Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
core.c
Go to the documentation of this file.
1 /*
2  * Copyright (C) 2006, Rusty Russell <[email protected]> IBM Corporation.
3  * Copyright (C) 2007, Jes Sorensen <[email protected]> SGI.
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 2 of the License, or
8  * (at your option) any later version.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13  * NON INFRINGEMENT. See the GNU General Public License for more
14  * details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19  */
20 /*P:450
21  * This file contains the x86-specific lguest code. It used to be all
22  * mixed in with drivers/lguest/core.c but several foolhardy code slashers
23  * wrestled most of the dependencies out to here in preparation for porting
24  * lguest to other architectures (see what I mean by foolhardy?).
25  *
26  * This also contains a couple of non-obvious setup and teardown pieces which
27  * were implemented after days of debugging pain.
28 :*/
29 #include <linux/kernel.h>
30 #include <linux/start_kernel.h>
31 #include <linux/string.h>
32 #include <linux/console.h>
33 #include <linux/screen_info.h>
34 #include <linux/irq.h>
35 #include <linux/interrupt.h>
36 #include <linux/clocksource.h>
37 #include <linux/clockchips.h>
38 #include <linux/cpu.h>
39 #include <linux/lguest.h>
40 #include <linux/lguest_launcher.h>
41 #include <asm/paravirt.h>
42 #include <asm/param.h>
43 #include <asm/page.h>
44 #include <asm/pgtable.h>
45 #include <asm/desc.h>
46 #include <asm/setup.h>
47 #include <asm/lguest.h>
48 #include <asm/uaccess.h>
49 #include <asm/i387.h>
50 #include "../lg.h"
51 
52 static int cpu_had_pge;
53 
54 static struct {
55  unsigned long offset;
56  unsigned short segment;
57 } lguest_entry;
58 
59 /* Offset from where switcher.S was compiled to where we've copied it */
60 static unsigned long switcher_offset(void)
61 {
62  return SWITCHER_ADDR - (unsigned long)start_switcher_text;
63 }
64 
65 /* This cpu's struct lguest_pages. */
66 static struct lguest_pages *lguest_pages(unsigned int cpu)
67 {
68  return &(((struct lguest_pages *)
70 }
71 
72 static DEFINE_PER_CPU(struct lg_cpu *, lg_last_cpu);
73 
74 /*S:010
75  * We approach the Switcher.
76  *
77  * Remember that each CPU has two pages which are visible to the Guest when it
78  * runs on that CPU. This has to contain the state for that Guest: we copy the
79  * state in just before we run the Guest.
80  *
81  * Each Guest has "changed" flags which indicate what has changed in the Guest
82  * since it last ran. We saw this set in interrupts_and_traps.c and
83  * segments.c.
84  */
85 static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages)
86 {
87  /*
88  * Copying all this data can be quite expensive. We usually run the
89  * same Guest we ran last time (and that Guest hasn't run anywhere else
90  * meanwhile). If that's not the case, we pretend everything in the
91  * Guest has changed.
92  */
93  if (__this_cpu_read(lg_last_cpu) != cpu || cpu->last_pages != pages) {
94  __this_cpu_write(lg_last_cpu, cpu);
95  cpu->last_pages = pages;
96  cpu->changed = CHANGED_ALL;
97  }
98 
99  /*
100  * These copies are pretty cheap, so we do them unconditionally: */
101  /* Save the current Host top-level page directory.
102  */
103  pages->state.host_cr3 = __pa(current->mm->pgd);
104  /*
105  * Set up the Guest's page tables to see this CPU's pages (and no
106  * other CPU's pages).
107  */
108  map_switcher_in_guest(cpu, pages);
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
112  * level 1).
113  */
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  /* This is a dummy value we need for GCC's sake. */
136  unsigned int clobber;
137 
138  /*
139  * Copy the guest-specific information into this CPU's "struct
140  * lguest_pages".
