boot.c 47.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12
/*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.
 *
13 14 15 16
 * 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.
R
Rusty Russell 已提交
17
 *
18 19 20 21 22
 * 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.
 *
23 24
 * 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
R
Rusty Russell 已提交
25 26
 * "paravirt" structures with our Guest versions, then boot like normal.
:*/
27 28

/*
R
Rusty Russell 已提交
29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52
 * 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>
R
Rusty Russell 已提交
53 54
#include <linux/clocksource.h>
#include <linux/clockchips.h>
R
Rusty Russell 已提交
55 56
#include <linux/lguest.h>
#include <linux/lguest_launcher.h>
R
Rusty Russell 已提交
57
#include <linux/virtio_console.h>
J
Jeff Garzik 已提交
58
#include <linux/pm.h>
I
Ingo Molnar 已提交
59
#include <asm/apic.h>
60
#include <asm/lguest.h>
R
Rusty Russell 已提交
61 62 63 64 65 66 67 68 69
#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>
70
#include <asm/i387.h>
71
#include <asm/stackprotector.h>
B
Balaji Rao 已提交
72
#include <asm/reboot.h>		/* for struct machine_ops */
R
Rusty Russell 已提交
73

R
Rusty Russell 已提交
74 75 76 77
/*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
R
Rusty Russell 已提交
78 79
 * same kernel as the Host (or at least, built from the same source code).
:*/
R
Rusty Russell 已提交
80

R
Rusty Russell 已提交
81 82 83 84
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,
85
	.kernel_address = PAGE_OFFSET,
R
Rusty Russell 已提交
86
	.blocked_interrupts = { 1 }, /* Block timer interrupts */
87
	.syscall_vec = SYSCALL_VECTOR,
R
Rusty Russell 已提交
88 89
};

R
Rusty Russell 已提交
90 91
/*G:037
 * async_hcall() is pretty simple: I'm quite proud of it really.  We have a
R
Rusty Russell 已提交
92
 * ring buffer of stored hypercalls which the Host will run though next time we
93
 * do a normal hypercall.  Each entry in the ring has 5 slots for the hypercall
R
Rusty Russell 已提交
94 95 96 97 98 99
 * 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
R
Rusty Russell 已提交
100 101
 * which empties it for next time!
 */
A
Adrian Bunk 已提交
102
static void async_hcall(unsigned long call, unsigned long arg1,
103 104
			unsigned long arg2, unsigned long arg3,
			unsigned long arg4)
R
Rusty Russell 已提交
105 106 107 108 109
{
	/* Note: This code assumes we're uniprocessor. */
	static unsigned int next_call;
	unsigned long flags;

R
Rusty Russell 已提交
110 111
	/*
	 * Disable interrupts if not already disabled: we don't want an
R
Rusty Russell 已提交
112
	 * interrupt handler making a hypercall while we're already doing
R
Rusty Russell 已提交
113 114
	 * one!
	 */
R
Rusty Russell 已提交
115 116 117
	local_irq_save(flags);
	if (lguest_data.hcall_status[next_call] != 0xFF) {
		/* Table full, so do normal hcall which will flush table. */
118
		hcall(call, arg1, arg2, arg3, arg4);
R
Rusty Russell 已提交
119
	} else {
J
Jes Sorensen 已提交
120 121 122 123
		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;
124
		lguest_data.hcalls[next_call].arg4 = arg4;
R
Rusty Russell 已提交
125
		/* Arguments must all be written before we mark it to go */
R
Rusty Russell 已提交
126 127 128 129 130 131 132
		wmb();
		lguest_data.hcall_status[next_call] = 0;
		if (++next_call == LHCALL_RING_SIZE)
			next_call = 0;
	}
	local_irq_restore(flags);
}
A
Adrian Bunk 已提交
133

R
Rusty Russell 已提交
134 135 136
/*G:035
 * Notice the lazy_hcall() above, rather than hcall().  This is our first real
 * optimization trick!
R
Rusty Russell 已提交
137 138 139 140 141 142 143 144 145
 *
 * 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
R
Rusty Russell 已提交
146 147
 * future processing:
 */
148
static void lazy_hcall1(unsigned long call, unsigned long arg1)
149 150
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
151
		hcall(call, arg1, 0, 0, 0);
152
	else
153
		async_hcall(call, arg1, 0, 0, 0);
154 155
}

R
Rusty Russell 已提交
156
/* You can imagine what lazy_hcall2, 3 and 4 look like. :*/
157
static void lazy_hcall2(unsigned long call,
158 159
			unsigned long arg1,
			unsigned long arg2)
160 161
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
162
		hcall(call, arg1, arg2, 0, 0);
163
	else
164
		async_hcall(call, arg1, arg2, 0, 0);
165 166 167
}

static void lazy_hcall3(unsigned long call,
168 169 170
			unsigned long arg1,
			unsigned long arg2,
			unsigned long arg3)
A
Adrian Bunk 已提交
171 172
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
173
		hcall(call, arg1, arg2, arg3, 0);
A
Adrian Bunk 已提交
174
	else
175 176 177
		async_hcall(call, arg1, arg2, arg3, 0);
}

M
Matias Zabaljauregui 已提交
178
#ifdef CONFIG_X86_PAE
179
static void lazy_hcall4(unsigned long call,
180 181 182 183
			unsigned long arg1,
			unsigned long arg2,
			unsigned long arg3,
			unsigned long arg4)
184 185
{
	if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE)
186
		hcall(call, arg1, arg2, arg3, arg4);
187 188
	else
		async_hcall(call, arg1, arg2, arg3, arg4);
A
Adrian Bunk 已提交
189
}
M
Matias Zabaljauregui 已提交
190
#endif
R
Rusty Russell 已提交
191

R
Rusty Russell 已提交
192 193 194 195
/*G:036
 * 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.
:*/
196
static void lguest_leave_lazy_mmu_mode(void)
R
Rusty Russell 已提交
197
{
198
	hcall(LHCALL_FLUSH_ASYNC, 0, 0, 0, 0);
199 200 201
	paravirt_leave_lazy_mmu();
}

202
static void lguest_end_context_switch(struct task_struct *next)
203
{
204
	hcall(LHCALL_FLUSH_ASYNC, 0, 0, 0, 0);
205
	paravirt_end_context_switch(next);
R
Rusty Russell 已提交
206
}
R
Rusty Russell 已提交
207

208
/*G:032
R
Rusty Russell 已提交
209 210
 * After that diversion we return to our first native-instruction
 * replacements: four functions for interrupt control.
R
Rusty Russell 已提交
211 212 213 214 215 216 217
 *
 * 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
218
 * check there before it tries to deliver an interrupt.
R
Rusty Russell 已提交
219 220
 */

