blob: cf8b3a93122b9cf616780ec08f08d363807873a8 [file] [log] [blame]
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001/*
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>
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -070018#include <linux/hardirq.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070019#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>
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -070025#include <linux/mm.h>
26#include <linux/page-flags.h>
27#include <linux/highmem.h>
Markus Armbrusterb8c2d3d2008-02-27 14:56:35 +010028#include <linux/console.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070029
30#include <xen/interface/xen.h>
31#include <xen/interface/physdev.h>
32#include <xen/interface/vcpu.h>
33#include <xen/features.h>
34#include <xen/page.h>
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -070035#include <xen/hvc-console.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070036
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>
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -070043#include <asm/msr-index.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070044#include <asm/setup.h>
45#include <asm/desc.h>
46#include <asm/pgtable.h>
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -070047#include <asm/tlbflush.h>
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -070048#include <asm/reboot.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070049
50#include "xen-ops.h"
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -070051#include "mmu.h"
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070052#include "multicalls.h"
53
54EXPORT_SYMBOL_GPL(hypercall_page);
55
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070056DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
57DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -070058
59/*
Jeremy Fitzhardingebf18bf92008-07-08 15:07:15 -070060 * Identity map, in addition to plain kernel map. This needs to be
61 * large enough to allocate page table pages to allocate the rest.
62 * Each page can map 2MB.
63 */
64static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
65
66#ifdef CONFIG_X86_64
67/* l3 pud for userspace vsyscall mapping */
68static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
69#endif /* CONFIG_X86_64 */
70
71/*
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -070072 * Note about cr3 (pagetable base) values:
73 *
74 * xen_cr3 contains the current logical cr3 value; it contains the
75 * last set cr3. This may not be the current effective cr3, because
76 * its update may be being lazily deferred. However, a vcpu looking
77 * at its own cr3 can use this value knowing that it everything will
78 * be self-consistent.
79 *
80 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
81 * hypercall to set the vcpu cr3 is complete (so it may be a little
82 * out of date, but it will never be set early). If one vcpu is
83 * looking at another vcpu's cr3 value, it should use this variable.
84 */
85DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
86DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070087
88struct start_info *xen_start_info;
89EXPORT_SYMBOL_GPL(xen_start_info);
90
Jeremy Fitzhardingea0d695c2008-05-26 23:31:21 +010091struct shared_info xen_dummy_shared_info;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -070092
93/*
94 * Point at some empty memory to start with. We map the real shared_info
95 * page as soon as fixmap is up and running.
96 */
Jeremy Fitzhardingea0d695c2008-05-26 23:31:21 +010097struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -070098
99/*
100 * Flag to determine whether vcpu info placement is available on all
101 * VCPUs. We assume it is to start with, and then set it to zero on
102 * the first failure. This is because it can succeed on some VCPUs
103 * and not others, since it can involve hypervisor memory allocation,
104 * or because the guest failed to guarantee all the appropriate
105 * constraints on all VCPUs (ie buffer can't cross a page boundary).
106 *
107 * Note that any particular CPU may be using a placed vcpu structure,
108 * but we can only optimise if the all are.
109 *
110 * 0: not available, 1: available
111 */
Jeremy Fitzhardinge04c44a02008-03-17 16:36:52 -0700112static int have_vcpu_info_placement = 1;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700113
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100114static void xen_vcpu_setup(int cpu)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700115{
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700116 struct vcpu_register_vcpu_info info;
117 int err;
118 struct vcpu_info *vcpup;
119
Jeremy Fitzhardingea0d695c2008-05-26 23:31:21 +0100120 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700121 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700122
123 if (!have_vcpu_info_placement)
124 return; /* already tested, not available */
125
126 vcpup = &per_cpu(xen_vcpu_info, cpu);
127
128 info.mfn = virt_to_mfn(vcpup);
129 info.offset = offset_in_page(vcpup);
130
Jeremy Fitzhardingee3d26972007-10-16 11:51:31 -0700131 printk(KERN_DEBUG "trying to map vcpu_info %d at %p, mfn %llx, offset %d\n",
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700132 cpu, vcpup, info.mfn, info.offset);
133
134 /* Check to see if the hypervisor will put the vcpu_info
135 structure where we want it, which allows direct access via
136 a percpu-variable. */
137 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
138
139 if (err) {
140 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
141 have_vcpu_info_placement = 0;
142 } else {
143 /* This cpu is using the registered vcpu info, even if
144 later ones fail to. */
145 per_cpu(xen_vcpu, cpu) = vcpup;
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -0700146
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700147 printk(KERN_DEBUG "cpu %d using vcpu_info at %p\n",
148 cpu, vcpup);
149 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700150}
151
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100152/*
153 * On restore, set the vcpu placement up again.
154 * If it fails, then we're in a bad state, since
155 * we can't back out from using it...
156 */
157void xen_vcpu_restore(void)
158{
159 if (have_vcpu_info_placement) {
160 int cpu;
161
162 for_each_online_cpu(cpu) {
163 bool other_cpu = (cpu != smp_processor_id());
164
165 if (other_cpu &&
166 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
167 BUG();
168
169 xen_vcpu_setup(cpu);
170
171 if (other_cpu &&
172 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
173 BUG();
174 }
175
176 BUG_ON(!have_vcpu_info_placement);
177 }
178}
179
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700180static void __init xen_banner(void)
181{
Jeremy Fitzhardinge95c7c232008-07-15 13:42:34 -0700182 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
183 struct xen_extraversion extra;
184 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
185
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700186 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700187 pv_info.name);
Jeremy Fitzhardinge95c7c232008-07-15 13:42:34 -0700188 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
189 version >> 16, version & 0xffff, extra.extraversion,
Jeremy Fitzhardingee57778a2008-06-16 04:30:02 -0700190 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700191}
192
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100193static void xen_cpuid(unsigned int *ax, unsigned int *bx,
194 unsigned int *cx, unsigned int *dx)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700195{
196 unsigned maskedx = ~0;
197
198 /*
199 * Mask out inconvenient features, to try and disable as many
200 * unsupported kernel subsystems as possible.
