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> | |
fefa629a | 33 | #include <xen/interface/sched.h> |
5ead97c8 JF |
34 | #include <xen/features.h> |
35 | #include <xen/page.h> | |
084a2a4e | 36 | #include <xen/hvc-console.h> |
5ead97c8 JF |
37 | |
38 | #include <asm/paravirt.h> | |
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 | |
bf18bf94 JF |
60 | /* |
61 | * Identity map, in addition to plain kernel map. This needs to be | |
62 | * large enough to allocate page table pages to allocate the rest. | |
63 | * Each page can map 2MB. | |
64 | */ | |
65 | static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss; | |
66 | ||
67 | #ifdef CONFIG_X86_64 | |
68 | /* l3 pud for userspace vsyscall mapping */ | |
69 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
70 | #endif /* CONFIG_X86_64 */ | |
71 | ||
9f79991d JF |
72 | /* |
73 | * Note about cr3 (pagetable base) values: | |
74 | * | |
75 | * xen_cr3 contains the current logical cr3 value; it contains the | |
76 | * last set cr3. This may not be the current effective cr3, because | |
77 | * its update may be being lazily deferred. However, a vcpu looking | |
78 | * at its own cr3 can use this value knowing that it everything will | |
79 | * be self-consistent. | |
80 | * | |
81 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
82 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
83 | * out of date, but it will never be set early). If one vcpu is | |
84 | * looking at another vcpu's cr3 value, it should use this variable. | |
85 | */ | |
86 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
87 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
5ead97c8 JF |
88 | |
89 | struct start_info *xen_start_info; | |
90 | EXPORT_SYMBOL_GPL(xen_start_info); | |
91 | ||
a0d695c8 | 92 | struct shared_info xen_dummy_shared_info; |
60223a32 JF |
93 | |
94 | /* | |
95 | * Point at some empty memory to start with. We map the real shared_info | |
96 | * page as soon as fixmap is up and running. | |
97 | */ | |
a0d695c8 | 98 | struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info; |
60223a32 JF |
99 | |
100 | /* | |
101 | * Flag to determine whether vcpu info placement is available on all | |
102 | * VCPUs. We assume it is to start with, and then set it to zero on | |
103 | * the first failure. This is because it can succeed on some VCPUs | |
104 | * and not others, since it can involve hypervisor memory allocation, | |
105 | * or because the guest failed to guarantee all the appropriate | |
106 | * constraints on all VCPUs (ie buffer can't cross a page boundary). | |
107 | * | |
108 | * Note that any particular CPU may be using a placed vcpu structure, | |
109 | * but we can only optimise if the all are. | |
110 | * | |
111 | * 0: not available, 1: available | |
112 | */ | |
04c44a08 | 113 | static int have_vcpu_info_placement = 1; |
60223a32 | 114 | |
9c7a7942 | 115 | static void xen_vcpu_setup(int cpu) |
5ead97c8 | 116 | { |
60223a32 JF |
117 | struct vcpu_register_vcpu_info info; |
118 | int err; | |
119 | struct vcpu_info *vcpup; | |
120 | ||
a0d695c8 | 121 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
5ead97c8 | 122 | per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; |
60223a32 JF |
123 | |
124 | if (!have_vcpu_info_placement) | |
125 | return; /* already tested, not available */ | |
126 | ||
127 | vcpup = &per_cpu(xen_vcpu_info, cpu); | |
128 | ||
129 | info.mfn = virt_to_mfn(vcpup); | |
130 | info.offset = offset_in_page(vcpup); | |
131 | ||
e3d26976 | 132 | printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", |
60223a32 JF |
133 | cpu, vcpup, info.mfn, info.offset); |
134 | ||
135 | /* Check to see if the hypervisor will put the vcpu_info | |
136 | structure where we want it, which allows direct access via | |
137 | a percpu-variable. */ | |
138 | err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); | |
139 | ||
140 | if (err) { | |
141 | printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); | |
142 | have_vcpu_info_placement = 0; | |
143 | } else { | |
144 | /* This cpu is using the registered vcpu info, even if | |
145 | later ones fail to. */ | |
146 | per_cpu(xen_vcpu, cpu) = vcpup; | |
6487673b | 147 | |
60223a32 JF |
148 | printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", |
149 | cpu, vcpup); | |
150 | } | |
5ead97c8 JF |
151 | } |
152 | ||
9c7a7942 JF |
153 | /* |
154 | * On restore, set the vcpu placement up again. | |
155 | * If it fails, then we're in a bad state, since | |
156 | * we can't back out from using it... | |
157 | */ | |
158 | void xen_vcpu_restore(void) | |
159 | { | |
160 | if (have_vcpu_info_placement) { | |
161 | int cpu; | |
162 | ||
163 | for_each_online_cpu(cpu) { | |
164 | bool other_cpu = (cpu != smp_processor_id()); | |
165 | ||
166 | if (other_cpu && | |
167 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL)) | |
168 | BUG(); | |
169 | ||
170 | xen_vcpu_setup(cpu); | |
171 | ||
172 | if (other_cpu && | |
173 | HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) | |
174 | BUG(); | |
175 | } | |
176 | ||
177 | BUG_ON(!have_vcpu_info_placement); | |
178 | } | |
179 | } | |
180 | ||
5ead97c8 JF |
181 | static void __init xen_banner(void) |
182 | { | |
95c7c23b JF |
183 | unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); |
184 | struct xen_extraversion extra; | |
185 | HYPERVISOR_xen_version(XENVER_extraversion, &extra); | |
186 | ||
5ead97c8 | 187 | printk(KERN_INFO "Booting paravirtualized kernel on %s\n", |
93b1eab3 | 188 | pv_info.name); |
95c7c23b JF |
189 | printk(KERN_INFO "Xen version: %d.%d%s%s\n", |
190 | version >> 16, version & 0xffff, extra.extraversion, | |
e57778a1 | 191 | xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); |
5ead97c8 JF |
192 | } |
193 | ||
65ea5b03 PA |
194 | static void xen_cpuid(unsigned int *ax, unsigned int *bx, |
195 | unsigned int *cx, unsigned int *dx) | |
5ead97c8 JF |
196 | { |
197 | unsigned maskedx = ~0; | |
198 | ||
199 | /* | |
200 | * Mask out inconvenient features, to try and disable as many | |
201 | * unsupported kernel subsystems as possible. | |
202 | */ | |
65ea5b03 | 203 | if (*ax == 1) |
5ead97c8 JF |
204 | maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */ |
205 | (1 << X86_FEATURE_ACPI) | /* disable ACPI */ | |
dbe9e994 JF |
206 | (1 << X86_FEATURE_MCE) | /* disable MCE */ |
207 | (1 << X86_FEATURE_MCA) | /* disable MCA */ | |
5ead97c8 JF |
208 | (1 << X86_FEATURE_ACC)); /* thermal monitoring */ |
209 | ||
210 | asm(XEN_EMULATE_PREFIX "cpuid" | |
65ea5b03 PA |
211 | : "=a" (*ax), |
212 | "=b" (*bx), | |
213 | "=c" (*cx), | |
214 | "=d" (*dx) | |
215 | : "0" (*ax), "2" (*cx)); | |
216 | *dx &= maskedx; | |
5ead97c8 JF |
217 | } |
218 | ||
219 | static void xen_set_debugreg(int reg, unsigned long val) | |
220 | { | |
221 | HYPERVISOR_set_debugreg(reg, val); | |
222 | } | |
223 | ||
224 | static unsigned long xen_get_debugreg(int reg) | |
225 | { | |
226 | return HYPERVISOR_get_debugreg(reg); | |
227 | } | |
228 | ||
229 | static unsigned long xen_save_fl(void) | |
230 | { | |
231 | struct vcpu_info *vcpu; | |
232 | unsigned long flags; | |
233 | ||
5ead97c8 | 234 | vcpu = x86_read_percpu(xen_vcpu); |
f120f13e | 235 | |
5ead97c8 JF |
236 | /* flag has opposite sense of mask */ |
237 | flags = !vcpu->evtchn_upcall_mask; | |
5ead97c8 JF |
238 | |
239 | /* convert to IF type flag | |
240 | -0 -> 0x00000000 | |
241 | -1 -> 0xffffffff | |
242 | */ | |
243 | return (-flags) & X86_EFLAGS_IF; | |
244 | } | |
245 | ||
