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> | |
31 | #include <xen/interface/physdev.h> | |
32 | #include <xen/interface/vcpu.h> | |
33 | #include <xen/features.h> | |
34 | #include <xen/page.h> | |
084a2a4e | 35 | #include <xen/hvc-console.h> |
5ead97c8 JF |
36 | |
37 | #include <asm/paravirt.h> | |
caf43bf7 | 38 | #include <asm/apic.h> |
5ead97c8 JF |
39 | #include <asm/page.h> |
40 | #include <asm/xen/hypercall.h> | |
41 | #include <asm/xen/hypervisor.h> | |
42 | #include <asm/fixmap.h> | |
43 | #include <asm/processor.h> | |
1153968a | 44 | #include <asm/msr-index.h> |
5ead97c8 JF |
45 | #include <asm/setup.h> |
46 | #include <asm/desc.h> | |
47 | #include <asm/pgtable.h> | |
f87e4cac | 48 | #include <asm/tlbflush.h> |
fefa629a | 49 | #include <asm/reboot.h> |
5ead97c8 JF |
50 | |
51 | #include "xen-ops.h" | |
3b827c1b | 52 | #include "mmu.h" |
5ead97c8 JF |
53 | #include "multicalls.h" |
54 | ||
55 | EXPORT_SYMBOL_GPL(hypercall_page); | |
56 | ||
5ead97c8 JF |
57 | DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); |
58 | DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); | |
9f79991d | 59 | |
6e833587 JF |
60 | enum xen_domain_type xen_domain_type = XEN_NATIVE; |
61 | EXPORT_SYMBOL_GPL(xen_domain_type); | |
62 | ||
bf18bf94 JF |
63 | /* |
64 | * Identity map, in addition to plain kernel map. This needs to be | |
65 | * large enough to allocate page table pages to allocate the rest. | |
66 | * Each page can map 2MB. | |
67 | */ | |
68 | static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss; | |
69 | ||
70 | #ifdef CONFIG_X86_64 | |
71 | /* l3 pud for userspace vsyscall mapping */ | |
72 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
73 | #endif /* CONFIG_X86_64 */ | |
74 | ||
9f79991d JF |
75 | /* |
76 | * Note about cr3 (pagetable base) values: | |
77 | * | |
78 | * xen_cr3 contains the current logical cr3 value; it contains the | |
79 | * last set cr3. This may not be the current effective cr3, because | |
80 | * its update may be being lazily deferred. However, a vcpu looking | |
81 | * at its own cr3 can use this value knowing that it everything will | |
82 | * be self-consistent. | |
83 | * | |
84 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
85 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
86 | * out of date, but it will never be set early). If one vcpu is | |
87 | * looking at another vcpu's cr3 value, it should use this variable. | |
88 | */ | |
89 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
90 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
5ead97c8 JF |
91 | |
92 | struct start_info *xen_start_info; | |
93 | EXPORT_SYMBOL_GPL(xen_start_info); | |
94 | ||
a0d695c8 | 95 | struct shared_info xen_dummy_shared_info; |
60223a32 JF |
96 | |
97 | /* | |
98 | * Point at some empty memory to start with. We map the real shared_info | |
99 | * page as soon as fixmap is up and running. | |
100 | */ | |
a0d695c8 | 101 | struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; |
60223a32 JF |
102 | |
103 | /* | |
104 | * Flag to determine whether vcpu info placement is available on all | |
105 | * VCPUs. We assume it is to start with, and then set it to zero on | |
106 | * the first failure. This is because it can succeed on some VCPUs | |
107 | * and not others, since it can involve hypervisor memory allocation, | |
108 | * or because the guest failed to guarantee all the appropriate | |
109 | * constraints on all VCPUs (ie buffer can't cross a page boundary). | |
110 | * | |
111 | * Note that any particular CPU may be using a placed vcpu structure, | |
112 | * but we can only optimise if the all are. | |
113 | * | |
114 | * 0: not available, 1: available | |
115 | */ | |
db053b86 JF |
116 | static int have_vcpu_info_placement = |
117 | #ifdef CONFIG_X86_32 | |
118 | 1 | |
119 | #else | |
120 | 0 | |
121 | #endif | |
122 | ; | |
123 | ||
60223a32 | 124 | |
9c7a7942 | 125 | static void xen_vcpu_setup(int cpu) |
5ead97c8 | 126 | { |
60223a32 JF |
127 | struct vcpu_register_vcpu_info info; |
128 | int err; | |
129 | struct vcpu_info *vcpup; | |
130 | ||
a0d695c8 | 131 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
5ead97c8 | 132 | per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; |
60223a32 JF |
133 | |
134 | if (!have_vcpu_info_placement) | |
135 | return; /* already tested, not available */ | |
136 | ||
137 | vcpup = &per_cpu(xen_vcpu_info, cpu); | |
138 | ||
139 | info.mfn = virt_to_mfn(vcpup); | |
140 | info.offset = offset_in_page(vcpup); | |
141 | ||
e3d26976 | 142 | printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", |
60223a32 JF |
143 | cpu, vcpup, info.mfn, info.offset); |
144 | ||
145 | /* Check to see if the hypervisor will put the vcpu_info | |
146 | structure where we want it, which allows direct access via | |
147 | a percpu-variable. */ | |
148 | err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); | |
149 | ||
150 | if (err) { | |
151 | printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); | |
152 | have_vcpu_info_placement = 0; | |
153 | } else { | |
154 | /* This cpu is using the registered vcpu info, even if | |
155 | later ones fail to. */ | |
156 | per_cpu(xen_vcpu, cpu) = vcpup; | |
6487673b | 157 | |
60223a32 JF |
158 | printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", |
159 | cpu, vcpup); | |
160 | } | |
5ead97c8 JF |
161 | } |
162 | ||
9c7a7942 JF |
163 | /* |
164 | * On restore, set the vcpu placement up again. | |
165 | * If it fails, then we're in a bad state, since | |
166 | * we can't back out from using it... | |
167 | */ | |
168 | void xen_vcpu_restore(void) | |
169 | { | |
170 | if (have_vcpu_info_placement) { | |
171 | int cpu; | |
172 | ||
173 | for_each_online_cpu(cpu) { | |
174 | bool other_cpu = (cpu != smp_processor_id()); | |
175 | ||
176 | if (other_cpu && | |
177 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) | |
178 | BUG(); | |
179 | ||
180 | xen_vcpu_setup(cpu); | |
181 | ||
182 | if (other_cpu && | |
183 | HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) | |
184 | BUG(); | |
185 | } | |
186 | ||
187 | BUG_ON(!have_vcpu_info_placement); | |
188 | } | |
189 | } | |
190 | ||
5ead97c8 JF |
191 | static void __init xen_banner(void) |
192 | { | |
95c7c23b JF |
193 | unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); |
194 | struct xen_extraversion extra; | |
195 | HYPERVISOR_xen_version(XENVER_extraversion, &extra); | |
196 | ||
5ead97c8 | 197 | printk(KERN_INFO "Booting paravirtualized kernel on %s\n", |
93b1eab3 | 198 | pv_info.name); |
95c7c23b JF |
199 | printk(KERN_INFO "Xen version: %d.%d%s%s\n", |
200 | version >> 16, version & 0xffff, extra.extraversion, | |
e57778a1 | 201 | xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); |
5ead97c8 JF |
202 | } |
203 | ||
65ea5b03 PA |
204 | static void xen_cpuid(unsigned int *ax, unsigned int *bx, |
205 | unsigned int *cx, unsigned int *dx) | |
5ead97c8 JF |
206 | { |
207 | unsigned maskedx = ~0; | |
208 | ||
209 | /* | |
210 | * Mask out inconvenient features, to try and disable as many | |
211 | * unsupported kernel subsystems as possible. | |
212 | */ | |
65ea5b03 | 213 | if (*ax == 1) |
5ead97c8 JF |
214 | maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */ |
215 | (1 << X86_FEATURE_ACPI) | /* disable ACPI */ | |
dbe9e994 JF |
216 | (1 << X86_FEATURE_MCE) | /* disable MCE */ |
217 | (1 << X86_FEATURE_MCA) | /* disable MCA */ | |
5ead97c8 JF |
218 | (1 << X86_FEATURE_ACC)); /* thermal monitoring */ |
219 | ||
220 | asm(XEN_EMULATE_PREFIX "cpuid" | |
65ea5b03 PA |
221 | : "=a" (*ax), |
222 | "=b" (*bx), | |
