | ||
969 | #ifdef CONFIG_X86_LOCAL_APIC | |
f87e4cac JF |
970 | .apic_write = xen_apic_write, |
971 | .apic_write_atomic = xen_apic_write, | |
5ead97c8 JF |
972 | .apic_read = xen_apic_read, |
973 | .setup_boot_clock = paravirt_nop, | |
974 | .setup_secondary_clock = paravirt_nop, | |
975 | .startup_ipi_hook = paravirt_nop, | |
976 | #endif | |
977 | ||
978 | .flush_tlb_user = xen_flush_tlb, | |
979 | .flush_tlb_kernel = xen_flush_tlb, | |
980 | .flush_tlb_single = xen_flush_tlb_single, | |
f87e4cac | 981 | .flush_tlb_others = xen_flush_tlb_others, |
5ead97c8 JF |
982 | |
983 | .pte_update = paravirt_nop, | |
984 | .pte_update_defer = paravirt_nop, | |
985 | ||
986 | .pagetable_setup_start = xen_pagetable_setup_start, | |
987 | .pagetable_setup_done = xen_pagetable_setup_done, | |
988 | ||
f4f97b3e | 989 | .alloc_pt = xen_alloc_pt_init, |
5ead97c8 | 990 | .release_pt = xen_release_pt, |
f4f97b3e JF |
991 | .alloc_pd = paravirt_nop, |
992 | .alloc_pd_clone = paravirt_nop, | |
993 | .release_pd = paravirt_nop, | |
994 | ||
995 | #ifdef CONFIG_HIGHPTE | |
996 | .kmap_atomic_pte = xen_kmap_atomic_pte, | |
997 | #endif | |
5ead97c8 | 998 | |
9a4029fd | 999 | .set_pte = NULL, /* see xen_pagetable_setup_* */ |
3b827c1b JF |
1000 | .set_pte_at = xen_set_pte_at, |
1001 | .set_pmd = xen_set_pmd, | |
1002 | ||
1003 | .pte_val = xen_pte_val, | |
1004 | .pgd_val = xen_pgd_val, | |
1005 | ||
1006 | .make_pte = xen_make_pte, | |
1007 | .make_pgd = xen_make_pgd, | |
1008 | ||
1009 | #ifdef CONFIG_X86_PAE | |
1010 | .set_pte_atomic = xen_set_pte_atomic, | |
1011 | .set_pte_present = xen_set_pte_at, | |
1012 | .set_pud = xen_set_pud, | |
1013 | .pte_clear = xen_pte_clear, | |
1014 | .pmd_clear = xen_pmd_clear, | |
1015 | ||
1016 | .make_pmd = xen_make_pmd, | |
1017 | .pmd_val = xen_pmd_val, | |
1018 | #endif /* PAE */ | |
1019 | ||
1020 | .activate_mm = xen_activate_mm, | |
1021 | .dup_mmap = xen_dup_mmap, | |
1022 | .exit_mmap = xen_exit_mmap, | |
1023 | ||
5ead97c8 JF |
1024 | .set_lazy_mode = xen_set_lazy_mode, |
1025 | }; | |
1026 | ||
f87e4cac JF |
1027 | #ifdef CONFIG_SMP |
1028 | static const struct smp_ops xen_smp_ops __initdata = { | |
1029 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, | |
1030 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
1031 | .cpu_up = xen_cpu_up, | |
1032 | .smp_cpus_done = xen_smp_cpus_done, | |
1033 | ||
1034 | .smp_send_stop = xen_smp_send_stop, | |
1035 | .smp_send_reschedule = xen_smp_send_reschedule, | |
1036 | .smp_call_function_mask = xen_smp_call_function_mask, | |
1037 | }; | |
1038 | #endif /* CONFIG_SMP */ | |
1039 | ||
fefa629a JF |
1040 | static void xen_reboot(int reason) |
1041 | { | |
1042 | #ifdef CONFIG_SMP | |
1043 | smp_send_stop(); | |
1044 | #endif | |
1045 | ||
1046 | if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason)) | |
1047 | BUG(); | |
1048 | } | |
1049 | ||
1050 | static void xen_restart(char *msg) | |
1051 | { | |
1052 | xen_reboot(SHUTDOWN_reboot); | |
1053 | } | |
1054 | ||
1055 | static void xen_emergency_restart(void) | |
1056 | { | |
