aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/lguest/boot.c
blob: 0188fd37b6c0936b4d6f5feb53f44a10b404abb9 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
/*P:010
 * A hypervisor allows multiple Operating Systems to run on a single machine.
 * To quote David Wheeler: "Any problem in computer science can be solved with
 * another layer of indirection."
 *
 * We keep things simple in two ways.  First, we start with a normal Linux
 * kernel and insert a module (lg.ko) which allows us to run other Linux
 * kernels the same way we'd run processes.  We call the first kernel the Host,
 * and the others the Guests.  The program which sets up and configures Guests
 * (such as the example in Documentation/lguest/lguest.c) is called the
 * Launcher.
 *
 * Secondly, we only run specially modified Guests, not normal kernels: setting
 * CONFIG_LGUEST_GUEST to "y" compiles this file into the kernel so it knows
 * how to be a Guest at boot time.  This means that you can use the same kernel
 * you boot normally (ie. as a Host) as a Guest.
 *
 * These Guests know that they cannot do privileged operations, such as disable
 * interrupts, and that they have to ask the Host to do such things explicitly.
 * This file consists of all the replacements for such low-level native
 * hardware operations: these special Guest versions call the Host.
 *
 * So how does the kernel know it's a Guest?  We'll see that later, but let's
 * just say that we end up here where we replace the native functions various
 * "paravirt" structures with our Guest versions, then boot like normal. :*/

/*
 * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 * NON INFRINGEMENT.  See the GNU General Public License for more
 * details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */
#include <linux/kernel.h>
#include <linux/start_kernel.h>
#include <linux/string.h>
#include <linux/console.h>
#include <linux/screen_info.h>
#include <linux/irq.h>
#include <linux/interrupt.h>
#include <linux/clocksource.h>
#include <linux/clockchips.h>
#include <linux/lguest.h>
#include <linux/lguest_launcher.h>
#include <linux/virtio_console.h>
#include <linux/pm.h>
#include <asm/apic.h>
#include <asm/lguest.h>
#include <asm/paravirt.h>
#include <asm/param.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/desc.h>
#include <asm/setup.h>
#include <asm/e820.h>
#include <asm/mce.h>
#include <asm/io.h>
#include <asm/i387.h>
#include <asm/stackprotector.h>
#include <asm/reboot.h>		/* for struct machine_ops */

/*G:010 Welcome to the Guest!
 *
 * The Guest in our tale is a simple creature: identical to the Host but
 * behaving in simplified but equivalent ways.  In particular, the Guest is the
 * same kernel as the Host (or at least, built from the same source code). :*/

struct lguest_data lguest_data = {
	.hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF },
	.noirq_start = (u32)lguest_noirq_start,
	.noirq_end = (u32)lguest_noirq_end,
	.kernel_address = PAGE_OFFSET,
	.blocked_interrupts = { 1 }, /* Block timer interrupts */
	.syscall_vec = SYSCALL_VECTOR,
};

/*G:037 async_hcall() is pretty simple: I'm quite proud of it really.  We have a
 * ring buffer of stored hypercalls which the Host will run though next time we
 * do a normal hypercall.  Each entry in the ring has 5 slots for the hypercall
 * arguments, and a "hcall_status" word which is 0 if the call is ready to go,
 * and 255 once the Host has finished with it.
 *
 * If we come around to a slot which hasn't been finished, then the table is
 * full and we just make the hypercall directly.  This has the nice side
 * effect of causing the Host to run all the stored calls in the ring buffer
 * which empties it for next time! */
static void async_hcall(unsigned long call, unsigned long arg1,
			unsigned long arg2, unsigned long arg3,
			unsigned long arg4)
{
	/* Note: This code assumes we're uniprocessor. */
	static unsigned int next_call;
	unsigned long flags;

	/* Disable interrupts if not already disabled: we don't want an
	 * interrupt handler making a hypercall while we're already doing
	 * one! */
	local_irq_save(flags);
	if (lguest_data.hcall_status[next_call] != 0xFF) {
		/* Table full, so do normal hcall which will flush table. */
		kvm_hypercall4(call, arg1, arg2, arg3, arg4);
	} else {
		lguest_data.hcalls[next_call].arg0 = call;
		lguest_data.hcalls[next_call].arg1 = arg1;
		lguest_data.hcalls[next_call].arg2 = arg2;
		lguest_data.hcalls[next_call].arg3 = arg3;
		lguest_data.hcalls[next_call].arg4 = arg4;
		/* Arguments must all be written before we mark it to go */
		wmb();
		lguest_data.hcall_status[next_call] = 0;
		if (++next_call == LHCALL_RING_SIZE)
			next_call = 0;
	}
	local_irq_restore(flags);
}

/*G:035 Notice the lazy_hcall() above, rather than hcall().  This is our first
 * real optimization trick!
 *
 * When lazy_mode is set, it means we're allowed to defer all hypercalls and do
 * them as a batch when lazy_mode is eventually turned off.  Because hypercalls
 * are reasonably expensive, batching them up makes sense.  For example, a
 * large munmap might update dozens of page table entries: that code calls
 * paravirt_enter_lazy_mmu(), does the dozen updates, then calls
 * lguest_leave_lazy_mode().
 *
 * So, when we're in lazy mode, we call async_hcall() to store the call for
 * future processing: */
static void lazy_hcall1(unsigned long call,
		       unsigned long arg1)
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
		kvm_hypercall1(call, arg1);
	else
		async_hcall(call, arg1, 0, 0, 0);
}

static void lazy_hcall2(unsigned long call,
		       unsigned long arg1,
		       unsigned long arg2)
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
		kvm_hypercall2(call, arg1, arg2);
	else
		async_hcall(call, arg1, arg2, 0, 0);
}

static void lazy_hcall3(unsigned long call,
		       unsigned long arg1,
		       unsigned long arg2,
		       unsigned long arg3)
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
		kvm_hypercall3(call, arg1, arg2, arg3);
	else
		async_hcall(call, arg1, arg2, arg3, 0);
}