223 | "=c" (*cx), | |
224 | "=d" (*dx) | |
225 | : "0" (*ax), "2" (*cx)); | |
226 | *dx &= maskedx; | |
5ead97c8 JF |
227 | } |
228 | ||
229 | static void xen_set_debugreg(int reg, unsigned long val) | |
230 | { | |
231 | HYPERVISOR_set_debugreg(reg, val); | |
232 | } | |
233 | ||
234 | static unsigned long xen_get_debugreg(int reg) | |
235 | { | |
236 | return HYPERVISOR_get_debugreg(reg); | |
237 | } | |
238 | ||
8965c1c0 | 239 | static void xen_leave_lazy(void) |
5ead97c8 | 240 | { |
8965c1c0 | 241 | paravirt_leave_lazy(paravirt_get_lazy_mode()); |
5ead97c8 | 242 | xen_mc_flush(); |
5ead97c8 JF |
243 | } |
244 | ||
245 | static unsigned long xen_store_tr(void) | |
246 | { | |
247 | return 0; | |
248 | } | |
249 | ||
a05d2eba | 250 | /* |
cef43bf6 JF |
251 | * Set the page permissions for a particular virtual address. If the |
252 | * address is a vmalloc mapping (or other non-linear mapping), then | |
253 | * find the linear mapping of the page and also set its protections to | |
254 | * match. | |
a05d2eba JF |
255 | */ |
256 | static void set_aliased_prot(void *v, pgprot_t prot) | |
257 | { | |
258 | int level; | |
259 | pte_t *ptep; | |
260 | pte_t pte; | |
261 | unsigned long pfn; | |
262 | struct page *page; | |
263 | ||
264 | ptep = lookup_address((unsigned long)v, &level); | |
265 | BUG_ON(ptep == NULL); | |
266 | ||
267 | pfn = pte_pfn(*ptep); | |
268 | page = pfn_to_page(pfn); | |
269 | ||
270 | pte = pfn_pte(pfn, prot); | |
271 | ||
272 | if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) | |
273 | BUG(); | |
274 | ||
275 | if (!PageHighMem(page)) { | |
276 | void *av = __va(PFN_PHYS(pfn)); | |
277 | ||
278 | if (av != v) | |
279 | if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) | |
280 | BUG(); | |
281 | } else | |
282 | kmap_flush_unused(); | |
283 | } | |
284 | ||
38ffbe66 JF |
285 | static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) |
286 | { | |
a05d2eba | 287 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
288 | int i; |
289 | ||
a05d2eba JF |
290 | for(i = 0; i < entries; i += entries_per_page) |
291 | set_aliased_prot(ldt + i, PAGE_KERNEL_RO); | |
38ffbe66 JF |
292 | } |
293 | ||
294 | static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) | |
295 | { | |
a05d2eba | 296 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; |
38ffbe66 JF |
297 | int i; |
298 | ||
a05d2eba JF |
299 | for(i = 0; i < entries; i += entries_per_page) |
300 | set_aliased_prot(ldt + i, PAGE_KERNEL); | |
38ffbe66 JF |
301 | } |
302 | ||
5ead97c8 JF |
303 | static void xen_set_ldt(const void *addr, unsigned entries) |
304 | { | |
5ead97c8 JF |
305 | struct mmuext_op *op; |
306 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
307 | ||
308 | op = mcs.args; | |
309 | op->cmd = MMUEXT_SET_LDT; | |
4dbf7af6 | 310 | op->arg1.linear_addr = (unsigned long)addr; |
5ead97c8 JF |
311 | op->arg2.nr_ents = entries; |
312 | ||
313 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
314 | ||
315 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
316 | } | |
317 | ||
6b68f01b | 318 | static void xen_load_gdt(const struct desc_ptr *dtr) |
5ead97c8 JF |
319 | { |
320 | unsigned long *frames; | |
321 | unsigned long va = dtr->address; | |
322 | unsigned int size = dtr->size + 1; | |
323 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
324 | int f; | |
325 | struct multicall_space mcs; | |
326 | ||
327 | /* A GDT can be up to 64k in size, which corresponds to 8192 | |
328 | 8-byte entries, or 16 4k pages.. */ | |
329 | ||
330 | BUG_ON(size > 65536); | |
331 | BUG_ON(va & ~PAGE_MASK); | |
332 | ||
333 | mcs = xen_mc_entry(sizeof(*frames) * pages); | |
334 | frames = mcs.args; | |
335 | ||
336 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
337 | frames[f] = virt_to_mfn(va); | |
338 | make_lowmem_page_readonly((void *)va); | |
339 | } | |
340 | ||
341 | MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct)); | |
342 | ||
343 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
344 | } | |
345 | ||
346 | static void load_TLS_descriptor(struct thread_struct *t, | |
347 | unsigned int cpu, unsigned int i) | |
348 | { | |
349 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
350 | xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); | |
351 | struct multicall_space mc = __xen_mc_entry(0); | |
352 | ||
353 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
354 | } | |
355 | ||
356 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
357 | { | |
8b84ad94 JF |
358 | /* |
359 | * XXX sleazy hack: If we're being called in a lazy-cpu zone, | |
360 | * it means we're in a context switch, and %gs has just been | |
361 | * saved. This means we can zero it out to prevent faults on | |
362 | * exit from the hypervisor if the next process has no %gs. | |
363 | * Either way, it has been saved, and the new value will get | |
364 | * loaded properly. This will go away as soon as Xen has been | |
365 | * modified to not save/restore %gs for normal hypercalls. | |
8a95408e EH |
366 | * |
367 | * On x86_64, this hack is not used for %gs, because gs points | |
368 | * to KERNEL_GS_BASE (and uses it for PDA references), so we | |
369 | * must not zero %gs on x86_64 | |
370 | * | |
371 | * For x86_64, we need to zero %fs, otherwise we may get an | |
372 | * exception between the new %fs descriptor being loaded and | |
373 | * %fs being effectively cleared at __switch_to(). | |
8b84ad94 | 374 | */ |
8a95408e EH |
375 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { |
376 | #ifdef CONFIG_X86_32 | |
8b84ad94 | 377 | loadsegment(gs, 0); |
8a95408e EH |
378 | #else |
379 | loadsegment(fs, 0); | |
380 | #endif | |
381 | } | |
382 | ||
383 | xen_mc_batch(); | |
384 | ||
385 | load_TLS_descriptor(t, cpu, 0); | |
386 | load_TLS_descriptor(t, cpu, 1); | |
387 | load_TLS_descriptor(t, cpu, 2); | |
388 | ||
389 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
5ead97c8 JF |
390 | } |
391 | ||
a8fc1089 EH |
392 | #ifdef CONFIG_X86_64 |
393 | static void xen_load_gs_index(unsigned int idx) | |
394 | { | |
395 | if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) | |
396 | BUG(); | |
5ead97c8 | 397 | } |
a8fc1089 | 398 | #endif |
5ead97c8 JF |
399 | |
400 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
75b8bb3e | 401 | const void *ptr) |
5ead97c8 | 402 | { |
cef43bf6 | 403 | xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); |
75b8bb3e | 404 | u64 entry = *(u64 *)ptr; |
5ead97c8 | 405 | |
f120f13e JF |
406 | preempt_disable(); |
407 | ||
5ead97c8 JF |
408 | xen_mc_flush(); |
409 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
410 | BUG(); | |
f120f13e JF |
411 | |
412 | preempt_enable(); | |
5ead97c8 JF |
413 | } |
414 | ||
e176d367 | 415 | static int cvt_gate_to_trap(int vector, const gate_desc *val, |
5ead97c8 JF |
416 | struct trap_info *info) |
417 | { | |
e176d367 | 418 | if (val->type != 0xf && val->type != 0xe) |
5ead97c8 JF |
419 | return 0; |
420 | ||
421 | info->vector = vector; | |
e176d367 EH |
422 | info->address = gate_offset(*val); |
423 | info->cs = gate_segment(*val); | |
424 | info->flags = val->dpl; | |
5ead97c8 | 425 | /* interrupt gates clear IF */ |
e176d367 | 426 | if (val->type == 0xe) |
5ead97c8 JF |
427 | info->flags |= 4; |
428 | ||
429 | return 1; | |
430 | } | |
431 | ||
432 | /* Locations of each CPU's IDT */ | |
6b68f01b | 433 | static DEFINE_PER_CPU(struct desc_ptr, idt_desc); |
5ead97c8 JF |
434 | |
435 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
436 | also update Xen. */ | |
8d947344 | 437 | static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) |
5ead97c8 | 438 | { |
