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> | |
36 | ||
37 | #include <asm/paravirt.h> | |
38 | #include <asm/page.h> | |
39 | #include <asm/xen/hypercall.h> | |
40 | #include <asm/xen/hypervisor.h> | |
41 | #include <asm/fixmap.h> | |
42 | #include <asm/processor.h> | |
43 | #include <asm/setup.h> | |
44 | #include <asm/desc.h> | |
45 | #include <asm/pgtable.h> | |
f87e4cac | 46 | #include <asm/tlbflush.h> |
fefa629a | 47 | #include <asm/reboot.h> |
5ead97c8 JF |
48 | |
49 | #include "xen-ops.h" | |
3b827c1b | 50 | #include "mmu.h" |
5ead97c8 JF |
51 | #include "multicalls.h" |
52 | ||
53 | EXPORT_SYMBOL_GPL(hypercall_page); | |
54 | ||
5ead97c8 JF |
55 | DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu); |
56 | DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info); | |
9f79991d JF |
57 | |
58 | /* | |
59 | * Note about cr3 (pagetable base) values: | |
60 | * | |
61 | * xen_cr3 contains the current logical cr3 value; it contains the | |
62 | * last set cr3. This may not be the current effective cr3, because | |
63 | * its update may be being lazily deferred. However, a vcpu looking | |
64 | * at its own cr3 can use this value knowing that it everything will | |
65 | * be self-consistent. | |
66 | * | |
67 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
68 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
69 | * out of date, but it will never be set early). If one vcpu is | |
70 | * looking at another vcpu's cr3 value, it should use this variable. | |
71 | */ | |
72 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
73 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
5ead97c8 JF |
74 | |
75 | struct start_info *xen_start_info; | |
76 | EXPORT_SYMBOL_GPL(xen_start_info); | |
77 | ||
60223a32 JF |
78 | static /* __initdata */ struct shared_info dummy_shared_info; |
79 | ||
80 | /* | |
81 | * Point at some empty memory to start with. We map the real shared_info | |
82 | * page as soon as fixmap is up and running. | |
83 | */ | |
84 | struct shared_info *HYPERVISOR_shared_info = (void *)&dummy_shared_info; | |
85 | ||
86 | /* | |
87 | * Flag to determine whether vcpu info placement is available on all | |
88 | * VCPUs. We assume it is to start with, and then set it to zero on | |
89 | * the first failure. This is because it can succeed on some VCPUs | |
90 | * and not others, since it can involve hypervisor memory allocation, | |
91 | * or because the guest failed to guarantee all the appropriate | |
92 | * constraints on all VCPUs (ie buffer can't cross a page boundary). | |
93 | * | |
94 | * Note that any particular CPU may be using a placed vcpu structure, | |
95 | * but we can only optimise if the all are. | |
96 | * | |
97 | * 0: not available, 1: available | |
98 | */ | |
04c44a08 | 99 | static int have_vcpu_info_placement = 1; |
60223a32 JF |
100 | |
101 | static void __init xen_vcpu_setup(int cpu) | |
5ead97c8 | 102 | { |
60223a32 JF |
103 | struct vcpu_register_vcpu_info info; |
104 | int err; | |
105 | struct vcpu_info *vcpup; | |
106 | ||
2e8fe719 | 107 | BUG_ON(HYPERVISOR_shared_info == &dummy_shared_info); |
5ead97c8 | 108 | per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu]; |
60223a32 JF |
109 | |
110 | if (!have_vcpu_info_placement) | |
111 | return; /* already tested, not available */ | |
112 | ||
113 | vcpup = &per_cpu(xen_vcpu_info, cpu); | |
114 | ||
115 | info.mfn = virt_to_mfn(vcpup); | |
116 | info.offset = offset_in_page(vcpup); | |
117 | ||
e3d26976 | 118 | printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n", |
60223a32 JF |
119 | cpu, vcpup, info.mfn, info.offset); |
120 | ||
121 | /* Check to see if the hypervisor will put the vcpu_info | |
122 | structure where we want it, which allows direct access via | |
123 | a percpu-variable. */ | |
124 | err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); | |
125 | ||
126 | if (err) { | |
127 | printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err); | |
128 | have_vcpu_info_placement = 0; | |
129 | } else { | |
130 | /* This cpu is using the registered vcpu info, even if | |
131 | later ones fail to. */ | |
132 | per_cpu(xen_vcpu, cpu) = vcpup; | |
6487673b | 133 | |
60223a32 JF |
134 | printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n", |
135 | cpu, vcpup); | |
136 | } | |
5ead97c8 JF |
137 | } |
138 | ||
139 | static void __init xen_banner(void) | |
140 | { | |
141 | printk(KERN_INFO "Booting paravirtualized kernel on %s\n", | |
93b1eab3 | 142 | pv_info.name); |
5ead97c8 JF |
143 | printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic); |
144 | } | |
145 | ||
65ea5b03 PA |
146 | static void xen_cpuid(unsigned int *ax, unsigned int *bx, |
147 | unsigned int *cx, unsigned int *dx) | |
5ead97c8 JF |
148 | { |
149 | unsigned maskedx = ~0; | |
150 | ||
151 | /* | |
152 | * Mask out inconvenient features, to try and disable as many | |
153 | * unsupported kernel subsystems as possible. | |
154 | */ | |
65ea5b03 | 155 | if (*ax == 1) |
5ead97c8 JF |
156 | maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */ |
157 | (1 << X86_FEATURE_ACPI) | /* disable ACPI */ | |
d40e7059 | 158 | (1 << X86_FEATURE_SEP) | /* disable SEP */ |
5ead97c8 JF |
159 | (1 << X86_FEATURE_ACC)); /* thermal monitoring */ |
160 | ||
161 | asm(XEN_EMULATE_PREFIX "cpuid" | |
65ea5b03 PA |
162 | : "=a" (*ax), |
163 | "=b" (*bx), | |
164 | "=c" (*cx), | |
165 | "=d" (*dx) | |
166 | : "0" (*ax), "2" (*cx)); | |
167 | *dx &= maskedx; | |
5ead97c8 JF |
168 | } |