#ifdef CONFIG_X86_PAE
static void lazy_hcall4(unsigned long call,
		       unsigned long arg1,
		       unsigned long arg2,
		       unsigned long arg3,
		       unsigned long arg4)
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
		kvm_hypercall4(call, arg1, arg2, arg3, arg4);
	else
		async_hcall(call, arg1, arg2, arg3, arg4);
}
#endif

/* When lazy mode is turned off reset the per-cpu lazy mode variable and then
 * issue the do-nothing hypercall to flush any stored calls. */
static void lguest_leave_lazy_mmu_mode(void)
{
	kvm_hypercall0(LHCALL_FLUSH_ASYNC);
	paravirt_leave_lazy_mmu();
}

static void lguest_end_context_switch(struct task_struct *next)
{
	kvm_hypercall0(LHCALL_FLUSH_ASYNC);
	paravirt_end_context_switch(next);
}

/*G:032
 * After that diversion we return to our first native-instruction
 * replacements: four functions for interrupt control.
 *
 * The simplest way of implementing these would be to have "turn interrupts
 * off" and "turn interrupts on" hypercalls.  Unfortunately, this is too slow:
 * these are by far the most commonly called functions of those we override.
 *
 * So instead we keep an "irq_enabled" field inside our "struct lguest_data",
 * which the Guest can update with a single instruction.  The Host knows to
 * check there before it tries to deliver an interrupt.
 */

/* save_flags() is expected to return the processor state (ie. "flags").  The
 * flags word contains all kind of stuff, but in practice Linux only cares
 * about the interrupt flag.  Our "save_flags()" just returns that. */
static unsigned long save_fl(void)
{
	return lguest_data.irq_enabled;
}

/* Interrupts go off... */
static void irq_disable(void)
{
	lguest_data.irq_enabled = 0;
}

/* Let's pause a moment.  Remember how I said these are called so often?
 * Jeremy Fitzhardinge optimized them so hard early in 2009 that he had to
 * break some rules.  In particular, these functions are assumed to save their
 * own registers if they need to: normal C functions assume they can trash the
 * eax register.  To use normal C functions, we use
 * PV_CALLEE_SAVE_REGS_THUNK(), which pushes %eax onto the stack, calls the
 * C function, then restores it. */
PV_CALLEE_SAVE_REGS_THUNK(save_fl);
PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
/*:*/

/* These are in i386_head.S */
extern void lg_irq_enable(void);
extern void lg_restore_fl(unsigned long flags);

/*M:003 Note that we don't check for outstanding interrupts when we re-enable
 * them (or when we unmask an interrupt).  This seems to work for the moment,
 * since interrupts are rare and we'll just get the interrupt on the next timer
 * tick, but now we can run with CONFIG_NO_HZ, we should revisit this.  One way
 * would be to put the "irq_enabled" field in a page by itself, and have the
 * Host write-protect it when an interrupt comes in when irqs are disabled.
 * There will then be a page fault as soon as interrupts are re-enabled.
 *
 * A better method is to implement soft interrupt disable generally for x86:
 * instead of disabling interrupts, we set a flag.  If an interrupt does come
 * in, we then disable them for real.  This is uncommon, so we could simply use
 * a hypercall for interrupt control and not worry about efficiency. :*/

/*G:034
 * The Interrupt Descriptor Table (IDT).
 *
 * The IDT tells the processor what to do when an interrupt comes in.  Each
 * entry in the table is a 64-bit descriptor: this holds the privilege level,
 * address of the handler, and... well, who cares?  The Guest just asks the
 * Host to make the change anyway, because the Host controls the real IDT.
 */
static void lguest_write_idt_entry(gate_desc *dt,
				   int entrynum, const gate_desc *g)
{
	/* The gate_desc structure is 8 bytes long: we hand it to the Host in
	 * two 32-bit chunks.  The whole 32-bit kernel used to hand descriptors
	 * around like this; typesafety wasn't a big concern in Linux's early
	 * years. */
	u32 *desc = (u32 *)g;
	/* Keep the local copy up to date. */
	native_write_idt_entry(dt, entrynum, g);
	/* Tell Host about this new entry. */
	kvm_hypercall3(LHCALL_LOAD_IDT_ENTRY, entrynum, desc[0], desc[1]);
}

/* Changing to a different IDT is very rare: we keep the IDT up-to-date every
 * time it is written, so we can simply loop through all entries and tell the
 * Host about them. */
static void lguest_load_idt(const struct desc_ptr *desc)
{
	unsigned int i;
	struct desc_struct *idt = (void *)desc->address;

	for (i = 0; i < (desc->size+1)/8; i++)
		kvm_hypercall3(LHCALL_LOAD_IDT_ENTRY, i, idt[i].a, idt[i].b);
}

/*
 * The Global Descriptor Table.
 *
 * The Intel architecture defines another table, called the Global Descriptor
 * Table (GDT).  You tell the CPU where it is (and its size) using the "lgdt"
 * instruction, and then several other instructions refer to entries in the
 * table.  There are three entries which the Switcher needs, so the Host simply
 * controls the entire thing and the Guest asks it to make changes using the
 * LOAD_GDT hypercall.
 *
 * This is the exactly like the IDT code.
 */
static void lguest_load_gdt(const struct desc_ptr *desc)
{
	unsigned int i;
	struct desc_struct *gdt = (void *)desc->address;

	for (i = 0; i < (desc->size+1)/8; i++)
		kvm_hypercall3(LHCALL_LOAD_GDT_ENTRY, i, gdt[i].a, gdt[i].b);
}

/* For a single GDT entry which changes, we do the lazy thing: alter our GDT,
 * then tell the Host to reload the entire thing.  This operation is so rare
 * that this naive implementation is reasonable. */
static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum,
				   const void *desc, int type)
{
	native_write_gdt_entry(dt, entrynum, desc, type);
	/* Tell Host about this new entry. */
	kvm_hypercall3(LHCALL_LOAD_GDT_ENTRY, entrynum,
		       dt[entrynum].a, dt[entrynum].b);
}