201 */
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100202 if (*ax == 1)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700203 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */
204 (1 << X86_FEATURE_ACPI) | /* disable ACPI */
Jeremy Fitzhardingedbe9e992008-03-17 16:37:21 -0700205 (1 << X86_FEATURE_MCE) | /* disable MCE */
206 (1 << X86_FEATURE_MCA) | /* disable MCA */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700207 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
208
209 asm(XEN_EMULATE_PREFIX "cpuid"
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100210 : "=a" (*ax),
211 "=b" (*bx),
212 "=c" (*cx),
213 "=d" (*dx)
214 : "0" (*ax), "2" (*cx));
215 *dx &= maskedx;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700216}
217
218static void xen_set_debugreg(int reg, unsigned long val)
219{
220 HYPERVISOR_set_debugreg(reg, val);
221}
222
223static unsigned long xen_get_debugreg(int reg)
224{
225 return HYPERVISOR_get_debugreg(reg);
226}
227
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700228static void xen_leave_lazy(void)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700229{
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700230 paravirt_leave_lazy(paravirt_get_lazy_mode());
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700231 xen_mc_flush();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700232}
233
234static unsigned long xen_store_tr(void)
235{
236 return 0;
237}
238
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700239/*
240 * If 'v' is a vmalloc mapping, then find the linear mapping of the
241 * page (if any) and also set its protections to match:
242 */
243static void set_aliased_prot(void *v, pgprot_t prot)
244{
245 int level;
246 pte_t *ptep;
247 pte_t pte;
248 unsigned long pfn;
249 struct page *page;
250
251 ptep = lookup_address((unsigned long)v, &level);
252 BUG_ON(ptep == NULL);
253
254 pfn = pte_pfn(*ptep);
255 page = pfn_to_page(pfn);
256
257 pte = pfn_pte(pfn, prot);
258
259 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
260 BUG();
261
262 if (!PageHighMem(page)) {
263 void *av = __va(PFN_PHYS(pfn));
264
265 if (av != v)
266 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
267 BUG();
268 } else
269 kmap_flush_unused();
270}
271
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700272static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
273{
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700274 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700275 int i;
276
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700277 for(i = 0; i < entries; i += entries_per_page)
278 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700279}
280
281static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
282{
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700283 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700284 int i;
285
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700286 for(i = 0; i < entries; i += entries_per_page)
287 set_aliased_prot(ldt + i, PAGE_KERNEL);
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700288}
289
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700290static void xen_set_ldt(const void *addr, unsigned entries)
291{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700292 struct mmuext_op *op;
293 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
294
295 op = mcs.args;
296 op->cmd = MMUEXT_SET_LDT;
Jan Beulich4dbf7af2008-01-30 13:33:14 +0100297 op->arg1.linear_addr = (unsigned long)addr;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700298 op->arg2.nr_ents = entries;
299
300 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
301
302 xen_mc_issue(PARAVIRT_LAZY_CPU);
303}
304
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100305static void xen_load_gdt(const struct desc_ptr *dtr)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700306{
307 unsigned long *frames;
308 unsigned long va = dtr->address;
309 unsigned int size = dtr->size + 1;
310 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
311 int f;
312 struct multicall_space mcs;
313
314 /* A GDT can be up to 64k in size, which corresponds to 8192
315 8-byte entries, or 16 4k pages.. */
316
317 BUG_ON(size > 65536);
318 BUG_ON(va & ~PAGE_MASK);
319
320 mcs = xen_mc_entry(sizeof(*frames) * pages);
321 frames = mcs.args;
322
323 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
324 frames[f] = virt_to_mfn(va);
325 make_lowmem_page_readonly((void *)va);
326 }
327
328 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
329
330 xen_mc_issue(PARAVIRT_LAZY_CPU);
331}
332
333static void load_TLS_descriptor(struct thread_struct *t,
334 unsigned int cpu, unsigned int i)
335{
336 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
337 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
338 struct multicall_space mc = __xen_mc_entry(0);
339
340 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
341}
342
343static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
344{
Jeremy Fitzhardinge8b84ad92007-07-17 18:37:06 -0700345 /*
346 * XXX sleazy hack: If we're being called in a lazy-cpu zone,
347 * it means we're in a context switch, and %gs has just been
348 * saved. This means we can zero it out to prevent faults on
349 * exit from the hypervisor if the next process has no %gs.
350 * Either way, it has been saved, and the new value will get
351 * loaded properly. This will go away as soon as Xen has been
352 * modified to not save/restore %gs for normal hypercalls.
Eduardo Habkost8a954082008-07-08 15:07:10 -0700353 *
354 * On x86_64, this hack is not used for %gs, because gs points
355 * to KERNEL_GS_BASE (and uses it for PDA references), so we
356 * must not zero %gs on x86_64
357 *
358 * For x86_64, we need to zero %fs, otherwise we may get an
359 * exception between the new %fs descriptor being loaded and
360 * %fs being effectively cleared at __switch_to().
Jeremy Fitzhardinge8b84ad92007-07-17 18:37:06 -0700361 */
Eduardo Habkost8a954082008-07-08 15:07:10 -0700362 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
363#ifdef CONFIG_X86_32
Jeremy Fitzhardinge8b84ad92007-07-17 18:37:06 -0700364 loadsegment(gs, 0);
Eduardo Habkost8a954082008-07-08 15:07:10 -0700365#else
366 loadsegment(fs, 0);
367#endif
368 }
369
370 xen_mc_batch();
371
372 load_TLS_descriptor(t, cpu, 0);
373 load_TLS_descriptor(t, cpu, 1);
374 load_TLS_descriptor(t, cpu, 2);
375
376 xen_mc_issue(PARAVIRT_LAZY_CPU);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700377}
378
Eduardo Habkosta8fc1082008-07-08 15:07:05 -0700379#ifdef CONFIG_X86_64
380static void xen_load_gs_index(unsigned int idx)
381{
382 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
383 BUG();
384}
385#endif
386
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700387static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100388 const void *ptr)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700389{
390 unsigned long lp = (unsigned long)&dt[entrynum];
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700391 xmaddr_t mach_lp = arbitrary_virt_to_machine(lp);
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100392 u64 entry = *(u64 *)ptr;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700393
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700394 preempt_disable();
395
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700396 xen_mc_flush();
397 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
398 BUG();
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700399
400 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700401}
402
Eduardo Habkoste176d362008-07-08 15:06:59 -0700403static int cvt_gate_to_trap(int vector, const gate_desc *val,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700404 struct trap_info *info)
405{
Eduardo Habkoste176d362008-07-08 15:06:59 -0700406 if (val->type != 0xf && val->type != 0xe)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700407 return 0;
408
409 info->vector = vector;
Eduardo Habkoste176d362008-07-08 15:06:59 -0700410 info->address = gate_offset(*val);
411 info->cs = gate_segment(*val);
412 info->flags = val->dpl;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700413 /* interrupt gates clear IF */
Eduardo Habkoste176d362008-07-08 15:06:59 -0700414 if (val->type == 0xe)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700415 info->flags |= 4;
416
417 return 1;
418}
419
420/* Locations of each CPU's IDT */
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100421static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700422
423/* Set an IDT entry. If the entry is part of the current IDT, then
424 also update Xen. */
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100425static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700426{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700427 unsigned long p = (unsigned long)&dt[entrynum];