5ead97c8 | 439 | unsigned long p = (unsigned long)&dt[entrynum]; |
f120f13e JF |
440 | unsigned long start, end; |
441 | ||
442 | preempt_disable(); | |
443 | ||
444 | start = __get_cpu_var(idt_desc).address; | |
445 | end = start + __get_cpu_var(idt_desc).size + 1; | |
5ead97c8 JF |
446 | |
447 | xen_mc_flush(); | |
448 | ||
8d947344 | 449 | native_write_idt_entry(dt, entrynum, g); |
5ead97c8 JF |
450 | |
451 | if (p >= start && (p + 8) <= end) { | |
452 | struct trap_info info[2]; | |
453 | ||
454 | info[1].address = 0; | |
455 | ||
e176d367 | 456 | if (cvt_gate_to_trap(entrynum, g, &info[0])) |
5ead97c8 JF |
457 | if (HYPERVISOR_set_trap_table(info)) |
458 | BUG(); | |
459 | } | |
f120f13e JF |
460 | |
461 | preempt_enable(); | |
5ead97c8 JF |
462 | } |
463 | ||
6b68f01b | 464 | static void xen_convert_trap_info(const struct desc_ptr *desc, |
f87e4cac | 465 | struct trap_info *traps) |
5ead97c8 | 466 | { |
5ead97c8 JF |
467 | unsigned in, out, count; |
468 | ||
e176d367 | 469 | count = (desc->size+1) / sizeof(gate_desc); |
5ead97c8 JF |
470 | BUG_ON(count > 256); |
471 | ||
5ead97c8 | 472 | for (in = out = 0; in < count; in++) { |
e176d367 | 473 | gate_desc *entry = (gate_desc*)(desc->address) + in; |
5ead97c8 | 474 | |
e176d367 | 475 | if (cvt_gate_to_trap(in, entry, &traps[out])) |
5ead97c8 JF |
476 | out++; |
477 | } | |
478 | traps[out].address = 0; | |
f87e4cac JF |
479 | } |
480 | ||
481 | void xen_copy_trap_info(struct trap_info *traps) | |
482 | { | |
6b68f01b | 483 | const struct desc_ptr *desc = &__get_cpu_var(idt_desc); |
f87e4cac JF |
484 | |
485 | xen_convert_trap_info(desc, traps); | |
f87e4cac JF |
486 | } |
487 | ||
488 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
489 | hold a spinlock to protect the static traps[] array (static because | |
490 | it avoids allocation, and saves stack space). */ | |
6b68f01b | 491 | static void xen_load_idt(const struct desc_ptr *desc) |
f87e4cac JF |
492 | { |
493 | static DEFINE_SPINLOCK(lock); | |
494 | static struct trap_info traps[257]; | |
f87e4cac JF |
495 | |
496 | spin_lock(&lock); | |
497 | ||
f120f13e JF |
498 | __get_cpu_var(idt_desc) = *desc; |
499 | ||
f87e4cac | 500 | xen_convert_trap_info(desc, traps); |
5ead97c8 JF |
501 | |
502 | xen_mc_flush(); | |
503 | if (HYPERVISOR_set_trap_table(traps)) | |
504 | BUG(); | |
505 | ||
506 | spin_unlock(&lock); | |
507 | } | |
508 | ||
509 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
510 | they're handled differently. */ | |
511 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
014b15be | 512 | const void *desc, int type) |
5ead97c8 | 513 | { |
f120f13e JF |
514 | preempt_disable(); |
515 | ||
014b15be GOC |
516 | switch (type) { |
517 | case DESC_LDT: | |
518 | case DESC_TSS: | |
5ead97c8 JF |
519 | /* ignore */ |
520 | break; | |
521 | ||
522 | default: { | |
523 | xmaddr_t maddr = virt_to_machine(&dt[entry]); | |
5ead97c8 JF |
524 | |
525 | xen_mc_flush(); | |
014b15be | 526 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) |
5ead97c8 JF |
527 | BUG(); |
528 | } | |
529 | ||
530 | } | |
f120f13e JF |
531 | |
532 | preempt_enable(); | |
5ead97c8 JF |
533 | } |
534 | ||
faca6227 | 535 | static void xen_load_sp0(struct tss_struct *tss, |
a05d2eba | 536 | struct thread_struct *thread) |
5ead97c8 JF |
537 | { |
538 | struct multicall_space mcs = xen_mc_entry(0); | |
faca6227 | 539 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); |
5ead97c8 JF |
540 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
541 | } | |
542 | ||
543 | static void xen_set_iopl_mask(unsigned mask) | |
544 | { | |
545 | struct physdev_set_iopl set_iopl; | |
546 | ||
547 | /* Force the change at ring 0. */ | |
548 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
549 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
550 | } | |
551 | ||
552 | static void xen_io_delay(void) | |
553 | { | |
554 | } | |
555 | ||
556 | #ifdef CONFIG_X86_LOCAL_APIC | |
ad66dd34 | 557 | static u32 xen_apic_read(u32 reg) |
5ead97c8 JF |
558 | { |
559 | return 0; | |
560 | } | |
f87e4cac | 561 | |
ad66dd34 | 562 | static void xen_apic_write(u32 reg, u32 val) |
f87e4cac JF |
563 | { |
564 | /* Warn to see if there's any stray references */ | |
565 | WARN_ON(1); | |
566 | } | |
ad66dd34 | 567 | |
ad66dd34 SS |
568 | static u64 xen_apic_icr_read(void) |
569 | { | |
570 | return 0; | |
571 | } | |
572 | ||
573 | static void xen_apic_icr_write(u32 low, u32 id) | |
574 | { | |
575 | /* Warn to see if there's any stray references */ | |
576 | WARN_ON(1); | |
577 | } | |
578 | ||
579 | static void xen_apic_wait_icr_idle(void) | |
580 | { | |
581 | return; | |
582 | } | |
583 | ||
94a8c3c2 YL |
584 | static u32 xen_safe_apic_wait_icr_idle(void) |
585 | { | |
586 | return 0; | |
587 | } | |
588 | ||
ad66dd34 SS |
589 | static struct apic_ops xen_basic_apic_ops = { |
590 | .read = xen_apic_read, | |
591 | .write = xen_apic_write, | |
ad66dd34 SS |
592 | .icr_read = xen_apic_icr_read, |
593 | .icr_write = xen_apic_icr_write, | |
594 | .wait_icr_idle = xen_apic_wait_icr_idle, | |
94a8c3c2 | 595 | .safe_wait_icr_idle = xen_safe_apic_wait_icr_idle, |
ad66dd34 | 596 | }; |
ad66dd34 | 597 | |
5ead97c8 JF |
598 | #endif |
599 | ||
600 | static void xen_flush_tlb(void) | |
601 | { | |
d66bf8fc | 602 | struct mmuext_op *op; |
41e332b2 JF |
603 | struct multicall_space mcs; |
604 | ||
605 | preempt_disable(); | |
606 | ||
607 | mcs = xen_mc_entry(sizeof(*op)); | |
5ead97c8 | 608 | |
d66bf8fc JF |
609 | op = mcs.args; |
610 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
611 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
612 | ||
613 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
41e332b2 JF |
614 | |
615 | preempt_enable(); | |
5ead97c8 JF |
616 | } |
617 | ||
618 | static void xen_flush_tlb_single(unsigned long addr) | |
619 | { | |
d66bf8fc | 620 | struct mmuext_op *op; |
41e332b2 JF |
621 | struct multicall_space mcs; |
622 | ||
623 | preempt_disable(); | |
5ead97c8 | 624 | |
41e332b2 | 625 | mcs = xen_mc_entry(sizeof(*op)); |
d66bf8fc JF |
626 | op = mcs.args; |
627 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
628 | op->arg1.linear_addr = addr & PAGE_MASK; | |
629 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
630 | ||
631 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
41e332b2 JF |
632 | |
633 | preempt_enable(); | |
5ead97c8 JF |
634 | } |
635 | ||
f87e4cac JF |
636 | static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm, |
637 | unsigned long va) | |
638 | { | |
d66bf8fc JF |
639 | struct { |
640 | struct mmuext_op op; | |
641 | cpumask_t mask; | |
642 | } *args; | |
f87e4cac | 643 | cpumask_t cpumask = *cpus; |
d66bf8fc | 644 | struct multicall_space mcs; |
f87e4cac JF |
645 | |
646 | /* | |
647 | * A couple of (to be removed) sanity checks: | |
648 | * | |
649 | * - current CPU must not be in mask | |
650 | * - mask must exist :) | |
651 | */ | |
652 | BUG_ON(cpus_empty(cpumask)); | |
653 | BUG_ON(cpu_isset(smp_processor_id(), cpumask)); | |
654 | BUG_ON(!mm); | |
655 | ||
656 | /* If a CPU which we ran on has gone down, OK. */ | |
657 | cpus_and(cpumask, cpumask, cpu_online_map); | |
658 | if (cpus_empty(cpumask)) | |
659 | return; | |
660 | ||
d66bf8fc JF |
661 | mcs = xen_mc_entry(sizeof(*args)); |
662 | args = mcs.args; | |
663 | args->mask = cpumask; | |
664 | args->op.arg2.vcpumask = &args->mask; | |
665 | ||
f87e4cac | 666 | if (va == TLB_FLUSH_ALL) { |
d66bf8fc | 667 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; |
f87e4cac | 668 | } else { |
d66bf8fc JF |
669 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