141  */
142  copy_in_guest_info(cpu, pages);
143 
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
147  * cause us to abort the Guest.
148  */
149  cpu->regs->trapnum = 256;
150 
151  /*
152  * Now: we push the "eflags" register on the stack, then do an "lcall".
153  * This is how we change from using the kernel code segment to using
154  * the dedicated lguest code segment, as well as jumping into the
155  * Switcher.
156  *
157  * The lcall also pushes the old code segment (KERNEL_CS) onto the
158  * stack, then the address of this call. This stack layout happens to
159  * exactly match the stack layout created by an interrupt...
160  */
161  asm volatile("pushf; lcall *lguest_entry"
162  /*
163  * This is how we tell GCC that %eax ("a") and %ebx ("b")
164  * are changed by this routine. The "=" means output.
165  */
166  : "=a"(clobber), "=b"(clobber)
167  /*
168  * %eax contains the pages pointer. ("0" refers to the
169  * 0-th argument above, ie "a"). %ebx contains the
170  * physical address of the Guest's top-level page
171  * directory.
172  */
173  : "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir))
174  /*
175  * We tell gcc that all these registers could change,
176  * which means we don't have to save and restore them in
177  * the Switcher.
178  */
179  : "memory", "%edx", "%ecx", "%edi", "%esi");
180 }
181 /*:*/
182 
183 /*M:002
184  * There are hooks in the scheduler which we can register to tell when we
185  * get kicked off the CPU (preempt_notifier_register()). This would allow us
186  * to lazily disable SYSENTER which would regain some performance, and should
187  * also simplify copy_in_guest_info(). Note that we'd still need to restore
188  * things when we exit to Launcher userspace, but that's fairly easy.
189  *
190  * We could also try using these hooks for PGE, but that might be too expensive.
191  *
192  * The hooks were designed for KVM, but we can also put them to good use.
193 :*/
194 
195 /*H:040
196  * This is the i386-specific code to setup and run the Guest. Interrupts
197  * are disabled: we own the CPU.
198  */
199 void lguest_arch_run_guest(struct lg_cpu *cpu)
200 {
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
204  * uses the FPU.
205  */
206  if (cpu->ts && user_has_fpu())
207  stts();
208 
209  /*
210  * SYSENTER is an optimized way of doing system calls. We can't allow
211  * it because it always jumps to privilege level 0. A normal Guest
212  * won't try it because we don't advertise it in CPUID, but a malicious
213  * Guest (or malicious Guest userspace program) could, so we tell the
214  * CPU to disable it before running the Guest.
215  */
216  if (boot_cpu_has(X86_FEATURE_SEP))
218 
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
222  * was doing.
223  */
224  run_guest_once(cpu, lguest_pages(raw_smp_processor_id()));
225 
226  /*
227  * Note that the "regs" structure contains two extra entries which are
228  * not really registers: a trap number which says what interrupt or
229  * trap made the switcher code come back, and an error code which some
230  * traps set.
231  */
232 
233  /* Restore SYSENTER if it's supposed to be on. */
234  if (boot_cpu_has(X86_FEATURE_SEP))
236 
237  /* Clear the host TS bit if it was set above. */
238  if (cpu->ts && user_has_fpu())
239  clts();
240 
241  /*
242  * If the Guest page faulted, then the cr2 register will tell us the
243  * bad virtual address. We have to grab this now, because once we
244  * re-enable interrupts an interrupt could fault and thus overwrite
245  * cr2, or we could even move off to a different CPU.
246  */
247  if (cpu->regs->trapnum == 14)
248  cpu->arch.last_pagefault = read_cr2();
249  /*
250  * Similarly, if we took a trap because the Guest used the FPU,
251  * we have to restore the FPU it expects to see.
252  * math_state_restore() may sleep and we may even move off to
253  * a different CPU. So all the critical stuff should be done
254  * before this.