R
Rusty Russell 已提交
221 222
/*
 * save_flags() is expected to return the processor state (ie. "flags").  The
223
 * flags word contains all kind of stuff, but in practice Linux only cares
R
Rusty Russell 已提交
224 225
 * about the interrupt flag.  Our "save_flags()" just returns that.
 */
R
Rusty Russell 已提交
226 227 228 229 230
static unsigned long save_fl(void)
{
	return lguest_data.irq_enabled;
}

R
Rusty Russell 已提交
231
/* Interrupts go off... */
R
Rusty Russell 已提交
232 233 234 235 236
static void irq_disable(void)
{
	lguest_data.irq_enabled = 0;
}

R
Rusty Russell 已提交
237 238
/*
 * Let's pause a moment.  Remember how I said these are called so often?
239 240 241 242 243
 * 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
R
Rusty Russell 已提交
244 245
 * C function, then restores it.
 */
246 247 248
PV_CALLEE_SAVE_REGS_THUNK(save_fl);
PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
/*:*/
249

250 251 252
/* These are in i386_head.S */
extern void lg_irq_enable(void);
extern void lg_restore_fl(unsigned long flags);
253

R
Rusty Russell 已提交
254
/*M:003
R
Rusty Russell 已提交
255 256 257 258 259
 * We could be more efficient in our checking of outstanding interrupts, rather
 * than using a branch.  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.
260 261 262 263
 *
 * 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
R
Rusty Russell 已提交
264 265
 * a hypercall for interrupt control and not worry about efficiency.
:*/
R
Rusty Russell 已提交
266

R
Rusty Russell 已提交
267 268 269 270 271 272 273 274
/*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.
 */
275 276
static void lguest_write_idt_entry(gate_desc *dt,
				   int entrynum, const gate_desc *g)
R
Rusty Russell 已提交
277
{
R
Rusty Russell 已提交
278 279
	/*
	 * The gate_desc structure is 8 bytes long: we hand it to the Host in
280 281
	 * 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
R
Rusty Russell 已提交
282 283
	 * years.
	 */
284
	u32 *desc = (u32 *)g;
R
Rusty Russell 已提交
285
	/* Keep the local copy up to date. */
286
	native_write_idt_entry(dt, entrynum, g);
R
Rusty Russell 已提交
287
	/* Tell Host about this new entry. */
288
	hcall(LHCALL_LOAD_IDT_ENTRY, entrynum, desc[0], desc[1], 0);
R
Rusty Russell 已提交
289 290
}

R
Rusty Russell 已提交
291 292
/*
 * Changing to a different IDT is very rare: we keep the IDT up-to-date every
R
Rusty Russell 已提交
293
 * time it is written, so we can simply loop through all entries and tell the
R
Rusty Russell 已提交
294 295
 * Host about them.
 */
296
static void lguest_load_idt(const struct desc_ptr *desc)
R
Rusty Russell 已提交
297 298 299 300 301
{
	unsigned int i;
	struct desc_struct *idt = (void *)desc->address;

	for (i = 0; i < (desc->size+1)/8; i++)
302
		hcall(LHCALL_LOAD_IDT_ENTRY, i, idt[i].a, idt[i].b, 0);
R
Rusty Russell 已提交
303 304
}

R
Rusty Russell 已提交
305 306 307 308 309 310 311 312 313 314
/*
 * 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.
 *
315
 * This is the exactly like the IDT code.
R
Rusty Russell 已提交
316
 */
317
static void lguest_load_gdt(const struct desc_ptr *desc)
R
Rusty Russell 已提交
318
{
319 320 321 322
	unsigned int i;
	struct desc_struct *gdt = (void *)desc->address;

	for (i = 0; i < (desc->size+1)/8; i++)
323
		hcall(LHCALL_LOAD_GDT_ENTRY, i, gdt[i].a, gdt[i].b, 0);
R
Rusty Russell 已提交
324 325
}

R
Rusty Russell 已提交
326
/*
327 328
 * For a single GDT entry which changes, we simply change our copy and
 * then tell the host about it.
R
Rusty Russell 已提交
329
 */
330 331
static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum,
				   const void *desc, int type)
R
Rusty Russell 已提交
332
{
333
	native_write_gdt_entry(dt, entrynum, desc, type);
334
	/* Tell Host about this new entry. */
335 336
	hcall(LHCALL_LOAD_GDT_ENTRY, entrynum,
	      dt[entrynum].a, dt[entrynum].b, 0);
R
Rusty Russell 已提交
337 338
}

R
Rusty Russell 已提交
339
/*
340
 * There are three "thread local storage" GDT entries which change
R
Rusty Russell 已提交
341
 * on every context switch (these three entries are how glibc implements
342 343 344 345 346
 * __thread variables).  As an optimization, we have a hypercall
 * specifically for this case.
 *
 * Wouldn't it be nicer to have a general LOAD_GDT_ENTRIES hypercall
 * which took a range of entries?
R
Rusty Russell 已提交
347
 */
R
Rusty Russell 已提交
348 349
static void lguest_load_tls(struct thread_struct *t, unsigned int cpu)
{
R
Rusty Russell 已提交
350 351
	/*
	 * There's one problem which normal hardware doesn't have: the Host
352
	 * can't handle us removing entries we're currently using.  So we clear
R
Rusty Russell 已提交
353 354
	 * the GS register here: if it's needed it'll be reloaded anyway.
	 */
355
	lazy_load_gs(0);
356
	lazy_hcall2(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu);
R
Rusty Russell 已提交
357 358
}

R
Rusty Russell 已提交
359 360 361
/*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).
R
Rusty Russell 已提交
362 363 364
 *
 * 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
R
Rusty Russell 已提交
365 366
 * here, so they'll get an informative and friendly Segmentation Fault.
 */
R
Rusty Russell 已提交
367 368 369 370
static void lguest_set_ldt(const void *addr, unsigned entries)
{
}

R
Rusty Russell 已提交
371 372
/*
 * This loads a GDT entry into the "Task Register": that entry points to a
R
Rusty Russell 已提交
373 374 375 376 377 378 379
 * 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
R
Rusty Russell 已提交
380 381
 * override the native version with a do-nothing version.
 */
R
Rusty Russell 已提交
382 383 384 385
static void lguest_load_tr_desc(void)
{
}

R
Rusty Russell 已提交
386 387
/*
 * The "cpuid" instruction is a way of querying both the CPU identity
R
Rusty Russell 已提交
388
 * (manufacturer, model, etc) and its features.  It was introduced before the
389 390 391
 * 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.
R
Rusty Russell 已提交
392 393
 *
 * This instruction even it has its own Wikipedia entry.  The Wikipedia entry
R
Rusty Russell 已提交
394
 * has been translated into 5 languages.  I am not making this up!
R
Rusty Russell 已提交
395 396 397 398 399 400 401 402 403 404 405
 *
 * 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
R
Rusty Russell 已提交
406 407
 * too worked up about it.
 */
408 409
static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
			 unsigned int *cx, unsigned int *dx)
R
Rusty Russell 已提交
410
{
411
	int function = *ax;
R
Rusty Russell 已提交
412