246 | static void xen_restore_fl(unsigned long flags) | |
247 | { | |
248 | struct vcpu_info *vcpu; | |
249 | ||
5ead97c8 JF |
250 | /* convert from IF type flag */ |
251 | flags = !(flags & X86_EFLAGS_IF); | |
f120f13e JF |
252 | |
253 | /* There's a one instruction preempt window here. We need to | |
254 | make sure we're don't switch CPUs between getting the vcpu | |
255 | pointer and updating the mask. */ | |
256 | preempt_disable(); | |
5ead97c8 JF |
257 | vcpu = x86_read_percpu(xen_vcpu); |
258 | vcpu->evtchn_upcall_mask = flags; | |
f120f13e | 259 | preempt_enable_no_resched(); |
5ead97c8 | 260 | |
f120f13e JF |
261 | /* Doesn't matter if we get preempted here, because any |
262 | pending event will get dealt with anyway. */ | |
5ead97c8 | 263 | |
f120f13e JF |
264 | if (flags == 0) { |
265 | preempt_check_resched(); | |
266 | barrier(); /* unmask then check (avoid races) */ | |
5ead97c8 JF |
267 | if (unlikely(vcpu->evtchn_upcall_pending)) |
268 | force_evtchn_callback(); | |
f120f13e | 269 | } |
5ead97c8 JF |
270 | } |
271 | ||
272 | static void xen_irq_disable(void) | |
273 | { | |
f120f13e JF |
274 | /* There's a one instruction preempt window here. We need to |
275 | make sure we're don't switch CPUs between getting the vcpu | |
276 | pointer and updating the mask. */ | |
5ead97c8 | 277 | preempt_disable(); |
f120f13e | 278 | x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1; |
5ead97c8 JF |
279 | preempt_enable_no_resched(); |
280 | } | |
281 | ||
282 | static void xen_irq_enable(void) | |
283 | { | |
284 | struct vcpu_info *vcpu; | |
285 | ||
239d1fc0 JF |
286 | /* We don't need to worry about being preempted here, since |
287 | either a) interrupts are disabled, so no preemption, or b) | |
288 | the caller is confused and is trying to re-enable interrupts | |
289 | on an indeterminate processor. */ | |
290 | ||
5ead97c8 JF |
291 | vcpu = x86_read_percpu(xen_vcpu); |
292 | vcpu->evtchn_upcall_mask = 0; | |
293 | ||
f120f13e JF |
294 | /* Doesn't matter if we get preempted here, because any |
295 | pending event will get dealt with anyway. */ | |
5ead97c8 | 296 | |
f120f13e | 297 | barrier(); /* unmask then check (avoid races) */ |
5ead97c8 JF |
298 | if (unlikely(vcpu->evtchn_upcall_pending)) |
299 | force_evtchn_callback(); | |
5ead97c8 JF |
300 | } |
301 | ||
302 | static void xen_safe_halt(void) | |
303 | { | |
304 | /* Blocking includes an implicit local_irq_enable(). */ | |
349c709f | 305 | if (HYPERVISOR_sched_op(SCHEDOP_block, NULL) != 0) |
5ead97c8 JF |
306 | BUG(); |
307 | } | |
308 | ||
309 | static void xen_halt(void) | |
310 | { | |
311 | if (irqs_disabled()) | |
312 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
313 | else | |
314 | xen_safe_halt(); | |
315 | } | |
316 | ||
8965c1c0 | 317 | static void xen_leave_lazy(void) |
5ead97c8 | 318 | { |
8965c1c0 | 319 | paravirt_leave_lazy(paravirt_get_lazy_mode()); |
5ead97c8 | 320 | xen_mc_flush(); |
5ead97c8 JF |
321 | } |
322 | ||
323 | static unsigned long xen_store_tr(void) | |
324 | { | |
325 | return 0; | |
326 | } | |
327 | ||
38ffbe66 JF |
328 | static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) |
329 | { | |
330 | unsigned pages = roundup(entries * LDT_ENTRY_SIZE, PAGE_SIZE); | |
331 | void *v = ldt; | |
332 | int i; | |
333 | ||
334 | for(i = 0; i < pages; i += PAGE_SIZE) | |
335 | make_lowmem_page_readonly(v + i); | |
336 | } | |
337 | ||
338 | static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) | |
339 | { | |
340 | unsigned pages = roundup(entries * LDT_ENTRY_SIZE, PAGE_SIZE); | |
341 | void *v = ldt; | |
342 | int i; | |
343 | ||
344 | for(i = 0; i < pages; i += PAGE_SIZE) | |
345 | make_lowmem_page_readwrite(v + i); | |
346 | } | |
347 | ||
5ead97c8 JF |
348 | static void xen_set_ldt(const void *addr, unsigned entries) |
349 | { | |
5ead97c8 JF |
350 | struct mmuext_op *op; |
351 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
352 | ||
353 | op = mcs.args; | |
354 | op->cmd = MMUEXT_SET_LDT; | |
4dbf7af6 | 355 | op->arg1.linear_addr = (unsigned long)addr; |
5ead97c8 JF |
356 | op->arg2.nr_ents = entries; |
357 | ||
358 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
359 | ||
360 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
361 | } | |
362 | ||
6b68f01b | 363 | static void xen_load_gdt(const struct desc_ptr *dtr) |
5ead97c8 JF |
364 | { |
365 | unsigned long *frames; | |
366 | unsigned long va = dtr->address; | |
367 | unsigned int size = dtr->size + 1; | |
368 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
369 | int f; | |
370 | struct multicall_space mcs; | |
371 | ||
372 | /* A GDT can be up to 64k in size, which corresponds to 8192 | |
373 | 8-byte entries, or 16 4k pages.. */ | |
374 | ||
375 | BUG_ON(size > 65536); | |
376 | BUG_ON(va & ~PAGE_MASK); | |
377 | ||
378 | mcs = xen_mc_entry(sizeof(*frames) * pages); | |
379 | frames = mcs.args; | |
380 | ||
381 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
382 | frames[f] = virt_to_mfn(va); | |
383 | make_lowmem_page_readonly((void *)va); | |
384 | } | |
385 | ||
386 | MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct)); | |
387 | ||
388 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
389 | } | |
390 | ||
391 | static void load_TLS_descriptor(struct thread_struct *t, | |
392 | unsigned int cpu, unsigned int i) | |
393 | { | |
394 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
395 | xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); | |
396 | struct multicall_space mc = __xen_mc_entry(0); | |
397 | ||
398 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
399 | } | |
400 | ||
401 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
402 | { | |
8b84ad94 JF |
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. | |
8a95408e EH |
411 | * |
412 | * On x86_64, this hack is not used for %gs, because gs points | |
413 | * to KERNEL_GS_BASE (and uses it for PDA references), so we | |
414 | * must not zero %gs on x86_64 | |
415 | * | |
416 | * For x86_64, we need to zero %fs, otherwise we may get an | |
417 | * exception between the new %fs descriptor being loaded and | |
418 | * %fs being effectively cleared at __switch_to(). | |
8b84ad94 | 419 | */ |
8a95408e EH |
420 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { |
421 | #ifdef CONFIG_X86_32 | |
8b84ad94 | 422 | loadsegment(gs, 0); |
8a95408e EH |
423 | #else |
424 | loadsegment(fs, 0); | |
425 | #endif | |
426 | } | |
427 | ||
428 | xen_mc_batch(); | |
429 | ||
430 | load_TLS_descriptor(t, cpu, 0); | |
431 | load_TLS_descriptor(t, cpu, 1); | |
432 | load_TLS_descriptor(t, cpu, 2); | |
433 | ||
434 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
5ead97c8 JF |
435 | } |
436 | ||
a8fc1089 EH |
437 | #ifdef CONFIG_X86_64 |
438 | static void xen_load_gs_index(unsigned int idx) | |
439 | { | |
440 | if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) | |
441 | BUG(); | |
5ead97c8 | 442 | } |
a8fc1089 | 443 | #endif |
5ead97c8 JF |
444 | |
445 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
75b8bb3e | 446 | const void *ptr) |
5ead97c8 JF |
447 | { |
448 | unsigned long lp = (unsigned long)&dt[entrynum]; | |
449 | xmaddr_t mach_lp = virt_to_machine(lp); | |
75b8bb3e | 450 | u64 entry = *(u64 *)ptr; |
5ead97c8 | 451 | |
f120f13e JF |
452 | preempt_disable(); |
453 | ||
5ead97c8 JF |
454 | xen_mc_flush(); |
455 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
456 | BUG(); | |
f120f13e JF |
457 | |
458 | preempt_enable(); | |
5ead97c8 JF |
459 | } |
460 | ||