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700428 unsigned long start, end;
429
430 preempt_disable();
431
432 start = __get_cpu_var(idt_desc).address;
433 end = start + __get_cpu_var(idt_desc).size + 1;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700434
435 xen_mc_flush();
436
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100437 native_write_idt_entry(dt, entrynum, g);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700438
439 if (p >= start && (p + 8) <= end) {
440 struct trap_info info[2];
441
442 info[1].address = 0;
443
Eduardo Habkoste176d362008-07-08 15:06:59 -0700444 if (cvt_gate_to_trap(entrynum, g, &info[0]))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700445 if (HYPERVISOR_set_trap_table(info))
446 BUG();
447 }
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700448
449 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700450}
451
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100452static void xen_convert_trap_info(const struct desc_ptr *desc,
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700453 struct trap_info *traps)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700454{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700455 unsigned in, out, count;
456
Eduardo Habkoste176d362008-07-08 15:06:59 -0700457 count = (desc->size+1) / sizeof(gate_desc);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700458 BUG_ON(count > 256);
459
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700460 for (in = out = 0; in < count; in++) {
Eduardo Habkoste176d362008-07-08 15:06:59 -0700461 gate_desc *entry = (gate_desc*)(desc->address) + in;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700462
Eduardo Habkoste176d362008-07-08 15:06:59 -0700463 if (cvt_gate_to_trap(in, entry, &traps[out]))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700464 out++;
465 }
466 traps[out].address = 0;
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700467}
468
469void xen_copy_trap_info(struct trap_info *traps)
470{
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100471 const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700472
473 xen_convert_trap_info(desc, traps);
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700474}
475
476/* Load a new IDT into Xen. In principle this can be per-CPU, so we
477 hold a spinlock to protect the static traps[] array (static because
478 it avoids allocation, and saves stack space). */
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100479static void xen_load_idt(const struct desc_ptr *desc)
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700480{
481 static DEFINE_SPINLOCK(lock);
482 static struct trap_info traps[257];
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700483
484 spin_lock(&lock);
485
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700486 __get_cpu_var(idt_desc) = *desc;
487
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700488 xen_convert_trap_info(desc, traps);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700489
490 xen_mc_flush();
491 if (HYPERVISOR_set_trap_table(traps))
492 BUG();
493
494 spin_unlock(&lock);
495}
496
497/* Write a GDT descriptor entry. Ignore LDT descriptors, since
498 they're handled differently. */
499static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100500 const void *desc, int type)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700501{
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700502 preempt_disable();
503
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100504 switch (type) {
505 case DESC_LDT:
506 case DESC_TSS:
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700507 /* ignore */
508 break;
509
510 default: {
511 xmaddr_t maddr = virt_to_machine(&dt[entry]);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700512
513 xen_mc_flush();
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100514 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700515 BUG();
516 }
517
518 }
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700519
520 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700521}
522
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100523static void xen_load_sp0(struct tss_struct *tss,
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700524 struct thread_struct *thread)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700525{
526 struct multicall_space mcs = xen_mc_entry(0);
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100527 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700528 xen_mc_issue(PARAVIRT_LAZY_CPU);
529}
530
531static void xen_set_iopl_mask(unsigned mask)
532{
533 struct physdev_set_iopl set_iopl;
534
535 /* Force the change at ring 0. */
536 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
537 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
538}
539
540static void xen_io_delay(void)
541{
542}
543
544#ifdef CONFIG_X86_LOCAL_APIC
Thomas Gleixner42e0a9a2008-01-30 13:30:15 +0100545static u32 xen_apic_read(unsigned long reg)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700546{
547 return 0;
548}
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700549
Thomas Gleixner42e0a9a2008-01-30 13:30:15 +0100550static void xen_apic_write(unsigned long reg, u32 val)
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700551{
552 /* Warn to see if there's any stray references */
553 WARN_ON(1);
554}
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700555#endif
556
557static void xen_flush_tlb(void)
558{
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700559 struct mmuext_op *op;
Jeremy Fitzhardinge41e332b2008-04-02 10:54:09 -0700560 struct multicall_space mcs;
561
562 preempt_disable();
563
564 mcs = xen_mc_entry(sizeof(*op));
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700565
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700566 op = mcs.args;
567 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
568 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
569
570 xen_mc_issue(PARAVIRT_LAZY_MMU);
Jeremy Fitzhardinge41e332b2008-04-02 10:54:09 -0700571
572 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700573}
574
575static void xen_flush_tlb_single(unsigned long addr)
576{
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700577 struct mmuext_op *op;
Jeremy Fitzhardinge41e332b2008-04-02 10:54:09 -0700578 struct multicall_space mcs;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700579
Jeremy Fitzhardinge41e332b2008-04-02 10:54:09 -0700580 preempt_disable();
581
582 mcs = xen_mc_entry(sizeof(*op));
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700583 op = mcs.args;
584 op->cmd = MMUEXT_INVLPG_LOCAL;
585 op->arg1.linear_addr = addr & PAGE_MASK;
586 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
587
588 xen_mc_issue(PARAVIRT_LAZY_MMU);
Jeremy Fitzhardinge41e332b2008-04-02 10:54:09 -0700589
590 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700591}
592
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700593static void xen_flush_tlb_others(const cpumask_t *cpus, struct mm_struct *mm,
594 unsigned long va)
595{
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700596 struct {
597 struct mmuext_op op;
598 cpumask_t mask;
599 } *args;
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700600 cpumask_t cpumask = *cpus;
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700601 struct multicall_space mcs;
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700602
603 /*
604 * A couple of (to be removed) sanity checks:
605 *
606 * - current CPU must not be in mask
607 * - mask must exist :)
608 */
609 BUG_ON(cpus_empty(cpumask));
610 BUG_ON(cpu_isset(smp_processor_id(), cpumask));
611 BUG_ON(!mm);
612
613 /* If a CPU which we ran on has gone down, OK. */
614 cpus_and(cpumask, cpumask, cpu_online_map);
615 if (cpus_empty(cpumask))
616 return;
617
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700618 mcs = xen_mc_entry(sizeof(*args));
619 args = mcs.args;
620 args->mask = cpumask;
621 args->op.arg2.vcpumask = &args->mask;
622
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700623 if (va == TLB_FLUSH_ALL) {
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700624 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700625 } else {
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700626 args->op.cmd = MMUEXT_INVLPG_MULTI;
627 args->op.arg1.linear_addr = va;
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700628 }
629
Jeremy Fitzhardinged66bf8f2007-07-17 18:37:06 -0700630 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
631
632 xen_mc_issue(PARAVIRT_LAZY_MMU);
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700633}
634
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +0100635static void xen_clts(void)
636{
637 struct multicall_space mcs;
638
639 mcs = xen_mc_entry(0);
640
641 MULTI_fpu_taskswitch(mcs.mc, 0);
642