255  */
256  else if (cpu->regs->trapnum == 7 && !user_has_fpu())
258 }
259 
260 /*H:130
261  * Now we've examined the hypercall code; our Guest can make requests.
262  * Our Guest is usually so well behaved; it never tries to do things it isn't
263  * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual
264  * infrastructure isn't quite complete, because it doesn't contain replacements
265  * for the Intel I/O instructions. As a result, the Guest sometimes fumbles
266  * across one during the boot process as it probes for various things which are
267  * usually attached to a PC.
268  *
269  * When the Guest uses one of these instructions, we get a trap (General
270  * Protection Fault) and come here. We see if it's one of those troublesome
271  * instructions and skip over it. We return true if we did.
272  */
273 static int emulate_insn(struct lg_cpu *cpu)
274 {
275  u8 insn;
276  unsigned int insnlen = 0, in = 0, small_operand = 0;
277  /*
278  * The eip contains the *virtual* address of the Guest's instruction:
279  * walk the Guest's page tables to find the "physical" address.
280  */
281  unsigned long physaddr = guest_pa(cpu, cpu->regs->eip);
282 
283  /*
284  * This must be the Guest kernel trying to do something, not userspace!
285  * The bottom two bits of the CS segment register are the privilege
286  * level.
287  */
288  if ((cpu->regs->cs & 3) != GUEST_PL)
289  return 0;
290 
291  /* Decoding x86 instructions is icky. */
292  insn = lgread(cpu, physaddr, u8);
293 
294  /*
295  * Around 2.6.33, the kernel started using an emulation for the
296  * cmpxchg8b instruction in early boot on many configurations. This
297  * code isn't paravirtualized, and it tries to disable interrupts.
298  * Ignore it, which will Mostly Work.
299  */
300  if (insn == 0xfa) {
301  /* "cli", or Clear Interrupt Enable instruction. Skip it. */
302  cpu->regs->eip++;
303  return 1;
304  }
305 
306  /*
307  * 0x66 is an "operand prefix". It means a 16, not 32 bit in/out.
308  */
309  if (insn == 0x66) {
310  small_operand = 1;
311  /* The instruction is 1 byte so far, read the next byte. */
312  insnlen = 1;
313  insn = lgread(cpu, physaddr + insnlen, u8);
314  }
315 
316  /*
317  * We can ignore the lower bit for the moment and decode the 4 opcodes
318  * we need to emulate.
319  */
320  switch (insn & 0xFE) {
321  case 0xE4: /* in <next byte>,%al */
322  insnlen += 2;
323  in = 1;
324  break;
325  case 0xEC: /* in (%dx),%al */
326  insnlen += 1;
327  in = 1;
328  break;
329  case 0xE6: /* out %al,<next byte> */
330  insnlen += 2;
331  break;
332  case 0xEE: /* out %al,(%dx) */
333  insnlen += 1;
334  break;
335  default:
336  /* OK, we don't know what this is, can't emulate. */
337  return 0;
338  }
339 
340  /*
341  * If it was an "IN" instruction, they expect the result to be read
342  * into %eax, so we change %eax. We always return all-ones, which
343  * traditionally means "there's nothing there".
344  */
345  if (in) {
346  /* Lower bit tells means it's a 32/16 bit access */
347  if (insn & 0x1) {
348  if (small_operand)
349  cpu->regs->eax |= 0xFFFF;
350  else
351  cpu->regs->eax = 0xFFFFFFFF;
352  } else
353  cpu->regs->eax |= 0xFF;
354  }
355  /* Finally, we've "done" the instruction, so move past it. */
356  cpu->regs->eip += insnlen;
357  /* Success! */
358  return 1;
359 }
360 
361 /*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */
363 {
364  switch (cpu->regs->trapnum) {
365  case 13: /* We've intercepted a General Protection Fault. */
366  /*
367  * Check if this was one of those annoying IN or OUT
368  * instructions which we need to emulate. If so, we just go
369  * back into the Guest after we've done it.