169 | ||
170 | static void xen_set_debugreg(int reg, unsigned long val) | |
171 | { | |
172 | HYPERVISOR_set_debugreg(reg, val); | |
173 | } | |
174 | ||
175 | static unsigned long xen_get_debugreg(int reg) | |
176 | { | |
177 | return HYPERVISOR_get_debugreg(reg); | |
178 | } | |
179 | ||
180 | static unsigned long xen_save_fl(void) | |
181 | { | |
182 | struct vcpu_info *vcpu; | |
183 | unsigned long flags; | |
184 | ||
5ead97c8 | 185 | vcpu = x86_read_percpu(xen_vcpu); |
f120f13e | 186 | |
5ead97c8 JF |
187 | /* flag has opposite sense of mask */ |
188 | flags = !vcpu->evtchn_upcall_mask; | |
5ead97c8 JF |
189 | |
190 | /* convert to IF type flag | |
191 | -0 -> 0x00000000 | |
192 | -1 -> 0xffffffff | |
193 | */ | |
194 | return (-flags) & X86_EFLAGS_IF; | |
195 | } | |
196 | ||
197 | static void xen_restore_fl(unsigned long flags) | |
198 | { | |
199 | struct vcpu_info *vcpu; | |
200 | ||
5ead97c8 JF |
201 | /* convert from IF type flag */ |
202 | flags = !(flags & X86_EFLAGS_IF); | |
f120f13e JF |
203 | |
204 | /* There's a one instruction preempt window here. We need to | |
205 | make sure we're don't switch CPUs between getting the vcpu | |
206 | pointer and updating the mask. */ | |
207 | preempt_disable(); | |
5ead97c8 JF |
208 | vcpu = x86_read_percpu(xen_vcpu); |
209 | vcpu->evtchn_upcall_mask = flags; | |
f120f13e | 210 | preempt_enable_no_resched(); |
5ead97c8 | 211 | |
f120f13e JF |
212 | /* Doesn't matter if we get preempted here, because any |
213 | pending event will get dealt with anyway. */ | |
5ead97c8 | 214 | |
f120f13e JF |
215 | if (flags == 0) { |
216 | preempt_check_resched(); | |
217 | barrier(); /* unmask then check (avoid races) */ | |
5ead97c8 JF |
218 | if (unlikely(vcpu->evtchn_upcall_pending)) |
219 | force_evtchn_callback(); | |
f120f13e | 220 | } |
5ead97c8 JF |
221 | } |
222 | ||
223 | static void xen_irq_disable(void) | |
224 | { | |
f120f13e JF |
225 | /* There's a one instruction preempt window here. We need to |
226 | make sure we're don't switch CPUs between getting the vcpu | |
227 | pointer and updating the mask. */ | |
5ead97c8 | 228 | preempt_disable(); |
f120f13e | 229 | x86_read_percpu(xen_vcpu)->evtchn_upcall_mask = 1; |
5ead97c8 JF |
230 | preempt_enable_no_resched(); |
231 | } | |
232 | ||
233 | static void xen_irq_enable(void) | |
234 | { | |
235 | struct vcpu_info *vcpu; | |
236 | ||
f120f13e JF |
237 | /* There's a one instruction preempt window here. We need to |
238 | make sure we're don't switch CPUs between getting the vcpu | |
239 | pointer and updating the mask. */ | |
5ead97c8 JF |
240 | preempt_disable(); |
241 | vcpu = x86_read_percpu(xen_vcpu); | |
242 | vcpu->evtchn_upcall_mask = 0; | |
f120f13e | 243 | preempt_enable_no_resched(); |
5ead97c8 | 244 | |
f120f13e JF |
245 | /* Doesn't matter if we get preempted here, because any |
246 | pending event will get dealt with anyway. */ | |
5ead97c8 | 247 | |
f120f13e | 248 | barrier(); /* unmask then check (avoid races) */ |
5ead97c8 JF |
249 | if (unlikely(vcpu->evtchn_upcall_pending)) |
250 | force_evtchn_callback(); | |
5ead97c8 JF |
251 | } |
252 | ||
253 | static void xen_safe_halt(void) | |
254 | { | |
255 | /* Blocking includes an implicit local_irq_enable(). */ | |
256 | if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0) | |
257 | BUG(); | |
258 | } | |
259 | ||
260 | static void xen_halt(void) | |
261 | { | |
262 | if (irqs_disabled()) | |
263 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
264 | else | |
265 | xen_safe_halt(); | |
266 | } | |
267 | ||
8965c1c0 | 268 | static void xen_leave_lazy(void) |
5ead97c8 | 269 | { |
8965c1c0 | 270 | paravirt_leave_lazy(paravirt_get_lazy_mode()); |
5ead97c8 | 271 | xen_mc_flush(); |
5ead97c8 JF |
272 | } |
273 | ||
274 | static unsigned long xen_store_tr(void) | |
275 | { | |
276 | return 0; | |
277 | } | |
278 | ||
279 | static void xen_set_ldt(const void *addr, unsigned entries) | |
280 | { | |
5ead97c8 JF |
281 | struct mmuext_op *op; |
282 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
283 | ||
284 | op = mcs.args; | |
285 | op->cmd = MMUEXT_SET_LDT; | |
4dbf7af6 | 286 | op->arg1.linear_addr = (unsigned long)addr; |
5ead97c8 JF |
287 | op->arg2.nr_ents = entries; |
288 | ||
289 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
290 | ||
291 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
292 | } | |
293 | ||
6b68f01b | 294 | static void xen_load_gdt(const struct desc_ptr *dtr) |
5ead97c8 JF |
295 | { |
296 | unsigned long *frames; | |
297 | unsigned long va = dtr->address; | |
298 | unsigned int size = dtr->size + 1; | |
299 | unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE; | |
300 | int f; | |
301 | struct multicall_space mcs; | |
302 | ||
303 | /* A GDT can be up to 64k in size, which corresponds to 8192 | |
304 | 8-byte entries, or 16 4k pages.. */ | |
305 | ||
306 | BUG_ON(size > 65536); | |
307 | BUG_ON(va & ~PAGE_MASK); | |
308 | ||
309 | mcs = xen_mc_entry(sizeof(*frames) * pages); | |
310 | frames = mcs.args; | |
311 | ||
312 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
313 | frames[f] = virt_to_mfn(va); | |
314 | make_lowmem_page_readonly((void *)va); | |
315 | } | |
316 | ||
317 | MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct)); | |
318 | ||
319 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
320 | } | |
321 | ||
322 | static void load_TLS_descriptor(struct thread_struct *t, | |
323 | unsigned int cpu, unsigned int i) | |
324 | { | |