/* OK, I lied.  There are three "thread local storage" GDT entries which change
 * on every context switch (these three entries are how glibc implements
 * __thread variables).  So we have a hypercall specifically for this case. */
static void lguest_load_tls(struct thread_struct *t, unsigned int cpu)
{
	/* There's one problem which normal hardware doesn't have: the Host
	 * can't handle us removing entries we're currently using.  So we clear
	 * the GS register here: if it's needed it'll be reloaded anyway. */
	lazy_load_gs(0);
	lazy_hcall2(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu);
}

/*G:038 That's enough excitement for now, back to ploughing through each of
 * the different pv_ops structures (we're about 1/3 of the way through).
 *
 * This is the Local Descriptor Table, another weird Intel thingy.  Linux only
 * uses this for some strange applications like Wine.  We don't do anything
 * here, so they'll get an informative and friendly Segmentation Fault. */
static void lguest_set_ldt(const void *addr, unsigned entries)
{
}

/* This loads a GDT entry into the "Task Register": that entry points to a
 * structure called the Task State Segment.  Some comments scattered though the
 * kernel code indicate that this used for task switching in ages past, along
 * with blood sacrifice and astrology.
 *
 * Now there's nothing interesting in here that we don't get told elsewhere.
 * But the native version uses the "ltr" instruction, which makes the Host
 * complain to the Guest about a Segmentation Fault and it'll oops.  So we
 * override the native version with a do-nothing version. */
static void lguest_load_tr_desc(void)
{
}

/* The "cpuid" instruction is a way of querying both the CPU identity
 * (manufacturer, model, etc) and its features.  It was introduced before the
 * Pentium in 1993 and keeps getting extended by both Intel, AMD and others.
 * As you might imagine, after a decade and a half this treatment, it is now a
 * giant ball of hair.  Its entry in the current Intel manual runs to 28 pages.
 *
 * This instruction even it has its own Wikipedia entry.  The Wikipedia entry
 * has been translated into 4 languages.  I am not making this up!
 *
 * We could get funky here and identify ourselves as "GenuineLguest", but
 * instead we just use the real "cpuid" instruction.  Then I pretty much turned
 * off feature bits until the Guest booted.  (Don't say that: you'll damage
 * lguest sales!)  Shut up, inner voice!  (Hey, just pointing out that this is
 * hardly future proof.)  Noone's listening!  They don't like you anyway,
 * parenthetic weirdo!
 *
 * Replacing the cpuid so we can turn features off is great for the kernel, but
 * anyone (including userspace) can just use the raw "cpuid" instruction and
 * the Host won't even notice since it isn't privileged.  So we try not to get
 * too worked up about it. */
static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
			 unsigned int *cx, unsigned int *dx)
{
	int function = *ax;

	native_cpuid(ax, bx, cx, dx);
	switch (function) {
	case 1:	/* Basic feature request. */
		/* We only allow kernel to see SSE3, CMPXCHG16B and SSSE3 */
		*cx &= 0x00002201;
		/* SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, TSC, FPU, PAE. */
		*dx &= 0x07808151;
		/* The Host can do a nice optimization if it knows that the
		 * kernel mappings (addresses above 0xC0000000 or whatever
		 * PAGE_OFFSET is set to) haven't changed.  But Linux calls
		 * flush_tlb_user() for both user and kernel mappings unless
		 * the Page Global Enable (PGE) feature bit is set. */
		*dx |= 0x00002000;
		/* We also lie, and say we're family id 5.  6 or greater
		 * leads to a rdmsr in early_init_intel which we can't handle.
		 * Family ID is returned as bits 8-12 in ax. */
		*ax &= 0xFFFFF0FF;
		*ax |= 0x00000500;
		break;
	case 0x80000000:
		/* Futureproof this a little: if they ask how much extended
		 * processor information there is, limit it to known fields. */
		if (*ax > 0x80000008)
			*ax = 0x80000008;
		break;
	case 0x80000001:
		/* Here we should fix nx cap depending on host. */
		/* For this version of PAE, we just clear NX bit. */
		*dx &= ~(1 << 20);
		break;
	}
}

/* Intel has four control registers, imaginatively named cr0, cr2, cr3 and cr4.
 * I assume there's a cr1, but it hasn't bothered us yet, so we'll not bother
 * it.  The Host needs to know when the Guest wants to change them, so we have
 * a whole series of functions like read_cr0() and write_cr0().
 *
 * We start with cr0.  cr0 allows you to turn on and off all kinds of basic
 * features, but Linux only really cares about one: the horrifically-named Task
 * Switched (TS) bit at bit 3 (ie. 8)
 *
 * What does the TS bit do?  Well, it causes the CPU to trap (interrupt 7) if
 * the floating point unit is used.  Which allows us to restore FPU state
 * lazily after a task switch, and Linux uses that gratefully, but wouldn't a
 * name like "FPUTRAP bit" be a little less cryptic?
 *
 * We store cr0 locally because the Host never changes it.  The Guest sometimes
 * wants to read it and we'd prefer not to bother the Host unnecessarily. */
static unsigned long current_cr0;
static void lguest_write_cr0(unsigned long val)
{
	lazy_hcall1(LHCALL_TS, val & X86_CR0_TS);
	current_cr0 = val;
}

static unsigned long lguest_read_cr0(void)
{
	return current_cr0;
}

/* Intel provided a special instruction to clear the TS bit for people too cool
 * to use write_cr0() to do it.  This "clts" instruction is faster, because all
 * the vowels have been optimized out. */
static void lguest_clts(void)
{
	lazy_hcall1(LHCALL_TS, 0);
	current_cr0 &= ~X86_CR0_TS;
}

/* cr2 is the virtual address of the last page fault, which the Guest only ever
 * reads.  The Host kindly writes this into our "struct lguest_data", so we
 * just read it out of there. */
static unsigned long lguest_read_cr2(void)
{
	return lguest_data.cr2;
}

/* See lguest_set_pte() below. */
static bool cr3_changed = false;

