core.c 23.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
/*
 * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
 * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> SGI.
 *
 * 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.
 */
R
Rusty Russell 已提交
20 21
/*P:450
 * This file contains the x86-specific lguest code.  It used to be all
22 23 24 25 26
 * mixed in with drivers/lguest/core.c but several foolhardy code slashers
 * wrestled most of the dependencies out to here in preparation for porting
 * lguest to other architectures (see what I mean by foolhardy?).
 *
 * This also contains a couple of non-obvious setup and teardown pieces which
R
Rusty Russell 已提交
27 28
 * were implemented after days of debugging pain.
:*/
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
#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/cpu.h>
#include <linux/lguest.h>
#include <linux/lguest_launcher.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/lguest.h>
#include <asm/uaccess.h>
#include <asm/i387.h>
#include "../lg.h"

static int cpu_had_pge;

static struct {
	unsigned long offset;
	unsigned short segment;
} lguest_entry;

/* Offset from where switcher.S was compiled to where we've copied it */
static unsigned long switcher_offset(void)
{
	return SWITCHER_ADDR - (unsigned long)start_switcher_text;
}

/* This cpu's struct lguest_pages. */
static struct lguest_pages *lguest_pages(unsigned int cpu)
{
	return &(((struct lguest_pages *)
		  (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
}

72
static DEFINE_PER_CPU(struct lg_cpu *, lg_last_cpu);
73 74

/*S:010
75
 * We approach the Switcher.
76 77 78 79 80 81 82 83 84
 *
 * Remember that each CPU has two pages which are visible to the Guest when it
 * runs on that CPU.  This has to contain the state for that Guest: we copy the
 * state in just before we run the Guest.
 *
 * Each Guest has "changed" flags which indicate what has changed in the Guest
 * since it last ran.  We saw this set in interrupts_and_traps.c and
 * segments.c.
 */
85
static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages)
86
{
R
Rusty Russell 已提交
87 88
	/*
	 * Copying all this data can be quite expensive.  We usually run the
89 90
	 * same Guest we ran last time (and that Guest hasn't run anywhere else
	 * meanwhile).  If that's not the case, we pretend everything in the
R
Rusty Russell 已提交
91 92
	 * Guest has changed.
	 */
93
	if (__this_cpu_read(lg_last_cpu) != cpu || cpu->last_pages != pages) {
R
Rusty Russell 已提交
94
		__this_cpu_write(lg_last_cpu, cpu);
95
		cpu->last_pages = pages;
96
		cpu->changed = CHANGED_ALL;
97 98
	}

R
Rusty Russell 已提交
99 100 101 102
	/*
	 * These copies are pretty cheap, so we do them unconditionally: */
	/* Save the current Host top-level page directory.
	 */
103
	pages->state.host_cr3 = __pa(current->mm->pgd);
R
Rusty Russell 已提交
104 105 106 107
	/*
	 * Set up the Guest's page tables to see this CPU's pages (and no
	 * other CPU's pages).
	 */
108
	map_switcher_in_guest(cpu, pages);
R
Rusty Russell 已提交
109 110
	/*
	 * Set up the two "TSS" members which tell the CPU what stack to use
111
	 * for traps which do directly into the Guest (ie. traps at privilege
R
Rusty Russell 已提交
112 113
	 * level 1).
	 */
114
	pages->state.guest_tss.sp1 = cpu->esp1;
115
	pages->state.guest_tss.ss1 = cpu->ss1;
116 117

	/* Copy direct-to-Guest trap entries. */
118
	if (cpu->changed & CHANGED_IDT)
119
		copy_traps(cpu, pages->state.guest_idt, default_idt_entries);
120 121

	/* Copy all GDT entries which the Guest can change. */
122
	if (cpu->changed & CHANGED_GDT)
123
		copy_gdt(cpu, pages->state.guest_gdt);
124
	/* If only the TLS entries have changed, copy them. */
125
	else if (cpu->changed & CHANGED_GDT_TLS)
126
		copy_gdt_tls(cpu, pages->state.guest_gdt);
127 128

	/* Mark the Guest as unchanged for next time. */
129
	cpu->changed = 0;
130 131 132
}

/* Finally: the code to actually call into the Switcher to run the Guest. */
133
static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages)
134 135 136 137
{
	/* This is a dummy value we need for GCC's sake. */
	unsigned int clobber;