325 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
326 | xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); | |
327 | struct multicall_space mc = __xen_mc_entry(0); | |
328 | ||
329 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
330 | } | |
331 | ||
332 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
333 | { | |
334 | xen_mc_batch(); | |
335 | ||
336 | load_TLS_descriptor(t, cpu, 0); | |
337 | load_TLS_descriptor(t, cpu, 1); | |
338 | load_TLS_descriptor(t, cpu, 2); | |
339 | ||
340 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
8b84ad94 JF |
341 | |
342 | /* | |
343 | * XXX sleazy hack: If we're being called in a lazy-cpu zone, | |
344 | * it means we're in a context switch, and %gs has just been | |
345 | * saved. This means we can zero it out to prevent faults on | |
346 | * exit from the hypervisor if the next process has no %gs. | |
347 | * Either way, it has been saved, and the new value will get | |
348 | * loaded properly. This will go away as soon as Xen has been | |
349 | * modified to not save/restore %gs for normal hypercalls. | |
350 | */ | |
8965c1c0 | 351 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) |
8b84ad94 | 352 | loadsegment(gs, 0); |
5ead97c8 JF |
353 | } |
354 | ||
355 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
75b8bb3e | 356 | const void *ptr) |
5ead97c8 JF |
357 | { |
358 | unsigned long lp = (unsigned long)&dt[entrynum]; | |
359 | xmaddr_t mach_lp = virt_to_machine(lp); | |
75b8bb3e | 360 | u64 entry = *(u64 *)ptr; |
5ead97c8 | 361 | |
f120f13e JF |
362 | preempt_disable(); |
363 | ||
5ead97c8 JF |
364 | xen_mc_flush(); |
365 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
366 | BUG(); | |
f120f13e JF |
367 | |
368 | preempt_enable(); | |
5ead97c8 JF |
369 | } |
370 | ||
371 | static int cvt_gate_to_trap(int vector, u32 low, u32 high, | |
372 | struct trap_info *info) | |
373 | { | |
374 | u8 type, dpl; | |
375 | ||
376 | type = (high >> 8) & 0x1f; | |
377 | dpl = (high >> 13) & 3; | |
378 | ||
379 | if (type != 0xf && type != 0xe) | |
380 | return 0; | |
381 | ||
382 | info->vector = vector; | |
383 | info->address = (high & 0xffff0000) | (low & 0x0000ffff); | |
384 | info->cs = low >> 16; | |
385 | info->flags = dpl; | |
386 | /* interrupt gates clear IF */ | |
387 | if (type == 0xe) | |
388 | info->flags |= 4; | |
389 | ||
390 | return 1; | |
391 | } | |
392 | ||
393 | /* Locations of each CPU's IDT */ | |
6b68f01b | 394 | static DEFINE_PER_CPU(struct desc_ptr, idt_desc); |
5ead97c8 JF |
395 | |
396 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
397 | also update Xen. */ | |
8d947344 | 398 | static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) |
5ead97c8 | 399 | { |
5ead97c8 | 400 | unsigned long p = (unsigned long)&dt[entrynum]; |
f120f13e JF |
401 | unsigned long start, end; |
402 | ||
403 | preempt_disable(); | |
404 | ||
405 | start = __get_cpu_var(idt_desc).address; | |
406 | end = start + __get_cpu_var(idt_desc).size + 1; | |
5ead97c8 JF |
407 | |
408 | xen_mc_flush(); | |
409 | ||
8d947344 | 410 | native_write_idt_entry(dt, entrynum, g); |
5ead97c8 JF |
411 | |
412 | if (p >= start && (p + 8) <= end) { | |
413 | struct trap_info info[2]; | |
8d947344 | 414 | u32 *desc = (u32 *)g; |
5ead97c8 JF |
415 | |
416 | info[1].address = 0; | |
417 | ||
8d947344 | 418 | if (cvt_gate_to_trap(entrynum, desc[0], desc[1], &info[0])) |
5ead97c8 JF |
419 | if (HYPERVISOR_set_trap_table(info)) |
420 | BUG(); | |
421 | } | |
f120f13e JF |
422 | |
423 | preempt_enable(); | |
5ead97c8 JF |
424 | } |
425 | ||
6b68f01b | 426 | static void xen_convert_trap_info(const struct desc_ptr *desc, |
f87e4cac | 427 | struct trap_info *traps) |
5ead97c8 | 428 | { |
5ead97c8 JF |
429 | unsigned in, out, count; |
430 | ||
5ead97c8 JF |
431 | count = (desc->size+1) / 8; |
432 | BUG_ON(count > 256); | |
433 | ||
5ead97c8 JF |
434 | for (in = out = 0; in < count; in++) { |
435 | const u32 *entry = (u32 *)(desc->address + in * 8); | |
436 | ||
437 | if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out])) | |
438 | out++; | |
439 | } | |
440 | traps[out].address = 0; | |
f87e4cac JF |
441 | } |
442 | ||
443 | void xen_copy_trap_info(struct trap_info *traps) | |
444 | { | |
6b68f01b | 445 | const struct desc_ptr *desc = &__get_cpu_var(idt_desc); |
f87e4cac JF |
446 | |
447 | xen_convert_trap_info(desc, traps); | |
f87e4cac JF |
448 | } |
449 | ||
450 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
451 | hold a spinlock to protect the static traps[] array (static because | |
452 | it avoids allocation, and saves stack space). */ | |
6b68f01b | 453 | static void xen_load_idt(const struct desc_ptr *desc) |
f87e4cac JF |
454 | { |
455 | static DEFINE_SPINLOCK(lock); | |
456 | static struct trap_info traps[257]; | |
f87e4cac JF |
457 | |
458 | spin_lock(&lock); | |
459 | ||
f120f13e JF |
460 | __get_cpu_var(idt_desc) = *desc; |
461 | ||
f87e4cac | 462 | xen_convert_trap_info(desc, traps); |
5ead97c8 JF |
463 | |
464 | xen_mc_flush(); | |
465 | if (HYPERVISOR_set_trap_table(traps)) | |
466 | BUG(); | |
467 | ||
468 | spin_unlock(&lock); | |
469 | } | |
470 | ||
471 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
472 | they're handled differently. */ | |
473 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
014b15be | 474 | const void *desc, int type) |
5ead97c8 | 475 | { |
f120f13e JF |
476 | preempt_disable(); |
477 | ||
014b15be GOC |
478 | switch (type) { |
479 | case DESC_LDT: | |
480 | case DESC_TSS: | |
