1 /*P:400 This contains run_guest() which actually calls into the Host<->Guest
2 * Switcher and analyzes the return, such as determining if the Guest wants the
3 * Host to do something. This file also contains useful helper routines, and a
4 * couple of non-obvious setup and teardown pieces which were implemented after
5 * days of debugging pain. :*/
6 #include <linux/module.h>
7 #include <linux/stringify.h>
8 #include <linux/stddef.h>
11 #include <linux/vmalloc.h>
12 #include <linux/cpu.h>
13 #include <linux/freezer.h>
14 #include <asm/paravirt.h>
16 #include <asm/pgtable.h>
17 #include <asm/uaccess.h>
19 #include <asm/highmem.h>
20 #include <asm/asm-offsets.h>
24 /* Found in switcher.S */
25 extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
26 extern unsigned long default_idt_entries[];
28 /* Every guest maps the core switcher code. */
29 #define SHARED_SWITCHER_PAGES \
30 DIV_ROUND_UP(end_switcher_text - start_switcher_text, PAGE_SIZE)
31 /* Pages for switcher itself, then two pages per cpu */
32 #define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * NR_CPUS)
34 /* We map at -4M for ease of mapping into the guest (one PTE page). */
35 #define SWITCHER_ADDR 0xFFC00000
37 static struct vm_struct *switcher_vma;
38 static struct page **switcher_page;
40 static int cpu_had_pge;
43 unsigned short segment;
46 /* This One Big lock protects all inter-guest data structures. */
47 DEFINE_MUTEX(lguest_lock);
48 static DEFINE_PER_CPU(struct lguest *, last_guest);
50 /* Offset from where switcher.S was compiled to where we've copied it */
51 static unsigned long switcher_offset(void)
53 return SWITCHER_ADDR - (unsigned long)start_switcher_text;
56 /* This cpu's struct lguest_pages. */
57 static struct lguest_pages *lguest_pages(unsigned int cpu)
59 return &(((struct lguest_pages *)
60 (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
63 /*H:010 We need to set up the Switcher at a high virtual address. Remember the
64 * Switcher is a few hundred bytes of assembler code which actually changes the
65 * CPU to run the Guest, and then changes back to the Host when a trap or
68 * The Switcher code must be at the same virtual address in the Guest as the
69 * Host since it will be running as the switchover occurs.
71 * Trying to map memory at a particular address is an unusual thing to do, so
72 * it's not a simple one-liner. We also set up the per-cpu parts of the
75 static __init int map_switcher(void)
81 * Map the Switcher in to high memory.
83 * It turns out that if we choose the address 0xFFC00000 (4MB under the
84 * top virtual address), it makes setting up the page tables really
88 /* We allocate an array of "struct page"s. map_vm_area() wants the
89 * pages in this form, rather than just an array of pointers. */
90 switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
97 /* Now we actually allocate the pages. The Guest will see these pages,
98 * so we make sure they're zeroed. */
99 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
100 unsigned long addr = get_zeroed_page(GFP_KERNEL);
103 goto free_some_pages;
105 switcher_page[i] = virt_to_page(addr);
108 /* Now we reserve the "virtual memory area" we want: 0xFFC00000
109 * (SWITCHER_ADDR). We might not get it in theory, but in practice
110 * it's worked so far. */
111 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
112 VM_ALLOC, SWITCHER_ADDR, VMALLOC_END);
115 printk("lguest: could not map switcher pages high\n");
119 /* This code actually sets up the pages we've allocated to appear at
120 * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
121 * kind of pages we're mapping (kernel pages), and a pointer to our
122 * array of struct pages. It increments that pointer, but we don't
124 pagep = switcher_page;
125 err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
127 printk("lguest: map_vm_area failed: %i\n", err);
131 /* Now the switcher is mapped at the right address, we can't fail!
132 * Copy in the compiled-in Switcher code (from switcher.S). */
133 memcpy(switcher_vma->addr, start_switcher_text,
134 end_switcher_text - start_switcher_text);
136 /* Most of the switcher.S doesn't care that it's been moved; on Intel,
137 * jumps are relative, and it doesn't access any references to external
140 * The only exception is the interrupt handlers in switcher.S: their
141 * addresses are placed in a table (default_idt_entries), so we need to
142 * update the table with the new addresses. switcher_offset() is a
143 * convenience function which returns the distance between the builtin
144 * switcher code and the high-mapped copy we just made. */
145 for (i = 0; i < IDT_ENTRIES; i++)
146 default_idt_entries[i] += switcher_offset();
149 * Set up the Switcher's per-cpu areas.
151 * Each CPU gets two pages of its own within the high-mapped region
152 * (aka. "struct lguest_pages"). Much of this can be initialized now,
153 * but some depends on what Guest we are running (which is set up in
154 * copy_in_guest_info()).