643 xen_mc_issue(PARAVIRT_LAZY_CPU);
644}
645
646static void xen_write_cr0(unsigned long cr0)
647{
648 struct multicall_space mcs;
649
650 /* Only pay attention to cr0.TS; everything else is
651 ignored. */
652 mcs = xen_mc_entry(0);
653
654 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
655
656 xen_mc_issue(PARAVIRT_LAZY_CPU);
657}
658
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700659static void xen_write_cr2(unsigned long cr2)
660{
661 x86_read_percpu(xen_vcpu)->arch.cr2 = cr2;
662}
663
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700664static unsigned long xen_read_cr2(void)
665{
666 return x86_read_percpu(xen_vcpu)->arch.cr2;
667}
668
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700669static unsigned long xen_read_cr2_direct(void)
670{
671 return x86_read_percpu(xen_vcpu_info.arch.cr2);
672}
673
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700674static void xen_write_cr4(unsigned long cr4)
675{
Jeremy Fitzhardinge2956a352008-05-26 23:31:04 +0100676 cr4 &= ~X86_CR4_PGE;
677 cr4 &= ~X86_CR4_PSE;
678
679 native_write_cr4(cr4);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700680}
681
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700682static unsigned long xen_read_cr3(void)
683{
684 return x86_read_percpu(xen_cr3);
685}
686
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -0700687static void set_current_cr3(void *v)
688{
689 x86_write_percpu(xen_current_cr3, (unsigned long)v);
690}
691
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700692static void __xen_write_cr3(bool kernel, unsigned long cr3)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700693{
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -0700694 struct mmuext_op *op;
695 struct multicall_space mcs;
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700696 unsigned long mfn;
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -0700697
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700698 if (cr3)
699 mfn = pfn_to_mfn(PFN_DOWN(cr3));
700 else
701 mfn = 0;
702
703 WARN_ON(mfn == 0 && kernel);
704
705 mcs = __xen_mc_entry(sizeof(*op));
706
707 op = mcs.args;
708 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
709 op->arg1.mfn = mfn;
710
711 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
712
713 if (kernel) {
714 x86_write_percpu(xen_cr3, cr3);
715
716 /* Update xen_current_cr3 once the batch has actually
717 been submitted. */
718 xen_mc_callback(set_current_cr3, (void *)cr3);
719 }
720}
721
722static void xen_write_cr3(unsigned long cr3)
723{
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700724 BUG_ON(preemptible());
725
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700726 xen_mc_batch(); /* disables interrupts */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700727
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -0700728 /* Update while interrupts are disabled, so its atomic with
729 respect to ipis */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700730 x86_write_percpu(xen_cr3, cr3);
731
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700732 __xen_write_cr3(true, cr3);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700733
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700734#ifdef CONFIG_X86_64
735 {
736 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
737 if (user_pgd)
738 __xen_write_cr3(false, __pa(user_pgd));
739 else
740 __xen_write_cr3(false, 0);
741 }
742#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700743
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -0700744 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700745}
746
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -0700747static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
748{
749 int ret;
750
751 ret = 0;
752
753 switch(msr) {
754#ifdef CONFIG_X86_64
755 unsigned which;
756 u64 base;
757
758 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
759 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
760 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
761
762 set:
763 base = ((u64)high << 32) | low;
764 if (HYPERVISOR_set_segment_base(which, base) != 0)
765 ret = -EFAULT;
766 break;
767#endif
Jeremy Fitzhardinged89961e2008-07-24 13:48:58 -0700768
769 case MSR_STAR:
770 case MSR_CSTAR:
771 case MSR_LSTAR:
772 case MSR_SYSCALL_MASK:
773 case MSR_IA32_SYSENTER_CS:
774 case MSR_IA32_SYSENTER_ESP:
775 case MSR_IA32_SYSENTER_EIP:
776 /* Fast syscall setup is all done in hypercalls, so
777 these are all ignored. Stub them out here to stop
778 Xen console noise. */
779 break;
780
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -0700781 default:
782 ret = native_write_msr_safe(msr, low, high);
783 }
784
785 return ret;
786}
787
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700788/* Early in boot, while setting up the initial pagetable, assume
789 everything is pinned. */
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700790static __init void xen_alloc_pte_init(struct mm_struct *mm, u32 pfn)
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700791{
Jeremy Fitzhardingeaf7ae3b2008-04-02 10:54:12 -0700792#ifdef CONFIG_FLATMEM
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700793 BUG_ON(mem_map); /* should only be used early */
Jeremy Fitzhardingeaf7ae3b2008-04-02 10:54:12 -0700794#endif
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700795 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
796}
797
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700798/* Early release_pte assumes that all pts are pinned, since there's
Jeremy Fitzhardinge1c70e9b2008-01-30 13:33:39 +0100799 only init_mm and anything attached to that is pinned. */
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700800static void xen_release_pte_init(u32 pfn)
Jeremy Fitzhardinge1c70e9b2008-01-30 13:33:39 +0100801{
802 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
803}
804
Mark McLoughlinf6433702008-04-02 15:36:36 +0100805static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
Jeremy Fitzhardinge74260712007-10-16 11:51:30 -0700806{
807 struct mmuext_op op;
Mark McLoughlinf6433702008-04-02 15:36:36 +0100808 op.cmd = cmd;
Jeremy Fitzhardinge74260712007-10-16 11:51:30 -0700809 op.arg1.mfn = pfn_to_mfn(pfn);
810 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
811 BUG();
812}
813
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700814/* This needs to make sure the new pte page is pinned iff its being
815 attached to a pinned pagetable. */
Jeremy Fitzhardinge1c70e9b2008-01-30 13:33:39 +0100816static void xen_alloc_ptpage(struct mm_struct *mm, u32 pfn, unsigned level)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700817{
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700818 struct page *page = pfn_to_page(pfn);
819
820 if (PagePinned(virt_to_page(mm->pgd))) {
821 SetPagePinned(page);
822
Jeremy Fitzhardinge74260712007-10-16 11:51:30 -0700823 if (!PageHighMem(page)) {
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700824 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
Mark McLoughlinf6433702008-04-02 15:36:36 +0100825 if (level == PT_PTE)
826 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
Jeremy Fitzhardinge74260712007-10-16 11:51:30 -0700827 } else
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700828 /* make sure there are no stray mappings of
829 this page */
830 kmap_flush_unused();
831 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700832}
833
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700834static void xen_alloc_pte(struct mm_struct *mm, u32 pfn)
Jeremy Fitzhardinge1c70e9b2008-01-30 13:33:39 +0100835{
Mark McLoughlinf6433702008-04-02 15:36:36 +0100836 xen_alloc_ptpage(mm, pfn, PT_PTE);
Jeremy Fitzhardinge1c70e9b2008-01-30 13:33:39 +0100837}
838
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700839static void xen_alloc_pmd(struct mm_struct *mm, u32 pfn)
Jeremy Fitzhardinge1c70e9b2008-01-30 13:33:39 +0100840{
Mark McLoughlinf6433702008-04-02 15:36:36 +0100841 xen_alloc_ptpage(mm, pfn, PT_PMD);
Jeremy Fitzhardinge1c70e9b2008-01-30 13:33:39 +0100842}
843
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700844static int xen_pgd_alloc(struct mm_struct *mm)
845{
846 pgd_t *pgd = mm->pgd;
847 int ret = 0;
848
849 BUG_ON(PagePinned(virt_to_page(pgd)));
850
851#ifdef CONFIG_X86_64
852 {
853 struct page *page = virt_to_page(pgd);
Jeremy Fitzhardingebf18bf92008-07-08 15:07:15 -0700854 pgd_t *user_pgd;
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700855
856 BUG_ON(page->private != 0);
857
Jeremy Fitzhardingebf18bf92008-07-08 15:07:15 -0700858 ret = -ENOMEM;
859
860 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
861 page->private = (unsigned long)user_pgd;
862
863 if (user_pgd != NULL) {
864 user_pgd[pgd_index(VSYSCALL_START)] =
865 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
866 ret = 0;
867 }