e176d367 | 461 | static int cvt_gate_to_trap(int vector, const gate_desc *val, |
5ead97c8 JF |
462 | struct trap_info *info) |
463 | { | |
e176d367 | 464 | if (val->type != 0xf && val->type != 0xe) |
5ead97c8 JF |
465 | return 0; |
466 | ||
467 | info->vector = vector; | |
e176d367 EH |
468 | info->address = gate_offset(*val); |
469 | info->cs = gate_segment(*val); | |
470 | info->flags = val->dpl; | |
5ead97c8 | 471 | /* interrupt gates clear IF */ |
e176d367 | 472 | if (val->type == 0xe) |
5ead97c8 JF |
473 | info->flags |= 4; |
474 | ||
475 | return 1; | |
476 | } | |
477 | ||
478 | /* Locations of each CPU's IDT */ | |
6b68f01b | 479 | static DEFINE_PER_CPU(struct desc_ptr, idt_desc); |
5ead97c8 JF |
480 | |
481 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
482 | also update Xen. */ | |
8d947344 | 483 | static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) |
5ead97c8 | 484 | { |
5ead97c8 | 485 | unsigned long p = (unsigned long)&dt[entrynum]; |
f120f13e JF |
486 | unsigned long start, end; |
487 | ||
488 | preempt_disable(); | |
489 | ||
490 | start = __get_cpu_var(idt_desc).address; | |
491 | end = start + __get_cpu_var(idt_desc).size + 1; | |
5ead97c8 JF |
492 | |
493 | xen_mc_flush(); | |
494 | ||
8d947344 | 495 | native_write_idt_entry(dt, entrynum, g); |
5ead97c8 JF |
496 | |
497 | if (p >= start && (p + 8) <= end) { | |
498 | struct trap_info info[2]; | |
499 | ||
500 | info[1].address = 0; | |
501 | ||
e176d367 | 502 | if (cvt_gate_to_trap(entrynum, g, &info[0])) |
5ead97c8 JF |
503 | if (HYPERVISOR_set_trap_table(info)) |
504 | BUG(); | |
505 | } | |
f120f13e JF |
506 | |
507 | preempt_enable(); | |
5ead97c8 JF |
508 | } |
509 | ||
6b68f01b | 510 | static void xen_convert_trap_info(const struct desc_ptr *desc, |
f87e4cac | 511 | struct trap_info *traps) |
5ead97c8 | 512 | { |
5ead97c8 JF |
513 | unsigned in, out, count; |
514 | ||
e176d367 | 515 | count = (desc->size+1) / sizeof(gate_desc); |
5ead97c8 JF |
516 | BUG_ON(count > 256); |
517 | ||
5ead97c8 | 518 | for (in = out = 0; in < count; in++) { |
e176d367 | 519 | gate_desc *entry = (gate_desc*)(desc->address) + in; |
5ead97c8 | 520 | |
e176d367 | 521 | if (cvt_gate_to_trap(in, entry, &traps[out])) |
5ead97c8 JF |
522 | out++; |
523 | } | |
524 | traps[out].address = 0; | |
f87e4cac JF |
525 | } |
526 | ||
527 | void xen_copy_trap_info(struct trap_info *traps) | |
528 | { | |
6b68f01b | 529 | const struct desc_ptr *desc = &__get_cpu_var(idt_desc); |
f87e4cac JF |
530 | |
531 | xen_convert_trap_info(desc, traps); | |
f87e4cac JF |
532 | } |
533 | ||
534 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
535 | hold a spinlock to protect the static traps[] array (static because | |
536 | it avoids allocation, and saves stack space). */ | |
6b68f01b | 537 | static void xen_load_idt(const struct desc_ptr *desc) |
f87e4cac JF |
538 | { |
539 | static DEFINE_SPINLOCK(lock); | |
540 | static struct trap_info traps[257]; | |
f87e4cac JF |
541 | |
542 | spin_lock(&lock); | |
543 | ||
f120f13e JF |
544 | __get_cpu_var(idt_desc) = *desc; |
545 | ||
f87e4cac | 546 | xen_convert_trap_info(desc, traps); |
5ead97c8 JF |
547 | |
548 | xen_mc_flush(); | |
549 | if (HYPERVISOR_set_trap_table(traps)) | |
550 | BUG(); | |
551 | ||
552 | spin_unlock(&lock); | |
553 | } | |
554 | ||
555 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
556 | they're handled differently. */ | |
557 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
014b15be | 558 | const void *desc, int type) |
5ead97c8 | 559 | { |
f120f13e JF |
560 | preempt_disable(); |
561 | ||
014b15be GOC |
562 | switch (type) { |
563 | case DESC_LDT: | |
564 | case DESC_TSS: | |
5ead97c8 JF |
565 | /* ignore */ |
566 | break; | |
567 | ||
568 | default: { | |
569 | xmaddr_t maddr = virt_to_machine(&dt[entry]); | |
5ead97c8 JF |
570 | |
571 | xen_mc_flush(); | |
014b15be | 572 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) |
5ead97c8 JF |
573 | BUG(); |
574 | } | |
575 | ||
576 | } | |
f120f13e JF |
577 | |
578 | preempt_enable(); | |
5ead97c8 JF |
579 | } |
580 | ||
faca6227 | 581 | static void xen_load_sp0(struct tss_struct *tss, |
f120f13e | 582 | struct thread_struct *thread) |
5ead97c8 JF |
583 | { |
584 | struct multicall_space mcs = xen_mc_entry(0); | |
faca6227 | 585 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); |
5ead97c8 JF |
586 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
587 | } | |
588 | ||
589 | static void xen_set_iopl_mask(unsigned mask) | |
590 | { | |
591 | struct physdev_set_iopl set_iopl; | |
592 | ||
593 | /* Force the change at ring 0. */ | |
594 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
595 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
596 | } | |
597 | ||
598 | static void xen_io_delay(void) | |
599 | { | |
600 | } | |
601 | ||
602 | #ifdef CONFIG_X86_LOCAL_APIC | |
42e0a9aa | 603 | static u32 xen_apic_read(unsigned long reg) |
5ead97c8 JF |
604 | { |
605 | return 0; | |
606 | } | |
f87e4cac | 607 | |
42e0a9aa | 608 | static void xen_apic_write(unsigned long reg, u32 val) |
f87e4cac JF |
609 | { |
610 | /* Warn to see if there's any stray references */ | |
611 | WARN_ON(1); | |
612 | } | |
5ead97c8 JF |
613 | #endif |
614 | ||
615 | static void xen_flush_tlb(void) | |
616 | { | |
d66bf8fc | 617 | struct mmuext_op *op; |
41e332b2 JF |
618 | struct multicall_space mcs; |
619 | ||
620 | preempt_disable(); | |
621 | ||
622 | mcs = xen_mc_entry(sizeof(*op)); | |
5ead97c8 | 623 | |
d66bf8fc JF |
624 | op = mcs.args; |
625 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
626 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
627 | ||
628 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
41e332b2 JF |
629 | |
630 | preempt_enable(); | |
5ead97c8 JF |
631 | } |
632 | ||
633 | static void xen_flush_tlb_single(unsigned long addr) | |
634 | { | |
d66bf8fc | 635 | struct mmuext_op *op; |
41e332b2 JF |
636 | struct multicall_space mcs; |
637 | ||
638 | preempt_disable(); | |
5ead97c8 | 639 | |
41e332b2 | 640 | mcs = xen_mc_entry(sizeof(*op)); |
d66bf8fc JF |
641 | op = mcs.args; |
642 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
643 | op->arg1.linear_addr = addr & PAGE_MASK; | |
644 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
645 | ||
646 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
41e332b2 JF |
647 | |
648 | preempt_enable(); | |
5ead97c8 JF |
649 | } |
650 | ||
f87e4cac JF |
651 | static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm, |
652 | unsigned long va) | |
653 | { | |
d66bf8fc JF |
654 | struct { |
655 | struct mmuext_op op; | |
656 | cpumask_t mask; | |
657 | } *args; | |
f87e4cac | 658 | cpumask_t cpumask = *cpus; |
d66bf8fc | 659 | struct multicall_space mcs; |
f87e4cac JF |
660 | |
661 | /* | |
662 | * A couple of (to be removed) sanity checks: | |
663 | * | |
664 | * - current CPU must not be in mask | |
665 | * - mask must exist :) | |
666 | */ | |
667 | BUG_ON(cpus_empty(cpumask)); | |
668 | BUG_ON(cpu_isset(smp_processor_id(), cpumask)); | |
669 | BUG_ON(!mm); | |
670 | ||
671 | /* If a CPU which we ran on has gone down, OK. */ | |
672 | cpus_and(cpumask, cpumask, cpu_online_map); | |
673 | if (cpus_empty(cpumask)) | |
674 | return; | |
675 | ||
d66bf8fc JF |
676 | mcs = xen_mc_entry(sizeof(*args)); |
677 | args = mcs.args; | |
678 | args->mask = cpumask; | |
679 | args->op.arg2.vcpumask = &args->mask; | |
680 | ||
f87e4cac | 681 | if (va == TLB_FLUSH_ALL) { |
d66bf8fc | 682 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; |
