drivers/lguest/x86/core.c

   1 /*
   2  * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
   3  * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> 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 This file contains the x86-specific lguest code.  It used to be all
  21  * mixed in with drivers/lguest/core.c but several foolhardy code slashers
  22  * wrestled most of the dependencies out to here in preparation for porting
  23  * lguest to other architectures (see what I mean by foolhardy?).
  24  *
  25  * This also contains a couple of non-obvious setup and teardown pieces which
  26  * were implemented after days of debugging pain. :*/
  27 #include <linux/kernel.h>
  28 #include <linux/start_kernel.h>
  29 #include <linux/string.h>
  30 #include <linux/console.h>
  31 #include <linux/screen_info.h>
  32 #include <linux/irq.h>
  33 #include <linux/interrupt.h>
  34 #include <linux/clocksource.h>
  35 #include <linux/clockchips.h>
  36 #include <linux/cpu.h>
  37 #include <linux/lguest.h>
  38 #include <linux/lguest_launcher.h>
  39 #include <asm/paravirt.h>
  40 #include <asm/param.h>
  41 #include <asm/page.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/desc.h>
  44 #include <asm/setup.h>
  45 #include <asm/lguest.h>
  46 #include <asm/uaccess.h>
  47 #include <asm/i387.h>
  48 #include "../lg.h"
  49
  50 static int cpu_had_pge;
  51
  52 static struct {
  53         unsigned long offset;
  54         unsigned short segment;
  55 } lguest_entry;
  56
  57 /* Offset from where switcher.S was compiled to where we've copied it */
  58 static unsigned long switcher_offset(void)
  59 {
  60         return SWITCHER_ADDR - (unsigned long)start_switcher_text;
  61 }
  62
  63 /* This cpu's struct lguest_pages. */
  64 static struct lguest_pages *lguest_pages(unsigned int cpu)
  65 {
  66         return &(((struct lguest_pages *)
  67                   (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
  68 }
  69
  70 static DEFINE_PER_CPU(struct lg_cpu *, last_cpu);
  71
  72 /*S:010
  73  * We approach the Switcher.
  74  *
  75  * Remember that each CPU has two pages which are visible to the Guest when it
  76  * runs on that CPU.  This has to contain the state for that Guest: we copy the
  77  * state in just before we run the Guest.
  78  *
  79  * Each Guest has "changed" flags which indicate what has changed in the Guest
  80  * since it last ran.  We saw this set in interrupts_and_traps.c and
  81  * segments.c.
  82  */
  83 static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages)
  84 {
  85         /* Copying all this data can be quite expensive.  We usually run the
  86          * same Guest we ran last time (and that Guest hasn't run anywhere else
  87          * meanwhile).  If that's not the case, we pretend everything in the
  88          * Guest has changed. */
  89         if (__get_cpu_var(last_cpu) != cpu || cpu->last_pages != pages) {
  90                 __get_cpu_var(last_cpu) = cpu;
  91                 cpu->last_pages = pages;
  92                 cpu->changed = CHANGED_ALL;
  93         }
  94
  95         /* These copies are pretty cheap, so we do them unconditionally: */
  96         /* Save the current Host top-level page directory. */
  97         pages->state.host_cr3 = __pa(current->mm->pgd);
  98         /* Set up the Guest's page tables to see this CPU's pages (and no
  99          * other CPU's pages). */
 100         map_switcher_in_guest(cpu, pages);
 101         /* Set up the two "TSS" members which tell the CPU what stack to use
 102          * for traps which do directly into the Guest (ie. traps at privilege
 103          * level 1). */
 104         pages->state.guest_tss.sp1 = cpu->esp1;
 105         pages->state.guest_tss.ss1 = cpu->ss1;
 106
 107         /* Copy direct-to-Guest trap entries. */
 108         if (cpu->changed & CHANGED_IDT)
 109                 copy_traps(cpu, pages->state.guest_idt, default_idt_entries);
 110
 111         /* Copy all GDT entries which the Guest can change. */
 112         if (cpu->changed & CHANGED_GDT)
 113                 copy_gdt(cpu, pages->state.guest_gdt);
 114         /* If only the TLS entries have changed, copy them. */
 115         else if (cpu->changed & CHANGED_GDT_TLS)
 116                 copy_gdt_tls(cpu, pages->state.guest_gdt);
 117
 118         /* Mark the Guest as unchanged for next time. */
 119         cpu->changed = 0;
 120 }
 121
 122 /* Finally: the code to actually call into the Switcher to run the Guest. */
 123 static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages)
 124 {
 125         /* This is a dummy value we need for GCC's sake. */
 126         unsigned int clobber;
 127
 128         /* Copy the guest-specific information into this CPU's "struct
 129          * lguest_pages". */
 130         copy_in_guest_info(cpu, pages);
 131
 132         /* Set the trap number to 256 (impossible value).  If we fault while
 133          * switching to the Guest (bad segment registers or bug), this will
 134          * cause us to abort the Guest. */
 135         cpu->regs->trapnum = 256;
 136
 137         /* Now: we push the "eflags" register on the stack, then do an "lcall".