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -0700868
869 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
870 }
871#endif
872
873 return ret;
874}
875
876static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
877{
878#ifdef CONFIG_X86_64
879 pgd_t *user_pgd = xen_get_user_pgd(pgd);
880
881 if (user_pgd)
882 free_page((unsigned long)user_pgd);
883#endif
884}
885
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700886/* This should never happen until we're OK to use struct page */
Mark McLoughlinf6433702008-04-02 15:36:36 +0100887static void xen_release_ptpage(u32 pfn, unsigned level)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700888{
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700889 struct page *page = pfn_to_page(pfn);
890
891 if (PagePinned(page)) {
Jeremy Fitzhardinge74260712007-10-16 11:51:30 -0700892 if (!PageHighMem(page)) {
Mark McLoughlina684d692008-04-02 15:36:37 +0100893 if (level == PT_PTE)
894 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700895 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
Jeremy Fitzhardinge74260712007-10-16 11:51:30 -0700896 }
Mark McLoughlinc946c7d2008-04-02 15:36:38 +0100897 ClearPagePinned(page);
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700898 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700899}
900
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700901static void xen_release_pte(u32 pfn)
Mark McLoughlinf6433702008-04-02 15:36:36 +0100902{
903 xen_release_ptpage(pfn, PT_PTE);
904}
905
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700906static void xen_release_pmd(u32 pfn)
Mark McLoughlinf6433702008-04-02 15:36:36 +0100907{
908 xen_release_ptpage(pfn, PT_PMD);
909}
910
Jeremy Fitzhardingef6e58732008-07-08 15:06:38 -0700911#if PAGETABLE_LEVELS == 4
912static void xen_alloc_pud(struct mm_struct *mm, u32 pfn)
913{
914 xen_alloc_ptpage(mm, pfn, PT_PUD);
915}
916
917static void xen_release_pud(u32 pfn)
918{
919 xen_release_ptpage(pfn, PT_PUD);
920}
921#endif
922
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700923#ifdef CONFIG_HIGHPTE
924static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700925{
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700926 pgprot_t prot = PAGE_KERNEL;
927
928 if (PagePinned(page))
929 prot = PAGE_KERNEL_RO;
930
931 if (0 && PageHighMem(page))
932 printk("mapping highpte %lx type %d prot %s\n",
933 page_to_pfn(page), type,
934 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
935
936 return kmap_atomic_prot(page, type, prot);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700937}
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700938#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700939
Jeremy Fitzhardinge9a4029f2007-07-17 18:37:05 -0700940static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
941{
942 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
943 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
944 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
945 pte_val_ma(pte));
946
947 return pte;
948}
949
950/* Init-time set_pte while constructing initial pagetables, which
951 doesn't allow RO pagetable pages to be remapped RW */
952static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
953{
954 pte = mask_rw_pte(ptep, pte);
955
956 xen_set_pte(ptep, pte);
957}
958
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700959static __init void xen_pagetable_setup_start(pgd_t *base)
960{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700961}
962
Jeremy Fitzhardinge0e913982008-05-26 23:31:27 +0100963void xen_setup_shared_info(void)
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -0700964{
965 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -0700966 set_fixmap(FIX_PARAVIRT_BOOTMAP,
967 xen_start_info->shared_info);
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -0700968
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -0700969 HYPERVISOR_shared_info =
970 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -0700971 } else
972 HYPERVISOR_shared_info =
973 (struct shared_info *)__va(xen_start_info->shared_info);
974
975#ifndef CONFIG_SMP
976 /* In UP this is as good a place as any to set up shared info */
977 xen_setup_vcpu_info_placement();
978#endif
Jeremy Fitzhardinged5edbc12008-05-26 23:31:22 +0100979
980 xen_setup_mfn_list_list();
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -0700981}
982
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700983static __init void xen_pagetable_setup_done(pgd_t *base)
984{
Jeremy Fitzhardinge8745f8b2008-07-08 15:06:57 -0700985 xen_setup_shared_info();
986}
987
988static __init void xen_post_allocator_init(void)
989{
990 pv_mmu_ops.set_pte = xen_set_pte;
991 pv_mmu_ops.set_pmd = xen_set_pmd;
992 pv_mmu_ops.set_pud = xen_set_pud;
993#if PAGETABLE_LEVELS == 4
994 pv_mmu_ops.set_pgd = xen_set_pgd;
995#endif
996
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -0700997 /* This will work as long as patching hasn't happened yet
998 (which it hasn't) */
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700999 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1000 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1001 pv_mmu_ops.release_pte = xen_release_pte;
1002 pv_mmu_ops.release_pmd = xen_release_pmd;
Jeremy Fitzhardingef6e58732008-07-08 15:06:38 -07001003#if PAGETABLE_LEVELS == 4
1004 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1005 pv_mmu_ops.release_pud = xen_release_pud;
1006#endif
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -07001007
Jeremy Fitzhardingebf18bf92008-07-08 15:07:15 -07001008#ifdef CONFIG_X86_64
1009 SetPagePinned(virt_to_page(level3_user_vsyscall));
1010#endif
Jeremy Fitzhardingee2426cf2008-05-31 01:24:27 +01001011 xen_mark_init_mm_pinned();
1012}
1013
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001014/* This is called once we have the cpu_possible_map */
Jeremy Fitzhardinge0e913982008-05-26 23:31:27 +01001015void xen_setup_vcpu_info_placement(void)
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001016{
1017 int cpu;
1018
1019 for_each_possible_cpu(cpu)
1020 xen_vcpu_setup(cpu);
1021
1022 /* xen_vcpu_setup managed to place the vcpu_info within the
1023 percpu area for all cpus, so make use of it */
Jeremy Fitzhardinge5b09b282008-07-08 15:06:42 -07001024#ifdef CONFIG_X86_32
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001025 if (have_vcpu_info_placement) {
1026 printk(KERN_INFO "Xen: using vcpu_info placement\n");
1027
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001028 pv_irq_ops.save_fl = xen_save_fl_direct;
1029 pv_irq_ops.restore_fl = xen_restore_fl_direct;
1030 pv_irq_ops.irq_disable = xen_irq_disable_direct;
1031 pv_irq_ops.irq_enable = xen_irq_enable_direct;
1032 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001033 }
Jeremy Fitzhardinge5b09b282008-07-08 15:06:42 -07001034#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001035}
1036
Andi Kleenab144f52007-08-10 22:31:03 +02001037static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1038 unsigned long addr, unsigned len)
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001039{
1040 char *start, *end, *reloc;
1041 unsigned ret;
1042
1043 start = end = reloc = NULL;
1044
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001045#define SITE(op, x) \
1046 case PARAVIRT_PATCH(op.x): \
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001047 if (have_vcpu_info_placement) { \
1048 start = (char *)xen_##x##_direct; \
1049 end = xen_##x##_direct_end; \
1050 reloc = xen_##x##_direct_reloc; \
1051 } \
1052 goto patch_site
1053
1054 switch (type) {
Jeremy Fitzhardinge5b09b282008-07-08 15:06:42 -07001055#ifdef CONFIG_X86_32
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001056 SITE(pv_irq_ops, irq_enable);
1057 SITE(pv_irq_ops, irq_disable);
1058 SITE(pv_irq_ops, save_fl);
1059 SITE(pv_irq_ops, restore_fl);
Jeremy Fitzhardinge5b09b282008-07-08 15:06:42 -07001060#endif /* CONFIG_X86_32 */
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001061#undef SITE
1062
1063 patch_site:
1064 if (start == NULL || (end-start) > len)
1065 goto default_patch;
1066
Andi Kleenab144f52007-08-10 22:31:03 +02001067 ret = paravirt_patch_insns(insnbuf, len, start, end);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001068
1069 /* Note: because reloc is assigned from something that
1070 appears to be an array, gcc assumes it's non-null,
1071 but doesn't know its relationship with start and
1072 end. */
1073 if (reloc > start && reloc < end) {
1074 int reloc_off = reloc - start;
Andi Kleenab144f52007-08-10 22:31:03 +02001075 long *relocp = (long *)(insnbuf + reloc_off);
1076 long delta = start - (char *)addr;
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001077
1078 *relocp += delta;
1079 }
1080 break;
1081
1082 default_patch:
1083 default:
Andi Kleenab144f52007-08-10 22:31:03 +02001084 ret = paravirt_patch_default(type, clobbers, insnbuf,