413
	native_cpuid(ax, bx, cx, dx);
R
Rusty Russell 已提交
414
	switch (function) {
R
Rusty Russell 已提交
415 416 417 418 419
	/*
	 * CPUID 0 gives the highest legal CPUID number (and the ID string).
	 * We futureproof our code a little by sticking to known CPUID values.
	 */
	case 0:
420 421 422
		if (*ax > 5)
			*ax = 5;
		break;
R
Rusty Russell 已提交
423 424 425 426 427 428 429 430

	/*
	 * CPUID 1 is a basic feature request.
	 *
	 * CX: we only allow kernel to see SSE3, CMPXCHG16B and SSSE3
	 * DX: SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, TSC, FPU and PAE.
	 */
	case 1:
431
		*cx &= 0x00002201;
M
Matias Zabaljauregui 已提交
432
		*dx &= 0x07808151;
R
Rusty Russell 已提交
433 434
		/*
		 * The Host can do a nice optimization if it knows that the
R
Rusty Russell 已提交
435 436 437
		 * 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
R
Rusty Russell 已提交
438 439
		 * the Page Global Enable (PGE) feature bit is set.
		 */
440
		*dx |= 0x00002000;
R
Rusty Russell 已提交
441 442
		/*
		 * We also lie, and say we're family id 5.  6 or greater
443
		 * leads to a rdmsr in early_init_intel which we can't handle.
R
Rusty Russell 已提交
444 445
		 * Family ID is returned as bits 8-12 in ax.
		 */
446 447
		*ax &= 0xFFFFF0FF;
		*ax |= 0x00000500;
R
Rusty Russell 已提交
448
		break;
R
Rusty Russell 已提交
449 450 451 452
	/*
	 * 0x80000000 returns the highest Extended Function, so we futureproof
	 * like we do above by limiting it to known fields.
	 */
R
Rusty Russell 已提交
453
	case 0x80000000:
454 455
		if (*ax > 0x80000008)
			*ax = 0x80000008;
R
Rusty Russell 已提交
456
		break;
R
Rusty Russell 已提交
457 458 459 460 461 462 463

	/*
	 * PAE systems can mark pages as non-executable.  Linux calls this the
	 * NX bit.  Intel calls it XD (eXecute Disable), AMD EVP (Enhanced
	 * Virus Protection).  We just switch turn if off here, since we don't
	 * support it.
	 */
M
Matias Zabaljauregui 已提交
464 465 466
	case 0x80000001:
		*dx &= ~(1 << 20);
		break;
R
Rusty Russell 已提交
467 468 469
	}
}

R
Rusty Russell 已提交
470 471
/*
 * Intel has four control registers, imaginatively named cr0, cr2, cr3 and cr4.
R
Rusty Russell 已提交
472 473 474 475
 * 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().
 *
R
Rusty Russell 已提交
476
 * We start with cr0.  cr0 allows you to turn on and off all kinds of basic
R
Rusty Russell 已提交
477 478 479 480 481 482 483 484
 * 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?
 *
R
Rusty Russell 已提交
485
 * We store cr0 locally because the Host never changes it.  The Guest sometimes
R
Rusty Russell 已提交
486 487
 * wants to read it and we'd prefer not to bother the Host unnecessarily.
 */
R
Rusty Russell 已提交
488
static unsigned long current_cr0;
R
Rusty Russell 已提交
489 490
static void lguest_write_cr0(unsigned long val)
{
491
	lazy_hcall1(LHCALL_TS, val & X86_CR0_TS);
R
Rusty Russell 已提交
492 493 494 495 496 497 498 499
	current_cr0 = val;
}

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

R
Rusty Russell 已提交
500 501
/*
 * Intel provided a special instruction to clear the TS bit for people too cool
R
Rusty Russell 已提交
502
 * to use write_cr0() to do it.  This "clts" instruction is faster, because all
R
Rusty Russell 已提交
503 504
 * the vowels have been optimized out.
 */
R
Rusty Russell 已提交
505 506
static void lguest_clts(void)
{
507
	lazy_hcall1(LHCALL_TS, 0);
508
	current_cr0 &= ~X86_CR0_TS;
R
Rusty Russell 已提交
509 510
}

R
Rusty Russell 已提交
511 512
/*
 * cr2 is the virtual address of the last page fault, which the Guest only ever
R
Rusty Russell 已提交
513
 * reads.  The Host kindly writes this into our "struct lguest_data", so we
R
Rusty Russell 已提交
514 515
 * just read it out of there.
 */
R
Rusty Russell 已提交
516 517 518 519 520
static unsigned long lguest_read_cr2(void)
{
	return lguest_data.cr2;
}

R
Rusty Russell 已提交
521 522 523
/* See lguest_set_pte() below. */
static bool cr3_changed = false;

R
Rusty Russell 已提交
524 525
/*
 * cr3 is the current toplevel pagetable page: the principle is the same as
R
Rusty Russell 已提交
526 527
 * 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
R
Rusty Russell 已提交
528 529
 * to set it upon our initial hypercall.
 */
R
Rusty Russell 已提交
530 531
static void lguest_write_cr3(unsigned long cr3)
{
R
Rusty Russell 已提交
532
	lguest_data.pgdir = cr3;
533
	lazy_hcall1(LHCALL_NEW_PGTABLE, cr3);
R
Rusty Russell 已提交
534
	cr3_changed = true;
R
Rusty Russell 已提交
535 536 537 538
}

static unsigned long lguest_read_cr3(void)
{
R
Rusty Russell 已提交
539
	return lguest_data.pgdir;
R
Rusty Russell 已提交
540 541
}

R
Rusty Russell 已提交
542
/* cr4 is used to enable and disable PGE, but we don't care. */
R
Rusty Russell 已提交
543 544 545 546 547 548 549 550 551
static unsigned long lguest_read_cr4(void)
{
	return 0;
}