370  */
371  if (cpu->regs->errcode == 0) {
372  if (emulate_insn(cpu))
373  return;
374  }
375  break;
376  case 14: /* We've intercepted a Page Fault. */
377  /*
378  * The Guest accessed a virtual address that wasn't mapped.
379  * This happens a lot: we don't actually set up most of the page
380  * tables for the Guest at all when we start: as it runs it asks
381  * for more and more, and we set them up as required. In this
382  * case, we don't even tell the Guest that the fault happened.
383  *
384  * The errcode tells whether this was a read or a write, and
385  * whether kernel or userspace code.
386  */
387  if (demand_page(cpu, cpu->arch.last_pagefault,
388  cpu->regs->errcode))
389  return;
390 
391  /*
392  * OK, it's really not there (or not OK): the Guest needs to
393  * know. We write out the cr2 value so it knows where the
394  * fault occurred.
395  *
396  * Note that if the Guest were really messed up, this could
397  * happen before it's done the LHCALL_LGUEST_INIT hypercall, so
398  * lg->lguest_data could be NULL
399  */
400  if (cpu->lg->lguest_data &&
401  put_user(cpu->arch.last_pagefault,
402  &cpu->lg->lguest_data->cr2))
403  kill_guest(cpu, "Writing cr2");
404  break;
405  case 7: /* We've intercepted a Device Not Available fault. */
406  /*
407  * If the Guest doesn't want to know, we already restored the
408  * Floating Point Unit, so we just continue without telling it.
409  */
410  if (!cpu->ts)
411  return;
412  break;
413  case 32 ... 255:
414  /*
415  * These values mean a real interrupt occurred, in which case
416  * the Host handler has already been run. We just do a
417  * friendly check if another process should now be run, then
418  * return to run the Guest again.
419  */
420  cond_resched();
421  return;
422  case LGUEST_TRAP_ENTRY:
423  /*
424  * Our 'struct hcall_args' maps directly over our regs: we set
425  * up the pointer now to indicate a hypercall is pending.
426  */
427  cpu->hcall = (struct hcall_args *)cpu->regs;
428  return;
429  }
430 
431  /* We didn't handle the trap, so it needs to go to the Guest. */
432  if (!deliver_trap(cpu, cpu->regs->trapnum))
433  /*
434  * If the Guest doesn't have a handler (either it hasn't
435  * registered any yet, or it's one of the faults we don't let
436  * it handle), it dies with this cryptic error message.
437  */
438  kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)",
439  cpu->regs->trapnum, cpu->regs->eip,
440  cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault
441  : cpu->regs->errcode);
442 }
443 
444 /*
445  * Now we can look at each of the routines this calls, in increasing order of
446  * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
447  * deliver_trap() and demand_page(). After all those, we'll be ready to
448  * examine the Switcher, and our philosophical understanding of the Host/Guest
449  * duality will be complete.
450 :*/
451 static void adjust_pge(void *on)
452 {
453  if (on)
454  write_cr4(read_cr4() | X86_CR4_PGE);
455  else
456  write_cr4(read_cr4() & ~X86_CR4_PGE);
457 }
458 
459 /*H:020
460  * Now the Switcher is mapped and every thing else is ready, we need to do
461  * some more i386-specific initialization.
462  */
464 {
465  int i;
466 
467  /*
468  * Most of the x86/switcher_32.S doesn't care that it's been moved; on
469  * Intel, jumps are relative, and it doesn't access any references to
470  * external code or data.
471  *
472  * The only exception is the interrupt handlers in switcher.S: their
473  * addresses are placed in a table (default_idt_entries), so we need to
474  * update the table with the new addresses. switcher_offset() is a
475  * convenience function which returns the distance between the
476  * compiled-in switcher code and the high-mapped copy we just made.
477  */
478  for (i = 0; i < IDT_ENTRIES; i++)
479  default_idt_entries[i] += switcher_offset();
480 
481  /*
482  * Set up the Switcher's per-cpu areas.