/* cr3 is the current toplevel pagetable page: the principle is the same as
 * cr0.  Keep a local copy, and tell the Host when it changes.  The only
 * difference is that our local copy is in lguest_data because the Host needs
 * to set it upon our initial hypercall. */
static void lguest_write_cr3(unsigned long cr3)
{
	lguest_data.pgdir = cr3;
	lazy_hcall1(LHCALL_NEW_PGTABLE, cr3);
	cr3_changed = true;
}

static unsigned long lguest_read_cr3(void)
{
	return lguest_data.pgdir;
}

/* cr4 is used to enable and disable PGE, but we don't care. */
static unsigned long lguest_read_cr4(void)
{
	return 0;
}

static void lguest_write_cr4(unsigned long val)
{
}

/*
 * Page Table Handling.
 *
 * Now would be a good time to take a rest and grab a coffee or similarly
 * relaxing stimulant.  The easy parts are behind us, and the trek gradually
 * winds uphill from here.
 *
 * Quick refresher: memory is divided into "pages" of 4096 bytes each.  The CPU
 * maps virtual addresses to physical addresses using "page tables".  We could
 * use one huge index of 1 million entries: each address is 4 bytes, so that's
 * 1024 pages just to hold the page tables.   But since most virtual addresses
 * are unused, we use a two level index which saves space.  The cr3 register
 * contains the physical address of the top level "page directory" page, which
 * contains physical addresses of up to 1024 second-level pages.  Each of these
 * second level pages contains up to 1024 physical addresses of actual pages,
 * or Page Table Entries (PTEs).
 *
 * Here's a diagram, where arrows indicate physical addresses:
 *
 * cr3 ---> +---------+
 *	    |  	   --------->+---------+
 *	    |	      |	     | PADDR1  |
 *	  Top-level   |	     | PADDR2  |
 *	  (PMD) page  |	     | 	       |
 *	    |	      |	   Lower-level |
 *	    |	      |	   (PTE) page  |
 *	    |	      |	     |	       |
 *	      ....    	     	 ....
 *
 * So to convert a virtual address to a physical address, we look up the top
 * level, which points us to the second level, which gives us the physical
 * address of that page.  If the top level entry was not present, or the second
 * level entry was not present, then the virtual address is invalid (we
 * say "the page was not mapped").
 *
 * Put another way, a 32-bit virtual address is divided up like so:
 *
 *  1 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 * |<---- 10 bits ---->|<---- 10 bits ---->|<------ 12 bits ------>|
 *    Index into top     Index into second      Offset within page
 *  page directory page    pagetable page
 *
 * The kernel spends a lot of time changing both the top-level page directory
 * and lower-level pagetable pages.  The Guest doesn't know physical addresses,
 * so while it maintains these page tables exactly like normal, it also needs
 * to keep the Host informed whenever it makes a change: the Host will create
 * the real page tables based on the Guests'.
 */

/* The Guest calls this to set a second-level entry (pte), ie. to map a page
 * into a process' address space.  We set the entry then tell the Host the
 * toplevel and address this corresponds to.  The Guest uses one pagetable per
 * process, so we need to tell the Host which one we're changing (mm->pgd). */
static void lguest_pte_update(struct mm_struct *mm, unsigned long addr,
			       pte_t *ptep)
{
#ifdef CONFIG_X86_PAE
	lazy_hcall4(LHCALL_SET_PTE, __pa(mm->pgd), addr,
		    ptep->pte_low, ptep->pte_high);
#else
	lazy_hcall3(LHCALL_SET_PTE, __pa(mm->pgd), addr, ptep->pte_low);
#endif
}

static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
			      pte_t *ptep, pte_t pteval)
{
	native_set_pte(ptep, pteval);
	lguest_pte_update(mm, addr, ptep);
}

/* The Guest calls lguest_set_pud to set a top-level entry and lguest_set_pmd
 * to set a middle-level entry when PAE is activated.
 * Again, we set the entry then tell the Host which page we changed,
 * and the index of the entry we changed. */
#ifdef CONFIG_X86_PAE
static void lguest_set_pud(pud_t *pudp, pud_t pudval)
{
	native_set_pud(pudp, pudval);

	/* 32 bytes aligned pdpt address and the index. */
	lazy_hcall2(LHCALL_SET_PGD, __pa(pudp) & 0xFFFFFFE0,
		   (__pa(pudp) & 0x1F) / sizeof(pud_t));
}

static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
{
	native_set_pmd(pmdp, pmdval);
	lazy_hcall2(LHCALL_SET_PMD, __pa(pmdp) & PAGE_MASK,
		   (__pa(pmdp) & (PAGE_SIZE - 1)) / sizeof(pmd_t));
}
#else

/* The Guest calls lguest_set_pmd to set a top-level entry when PAE is not
 * activated. */
static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
{
	native_set_pmd(pmdp, pmdval);
	lazy_hcall2(LHCALL_SET_PGD, __pa(pmdp) & PAGE_MASK,
		   (__pa(pmdp) & (PAGE_SIZE - 1)) / sizeof(pmd_t));
}
#endif

/* There are a couple of legacy places where the kernel sets a PTE, but we
 * don't know the top level any more.  This is useless for us, since we don't
 * know which pagetable is changing or what address, so we just tell the Host
 * to forget all of them.  Fortunately, this is very rare.
 *
 * ... except in early boot when the kernel sets up the initial pagetables,
 * which makes booting astonishingly slow: 1.83 seconds!  So we don't even tell
 * the Host anything changed until we've done the first page table switch,
 * which brings boot back to 0.25 seconds. */
static void lguest_set_pte(pte_t *ptep, pte_t pteval)
{
	native_set_pte(ptep, pteval);
	if (cr3_changed)
		lazy_hcall1(LHCALL_FLUSH_TLB, 1);
}