static void lguest_write_cr4(unsigned long val)
{
}

R
Rusty Russell 已提交
552 553 554 555 556 557 558 559 560 561 562
/*
 * 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
R
Rusty Russell 已提交
563
 * are unused, we use a two level index which saves space.  The cr3 register
R
Rusty Russell 已提交
564 565 566 567 568 569 570
 * 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:
 *
R
Rusty Russell 已提交
571
 * cr3 ---> +---------+
R
Rusty Russell 已提交
572 573
 *	    |  	   --------->+---------+
 *	    |	      |	     | PADDR1  |
R
Rusty Russell 已提交
574
 *	  Mid-level   |	     | PADDR2  |
R
Rusty Russell 已提交
575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593
 *	  (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
 *
R
Rusty Russell 已提交
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
 * Now, unfortunately, this isn't the whole story: Intel added Physical Address
 * Extension (PAE) to allow 32 bit systems to use 64GB of memory (ie. 36 bits).
 * These are held in 64-bit page table entries, so we can now only fit 512
 * entries in a page, and the neat three-level tree breaks down.
 *
 * The result is a four level page table:
 *
 * cr3 --> [ 4 Upper  ]
 *	   [   Level  ]
 *	   [  Entries ]
 *	   [(PUD Page)]---> +---------+
 *	 		    |  	   --------->+---------+
 *	 		    |	      |	     | PADDR1  |
 *	 		  Mid-level   |	     | PADDR2  |
 *	 		  (PMD) page  |	     | 	       |
 *	 		    |	      |	   Lower-level |
 *	 		    |	      |	   (PTE) page  |
 *	 		    |	      |	     |	       |
 *	 		      ....    	     	 ....
 *
 *
 * And the virtual address is decoded as:
 *
 *         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
 *      |<-2->|<--- 9 bits ---->|<---- 9 bits --->|<------ 12 bits ------>|
 * Index into    Index into mid    Index into lower    Offset within page
 * top entries   directory page     pagetable page
 *
 * It's too hard to switch between these two formats at runtime, so Linux only
 * supports one or the other depending on whether CONFIG_X86_PAE is set.  Many
 * distributions turn it on, and not just for people with silly amounts of
 * memory: the larger PTE entries allow room for the NX bit, which lets the
 * kernel disable execution of pages and increase security.
 *
 * This was a problem for lguest, which couldn't run on these distributions;
 * then Matias Zabaljauregui figured it all out and implemented it, and only a
 * handful of puppies were crushed in the process!
 *
 * Back to our point: 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'.
R
Rusty Russell 已提交
637 638
 */

R
Rusty Russell 已提交
639
/*
R
Rusty Russell 已提交
640 641 642 643
 * The Guest calls this after it has set a second-level entry (pte), ie. to map
 * a page into a process' address space.  Wetell 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).
R
Rusty Russell 已提交
644
 */
645 646 647
static void lguest_pte_update(struct mm_struct *mm, unsigned long addr,
			       pte_t *ptep)
{
M
Matias Zabaljauregui 已提交
648
#ifdef CONFIG_X86_PAE
R
Rusty Russell 已提交
649
	/* PAE needs to hand a 64 bit page table entry, so it uses two args. */
M
Matias Zabaljauregui 已提交
650 651 652
	lazy_hcall4(LHCALL_SET_PTE, __pa(mm->pgd), addr,
		    ptep->pte_low, ptep->pte_high);
#else
653
	lazy_hcall3(LHCALL_SET_PTE, __pa(mm->pgd), addr, ptep->pte_low);
M
Matias Zabaljauregui 已提交
654
#endif
655 656
}

R
Rusty Russell 已提交
657
/* This is the "set and update" combo-meal-deal version. */
R
Rusty Russell 已提交
658 659 660
static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
			      pte_t *ptep, pte_t pteval)
{
661
	native_set_pte(ptep, pteval);
662
	lguest_pte_update(mm, addr, ptep);
R
Rusty Russell 已提交
663 664
}

R
Rusty Russell 已提交
665 666
/*
 * The Guest calls lguest_set_pud to set a top-level entry and lguest_set_pmd
M
Matias Zabaljauregui 已提交
667
 * to set a middle-level entry when PAE is activated.
R
Rusty Russell 已提交
668
 *
M
Matias Zabaljauregui 已提交
669
 * Again, we set the entry then tell the Host which page we changed,
R
Rusty Russell 已提交
670 671
 * and the index of the entry we changed.
 */
M
Matias Zabaljauregui 已提交
672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689
#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

R
Rusty Russell 已提交
690
/* The Guest calls lguest_set_pmd to set a top-level entry when !PAE. */
R
Rusty Russell 已提交
691 692
static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
{
693
	native_set_pmd(pmdp, pmdval);
694
	lazy_hcall2(LHCALL_SET_PGD, __pa(pmdp) & PAGE_MASK,
695
		   (__pa(pmdp) & (PAGE_SIZE - 1)) / sizeof(pmd_t));
R
Rusty Russell 已提交
696
}
M
Matias Zabaljauregui 已提交
697
#endif
R
Rusty Russell 已提交
698

R
Rusty Russell 已提交
699 700
/*
 * There are a couple of legacy places where the kernel sets a PTE, but we
R
Rusty Russell 已提交
701 702 703 704 705
 * 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,
R
Rusty Russell 已提交
706 707
 * 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,
R
Rusty Russell 已提交
708 709
 * which brings boot back to 0.25 seconds.
 */
R
Rusty Russell 已提交
710 711
static void lguest_set_pte(pte_t *ptep, pte_t pteval)
{
712
	native_set_pte(ptep, pteval);
R
Rusty Russell 已提交
713
	if (cr3_changed)
714
		lazy_hcall1(LHCALL_FLUSH_TLB, 1);
R
Rusty Russell 已提交
715 716
}

M
Matias Zabaljauregui 已提交
717
#ifdef CONFIG_X86_PAE
R
Rusty Russell 已提交
718 719 720 721 722
/*
 * With 64-bit PTE values, we need to be careful setting them: if we set 32
 * bits at a time, the hardware could see a weird half-set entry.  These
 * versions ensure we update all 64 bits at once.
 */
M
Matias Zabaljauregui 已提交
723 724 725 726 727 728 729
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);
}

R
Rusty Russell 已提交
730 731
static void lguest_pte_clear(struct mm_struct *mm, unsigned long addr,
			     pte_t *ptep)
M
Matias Zabaljauregui 已提交
732 733 734 735 736
{
	native_pte_clear(mm, addr, ptep);
	lguest_pte_update(mm, addr, ptep);
}

R
Rusty Russell 已提交
737
static void lguest_pmd_clear(pmd_t *pmdp)
M
Matias Zabaljauregui 已提交
738 739 740 741 742
{
	lguest_set_pmd(pmdp, __pmd(0));
}
#endif

R
Rusty Russell 已提交
743 744
/*
 * Unfortunately for Lguest, the pv_mmu_ops for page tables were based on
R
Rusty Russell 已提交
745 746 747 748 749 750 751 752
 * 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
R
Rusty Russell 已提交
753 754
 * bit is zero).
 */
R
Rusty Russell 已提交
755 756
static void lguest_flush_tlb_single(unsigned long addr)
{
R
Rusty Russell 已提交
757
	/* Simply set it to zero: if it was not, it will fault back in. */
758
	lazy_hcall3(LHCALL_SET_PTE, lguest_data.pgdir, addr, 0);
R
Rusty Russell 已提交
759 760
}