 138          * This is how we change from using the kernel code segment to using
 139          * the dedicated lguest code segment, as well as jumping into the
 140          * Switcher.
 141          *
 142          * The lcall also pushes the old code segment (KERNEL_CS) onto the
 143          * stack, then the address of this call.  This stack layout happens to
 144          * exactly match the stack layout created by an interrupt... */
 145         asm volatile("pushf; lcall *lguest_entry"
 146                      /* This is how we tell GCC that %eax ("a") and %ebx ("b")
 147                       * are changed by this routine.  The "=" means output. */
 148                      : "=a"(clobber), "=b"(clobber)
 149                      /* %eax contains the pages pointer.  ("0" refers to the
 150                       * 0-th argument above, ie "a").  %ebx contains the
 151                       * physical address of the Guest's top-level page
 152                       * directory. */
 153                      : "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir))
 154                      /* We tell gcc that all these registers could change,
 155                       * which means we don't have to save and restore them in
 156                       * the Switcher. */
 157                      : "memory", "%edx", "%ecx", "%edi", "%esi");
 158 }
 159 /*:*/
 160
 161 /*M:002 There are hooks in the scheduler which we can register to tell when we
 162  * get kicked off the CPU (preempt_notifier_register()).  This would allow us
 163  * to lazily disable SYSENTER which would regain some performance, and should
 164  * also simplify copy_in_guest_info().  Note that we'd still need to restore
 165  * things when we exit to Launcher userspace, but that's fairly easy.
 166  *
 167  * We could also try using this hooks for PGE, but that might be too expensive.
 168  *
 169  * The hooks were designed for KVM, but we can also put them to good use. :*/
 170
 171 /*H:040 This is the i386-specific code to setup and run the Guest.  Interrupts
 172  * are disabled: we own the CPU. */
 173 void lguest_arch_run_guest(struct lg_cpu *cpu)
 174 {
 175         /* Remember the awfully-named TS bit?  If the Guest has asked to set it
 176          * we set it now, so we can trap and pass that trap to the Guest if it
 177          * uses the FPU. */
 178         if (cpu->ts)
 179                 lguest_set_ts();
 180
 181         /* SYSENTER is an optimized way of doing system calls.  We can't allow
 182          * it because it always jumps to privilege level 0.  A normal Guest
 183          * won't try it because we don't advertise it in CPUID, but a malicious
 184          * Guest (or malicious Guest userspace program) could, so we tell the
 185          * CPU to disable it before running the Guest. */
 186         if (boot_cpu_has(X86_FEATURE_SEP))
 187                 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
 188
 189         /* Now we actually run the Guest.  It will return when something
 190          * interesting happens, and we can examine its registers to see what it
 191          * was doing. */
 192         run_guest_once(cpu, lguest_pages(raw_smp_processor_id()));
 193
 194         /* Note that the "regs" structure contains two extra entries which are
 195          * not really registers: a trap number which says what interrupt or
 196          * trap made the switcher code come back, and an error code which some
 197          * traps set.  */
 198
 199         /* If the Guest page faulted, then the cr2 register will tell us the
 200          * bad virtual address.  We have to grab this now, because once we
 201          * re-enable interrupts an interrupt could fault and thus overwrite
 202          * cr2, or we could even move off to a different CPU. */
 203         if (cpu->regs->trapnum == 14)
 204                 cpu->arch.last_pagefault = read_cr2();
 205         /* Similarly, if we took a trap because the Guest used the FPU,
 206          * we have to restore the FPU it expects to see. */
 207         else if (cpu->regs->trapnum == 7)
 208                 math_state_restore();
 209
 210         /* Restore SYSENTER if it's supposed to be on. */
 211         if (boot_cpu_has(X86_FEATURE_SEP))
 212                 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
 213 }
 214
 215 /*H:130 Now we've examined the hypercall code; our Guest can make requests.