R
Rusty Russell 已提交
138 139 140 141
	/*
	 * Copy the guest-specific information into this CPU's "struct
	 * lguest_pages".
	 */
142
	copy_in_guest_info(cpu, pages);
143

R
Rusty Russell 已提交
144 145
	/*
	 * Set the trap number to 256 (impossible value).  If we fault while
146
	 * switching to the Guest (bad segment registers or bug), this will
R
Rusty Russell 已提交
147 148
	 * cause us to abort the Guest.
	 */
149
	cpu->regs->trapnum = 256;
150

R
Rusty Russell 已提交
151 152
	/*
	 * Now: we push the "eflags" register on the stack, then do an "lcall".
153 154 155 156 157 158
	 * This is how we change from using the kernel code segment to using
	 * the dedicated lguest code segment, as well as jumping into the
	 * Switcher.
	 *
	 * The lcall also pushes the old code segment (KERNEL_CS) onto the
	 * stack, then the address of this call.  This stack layout happens to
R
Rusty Russell 已提交
159 160
	 * exactly match the stack layout created by an interrupt...
	 */
161
	asm volatile("pushf; lcall *lguest_entry"
R
Rusty Russell 已提交
162 163 164 165
		     /*
		      * This is how we tell GCC that %eax ("a") and %ebx ("b")
		      * are changed by this routine.  The "=" means output.
		      */
166
		     : "=a"(clobber), "=b"(clobber)
R
Rusty Russell 已提交
167 168
		     /*
		      * %eax contains the pages pointer.  ("0" refers to the
169 170
		      * 0-th argument above, ie "a").  %ebx contains the
		      * physical address of the Guest's top-level page
R
Rusty Russell 已提交
171 172
		      * directory.
		      */
173
		     : "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir))
R
Rusty Russell 已提交
174 175
		     /*
		      * We tell gcc that all these registers could change,
176
		      * which means we don't have to save and restore them in
R
Rusty Russell 已提交
177 178
		      * the Switcher.
		      */
179 180 181 182
		     : "memory", "%edx", "%ecx", "%edi", "%esi");
}
/*:*/

R
Rusty Russell 已提交
183 184
/*M:002
 * There are hooks in the scheduler which we can register to tell when we
185 186 187 188 189
 * get kicked off the CPU (preempt_notifier_register()).  This would allow us
 * to lazily disable SYSENTER which would regain some performance, and should
 * also simplify copy_in_guest_info().  Note that we'd still need to restore
 * things when we exit to Launcher userspace, but that's fairly easy.
 *
R
Rusty Russell 已提交
190
 * We could also try using these hooks for PGE, but that might be too expensive.
191
 *
R
Rusty Russell 已提交
192 193
 * The hooks were designed for KVM, but we can also put them to good use.
:*/
194

R
Rusty Russell 已提交
195 196 197 198
/*H:040
 * This is the i386-specific code to setup and run the Guest.  Interrupts
 * are disabled: we own the CPU.
 */
199
void lguest_arch_run_guest(struct lg_cpu *cpu)
200
{
R
Rusty Russell 已提交
201 202
	/*
	 * Remember the awfully-named TS bit?  If the Guest has asked to set it
203
	 * we set it now, so we can trap and pass that trap to the Guest if it
R
Rusty Russell 已提交
204 205
	 * uses the FPU.
	 */
206
	if (cpu->ts)
207
		unlazy_fpu(current);
208

R
Rusty Russell 已提交
209 210
	/*
	 * SYSENTER is an optimized way of doing system calls.  We can't allow
211 212 213
	 * it because it always jumps to privilege level 0.  A normal Guest
	 * won't try it because we don't advertise it in CPUID, but a malicious
	 * Guest (or malicious Guest userspace program) could, so we tell the
R
Rusty Russell 已提交
214 215
	 * CPU to disable it before running the Guest.
	 */
216 217 218
	if (boot_cpu_has(X86_FEATURE_SEP))
		wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);

R
Rusty Russell 已提交
219 220
	/*
	 * Now we actually run the Guest.  It will return when something
221
	 * interesting happens, and we can examine its registers to see what it
R
Rusty Russell 已提交
222 223
	 * was doing.
	 */
224
	run_guest_once(cpu, lguest_pages(raw_smp_processor_id()));
225