R
Rusty Russell 已提交
761 762
/*
 * This is what happens after the Guest has removed a large number of entries.
R
Rusty Russell 已提交
763
 * This tells the Host that any of the page table entries for userspace might
R
Rusty Russell 已提交
764 765
 * have changed, ie. virtual addresses below PAGE_OFFSET.
 */
R
Rusty Russell 已提交
766 767
static void lguest_flush_tlb_user(void)
{
768
	lazy_hcall1(LHCALL_FLUSH_TLB, 0);
R
Rusty Russell 已提交
769 770
}

R
Rusty Russell 已提交
771 772
/*
 * This is called when the kernel page tables have changed.  That's not very
R
Rusty Russell 已提交
773
 * common (unless the Guest is using highmem, which makes the Guest extremely
R
Rusty Russell 已提交
774 775
 * slow), so it's worth separating this from the user flushing above.
 */
R
Rusty Russell 已提交
776 777
static void lguest_flush_tlb_kernel(void)
{
778
	lazy_hcall1(LHCALL_FLUSH_TLB, 1);
R
Rusty Russell 已提交
779 780
}

R
Rusty Russell 已提交
781 782 783 784 785 786 787 788 789 790 791 792 793
/*
 * 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?
 */
794
static void disable_lguest_irq(struct irq_data *data)
R
Rusty Russell 已提交
795
{
796
	set_bit(data->irq, lguest_data.blocked_interrupts);
R
Rusty Russell 已提交
797 798
}

799
static void enable_lguest_irq(struct irq_data *data)
R
Rusty Russell 已提交
800
{
801
	clear_bit(data->irq, lguest_data.blocked_interrupts);
R
Rusty Russell 已提交
802 803
}

R
Rusty Russell 已提交
804
/* This structure describes the lguest IRQ controller. */
R
Rusty Russell 已提交
805 806
static struct irq_chip lguest_irq_controller = {
	.name		= "lguest",
807 808 809
	.irq_mask	= disable_lguest_irq,
	.irq_mask_ack	= disable_lguest_irq,
	.irq_unmask	= enable_lguest_irq,
R
Rusty Russell 已提交
810 811
};

R
Rusty Russell 已提交
812 813
/*
 * This sets up the Interrupt Descriptor Table (IDT) entry for each hardware
R
Rusty Russell 已提交
814 815
 * interrupt (except 128, which is used for system calls), and then tells the
 * Linux infrastructure that each interrupt is controlled by our level-based
R
Rusty Russell 已提交
816 817
 * lguest interrupt controller.
 */
R
Rusty Russell 已提交
818 819 820 821
static void __init lguest_init_IRQ(void)
{
	unsigned int i;

R
Rusty Russell 已提交
822
	for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) {
R
Rusty Russell 已提交
823
		/* Some systems map "vectors" to interrupts weirdly.  Not us! */
R
Rusty Russell 已提交
824 825 826
		__get_cpu_var(vector_irq)[i] = i - FIRST_EXTERNAL_VECTOR;
		if (i != SYSCALL_VECTOR)
			set_intr_gate(i, interrupt[i - FIRST_EXTERNAL_VECTOR]);
R
Rusty Russell 已提交
827
	}
R
Rusty Russell 已提交
828 829 830 831 832

	/*
	 * This call is required to set up for 4k stacks, where we have
	 * separate stacks for hard and soft interrupts.
	 */
R
Rusty Russell 已提交
833 834 835
	irq_ctx_init(smp_processor_id());
}

R
Rusty Russell 已提交
836 837 838 839 840 841 842 843
/*
 * With CONFIG_SPARSE_IRQ, interrupt descriptors are allocated as-needed, so
 * rather than set them in lguest_init_IRQ we are called here every time an
 * lguest device needs an interrupt.
 *
 * FIXME: irq_to_desc_alloc_node() can fail due to lack of memory, we should
 * pass that up!
 */
844 845
void lguest_setup_irq(unsigned int irq)
{
846
	irq_to_desc_alloc_node(irq, 0);
847 848 849 850
	set_irq_chip_and_handler_name(irq, &lguest_irq_controller,
				      handle_level_irq, "level");
}

R
Rusty Russell 已提交
851 852 853 854
/*
 * Time.
 *
 * It would be far better for everyone if the Guest had its own clock, but
855
 * until then the Host gives us the time on every interrupt.
R
Rusty Russell 已提交
856
 */
R
Rusty Russell 已提交
857 858
static unsigned long lguest_get_wallclock(void)
{
859
	return lguest_data.time.tv_sec;
R
Rusty Russell 已提交
860 861
}

R
Rusty Russell 已提交
862 863
/*
 * The TSC is an Intel thing called the Time Stamp Counter.  The Host tells us
864 865
 * 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
R
Rusty Russell 已提交
866 867
 * TSC clock will give up and not register itself.
 */
868
static unsigned long lguest_tsc_khz(void)
869 870 871 872
{
	return lguest_data.tsc_khz;
}

R
Rusty Russell 已提交
873 874 875 876
/*
 * 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.
 */
877
static cycle_t lguest_clock_read(struct clocksource *cs)
R
Rusty Russell 已提交
878
{
879 880
	unsigned long sec, nsec;

R
Rusty Russell 已提交
881 882
	/*
	 * Since the time is in two parts (seconds and nanoseconds), we risk
883 884
	 * 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
R
Rusty Russell 已提交
885 886
	 * of time travel, we must be careful:
	 */
887 888 889
	do {
		/* First we read the seconds part. */
		sec = lguest_data.time.tv_sec;
R
Rusty Russell 已提交
890 891
		/*
		 * This read memory barrier tells the compiler and the CPU that
892
		 * this can't be reordered: we have to complete the above
R
Rusty Russell 已提交
893 894
		 * before going on.
		 */
895 896 897 898 899 900 901 902
		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));

903
	/* Our lguest clock is in real nanoseconds. */
904
	return sec*1000000000ULL + nsec;
R
Rusty Russell 已提交
905 906
}

907
/* This is the fallback clocksource: lower priority than the TSC clocksource. */
R
Rusty Russell 已提交
908 909
static struct clocksource lguest_clock = {
	.name		= "lguest",
910
	.rating		= 200,
R
Rusty Russell 已提交
911
	.read		= lguest_clock_read,
912
	.mask		= CLOCKSOURCE_MASK(64),
913 914
	.mult		= 1 << 22,
	.shift		= 22,
915
	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
R
Rusty Russell 已提交
916 917
};

R
Rusty Russell 已提交
918 919
/*
 * We also need a "struct clock_event_device": Linux asks us to set it to go
R
Rusty Russell 已提交
920
 * off some time in the future.  Actually, James Morris figured all this out, I
R
Rusty Russell 已提交
921 922
 * just applied the patch.
 */
R
Rusty Russell 已提交
923 924 925
static int lguest_clockevent_set_next_event(unsigned long delta,
                                           struct clock_event_device *evt)
{
926 927
	/* 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? */
R
Rusty Russell 已提交
928 929 930
	if (delta < LG_CLOCK_MIN_DELTA) {
		if (printk_ratelimit())
			printk(KERN_DEBUG "%s: small delta %lu ns\n",
931
			       __func__, delta);
R
Rusty Russell 已提交
932 933
		return -ETIME;
	}
934 935