#ifdef CONFIG_X86_PAE
static void lguest_set_pte_atomic(pte_t *ptep, pte_t pte)
{
	native_set_pte_atomic(ptep, pte);
	if (cr3_changed)
		lazy_hcall1(LHCALL_FLUSH_TLB, 1);
}

void lguest_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
{
	native_pte_clear(mm, addr, ptep);
	lguest_pte_update(mm, addr, ptep);
}

void lguest_pmd_clear(pmd_t *pmdp)
{
	lguest_set_pmd(pmdp, __pmd(0));
}
#endif

/* Unfortunately for Lguest, the pv_mmu_ops for page tables were based on
 * native page table operations.  On native hardware you can set a new page
 * table entry whenever you want, but if you want to remove one you have to do
 * a TLB flush (a TLB is a little cache of page table entries kept by the CPU).
 *
 * So the lguest_set_pte_at() and lguest_set_pmd() functions above are only
 * called when a valid entry is written, not when it's removed (ie. marked not
 * present).  Instead, this is where we come when the Guest wants to remove a
 * page table entry: we tell the Host to set that entry to 0 (ie. the present
 * bit is zero). */
static void lguest_flush_tlb_single(unsigned long addr)
{
	/* Simply set it to zero: if it was not, it will fault back in. */
	lazy_hcall3(LHCALL_SET_PTE, lguest_data.pgdir, addr, 0);
}

/* This is what happens after the Guest has removed a large number of entries.
 * This tells the Host that any of the page table entries for userspace might
 * have changed, ie. virtual addresses below PAGE_OFFSET. */
static void lguest_flush_tlb_user(void)
{
	lazy_hcall1(LHCALL_FLUSH_TLB, 0);
}

/* This is called when the kernel page tables have changed.  That's not very
 * common (unless the Guest is using highmem, which makes the Guest extremely
 * slow), so it's worth separating this from the user flushing above. */
static void lguest_flush_tlb_kernel(void)
{
	lazy_hcall1(LHCALL_FLUSH_TLB, 1);
}

/*
 * The Unadvanced Programmable Interrupt Controller.
 *
 * This is an attempt to implement the simplest possible interrupt controller.
 * I spent some time looking though routines like set_irq_chip_and_handler,
 * set_irq_chip_and_handler_name, set_irq_chip_data and set_phasers_to_stun and
 * I *think* this is as simple as it gets.
 *
 * We can tell the Host what interrupts we want blocked ready for using the
 * lguest_data.interrupts bitmap, so disabling (aka "masking") them is as
 * simple as setting a bit.  We don't actually "ack" interrupts as such, we
 * just mask and unmask them.  I wonder if we should be cleverer?
 */
static void disable_lguest_irq(unsigned int irq)
{
	set_bit(irq, lguest_data.blocked_interrupts);
}

static void enable_lguest_irq(unsigned int irq)
{
	clear_bit(irq, lguest_data.blocked_interrupts);
}

/* This structure describes the lguest IRQ controller. */
static struct irq_chip lguest_irq_controller = {
	.name		= "lguest",
	.mask		= disable_lguest_irq,
	.mask_ack	= disable_lguest_irq,
	.unmask		= enable_lguest_irq,
};

/* This sets up the Interrupt Descriptor Table (IDT) entry for each hardware
 * interrupt (except 128, which is used for system calls), and then tells the
 * Linux infrastructure that each interrupt is controlled by our level-based
 * lguest interrupt controller. */
static void __init lguest_init_IRQ(void)
{
	unsigned int i;

	for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) {
		/* Some systems map "vectors" to interrupts weirdly.  Lguest has
		 * a straightforward 1 to 1 mapping, so force that here. */
		__get_cpu_var(vector_irq)[i] = i - FIRST_EXTERNAL_VECTOR;
		if (i != SYSCALL_VECTOR)
			set_intr_gate(i, interrupt[i - FIRST_EXTERNAL_VECTOR]);
	}
	/* This call is required to set up for 4k stacks, where we have
	 * separate stacks for hard and soft interrupts. */
	irq_ctx_init(smp_processor_id());
}

void lguest_setup_irq(unsigned int irq)
{
	irq_to_desc_alloc_node(irq, 0);
	set_irq_chip_and_handler_name(irq, &lguest_irq_controller,
				      handle_level_irq, "level");
}

/*
 * Time.
 *
 * It would be far better for everyone if the Guest had its own clock, but
 * until then the Host gives us the time on every interrupt.
 */
static unsigned long lguest_get_wallclock(void)
{
	return lguest_data.time.tv_sec;
}

/* The TSC is an Intel thing called the Time Stamp Counter.  The Host tells us
 * what speed it runs at, or 0 if it's unusable as a reliable clock source.
 * This matches what we want here: if we return 0 from this function, the x86
 * TSC clock will give up and not register itself. */
static unsigned long lguest_tsc_khz(void)
{
	return lguest_data.tsc_khz;
}

/* If we can't use the TSC, the kernel falls back to our lower-priority
 * "lguest_clock", where we read the time value given to us by the Host. */
static cycle_t lguest_clock_read(struct clocksource *cs)
{
	unsigned long sec, nsec;

	/* Since the time is in two parts (seconds and nanoseconds), we risk
	 * reading it just as it's changing from 99 & 0.999999999 to 100 and 0,
	 * and getting 99 and 0.  As Linux tends to come apart under the stress
	 * of time travel, we must be careful: */
	do {
		/* First we read the seconds part. */
		sec = lguest_data.time.tv_sec;
		/* This read memory barrier tells the compiler and the CPU that
		 * this can't be reordered: we have to complete the above
		 * before going on. */
		rmb();
		/* Now we read the nanoseconds part. */
		nsec = lguest_data.time.tv_nsec;
		/* Make sure we've done that. */
		rmb();
		/* Now if the seconds part has changed, try again. */
	} while (unlikely(lguest_data.time.tv_sec != sec));

	/* Our lguest clock is in real nanoseconds. */
	return sec*1000000000ULL + nsec;
}

/* This is the fallback clocksource: lower priority than the TSC clocksource. */
static struct clocksource lguest_clock = {
	.name		= "lguest",
	.rating		= 200,
	.read		= lguest_clock_read,
	.mask		= CLOCKSOURCE_MASK(64),
	.mult		= 1 << 22,
	.shift		= 22,
	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
};

/* We also need a "struct clock_event_device": Linux asks us to set it to go
 * off some time in the future.  Actually, James Morris figured all this out, I
 * just applied the patch. */
static int lguest_clockevent_set_next_event(unsigned long delta,
                                           struct clock_event_device *evt)
{
	/* FIXME: I don't think this can ever happen, but James tells me he had
	 * to put this code in.  Maybe we should remove it now.  Anyone? */
	if (delta < LG_CLOCK_MIN_DELTA) {
		if (printk_ratelimit())
			printk(KERN_DEBUG "%s: small delta %lu ns\n",
			       __func__, delta);
		return -ETIME;
	}