5ead97c8 JF |
481 | /* ignore */ |
482 | break; | |
483 | ||
484 | default: { | |
485 | xmaddr_t maddr = virt_to_machine(&dt[entry]); | |
5ead97c8 JF |
486 | |
487 | xen_mc_flush(); | |
014b15be | 488 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) |
5ead97c8 JF |
489 | BUG(); |
490 | } | |
491 | ||
492 | } | |
f120f13e JF |
493 | |
494 | preempt_enable(); | |
5ead97c8 JF |
495 | } |
496 | ||
faca6227 | 497 | static void xen_load_sp0(struct tss_struct *tss, |
f120f13e | 498 | struct thread_struct *thread) |
5ead97c8 JF |
499 | { |
500 | struct multicall_space mcs = xen_mc_entry(0); | |
faca6227 | 501 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); |
5ead97c8 JF |
502 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
503 | } | |
504 | ||
505 | static void xen_set_iopl_mask(unsigned mask) | |
506 | { | |
507 | struct physdev_set_iopl set_iopl; | |
508 | ||
509 | /* Force the change at ring 0. */ | |
510 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
511 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
512 | } | |
513 | ||
514 | static void xen_io_delay(void) | |
515 | { | |
516 | } | |
517 | ||
518 | #ifdef CONFIG_X86_LOCAL_APIC | |
42e0a9aa | 519 | static u32 xen_apic_read(unsigned long reg) |
5ead97c8 JF |
520 | { |
521 | return 0; | |
522 | } | |
f87e4cac | 523 | |
42e0a9aa | 524 | static void xen_apic_write(unsigned long reg, u32 val) |
f87e4cac JF |
525 | { |
526 | /* Warn to see if there's any stray references */ | |
527 | WARN_ON(1); | |
528 | } | |
5ead97c8 JF |
529 | #endif |
530 | ||
531 | static void xen_flush_tlb(void) | |
532 | { | |
d66bf8fc JF |
533 | struct mmuext_op *op; |
534 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
5ead97c8 | 535 | |
d66bf8fc JF |
536 | op = mcs.args; |
537 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
538 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
539 | ||
540 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
5ead97c8 JF |
541 | } |
542 | ||
543 | static void xen_flush_tlb_single(unsigned long addr) | |
544 | { | |
d66bf8fc JF |
545 | struct mmuext_op *op; |
546 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
5ead97c8 | 547 | |
d66bf8fc JF |
548 | op = mcs.args; |
549 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
550 | op->arg1.linear_addr = addr & PAGE_MASK; | |
551 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
552 | ||
553 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
5ead97c8 JF |
554 | } |
555 | ||
f87e4cac JF |
556 | static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm, |
557 | unsigned long va) | |
558 | { | |
d66bf8fc JF |
559 | struct { |
560 | struct mmuext_op op; | |
561 | cpumask_t mask; | |
562 | } *args; | |
f87e4cac | 563 | cpumask_t cpumask = *cpus; |
d66bf8fc | 564 | struct multicall_space mcs; |
f87e4cac JF |
565 | |
566 | /* | |
567 | * A couple of (to be removed) sanity checks: | |
568 | * | |
569 | * - current CPU must not be in mask | |
570 | * - mask must exist :) | |
571 | */ | |
572 | BUG_ON(cpus_empty(cpumask)); | |
573 | BUG_ON(cpu_isset(smp_processor_id(), cpumask)); | |
574 | BUG_ON(!mm); | |
575 | ||
576 | /* If a CPU which we ran on has gone down, OK. */ | |
577 | cpus_and(cpumask, cpumask, cpu_online_map); | |
578 | if (cpus_empty(cpumask)) | |
579 | return; | |
580 | ||
d66bf8fc JF |
581 | mcs = xen_mc_entry(sizeof(*args)); |
582 | args = mcs.args; | |
583 | args->mask = cpumask; | |
584 | args->op.arg2.vcpumask = &args->mask; | |
585 | ||
f87e4cac | 586 | if (va == TLB_FLUSH_ALL) { |
d66bf8fc | 587 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; |
f87e4cac | 588 | } else { |
d66bf8fc JF |
589 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
590 | args->op.arg1.linear_addr = va; | |
f87e4cac JF |
591 | } |
592 | ||
d66bf8fc JF |
593 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); |
594 | ||
595 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f87e4cac JF |
596 | } |
597 | ||
60223a32 JF |
598 | static void xen_write_cr2(unsigned long cr2) |
599 | { | |
600 | x86_read_percpu(xen_vcpu)->arch.cr2 = cr2; | |
601 | } | |
602 | ||
5ead97c8 JF |
603 | static unsigned long xen_read_cr2(void) |
604 | { | |
605 | return x86_read_percpu(xen_vcpu)->arch.cr2; | |
606 | } | |
607 | ||
60223a32 JF |
608 | static unsigned long xen_read_cr2_direct(void) |
609 | { | |
610 | return x86_read_percpu(xen_vcpu_info.arch.cr2); | |
611 | } | |
612 | ||
5ead97c8 JF |
613 | static void xen_write_cr4(unsigned long cr4) |
614 | { | |
389a3c02 JF |
615 | /* Just ignore cr4 changes; Xen doesn't allow us to do |
616 | anything anyway. */ | |
5ead97c8 JF |
617 | } |
618 | ||
5ead97c8 JF |
619 | static unsigned long xen_read_cr3(void) |
620 | { | |
621 | return x86_read_percpu(xen_cr3); | |
622 | } | |
623 | ||
9f79991d JF |
624 | static void set_current_cr3(void *v) |
625 | { | |
626 | x86_write_percpu(xen_current_cr3, (unsigned long)v); | |
627 | } | |
628 | ||
5ead97c8 JF |
629 | static void xen_write_cr3(unsigned long cr3) |
630 | { | |
9f79991d JF |
631 | struct mmuext_op *op; |
632 | struct multicall_space mcs; | |
633 | unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
634 | ||
f120f13e JF |
635 | BUG_ON(preemptible()); |
636 | ||
9f79991d | 637 | mcs = xen_mc_entry(sizeof(*op)); /* disables interrupts */ |
5ead97c8 | 638 | |
9f79991d JF |
639 | /* Update while interrupts are disabled, so its atomic with |
640 | respect to ipis */ | |
5ead97c8 JF |
641 | x86_write_percpu(xen_cr3, cr3); |
642 | ||
9f79991d JF |