483  *
484  * Each CPU gets two pages of its own within the high-mapped region
485  * (aka. "struct lguest_pages"). Much of this can be initialized now,
486  * but some depends on what Guest we are running (which is set up in
487  * copy_in_guest_info()).
488  */
490  /* lguest_pages() returns this CPU's two pages. */
491  struct lguest_pages *pages = lguest_pages(i);
492  /* This is a convenience pointer to make the code neater. */
493  struct lguest_ro_state *state = &pages->state;
494 
495  /*
496  * The Global Descriptor Table: the Host has a different one
497  * for each CPU. We keep a descriptor for the GDT which says
498  * where it is and how big it is (the size is actually the last
499  * byte, not the size, hence the "-1").
500  */
501  state->host_gdt_desc.size = GDT_SIZE-1;
502  state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
503 
504  /*
505  * All CPUs on the Host use the same Interrupt Descriptor
506  * Table, so we just use store_idt(), which gets this CPU's IDT
507  * descriptor.
508  */
509  store_idt(&state->host_idt_desc);
510 
511  /*
512  * The descriptors for the Guest's GDT and IDT can be filled
513  * out now, too. We copy the GDT & IDT into ->guest_gdt and
514  * ->guest_idt before actually running the Guest.
515  */
516  state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
517  state->guest_idt_desc.address = (long)&state->guest_idt;
518  state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
519  state->guest_gdt_desc.address = (long)&state->guest_gdt;
520 
521  /*
522  * We know where we want the stack to be when the Guest enters
523  * the Switcher: in pages->regs. The stack grows upwards, so
524  * we start it at the end of that structure.
525  */
526  state->guest_tss.sp0 = (long)(&pages->regs + 1);
527  /*
528  * And this is the GDT entry to use for the stack: we keep a
529  * couple of special LGUEST entries.
530  */
531  state->guest_tss.ss0 = LGUEST_DS;
532 
533  /*
534  * x86 can have a finegrained bitmap which indicates what I/O
535  * ports the process can use. We set it to the end of our
536  * structure, meaning "none".
537  */
538  state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
539 
540  /*
541  * Some GDT entries are the same across all Guests, so we can
542  * set them up now.
543  */
545  /* Most IDT entries are the same for all Guests, too.*/
547 
548  /*
549  * The Host needs to be able to use the LGUEST segments on this
550  * CPU, too, so put them in the Host GDT.
551  */
552  get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
553  get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
554  }
555 
556  /*
557  * In the Switcher, we want the %cs segment register to use the
558  * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
559  * it will be undisturbed when we switch. To change %cs and jump we
560  * need this structure to feed to Intel's "lcall" instruction.
561  */
562  lguest_entry.offset = (long)switch_to_guest + switcher_offset();
563  lguest_entry.segment = LGUEST_CS;
564 
565  /*
566  * Finally, we need to turn off "Page Global Enable". PGE is an
567  * optimization where page table entries are specially marked to show
568  * they never change. The Host kernel marks all the kernel pages this
569  * way because it's always present, even when userspace is running.
570  *
571  * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
572  * switch to the Guest kernel. If you don't disable this on all CPUs,
573  * you'll get really weird bugs that you'll chase for two days.
574  *
575  * I used to turn PGE off every time we switched to the Guest and back
576  * on when we return, but that slowed the Switcher down noticibly.
577  */
578 
579  /*
580  * We don't need the complexity of CPUs coming and going while we're
581  * doing this.
582  */
583  get_online_cpus();
584  if (cpu_has_pge) { /* We have a broader idea of "global". */
585  /* Remember that this was originally set (for cleanup). */
586  cpu_had_pge = 1;
587  /*
588  * adjust_pge is a helper function which sets or unsets the PGE
589  * bit on its CPU, depending on the argument (0 == unset).