	/* Please wake us this far in the future. */
	kvm_hypercall1(LHCALL_SET_CLOCKEVENT, delta);
	return 0;
}

static void lguest_clockevent_set_mode(enum clock_event_mode mode,
                                      struct clock_event_device *evt)
{
	switch (mode) {
	case CLOCK_EVT_MODE_UNUSED:
	case CLOCK_EVT_MODE_SHUTDOWN:
		/* A 0 argument shuts the clock down. */
		kvm_hypercall0(LHCALL_SET_CLOCKEVENT);
		break;
	case CLOCK_EVT_MODE_ONESHOT:
		/* This is what we expect. */
		break;
	case CLOCK_EVT_MODE_PERIODIC:
		BUG();
	case CLOCK_EVT_MODE_RESUME:
		break;
	}
}

/* This describes our primitive timer chip. */
static struct clock_event_device lguest_clockevent = {
	.name                   = "lguest",
	.features               = CLOCK_EVT_FEAT_ONESHOT,
	.set_next_event         = lguest_clockevent_set_next_event,
	.set_mode               = lguest_clockevent_set_mode,
	.rating                 = INT_MAX,
	.mult                   = 1,
	.shift                  = 0,
	.min_delta_ns           = LG_CLOCK_MIN_DELTA,
	.max_delta_ns           = LG_CLOCK_MAX_DELTA,
};

/* This is the Guest timer interrupt handler (hardware interrupt 0).  We just
 * call the clockevent infrastructure and it does whatever needs doing. */
static void lguest_time_irq(unsigned int irq, struct irq_desc *desc)
{
	unsigned long flags;

	/* Don't interrupt us while this is running. */
	local_irq_save(flags);
	lguest_clockevent.event_handler(&lguest_clockevent);
	local_irq_restore(flags);
}

/* At some point in the boot process, we get asked to set up our timing
 * infrastructure.  The kernel doesn't expect timer interrupts before this, but
 * we cleverly initialized the "blocked_interrupts" field of "struct
 * lguest_data" so that timer interrupts were blocked until now. */
static void lguest_time_init(void)
{
	/* Set up the timer interrupt (0) to go to our simple timer routine */
	set_irq_handler(0, lguest_time_irq);

	clocksource_register(&lguest_clock);

	/* We can't set cpumask in the initializer: damn C limitations!  Set it
	 * here and register our timer device. */
	lguest_clockevent.cpumask = cpumask_of(0);
	clockevents_register_device(&lguest_clockevent);

	/* Finally, we unblock the timer interrupt. */
	enable_lguest_irq(0);
}

/*
 * Miscellaneous bits and pieces.
 *
 * Here is an oddball collection of functions which the Guest needs for things
 * to work.  They're pretty simple.
 */

/* The Guest needs to tell the Host what stack it expects traps to use.  For
 * native hardware, this is part of the Task State Segment mentioned above in
 * lguest_load_tr_desc(), but to help hypervisors there's this special call.
 *
 * We tell the Host the segment we want to use (__KERNEL_DS is the kernel data
 * segment), the privilege level (we're privilege level 1, the Host is 0 and
 * will not tolerate us trying to use that), the stack pointer, and the number
 * of pages in the stack. */
static void lguest_load_sp0(struct tss_struct *tss,
			    struct thread_struct *thread)
{
	lazy_hcall3(LHCALL_SET_STACK, __KERNEL_DS | 0x1, thread->sp0,
		   THREAD_SIZE / PAGE_SIZE);
}

/* Let's just say, I wouldn't do debugging under a Guest. */
static void lguest_set_debugreg(int regno, unsigned long value)
{
	/* FIXME: Implement */
}

/* There are times when the kernel wants to make sure that no memory writes are
 * caught in the cache (that they've all reached real hardware devices).  This
 * doesn't matter for the Guest which has virtual hardware.
 *
 * On the Pentium 4 and above, cpuid() indicates that the Cache Line Flush
 * (clflush) instruction is available and the kernel uses that.  Otherwise, it
 * uses the older "Write Back and Invalidate Cache" (wbinvd) instruction.
 * Unlike clflush, wbinvd can only be run at privilege level 0.  So we can
 * ignore clflush, but replace wbinvd.
 */
static void lguest_wbinvd(void)
{
}

/* If the Guest expects to have an Advanced Programmable Interrupt Controller,
 * we play dumb by ignoring writes and returning 0 for reads.  So it's no
 * longer Programmable nor Controlling anything, and I don't think 8 lines of
 * code qualifies for Advanced.  It will also never interrupt anything.  It
 * does, however, allow us to get through the Linux boot code. */
#ifdef CONFIG_X86_LOCAL_APIC
static void lguest_apic_write(u32 reg, u32 v)
{
}

static u32 lguest_apic_read(u32 reg)
{
	return 0;
}

static u64 lguest_apic_icr_read(void)
{
	return 0;
}

static void lguest_apic_icr_write(u32 low, u32 id)
{
	/* Warn to see if there's any stray references */
	WARN_ON(1);
}

static void lguest_apic_wait_icr_idle(void)
{
	return;
}

static u32 lguest_apic_safe_wait_icr_idle(void)
{
	return 0;
}

static void set_lguest_basic_apic_ops(void)
{
	apic->read = lguest_apic_read;
	apic->write = lguest_apic_write;
	apic->icr_read = lguest_apic_icr_read;
	apic->icr_write = lguest_apic_icr_write;
	apic->wait_icr_idle = lguest_apic_wait_icr_idle;
	apic->safe_wait_icr_idle = lguest_apic_safe_wait_icr_idle;
};
#endif

/* STOP!  Until an interrupt comes in. */
static void lguest_safe_halt(void)
{
	kvm_hypercall0(LHCALL_HALT);
}