f87e4cac | 683 | } else { |
d66bf8fc JF |
684 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
685 | args->op.arg1.linear_addr = va; | |
f87e4cac JF |
686 | } |
687 | ||
d66bf8fc JF |
688 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); |
689 | ||
690 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f87e4cac JF |
691 | } |
692 | ||
7b1333aa JF |
693 | static void xen_clts(void) |
694 | { | |
695 | struct multicall_space mcs; | |
696 | ||
697 | mcs = xen_mc_entry(0); | |
698 | ||
699 | MULTI_fpu_taskswitch(mcs.mc, 0); | |
700 | ||
701 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
702 | } | |
703 | ||
704 | static void xen_write_cr0(unsigned long cr0) | |
705 | { | |
706 | struct multicall_space mcs; | |
707 | ||
708 | /* Only pay attention to cr0.TS; everything else is | |
709 | ignored. */ | |
710 | mcs = xen_mc_entry(0); | |
711 | ||
712 | MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); | |
713 | ||
714 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
715 | } | |
716 | ||
60223a32 JF |
717 | static void xen_write_cr2(unsigned long cr2) |
718 | { | |
719 | x86_read_percpu(xen_vcpu)->arch.cr2 = cr2; | |
720 | } | |
721 | ||
5ead97c8 JF |
722 | static unsigned long xen_read_cr2(void) |
723 | { | |
724 | return x86_read_percpu(xen_vcpu)->arch.cr2; | |
725 | } | |
726 | ||
60223a32 JF |
727 | static unsigned long xen_read_cr2_direct(void) |
728 | { | |
729 | return x86_read_percpu(xen_vcpu_info.arch.cr2); | |
730 | } | |
731 | ||
5ead97c8 JF |
732 | static void xen_write_cr4(unsigned long cr4) |
733 | { | |
2956a351 JF |
734 | cr4 &= ~X86_CR4_PGE; |
735 | cr4 &= ~X86_CR4_PSE; | |
736 | ||
737 | native_write_cr4(cr4); | |
5ead97c8 JF |
738 | } |
739 | ||
5ead97c8 JF |
740 | static unsigned long xen_read_cr3(void) |
741 | { | |
742 | return x86_read_percpu(xen_cr3); | |
743 | } | |
744 | ||
9f79991d JF |
745 | static void set_current_cr3(void *v) |
746 | { | |
747 | x86_write_percpu(xen_current_cr3, (unsigned long)v); | |
748 | } | |
749 | ||
d6182fbf | 750 | static void __xen_write_cr3(bool kernel, unsigned long cr3) |
5ead97c8 | 751 | { |
9f79991d JF |
752 | struct mmuext_op *op; |
753 | struct multicall_space mcs; | |
d6182fbf | 754 | unsigned long mfn; |
9f79991d | 755 | |
d6182fbf JF |
756 | if (cr3) |
757 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
758 | else | |
759 | mfn = 0; | |
f120f13e | 760 | |
d6182fbf | 761 | WARN_ON(mfn == 0 && kernel); |
5ead97c8 | 762 | |
d6182fbf | 763 | mcs = __xen_mc_entry(sizeof(*op)); |
5ead97c8 | 764 | |
9f79991d | 765 | op = mcs.args; |
d6182fbf | 766 | op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; |
9f79991d | 767 | op->arg1.mfn = mfn; |
5ead97c8 | 768 | |
9f79991d | 769 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); |
5ead97c8 | 770 | |
d6182fbf JF |
771 | if (kernel) { |
772 | x86_write_percpu(xen_cr3, cr3); | |
773 | ||
774 | /* Update xen_current_cr3 once the batch has actually | |
775 | been submitted. */ | |
776 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
777 | } | |
778 | } | |
779 | ||
780 | static void xen_write_cr3(unsigned long cr3) | |
781 | { | |
782 | BUG_ON(preemptible()); | |
783 | ||
784 | xen_mc_batch(); /* disables interrupts */ | |
785 | ||
786 | /* Update while interrupts are disabled, so its atomic with | |
787 | respect to ipis */ | |
788 | x86_write_percpu(xen_cr3, cr3); | |
789 | ||
790 | __xen_write_cr3(true, cr3); | |
791 | ||
792 | #ifdef CONFIG_X86_64 | |
793 | { | |
794 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
795 | if (user_pgd) | |
796 | __xen_write_cr3(false, __pa(user_pgd)); | |
797 | else | |
798 | __xen_write_cr3(false, 0); | |
799 | } | |
800 | #endif | |
5ead97c8 | 801 | |
9f79991d | 802 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ |
5ead97c8 JF |
803 | } |
804 | ||
1153968a JF |
805 | static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) |
806 | { | |
807 | int ret; | |
808 | ||
809 | ret = 0; | |
810 | ||
811 | switch(msr) { | |
812 | #ifdef CONFIG_X86_64 | |
813 | unsigned which; | |
814 | u64 base; | |
815 | ||
816 | case MSR_FS_BASE: which = SEGBASE_FS; goto set; | |
817 | case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; | |
818 | case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; | |
819 | ||
820 | set: | |
821 | base = ((u64)high << 32) | low; | |
822 | if (HYPERVISOR_set_segment_base(which, base) != 0) | |
823 | ret = -EFAULT; | |
824 | break; | |
825 | #endif | |
826 | default: | |
827 | ret = native_write_msr_safe(msr, low, high); | |
828 | } | |
829 | ||
830 | return ret; | |
831 | } | |
832 | ||
f4f97b3e JF |
833 | /* Early in boot, while setting up the initial pagetable, assume |
834 | everything is pinned. */ | |
6944a9c8 | 835 | static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn) |
5ead97c8 | 836 | { |
af7ae3b9 | 837 | #ifdef CONFIG_FLATMEM |
f4f97b3e | 838 | BUG_ON(mem_map); /* should only be used early */ |
af7ae3b9 | 839 | #endif |
5ead97c8 JF |
840 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
841 | } | |
842 | ||
6944a9c8 | 843 | /* Early release_pte assumes that all pts are pinned, since there's |
1c70e9bd | 844 | only init_mm and anything attached to that is pinned. */ |
6944a9c8 | 845 | static void xen_release_pte_init(u32 pfn) |
1c70e9bd JF |
846 | { |
847 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
848 | } | |
849 | ||
f6433706 | 850 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
74260714 JF |
851 | { |
852 | struct mmuext_op op; | |
f6433706 | 853 | op.cmd = cmd; |
74260714 JF |
854 | op.arg1.mfn = pfn_to_mfn(pfn); |
855 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
856 | BUG(); | |
857 | } | |
858 | ||
f4f97b3e JF |
859 | /* This needs to make sure the new pte page is pinned iff its being |
860 | attached to a pinned pagetable. */ | |
1c70e9bd | 861 | static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level) |
5ead97c8 | 862 | { |
f4f97b3e | 863 | struct page *page = pfn_to_page(pfn); |
5ead97c8 | 864 | |
f4f97b3e JF |
865 | if (PagePinned(virt_to_page(mm->pgd))) { |
866 | SetPagePinned(page); | |
867 | ||
74260714 | 868 | if (!PageHighMem(page)) { |
f4f97b3e | 869 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
f6433706 MM |
870 | if (level == PT_PTE) |
871 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
74260714 | 872 | } else |
f4f97b3e JF |
873 | /* make sure there are no stray mappings of |
874 | this page */ | |
875 | kmap_flush_unused(); | |
876 | } | |
5ead97c8 JF |
877 | } |
878 | ||
6944a9c8 | 879 | static void xen_alloc_pte(struct mm_struct *mm, u32 pfn) |
1c70e9bd | 880 | { |
f6433706 | 881 | xen_alloc_ptpage(mm, pfn, PT_PTE); |
1c70e9bd JF |
882 | } |
883 | ||
6944a9c8 | 884 | static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn) |
1c70e9bd | 885 | { |
f6433706 | 886 | xen_alloc_ptpage(mm, pfn, PT_PMD); |
1c70e9bd JF |
887 | } |
888 | ||
d6182fbf JF |
889 | static int xen_pgd_alloc(struct mm_struct *mm) |
890 | { | |
891 | pgd_t *pgd = mm->pgd; | |
892 | int ret = 0; | |
893 | ||
894 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
895 | ||
896 | #ifdef CONFIG_X86_64 | |
897 | { | |
898 | struct page *page = virt_to_page(pgd); | |
bf18bf94 | 899 | pgd_t *user_pgd; |
d6182fbf JF |
900 | |
901 | BUG_ON(page->private != 0); | |
902 | ||
bf18bf94 JF |
903 | ret = -ENOMEM; |
904 | ||
905 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
906 | page->private = (unsigned long)user_pgd; | |
907 | ||
908 | if (user_pgd != NULL) { | |