R
Rusty Russell 已提交
226 227
	/*
	 * Note that the "regs" structure contains two extra entries which are
228 229
	 * not really registers: a trap number which says what interrupt or
	 * trap made the switcher code come back, and an error code which some
R
Rusty Russell 已提交
230 231
	 * traps set.
	 */
232

233 234 235 236
	 /* Restore SYSENTER if it's supposed to be on. */
	 if (boot_cpu_has(X86_FEATURE_SEP))
		wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);

R
Rusty Russell 已提交
237 238
	/*
	 * If the Guest page faulted, then the cr2 register will tell us the
239 240
	 * bad virtual address.  We have to grab this now, because once we
	 * re-enable interrupts an interrupt could fault and thus overwrite
R
Rusty Russell 已提交
241 242
	 * cr2, or we could even move off to a different CPU.
	 */
243
	if (cpu->regs->trapnum == 14)
244
		cpu->arch.last_pagefault = read_cr2();
R
Rusty Russell 已提交
245 246
	/*
	 * Similarly, if we took a trap because the Guest used the FPU,
247 248 249
	 * we have to restore the FPU it expects to see.
	 * math_state_restore() may sleep and we may even move off to
	 * a different CPU. So all the critical stuff should be done
R
Rusty Russell 已提交
250 251
	 * before this.
	 */
252
	else if (cpu->regs->trapnum == 7)
253 254 255
		math_state_restore();
}

R
Rusty Russell 已提交
256 257
/*H:130
 * Now we've examined the hypercall code; our Guest can make requests.
258 259 260 261 262 263
 * Our Guest is usually so well behaved; it never tries to do things it isn't
 * allowed to, and uses hypercalls instead.  Unfortunately, Linux's paravirtual
 * infrastructure isn't quite complete, because it doesn't contain replacements
 * for the Intel I/O instructions.  As a result, the Guest sometimes fumbles
 * across one during the boot process as it probes for various things which are
 * usually attached to a PC.
264
 *
265
 * When the Guest uses one of these instructions, we get a trap (General
266
 * Protection Fault) and come here.  We see if it's one of those troublesome
R
Rusty Russell 已提交
267 268
 * instructions and skip over it.  We return true if we did.
 */
269
static int emulate_insn(struct lg_cpu *cpu)
270 271
{
	u8 insn;
272
	unsigned int insnlen = 0, in = 0, small_operand = 0;
R
Rusty Russell 已提交
273 274
	/*
	 * The eip contains the *virtual* address of the Guest's instruction:
R
Rusty Russell 已提交
275
	 * walk the Guest's page tables to find the "physical" address.
R
Rusty Russell 已提交
276
	 */
277
	unsigned long physaddr = guest_pa(cpu, cpu->regs->eip);
278

R
Rusty Russell 已提交
279 280
	/*
	 * This must be the Guest kernel trying to do something, not userspace!
281
	 * The bottom two bits of the CS segment register are the privilege
R
Rusty Russell 已提交
282 283
	 * level.
	 */
284
	if ((cpu->regs->cs & 3) != GUEST_PL)
285 286 287
		return 0;

	/* Decoding x86 instructions is icky. */
288
	insn = lgread(cpu, physaddr, u8);
289

290 291 292 293 294 295 296 297 298 299 300 301
	/*
	 * Around 2.6.33, the kernel started using an emulation for the
	 * cmpxchg8b instruction in early boot on many configurations.  This
	 * code isn't paravirtualized, and it tries to disable interrupts.
	 * Ignore it, which will Mostly Work.
	 */
	if (insn == 0xfa) {
		/* "cli", or Clear Interrupt Enable instruction.  Skip it. */
		cpu->regs->eip++;
		return 1;
	}

R
Rusty Russell 已提交
302
	/*
303
	 * 0x66 is an "operand prefix".  It means a 16, not 32 bit in/out.
R
Rusty Russell 已提交
304
	 */
305
	if (insn == 0x66) {
306
		small_operand = 1;
307 308
		/* The instruction is 1 byte so far, read the next byte. */
		insnlen = 1;
309
		insn = lgread(cpu, physaddr + insnlen, u8);
310 311
	}