 216  * Our Guest is usually so well behaved; it never tries to do things it isn't
 217  * allowed to, and uses hypercalls instead.  Unfortunately, Linux's paravirtual
 218  * infrastructure isn't quite complete, because it doesn't contain replacements
 219  * for the Intel I/O instructions.  As a result, the Guest sometimes fumbles
 220  * across one during the boot process as it probes for various things which are
 221  * usually attached to a PC.
 222  *
 223  * When the Guest uses one of these instructions, we get a trap (General
 224  * Protection Fault) and come here.  We see if it's one of those troublesome
 225  * instructions and skip over it.  We return true if we did. */
 226 static int emulate_insn(struct lg_cpu *cpu)
 227 {
 228         u8 insn;
 229         unsigned int insnlen = 0, in = 0, shift = 0;
 230         /* The eip contains the *virtual* address of the Guest's instruction:
 231          * guest_pa just subtracts the Guest's page_offset. */
 232         unsigned long physaddr = guest_pa(cpu, cpu->regs->eip);
 233
 234         /* This must be the Guest kernel trying to do something, not userspace!
 235          * The bottom two bits of the CS segment register are the privilege
 236          * level. */
 237         if ((cpu->regs->cs & 3) != GUEST_PL)
 238                 return 0;
 239
 240         /* Decoding x86 instructions is icky. */
 241         insn = lgread(cpu, physaddr, u8);
 242
 243         /* 0x66 is an "operand prefix".  It means it's using the upper 16 bits
 244            of the eax register. */
 245         if (insn == 0x66) {
 246                 shift = 16;
 247                 /* The instruction is 1 byte so far, read the next byte. */
 248                 insnlen = 1;
 249                 insn = lgread(cpu, physaddr + insnlen, u8);
 250         }
 251
 252         /* We can ignore the lower bit for the moment and decode the 4 opcodes
 253          * we need to emulate. */
 254         switch (insn & 0xFE) {
 255         case 0xE4: /* in     <next byte>,%al */
 256                 insnlen += 2;
 257                 in = 1;
 258                 break;
 259         case 0xEC: /* in     (%dx),%al */
 260                 insnlen += 1;
 261                 in = 1;
 262                 break;
 263         case 0xE6: /* out    %al,<next byte> */
 264                 insnlen += 2;
 265                 break;
 266         case 0xEE: /* out    %al,(%dx) */
 267                 insnlen += 1;
 268                 break;
 269         default:
 270                 /* OK, we don't know what this is, can't emulate. */
 271                 return 0;
 272         }
 273
 274         /* If it was an "IN" instruction, they expect the result to be read
 275          * into %eax, so we change %eax.  We always return all-ones, which
 276          * traditionally means "there's nothing there". */
 277         if (in) {
 278                 /* Lower bit tells is whether it's a 16 or 32 bit access */
 279                 if (insn & 0x1)
 280                         cpu->regs->eax = 0xFFFFFFFF;
 281                 else
 282                         cpu->regs->eax |= (0xFFFF << shift);
 283         }
 284         /* Finally, we've "done" the instruction, so move past it. */
 285         cpu->regs->eip += insnlen;
 286         /* Success! */
 287         return 1;
 288 }
 289
 290 /*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */
 291 void lguest_arch_handle_trap(struct lg_cpu *cpu)
 292 {
 293         switch (cpu->regs->trapnum) {
 294         case 13: /* We've intercepted a General Protection Fault. */
 295                 /* Check if this was one of those annoying IN or OUT
 296                  * instructions which we need to emulate.  If so, we just go
 297                  * back into the Guest after we've done it. */
 298                 if (cpu->regs->errcode == 0) {
 299                         if (emulate_insn(cpu))
 300                                 return;
 301                 }
 302                 break;
 303         case 14: /* We've intercepted a Page Fault. */
 304                 /* The Guest accessed a virtual address that wasn't mapped.