909 | user_pgd[pgd_index(VSYSCALL_START)] = | |
910 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
911 | ret = 0; | |
912 | } | |
d6182fbf JF |
913 | |
914 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
915 | } | |
916 | #endif | |
917 | ||
918 | return ret; | |
919 | } | |
920 | ||
921 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
922 | { | |
923 | #ifdef CONFIG_X86_64 | |
924 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
925 | ||
926 | if (user_pgd) | |
927 | free_page((unsigned long)user_pgd); | |
928 | #endif | |
929 | } | |
930 | ||
f4f97b3e | 931 | /* This should never happen until we're OK to use struct page */ |
f6433706 | 932 | static void xen_release_ptpage(u32 pfn, unsigned level) |
5ead97c8 | 933 | { |
f4f97b3e JF |
934 | struct page *page = pfn_to_page(pfn); |
935 | ||
936 | if (PagePinned(page)) { | |
74260714 | 937 | if (!PageHighMem(page)) { |
a684d69d MM |
938 | if (level == PT_PTE) |
939 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
f4f97b3e | 940 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
74260714 | 941 | } |
c946c7de | 942 | ClearPagePinned(page); |
f4f97b3e | 943 | } |
5ead97c8 JF |
944 | } |
945 | ||
6944a9c8 | 946 | static void xen_release_pte(u32 pfn) |
f6433706 MM |
947 | { |
948 | xen_release_ptpage(pfn, PT_PTE); | |
949 | } | |
950 | ||
6944a9c8 | 951 | static void xen_release_pmd(u32 pfn) |
f6433706 MM |
952 | { |
953 | xen_release_ptpage(pfn, PT_PMD); | |
954 | } | |
955 | ||
f6e58732 JF |
956 | #if PAGETABLE_LEVELS == 4 |
957 | static void xen_alloc_pud(struct mm_struct *mm, u32 pfn) | |
958 | { | |
959 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
960 | } | |
961 | ||
962 | static void xen_release_pud(u32 pfn) | |
963 | { | |
964 | xen_release_ptpage(pfn, PT_PUD); | |
965 | } | |
966 | #endif | |
967 | ||
f4f97b3e JF |
968 | #ifdef CONFIG_HIGHPTE |
969 | static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) | |
5ead97c8 | 970 | { |
f4f97b3e JF |
971 | pgprot_t prot = PAGE_KERNEL; |
972 | ||
973 | if (PagePinned(page)) | |
974 | prot = PAGE_KERNEL_RO; | |
975 | ||
976 | if (0 && PageHighMem(page)) | |
977 | printk("mapping highpte %lx type %d prot %s\n", | |
978 | page_to_pfn(page), type, | |
979 | (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); | |
980 | ||
981 | return kmap_atomic_prot(page, type, prot); | |
5ead97c8 | 982 | } |
f4f97b3e | 983 | #endif |
5ead97c8 | 984 | |
9a4029fd JF |
985 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) |
986 | { | |
987 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
988 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
989 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
990 | pte_val_ma(pte)); | |
991 | ||
992 | return pte; | |
993 | } | |
994 | ||
995 | /* Init-time set_pte while constructing initial pagetables, which | |
996 | doesn't allow RO pagetable pages to be remapped RW */ | |
997 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
998 | { | |
999 | pte = mask_rw_pte(ptep, pte); | |
1000 | ||
1001 | xen_set_pte(ptep, pte); | |
1002 | } | |
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 */ | |
5b09b287 | 1069 | #ifdef CONFIG_X86_32 |
60223a32 JF |
1070 | if (have_vcpu_info_placement) { |
1071 | printk(KERN_INFO "Xen: using vcpu_info placement\n"); | |
1072 | ||
93b1eab3 JF |
1073 | pv_irq_ops.save_fl = xen_save_fl_direct; |
1074 | pv_irq_ops.restore_fl = xen_restore_fl_direct; | |
1075 | pv_irq_ops.irq_disable = xen_irq_disable_direct; | |
1076 | pv_irq_ops.irq_enable = xen_irq_enable_direct; | |
1077 | pv_mmu_ops.read_cr2 = xen_read_cr2_direct; | |
60223a32 | 1078 | } |
5b09b287 | 1079 | #endif |
5ead97c8 JF |
1080 | } |
1081 | ||
ab144f5e AK |
1082 | static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, |
1083 | unsigned long addr, unsigned len) | |
6487673b JF |
1084 | { |
1085 | char *start, *end, *reloc; | |
1086 | unsigned ret; | |
1087 | ||
1088 | start = end = reloc = NULL; | |
1089 | ||
93b1eab3 JF |
1090 | #define SITE(op, x) \ |
1091 | case PARAVIRT_PATCH(op.x): \ | |
6487673b JF |
1092 | if (have_vcpu_info_placement) { \ |
1093 | start = (char *)xen_##x##_direct; \ | |
1094 | end = xen_##x##_direct_end; \ | |
1095 | reloc = xen_##x##_direct_reloc; \ | |
1096 | } \ | |
1097 | goto patch_site | |
1098 | ||
1099 | switch (type) { | |
5b09b287 | 1100 | #ifdef CONFIG_X86_32 |
93b1eab3 JF |
1101 | SITE(pv_irq_ops, irq_enable); |
1102 | SITE(pv_irq_ops, irq_disable); | |
1103 | SITE(pv_irq_ops, save_fl); | |
1104 | SITE(pv_irq_ops, restore_fl); | |
5b09b287 | 1105 | #endif /* CONFIG_X86_32 */ |
6487673b JF |
1106 | #undef SITE |
1107 | ||
1108 | patch_site: | |
1109 | if (start == NULL || (end-start) > len) | |
1110 | goto default_patch; | |
1111 | ||
ab144f5e | 1112 | ret = paravirt_patch_insns(insnbuf, len, start, end); |
6487673b JF |
1113 | |
1114 | /* Note: because reloc is assigned from something that | |
1115 | appears to be an array, gcc assumes it's non-null, | |
1116 | but doesn't know its relationship with start and | |
1117 | end. */ | |
1118 | if (reloc > start && reloc < end) { | |
1119 | int reloc_off = reloc - start; | |
ab144f5e AK |
1120 | long *relocp = (long *)(insnbuf + reloc_off); |
1121 | long delta = start - (char *)addr; | |
6487673b JF |
1122 | |
1123 | *relocp += delta; | |
1124 | } | |
1125 | break; | |
1126 | ||
1127 | default_patch: | |
1128 | default: | |
ab144f5e AK |
1129 | ret = paravirt_patch_default(type, clobbers, insnbuf, |
1130 | addr, len); | |
6487673b JF |
1131 | break; |
1132 | } | |
1133 | ||
1134 | return ret; | |
1135 | } | |
1136 | ||
aeaaa59c JF |
1137 | static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot) |
1138 | { | |
1139 | pte_t pte; | |
1140 | ||
1141 | phys >>= PAGE_SHIFT; | |
1142 | ||
1143 | switch (idx) { | |
1144 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
1145 | #ifdef CONFIG_X86_F00F_BUG | |
1146 | case FIX_F00F_IDT: | |
1147 | #endif | |
15664f96 | 1148 | #ifdef CONFIG_X86_32 |
aeaaa59c JF |
1149 | case FIX_WP_TEST: |
1150 | case FIX_VDSO: | |
b3fe1243 | 1151 | # ifdef CONFIG_HIGHMEM |
15664f96 | 1152 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: |
b3fe1243 | 1153 | # endif |
15664f96 JF |
1154 | #else |
1155 | case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE: | |
1156 | #endif | |
aeaaa59c JF |
1157 | #ifdef CONFIG_X86_LOCAL_APIC |
1158 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1159 | #endif | |
1160 | pte = pfn_pte(phys, prot); | |
1161 | break; | |
1162 | ||
1163 | default: | |
1164 | pte = mfn_pte(phys, prot); | |
1165 | break; | |
1166 | } | |
1167 | ||
1168 | __native_set_fixmap(idx, pte); | |
bf18bf94 JF |
1169 | |
1170 | #ifdef CONFIG_X86_64 | |
1171 | /* Replicate changes to map the vsyscall page into the user | |
1172 | pagetable vsyscall mapping. */ | |
1173 | if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) { | |
1174 | unsigned long vaddr = __fix_to_virt(idx); | |
1175 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
1176 | } | |
1177 | #endif | |
aeaaa59c JF |
1178 | } |
1179 | ||
93b1eab3 | 1180 | static const struct pv_info xen_info __initdata = { |
5ead97c8 JF |
1181 | .paravirt_enabled = 1, |
1182 | .shared_kernel_pmd = 0, | |
1183 | ||
1184 | .name = "Xen", | |
93b1eab3 | 1185 | }; |
5ead97c8 | 1186 | |
93b1eab3 | 1187 | static const struct pv_init_ops xen_init_ops __initdata = { |
6487673b | 1188 | .patch = xen_patch, |
5ead97c8 | 1189 | |
93b1eab3 | 1190 | .banner = xen_banner, |
5ead97c8 JF |
1191 | .memory_setup = xen_memory_setup, |