670 | args->op.arg1.linear_addr = va; | |
f87e4cac JF |
671 | } |
672 | ||
d66bf8fc JF |
673 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); |
674 | ||
675 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f87e4cac JF |
676 | } |
677 | ||
7b1333aa JF |
678 | static void xen_clts(void) |
679 | { | |
680 | struct multicall_space mcs; | |
681 | ||
682 | mcs = xen_mc_entry(0); | |
683 | ||
684 | MULTI_fpu_taskswitch(mcs.mc, 0); | |
685 | ||
686 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
687 | } | |
688 | ||
689 | static void xen_write_cr0(unsigned long cr0) | |
690 | { | |
691 | struct multicall_space mcs; | |
692 | ||
693 | /* Only pay attention to cr0.TS; everything else is | |
694 | ignored. */ | |
695 | mcs = xen_mc_entry(0); | |
696 | ||
697 | MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); | |
698 | ||
699 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
700 | } | |
701 | ||
60223a32 JF |
702 | static void xen_write_cr2(unsigned long cr2) |
703 | { | |
704 | x86_read_percpu(xen_vcpu)->arch.cr2 = cr2; | |
705 | } | |
706 | ||
5ead97c8 JF |
707 | static unsigned long xen_read_cr2(void) |
708 | { | |
709 | return x86_read_percpu(xen_vcpu)->arch.cr2; | |
710 | } | |
711 | ||
60223a32 JF |
712 | static unsigned long xen_read_cr2_direct(void) |
713 | { | |
714 | return x86_read_percpu(xen_vcpu_info.arch.cr2); | |
715 | } | |
716 | ||
5ead97c8 JF |
717 | static void xen_write_cr4(unsigned long cr4) |
718 | { | |
2956a351 JF |
719 | cr4 &= ~X86_CR4_PGE; |
720 | cr4 &= ~X86_CR4_PSE; | |
721 | ||
722 | native_write_cr4(cr4); | |
5ead97c8 JF |
723 | } |
724 | ||
5ead97c8 JF |
725 | static unsigned long xen_read_cr3(void) |
726 | { | |
727 | return x86_read_percpu(xen_cr3); | |
728 | } | |
729 | ||
9f79991d JF |
730 | static void set_current_cr3(void *v) |
731 | { | |
732 | x86_write_percpu(xen_current_cr3, (unsigned long)v); | |
733 | } | |
734 | ||
d6182fbf | 735 | static void __xen_write_cr3(bool kernel, unsigned long cr3) |
5ead97c8 | 736 | { |
9f79991d JF |
737 | struct mmuext_op *op; |
738 | struct multicall_space mcs; | |
d6182fbf | 739 | unsigned long mfn; |
9f79991d | 740 | |
d6182fbf JF |
741 | if (cr3) |
742 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
743 | else | |
744 | mfn = 0; | |
f120f13e | 745 | |
d6182fbf | 746 | WARN_ON(mfn == 0 && kernel); |
5ead97c8 | 747 | |
d6182fbf | 748 | mcs = __xen_mc_entry(sizeof(*op)); |
5ead97c8 | 749 | |
9f79991d | 750 | op = mcs.args; |
d6182fbf | 751 | op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; |
9f79991d | 752 | op->arg1.mfn = mfn; |
5ead97c8 | 753 | |
9f79991d | 754 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); |
5ead97c8 | 755 | |
d6182fbf JF |
756 | if (kernel) { |
757 | x86_write_percpu(xen_cr3, cr3); | |
758 | ||
759 | /* Update xen_current_cr3 once the batch has actually | |
760 | been submitted. */ | |
761 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
762 | } | |
763 | } | |
764 | ||
765 | static void xen_write_cr3(unsigned long cr3) | |
766 | { | |
767 | BUG_ON(preemptible()); | |
768 | ||
769 | xen_mc_batch(); /* disables interrupts */ | |
770 | ||
771 | /* Update while interrupts are disabled, so its atomic with | |
772 | respect to ipis */ | |
773 | x86_write_percpu(xen_cr3, cr3); | |
774 | ||
775 | __xen_write_cr3(true, cr3); | |
776 | ||
777 | #ifdef CONFIG_X86_64 | |
778 | { | |
779 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
780 | if (user_pgd) | |
781 | __xen_write_cr3(false, __pa(user_pgd)); | |
782 | else | |
783 | __xen_write_cr3(false, 0); | |
784 | } | |
785 | #endif | |
5ead97c8 | 786 | |
9f79991d | 787 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ |
5ead97c8 JF |
788 | } |
789 | ||
1153968a JF |
790 | static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) |
791 | { | |
792 | int ret; | |
793 | ||
794 | ret = 0; | |
795 | ||
796 | switch(msr) { | |
797 | #ifdef CONFIG_X86_64 | |
798 | unsigned which; | |
799 | u64 base; | |
800 | ||
801 | case MSR_FS_BASE: which = SEGBASE_FS; goto set; | |
802 | case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; | |
803 | case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; | |
804 | ||
805 | set: | |
806 | base = ((u64)high << 32) | low; | |
807 | if (HYPERVISOR_set_segment_base(which, base) != 0) | |
808 | ret = -EFAULT; | |
809 | break; | |
810 | #endif | |
d89961e2 JF |
811 | |
812 | case MSR_STAR: | |
813 | case MSR_CSTAR: | |
814 | case MSR_LSTAR: | |
815 | case MSR_SYSCALL_MASK: | |
816 | case MSR_IA32_SYSENTER_CS: | |
817 | case MSR_IA32_SYSENTER_ESP: | |
818 | case MSR_IA32_SYSENTER_EIP: | |
819 | /* Fast syscall setup is all done in hypercalls, so | |
820 | these are all ignored. Stub them out here to stop | |
821 | Xen console noise. */ | |
822 | break; | |
823 | ||
1153968a JF |
824 | default: |
825 | ret = native_write_msr_safe(msr, low, high); | |
826 | } | |
827 | ||
828 | return ret; | |
829 | } | |
830 | ||
f4f97b3e JF |
831 | /* Early in boot, while setting up the initial pagetable, assume |
832 | everything is pinned. */ | |
f8639939 | 833 | static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) |
5ead97c8 | 834 | { |
af7ae3b9 | 835 | #ifdef CONFIG_FLATMEM |
f4f97b3e | 836 | BUG_ON(mem_map); /* should only be used early */ |
af7ae3b9 | 837 | #endif |
5ead97c8 JF |
838 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
839 | } | |
840 | ||
6944a9c8 | 841 | /* Early release_pte assumes that all pts are pinned, since there's |
1c70e9bd | 842 | only init_mm and anything attached to that is pinned. */ |
f8639939 | 843 | static void xen_release_pte_init(unsigned long pfn) |
1c70e9bd JF |
844 | { |
845 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
846 | } | |
847 | ||
f6433706 | 848 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
74260714 JF |
849 | { |
850 | struct mmuext_op op; | |
f6433706 | 851 | op.cmd = cmd; |
74260714 JF |
852 | op.arg1.mfn = pfn_to_mfn(pfn); |
853 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
854 | BUG(); | |
855 | } | |
856 | ||
f4f97b3e JF |
857 | /* This needs to make sure the new pte page is pinned iff its being |
858 | attached to a pinned pagetable. */ | |
f8639939 | 859 | static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level) |
5ead97c8 | 860 | { |
f4f97b3e | 861 | struct page *page = pfn_to_page(pfn); |
5ead97c8 | 862 | |
f4f97b3e JF |
863 | if (PagePinned(virt_to_page(mm->pgd))) { |
864 | SetPagePinned(page); | |
865 | ||
74260714 | 866 | if (!PageHighMem(page)) { |
169ad16b | 867 | make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn))); |
6a9e9184 | 868 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) |
f6433706 | 869 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); |
74260714 | 870 | } else |
f4f97b3e JF |
871 | /* make sure there are no stray mappings of |
872 | this page */ | |
873 | kmap_flush_unused(); | |
874 | } | |
5ead97c8 JF |
875 | } |
876 | ||
f8639939 | 877 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) |
1c70e9bd | 878 | { |
f6433706 | 879 | xen_alloc_ptpage(mm, pfn, PT_PTE); |
1c70e9bd JF |
880 | } |
881 | ||
f8639939 | 882 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) |
1c70e9bd | 883 | { |
f6433706 | 884 | xen_alloc_ptpage(mm, pfn, PT_PMD); |
1c70e9bd JF |
885 | } |
886 | ||
d6182fbf JF |
887 | static int xen_pgd_alloc(struct mm_struct *mm) |
888 | { | |
889 | pgd_t *pgd = mm->pgd; | |
890 | int ret = 0; | |
891 | ||
892 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
893 | ||
894 | #ifdef CONFIG_X86_64 | |
895 | { | |
896 | struct page *page = virt_to_page(pgd); | |