 305                  * This happens a lot: we don't actually set up most of the page
 306                  * tables for the Guest at all when we start: as it runs it asks
 307                  * for more and more, and we set them up as required. In this
 308                  * case, we don't even tell the Guest that the fault happened.
 309                  *
 310                  * The errcode tells whether this was a read or a write, and
 311                  * whether kernel or userspace code. */
 312                 if (demand_page(cpu, cpu->arch.last_pagefault,
 313                                 cpu->regs->errcode))
 314                         return;
 315
 316                 /* OK, it's really not there (or not OK): the Guest needs to
 317                  * know.  We write out the cr2 value so it knows where the
 318                  * fault occurred.
 319                  *
 320                  * Note that if the Guest were really messed up, this could
 321                  * happen before it's done the LHCALL_LGUEST_INIT hypercall, so
 322                  * lg->lguest_data could be NULL */
 323                 if (cpu->lg->lguest_data &&
 324                     put_user(cpu->arch.last_pagefault,
 325                              &cpu->lg->lguest_data->cr2))
 326                         kill_guest(cpu, "Writing cr2");
 327                 break;
 328         case 7: /* We've intercepted a Device Not Available fault. */
 329                 /* If the Guest doesn't want to know, we already restored the
 330                  * Floating Point Unit, so we just continue without telling
 331                  * it. */
 332                 if (!cpu->ts)
 333                         return;
 334                 break;
 335         case 32 ... 255:
 336                 /* These values mean a real interrupt occurred, in which case
 337                  * the Host handler has already been run.  We just do a
 338                  * friendly check if another process should now be run, then
 339                  * return to run the Guest again */
 340                 cond_resched();
 341                 return;
 342         case LGUEST_TRAP_ENTRY:
 343                 /* Our 'struct hcall_args' maps directly over our regs: we set
 344                  * up the pointer now to indicate a hypercall is pending. */
 345                 cpu->hcall = (struct hcall_args *)cpu->regs;
 346                 return;
 347         }
 348
 349         /* We didn't handle the trap, so it needs to go to the Guest. */
 350         if (!deliver_trap(cpu, cpu->regs->trapnum))
 351                 /* If the Guest doesn't have a handler (either it hasn't
 352                  * registered any yet, or it's one of the faults we don't let
 353                  * it handle), it dies with this cryptic error message. */
 354                 kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)",
 355                            cpu->regs->trapnum, cpu->regs->eip,
 356                            cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault
 357                            : cpu->regs->errcode);
 358 }
 359
 360 /* Now we can look at each of the routines this calls, in increasing order of
 361  * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
 362  * deliver_trap() and demand_page().  After all those, we'll be ready to
 363  * examine the Switcher, and our philosophical understanding of the Host/Guest
 364  * duality will be complete. :*/
 365 static void adjust_pge(void *on)
 366 {
 367         if (on)
 368                 write_cr4(read_cr4() | X86_CR4_PGE);
 369         else
 370                 write_cr4(read_cr4() & ~X86_CR4_PGE);
 371 }
 372
 373 /*H:020 Now the Switcher is mapped and every thing else is ready, we need to do
 374  * some more i386-specific initialization. */
 375 void __init lguest_arch_host_init(void)
 376 {
 377         int i;
 378
 379         /* Most of the i386/switcher.S doesn't care that it's been moved; on
 380          * Intel, jumps are relative, and it doesn't access any references to
 381          * external code or data.
 382          *
 383          * The only exception is the interrupt handlers in switcher.S: their
 384          * addresses are placed in a table (default_idt_entries), so we need to
 385          * update the table with the new addresses.  switcher_offset() is a
 386          * convenience function which returns the distance between the
 387          * compiled-in switcher code and the high-mapped copy we just made. */
 388         for (i = 0; i < IDT_ENTRIES; i++)
 389                 default_idt_entries[i] += switcher_offset();
 390
 391         /*
 392          * Set up the Switcher's per-cpu areas.