1085 addr, len);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001086 break;
1087 }
1088
1089 return ret;
1090}
1091
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -07001092static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot)
1093{
1094 pte_t pte;
1095
1096 phys >>= PAGE_SHIFT;
1097
1098 switch (idx) {
1099 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1100#ifdef CONFIG_X86_F00F_BUG
1101 case FIX_F00F_IDT:
1102#endif
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -07001103#ifdef CONFIG_X86_32
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -07001104 case FIX_WP_TEST:
1105 case FIX_VDSO:
Ingo Molnarb3fe1242008-07-09 13:45:33 +02001106# ifdef CONFIG_HIGHMEM
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -07001107 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
Ingo Molnarb3fe1242008-07-09 13:45:33 +02001108# endif
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -07001109#else
1110 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1111#endif
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -07001112#ifdef CONFIG_X86_LOCAL_APIC
1113 case FIX_APIC_BASE: /* maps dummy local APIC */
1114#endif
1115 pte = pfn_pte(phys, prot);
1116 break;
1117
1118 default:
1119 pte = mfn_pte(phys, prot);
1120 break;
1121 }
1122
1123 __native_set_fixmap(idx, pte);
Jeremy Fitzhardingebf18bf92008-07-08 15:07:15 -07001124
1125#ifdef CONFIG_X86_64
1126 /* Replicate changes to map the vsyscall page into the user
1127 pagetable vsyscall mapping. */
1128 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1129 unsigned long vaddr = __fix_to_virt(idx);
1130 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1131 }
1132#endif
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -07001133}
1134
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001135static const struct pv_info xen_info __initdata = {
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001136 .paravirt_enabled = 1,
1137 .shared_kernel_pmd = 0,
1138
1139 .name = "Xen",
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001140};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001141
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001142static const struct pv_init_ops xen_init_ops __initdata = {
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001143 .patch = xen_patch,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001144
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001145 .banner = xen_banner,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001146 .memory_setup = xen_memory_setup,
1147 .arch_setup = xen_arch_setup,
Jeremy Fitzhardingee2426cf2008-05-31 01:24:27 +01001148 .post_allocator_init = xen_post_allocator_init,
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001149};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001150
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001151static const struct pv_time_ops xen_time_ops __initdata = {
Jeremy Fitzhardinge15c84732007-07-17 18:37:05 -07001152 .time_init = xen_time_init,
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001153
Jeremy Fitzhardinge15c84732007-07-17 18:37:05 -07001154 .set_wallclock = xen_set_wallclock,
1155 .get_wallclock = xen_get_wallclock,
Alok Katariae93ef942008-07-01 11:43:36 -07001156 .get_tsc_khz = xen_tsc_khz,
Jeremy Fitzhardingeab550282007-07-17 18:37:05 -07001157 .sched_clock = xen_sched_clock,
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001158};
Jeremy Fitzhardinge15c84732007-07-17 18:37:05 -07001159
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001160static const struct pv_cpu_ops xen_cpu_ops __initdata = {
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001161 .cpuid = xen_cpuid,
1162
1163 .set_debugreg = xen_set_debugreg,
1164 .get_debugreg = xen_get_debugreg,
1165
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001166 .clts = xen_clts,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001167
1168 .read_cr0 = native_read_cr0,
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001169 .write_cr0 = xen_write_cr0,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001170
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001171 .read_cr4 = native_read_cr4,
1172 .read_cr4_safe = native_read_cr4_safe,
1173 .write_cr4 = xen_write_cr4,
1174
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001175 .wbinvd = native_wbinvd,
1176
1177 .read_msr = native_read_msr_safe,
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001178 .write_msr = xen_write_msr_safe,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001179 .read_tsc = native_read_tsc,
1180 .read_pmc = native_read_pmc,
1181
Jeremy Fitzhardinge81e103f2008-04-17 17:40:51 +02001182 .iret = xen_iret,
Jeremy Fitzhardinged75cd222008-06-25 00:19:26 -04001183 .irq_enable_sysexit = xen_sysexit,
Jeremy Fitzhardinge6fcac6d2008-07-08 15:07:14 -07001184#ifdef CONFIG_X86_64
1185 .usergs_sysret32 = xen_sysret32,
1186 .usergs_sysret64 = xen_sysret64,
1187#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001188
1189 .load_tr_desc = paravirt_nop,
1190 .set_ldt = xen_set_ldt,
1191 .load_gdt = xen_load_gdt,
1192 .load_idt = xen_load_idt,
1193 .load_tls = xen_load_tls,
Eduardo Habkosta8fc1082008-07-08 15:07:05 -07001194#ifdef CONFIG_X86_64
1195 .load_gs_index = xen_load_gs_index,
1196#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001197
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -07001198 .alloc_ldt = xen_alloc_ldt,
1199 .free_ldt = xen_free_ldt,
1200
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001201 .store_gdt = native_store_gdt,
1202 .store_idt = native_store_idt,
1203 .store_tr = xen_store_tr,
1204
1205 .write_ldt_entry = xen_write_ldt_entry,
1206 .write_gdt_entry = xen_write_gdt_entry,
1207 .write_idt_entry = xen_write_idt_entry,
H. Peter Anvinfaca6222008-01-30 13:31:02 +01001208 .load_sp0 = xen_load_sp0,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001209
1210 .set_iopl_mask = xen_set_iopl_mask,
1211 .io_delay = xen_io_delay,
1212
Jeremy Fitzhardinge952d1d72008-07-08 15:07:01 -07001213 /* Xen takes care of %gs when switching to usermode for us */
1214 .swapgs = paravirt_nop,
1215
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -07001216 .lazy_mode = {
1217 .enter = paravirt_enter_lazy_cpu,
1218 .leave = xen_leave_lazy,
1219 },
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001220};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001221
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001222static const struct pv_apic_ops xen_apic_ops __initdata = {
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001223#ifdef CONFIG_X86_LOCAL_APIC
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -07001224 .apic_write = xen_apic_write,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001225 .apic_read = xen_apic_read,
1226 .setup_boot_clock = paravirt_nop,
1227 .setup_secondary_clock = paravirt_nop,
1228 .startup_ipi_hook = paravirt_nop,
1229#endif
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001230};
1231
1232static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1233 .pagetable_setup_start = xen_pagetable_setup_start,
1234 .pagetable_setup_done = xen_pagetable_setup_done,
1235
1236 .read_cr2 = xen_read_cr2,
1237 .write_cr2 = xen_write_cr2,
1238
1239 .read_cr3 = xen_read_cr3,
1240 .write_cr3 = xen_write_cr3,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001241
1242 .flush_tlb_user = xen_flush_tlb,
1243 .flush_tlb_kernel = xen_flush_tlb,
1244 .flush_tlb_single = xen_flush_tlb_single,
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -07001245 .flush_tlb_others = xen_flush_tlb_others,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001246
1247 .pte_update = paravirt_nop,
1248 .pte_update_defer = paravirt_nop,
1249
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -07001250 .pgd_alloc = xen_pgd_alloc,
1251 .pgd_free = xen_pgd_free,
Jeremy Fitzhardingeeba00452008-06-25 00:19:12 -04001252
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -07001253 .alloc_pte = xen_alloc_pte_init,
1254 .release_pte = xen_release_pte_init,
1255 .alloc_pmd = xen_alloc_pte_init,
1256 .alloc_pmd_clone = paravirt_nop,
1257 .release_pmd = xen_release_pte_init,
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -07001258
1259#ifdef CONFIG_HIGHPTE
1260 .kmap_atomic_pte = xen_kmap_atomic_pte,
1261#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001262
Jeremy Fitzhardinge22911b32008-07-08 15:06:51 -07001263#ifdef CONFIG_X86_64
1264 .set_pte = xen_set_pte,
1265#else
Jeremy Fitzhardinge851fa3c2008-07-08 15:06:33 -07001266 .set_pte = xen_set_pte_init,
Jeremy Fitzhardinge22911b32008-07-08 15:06:51 -07001267#endif
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001268 .set_pte_at = xen_set_pte_at,
Jeremy Fitzhardingee2426cf2008-05-31 01:24:27 +01001269 .set_pmd = xen_set_pmd_hyper,
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001270