156 for_each_possible_cpu(i) {
157 /* lguest_pages() returns this CPU's two pages. */
158 struct lguest_pages *pages = lguest_pages(i);
159 /* This is a convenience pointer to make the code fit one
160 * statement to a line. */
161 struct lguest_ro_state *state = &pages->state;
163 /* The Global Descriptor Table: the Host has a different one
164 * for each CPU. We keep a descriptor for the GDT which says
165 * where it is and how big it is (the size is actually the last
166 * byte, not the size, hence the "-1"). */
167 state->host_gdt_desc.size = GDT_SIZE-1;
168 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
170 /* All CPUs on the Host use the same Interrupt Descriptor
171 * Table, so we just use store_idt(), which gets this CPU's IDT
173 store_idt(&state->host_idt_desc);
175 /* The descriptors for the Guest's GDT and IDT can be filled
176 * out now, too. We copy the GDT & IDT into ->guest_gdt and
177 * ->guest_idt before actually running the Guest. */
178 state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
179 state->guest_idt_desc.address = (long)&state->guest_idt;
180 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
181 state->guest_gdt_desc.address = (long)&state->guest_gdt;
183 /* We know where we want the stack to be when the Guest enters
184 * the switcher: in pages->regs. The stack grows upwards, so
185 * we start it at the end of that structure. */
186 state->guest_tss.esp0 = (long)(&pages->regs + 1);
187 /* And this is the GDT entry to use for the stack: we keep a
188 * couple of special LGUEST entries. */
189 state->guest_tss.ss0 = LGUEST_DS;
191 /* x86 can have a finegrained bitmap which indicates what I/O
192 * ports the process can use. We set it to the end of our
193 * structure, meaning "none". */
194 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
196 /* Some GDT entries are the same across all Guests, so we can
197 * set them up now. */
198 setup_default_gdt_entries(state);
199 /* Most IDT entries are the same for all Guests, too.*/
200 setup_default_idt_entries(state, default_idt_entries);
202 /* The Host needs to be able to use the LGUEST segments on this
203 * CPU, too, so put them in the Host GDT. */
204 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
205 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
208 /* In the Switcher, we want the %cs segment register to use the
209 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
210 * it will be undisturbed when we switch. To change %cs and jump we
211 * need this structure to feed to Intel's "lcall" instruction. */
212 lguest_entry.offset = (long)switch_to_guest + switcher_offset();
213 lguest_entry.segment = LGUEST_CS;
215 printk(KERN_INFO "lguest: mapped switcher at %p\n",
217 /* And we succeeded... */
221 vunmap(switcher_vma->addr);
223 i = TOTAL_SWITCHER_PAGES;
225 for (--i; i >= 0; i--)
226 __free_pages(switcher_page[i], 0);
227 kfree(switcher_page);
233 /* Cleaning up the mapping when the module is unloaded is almost...
235 static void unmap_switcher(void)
239 /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
240 vunmap(switcher_vma->addr);
241 /* Now we just need to free the pages we copied the switcher into */
242 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
243 __free_pages(switcher_page[i], 0);
246 /*H:130 Our Guest is usually so well behaved; it never tries to do things it
247 * isn't allowed to. Unfortunately, Linux's paravirtual infrastructure isn't
248 * quite complete, because it doesn't contain replacements for the Intel I/O
249 * instructions. As a result, the Guest sometimes fumbles across one during
250 * the boot process as it probes for various things which are usually attached
253 * When the Guest uses one of these instructions, we get trap #13 (General
254 * Protection Fault) and come here. We see if it's one of those troublesome
255 * instructions and skip over it. We return true if we did. */
256 static int emulate_insn(struct lguest *lg)
259 unsigned int insnlen = 0, in = 0, shift = 0;
260 /* The eip contains the *virtual* address of the Guest's instruction:
261 * guest_pa just subtracts the Guest's page_offset. */
262 unsigned long physaddr = guest_pa(lg, lg->regs->eip);
264 /* The guest_pa() function only works for Guest kernel addresses, but
265 * that's all we're trying to do anyway. */
266 if (lg->regs->eip < lg->page_offset)
269 /* Decoding x86 instructions is icky. */
270 lgread(lg, &insn, physaddr, 1);
272 /* 0x66 is an "operand prefix". It means it's using the upper 16 bits
273 of the eax register. */
276 /* The instruction is 1 byte so far, read the next byte. */
278 lgread(lg, &insn, physaddr + insnlen, 1);
281 /* We can ignore the lower bit for the moment and decode the 4 opcodes
282 * we need to emulate. */
283 switch (insn & 0xFE) {
284 case 0xE4: /* in <next byte>,%al */
288 case 0xEC: /* in (%dx),%al */
292 case 0xE6: /* out %al,<next byte> */
295 case 0xEE: /* out %al,(%dx) */
299 /* OK, we don't know what this is, can't emulate. */
303 /* If it was an "IN" instruction, they expect the result to be read
304 * into %eax, so we change %eax. We always return all-ones, which
305 * traditionally means "there's nothing there". */
307 /* Lower bit tells is whether it's a 16 or 32 bit access */
309 lg->regs->eax = 0xFFFFFFFF;
311 lg->regs->eax |= (0xFFFF << shift);
313 /* Finally, we've "done" the instruction, so move past it. */
314 lg->regs->eip += insnlen;
321 * Dealing With Guest Memory.