1192 | .arch_setup = xen_arch_setup, | |
e2426cf8 | 1193 | .post_allocator_init = xen_post_allocator_init, |
93b1eab3 | 1194 | }; |
5ead97c8 | 1195 | |
93b1eab3 | 1196 | static const struct pv_time_ops xen_time_ops __initdata = { |
15c84731 | 1197 | .time_init = xen_time_init, |
93b1eab3 | 1198 | |
15c84731 JF |
1199 | .set_wallclock = xen_set_wallclock, |
1200 | .get_wallclock = xen_get_wallclock, | |
e93ef949 | 1201 | .get_tsc_khz = xen_tsc_khz, |
ab550288 | 1202 | .sched_clock = xen_sched_clock, |
93b1eab3 | 1203 | }; |
15c84731 | 1204 | |
93b1eab3 | 1205 | static const struct pv_cpu_ops xen_cpu_ops __initdata = { |
5ead97c8 JF |
1206 | .cpuid = xen_cpuid, |
1207 | ||
1208 | .set_debugreg = xen_set_debugreg, | |
1209 | .get_debugreg = xen_get_debugreg, | |
1210 | ||
7b1333aa | 1211 | .clts = xen_clts, |
5ead97c8 JF |
1212 | |
1213 | .read_cr0 = native_read_cr0, | |
7b1333aa | 1214 | .write_cr0 = xen_write_cr0, |
5ead97c8 | 1215 | |
5ead97c8 JF |
1216 | .read_cr4 = native_read_cr4, |
1217 | .read_cr4_safe = native_read_cr4_safe, | |
1218 | .write_cr4 = xen_write_cr4, | |
1219 | ||
5ead97c8 JF |
1220 | .wbinvd = native_wbinvd, |
1221 | ||
1222 | .read_msr = native_read_msr_safe, | |
1153968a | 1223 | .write_msr = xen_write_msr_safe, |
5ead97c8 JF |
1224 | .read_tsc = native_read_tsc, |
1225 | .read_pmc = native_read_pmc, | |
1226 | ||
81e103f1 | 1227 | .iret = xen_iret, |
d75cd22f | 1228 | .irq_enable_sysexit = xen_sysexit, |
6fcac6d3 JF |
1229 | #ifdef CONFIG_X86_64 |
1230 | .usergs_sysret32 = xen_sysret32, | |
1231 | .usergs_sysret64 = xen_sysret64, | |
1232 | #endif | |
5ead97c8 JF |
1233 | |
1234 | .load_tr_desc = paravirt_nop, | |
1235 | .set_ldt = xen_set_ldt, | |
1236 | .load_gdt = xen_load_gdt, | |
1237 | .load_idt = xen_load_idt, | |
1238 | .load_tls = xen_load_tls, | |
a8fc1089 EH |
1239 | #ifdef CONFIG_X86_64 |
1240 | .load_gs_index = xen_load_gs_index, | |
1241 | #endif | |
5ead97c8 | 1242 | |
38ffbe66 JF |
1243 | .alloc_ldt = xen_alloc_ldt, |
1244 | .free_ldt = xen_free_ldt, | |
1245 | ||
5ead97c8 JF |
1246 | .store_gdt = native_store_gdt, |
1247 | .store_idt = native_store_idt, | |
1248 | .store_tr = xen_store_tr, | |
1249 | ||
1250 | .write_ldt_entry = xen_write_ldt_entry, | |
1251 | .write_gdt_entry = xen_write_gdt_entry, | |
1252 | .write_idt_entry = xen_write_idt_entry, | |
faca6227 | 1253 | .load_sp0 = xen_load_sp0, |
5ead97c8 JF |
1254 | |
1255 | .set_iopl_mask = xen_set_iopl_mask, | |
1256 | .io_delay = xen_io_delay, | |
1257 | ||
952d1d70 JF |
1258 | /* Xen takes care of %gs when switching to usermode for us */ |
1259 | .swapgs = paravirt_nop, | |
1260 | ||
8965c1c0 JF |
1261 | .lazy_mode = { |
1262 | .enter = paravirt_enter_lazy_cpu, | |
1263 | .leave = xen_leave_lazy, | |
1264 | }, | |
93b1eab3 JF |
1265 | }; |
1266 | ||
0725cbb9 JF |
1267 | static void __init __xen_init_IRQ(void) |
1268 | { | |
1269 | #ifdef CONFIG_X86_64 | |
1270 | int i; | |
1271 | ||
1272 | /* Create identity vector->irq map */ | |
1273 | for(i = 0; i < NR_VECTORS; i++) { | |
1274 | int cpu; | |
1275 | ||
1276 | for_each_possible_cpu(cpu) | |
1277 | per_cpu(vector_irq, cpu)[i] = i; | |
1278 | } | |
1279 | #endif /* CONFIG_X86_64 */ | |
1280 | ||
1281 | xen_init_IRQ(); | |
1282 | } | |
1283 | ||
93b1eab3 | 1284 | static const struct pv_irq_ops xen_irq_ops __initdata = { |
0725cbb9 | 1285 | .init_IRQ = __xen_init_IRQ, |
93b1eab3 JF |
1286 | .save_fl = xen_save_fl, |
1287 | .restore_fl = xen_restore_fl, | |
1288 | .irq_disable = xen_irq_disable, | |
1289 | .irq_enable = xen_irq_enable, | |
1290 | .safe_halt = xen_safe_halt, | |
1291 | .halt = xen_halt, | |
fab58420 | 1292 | #ifdef CONFIG_X86_64 |
997409d3 | 1293 | .adjust_exception_frame = xen_adjust_exception_frame, |
fab58420 | 1294 | #endif |
93b1eab3 | 1295 | }; |
5ead97c8 | 1296 | |
93b1eab3 | 1297 | static const struct pv_apic_ops xen_apic_ops __initdata = { |
5ead97c8 | 1298 | #ifdef CONFIG_X86_LOCAL_APIC |
f87e4cac | 1299 | .apic_write = xen_apic_write, |
5ead97c8 JF |
1300 | .apic_read = xen_apic_read, |
1301 | .setup_boot_clock = paravirt_nop, | |
1302 | .setup_secondary_clock = paravirt_nop, | |
1303 | .startup_ipi_hook = paravirt_nop, | |
1304 | #endif | |
93b1eab3 JF |
1305 | }; |
1306 | ||
1307 | static const struct pv_mmu_ops xen_mmu_ops __initdata = { | |
1308 | .pagetable_setup_start = xen_pagetable_setup_start, | |
1309 | .pagetable_setup_done = xen_pagetable_setup_done, | |
1310 | ||
1311 | .read_cr2 = xen_read_cr2, | |
1312 | .write_cr2 = xen_write_cr2, | |
1313 | ||
1314 | .read_cr3 = xen_read_cr3, | |
1315 | .write_cr3 = xen_write_cr3, | |
5ead97c8 JF |
1316 | |
1317 | .flush_tlb_user = xen_flush_tlb, | |
1318 | .flush_tlb_kernel = xen_flush_tlb, | |
1319 | .flush_tlb_single = xen_flush_tlb_single, | |
f87e4cac | 1320 | .flush_tlb_others = xen_flush_tlb_others, |
5ead97c8 JF |
1321 | |
1322 | .pte_update = paravirt_nop, | |
1323 | .pte_update_defer = paravirt_nop, | |
1324 | ||
d6182fbf JF |
1325 | .pgd_alloc = xen_pgd_alloc, |
1326 | .pgd_free = xen_pgd_free, | |
eba0045f | 1327 | |
6944a9c8 JF |
1328 | .alloc_pte = xen_alloc_pte_init, |
1329 | .release_pte = xen_release_pte_init, | |
1330 | .alloc_pmd = xen_alloc_pte_init, | |
1331 | .alloc_pmd_clone = paravirt_nop, | |
1332 | .release_pmd = xen_release_pte_init, | |
f4f97b3e JF |
1333 | |
1334 | #ifdef CONFIG_HIGHPTE | |
1335 | .kmap_atomic_pte = xen_kmap_atomic_pte, | |
1336 | #endif | |
5ead97c8 | 1337 | |
22911b3f JF |
1338 | #ifdef CONFIG_X86_64 |
1339 | .set_pte = xen_set_pte, | |
1340 | #else | |
851fa3c4 | 1341 | .set_pte = xen_set_pte_init, |
22911b3f | 1342 | #endif |
3b827c1b | 1343 | .set_pte_at = xen_set_pte_at, |
e2426cf8 | 1344 | .set_pmd = xen_set_pmd_hyper, |
3b827c1b | 1345 | |
08b882c6 JF |
1346 | .ptep_modify_prot_start = __ptep_modify_prot_start, |
1347 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
1348 | ||
3b827c1b | 1349 | .pte_val = xen_pte_val, |
b56afe1d | 1350 | .pte_flags = native_pte_flags, |
3b827c1b JF |
1351 | .pgd_val = xen_pgd_val, |
1352 | ||
1353 | .make_pte = xen_make_pte, | |
1354 | .make_pgd = xen_make_pgd, | |
1355 | ||
f6e58732 | 1356 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
1357 | .set_pte_atomic = xen_set_pte_atomic, |
1358 | .set_pte_present = xen_set_pte_at, | |
3b827c1b JF |
1359 | .pte_clear = xen_pte_clear, |
1360 | .pmd_clear = xen_pmd_clear, | |
f6e58732 JF |
1361 | #endif /* CONFIG_X86_PAE */ |
1362 | .set_pud = xen_set_pud_hyper, | |
3b827c1b JF |
1363 | |
1364 | .make_pmd = xen_make_pmd, | |
1365 | .pmd_val = xen_pmd_val, | |
3b827c1b | 1366 | |
f6e58732 JF |
1367 | #if PAGETABLE_LEVELS == 4 |
1368 | .pud_val = xen_pud_val, | |
1369 | .make_pud = xen_make_pud, | |
1370 | .set_pgd = xen_set_pgd_hyper, | |
1371 | ||
1372 | .alloc_pud = xen_alloc_pte_init, | |
1373 | .release_pud = xen_release_pte_init, | |
1374 | #endif /* PAGETABLE_LEVELS == 4 */ | |
1375 | ||
3b827c1b JF |
1376 | .activate_mm = xen_activate_mm, |
1377 | .dup_mmap = xen_dup_mmap, | |
1378 | .exit_mmap = xen_exit_mmap, | |
1379 | ||
8965c1c0 JF |
1380 | .lazy_mode = { |
1381 | .enter = paravirt_enter_lazy_mmu, | |
1382 | .leave = xen_leave_lazy, | |
1383 | }, | |
aeaaa59c JF |
1384 | |
1385 | .set_fixmap = xen_set_fixmap, | |
5ead97c8 JF |
1386 | }; |
1387 | ||
fefa629a JF |
1388 | static void xen_reboot(int reason) |
1389 | { | |
349c709f JF |
1390 | struct sched_shutdown r = { .reason = reason }; |
1391 | ||
fefa629a JF |
1392 | #ifdef CONFIG_SMP |
1393 | smp_send_stop(); | |
1394 | #endif | |