590  */
591  on_each_cpu(adjust_pge, (void *)0, 1);
592  /* Turn off the feature in the global feature set. */
593  clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
594  }
595  put_online_cpus();
596 }
597 /*:*/
598 
600 {
601  /* If we had PGE before we started, turn it back on now. */
602  get_online_cpus();
603  if (cpu_had_pge) {
604  set_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
605  /* adjust_pge's argument "1" means set PGE. */
606  on_each_cpu(adjust_pge, (void *)1, 1);
607  }
608  put_online_cpus();
609 }
610 
611 
612 /*H:122 The i386-specific hypercalls simply farm out to the right functions. */
613 int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
614 {
615  switch (args->arg0) {
617  load_guest_gdt_entry(cpu, args->arg1, args->arg2, args->arg3);
618  break;
620  load_guest_idt_entry(cpu, args->arg1, args->arg2, args->arg3);
621  break;
622  case LHCALL_LOAD_TLS:
623  guest_load_tls(cpu, args->arg1);
624  break;
625  default:
626  /* Bad Guest. Bad! */
627  return -EIO;
628  }
629  return 0;
630 }
631 
632 /*H:126 i386-specific hypercall initialization: */
634 {
635  u32 tsc_speed;
636 
637  /*
638  * The pointer to the Guest's "struct lguest_data" is the only argument.
639  * We check that address now.
640  */
641  if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1,
642  sizeof(*cpu->lg->lguest_data)))
643  return -EFAULT;
644 
645  /*
646  * Having checked it, we simply set lg->lguest_data to point straight
647  * into the Launcher's memory at the right place and then use
648  * copy_to_user/from_user from now on, instead of lgread/write. I put
649  * this in to show that I'm not immune to writing stupid
650  * optimizations.
651  */
652  cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1;
653 
654  /*
655  * We insist that the Time Stamp Counter exist and doesn't change with
656  * cpu frequency. Some devious chip manufacturers decided that TSC
657  * changes could be handled in software. I decided that time going
658  * backwards might be good for benchmarks, but it's bad for users.
659  *
660  * We also insist that the TSC be stable: the kernel detects unreliable
661  * TSCs for its own purposes, and we use that here.
662  */
663  if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
664  tsc_speed = tsc_khz;
665  else
666  tsc_speed = 0;
667  if (put_user(tsc_speed, &cpu->lg->lguest_data->tsc_khz))
668  return -EFAULT;
669 
670  /* The interrupt code might not like the system call vector. */
671  if (!check_syscall_vector(cpu->lg))
672  kill_guest(cpu, "bad syscall vector");
673 
674  return 0;
675 }
676 /*:*/
677 
678 /*L:030
679  * Most of the Guest's registers are left alone: we used get_zeroed_page() to
680  * allocate the structure, so they will be 0.
681  */
682 void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start)
683 {
684  struct lguest_regs *regs = cpu->regs;
685 
686  /*
687  * There are four "segment" registers which the Guest needs to boot:
688  * The "code segment" register (cs) refers to the kernel code segment
689  * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
690  * refer to the kernel data segment __KERNEL_DS.
691  *
692  * The privilege level is packed into the lower bits. The Guest runs
693  * at privilege level 1 (GUEST_PL).
694  */
695  regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
696  regs->cs = __KERNEL_CS|GUEST_PL;
697 
698  /*
699  * The "eflags" register contains miscellaneous flags. Bit 1 (0x002)
700  * is supposed to always be "1". Bit 9 (0x200) controls whether
701  * interrupts are enabled. We always leave interrupts enabled while
702  * running the Guest.
703  */
705 
706  /*
707  * The "Extended Instruction Pointer" register says where the Guest is
708  * running.
709  */
710  regs->eip = start;
711 
712  /*
713  * %esi points to our boot information, at physical address 0, so don't
714  * touch it.
715  */
716 
717  /* There are a couple of GDT entries the Guest expects at boot. */
718  setup_guest_gdt(cpu);
719 }