R
Rusty Russell 已提交
312 313 314 315
	/*
	 * We can ignore the lower bit for the moment and decode the 4 opcodes
	 * we need to emulate.
	 */
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335
	switch (insn & 0xFE) {
	case 0xE4: /* in     <next byte>,%al */
		insnlen += 2;
		in = 1;
		break;
	case 0xEC: /* in     (%dx),%al */
		insnlen += 1;
		in = 1;
		break;
	case 0xE6: /* out    %al,<next byte> */
		insnlen += 2;
		break;
	case 0xEE: /* out    %al,(%dx) */
		insnlen += 1;
		break;
	default:
		/* OK, we don't know what this is, can't emulate. */
		return 0;
	}

R
Rusty Russell 已提交
336 337
	/*
	 * If it was an "IN" instruction, they expect the result to be read
338
	 * into %eax, so we change %eax.  We always return all-ones, which
R
Rusty Russell 已提交
339 340
	 * traditionally means "there's nothing there".
	 */
341
	if (in) {
342 343 344 345 346 347 348 349
		/* Lower bit tells means it's a 32/16 bit access */
		if (insn & 0x1) {
			if (small_operand)
				cpu->regs->eax |= 0xFFFF;
			else
				cpu->regs->eax = 0xFFFFFFFF;
		} else
			cpu->regs->eax |= 0xFF;
350 351
	}
	/* Finally, we've "done" the instruction, so move past it. */
352
	cpu->regs->eip += insnlen;
353 354 355 356 357
	/* Success! */
	return 1;
}

/*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */
358
void lguest_arch_handle_trap(struct lg_cpu *cpu)
359
{
360
	switch (cpu->regs->trapnum) {
361
	case 13: /* We've intercepted a General Protection Fault. */
R
Rusty Russell 已提交
362 363
		/*
		 * Check if this was one of those annoying IN or OUT
364
		 * instructions which we need to emulate.  If so, we just go
R
Rusty Russell 已提交
365 366
		 * back into the Guest after we've done it.
		 */
367
		if (cpu->regs->errcode == 0) {
368
			if (emulate_insn(cpu))
369 370 371
				return;
		}
		break;
372
	case 14: /* We've intercepted a Page Fault. */
R
Rusty Russell 已提交
373 374
		/*
		 * The Guest accessed a virtual address that wasn't mapped.
375 376 377 378
		 * This happens a lot: we don't actually set up most of the page
		 * tables for the Guest at all when we start: as it runs it asks
		 * for more and more, and we set them up as required. In this
		 * case, we don't even tell the Guest that the fault happened.
379 380
		 *
		 * The errcode tells whether this was a read or a write, and
R
Rusty Russell 已提交
381 382
		 * whether kernel or userspace code.
		 */
383 384
		if (demand_page(cpu, cpu->arch.last_pagefault,
				cpu->regs->errcode))
385 386
			return;

R
Rusty Russell 已提交
387 388
		/*
		 * OK, it's really not there (or not OK): the Guest needs to
389 390 391 392 393
		 * know.  We write out the cr2 value so it knows where the
		 * fault occurred.
		 *
		 * Note that if the Guest were really messed up, this could
		 * happen before it's done the LHCALL_LGUEST_INIT hypercall, so
R
Rusty Russell 已提交
394 395
		 * lg->lguest_data could be NULL
		 */
396 397 398 399
		if (cpu->lg->lguest_data &&
		    put_user(cpu->arch.last_pagefault,
			     &cpu->lg->lguest_data->cr2))
			kill_guest(cpu, "Writing cr2");
400 401
		break;
	case 7: /* We've intercepted a Device Not Available fault. */
R
Rusty Russell 已提交
402 403 404 405
		/*
		 * If the Guest doesn't want to know, we already restored the
		 * Floating Point Unit, so we just continue without telling it.
		 */
406
		if (!cpu->ts)
407 408 409
			return;
		break;
	case 32 ... 255:
R
Rusty Russell 已提交
410 411
		/*
		 * These values mean a real interrupt occurred, in which case
412
		 * the Host handler has already been run. We just do a
413
		 * friendly check if another process should now be run, then
R
Rusty Russell 已提交
414
		 * return to run the Guest again.
R
Rusty Russell 已提交
415
		 */
416
		cond_resched();
417 418
		return;
	case LGUEST_TRAP_ENTRY:
R
Rusty Russell 已提交
419 420 421 422
		/*
		 * Our 'struct hcall_args' maps directly over our regs: we set
		 * up the pointer now to indicate a hypercall is pending.
		 */
423
		cpu->hcall = (struct hcall_args *)cpu->regs;
424 425 426 427
		return;
	}