1395 | ||
349c709f | 1396 | if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r)) |
fefa629a JF |
1397 | BUG(); |
1398 | } | |
1399 | ||
1400 | static void xen_restart(char *msg) | |
1401 | { | |
1402 | xen_reboot(SHUTDOWN_reboot); | |
1403 | } | |
1404 | ||
1405 | static void xen_emergency_restart(void) | |
1406 | { | |
1407 | xen_reboot(SHUTDOWN_reboot); | |
1408 | } | |
1409 | ||
1410 | static void xen_machine_halt(void) | |
1411 | { | |
1412 | xen_reboot(SHUTDOWN_poweroff); | |
1413 | } | |
1414 | ||
1415 | static void xen_crash_shutdown(struct pt_regs *regs) | |
1416 | { | |
1417 | xen_reboot(SHUTDOWN_crash); | |
1418 | } | |
1419 | ||
1420 | static const struct machine_ops __initdata xen_machine_ops = { | |
1421 | .restart = xen_restart, | |
1422 | .halt = xen_machine_halt, | |
1423 | .power_off = xen_machine_halt, | |
1424 | .shutdown = xen_machine_halt, | |
1425 | .crash_shutdown = xen_crash_shutdown, | |
1426 | .emergency_restart = xen_emergency_restart, | |
1427 | }; | |
1428 | ||
6487673b | 1429 | |
fb1d8404 JF |
1430 | static void __init xen_reserve_top(void) |
1431 | { | |
f5d36de0 | 1432 | #ifdef CONFIG_X86_32 |
fb1d8404 JF |
1433 | unsigned long top = HYPERVISOR_VIRT_START; |
1434 | struct xen_platform_parameters pp; | |
1435 | ||
1436 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1437 | top = pp.virt_start; | |
1438 | ||
1439 | reserve_top_address(-top + 2 * PAGE_SIZE); | |
f5d36de0 | 1440 | #endif /* CONFIG_X86_32 */ |
fb1d8404 JF |
1441 | } |
1442 | ||
084a2a4e JF |
1443 | /* |
1444 | * Like __va(), but returns address in the kernel mapping (which is | |
1445 | * all we have until the physical memory mapping has been set up. | |
1446 | */ | |
1447 | static void *__ka(phys_addr_t paddr) | |
1448 | { | |
39dbc5bd | 1449 | #ifdef CONFIG_X86_64 |
084a2a4e | 1450 | return (void *)(paddr + __START_KERNEL_map); |
39dbc5bd JF |
1451 | #else |
1452 | return __va(paddr); | |
1453 | #endif | |
fb1d8404 JF |
1454 | } |
1455 | ||
084a2a4e JF |
1456 | /* Convert a machine address to physical address */ |
1457 | static unsigned long m2p(phys_addr_t maddr) | |
1458 | { | |
1459 | phys_addr_t paddr; | |
1460 | ||
59438c9f | 1461 | maddr &= PTE_PFN_MASK; |
084a2a4e JF |
1462 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; |
1463 | ||
1464 | return paddr; | |
fb1d8404 JF |
1465 | } |
1466 | ||
084a2a4e JF |
1467 | /* Convert a machine address to kernel virtual */ |
1468 | static void *m2v(phys_addr_t maddr) | |
1469 | { | |
1470 | return __ka(m2p(maddr)); | |
1471 | } | |
1472 | ||
39dbc5bd | 1473 | #ifdef CONFIG_X86_64 |
084a2a4e JF |
1474 | static void walk(pgd_t *pgd, unsigned long addr) |
1475 | { | |
1476 | unsigned l4idx = pgd_index(addr); | |
1477 | unsigned l3idx = pud_index(addr); | |
1478 | unsigned l2idx = pmd_index(addr); | |
1479 | unsigned l1idx = pte_index(addr); | |
1480 | pgd_t l4; | |
1481 | pud_t l3; | |
1482 | pmd_t l2; | |
1483 | pte_t l1; | |
1484 | ||
1485 | xen_raw_printk("walk %p, %lx -> %d %d %d %d\n", | |
1486 | pgd, addr, l4idx, l3idx, l2idx, l1idx); | |
1487 | ||
1488 | l4 = pgd[l4idx]; | |
1489 | xen_raw_printk(" l4: %016lx\n", l4.pgd); | |
1490 | xen_raw_printk(" %016lx\n", pgd_val(l4)); | |
1491 | ||
1492 | l3 = ((pud_t *)(m2v(l4.pgd)))[l3idx]; | |
1493 | xen_raw_printk(" l3: %016lx\n", l3.pud); | |
1494 | xen_raw_printk(" %016lx\n", pud_val(l3)); | |
1495 | ||
1496 | l2 = ((pmd_t *)(m2v(l3.pud)))[l2idx]; | |
1497 | xen_raw_printk(" l2: %016lx\n", l2.pmd); | |
1498 | xen_raw_printk(" %016lx\n", pmd_val(l2)); | |
1499 | ||
1500 | l1 = ((pte_t *)(m2v(l2.pmd)))[l1idx]; | |
1501 | xen_raw_printk(" l1: %016lx\n", l1.pte); | |
1502 | xen_raw_printk(" %016lx\n", pte_val(l1)); | |
1503 | } | |
39dbc5bd | 1504 | #endif |
084a2a4e JF |
1505 | |
1506 | static void set_page_prot(void *addr, pgprot_t prot) | |
1507 | { | |
1508 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1509 | pte_t pte = pfn_pte(pfn, prot); | |
1510 | ||
39dbc5bd | 1511 | xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n", |
084a2a4e JF |
1512 | addr, pfn, get_phys_to_machine(pfn), |
1513 | pgprot_val(prot), pte.pte); | |
1514 | ||
1515 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) | |
1516 | BUG(); | |
1517 | } | |
1518 | ||
39dbc5bd | 1519 | static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) |
d114e198 JF |
1520 | { |
1521 | unsigned pmdidx, pteidx; | |
1522 | unsigned ident_pte; | |
1523 | unsigned long pfn; | |
1524 | ||
1525 | ident_pte = 0; | |
1526 | pfn = 0; | |
1527 | for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1528 | pte_t *pte_page; | |
1529 | ||
d114e198 | 1530 | /* Reuse or allocate a page of ptes */ |
39dbc5bd JF |
1531 | if (pmd_present(pmd[pmdidx])) |
1532 | pte_page = m2v(pmd[pmdidx].pmd); | |
d114e198 JF |
1533 | else { |
1534 | /* Check for free pte pages */ | |
1535 | if (ident_pte == ARRAY_SIZE(level1_ident_pgt)) | |
1536 | break; | |
1537 | ||
1538 | pte_page = &level1_ident_pgt[ident_pte]; | |
1539 | ident_pte += PTRS_PER_PTE; | |
1540 | ||
39dbc5bd | 1541 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); |
d114e198 JF |
1542 | } |
1543 | ||
1544 | /* Install mappings */ | |
1545 | for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1546 | pte_t pte; | |
1547 | ||
1548 | if (pfn > max_pfn_mapped) | |
1549 | max_pfn_mapped = pfn; | |
1550 | ||
1551 | if (!pte_none(pte_page[pteidx])) | |
1552 | continue; | |
1553 | ||
1554 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1555 | pte_page[pteidx] = pte; | |
1556 | } | |
1557 | } | |
1558 | ||
1559 | for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1560 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
39dbc5bd JF |
1561 | |
1562 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1563 | } | |
1564 | ||
1565 | #ifdef CONFIG_X86_64 | |
1566 | static void convert_pfn_mfn(void *v) | |
1567 | { | |
1568 | pte_t *pte = v; | |
1569 | int i; | |
1570 | ||
1571 | /* All levels are converted the same way, so just treat them | |
1572 | as ptes. */ | |
1573 | for(i = 0; i < PTRS_PER_PTE; i++) | |
1574 | pte[i] = xen_make_pte(pte[i].pte); | |
d114e198 JF |
1575 | } |
1576 | ||
084a2a4e JF |
1577 | /* |
1578 | * Set up the inital kernel pagetable. | |
1579 | * | |
1580 | * We can construct this by grafting the Xen provided pagetable into | |
1581 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1582 | * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This | |
1583 | * means that only the kernel has a physical mapping to start with - | |
1584 | * but that's enough to get __va working. We need to fill in the rest | |
1585 | * of the physical mapping once some sort of allocator has been set | |
1586 | * up. | |
1587 | */ | |
d114e198 | 1588 | static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) |
084a2a4e JF |
1589 | { |
1590 | pud_t *l3; | |
1591 | pmd_t *l2; | |
1592 | ||
1593 | /* Zap identity mapping */ | |
1594 | init_level4_pgt[0] = __pgd(0); | |
1595 | ||
1596 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1597 | convert_pfn_mfn(init_level4_pgt); | |
1598 | convert_pfn_mfn(level3_ident_pgt); | |
1599 | convert_pfn_mfn(level3_kernel_pgt); | |
1600 | ||
1601 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1602 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1603 | ||
1604 | memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1605 | memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1606 | ||
1607 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd); | |