 393          *
 394          * Each CPU gets two pages of its own within the high-mapped region
 395          * (aka. "struct lguest_pages").  Much of this can be initialized now,
 396          * but some depends on what Guest we are running (which is set up in
 397          * copy_in_guest_info()).
 398          */
 399         for_each_possible_cpu(i) {
 400                 /* lguest_pages() returns this CPU's two pages. */
 401                 struct lguest_pages *pages = lguest_pages(i);
 402                 /* This is a convenience pointer to make the code fit one
 403                  * statement to a line. */
 404                 struct lguest_ro_state *state = &pages->state;
 405
 406                 /* The Global Descriptor Table: the Host has a different one
 407                  * for each CPU.  We keep a descriptor for the GDT which says
 408                  * where it is and how big it is (the size is actually the last
 409                  * byte, not the size, hence the "-1"). */
 410                 state->host_gdt_desc.size = GDT_SIZE-1;
 411                 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
 412
 413                 /* All CPUs on the Host use the same Interrupt Descriptor
 414                  * Table, so we just use store_idt(), which gets this CPU's IDT
 415                  * descriptor. */
 416                 store_idt(&state->host_idt_desc);
 417
 418                 /* The descriptors for the Guest's GDT and IDT can be filled
 419                  * out now, too.  We copy the GDT & IDT into ->guest_gdt and
 420                  * ->guest_idt before actually running the Guest. */
 421                 state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
 422                 state->guest_idt_desc.address = (long)&state->guest_idt;
 423                 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
 424                 state->guest_gdt_desc.address = (long)&state->guest_gdt;
 425
 426                 /* We know where we want the stack to be when the Guest enters
 427                  * the Switcher: in pages->regs.  The stack grows upwards, so
 428                  * we start it at the end of that structure. */
 429                 state->guest_tss.sp0 = (long)(&pages->regs + 1);
 430                 /* And this is the GDT entry to use for the stack: we keep a
 431                  * couple of special LGUEST entries. */
 432                 state->guest_tss.ss0 = LGUEST_DS;
 433
 434                 /* x86 can have a finegrained bitmap which indicates what I/O
 435                  * ports the process can use.  We set it to the end of our
 436                  * structure, meaning "none". */
 437                 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
 438
 439                 /* Some GDT entries are the same across all Guests, so we can
 440                  * set them up now. */
 441                 setup_default_gdt_entries(state);
 442                 /* Most IDT entries are the same for all Guests, too.*/
 443                 setup_default_idt_entries(state, default_idt_entries);
 444
 445                 /* The Host needs to be able to use the LGUEST segments on this
 446                  * CPU, too, so put them in the Host GDT. */
 447                 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
 448                 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
 449         }
 450
 451         /* In the Switcher, we want the %cs segment register to use the
 452          * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
 453          * it will be undisturbed when we switch.  To change %cs and jump we
 454          * need this structure to feed to Intel's "lcall" instruction. */
 455         lguest_entry.offset = (long)switch_to_guest + switcher_offset();
 456         lguest_entry.segment = LGUEST_CS;
 457
 458         /* Finally, we need to turn off "Page Global Enable".  PGE is an
 459          * optimization where page table entries are specially marked to show
 460          * they never change.  The Host kernel marks all the kernel pages this
 461          * way because it's always present, even when userspace is running.
 462          *
 463          * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
 464          * switch to the Guest kernel.  If you don't disable this on all CPUs,
 465          * you'll get really weird bugs that you'll chase for two days.