/* The SHUTDOWN hypercall takes a string to describe what's happening, and
 * an argument which says whether this to restart (reboot) the Guest or not.
 *
 * Note that the Host always prefers that the Guest speak in physical addresses
 * rather than virtual addresses, so we use __pa() here. */
static void lguest_power_off(void)
{
	kvm_hypercall2(LHCALL_SHUTDOWN, __pa("Power down"),
					LGUEST_SHUTDOWN_POWEROFF);
}

/*
 * Panicing.
 *
 * Don't.  But if you did, this is what happens.
 */
static int lguest_panic(struct notifier_block *nb, unsigned long l, void *p)
{
	kvm_hypercall2(LHCALL_SHUTDOWN, __pa(p), LGUEST_SHUTDOWN_POWEROFF);
	/* The hcall won't return, but to keep gcc happy, we're "done". */
	return NOTIFY_DONE;
}

static struct notifier_block paniced = {
	.notifier_call = lguest_panic
};

/* Setting up memory is fairly easy. */
static __init char *lguest_memory_setup(void)
{
	/* We do this here and not earlier because lockcheck used to barf if we
	 * did it before start_kernel().  I think we fixed that, so it'd be
	 * nice to move it back to lguest_init.  Patch welcome... */
	atomic_notifier_chain_register(&panic_notifier_list, &paniced);

	/* The Linux bootloader header contains an "e820" memory map: the
	 * Launcher populated the first entry with our memory limit. */
	e820_add_region(boot_params.e820_map[0].addr,
			  boot_params.e820_map[0].size,
			  boot_params.e820_map[0].type);

	/* This string is for the boot messages. */
	return "LGUEST";
}

/* We will eventually use the virtio console device to produce console output,
 * but before that is set up we use LHCALL_NOTIFY on normal memory to produce
 * console output. */
static __init int early_put_chars(u32 vtermno, const char *buf, int count)
{
	char scratch[17];
	unsigned int len = count;

	/* We use a nul-terminated string, so we have to make a copy.  Icky,
	 * huh? */
	if (len > sizeof(scratch) - 1)
		len = sizeof(scratch) - 1;
	scratch[len] = '\0';
	memcpy(scratch, buf, len);
	kvm_hypercall1(LHCALL_NOTIFY, __pa(scratch));

	/* This routine returns the number of bytes actually written. */
	return len;
}

/* Rebooting also tells the Host we're finished, but the RESTART flag tells the
 * Launcher to reboot us. */
static void lguest_restart(char *reason)
{
	kvm_hypercall2(LHCALL_SHUTDOWN, __pa(reason), LGUEST_SHUTDOWN_RESTART);
}

/*G:050
 * Patching (Powerfully Placating Performance Pedants)
 *
 * We have already seen that pv_ops structures let us replace simple native
 * instructions with calls to the appropriate back end all throughout the
 * kernel.  This allows the same kernel to run as a Guest and as a native
 * kernel, but it's slow because of all the indirect branches.
 *
 * Remember that David Wheeler quote about "Any problem in computer science can
 * be solved with another layer of indirection"?  The rest of that quote is
 * "... But that usually will create another problem."  This is the first of
 * those problems.
 *
 * Our current solution is to allow the paravirt back end to optionally patch
 * over the indirect calls to replace them with something more efficient.  We
 * patch two of the simplest of the most commonly called functions: disable
 * interrupts and save interrupts.  We usually have 6 or 10 bytes to patch
 * into: the Guest versions of these operations are small enough that we can
 * fit comfortably.
 *
 * First we need assembly templates of each of the patchable Guest operations,
 * and these are in i386_head.S. */

/*G:060 We construct a table from the assembler templates: */
static const struct lguest_insns
{
	const char *start, *end;
} lguest_insns[] = {
	[PARAVIRT_PATCH(pv_irq_ops.irq_disable)] = { lgstart_cli, lgend_cli },
	[PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf },
};

/* Now our patch routine is fairly simple (based on the native one in
 * paravirt.c).  If we have a replacement, we copy it in and return how much of
 * the available space we used. */
static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
			     unsigned long addr, unsigned len)
{
	unsigned int insn_len;

	/* Don't do anything special if we don't have a replacement */
	if (type >= ARRAY_SIZE(lguest_insns) || !lguest_insns[type].start)
		return paravirt_patch_default(type, clobber, ibuf, addr, len);

	insn_len = lguest_insns[type].end - lguest_insns[type].start;

	/* Similarly if we can't fit replacement (shouldn't happen, but let's
	 * be thorough). */
	if (len < insn_len)
		return paravirt_patch_default(type, clobber, ibuf, addr, len);

	/* Copy in our instructions. */
	memcpy(ibuf, lguest_insns[type].start, insn_len);
	return insn_len;
}

/*G:029 Once we get to lguest_init(), we know we're a Guest.  The various
 * pv_ops structures in the kernel provide points for (almost) every routine we
 * have to override to avoid privileged instructions. */
__init void lguest_init(void)
{
	/* We're under lguest, paravirt is enabled, and we're running at
	 * privilege level 1, not 0 as normal. */
	pv_info.name = "lguest";
	pv_info.paravirt_enabled = 1;
	pv_info.kernel_rpl = 1;
	pv_info.shared_kernel_pmd = 1;

	/* We set up all the lguest overrides for sensitive operations.  These
	 * are detailed with the operations themselves. */

	/* interrupt-related operations */
	pv_irq_ops.init_IRQ = lguest_init_IRQ;
	pv_irq_ops.save_fl = PV_CALLEE_SAVE(save_fl);
	pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(lg_restore_fl);
	pv_irq_ops.irq_disable = PV_CALLEE_SAVE(irq_disable);
	pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(lg_irq_enable);
	pv_irq_ops.safe_halt = lguest_safe_halt;

	/* init-time operations */
	pv_init_ops.memory_setup = lguest_memory_setup;
	pv_init_ops.patch = lguest_patch;