	/* Please wake us this far in the future. */
936
	hcall(LHCALL_SET_CLOCKEVENT, delta, 0, 0, 0);
R
Rusty Russell 已提交
937 938 939 940 941 942 943 944 945 946
	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. */
947
		hcall(LHCALL_SET_CLOCKEVENT, 0, 0, 0, 0);
R
Rusty Russell 已提交
948 949 950 951 952 953
		break;
	case CLOCK_EVT_MODE_ONESHOT:
		/* This is what we expect. */
		break;
	case CLOCK_EVT_MODE_PERIODIC:
		BUG();
T
Thomas Gleixner 已提交
954 955
	case CLOCK_EVT_MODE_RESUME:
		break;
R
Rusty Russell 已提交
956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971
	}
}

/* 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,
};

R
Rusty Russell 已提交
972 973 974 975
/*
 * This is the Guest timer interrupt handler (hardware interrupt 0).  We just
 * call the clockevent infrastructure and it does whatever needs doing.
 */
R
Rusty Russell 已提交
976 977
static void lguest_time_irq(unsigned int irq, struct irq_desc *desc)
{
R
Rusty Russell 已提交
978 979 980 981 982 983
	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);
R
Rusty Russell 已提交
984 985
}

R
Rusty Russell 已提交
986 987
/*
 * At some point in the boot process, we get asked to set up our timing
R
Rusty Russell 已提交
988 989
 * infrastructure.  The kernel doesn't expect timer interrupts before this, but
 * we cleverly initialized the "blocked_interrupts" field of "struct
R
Rusty Russell 已提交
990 991
 * lguest_data" so that timer interrupts were blocked until now.
 */
R
Rusty Russell 已提交
992 993
static void lguest_time_init(void)
{
R
Rusty Russell 已提交
994
	/* Set up the timer interrupt (0) to go to our simple timer routine */
R
Rusty Russell 已提交
995 996
	set_irq_handler(0, lguest_time_irq);

R
Rusty Russell 已提交
997 998
	clocksource_register(&lguest_clock);

R
Rusty Russell 已提交
999 1000
	/* We can't set cpumask in the initializer: damn C limitations!  Set it
	 * here and register our timer device. */
1001
	lguest_clockevent.cpumask = cpumask_of(0);
R
Rusty Russell 已提交
1002 1003
	clockevents_register_device(&lguest_clockevent);

R
Rusty Russell 已提交
1004
	/* Finally, we unblock the timer interrupt. */
R
Rusty Russell 已提交
1005
	enable_lguest_irq(0);
R
Rusty Russell 已提交
1006 1007
}

R
Rusty Russell 已提交
1008 1009 1010 1011 1012 1013 1014
/*
 * Miscellaneous bits and pieces.
 *
 * Here is an oddball collection of functions which the Guest needs for things
 * to work.  They're pretty simple.
 */

R
Rusty Russell 已提交
1015 1016
/*
 * The Guest needs to tell the Host what stack it expects traps to use.  For
R
Rusty Russell 已提交
1017 1018 1019 1020 1021 1022
 * 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
R
Rusty Russell 已提交
1023 1024
 * of pages in the stack.
 */
1025
static void lguest_load_sp0(struct tss_struct *tss,
1026
			    struct thread_struct *thread)
R
Rusty Russell 已提交
1027
{
1028 1029
	lazy_hcall3(LHCALL_SET_STACK, __KERNEL_DS | 0x1, thread->sp0,
		   THREAD_SIZE / PAGE_SIZE);
R
Rusty Russell 已提交
1030 1031
}

R
Rusty Russell 已提交
1032
/* Let's just say, I wouldn't do debugging under a Guest. */
R
Rusty Russell 已提交
1033 1034 1035 1036 1037
static void lguest_set_debugreg(int regno, unsigned long value)
{
	/* FIXME: Implement */
}

R
Rusty Russell 已提交
1038 1039
/*
 * There are times when the kernel wants to make sure that no memory writes are
R
Rusty Russell 已提交
1040 1041 1042 1043 1044 1045 1046 1047 1048
 * 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.
 */
R
Rusty Russell 已提交
1049 1050 1051 1052
static void lguest_wbinvd(void)
{
}

R
Rusty Russell 已提交
1053 1054
/*
 * If the Guest expects to have an Advanced Programmable Interrupt Controller,
R
Rusty Russell 已提交
1055 1056 1057
 * 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
R
Rusty Russell 已提交
1058 1059
 * does, however, allow us to get through the Linux boot code.
 */
R
Rusty Russell 已提交
1060
#ifdef CONFIG_X86_LOCAL_APIC
1061
static void lguest_apic_write(u32 reg, u32 v)
R
Rusty Russell 已提交
1062 1063 1064
{
}

1065
static u32 lguest_apic_read(u32 reg)
R
Rusty Russell 已提交
1066 1067 1068
{
	return 0;
}
S
Suresh Siddha 已提交
1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090

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;
}

Y
Yinghai Lu 已提交
1091 1092 1093 1094 1095 1096 1097 1098
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;
S
Suresh Siddha 已提交
1099
};
R
Rusty Russell 已提交
1100 1101
#endif

R
Rusty Russell 已提交
1102
/* STOP!  Until an interrupt comes in. */
R
Rusty Russell 已提交
1103 1104
static void lguest_safe_halt(void)
{
1105
	hcall(LHCALL_HALT, 0, 0, 0, 0);
R
Rusty Russell 已提交
1106 1107
}

R
Rusty Russell 已提交
1108 1109
/*
 * The SHUTDOWN hypercall takes a string to describe what's happening, and
1110
 * an argument which says whether this to restart (reboot) the Guest or not.
R
Rusty Russell 已提交
1111 1112
 *
 * Note that the Host always prefers that the Guest speak in physical addresses
R
Rusty Russell 已提交
1113 1114
 * rather than virtual addresses, so we use __pa() here.
 */
R
Rusty Russell 已提交
1115 1116
static void lguest_power_off(void)
{
1117 1118
	hcall(LHCALL_SHUTDOWN, __pa("Power down"),
	      LGUEST_SHUTDOWN_POWEROFF, 0, 0);
R
Rusty Russell 已提交
1119 1120
}