643 | op = mcs.args; |
644 | op->cmd = MMUEXT_NEW_BASEPTR; | |
645 | op->arg1.mfn = mfn; | |
5ead97c8 | 646 | |
9f79991d | 647 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); |
5ead97c8 | 648 | |
9f79991d JF |
649 | /* Update xen_update_cr3 once the batch has actually |
650 | been submitted. */ | |
651 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
5ead97c8 | 652 | |
9f79991d | 653 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ |
5ead97c8 JF |
654 | } |
655 | ||
f4f97b3e JF |
656 | /* Early in boot, while setting up the initial pagetable, assume |
657 | everything is pinned. */ | |
9a4029fd | 658 | static __init void xen_alloc_pt_init(struct mm_struct *mm, u32 pfn) |
5ead97c8 | 659 | { |
f4f97b3e | 660 | BUG_ON(mem_map); /* should only be used early */ |
5ead97c8 JF |
661 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
662 | } | |
663 | ||
1c70e9bd JF |
664 | /* Early release_pt assumes that all pts are pinned, since there's |
665 | only init_mm and anything attached to that is pinned. */ | |
666 | static void xen_release_pt_init(u32 pfn) | |
667 | { | |
668 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
669 | } | |
670 | ||
f6433706 | 671 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
74260714 JF |
672 | { |
673 | struct mmuext_op op; | |
f6433706 | 674 | op.cmd = cmd; |
74260714 JF |
675 | op.arg1.mfn = pfn_to_mfn(pfn); |
676 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
677 | BUG(); | |
678 | } | |
679 | ||
f4f97b3e JF |
680 | /* This needs to make sure the new pte page is pinned iff its being |
681 | attached to a pinned pagetable. */ | |
1c70e9bd | 682 | static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level) |
5ead97c8 | 683 | { |
f4f97b3e | 684 | struct page *page = pfn_to_page(pfn); |
5ead97c8 | 685 | |
f4f97b3e JF |
686 | if (PagePinned(virt_to_page(mm->pgd))) { |
687 | SetPagePinned(page); | |
688 | ||
74260714 | 689 | if (!PageHighMem(page)) { |
f4f97b3e | 690 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
f6433706 MM |
691 | if (level == PT_PTE) |
692 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
74260714 | 693 | } else |
f4f97b3e JF |
694 | /* make sure there are no stray mappings of |
695 | this page */ | |
696 | kmap_flush_unused(); | |
697 | } | |
5ead97c8 JF |
698 | } |
699 | ||
1c70e9bd JF |
700 | static void xen_alloc_pt(struct mm_struct *mm, u32 pfn) |
701 | { | |
f6433706 | 702 | xen_alloc_ptpage(mm, pfn, PT_PTE); |
1c70e9bd JF |
703 | } |
704 | ||
705 | static void xen_alloc_pd(struct mm_struct *mm, u32 pfn) | |
706 | { | |
f6433706 | 707 | xen_alloc_ptpage(mm, pfn, PT_PMD); |
1c70e9bd JF |
708 | } |
709 | ||
f4f97b3e | 710 | /* This should never happen until we're OK to use struct page */ |
f6433706 | 711 | static void xen_release_ptpage(u32 pfn, unsigned level) |
5ead97c8 | 712 | { |
f4f97b3e JF |
713 | struct page *page = pfn_to_page(pfn); |
714 | ||
715 | if (PagePinned(page)) { | |
74260714 | 716 | if (!PageHighMem(page)) { |
a684d69d MM |
717 | if (level == PT_PTE) |
718 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
f4f97b3e | 719 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
74260714 | 720 | } |
c946c7de | 721 | ClearPagePinned(page); |
f4f97b3e | 722 | } |
5ead97c8 JF |
723 | } |
724 | ||
f6433706 MM |
725 | static void xen_release_pt(u32 pfn) |
726 | { | |
727 | xen_release_ptpage(pfn, PT_PTE); | |
728 | } | |
729 | ||
730 | static void xen_release_pd(u32 pfn) | |
731 | { | |
732 | xen_release_ptpage(pfn, PT_PMD); | |
733 | } | |
734 | ||
f4f97b3e JF |
735 | #ifdef CONFIG_HIGHPTE |
736 | static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) | |
5ead97c8 | 737 | { |
f4f97b3e JF |
738 | pgprot_t prot = PAGE_KERNEL; |
739 | ||
740 | if (PagePinned(page)) | |
741 | prot = PAGE_KERNEL_RO; | |
742 | ||
743 | if (0 && PageHighMem(page)) | |
744 | printk("mapping highpte %lx type %d prot %s\n", | |
745 | page_to_pfn(page), type, | |
746 | (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); | |
747 | ||
748 | return kmap_atomic_prot(page, type, prot); | |
5ead97c8 | 749 | } |
f4f97b3e | 750 | #endif |
5ead97c8 | 751 | |
9a4029fd JF |
752 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) |
753 | { | |
754 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
755 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
756 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
757 | pte_val_ma(pte)); | |
758 | ||
759 | return pte; | |
760 | } | |
761 | ||
762 | /* Init-time set_pte while constructing initial pagetables, which | |
763 | doesn't allow RO pagetable pages to be remapped RW */ | |
764 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
765 | { | |
766 | pte = mask_rw_pte(ptep, pte); | |
767 | ||
768 | xen_set_pte(ptep, pte); | |
769 | } | |
770 | ||
5ead97c8 JF |
771 | static __init void xen_pagetable_setup_start(pgd_t *base) |
772 | { | |
773 | pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base; | |
774 | ||
9a4029fd | 775 | /* special set_pte for pagetable initialization */ |
93b1eab3 | 776 | pv_mmu_ops.set_pte = xen_set_pte_init; |
9a4029fd | 777 | |
5ead97c8 JF |
778 | init_mm.pgd = base; |
779 | /* | |
780 | * copy top-level of Xen-supplied pagetable into place. For | |
781 | * !PAE we can use this as-is, but for PAE it is a stand-in | |
782 | * while we copy the pmd pages. | |
783 | */ | |
784 | memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t)); | |
785 | ||
786 | if (PTRS_PER_PMD > 1) { | |
787 | int i; | |
788 | /* | |