	/* We didn't handle the trap, so it needs to go to the Guest. */
428
	if (!deliver_trap(cpu, cpu->regs->trapnum))
R
Rusty Russell 已提交
429 430
		/*
		 * If the Guest doesn't have a handler (either it hasn't
431
		 * registered any yet, or it's one of the faults we don't let
R
Rusty Russell 已提交
432 433
		 * it handle), it dies with this cryptic error message.
		 */
434
		kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)",
435
			   cpu->regs->trapnum, cpu->regs->eip,
436
			   cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault
437
			   : cpu->regs->errcode);
438 439
}

R
Rusty Russell 已提交
440 441
/*
 * Now we can look at each of the routines this calls, in increasing order of
442 443 444
 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
 * deliver_trap() and demand_page().  After all those, we'll be ready to
 * examine the Switcher, and our philosophical understanding of the Host/Guest
R
Rusty Russell 已提交
445 446
 * duality will be complete.
:*/
447 448 449 450 451 452 453 454
static void adjust_pge(void *on)
{
	if (on)
		write_cr4(read_cr4() | X86_CR4_PGE);
	else
		write_cr4(read_cr4() & ~X86_CR4_PGE);
}

R
Rusty Russell 已提交
455 456 457 458
/*H:020
 * Now the Switcher is mapped and every thing else is ready, we need to do
 * some more i386-specific initialization.
 */
459 460 461 462
void __init lguest_arch_host_init(void)
{
	int i;

R
Rusty Russell 已提交
463
	/*
R
Rusty Russell 已提交
464
	 * Most of the x86/switcher_32.S doesn't care that it's been moved; on
465 466 467 468 469 470
	 * Intel, jumps are relative, and it doesn't access any references to
	 * external code or data.
	 *
	 * The only exception is the interrupt handlers in switcher.S: their
	 * addresses are placed in a table (default_idt_entries), so we need to
	 * update the table with the new addresses.  switcher_offset() is a
471
	 * convenience function which returns the distance between the
R
Rusty Russell 已提交
472 473
	 * compiled-in switcher code and the high-mapped copy we just made.
	 */
474 475 476 477 478 479 480 481 482 483 484 485 486 487
	for (i = 0; i < IDT_ENTRIES; i++)
		default_idt_entries[i] += switcher_offset();

	/*
	 * Set up the Switcher's per-cpu areas.
	 *
	 * Each CPU gets two pages of its own within the high-mapped region
	 * (aka. "struct lguest_pages").  Much of this can be initialized now,
	 * but some depends on what Guest we are running (which is set up in
	 * copy_in_guest_info()).
	 */
	for_each_possible_cpu(i) {
		/* lguest_pages() returns this CPU's two pages. */
		struct lguest_pages *pages = lguest_pages(i);
R
Rusty Russell 已提交
488
		/* This is a convenience pointer to make the code neater. */
489 490
		struct lguest_ro_state *state = &pages->state;

R
Rusty Russell 已提交
491 492
		/*
		 * The Global Descriptor Table: the Host has a different one
493 494
		 * for each CPU.  We keep a descriptor for the GDT which says
		 * where it is and how big it is (the size is actually the last
R
Rusty Russell 已提交
495 496
		 * byte, not the size, hence the "-1").
		 */
497 498 499
		state->host_gdt_desc.size = GDT_SIZE-1;
		state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);

R
Rusty Russell 已提交
500 501
		/*
		 * All CPUs on the Host use the same Interrupt Descriptor
502
		 * Table, so we just use store_idt(), which gets this CPU's IDT
R
Rusty Russell 已提交
503 504
		 * descriptor.
		 */
505 506
		store_idt(&state->host_idt_desc);

R
Rusty Russell 已提交
507 508
		/*
		 * The descriptors for the Guest's GDT and IDT can be filled
509
		 * out now, too.  We copy the GDT & IDT into ->guest_gdt and
R
Rusty Russell 已提交
510 511
		 * ->guest_idt before actually running the Guest.
		 */
512 513 514 515 516
		state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
		state->guest_idt_desc.address = (long)&state->guest_idt;
		state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
		state->guest_gdt_desc.address = (long)&state->guest_gdt;