R
Rusty Russell 已提交
1121 1122 1123 1124 1125
/*
 * Panicing.
 *
 * Don't.  But if you did, this is what happens.
 */
R
Rusty Russell 已提交
1126 1127
static int lguest_panic(struct notifier_block *nb, unsigned long l, void *p)
{
1128
	hcall(LHCALL_SHUTDOWN, __pa(p), LGUEST_SHUTDOWN_POWEROFF, 0, 0);
R
Rusty Russell 已提交
1129
	/* The hcall won't return, but to keep gcc happy, we're "done". */
R
Rusty Russell 已提交
1130 1131 1132 1133 1134 1135 1136
	return NOTIFY_DONE;
}

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

R
Rusty Russell 已提交
1137
/* Setting up memory is fairly easy. */
R
Rusty Russell 已提交
1138 1139
static __init char *lguest_memory_setup(void)
{
R
Rusty Russell 已提交
1140 1141 1142 1143
	/*
	 *The Linux bootloader header contains an "e820" memory map: the
	 * Launcher populated the first entry with our memory limit.
	 */
1144
	e820_add_region(boot_params.e820_map[0].addr,
1145 1146
			  boot_params.e820_map[0].size,
			  boot_params.e820_map[0].type);
R
Rusty Russell 已提交
1147 1148

	/* This string is for the boot messages. */
R
Rusty Russell 已提交
1149 1150 1151
	return "LGUEST";
}

R
Rusty Russell 已提交
1152 1153
/*
 * We will eventually use the virtio console device to produce console output,
R
Rusty Russell 已提交
1154
 * but before that is set up we use LHCALL_NOTIFY on normal memory to produce
R
Rusty Russell 已提交
1155 1156
 * console output.
 */
R
Rusty Russell 已提交
1157 1158 1159 1160 1161
static __init int early_put_chars(u32 vtermno, const char *buf, int count)
{
	char scratch[17];
	unsigned int len = count;

R
Rusty Russell 已提交
1162
	/* We use a nul-terminated string, so we make a copy.  Icky, huh? */
R
Rusty Russell 已提交
1163 1164 1165 1166
	if (len > sizeof(scratch) - 1)
		len = sizeof(scratch) - 1;
	scratch[len] = '\0';
	memcpy(scratch, buf, len);
1167
	hcall(LHCALL_NOTIFY, __pa(scratch), 0, 0, 0);
R
Rusty Russell 已提交
1168 1169 1170 1171 1172

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

R
Rusty Russell 已提交
1173 1174 1175 1176
/*
 * Rebooting also tells the Host we're finished, but the RESTART flag tells the
 * Launcher to reboot us.
 */
1177 1178
static void lguest_restart(char *reason)
{
1179
	hcall(LHCALL_SHUTDOWN, __pa(reason), LGUEST_SHUTDOWN_RESTART, 0, 0);
1180 1181
}

R
Rusty Russell 已提交
1182 1183 1184
/*G:050
 * Patching (Powerfully Placating Performance Pedants)
 *
1185 1186 1187
 * 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
R
Rusty Russell 已提交
1188 1189 1190 1191 1192 1193 1194 1195 1196
 * 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
1197 1198 1199 1200
 * 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.
R
Rusty Russell 已提交
1201 1202
 *
 * First we need assembly templates of each of the patchable Guest operations,
R
Rusty Russell 已提交
1203 1204
 * and these are in i386_head.S.
 */
R
Rusty Russell 已提交
1205 1206

/*G:060 We construct a table from the assembler templates: */
R
Rusty Russell 已提交
1207 1208 1209 1210
static const struct lguest_insns
{
	const char *start, *end;
} lguest_insns[] = {
1211 1212
	[PARAVIRT_PATCH(pv_irq_ops.irq_disable)] = { lgstart_cli, lgend_cli },
	[PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf },
R
Rusty Russell 已提交
1213
};
R
Rusty Russell 已提交
1214

R
Rusty Russell 已提交
1215 1216
/*
 * Now our patch routine is fairly simple (based on the native one in
R
Rusty Russell 已提交
1217
 * paravirt.c).  If we have a replacement, we copy it in and return how much of
R
Rusty Russell 已提交
1218 1219
 * the available space we used.
 */
1220 1221
static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
			     unsigned long addr, unsigned len)
R
Rusty Russell 已提交
1222 1223 1224
{
	unsigned int insn_len;

R
Rusty Russell 已提交
1225
	/* Don't do anything special if we don't have a replacement */
R
Rusty Russell 已提交
1226
	if (type >= ARRAY_SIZE(lguest_insns) || !lguest_insns[type].start)
1227
		return paravirt_patch_default(type, clobber, ibuf, addr, len);
R
Rusty Russell 已提交
1228 1229 1230

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

R
Rusty Russell 已提交
1231
	/* Similarly if it can't fit (doesn't happen, but let's be thorough). */
R
Rusty Russell 已提交
1232
	if (len < insn_len)
1233
		return paravirt_patch_default(type, clobber, ibuf, addr, len);
R
Rusty Russell 已提交
1234

R
Rusty Russell 已提交
1235
	/* Copy in our instructions. */
1236
	memcpy(ibuf, lguest_insns[type].start, insn_len);
R
Rusty Russell 已提交
1237 1238 1239
	return insn_len;
}

R
Rusty Russell 已提交
1240 1241
/*G:029
 * Once we get to lguest_init(), we know we're a Guest.  The various
1242
 * pv_ops structures in the kernel provide points for (almost) every routine we
R
Rusty Russell 已提交
1243 1244
 * have to override to avoid privileged instructions.
 */
1245
__init void lguest_init(void)
R
Rusty Russell 已提交
1246
{
R
Rusty Russell 已提交
1247
	/* We're under lguest. */
1248
	pv_info.name = "lguest";
R
Rusty Russell 已提交
1249
	/* Paravirt is enabled. */
1250
	pv_info.paravirt_enabled = 1;
R
Rusty Russell 已提交
1251
	/* We're running at privilege level 1, not 0 as normal. */
1252
	pv_info.kernel_rpl = 1;
R
Rusty Russell 已提交
1253
	/* Everyone except Xen runs with this set. */
M
Matias Zabaljauregui 已提交
1254
	pv_info.shared_kernel_pmd = 1;
R
Rusty Russell 已提交
1255

R
Rusty Russell 已提交
1256 1257 1258 1259
	/*
	 * We set up all the lguest overrides for sensitive operations.  These
	 * are detailed with the operations themselves.
	 */
1260

R
Rusty Russell 已提交
1261
	/* Interrupt-related operations */
1262
	pv_irq_ops.save_fl = PV_CALLEE_SAVE(save_fl);
1263
	pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(lg_restore_fl);
1264
	pv_irq_ops.irq_disable = PV_CALLEE_SAVE(irq_disable);
1265
	pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(lg_irq_enable);
1266 1267
	pv_irq_ops.safe_halt = lguest_safe_halt;

R
Rusty Russell 已提交
1268
	/* Setup operations */
1269 1270
	pv_init_ops.patch = lguest_patch;