789 | * For PAE, need to allocate new pmds, rather than | |
790 | * share Xen's, since Xen doesn't like pmd's being | |
791 | * shared between address spaces. | |
792 | */ | |
793 | for (i = 0; i < PTRS_PER_PGD; i++) { | |
794 | if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) { | |
795 | pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE); | |
796 | ||
797 | memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]), | |
798 | PAGE_SIZE); | |
799 | ||
f4f97b3e | 800 | make_lowmem_page_readonly(pmd); |
5ead97c8 JF |
801 | |
802 | set_pgd(&base[i], __pgd(1 + __pa(pmd))); | |
803 | } else | |
804 | pgd_clear(&base[i]); | |
805 | } | |
806 | } | |
807 | ||
808 | /* make sure zero_page is mapped RO so we can use it in pagetables */ | |
809 | make_lowmem_page_readonly(empty_zero_page); | |
810 | make_lowmem_page_readonly(base); | |
811 | /* | |
812 | * Switch to new pagetable. This is done before | |
813 | * pagetable_init has done anything so that the new pages | |
814 | * added to the table can be prepared properly for Xen. | |
815 | */ | |
816 | xen_write_cr3(__pa(base)); | |
2b540781 JF |
817 | |
818 | /* Unpin initial Xen pagetable */ | |
819 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, | |
820 | PFN_DOWN(__pa(xen_start_info->pt_base))); | |
5ead97c8 JF |
821 | } |
822 | ||
2e8fe719 | 823 | static __init void setup_shared_info(void) |
5ead97c8 JF |
824 | { |
825 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
2e8fe719 JF |
826 | unsigned long addr = fix_to_virt(FIX_PARAVIRT_BOOTMAP); |
827 | ||
5ead97c8 JF |
828 | /* |
829 | * Create a mapping for the shared info page. | |
830 | * Should be set_fixmap(), but shared_info is a machine | |
831 | * address with no corresponding pseudo-phys address. | |
832 | */ | |
2e8fe719 | 833 | set_pte_mfn(addr, |
5ead97c8 JF |
834 | PFN_DOWN(xen_start_info->shared_info), |
835 | PAGE_KERNEL); | |
5ead97c8 | 836 | |
2e8fe719 | 837 | HYPERVISOR_shared_info = (struct shared_info *)addr; |
5ead97c8 JF |
838 | } else |
839 | HYPERVISOR_shared_info = | |
840 | (struct shared_info *)__va(xen_start_info->shared_info); | |
841 | ||
2e8fe719 JF |
842 | #ifndef CONFIG_SMP |
843 | /* In UP this is as good a place as any to set up shared info */ | |
844 | xen_setup_vcpu_info_placement(); | |
845 | #endif | |
846 | } | |
847 | ||
848 | static __init void xen_pagetable_setup_done(pgd_t *base) | |
849 | { | |
850 | /* This will work as long as patching hasn't happened yet | |
851 | (which it hasn't) */ | |
852 | pv_mmu_ops.alloc_pt = xen_alloc_pt; | |
853 | pv_mmu_ops.alloc_pd = xen_alloc_pd; | |
854 | pv_mmu_ops.release_pt = xen_release_pt; | |
f6433706 | 855 | pv_mmu_ops.release_pd = xen_release_pd; |
2e8fe719 JF |
856 | pv_mmu_ops.set_pte = xen_set_pte; |
857 | ||
858 | setup_shared_info(); | |
859 | ||
f4f97b3e JF |
860 | /* Actually pin the pagetable down, but we can't set PG_pinned |
861 | yet because the page structures don't exist yet. */ | |
862 | { | |
74260714 JF |
863 | unsigned level; |
864 | ||
f4f97b3e | 865 | #ifdef CONFIG_X86_PAE |
74260714 | 866 | level = MMUEXT_PIN_L3_TABLE; |
f4f97b3e | 867 | #else |
74260714 | 868 | level = MMUEXT_PIN_L2_TABLE; |
f4f97b3e | 869 | #endif |
74260714 JF |
870 | |
871 | pin_pagetable_pfn(level, PFN_DOWN(__pa(base))); | |
f4f97b3e | 872 | } |
60223a32 | 873 | } |
5ead97c8 | 874 | |
60223a32 JF |
875 | /* This is called once we have the cpu_possible_map */ |
876 | void __init xen_setup_vcpu_info_placement(void) | |
877 | { | |
878 | int cpu; | |
879 | ||
880 | for_each_possible_cpu(cpu) | |
881 | xen_vcpu_setup(cpu); | |
882 | ||
883 | /* xen_vcpu_setup managed to place the vcpu_info within the | |
884 | percpu area for all cpus, so make use of it */ | |
885 | if (have_vcpu_info_placement) { | |
886 | printk(KERN_INFO "Xen: using vcpu_info placement\n"); | |
887 | ||
93b1eab3 JF |
888 | pv_irq_ops.save_fl = xen_save_fl_direct; |
889 | pv_irq_ops.restore_fl = xen_restore_fl_direct; | |
890 | pv_irq_ops.irq_disable = xen_irq_disable_direct; | |
891 | pv_irq_ops.irq_enable = xen_irq_enable_direct; | |
892 | pv_mmu_ops.read_cr2 = xen_read_cr2_direct; | |
893 | pv_cpu_ops.iret = xen_iret_direct; | |
60223a32 | 894 | } |
5ead97c8 JF |
895 | } |
896 | ||
ab144f5e AK |
897 | static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf, |
898 | unsigned long addr, unsigned len) | |
6487673b JF |
899 | { |
900 | char *start, *end, *reloc; | |
901 | unsigned ret; | |
902 | ||
903 | start = end = reloc = NULL; | |
904 | ||
93b1eab3 JF |
905 | #define SITE(op, x) \ |
906 | case PARAVIRT_PATCH(op.x): \ | |
6487673b JF |
907 | if (have_vcpu_info_placement) { \ |
908 | start = (char *)xen_##x##_direct; \ | |
909 | end = xen_##x##_direct_end; \ | |
910 | reloc = xen_##x##_direct_reloc; \ | |
911 | } \ | |
912 | goto patch_site | |
913 | ||
914 | switch (type) { | |
93b1eab3 JF |
915 | SITE(pv_irq_ops, irq_enable); |
916 | SITE(pv_irq_ops, irq_disable); | |
917 | SITE(pv_irq_ops, save_fl); | |
918 | SITE(pv_irq_ops, restore_fl); | |
6487673b JF |
919 | #undef SITE |
920 | ||
921 | patch_site: | |
922 | if (start == NULL || (end-start) > len) | |
923 | goto default_patch; | |
924 | ||
ab144f5e | 925 | ret = paravirt_patch_insns(insnbuf, len, start, end); |
6487673b JF |
926 | |
927 | /* Note: because reloc is assigned from something that | |
928 | appears to be an array, gcc assumes it's non-null, | |
929 | but doesn't know its relationship with start and | |
930 | end. */ | |