bf18bf94 | 897 | pgd_t *user_pgd; |
d6182fbf JF |
898 | |
899 | BUG_ON(page->private != 0); | |
900 | ||
bf18bf94 JF |
901 | ret = -ENOMEM; |
902 | ||
903 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
904 | page->private = (unsigned long)user_pgd; | |
905 | ||
906 | if (user_pgd != NULL) { | |
907 | user_pgd[pgd_index(VSYSCALL_START)] = | |
908 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
909 | ret = 0; | |
910 | } | |
d6182fbf JF |
911 | |
912 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
913 | } | |
914 | #endif | |
915 | ||
916 | return ret; | |
917 | } | |
918 | ||
919 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
920 | { | |
921 | #ifdef CONFIG_X86_64 | |
922 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
923 | ||
924 | if (user_pgd) | |
925 | free_page((unsigned long)user_pgd); | |
926 | #endif | |
927 | } | |
928 | ||
f4f97b3e | 929 | /* This should never happen until we're OK to use struct page */ |
f8639939 | 930 | static void xen_release_ptpage(unsigned long pfn, unsigned level) |
5ead97c8 | 931 | { |
f4f97b3e JF |
932 | struct page *page = pfn_to_page(pfn); |
933 | ||
934 | if (PagePinned(page)) { | |
74260714 | 935 | if (!PageHighMem(page)) { |
6a9e9184 | 936 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) |
a684d69d | 937 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
f4f97b3e | 938 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
74260714 | 939 | } |
c946c7de | 940 | ClearPagePinned(page); |
f4f97b3e | 941 | } |
5ead97c8 JF |
942 | } |
943 | ||
f8639939 | 944 | static void xen_release_pte(unsigned long pfn) |
f6433706 MM |
945 | { |
946 | xen_release_ptpage(pfn, PT_PTE); | |
947 | } | |
948 | ||
f8639939 | 949 | static void xen_release_pmd(unsigned long pfn) |
f6433706 MM |
950 | { |
951 | xen_release_ptpage(pfn, PT_PMD); | |
952 | } | |
953 | ||
f6e58732 | 954 | #if PAGETABLE_LEVELS == 4 |
f8639939 | 955 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) |
f6e58732 JF |
956 | { |
957 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
958 | } | |
959 | ||
f8639939 | 960 | static void xen_release_pud(unsigned long pfn) |
f6e58732 JF |
961 | { |
962 | xen_release_ptpage(pfn, PT_PUD); | |
963 | } | |
964 | #endif | |
965 | ||
f4f97b3e JF |
966 | #ifdef CONFIG_HIGHPTE |
967 | static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) | |
5ead97c8 | 968 | { |
f4f97b3e JF |
969 | pgprot_t prot = PAGE_KERNEL; |
970 | ||
971 | if (PagePinned(page)) | |
972 | prot = PAGE_KERNEL_RO; | |
973 | ||
974 | if (0 && PageHighMem(page)) | |
975 | printk("mapping highpte %lx type %d prot %s\n", | |
976 | page_to_pfn(page), type, | |
977 | (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); | |
978 | ||
979 | return kmap_atomic_prot(page, type, prot); | |
5ead97c8 | 980 | } |
f4f97b3e | 981 | #endif |
5ead97c8 | 982 | |
db053b86 | 983 | #ifdef CONFIG_X86_32 |
9a4029fd JF |
984 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) |
985 | { | |
986 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
987 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
988 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
989 | pte_val_ma(pte)); | |
990 | ||
991 | return pte; | |
992 | } | |
993 | ||
994 | /* Init-time set_pte while constructing initial pagetables, which | |
995 | doesn't allow RO pagetable pages to be remapped RW */ | |
996 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
997 | { | |
998 | pte = mask_rw_pte(ptep, pte); | |
999 | ||
1000 | xen_set_pte(ptep, pte); | |
1001 | } | |
db053b86 | 1002 | #endif |
9a4029fd | 1003 | |
5ead97c8 JF |
1004 | static __init void xen_pagetable_setup_start(pgd_t *base) |
1005 | { | |
5ead97c8 JF |
1006 | } |
1007 | ||
0e91398f | 1008 | void xen_setup_shared_info(void) |
5ead97c8 JF |
1009 | { |
1010 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
15664f96 JF |
1011 | set_fixmap(FIX_PARAVIRT_BOOTMAP, |
1012 | xen_start_info->shared_info); | |
1013 | ||
1014 | HYPERVISOR_shared_info = | |
1015 | (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); | |
5ead97c8 JF |
1016 | } else |
1017 | HYPERVISOR_shared_info = | |
1018 | (struct shared_info *)__va(xen_start_info->shared_info); | |
1019 | ||
2e8fe719 JF |
1020 | #ifndef CONFIG_SMP |
1021 | /* In UP this is as good a place as any to set up shared info */ | |
1022 | xen_setup_vcpu_info_placement(); | |
1023 | #endif | |
d5edbc1f JF |
1024 | |
1025 | xen_setup_mfn_list_list(); | |
2e8fe719 JF |
1026 | } |
1027 | ||
1028 | static __init void xen_pagetable_setup_done(pgd_t *base) | |
1029 | { | |
0e91398f | 1030 | xen_setup_shared_info(); |
60223a32 | 1031 | } |
5ead97c8 | 1032 | |
e2426cf8 JF |
1033 | static __init void xen_post_allocator_init(void) |
1034 | { | |
8745f8b0 | 1035 | pv_mmu_ops.set_pte = xen_set_pte; |
e2426cf8 JF |
1036 | pv_mmu_ops.set_pmd = xen_set_pmd; |
1037 | pv_mmu_ops.set_pud = xen_set_pud; | |
f6e58732 JF |
1038 | #if PAGETABLE_LEVELS == 4 |
1039 | pv_mmu_ops.set_pgd = xen_set_pgd; | |
1040 | #endif | |
e2426cf8 | 1041 | |
2e8fe719 JF |
1042 | /* This will work as long as patching hasn't happened yet |
1043 | (which it hasn't) */ | |
6944a9c8 JF |
1044 | pv_mmu_ops.alloc_pte = xen_alloc_pte; |
1045 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
1046 | pv_mmu_ops.release_pte = xen_release_pte; | |
1047 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
8745f8b0 JF |
1048 | #if PAGETABLE_LEVELS == 4 |
1049 | pv_mmu_ops.alloc_pud = xen_alloc_pud; | |
1050 | pv_mmu_ops.release_pud = xen_release_pud; | |
1051 | #endif | |
e2426cf8 | 1052 | |
bf18bf94 JF |
1053 | #ifdef CONFIG_X86_64 |
1054 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
1055 | #endif | |
e2426cf8 JF |
1056 | xen_mark_init_mm_pinned(); |
1057 | } | |
1058 | ||
60223a32 | 1059 | /* This is called once we have the cpu_possible_map */ |
0e91398f | 1060 | void xen_setup_vcpu_info_placement(void) |
60223a32 JF |
1061 | { |
1062 | int cpu; | |
1063 | ||
1064 | for_each_possible_cpu(cpu) | |
1065 | xen_vcpu_setup(cpu); | |
1066 | ||
1067 | /* xen_vcpu_setup managed to place the vcpu_info within the | |
1068 | percpu area for all cpus, so make use of it */ | |
1069 | if (have_vcpu_info_placement) { | |
1070 | printk(KERN_INFO "Xen: using vcpu_info placement\n"); | |
1071 | ||
93b1eab3 JF |
1072 | pv_irq_ops.save_fl = xen_save_fl_direct; |
1073 | pv_irq_ops.restore_fl = xen_restore_fl_direct; | |
1074 | pv_irq_ops.irq_disable = xen_irq_disable_direct; | |
1075 | pv_irq_ops.irq_enable = xen_irq_enable_direct; | |
1076 | pv_mmu_ops.read_cr2 = xen_read_cr2_direct; | |
60223a32 | 1077 | } |
5ead97c8 JF |
1078 | } |
1079 | ||
ab144f5e AK |
1080 | static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, |
1081 | unsigned long addr, unsigned len) | |
6487673b JF |
1082 | { |
1083 | char *start, *end, *reloc; | |
1084 | unsigned ret; | |
1085 | ||
1086 | start = end = reloc = NULL; | |
1087 | ||
93b1eab3 JF |
1088 | #define SITE(op, x) \ |
1089 | case PARAVIRT_PATCH(op.x): \ | |
6487673b JF |
1090 | if (have_vcpu_info_placement) { \ |
1091 | start = (char *)xen_##x##_direct; \ | |
1092 | end = xen_##x##_direct_end; \ | |
1093 | reloc = xen_##x##_direct_reloc; \ | |
1094 | } \ | |
1095 | goto patch_site | |
1096 | ||
1097 | switch (type) { | |
93b1eab3 JF |
1098 | SITE(pv_irq_ops, irq_enable); |
1099 | SITE(pv_irq_ops, irq_disable); | |
1100 | SITE(pv_irq_ops, save_fl); | |
1101 | SITE(pv_irq_ops, restore_fl); | |
6487673b JF |
1102 | #undef SITE |
1103 | ||
1104 | patch_site: | |
1105 | if (start == NULL || (end-start) > len) | |
1106 | goto default_patch; | |
1107 | ||
ab144f5e | 1108 | ret = paravirt_patch_insns(insnbuf, len, start, end); |