Jeremy Fitzhardinge08b882c2008-06-16 04:30:01 -07001271 .ptep_modify_prot_start = __ptep_modify_prot_start,
1272 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1273
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001274 .pte_val = xen_pte_val,
Eduardo Habkostb56afe12008-07-24 12:15:45 -03001275 .pte_flags = native_pte_flags,
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001276 .pgd_val = xen_pgd_val,
1277
1278 .make_pte = xen_make_pte,
1279 .make_pgd = xen_make_pgd,
1280
Jeremy Fitzhardingef6e58732008-07-08 15:06:38 -07001281#ifdef CONFIG_X86_PAE
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001282 .set_pte_atomic = xen_set_pte_atomic,
1283 .set_pte_present = xen_set_pte_at,
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001284 .pte_clear = xen_pte_clear,
1285 .pmd_clear = xen_pmd_clear,
Jeremy Fitzhardingef6e58732008-07-08 15:06:38 -07001286#endif /* CONFIG_X86_PAE */
1287 .set_pud = xen_set_pud_hyper,
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001288
1289 .make_pmd = xen_make_pmd,
1290 .pmd_val = xen_pmd_val,
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001291
Jeremy Fitzhardingef6e58732008-07-08 15:06:38 -07001292#if PAGETABLE_LEVELS == 4
1293 .pud_val = xen_pud_val,
1294 .make_pud = xen_make_pud,
1295 .set_pgd = xen_set_pgd_hyper,
1296
1297 .alloc_pud = xen_alloc_pte_init,
1298 .release_pud = xen_release_pte_init,
1299#endif /* PAGETABLE_LEVELS == 4 */
1300
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -07001301 .activate_mm = xen_activate_mm,
1302 .dup_mmap = xen_dup_mmap,
1303 .exit_mmap = xen_exit_mmap,
1304
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -07001305 .lazy_mode = {
1306 .enter = paravirt_enter_lazy_mmu,
1307 .leave = xen_leave_lazy,
1308 },
Jeremy Fitzhardingeaeaaa592008-06-17 11:42:01 -07001309
1310 .set_fixmap = xen_set_fixmap,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001311};
1312
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001313static void xen_reboot(int reason)
1314{
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001315 struct sched_shutdown r = { .reason = reason };
1316
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001317#ifdef CONFIG_SMP
1318 smp_send_stop();
1319#endif
1320
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001321 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001322 BUG();
1323}
1324
1325static void xen_restart(char *msg)
1326{
1327 xen_reboot(SHUTDOWN_reboot);
1328}
1329
1330static void xen_emergency_restart(void)
1331{
1332 xen_reboot(SHUTDOWN_reboot);
1333}
1334
1335static void xen_machine_halt(void)
1336{
1337 xen_reboot(SHUTDOWN_poweroff);
1338}
1339
1340static void xen_crash_shutdown(struct pt_regs *regs)
1341{
1342 xen_reboot(SHUTDOWN_crash);
1343}
1344
1345static const struct machine_ops __initdata xen_machine_ops = {
1346 .restart = xen_restart,
1347 .halt = xen_machine_halt,
1348 .power_off = xen_machine_halt,
1349 .shutdown = xen_machine_halt,
1350 .crash_shutdown = xen_crash_shutdown,
1351 .emergency_restart = xen_emergency_restart,
1352};
1353
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001354
Jeremy Fitzhardingefb1d8402007-10-16 11:51:31 -07001355static void __init xen_reserve_top(void)
1356{
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001357#ifdef CONFIG_X86_32
Jeremy Fitzhardingefb1d8402007-10-16 11:51:31 -07001358 unsigned long top = HYPERVISOR_VIRT_START;
1359 struct xen_platform_parameters pp;
1360
1361 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1362 top = pp.virt_start;
1363
1364 reserve_top_address(-top + 2 * PAGE_SIZE);
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001365#endif /* CONFIG_X86_32 */
Jeremy Fitzhardingefb1d8402007-10-16 11:51:31 -07001366}
1367
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001368/*
1369 * Like __va(), but returns address in the kernel mapping (which is
1370 * all we have until the physical memory mapping has been set up.
1371 */
1372static void *__ka(phys_addr_t paddr)
1373{
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001374#ifdef CONFIG_X86_64
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001375 return (void *)(paddr + __START_KERNEL_map);
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001376#else
1377 return __va(paddr);
1378#endif
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001379}
1380
1381/* Convert a machine address to physical address */
1382static unsigned long m2p(phys_addr_t maddr)
1383{
1384 phys_addr_t paddr;
1385
Jeremy Fitzhardinge59438c92008-07-21 22:59:42 -07001386 maddr &= PTE_PFN_MASK;
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001387 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1388
1389 return paddr;
1390}
1391
1392/* Convert a machine address to kernel virtual */
1393static void *m2v(phys_addr_t maddr)
1394{
1395 return __ka(m2p(maddr));
1396}
1397
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001398#ifdef CONFIG_X86_64
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001399static void walk(pgd_t *pgd, unsigned long addr)
1400{
1401 unsigned l4idx = pgd_index(addr);
1402 unsigned l3idx = pud_index(addr);
1403 unsigned l2idx = pmd_index(addr);
1404 unsigned l1idx = pte_index(addr);
1405 pgd_t l4;
1406 pud_t l3;
1407 pmd_t l2;
1408 pte_t l1;
1409
1410 xen_raw_printk("walk %p, %lx -> %d %d %d %d\n",
1411 pgd, addr, l4idx, l3idx, l2idx, l1idx);
1412
1413 l4 = pgd[l4idx];
1414 xen_raw_printk(" l4: %016lx\n", l4.pgd);
1415 xen_raw_printk(" %016lx\n", pgd_val(l4));
1416
1417 l3 = ((pud_t *)(m2v(l4.pgd)))[l3idx];
1418 xen_raw_printk(" l3: %016lx\n", l3.pud);
1419 xen_raw_printk(" %016lx\n", pud_val(l3));
1420
1421 l2 = ((pmd_t *)(m2v(l3.pud)))[l2idx];
1422 xen_raw_printk(" l2: %016lx\n", l2.pmd);
1423 xen_raw_printk(" %016lx\n", pmd_val(l2));
1424
1425 l1 = ((pte_t *)(m2v(l2.pmd)))[l1idx];
1426 xen_raw_printk(" l1: %016lx\n", l1.pte);
1427 xen_raw_printk(" %016lx\n", pte_val(l1));
1428}
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001429#endif
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001430
1431static void set_page_prot(void *addr, pgprot_t prot)
1432{
1433 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1434 pte_t pte = pfn_pte(pfn, prot);
1435
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001436 xen_raw_printk("addr=%p pfn=%lx mfn=%lx prot=%016llx pte=%016llx\n",
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001437 addr, pfn, get_phys_to_machine(pfn),
1438 pgprot_val(prot), pte.pte);
1439
1440 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1441 BUG();
1442}
1443
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001444static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001445{
1446 unsigned pmdidx, pteidx;
1447 unsigned ident_pte;
1448 unsigned long pfn;
1449
1450 ident_pte = 0;
1451 pfn = 0;
1452 for(pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1453 pte_t *pte_page;
1454
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001455 /* Reuse or allocate a page of ptes */
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001456 if (pmd_present(pmd[pmdidx]))
1457 pte_page = m2v(pmd[pmdidx].pmd);
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001458 else {
1459 /* Check for free pte pages */
1460 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1461 break;
1462
1463 pte_page = &level1_ident_pgt[ident_pte];
1464 ident_pte += PTRS_PER_PTE;
1465
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001466 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001467 }
1468
1469 /* Install mappings */
1470 for(pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1471 pte_t pte;
1472
1473 if (pfn > max_pfn_mapped)
1474 max_pfn_mapped = pfn;
1475
1476 if (!pte_none(pte_page[pteidx]))
1477 continue;
1478
1479 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1480 pte_page[pteidx] = pte;
1481 }
1482 }
1483
1484 for(pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1485 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001486
1487 set_page_prot(pmd, PAGE_KERNEL_RO);
1488}
1489
1490#ifdef CONFIG_X86_64
1491static void convert_pfn_mfn(void *v)
1492{
1493 pte_t *pte = v;
1494 int i;
1495
1496 /* All levels are converted the same way, so just treat them
1497 as ptes. */
1498 for(i = 0; i < PTRS_PER_PTE; i++)
1499 pte[i] = xen_make_pte(pte[i].pte);
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001500}
1501
1502/*
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001503 * Set up the inital kernel pagetable.
1504 *
1505 * We can construct this by grafting the Xen provided pagetable into
1506 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1507 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1508 * means that only the kernel has a physical mapping to start with -
1509 * but that's enough to get __va working. We need to fill in the rest
1510 * of the physical mapping once some sort of allocator has been set
1511 * up.