	/* Intercepts of various cpu instructions */
	pv_cpu_ops.load_gdt = lguest_load_gdt;
	pv_cpu_ops.cpuid = lguest_cpuid;
	pv_cpu_ops.load_idt = lguest_load_idt;
	pv_cpu_ops.iret = lguest_iret;
	pv_cpu_ops.load_sp0 = lguest_load_sp0;
	pv_cpu_ops.load_tr_desc = lguest_load_tr_desc;
	pv_cpu_ops.set_ldt = lguest_set_ldt;
	pv_cpu_ops.load_tls = lguest_load_tls;
	pv_cpu_ops.set_debugreg = lguest_set_debugreg;
	pv_cpu_ops.clts = lguest_clts;
	pv_cpu_ops.read_cr0 = lguest_read_cr0;
	pv_cpu_ops.write_cr0 = lguest_write_cr0;
	pv_cpu_ops.read_cr4 = lguest_read_cr4;
	pv_cpu_ops.write_cr4 = lguest_write_cr4;
	pv_cpu_ops.write_gdt_entry = lguest_write_gdt_entry;
	pv_cpu_ops.write_idt_entry = lguest_write_idt_entry;
	pv_cpu_ops.wbinvd = lguest_wbinvd;
	pv_cpu_ops.start_context_switch = paravirt_start_context_switch;
	pv_cpu_ops.end_context_switch = lguest_end_context_switch;

	/* pagetable management */
	pv_mmu_ops.write_cr3 = lguest_write_cr3;
	pv_mmu_ops.flush_tlb_user = lguest_flush_tlb_user;
	pv_mmu_ops.flush_tlb_single = lguest_flush_tlb_single;
	pv_mmu_ops.flush_tlb_kernel = lguest_flush_tlb_kernel;
	pv_mmu_ops.set_pte = lguest_set_pte;
	pv_mmu_ops.set_pte_at = lguest_set_pte_at;
	pv_mmu_ops.set_pmd = lguest_set_pmd;
#ifdef CONFIG_X86_PAE
	pv_mmu_ops.set_pte_atomic = lguest_set_pte_atomic;
	pv_mmu_ops.pte_clear = lguest_pte_clear;
	pv_mmu_ops.pmd_clear = lguest_pmd_clear;
	pv_mmu_ops.set_pud = lguest_set_pud;
#endif
	pv_mmu_ops.read_cr2 = lguest_read_cr2;
	pv_mmu_ops.read_cr3 = lguest_read_cr3;
	pv_mmu_ops.lazy_mode.enter = paravirt_enter_lazy_mmu;
	pv_mmu_ops.lazy_mode.leave = lguest_leave_lazy_mmu_mode;
	pv_mmu_ops.pte_update = lguest_pte_update;
	pv_mmu_ops.pte_update_defer = lguest_pte_update;

#ifdef CONFIG_X86_LOCAL_APIC
	/* apic read/write intercepts */
	set_lguest_basic_apic_ops();
#endif

	/* time operations */
	pv_time_ops.get_wallclock = lguest_get_wallclock;
	pv_time_ops.time_init = lguest_time_init;
	pv_time_ops.get_tsc_khz = lguest_tsc_khz;

	/* Now is a good time to look at the implementations of these functions
	 * before returning to the rest of lguest_init(). */

	/*G:070 Now we've seen all the paravirt_ops, we return to
	 * lguest_init() where the rest of the fairly chaotic boot setup
	 * occurs. */

	/* The stack protector is a weird thing where gcc places a canary
	 * value on the stack and then checks it on return.  This file is
	 * compiled with -fno-stack-protector it, so we got this far without
	 * problems.  The value of the canary is kept at offset 20 from the
	 * %gs register, so we need to set that up before calling C functions
	 * in other files. */
	setup_stack_canary_segment(0);
	/* We could just call load_stack_canary_segment(), but we might as
	 * call switch_to_new_gdt() which loads the whole table and sets up
	 * the per-cpu segment descriptor register %fs as well. */
	switch_to_new_gdt(0);

	/* As described in head_32.S, we map the first 128M of memory. */
	max_pfn_mapped = (128*1024*1024) >> PAGE_SHIFT;

	/* The Host<->Guest Switcher lives at the top of our address space, and
	 * the Host told us how big it is when we made LGUEST_INIT hypercall:
	 * it put the answer in lguest_data.reserve_mem  */
	reserve_top_address(lguest_data.reserve_mem);

	/* If we don't initialize the lock dependency checker now, it crashes
	 * paravirt_disable_iospace. */
	lockdep_init();

	/* The IDE code spends about 3 seconds probing for disks: if we reserve
	 * all the I/O ports up front it can't get them and so doesn't probe.
	 * Other device drivers are similar (but less severe).  This cuts the
	 * kernel boot time on my machine from 4.1 seconds to 0.45 seconds. */
	paravirt_disable_iospace();

	/* This is messy CPU setup stuff which the native boot code does before
	 * start_kernel, so we have to do, too: */
	cpu_detect(&new_cpu_data);
	/* head.S usually sets up the first capability word, so do it here. */
	new_cpu_data.x86_capability[0] = cpuid_edx(1);

	/* Math is always hard! */
	new_cpu_data.hard_math = 1;

	/* We don't have features.  We have puppies!  Puppies! */
#ifdef CONFIG_X86_MCE
	mce_disabled = 1;
#endif
#ifdef CONFIG_ACPI
	acpi_disabled = 1;
	acpi_ht = 0;
#endif

	/* We set the preferred console to "hvc".  This is the "hypervisor
	 * virtual console" driver written by the PowerPC people, which we also
	 * adapted for lguest's use. */
	add_preferred_console("hvc", 0, NULL);

	/* Register our very early console. */
	virtio_cons_early_init(early_put_chars);

	/* Last of all, we set the power management poweroff hook to point to
	 * the Guest routine to power off, and the reboot hook to our restart
	 * routine. */
	pm_power_off = lguest_power_off;
	machine_ops.restart = lguest_restart;

	/* Now we're set up, call i386_start_kernel() in head32.c and we proceed
	 * to boot as normal.  It never returns. */
	i386_start_kernel();
}
/*
 * This marks the end of stage II of our journey, The Guest.
 *
 * It is now time for us to explore the layer of virtual drivers and complete
 * our understanding of the Guest in "make Drivers".
 */