 466          *
 467          * I used to turn PGE off every time we switched to the Guest and back
 468          * on when we return, but that slowed the Switcher down noticibly. */
 469
 470         /* We don't need the complexity of CPUs coming and going while we're
 471          * doing this. */
 472         get_online_cpus();
 473         if (cpu_has_pge) { /* We have a broader idea of "global". */
 474                 /* Remember that this was originally set (for cleanup). */
 475                 cpu_had_pge = 1;
 476                 /* adjust_pge is a helper function which sets or unsets the PGE
 477                  * bit on its CPU, depending on the argument (0 == unset). */
 478                 on_each_cpu(adjust_pge, (void *)0, 0, 1);
 479                 /* Turn off the feature in the global feature set. */
 480                 clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
 481         }
 482         put_online_cpus();
 483 };
 484 /*:*/
 485
 486 void __exit lguest_arch_host_fini(void)
 487 {
 488         /* If we had PGE before we started, turn it back on now. */
 489         get_online_cpus();
 490         if (cpu_had_pge) {
 491                 set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
 492                 /* adjust_pge's argument "1" means set PGE. */
 493                 on_each_cpu(adjust_pge, (void *)1, 0, 1);
 494         }
 495         put_online_cpus();
 496 }
 497
 498
 499 /*H:122 The i386-specific hypercalls simply farm out to the right functions. */
 500 int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
 501 {
 502         switch (args->arg0) {
 503         case LHCALL_LOAD_GDT:
 504                 load_guest_gdt(cpu, args->arg1, args->arg2);
 505                 break;
 506         case LHCALL_LOAD_IDT_ENTRY:
 507                 load_guest_idt_entry(cpu, args->arg1, args->arg2, args->arg3);
 508                 break;
 509         case LHCALL_LOAD_TLS:
 510                 guest_load_tls(cpu, args->arg1);
 511                 break;
 512         default:
 513                 /* Bad Guest.  Bad! */
 514                 return -EIO;
 515         }
 516         return 0;
 517 }
 518
 519 /*H:126 i386-specific hypercall initialization: */
 520 int lguest_arch_init_hypercalls(struct lg_cpu *cpu)
 521 {
 522         u32 tsc_speed;
 523
 524         /* The pointer to the Guest's "struct lguest_data" is the only argument.
 525          * We check that address now. */
 526         if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1,
 527                                sizeof(*cpu->lg->lguest_data)))
 528                 return -EFAULT;
 529
 530         /* Having checked it, we simply set lg->lguest_data to point straight
 531          * into the Launcher's memory at the right place and then use
 532          * copy_to_user/from_user from now on, instead of lgread/write.  I put
 533          * this in to show that I'm not immune to writing stupid
 534          * optimizations. */
 535         cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1;
 536
 537         /* We insist that the Time Stamp Counter exist and doesn't change with
 538          * cpu frequency.  Some devious chip manufacturers decided that TSC
 539          * changes could be handled in software.  I decided that time going
 540          * backwards might be good for benchmarks, but it's bad for users.
 541          *
 542          * We also insist that the TSC be stable: the kernel detects unreliable
 543          * TSCs for its own purposes, and we use that here. */
 544         if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
 545                 tsc_speed = tsc_khz;
 546         else
 547                 tsc_speed = 0;
 548         if (put_user(tsc_speed, &cpu->lg->lguest_data->tsc_khz))
 549                 return -EFAULT;
 550
 551         /* The interrupt code might not like the system call vector. */
 552         if (!check_syscall_vector(cpu->lg))
 553                 kill_guest(cpu, "bad syscall vector");
 554
 555         return 0;
 556 }
 557 /*:*/
 558
 559 /*L:030 lguest_arch_setup_regs()
 560  *
 561  * Most of the Guest's registers are left alone: we used get_zeroed_page() to
 562  * allocate the structure, so they will be 0. */
 563 void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start)
 564 {
 565         struct lguest_regs *regs = cpu->regs;
 566
 567         /* There are four "segment" registers which the Guest needs to boot:
 568          * The "code segment" register (cs) refers to the kernel code segment
 569          * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
 570          * refer to the kernel data segment __KERNEL_DS.
 571          *
 572          * The privilege level is packed into the lower bits.  The Guest runs
 573          * at privilege level 1 (GUEST_PL).*/
 574         regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
 575         regs->cs = __KERNEL_CS|GUEST_PL;
 576
 577         /* The "eflags" register contains miscellaneous flags.  Bit 1 (0x002)
 578          * is supposed to always be "1".  Bit 9 (0x200) controls whether
 579          * interrupts are enabled.  We always leave interrupts enabled while
 580          * running the Guest. */
 581         regs->eflags = X86_EFLAGS_IF | 0x2;
 582
 583         /* The "Extended Instruction Pointer" register says where the Guest is
 584          * running. */
 585         regs->eip = start;
 586
 587         /* %esi points to our boot information, at physical address 0, so don't
 588          * touch it. */
 589
 590         /* There are a couple of GDT entries the Guest expects when first
 591          * booting. */
 592         setup_guest_gdt(cpu);
 593 }