6487673b JF |
1109 | |
1110 | /* Note: because reloc is assigned from something that | |
1111 | appears to be an array, gcc assumes it's non-null, | |
1112 | but doesn't know its relationship with start and | |
1113 | end. */ | |
1114 | if (reloc > start && reloc < end) { | |
1115 | int reloc_off = reloc - start; | |
ab144f5e AK |
1116 | long *relocp = (long *)(insnbuf + reloc_off); |
1117 | long delta = start - (char *)addr; | |
6487673b JF |
1118 | |
1119 | *relocp += delta; | |
1120 | } | |
1121 | break; | |
1122 | ||
1123 | default_patch: | |
1124 | default: | |
ab144f5e AK |
1125 | ret = paravirt_patch_default(type, clobbers, insnbuf, |
1126 | addr, len); | |
6487673b JF |
1127 | break; |
1128 | } | |
1129 | ||
1130 | return ret; | |
1131 | } | |
1132 | ||
aeaaa59c JF |
1133 | static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot) |
1134 | { | |
1135 | pte_t pte; | |
1136 | ||
1137 | phys >>= PAGE_SHIFT; | |
1138 | ||
1139 | switch (idx) { | |
1140 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
1141 | #ifdef CONFIG_X86_F00F_BUG | |
1142 | case FIX_F00F_IDT: | |
1143 | #endif | |
15664f96 | 1144 | #ifdef CONFIG_X86_32 |
aeaaa59c JF |
1145 | case FIX_WP_TEST: |
1146 | case FIX_VDSO: | |
b3fe1243 | 1147 | # ifdef CONFIG_HIGHMEM |
15664f96 | 1148 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: |
b3fe1243 | 1149 | # endif |
15664f96 JF |
1150 | #else |
1151 | case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE: | |
1152 | #endif | |
aeaaa59c JF |
1153 | #ifdef CONFIG_X86_LOCAL_APIC |
1154 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1155 | #endif | |
1156 | pte = pfn_pte(phys, prot); | |
1157 | break; | |
1158 | ||
1159 | default: | |
1160 | pte = mfn_pte(phys, prot); | |
1161 | break; | |
1162 | } | |
1163 | ||
1164 | __native_set_fixmap(idx, pte); | |
bf18bf94 JF |
1165 | |
1166 | #ifdef CONFIG_X86_64 | |
1167 | /* Replicate changes to map the vsyscall page into the user | |
1168 | pagetable vsyscall mapping. */ | |
1169 | if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) { | |
1170 | unsigned long vaddr = __fix_to_virt(idx); | |
1171 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
1172 | } | |
1173 | #endif | |
aeaaa59c JF |
1174 | } |
1175 | ||
93b1eab3 | 1176 | static const struct pv_info xen_info __initdata = { |
5ead97c8 JF |
1177 | .paravirt_enabled = 1, |
1178 | .shared_kernel_pmd = 0, | |
1179 | ||
1180 | .name = "Xen", | |
93b1eab3 | 1181 | }; |
5ead97c8 | 1182 | |
93b1eab3 | 1183 | static const struct pv_init_ops xen_init_ops __initdata = { |
6487673b | 1184 | .patch = xen_patch, |
5ead97c8 | 1185 | |
93b1eab3 | 1186 | .banner = xen_banner, |
5ead97c8 JF |
1187 | .memory_setup = xen_memory_setup, |
1188 | .arch_setup = xen_arch_setup, | |
e2426cf8 | 1189 | .post_allocator_init = xen_post_allocator_init, |
93b1eab3 | 1190 | }; |
5ead97c8 | 1191 | |
93b1eab3 | 1192 | static const struct pv_time_ops xen_time_ops __initdata = { |
15c84731 | 1193 | .time_init = xen_time_init, |
93b1eab3 | 1194 | |
15c84731 JF |
1195 | .set_wallclock = xen_set_wallclock, |
1196 | .get_wallclock = xen_get_wallclock, | |
e93ef949 | 1197 | .get_tsc_khz = xen_tsc_khz, |
ab550288 | 1198 | .sched_clock = xen_sched_clock, |
93b1eab3 | 1199 | }; |
15c84731 | 1200 | |
93b1eab3 | 1201 | static const struct pv_cpu_ops xen_cpu_ops __initdata = { |
5ead97c8 JF |
1202 | .cpuid = xen_cpuid, |
1203 | ||
1204 | .set_debugreg = xen_set_debugreg, | |
1205 | .get_debugreg = xen_get_debugreg, | |
1206 | ||
7b1333aa | 1207 | .clts = xen_clts, |
5ead97c8 JF |
1208 | |
1209 | .read_cr0 = native_read_cr0, | |
7b1333aa | 1210 | .write_cr0 = xen_write_cr0, |
5ead97c8 | 1211 | |
5ead97c8 JF |
1212 | .read_cr4 = native_read_cr4, |
1213 | .read_cr4_safe = native_read_cr4_safe, | |
1214 | .write_cr4 = xen_write_cr4, | |
1215 | ||
5ead97c8 JF |
1216 | .wbinvd = native_wbinvd, |
1217 | ||
1218 | .read_msr = native_read_msr_safe, | |
1153968a | 1219 | .write_msr = xen_write_msr_safe, |
5ead97c8 JF |
1220 | .read_tsc = native_read_tsc, |
1221 | .read_pmc = native_read_pmc, | |
1222 | ||
81e103f1 | 1223 | .iret = xen_iret, |
d75cd22f | 1224 | .irq_enable_sysexit = xen_sysexit, |
6fcac6d3 JF |
1225 | #ifdef CONFIG_X86_64 |
1226 | .usergs_sysret32 = xen_sysret32, | |
1227 | .usergs_sysret64 = xen_sysret64, | |
1228 | #endif | |
5ead97c8 JF |
1229 | |
1230 | .load_tr_desc = paravirt_nop, | |
1231 | .set_ldt = xen_set_ldt, | |
1232 | .load_gdt = xen_load_gdt, | |
1233 | .load_idt = xen_load_idt, | |
1234 | .load_tls = xen_load_tls, | |
a8fc1089 EH |
1235 | #ifdef CONFIG_X86_64 |
1236 | .load_gs_index = xen_load_gs_index, | |
1237 | #endif | |
5ead97c8 | 1238 | |
38ffbe66 JF |
1239 | .alloc_ldt = xen_alloc_ldt, |
1240 | .free_ldt = xen_free_ldt, | |
1241 | ||
5ead97c8 JF |
1242 | .store_gdt = native_store_gdt, |
1243 | .store_idt = native_store_idt, | |
1244 | .store_tr = xen_store_tr, | |
1245 | ||
1246 | .write_ldt_entry = xen_write_ldt_entry, | |
1247 | .write_gdt_entry = xen_write_gdt_entry, | |
1248 | .write_idt_entry = xen_write_idt_entry, | |
faca6227 | 1249 | .load_sp0 = xen_load_sp0, |
5ead97c8 JF |
1250 | |
1251 | .set_iopl_mask = xen_set_iopl_mask, | |
1252 | .io_delay = xen_io_delay, | |
1253 | ||
952d1d70 JF |
1254 | /* Xen takes care of %gs when switching to usermode for us */ |
1255 | .swapgs = paravirt_nop, | |
1256 | ||
8965c1c0 JF |
1257 | .lazy_mode = { |
1258 | .enter = paravirt_enter_lazy_cpu, | |
1259 | .leave = xen_leave_lazy, | |
1260 | }, | |
93b1eab3 JF |
1261 | }; |
1262 | ||
93b1eab3 | 1263 | static const struct pv_apic_ops xen_apic_ops __initdata = { |
5ead97c8 | 1264 | #ifdef CONFIG_X86_LOCAL_APIC |
5ead97c8 JF |
1265 | .setup_boot_clock = paravirt_nop, |
1266 | .setup_secondary_clock = paravirt_nop, | |
1267 | .startup_ipi_hook = paravirt_nop, | |
1268 | #endif | |
93b1eab3 JF |
1269 | }; |
1270 | ||
1271 | static const struct pv_mmu_ops xen_mmu_ops __initdata = { | |
1272 | .pagetable_setup_start = xen_pagetable_setup_start, | |
1273 | .pagetable_setup_done = xen_pagetable_setup_done, | |
1274 | ||
1275 | .read_cr2 = xen_read_cr2, | |
1276 | .write_cr2 = xen_write_cr2, | |
1277 | ||
1278 | .read_cr3 = xen_read_cr3, | |
1279 | .write_cr3 = xen_write_cr3, | |
5ead97c8 JF |
1280 | |
1281 | .flush_tlb_user = xen_flush_tlb, | |
1282 | .flush_tlb_kernel = xen_flush_tlb, | |
1283 | .flush_tlb_single = xen_flush_tlb_single, | |
f87e4cac | 1284 | .flush_tlb_others = xen_flush_tlb_others, |
5ead97c8 JF |
1285 | |
1286 | .pte_update = paravirt_nop, | |
1287 | .pte_update_defer = paravirt_nop, | |
1288 | ||
d6182fbf JF |
1289 | .pgd_alloc = xen_pgd_alloc, |
1290 | .pgd_free = xen_pgd_free, | |
eba0045f | 1291 | |
6944a9c8 JF |
1292 | .alloc_pte = xen_alloc_pte_init, |
1293 | .release_pte = xen_release_pte_init, | |
1294 | .alloc_pmd = xen_alloc_pte_init, | |
1295 | .alloc_pmd_clone = paravirt_nop, | |
1296 | .release_pmd = xen_release_pte_init, | |
f4f97b3e JF |
1297 | |
1298 | #ifdef CONFIG_HIGHPTE | |
1299 | .kmap_atomic_pte = xen_kmap_atomic_pte, | |
1300 | #endif | |
5ead97c8 | 1301 | |
22911b3f JF |
1302 | #ifdef CONFIG_X86_64 |
1303 | .set_pte = xen_set_pte, | |
1304 | #else | |
851fa3c4 | 1305 | .set_pte = xen_set_pte_init, |
22911b3f | 1306 | #endif |
3b827c1b | 1307 | .set_pte_at = xen_set_pte_at, |
e2426cf8 | 1308 | .set_pmd = xen_set_pmd_hyper, |
3b827c1b | 1309 | |
08b882c6 JF |
1310 | .ptep_modify_prot_start = __ptep_modify_prot_start, |
1311 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
1312 | ||
3b827c1b | 1313 | .pte_val = xen_pte_val, |
b56afe1d | 1314 | .pte_flags = native_pte_flags, |