1608 | l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud); | |
1609 | memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1610 | ||
d114e198 | 1611 | /* Set up identity map */ |
39dbc5bd | 1612 | xen_map_identity_early(level2_ident_pgt, max_pfn); |
d114e198 | 1613 | |
084a2a4e JF |
1614 | /* Make pagetable pieces RO */ |
1615 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1616 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1617 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
bf18bf94 | 1618 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); |
084a2a4e JF |
1619 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); |
1620 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1621 | ||
1622 | /* Pin down new L4 */ | |
39dbc5bd JF |
1623 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, |
1624 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
084a2a4e JF |
1625 | |
1626 | /* Unpin Xen-provided one */ | |
1627 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1628 | ||
1629 | /* Switch over */ | |
1630 | pgd = init_level4_pgt; | |
d6182fbf JF |
1631 | |
1632 | /* | |
1633 | * At this stage there can be no user pgd, and no page | |
1634 | * structure to attach it to, so make sure we just set kernel | |
1635 | * pgd. | |
1636 | */ | |
1637 | xen_mc_batch(); | |
1638 | __xen_write_cr3(true, __pa(pgd)); | |
1639 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
084a2a4e | 1640 | |
d114e198 JF |
1641 | reserve_early(__pa(xen_start_info->pt_base), |
1642 | __pa(xen_start_info->pt_base + | |
1643 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1644 | "XEN PAGETABLES"); | |
084a2a4e JF |
1645 | |
1646 | return pgd; | |
1647 | } | |
39dbc5bd JF |
1648 | #else /* !CONFIG_X86_64 */ |
1649 | static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss; | |
1650 | ||
d114e198 | 1651 | static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) |
084a2a4e | 1652 | { |
39dbc5bd JF |
1653 | pmd_t *kernel_pmd; |
1654 | ||
084a2a4e JF |
1655 | init_pg_tables_start = __pa(pgd); |
1656 | init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE; | |
1657 | max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024); | |
1658 | ||
39dbc5bd JF |
1659 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); |
1660 | memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD); | |
d114e198 | 1661 | |
39dbc5bd JF |
1662 | xen_map_identity_early(level2_kernel_pgt, max_pfn); |
1663 | ||
1664 | memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD); | |
1665 | set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY], | |
1666 | __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT)); | |
1667 | ||
1668 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1669 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1670 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
1671 | ||
1672 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1673 | ||
1674 | xen_write_cr3(__pa(swapper_pg_dir)); | |
1675 | ||
1676 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir))); | |
1677 | ||
1678 | return swapper_pg_dir; | |
fb1d8404 | 1679 | } |
084a2a4e | 1680 | #endif /* CONFIG_X86_64 */ |
fb1d8404 | 1681 | |
5ead97c8 JF |
1682 | /* First C function to be called on Xen boot */ |
1683 | asmlinkage void __init xen_start_kernel(void) | |
1684 | { | |
1685 | pgd_t *pgd; | |
1686 | ||
1687 | if (!xen_start_info) | |
1688 | return; | |
1689 | ||
7999f4b4 | 1690 | BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0); |
5ead97c8 | 1691 | |
e57778a1 JF |
1692 | xen_setup_features(); |
1693 | ||
5ead97c8 | 1694 | /* Install Xen paravirt ops */ |
93b1eab3 JF |
1695 | pv_info = xen_info; |
1696 | pv_init_ops = xen_init_ops; | |
1697 | pv_time_ops = xen_time_ops; | |
1698 | pv_cpu_ops = xen_cpu_ops; | |
1699 | pv_irq_ops = xen_irq_ops; | |
1700 | pv_apic_ops = xen_apic_ops; | |
1701 | pv_mmu_ops = xen_mmu_ops; | |
93b1eab3 | 1702 | |
e57778a1 JF |
1703 | if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { |
1704 | pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; | |
1705 | pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; | |
1706 | } | |
1707 | ||
fefa629a JF |
1708 | machine_ops = xen_machine_ops; |
1709 | ||
f5d36de0 JF |
1710 | #ifdef CONFIG_X86_64 |
1711 | /* Disable until direct per-cpu data access. */ | |
1712 | have_vcpu_info_placement = 0; | |
5b09b287 | 1713 | x86_64_init_pda(); |
f87e4cac | 1714 | #endif |
5ead97c8 | 1715 | |
a9e7062d | 1716 | xen_smp_init(); |
5ead97c8 JF |
1717 | |
1718 | /* Get mfn list */ | |
1719 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
d451bb7a | 1720 | xen_build_dynamic_phys_to_machine(); |
5ead97c8 JF |
1721 | |
1722 | pgd = (pgd_t *)xen_start_info->pt_base; | |
1723 | ||
084a2a4e JF |
1724 | /* Prevent unwanted bits from being set in PTEs. */ |
1725 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
1726 | if (!is_initial_xendomain()) | |
1727 | __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD); | |
60223a32 | 1728 | |
60223a32 | 1729 | /* Don't do the full vcpu_info placement stuff until we have a |
2e8fe719 | 1730 | possible map and a non-dummy shared_info. */ |
60223a32 | 1731 | per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; |
5ead97c8 | 1732 | |
084a2a4e | 1733 | xen_raw_console_write("mapping kernel into physical memory\n"); |
d114e198 | 1734 | pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages); |
5ead97c8 | 1735 | |
084a2a4e | 1736 | init_mm.pgd = pgd; |
5ead97c8 JF |
1737 | |
1738 | /* keep using Xen gdt for now; no urgent need to change it */ | |
1739 | ||
93b1eab3 | 1740 | pv_info.kernel_rpl = 1; |
5ead97c8 | 1741 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) |
93b1eab3 | 1742 | pv_info.kernel_rpl = 0; |
5ead97c8 JF |
1743 | |
1744 | /* set the limit of our address space */ | |
fb1d8404 | 1745 | xen_reserve_top(); |
5ead97c8 | 1746 | |
7d087b68 | 1747 | #ifdef CONFIG_X86_32 |
5ead97c8 JF |
1748 | /* set up basic CPUID stuff */ |
1749 | cpu_detect(&new_cpu_data); | |
1750 | new_cpu_data.hard_math = 1; | |
1751 | new_cpu_data.x86_capability[0] = cpuid_edx(1); | |
7d087b68 | 1752 | #endif |
5ead97c8 JF |
1753 | |
1754 | /* Poke various useful things into boot_params */ | |
30c82645 PA |
1755 | boot_params.hdr.type_of_loader = (9 << 4) | 0; |
1756 | boot_params.hdr.ramdisk_image = xen_start_info->mod_start | |
1757 | ? __pa(xen_start_info->mod_start) : 0; | |
1758 | boot_params.hdr.ramdisk_size = xen_start_info->mod_len; | |
b7c3c5c1 | 1759 | boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); |
5ead97c8 | 1760 | |
9e124fe1 | 1761 | if (!is_initial_xendomain()) { |
83abc70a | 1762 | add_preferred_console("xenboot", 0, NULL); |
9e124fe1 | 1763 | add_preferred_console("tty", 0, NULL); |
b8c2d3df | 1764 | add_preferred_console("hvc", 0, NULL); |
9e124fe1 | 1765 | } |
b8c2d3df | 1766 | |
084a2a4e JF |
1767 | xen_raw_console_write("about to get started...\n"); |
1768 | ||
1769 | #if 0 | |
1770 | xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n", | |
1771 | &boot_params, __pa_symbol(&boot_params), | |
1772 | __va(__pa_symbol(&boot_params))); | |
1773 | ||
1774 | walk(pgd, &boot_params); | |
1775 | walk(pgd, __va(__pa(&boot_params))); | |
1776 | #endif | |
b8c2d3df | 1777 | |
5ead97c8 | 1778 | /* Start the world */ |
f5d36de0 | 1779 | #ifdef CONFIG_X86_32 |
f0d43100 | 1780 | i386_start_kernel(); |
f5d36de0 | 1781 | #else |
084a2a4e | 1782 | x86_64_start_reservations((char *)__pa_symbol(&boot_params)); |
f5d36de0 | 1783 | #endif |
5ead97c8 | 1784 | } |