R
Rusty Russell 已提交
517 518
		/*
		 * We know where we want the stack to be when the Guest enters
519
		 * the Switcher: in pages->regs.  The stack grows upwards, so
R
Rusty Russell 已提交
520 521
		 * we start it at the end of that structure.
		 */
522
		state->guest_tss.sp0 = (long)(&pages->regs + 1);
R
Rusty Russell 已提交
523 524 525 526
		/*
		 * And this is the GDT entry to use for the stack: we keep a
		 * couple of special LGUEST entries.
		 */
527 528
		state->guest_tss.ss0 = LGUEST_DS;

R
Rusty Russell 已提交
529 530
		/*
		 * x86 can have a finegrained bitmap which indicates what I/O
531
		 * ports the process can use.  We set it to the end of our
R
Rusty Russell 已提交
532 533
		 * structure, meaning "none".
		 */
534 535
		state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);

R
Rusty Russell 已提交
536 537 538 539
		/*
		 * Some GDT entries are the same across all Guests, so we can
		 * set them up now.
		 */
540 541 542 543
		setup_default_gdt_entries(state);
		/* Most IDT entries are the same for all Guests, too.*/
		setup_default_idt_entries(state, default_idt_entries);

R
Rusty Russell 已提交
544 545 546 547
		/*
		 * The Host needs to be able to use the LGUEST segments on this
		 * CPU, too, so put them in the Host GDT.
		 */
548 549 550 551
		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
	}

R
Rusty Russell 已提交
552 553
	/*
	 * In the Switcher, we want the %cs segment register to use the
554 555
	 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
	 * it will be undisturbed when we switch.  To change %cs and jump we
R
Rusty Russell 已提交
556 557
	 * need this structure to feed to Intel's "lcall" instruction.
	 */
558 559 560
	lguest_entry.offset = (long)switch_to_guest + switcher_offset();
	lguest_entry.segment = LGUEST_CS;

R
Rusty Russell 已提交
561 562
	/*
	 * Finally, we need to turn off "Page Global Enable".  PGE is an
563 564 565 566 567 568 569 570 571
	 * optimization where page table entries are specially marked to show
	 * they never change.  The Host kernel marks all the kernel pages this
	 * way because it's always present, even when userspace is running.
	 *
	 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
	 * switch to the Guest kernel.  If you don't disable this on all CPUs,
	 * you'll get really weird bugs that you'll chase for two days.
	 *
	 * I used to turn PGE off every time we switched to the Guest and back
R
Rusty Russell 已提交
572 573
	 * on when we return, but that slowed the Switcher down noticibly.
	 */
574

R
Rusty Russell 已提交
575 576 577 578
	/*
	 * We don't need the complexity of CPUs coming and going while we're
	 * doing this.
	 */
579
	get_online_cpus();
580 581 582
	if (cpu_has_pge) { /* We have a broader idea of "global". */
		/* Remember that this was originally set (for cleanup). */
		cpu_had_pge = 1;
R
Rusty Russell 已提交
583 584 585 586
		/*
		 * adjust_pge is a helper function which sets or unsets the PGE
		 * bit on its CPU, depending on the argument (0 == unset).
		 */
587
		on_each_cpu(adjust_pge, (void *)0, 1);
588
		/* Turn off the feature in the global feature set. */
589
		clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
590
	}
591
	put_online_cpus();
R
Rusty Russell 已提交
592
}
593 594 595 596 597
/*:*/

void __exit lguest_arch_host_fini(void)
{
	/* If we had PGE before we started, turn it back on now. */
598
	get_online_cpus();
599
	if (cpu_had_pge) {
600
		set_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
601
		/* adjust_pge's argument "1" means set PGE. */
602
		on_each_cpu(adjust_pge, (void *)1, 1);
603
	}
604
	put_online_cpus();
605
}
606 607 608