3b827c1b JF |
1315 | .pgd_val = xen_pgd_val, |
1316 | ||
1317 | .make_pte = xen_make_pte, | |
1318 | .make_pgd = xen_make_pgd, | |
1319 | ||
f6e58732 | 1320 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
1321 | .set_pte_atomic = xen_set_pte_atomic, |
1322 | .set_pte_present = xen_set_pte_at, | |
3b827c1b JF |
1323 | .pte_clear = xen_pte_clear, |
1324 | .pmd_clear = xen_pmd_clear, | |
f6e58732 JF |
1325 | #endif /* CONFIG_X86_PAE */ |
1326 | .set_pud = xen_set_pud_hyper, | |
3b827c1b JF |
1327 | |
1328 | .make_pmd = xen_make_pmd, | |
1329 | .pmd_val = xen_pmd_val, | |
3b827c1b | 1330 | |
f6e58732 JF |
1331 | #if PAGETABLE_LEVELS == 4 |
1332 | .pud_val = xen_pud_val, | |
1333 | .make_pud = xen_make_pud, | |
1334 | .set_pgd = xen_set_pgd_hyper, | |
1335 | ||
1336 | .alloc_pud = xen_alloc_pte_init, | |
1337 | .release_pud = xen_release_pte_init, | |
1338 | #endif /* PAGETABLE_LEVELS == 4 */ | |
1339 | ||
3b827c1b JF |
1340 | .activate_mm = xen_activate_mm, |
1341 | .dup_mmap = xen_dup_mmap, | |
1342 | .exit_mmap = xen_exit_mmap, | |
1343 | ||
8965c1c0 JF |
1344 | .lazy_mode = { |
1345 | .enter = paravirt_enter_lazy_mmu, | |
1346 | .leave = xen_leave_lazy, | |
1347 | }, | |
aeaaa59c JF |
1348 | |
1349 | .set_fixmap = xen_set_fixmap, | |
5ead97c8 JF |
1350 | }; |
1351 | ||
fefa629a JF |
1352 | static void xen_reboot(int reason) |
1353 | { | |
349c709f JF |
1354 | struct sched_shutdown r = { .reason = reason }; |
1355 | ||
fefa629a JF |
1356 | #ifdef CONFIG_SMP |
1357 | smp_send_stop(); | |
1358 | #endif | |
1359 | ||
349c709f | 1360 | if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) |
fefa629a JF |
1361 | BUG(); |
1362 | } | |
1363 | ||
1364 | static void xen_restart(char *msg) | |
1365 | { | |
1366 | xen_reboot(SHUTDOWN_reboot); | |
1367 | } | |
1368 | ||
1369 | static void xen_emergency_restart(void) | |
1370 | { | |
1371 | xen_reboot(SHUTDOWN_reboot); | |
1372 | } | |
1373 | ||
1374 | static void xen_machine_halt(void) | |
1375 | { | |
1376 | xen_reboot(SHUTDOWN_poweroff); | |
1377 | } | |
1378 | ||
1379 | static void xen_crash_shutdown(struct pt_regs *regs) | |
1380 | { | |
1381 | xen_reboot(SHUTDOWN_crash); | |
1382 | } | |
1383 | ||
1384 | static const struct machine_ops __initdata xen_machine_ops = { | |
1385 | .restart = xen_restart, | |
1386 | .halt = xen_machine_halt, | |
1387 | .power_off = xen_machine_halt, | |
1388 | .shutdown = xen_machine_halt, | |
1389 | .crash_shutdown = xen_crash_shutdown, | |
1390 | .emergency_restart = xen_emergency_restart, | |
1391 | }; | |
1392 | ||
6487673b | 1393 | |
fb1d8404 JF |
1394 | static void __init xen_reserve_top(void) |
1395 | { | |
f5d36de0 | 1396 | #ifdef CONFIG_X86_32 |
fb1d8404 JF |
1397 | unsigned long top = HYPERVISOR_VIRT_START; |
1398 | struct xen_platform_parameters pp; | |
1399 | ||
1400 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1401 | top = pp.virt_start; | |
1402 | ||
5dc64a34 | 1403 | reserve_top_address(-top); |
f5d36de0 | 1404 | #endif /* CONFIG_X86_32 */ |
fb1d8404 JF |
1405 | } |
1406 | ||
084a2a4e JF |
1407 | /* |
1408 | * Like __va(), but returns address in the kernel mapping (which is | |
1409 | * all we have until the physical memory mapping has been set up. | |
1410 | */ | |
1411 | static void *__ka(phys_addr_t paddr) | |
1412 | { | |
39dbc5bd | 1413 | #ifdef CONFIG_X86_64 |
084a2a4e | 1414 | return (void *)(paddr + __START_KERNEL_map); |
39dbc5bd JF |
1415 | #else |
1416 | return __va(paddr); | |
1417 | #endif | |
fb1d8404 JF |
1418 | } |
1419 | ||
084a2a4e JF |
1420 | /* Convert a machine address to physical address */ |
1421 | static unsigned long m2p(phys_addr_t maddr) | |
1422 | { | |
1423 | phys_addr_t paddr; | |
1424 | ||
59438c9f | 1425 | maddr &= PTE_PFN_MASK; |
084a2a4e JF |
1426 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; |
1427 | ||
1428 | return paddr; | |
fb1d8404 JF |
1429 | } |
1430 | ||
084a2a4e JF |
1431 | /* Convert a machine address to kernel virtual */ |
1432 | static void *m2v(phys_addr_t maddr) | |
1433 | { | |
1434 | return __ka(m2p(maddr)); | |
1435 | } | |
1436 | ||
084a2a4e JF |
1437 | static void set_page_prot(void *addr, pgprot_t prot) |
1438 | { | |
1439 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1440 | pte_t pte = pfn_pte(pfn, prot); | |
1441 | ||
084a2a4e JF |
1442 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) |
1443 | BUG(); | |
1444 | } | |
1445 | ||
39dbc5bd | 1446 | static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) |
d114e198 JF |
1447 | { |
1448 | unsigned pmdidx, pteidx; | |
1449 | unsigned ident_pte; | |
1450 | unsigned long pfn; | |
1451 | ||
1452 | ident_pte = 0; | |
1453 | pfn = 0; | |
1454 | for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1455 | pte_t *pte_page; | |
1456 | ||
d114e198 | 1457 | /* Reuse or allocate a page of ptes */ |
39dbc5bd JF |
1458 | if (pmd_present(pmd[pmdidx])) |
1459 | pte_page = m2v(pmd[pmdidx].pmd); | |
d114e198 JF |
1460 | else { |
1461 | /* Check for free pte pages */ | |
1462 | if (ident_pte == ARRAY_SIZE(level1_ident_pgt)) | |
1463 | break; | |
1464 | ||
1465 | pte_page = &level1_ident_pgt[ident_pte]; | |
1466 | ident_pte += PTRS_PER_PTE; | |
1467 | ||
39dbc5bd | 1468 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); |
d114e198 JF |
1469 | } |
1470 | ||
1471 | /* Install mappings */ | |
1472 | for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1473 | pte_t pte; | |
1474 | ||
1475 | if (pfn > max_pfn_mapped) | |
1476 | max_pfn_mapped = pfn; | |
1477 | ||
1478 | if (!pte_none(pte_page[pteidx])) | |
1479 | continue; | |
1480 | ||
1481 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1482 | pte_page[pteidx] = pte; | |
1483 | } | |
1484 | } | |
1485 | ||
1486 | for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1487 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
39dbc5bd JF |
1488 | |
1489 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1490 | } | |
1491 | ||
1492 | #ifdef CONFIG_X86_64 | |
1493 | static void convert_pfn_mfn(void *v) | |
1494 | { | |
1495 | pte_t *pte = v; | |
1496 | int i; | |
1497 | ||
1498 | /* All levels are converted the same way, so just treat them | |
1499 | as ptes. */ | |
1500 | for(i = 0; i < PTRS_PER_PTE; i++) | |
1501 | pte[i] = xen_make_pte(pte[i].pte); | |
d114e198 JF |
1502 | } |
1503 | ||
084a2a4e JF |
1504 | /* |
1505 | * Set up the inital kernel pagetable. | |
1506 | * | |
1507 | * We can construct this by grafting the Xen provided pagetable into | |
1508 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1509 | * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This | |
1510 | * means that only the kernel has a physical mapping to start with - | |
1511 | * but that's enough to get __va working. We need to fill in the rest | |
1512 | * of the physical mapping once some sort of allocator has been set | |
1513 | * up. | |
1514 | */ | |
d114e198 | 1515 | static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) |
084a2a4e JF |
1516 | { |
1517 | pud_t *l3; | |
1518 | pmd_t *l2; | |
1519 | ||
1520 | /* Zap identity mapping */ | |
1521 | init_level4_pgt[0] = __pgd(0); | |
1522 | ||
1523 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1524 | convert_pfn_mfn(init_level4_pgt); | |
1525 | convert_pfn_mfn(level3_ident_pgt); | |
1526 | convert_pfn_mfn(level3_kernel_pgt); | |
1527 | ||
1528 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1529 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1530 | ||
1531 | memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1532 | memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1533 | ||