R
Rusty Russell 已提交
1271
	/* Intercepts of various CPU instructions */
1272 1273 1274 1275
	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;
1276
	pv_cpu_ops.load_sp0 = lguest_load_sp0;
1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288
	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;
1289 1290
	pv_cpu_ops.start_context_switch = paravirt_start_context_switch;
	pv_cpu_ops.end_context_switch = lguest_end_context_switch;
1291

R
Rusty Russell 已提交
1292
	/* Pagetable management */
1293 1294 1295 1296 1297 1298 1299
	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;
M
Matias Zabaljauregui 已提交
1300 1301 1302 1303 1304 1305
#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
1306 1307
	pv_mmu_ops.read_cr2 = lguest_read_cr2;
	pv_mmu_ops.read_cr3 = lguest_read_cr3;
1308
	pv_mmu_ops.lazy_mode.enter = paravirt_enter_lazy_mmu;
1309
	pv_mmu_ops.lazy_mode.leave = lguest_leave_lazy_mmu_mode;
1310 1311
	pv_mmu_ops.pte_update = lguest_pte_update;
	pv_mmu_ops.pte_update_defer = lguest_pte_update;
1312

R
Rusty Russell 已提交
1313
#ifdef CONFIG_X86_LOCAL_APIC
R
Rusty Russell 已提交
1314
	/* APIC read/write intercepts */
Y
Yinghai Lu 已提交
1315
	set_lguest_basic_apic_ops();
R
Rusty Russell 已提交
1316
#endif
1317

1318
	x86_init.resources.memory_setup = lguest_memory_setup;
1319
	x86_init.irqs.intr_init = lguest_init_IRQ;
1320
	x86_init.timers.timer_init = lguest_time_init;
1321
	x86_platform.calibrate_tsc = lguest_tsc_khz;
1322
	x86_platform.get_wallclock =  lguest_get_wallclock;
1323

R
Rusty Russell 已提交
1324 1325 1326 1327
	/*
	 * Now is a good time to look at the implementations of these functions
	 * before returning to the rest of lguest_init().
	 */
R
Rusty Russell 已提交
1328

R
Rusty Russell 已提交
1329 1330
	/*G:070
	 * Now we've seen all the paravirt_ops, we return to
R
Rusty Russell 已提交
1331
	 * lguest_init() where the rest of the fairly chaotic boot setup
R
Rusty Russell 已提交
1332 1333
	 * occurs.
	 */
R
Rusty Russell 已提交
1334

R
Rusty Russell 已提交
1335 1336
	/*
	 * The stack protector is a weird thing where gcc places a canary
1337 1338 1339 1340
	 * 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
R
Rusty Russell 已提交
1341 1342
	 * in other files.
	 */
1343
	setup_stack_canary_segment(0);
R
Rusty Russell 已提交
1344 1345 1346 1347 1348 1349

	/*
	 * We could just call load_stack_canary_segment(), but we might as well
	 * call switch_to_new_gdt() which loads the whole table and sets up the
	 * per-cpu segment descriptor register %fs as well.
	 */
1350 1351
	switch_to_new_gdt(0);

R
Rusty Russell 已提交
1352
	/* We actually boot with all memory mapped, but let's say 128MB. */
1353 1354
	max_pfn_mapped = (128*1024*1024) >> PAGE_SHIFT;

R
Rusty Russell 已提交
1355 1356
	/*
	 * The Host<->Guest Switcher lives at the top of our address space, and
1357
	 * the Host told us how big it is when we made LGUEST_INIT hypercall:
R
Rusty Russell 已提交
1358 1359
	 * it put the answer in lguest_data.reserve_mem
	 */
R
Rusty Russell 已提交
1360 1361
	reserve_top_address(lguest_data.reserve_mem);

R
Rusty Russell 已提交
1362 1363
	/*
	 * If we don't initialize the lock dependency checker now, it crashes
1364
	 * atomic_notifier_chain_register, then paravirt_disable_iospace.
R
Rusty Russell 已提交
1365
	 */
R
Rusty Russell 已提交
1366 1367
	lockdep_init();

1368 1369 1370
	/* Hook in our special panic hypercall code. */
	atomic_notifier_chain_register(&panic_notifier_list, &paniced);

R
Rusty Russell 已提交
1371 1372
	/*
	 * The IDE code spends about 3 seconds probing for disks: if we reserve
R
Rusty Russell 已提交
1373 1374
	 * 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
R
Rusty Russell 已提交
1375 1376
	 * kernel boot time on my machine from 4.1 seconds to 0.45 seconds.
	 */
R
Rusty Russell 已提交
1377 1378
	paravirt_disable_iospace();

R
Rusty Russell 已提交
1379 1380 1381 1382
	/*
	 * This is messy CPU setup stuff which the native boot code does before
	 * start_kernel, so we have to do, too:
	 */
R
Rusty Russell 已提交
1383 1384 1385 1386 1387 1388 1389
	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;

1390
	/* We don't have features.  We have puppies!  Puppies! */
R
Rusty Russell 已提交
1391 1392 1393 1394 1395 1396 1397
#ifdef CONFIG_X86_MCE
	mce_disabled = 1;
#endif
#ifdef CONFIG_ACPI
	acpi_disabled = 1;
#endif

R
Rusty Russell 已提交
1398 1399
	/*
	 * We set the preferred console to "hvc".  This is the "hypervisor
R
Rusty Russell 已提交
1400
	 * virtual console" driver written by the PowerPC people, which we also
R
Rusty Russell 已提交
1401 1402
	 * adapted for lguest's use.
	 */
R
Rusty Russell 已提交
1403 1404
	add_preferred_console("hvc", 0, NULL);

R
Rusty Russell 已提交
1405 1406 1407
	/* Register our very early console. */
	virtio_cons_early_init(early_put_chars);

R
Rusty Russell 已提交
1408 1409
	/*
	 * Last of all, we set the power management poweroff hook to point to
1410
	 * the Guest routine to power off, and the reboot hook to our restart
R
Rusty Russell 已提交
1411 1412
	 * routine.
	 */
R
Rusty Russell 已提交
1413
	pm_power_off = lguest_power_off;
B
Balaji Rao 已提交
1414
	machine_ops.restart = lguest_restart;
1415

R
Rusty Russell 已提交
1416 1417 1418 1419
	/*
	 * Now we're set up, call i386_start_kernel() in head32.c and we proceed
	 * to boot as normal.  It never returns.
	 */
1420
	i386_start_kernel();
R
Rusty Russell 已提交
1421
}
R
Rusty Russell 已提交
1422 1423 1424
/*
 * This marks the end of stage II of our journey, The Guest.
 *
R
Rusty Russell 已提交
1425 1426
 * It is now time for us to explore the layer of virtual drivers and complete
 * our understanding of the Guest in "make Drivers".
R
Rusty Russell 已提交
1427
 */