/*H:122 The i386-specific hypercalls simply farm out to the right functions. */
609
int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
610 611
{
	switch (args->arg0) {
612 613
	case LHCALL_LOAD_GDT_ENTRY:
		load_guest_gdt_entry(cpu, args->arg1, args->arg2, args->arg3);
614 615
		break;
	case LHCALL_LOAD_IDT_ENTRY:
616
		load_guest_idt_entry(cpu, args->arg1, args->arg2, args->arg3);
617 618
		break;
	case LHCALL_LOAD_TLS:
619
		guest_load_tls(cpu, args->arg1);
620 621 622 623 624 625 626 627 628
		break;
	default:
		/* Bad Guest.  Bad! */
		return -EIO;
	}
	return 0;
}

/*H:126 i386-specific hypercall initialization: */
629
int lguest_arch_init_hypercalls(struct lg_cpu *cpu)
630 631 632
{
	u32 tsc_speed;

R
Rusty Russell 已提交
633 634 635 636
	/*
	 * The pointer to the Guest's "struct lguest_data" is the only argument.
	 * We check that address now.
	 */
637 638
	if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1,
			       sizeof(*cpu->lg->lguest_data)))
639 640
		return -EFAULT;

R
Rusty Russell 已提交
641 642
	/*
	 * Having checked it, we simply set lg->lguest_data to point straight
643 644 645
	 * into the Launcher's memory at the right place and then use
	 * copy_to_user/from_user from now on, instead of lgread/write.  I put
	 * this in to show that I'm not immune to writing stupid
R
Rusty Russell 已提交
646 647
	 * optimizations.
	 */
648
	cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1;
649

R
Rusty Russell 已提交
650 651
	/*
	 * We insist that the Time Stamp Counter exist and doesn't change with
652 653 654 655 656
	 * cpu frequency.  Some devious chip manufacturers decided that TSC
	 * changes could be handled in software.  I decided that time going
	 * backwards might be good for benchmarks, but it's bad for users.
	 *
	 * We also insist that the TSC be stable: the kernel detects unreliable
R
Rusty Russell 已提交
657 658
	 * TSCs for its own purposes, and we use that here.
	 */
659 660 661 662
	if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
		tsc_speed = tsc_khz;
	else
		tsc_speed = 0;
663
	if (put_user(tsc_speed, &cpu->lg->lguest_data->tsc_khz))
664 665
		return -EFAULT;

666
	/* The interrupt code might not like the system call vector. */
667 668
	if (!check_syscall_vector(cpu->lg))
		kill_guest(cpu, "bad syscall vector");
669

670 671
	return 0;
}
672
/*:*/
673

R
Rusty Russell 已提交
674
/*L:030
675
 * Most of the Guest's registers are left alone: we used get_zeroed_page() to
R
Rusty Russell 已提交
676 677
 * allocate the structure, so they will be 0.
 */
678
void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start)
679
{
680
	struct lguest_regs *regs = cpu->regs;
681

R
Rusty Russell 已提交
682 683
	/*
	 * There are four "segment" registers which the Guest needs to boot:
684 685 686 687 688
	 * The "code segment" register (cs) refers to the kernel code segment
	 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
	 * refer to the kernel data segment __KERNEL_DS.
	 *
	 * The privilege level is packed into the lower bits.  The Guest runs
R
Rusty Russell 已提交
689 690
	 * at privilege level 1 (GUEST_PL).
	 */
691 692 693
	regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
	regs->cs = __KERNEL_CS|GUEST_PL;

R
Rusty Russell 已提交
694 695
	/*
	 * The "eflags" register contains miscellaneous flags.  Bit 1 (0x002)
696 697
	 * is supposed to always be "1".  Bit 9 (0x200) controls whether
	 * interrupts are enabled.  We always leave interrupts enabled while
R
Rusty Russell 已提交
698 699
	 * running the Guest.
	 */
700
	regs->eflags = X86_EFLAGS_IF | 0x2;
701

R
Rusty Russell 已提交
702 703 704 705
	/*
	 * The "Extended Instruction Pointer" register says where the Guest is
	 * running.
	 */
706 707
	regs->eip = start;

R
Rusty Russell 已提交
708 709 710 711
	/*
	 * %esi points to our boot information, at physical address 0, so don't
	 * touch it.
	 */
712

R
Rusty Russell 已提交
713
	/* There are a couple of GDT entries the Guest expects at boot. */
714
	setup_guest_gdt(cpu);
715
}
新手
引导
客服 返回
顶部