1534 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd); | |
1535 | l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud); | |
1536 | memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1537 | ||
d114e198 | 1538 | /* Set up identity map */ |
39dbc5bd | 1539 | xen_map_identity_early(level2_ident_pgt, max_pfn); |
d114e198 | 1540 | |
084a2a4e JF |
1541 | /* Make pagetable pieces RO */ |
1542 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1543 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1544 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
bf18bf94 | 1545 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); |
084a2a4e JF |
1546 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); |
1547 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1548 | ||
1549 | /* Pin down new L4 */ | |
39dbc5bd JF |
1550 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, |
1551 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
084a2a4e JF |
1552 | |
1553 | /* Unpin Xen-provided one */ | |
1554 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1555 | ||
1556 | /* Switch over */ | |
1557 | pgd = init_level4_pgt; | |
d6182fbf JF |
1558 | |
1559 | /* | |
1560 | * At this stage there can be no user pgd, and no page | |
1561 | * structure to attach it to, so make sure we just set kernel | |
1562 | * pgd. | |
1563 | */ | |
1564 | xen_mc_batch(); | |
1565 | __xen_write_cr3(true, __pa(pgd)); | |
1566 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
084a2a4e | 1567 | |
d114e198 JF |
1568 | reserve_early(__pa(xen_start_info->pt_base), |
1569 | __pa(xen_start_info->pt_base + | |
1570 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1571 | "XEN PAGETABLES"); | |
084a2a4e JF |
1572 | |
1573 | return pgd; | |
1574 | } | |
39dbc5bd JF |
1575 | #else /* !CONFIG_X86_64 */ |
1576 | static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss; | |
1577 | ||
d114e198 | 1578 | static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) |
084a2a4e | 1579 | { |
39dbc5bd JF |
1580 | pmd_t *kernel_pmd; |
1581 | ||
084a2a4e JF |
1582 | init_pg_tables_start = __pa(pgd); |
1583 | init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE; | |
1584 | max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024); | |
1585 | ||
39dbc5bd JF |
1586 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); |
1587 | memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD); | |
d114e198 | 1588 | |
39dbc5bd JF |
1589 | xen_map_identity_early(level2_kernel_pgt, max_pfn); |
1590 | ||
1591 | memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD); | |
1592 | set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY], | |
1593 | __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT)); | |
1594 | ||
1595 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1596 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1597 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
1598 | ||
1599 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1600 | ||
1601 | xen_write_cr3(__pa(swapper_pg_dir)); | |
1602 | ||
1603 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir))); | |
1604 | ||
1605 | return swapper_pg_dir; | |
fb1d8404 | 1606 | } |
084a2a4e | 1607 | #endif /* CONFIG_X86_64 */ |
fb1d8404 | 1608 | |
5ead97c8 JF |
1609 | /* First C function to be called on Xen boot */ |
1610 | asmlinkage void __init xen_start_kernel(void) | |
1611 | { | |
1612 | pgd_t *pgd; | |
1613 | ||
1614 | if (!xen_start_info) | |
1615 | return; | |
1616 | ||
6e833587 JF |
1617 | xen_domain_type = XEN_PV_DOMAIN; |
1618 | ||
7999f4b4 | 1619 | BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0); |
5ead97c8 | 1620 | |
e57778a1 JF |
1621 | xen_setup_features(); |
1622 | ||
5ead97c8 | 1623 | /* Install Xen paravirt ops */ |
93b1eab3 JF |
1624 | pv_info = xen_info; |
1625 | pv_init_ops = xen_init_ops; | |
1626 | pv_time_ops = xen_time_ops; | |
1627 | pv_cpu_ops = xen_cpu_ops; | |
93b1eab3 JF |
1628 | pv_apic_ops = xen_apic_ops; |
1629 | pv_mmu_ops = xen_mmu_ops; | |
93b1eab3 | 1630 | |
0d1edf46 JF |
1631 | xen_init_irq_ops(); |
1632 | ||
94a8c3c2 | 1633 | #ifdef CONFIG_X86_LOCAL_APIC |
ad66dd34 | 1634 | /* |
94a8c3c2 | 1635 | * set up the basic apic ops. |
ad66dd34 SS |
1636 | */ |
1637 | apic_ops = &xen_basic_apic_ops; | |
1638 | #endif | |
93b1eab3 | 1639 | |
e57778a1 JF |
1640 | if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { |
1641 | pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; | |
1642 | pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; | |
1643 | } | |
1644 | ||
fefa629a JF |
1645 | machine_ops = xen_machine_ops; |
1646 | ||
f5d36de0 JF |
1647 | #ifdef CONFIG_X86_64 |
1648 | /* Disable until direct per-cpu data access. */ | |
1649 | have_vcpu_info_placement = 0; | |
5b09b287 | 1650 | x86_64_init_pda(); |
f87e4cac | 1651 | #endif |
5ead97c8 | 1652 | |
a9e7062d | 1653 | xen_smp_init(); |
5ead97c8 JF |
1654 | |
1655 | /* Get mfn list */ | |
1656 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
d451bb7a | 1657 | xen_build_dynamic_phys_to_machine(); |
5ead97c8 JF |
1658 | |
1659 | pgd = (pgd_t *)xen_start_info->pt_base; | |
1660 | ||
084a2a4e JF |
1661 | /* Prevent unwanted bits from being set in PTEs. */ |
1662 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
6e833587 | 1663 | if (!xen_initial_domain()) |
084a2a4e | 1664 | __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); |
60223a32 | 1665 | |
60223a32 | 1666 | /* Don't do the full vcpu_info placement stuff until we have a |
2e8fe719 | 1667 | possible map and a non-dummy shared_info. */ |
60223a32 | 1668 | per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; |
5ead97c8 | 1669 | |
084a2a4e | 1670 | xen_raw_console_write("mapping kernel into physical memory\n"); |
d114e198 | 1671 | pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages); |
5ead97c8 | 1672 | |
084a2a4e | 1673 | init_mm.pgd = pgd; |
5ead97c8 JF |
1674 | |
1675 | /* keep using Xen gdt for now; no urgent need to change it */ | |
1676 | ||
93b1eab3 | 1677 | pv_info.kernel_rpl = 1; |
5ead97c8 | 1678 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) |
93b1eab3 | 1679 | pv_info.kernel_rpl = 0; |
5ead97c8 JF |
1680 | |
1681 | /* set the limit of our address space */ | |
fb1d8404 | 1682 | xen_reserve_top(); |
5ead97c8 | 1683 | |
7d087b68 | 1684 | #ifdef CONFIG_X86_32 |
5ead97c8 JF |
1685 | /* set up basic CPUID stuff */ |
1686 | cpu_detect(&new_cpu_data); | |
1687 | new_cpu_data.hard_math = 1; | |
1688 | new_cpu_data.x86_capability[0] = cpuid_edx(1); | |
7d087b68 | 1689 | #endif |
5ead97c8 JF |
1690 | |
1691 | /* Poke various useful things into boot_params */ | |
30c82645 PA |
1692 | boot_params.hdr.type_of_loader = (9 << 4) | 0; |
1693 | boot_params.hdr.ramdisk_image = xen_start_info->mod_start | |
1694 | ? __pa(xen_start_info->mod_start) : 0; | |
1695 | boot_params.hdr.ramdisk_size = xen_start_info->mod_len; | |
b7c3c5c1 | 1696 | boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); |
5ead97c8 | 1697 | |
6e833587 | 1698 | if (!xen_initial_domain()) { |
83abc70a | 1699 | add_preferred_console("xenboot", 0, NULL); |
9e124fe1 | 1700 | add_preferred_console("tty", 0, NULL); |
b8c2d3df | 1701 | add_preferred_console("hvc", 0, NULL); |
9e124fe1 | 1702 | } |
b8c2d3df | 1703 | |
084a2a4e JF |
1704 | xen_raw_console_write("about to get started...\n"); |
1705 | ||
5ead97c8 | 1706 | /* Start the world */ |
f5d36de0 | 1707 | #ifdef CONFIG_X86_32 |
f0d43100 | 1708 | i386_start_kernel(); |
f5d36de0 | 1709 | #else |
084a2a4e | 1710 | x86_64_start_reservations((char *)__pa_symbol(&boot_params)); |
f5d36de0 | 1711 | #endif |
5ead97c8 | 1712 | } |