1057 | xen_reboot(SHUTDOWN_reboot); | |
1058 | } | |
1059 | ||
1060 | static void xen_machine_halt(void) | |
1061 | { | |
1062 | xen_reboot(SHUTDOWN_poweroff); | |
1063 | } | |
1064 | ||
1065 | static void xen_crash_shutdown(struct pt_regs *regs) | |
1066 | { | |
1067 | xen_reboot(SHUTDOWN_crash); | |
1068 | } | |
1069 | ||
1070 | static const struct machine_ops __initdata xen_machine_ops = { | |
1071 | .restart = xen_restart, | |
1072 | .halt = xen_machine_halt, | |
1073 | .power_off = xen_machine_halt, | |
1074 | .shutdown = xen_machine_halt, | |
1075 | .crash_shutdown = xen_crash_shutdown, | |
1076 | .emergency_restart = xen_emergency_restart, | |
1077 | }; | |
1078 | ||
5ead97c8 JF |
1079 | /* First C function to be called on Xen boot */ |
1080 | asmlinkage void __init xen_start_kernel(void) | |
1081 | { | |
1082 | pgd_t *pgd; | |
1083 | ||
1084 | if (!xen_start_info) | |
1085 | return; | |
1086 | ||
1087 | BUG_ON(memcmp(xen_start_info->magic, "xen-3.0", 7) != 0); | |
1088 | ||
1089 | /* Install Xen paravirt ops */ | |
1090 | paravirt_ops = xen_paravirt_ops; | |
fefa629a JF |
1091 | machine_ops = xen_machine_ops; |
1092 | ||
f87e4cac JF |
1093 | #ifdef CONFIG_SMP |
1094 | smp_ops = xen_smp_ops; | |
1095 | #endif | |
5ead97c8 JF |
1096 | |
1097 | xen_setup_features(); | |
1098 | ||
1099 | /* Get mfn list */ | |
1100 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
1101 | phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list; | |
1102 | ||
1103 | pgd = (pgd_t *)xen_start_info->pt_base; | |
1104 | ||
1105 | init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE; | |
1106 | ||
1107 | init_mm.pgd = pgd; /* use the Xen pagetables to start */ | |
1108 | ||
1109 | /* keep using Xen gdt for now; no urgent need to change it */ | |
1110 | ||
1111 | x86_write_percpu(xen_cr3, __pa(pgd)); | |
60223a32 JF |
1112 | |
1113 | #ifdef CONFIG_SMP | |
1114 | /* Don't do the full vcpu_info placement stuff until we have a | |
1115 | possible map. */ | |
1116 | per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; | |
1117 | #else | |
1118 | /* May as well do it now, since there's no good time to call | |
1119 | it later on UP. */ | |
1120 | xen_setup_vcpu_info_placement(); | |
1121 | #endif | |
5ead97c8 JF |
1122 | |
1123 | paravirt_ops.kernel_rpl = 1; | |
1124 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) | |
1125 | paravirt_ops.kernel_rpl = 0; | |
1126 | ||
1127 | /* set the limit of our address space */ | |
1128 | reserve_top_address(-HYPERVISOR_VIRT_START + 2 * PAGE_SIZE); | |
1129 | ||
1130 | /* set up basic CPUID stuff */ | |
1131 | cpu_detect(&new_cpu_data); | |
1132 | new_cpu_data.hard_math = 1; | |
1133 | new_cpu_data.x86_capability[0] = cpuid_edx(1); | |
1134 | ||
1135 | /* Poke various useful things into boot_params */ | |
1136 | LOADER_TYPE = (9 << 4) | 0; | |
1137 | INITRD_START = xen_start_info->mod_start ? __pa(xen_start_info->mod_start) : 0; | |
1138 | INITRD_SIZE = xen_start_info->mod_len; | |
1139 | ||
1140 | /* Start the world */ | |
1141 | start_kernel(); | |
1142 | } |