1512 */
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001513static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001514{
1515 pud_t *l3;
1516 pmd_t *l2;
1517
1518 /* Zap identity mapping */
1519 init_level4_pgt[0] = __pgd(0);
1520
1521 /* Pre-constructed entries are in pfn, so convert to mfn */
1522 convert_pfn_mfn(init_level4_pgt);
1523 convert_pfn_mfn(level3_ident_pgt);
1524 convert_pfn_mfn(level3_kernel_pgt);
1525
1526 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1527 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1528
1529 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1530 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1531
1532 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1533 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1534 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1535
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001536 /* Set up identity map */
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001537 xen_map_identity_early(level2_ident_pgt, max_pfn);
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001538
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001539 /* Make pagetable pieces RO */
1540 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1541 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1542 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
Jeremy Fitzhardingebf18bf92008-07-08 15:07:15 -07001543 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001544 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1545 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1546
1547 /* Pin down new L4 */
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001548 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1549 PFN_DOWN(__pa_symbol(init_level4_pgt)));
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001550
1551 /* Unpin Xen-provided one */
1552 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1553
1554 /* Switch over */
1555 pgd = init_level4_pgt;
Jeremy Fitzhardinged6182fb2008-07-08 15:07:13 -07001556
1557 /*
1558 * At this stage there can be no user pgd, and no page
1559 * structure to attach it to, so make sure we just set kernel
1560 * pgd.
1561 */
1562 xen_mc_batch();
1563 __xen_write_cr3(true, __pa(pgd));
1564 xen_mc_issue(PARAVIRT_LAZY_CPU);
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001565
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001566 reserve_early(__pa(xen_start_info->pt_base),
1567 __pa(xen_start_info->pt_base +
1568 xen_start_info->nr_pt_frames * PAGE_SIZE),
1569 "XEN PAGETABLES");
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001570
1571 return pgd;
1572}
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001573#else /* !CONFIG_X86_64 */
1574static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1575
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001576static __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001577{
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001578 pmd_t *kernel_pmd;
1579
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001580 init_pg_tables_start = __pa(pgd);
1581 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
1582 max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024);
1583
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001584 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1585 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001586
Jeremy Fitzhardinge39dbc5b2008-07-08 15:06:53 -07001587 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1588
1589 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1590 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1591 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1592
1593 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1594 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1595 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1596
1597 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1598
1599 xen_write_cr3(__pa(swapper_pg_dir));
1600
1601 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1602
1603 return swapper_pg_dir;
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001604}
1605#endif /* CONFIG_X86_64 */
1606
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001607/* First C function to be called on Xen boot */
1608asmlinkage void __init xen_start_kernel(void)
1609{
1610 pgd_t *pgd;
1611
1612 if (!xen_start_info)
1613 return;
1614
Jeremy Fitzhardinge7999f4b2007-12-10 13:00:41 -08001615 BUG_ON(memcmp(xen_start_info->magic, "xen-3", 5) != 0);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001616
Jeremy Fitzhardingee57778a2008-06-16 04:30:02 -07001617 xen_setup_features();
1618
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001619 /* Install Xen paravirt ops */
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001620 pv_info = xen_info;
1621 pv_init_ops = xen_init_ops;
1622 pv_time_ops = xen_time_ops;
1623 pv_cpu_ops = xen_cpu_ops;
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001624 pv_apic_ops = xen_apic_ops;
1625 pv_mmu_ops = xen_mmu_ops;
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001626
Jeremy Fitzhardinge0d1edf42008-07-28 11:53:57 -07001627 xen_init_irq_ops();
1628
Jeremy Fitzhardingee57778a2008-06-16 04:30:02 -07001629 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1630 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1631 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1632 }
1633
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001634 machine_ops = xen_machine_ops;
1635
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001636#ifdef CONFIG_X86_64
1637 /* Disable until direct per-cpu data access. */
1638 have_vcpu_info_placement = 0;
Jeremy Fitzhardinge5b09b282008-07-08 15:06:42 -07001639 x86_64_init_pda();
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -07001640#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001641
Jeremy Fitzhardingea9e70622008-07-08 15:06:41 -07001642 xen_smp_init();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001643
1644 /* Get mfn list */
1645 if (!xen_feature(XENFEAT_auto_translated_physmap))
Jeremy Fitzhardinged451bb72008-05-26 23:31:18 +01001646 xen_build_dynamic_phys_to_machine();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001647
1648 pgd = (pgd_t *)xen_start_info->pt_base;
1649
Jeremy Fitzhardingeeb179e42008-06-16 15:01:53 -07001650 /* Prevent unwanted bits from being set in PTEs. */
1651 __supported_pte_mask &= ~_PAGE_GLOBAL;
1652 if (!is_initial_xendomain())
1653 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1654
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001655 /* Don't do the full vcpu_info placement stuff until we have a
1656 possible map and a non-dummy shared_info. */
1657 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1658
1659 xen_raw_console_write("mapping kernel into physical memory\n");
Jeremy Fitzhardinged114e1982008-07-08 15:06:52 -07001660 pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001661
1662 init_mm.pgd = pgd;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001663
1664 /* keep using Xen gdt for now; no urgent need to change it */
1665
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001666 pv_info.kernel_rpl = 1;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001667 if (xen_feature(XENFEAT_supervisor_mode_kernel))
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001668 pv_info.kernel_rpl = 0;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001669
1670 /* set the limit of our address space */
Jeremy Fitzhardingefb1d8402007-10-16 11:51:31 -07001671 xen_reserve_top();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001672
Jeremy Fitzhardinge7d087b62008-07-08 15:06:48 -07001673#ifdef CONFIG_X86_32
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001674 /* set up basic CPUID stuff */
1675 cpu_detect(&new_cpu_data);
1676 new_cpu_data.hard_math = 1;
1677 new_cpu_data.x86_capability[0] = cpuid_edx(1);
Jeremy Fitzhardinge7d087b62008-07-08 15:06:48 -07001678#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001679
1680 /* Poke various useful things into boot_params */
H. Peter Anvin30c82642007-10-15 17:13:22 -07001681 boot_params.hdr.type_of_loader = (9 << 4) | 0;
1682 boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1683 ? __pa(xen_start_info->mod_start) : 0;
1684 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
Jeremy Fitzhardingeb7c3c5c2008-07-08 15:07:07 -07001685 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001686
Markus Armbruster9e124fe2008-05-26 23:31:07 +01001687 if (!is_initial_xendomain()) {
Jeremy Fitzhardinge83abc702008-05-26 23:31:12 +01001688 add_preferred_console("xenboot", 0, NULL);
Markus Armbruster9e124fe2008-05-26 23:31:07 +01001689 add_preferred_console("tty", 0, NULL);
Markus Armbrusterb8c2d3d2008-02-27 14:56:35 +01001690 add_preferred_console("hvc", 0, NULL);
Markus Armbruster9e124fe2008-05-26 23:31:07 +01001691 }
Markus Armbrusterb8c2d3d2008-02-27 14:56:35 +01001692
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001693 xen_raw_console_write("about to get started...\n");
1694
1695#if 0
1696 xen_raw_printk("&boot_params=%p __pa(&boot_params)=%lx __va(__pa(&boot_params))=%lx\n",
1697 &boot_params, __pa_symbol(&boot_params),
1698 __va(__pa_symbol(&boot_params)));
1699
1700 walk(pgd, &boot_params);
1701 walk(pgd, __va(__pa(&boot_params)));
1702#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001703
1704 /* Start the world */
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001705#ifdef CONFIG_X86_32
Yinghai Luf0d43102008-05-29 12:56:36 -07001706 i386_start_kernel();
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001707#else
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001708 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001709#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001710}