931 | if (reloc > start && reloc < end) { | |
932 | int reloc_off = reloc - start; | |
ab144f5e AK |
933 | long *relocp = (long *)(insnbuf + reloc_off); |
934 | long delta = start - (char *)addr; | |
6487673b JF |
935 | |
936 | *relocp += delta; | |
937 | } | |
938 | break; | |
939 | ||
940 | default_patch: | |
941 | default: | |
ab144f5e AK |
942 | ret = paravirt_patch_default(type, clobbers, insnbuf, |
943 | addr, len); | |
6487673b JF |
944 | break; |
945 | } | |
946 | ||
947 | return ret; | |
948 | } | |
949 | ||
93b1eab3 | 950 | static const struct pv_info xen_info __initdata = { |
5ead97c8 JF |
951 | .paravirt_enabled = 1, |
952 | .shared_kernel_pmd = 0, | |
953 | ||
954 | .name = "Xen", | |
93b1eab3 | 955 | }; |
5ead97c8 | 956 | |
93b1eab3 | 957 | static const struct pv_init_ops xen_init_ops __initdata = { |
6487673b | 958 | .patch = xen_patch, |
5ead97c8 | 959 | |
93b1eab3 | 960 | .banner = xen_banner, |
5ead97c8 JF |
961 | .memory_setup = xen_memory_setup, |
962 | .arch_setup = xen_arch_setup, | |
f4f97b3e | 963 | .post_allocator_init = xen_mark_init_mm_pinned, |
93b1eab3 | 964 | }; |
5ead97c8 | 965 | |
93b1eab3 | 966 | static const struct pv_time_ops xen_time_ops __initdata = { |
15c84731 | 967 | .time_init = xen_time_init, |
93b1eab3 | 968 | |
15c84731 JF |
969 | .set_wallclock = xen_set_wallclock, |
970 | .get_wallclock = xen_get_wallclock, | |
971 | .get_cpu_khz = xen_cpu_khz, | |
ab550288 | 972 | .sched_clock = xen_sched_clock, |
93b1eab3 | 973 | }; |
15c84731 | 974 | |
93b1eab3 | 975 | static const struct pv_cpu_ops xen_cpu_ops __initdata = { |
5ead97c8 JF |
976 | .cpuid = xen_cpuid, |
977 | ||
978 | .set_debugreg = xen_set_debugreg, | |
979 | .get_debugreg = xen_get_debugreg, | |
980 | ||
981 | .clts = native_clts, | |
982 | ||
983 | .read_cr0 = native_read_cr0, | |
984 | .write_cr0 = native_write_cr0, | |
985 | ||
5ead97c8 JF |
986 | .read_cr4 = native_read_cr4, |
987 | .read_cr4_safe = native_read_cr4_safe, | |
988 | .write_cr4 = xen_write_cr4, | |
989 | ||
5ead97c8 JF |
990 | .wbinvd = native_wbinvd, |
991 | ||
992 | .read_msr = native_read_msr_safe, | |
993 | .write_msr = native_write_msr_safe, | |
994 | .read_tsc = native_read_tsc, | |
995 | .read_pmc = native_read_pmc, | |
996 | ||
997 | .iret = (void *)&hypercall_page[__HYPERVISOR_iret], | |
6abcd98f | 998 | .irq_enable_syscall_ret = NULL, /* never called */ |
5ead97c8 JF |
999 | |
1000 | .load_tr_desc = paravirt_nop, | |
1001 | .set_ldt = xen_set_ldt, | |
1002 | .load_gdt = xen_load_gdt, | |
1003 | .load_idt = xen_load_idt, | |
1004 | .load_tls = xen_load_tls, | |
1005 | ||
1006 | .store_gdt = native_store_gdt, | |
1007 | .store_idt = native_store_idt, | |
1008 | .store_tr = xen_store_tr, | |
1009 | ||
1010 | .write_ldt_entry = xen_write_ldt_entry, | |
1011 | .write_gdt_entry = xen_write_gdt_entry, | |
1012 | .write_idt_entry = xen_write_idt_entry, | |
faca6227 | 1013 | .load_sp0 = xen_load_sp0, |
5ead97c8 JF |
1014 | |
1015 | .set_iopl_mask = xen_set_iopl_mask, | |
1016 | .io_delay = xen_io_delay, | |
1017 | ||
8965c1c0 JF |
1018 | .lazy_mode = { |
1019 | .enter = paravirt_enter_lazy_cpu, | |
1020 | .leave = xen_leave_lazy, | |
1021 | }, | |
93b1eab3 JF |
1022 | }; |
1023 | ||
1024 | static const struct pv_irq_ops xen_irq_ops __initdata = { | |
1025 | .init_IRQ = xen_init_IRQ, | |
1026 | .save_fl = xen_save_fl, | |
1027 | .restore_fl = xen_restore_fl, | |
1028 | .irq_disable = xen_irq_disable, | |
1029 | .irq_enable = xen_irq_enable, | |
1030 | .safe_halt = xen_safe_halt, | |
1031 | .halt = xen_halt, | |
1032 | }; | |
5ead97c8 | 1033 | |
93b1eab3 | 1034 | static const struct pv_apic_ops xen_apic_ops __initdata = { |
5ead97c8 | 1035 | #ifdef CONFIG_X86_LOCAL_APIC |
f87e4cac JF |
1036 | .apic_write = xen_apic_write, |
1037 | .apic_write_atomic = xen_apic_write, | |
5ead97c8 JF |
1038 | .apic_read = xen_apic_read, |
1039 | .setup_boot_clock = paravirt_nop, | |
1040 | .setup_secondary_clock = paravirt_nop, | |
1041 | .startup_ipi_hook = paravirt_nop, | |
1042 | #endif | |
93b1eab3 JF |
1043 | }; |
1044 | ||
1045 | static const struct pv_mmu_ops xen_mmu_ops __initdata = { | |
1046 | .pagetable_setup_start = xen_pagetable_setup_start, | |
1047 | .pagetable_setup_done = xen_pagetable_setup_done, | |
1048 | ||
1049 | .read_cr2 = xen_read_cr2, | |
1050 | .write_cr2 = xen_write_cr2, | |
1051 | ||
1052 | .read_cr3 = xen_read_cr3, | |
1053 | .write_cr3 = xen_write_cr3, | |
5ead97c8 JF |
1054 | |
1055 | .flush_tlb_user = xen_flush_tlb, | |
1056 | .flush_tlb_kernel = xen_flush_tlb, | |
1057 | .flush_tlb_single = xen_flush_tlb_single, | |
f87e4cac | 1058 | .flush_tlb_others = xen_flush_tlb_others, |
5ead97c8 JF |
1059 | |
1060 | .pte_update = paravirt_nop, | |
1061 | .pte_update_defer = paravirt_nop, | |
1062 | ||
f4f97b3e | 1063 | .alloc_pt = xen_alloc_pt_init, |
1c70e9bd JF |
1064 | .release_pt = xen_release_pt_init, |
1065 | .alloc_pd = xen_alloc_pt_init, | |
f4f97b3e | 1066 | .alloc_pd_clone = paravirt_nop, |
1c70e9bd | 1067 | .release_pd = xen_release_pt_init, |
f4f97b3e JF |
1068 | |
1069 | #ifdef CONFIG_HIGHPTE | |
1070 | .kmap_atomic_pte = xen_kmap_atomic_pte, | |
1071 | #endif | |
5ead97c8 | 1072 | |
9a4029fd | 1073 | .set_pte = NULL, /* see xen_pagetable_setup_* */ |
3b827c1b JF |
1074 | .set_pte_at = xen_set_pte_at, |
1075 | .set_pmd = xen_set_pmd, | |
1076 | ||
1077 | .pte_val = xen_pte_val, | |
1078 | .pgd_val = xen_pgd_val, | |
1079 | ||
1080 | .make_pte = xen_make_pte, | |
1081 | .make_pgd = xen_make_pgd, | |
1082 | ||
1083 | #ifdef CONFIG_X86_PAE | |
1084 | .set_pte_atomic = xen_set_pte_atomic, | |
1085 | .set_pte_present = xen_set_pte_at, | |
1086 | .set_pud = xen_set_pud, | |
1087 | .pte_clear = xen_pte_clear, | |
1088 | .pmd_clear = xen_pmd_clear, | |
1089 | ||
1090 | .make_pmd = xen_make_pmd, | |
1091 | .pmd_val = xen_pmd_val, | |
1092 | #endif /* PAE */ | |
1093 | ||
1094 | .activate_mm = xen_activate_mm, | |
1095 | .dup_mmap = xen_dup_mmap, | |
1096 | .exit_mmap = xen_exit_mmap, | |
1097 | ||
8965c1c0 JF |
1098 | .lazy_mode = { |
1099 | .enter = paravirt_enter_lazy_mmu, | |
1100 | .leave = xen_leave_lazy, | |
1101 | }, | |
5ead97c8 JF |
1102 | }; |
1103 | ||
f87e4cac JF |
1104 | #ifdef CONFIG_SMP |
1105 | static const struct smp_ops xen_smp_ops __initdata = { | |
1106 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, | |
1107 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
1108 | .cpu_up = xen_cpu_up, | |
1109 | .smp_cpus_done = xen_smp_cpus_done, | |
1110 | ||
1111 | .smp_send_stop = xen_smp_send_stop, | |
1112 | .smp_send_reschedule = xen_smp_send_reschedule, | |
1113 | .smp_call_function_mask = xen_smp_call_function_mask, | |
1114 | }; | |
1115 | #endif /* CONFIG_SMP */ | |
1116 | ||
fefa629a JF |
1117 | static void xen_reboot(int reason) |
1118 | { | |
1119 | #ifdef CONFIG_SMP | |
1120 | smp_send_stop(); | |
1121 | #endif | |
1122 | ||
1123 | if (HYPERVISOR_sched_op(SCHEDOP_shutdown, reason)) | |
1124 | BUG(); | |
1125 | } | |
1126 | ||
1127 | static void xen_restart(char *msg) | |
1128 | { | |
1129 | xen_reboot(SHUTDOWN_reboot); | |
1130 | } | |
1131 | ||
1132 | static void xen_emergency_restart(void) | |
1133 | { | |
1134 | xen_reboot(SHUTDOWN_reboot); | |
1135 | } | |
1136 | ||
1137 | static void xen_machine_halt(void) | |
1138 | { | |
1139 | xen_reboot(SHUTDOWN_poweroff); | |
1140 | } | |
1141 | ||
1142 | static void xen_crash_shutdown(struct pt_regs *regs) | |
1143 | { | |
1144 | xen_reboot(SHUTDOWN_crash); | |
1145 | } | |
1146 | ||
1147 | static const struct machine_ops __initdata xen_machine_ops = { | |
1148 | .restart = xen_restart, | |
1149 | .halt = xen_machine_halt, | |
1150 | .power_off = xen_machine_halt, | |
1151 | .shutdown = xen_machine_halt, | |
1152 | .crash_shutdown = xen_crash_shutdown, | |
1153 | .emergency_restart = xen_emergency_restart, | |
1154 | }; | |
1155 | ||
6487673b | 1156 | |
fb1d8404 JF |
1157 | static void __init xen_reserve_top(void) |
1158 | { | |
1159 | unsigned long top = HYPERVISOR_VIRT_START; | |
1160 | struct xen_platform_parameters pp; | |
1161 | ||
1162 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1163 | top = pp.virt_start; | |
1164 | ||
1165 | reserve_top_address(-top + 2 * PAGE_SIZE); | |
1166 | } | |
1167 | ||
5ead97c8 JF |
1168 | /* First C function to be called on Xen boot */ |
1169 | asmlinkage void __init xen_start_kernel(void) | |
1170 | { | |
1171 | pgd_t *pgd; | |
1172 | ||
1173 | if (!xen_start_info) | |
1174 | return; | |
1175 | ||
7999f4b4 | 1176 | BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0); |
5ead97c8 JF |
1177 | |
1178 | /* Install Xen paravirt ops */ | |
93b1eab3 JF |
1179 | pv_info = xen_info; |
1180 | pv_init_ops = xen_init_ops; | |
1181 | pv_time_ops = xen_time_ops; | |
1182 | pv_cpu_ops = xen_cpu_ops; | |
1183 | pv_irq_ops = xen_irq_ops; | |
1184 | pv_apic_ops = xen_apic_ops; | |
1185 | pv_mmu_ops = xen_mmu_ops; | |
93b1eab3 | 1186 | |
fefa629a JF |
1187 | machine_ops = xen_machine_ops; |
1188 | ||
f87e4cac JF |
1189 | #ifdef CONFIG_SMP |
1190 | smp_ops = xen_smp_ops; | |
1191 | #endif | |
5ead97c8 JF |
1192 | |
1193 | xen_setup_features(); | |
1194 | ||
1195 | /* Get mfn list */ | |
1196 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
1197 | phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list; | |
1198 | ||
1199 | pgd = (pgd_t *)xen_start_info->pt_base; | |
1200 | ||
1201 | init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE; | |
1202 | ||
1203 | init_mm.pgd = pgd; /* use the Xen pagetables to start */ | |
1204 | ||
1205 | /* keep using Xen gdt for now; no urgent need to change it */ | |
1206 | ||
1207 | x86_write_percpu(xen_cr3, __pa(pgd)); | |
9f79991d | 1208 | x86_write_percpu(xen_current_cr3, __pa(pgd)); |
60223a32 | 1209 | |
60223a32 | 1210 | /* Don't do the full vcpu_info placement stuff until we have a |
2e8fe719 | 1211 | possible map and a non-dummy shared_info. */ |
60223a32 | 1212 | per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0]; |
5ead97c8 | 1213 | |
93b1eab3 | 1214 | pv_info.kernel_rpl = 1; |
5ead97c8 | 1215 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) |
93b1eab3 | 1216 | pv_info.kernel_rpl = 0; |
5ead97c8 JF |
1217 | |
1218 | /* set the limit of our address space */ | |
fb1d8404 | 1219 | xen_reserve_top(); |
5ead97c8 JF |
1220 | |
1221 | /* set up basic CPUID stuff */ | |
1222 | cpu_detect(&new_cpu_data); | |
1223 | new_cpu_data.hard_math = 1; | |
1224 | new_cpu_data.x86_capability[0] = cpuid_edx(1); | |
1225 | ||
1226 | /* Poke various useful things into boot_params */ | |
30c82645 PA |
1227 | boot_params.hdr.type_of_loader = (9 << 4) | 0; |
1228 | boot_params.hdr.ramdisk_image = xen_start_info->mod_start | |
1229 | ? __pa(xen_start_info->mod_start) : 0; | |
1230 | boot_params.hdr.ramdisk_size = xen_start_info->mod_len; | |
5ead97c8 | 1231 | |
b8c2d3df MA |
1232 | if (!is_initial_xendomain()) |
1233 | add_preferred_console("hvc", 0, NULL); | |
1234 | ||
5ead97c8 JF |
1235 | /* Start the world */ |
1236 | start_kernel(); | |
1237 | } |