323 * When the Guest gives us (what it thinks is) a physical address, we can use
324 * the normal copy_from_user() & copy_to_user() on the corresponding place in
325 * the memory region allocated by the Launcher.
327 * But we can't trust the Guest: it might be trying to access the Launcher
328 * code. We have to check that the range is below the pfn_limit the Launcher
329 * gave us. We have to make sure that addr + len doesn't give us a false
330 * positive by overflowing, too. */
331 int lguest_address_ok(const struct lguest *lg,
332 unsigned long addr, unsigned long len)
334 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
337 /* This is a convenient routine to get a 32-bit value from the Guest (a very
338 * common operation). Here we can see how useful the kill_lguest() routine we
339 * met in the Launcher can be: we return a random value (0) instead of needing
340 * to return an error. */
341 u32 lgread_u32(struct lguest *lg, unsigned long addr)
345 /* Don't let them access lguest binary. */
346 if (!lguest_address_ok(lg, addr, sizeof(val))
347 || get_user(val, (u32 *)(lg->mem_base + addr)) != 0)
348 kill_guest(lg, "bad read address %#lx: pfn_limit=%u membase=%p", addr, lg->pfn_limit, lg->mem_base);
352 /* Same thing for writing a value. */
353 void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val)
355 if (!lguest_address_ok(lg, addr, sizeof(val))
356 || put_user(val, (u32 *)(lg->mem_base + addr)) != 0)
357 kill_guest(lg, "bad write address %#lx", addr);
360 /* This routine is more generic, and copies a range of Guest bytes into a
361 * buffer. If the copy_from_user() fails, we fill the buffer with zeroes, so
362 * the caller doesn't end up using uninitialized kernel memory. */
363 void lgread(struct lguest *lg, void *b, unsigned long addr, unsigned bytes)
365 if (!lguest_address_ok(lg, addr, bytes)
366 || copy_from_user(b, lg->mem_base + addr, bytes) != 0) {
367 /* copy_from_user should do this, but as we rely on it... */
369 kill_guest(lg, "bad read address %#lx len %u", addr, bytes);
373 /* Similarly, our generic routine to copy into a range of Guest bytes. */
374 void lgwrite(struct lguest *lg, unsigned long addr, const void *b,
377 if (!lguest_address_ok(lg, addr, bytes)
378 || copy_to_user(lg->mem_base + addr, b, bytes) != 0)
379 kill_guest(lg, "bad write address %#lx len %u", addr, bytes);
381 /* (end of memory access helper routines) :*/
383 static void set_ts(void)
393 * We are getting close to the Switcher.
395 * Remember that each CPU has two pages which are visible to the Guest when it
396 * runs on that CPU. This has to contain the state for that Guest: we copy the
397 * state in just before we run the Guest.
399 * Each Guest has "changed" flags which indicate what has changed in the Guest
400 * since it last ran. We saw this set in interrupts_and_traps.c and
403 static void copy_in_guest_info(struct lguest *lg, struct lguest_pages *pages)
405 /* Copying all this data can be quite expensive. We usually run the
406 * same Guest we ran last time (and that Guest hasn't run anywhere else
407 * meanwhile). If that's not the case, we pretend everything in the
408 * Guest has changed. */
409 if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
410 __get_cpu_var(last_guest) = lg;
411 lg->last_pages = pages;
412 lg->changed = CHANGED_ALL;
415 /* These copies are pretty cheap, so we do them unconditionally: */
416 /* Save the current Host top-level page directory. */
417 pages->state.host_cr3 = __pa(current->mm->pgd);
418 /* Set up the Guest's page tables to see this CPU's pages (and no
419 * other CPU's pages). */
420 map_switcher_in_guest(lg, pages);
421 /* Set up the two "TSS" members which tell the CPU what stack to use
422 * for traps which do directly into the Guest (ie. traps at privilege
424 pages->state.guest_tss.esp1 = lg->esp1;
425 pages->state.guest_tss.ss1 = lg->ss1;
427 /* Copy direct-to-Guest trap entries. */
428 if (lg->changed & CHANGED_IDT)
429 copy_traps(lg, pages->state.guest_idt, default_idt_entries);
431 /* Copy all GDT entries which the Guest can change. */
432 if (lg->changed & CHANGED_GDT)
433 copy_gdt(lg, pages->state.guest_gdt);
434 /* If only the TLS entries have changed, copy them. */
435 else if (lg->changed & CHANGED_GDT_TLS)
436 copy_gdt_tls(lg, pages->state.guest_gdt);
438 /* Mark the Guest as unchanged for next time. */
442 /* Finally: the code to actually call into the Switcher to run the Guest. */
443 static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
445 /* This is a dummy value we need for GCC's sake. */
446 unsigned int clobber;
448 /* Copy the guest-specific information into this CPU's "struct
450 copy_in_guest_info(lg, pages);
452 /* Set the trap number to 256 (impossible value). If we fault while
453 * switching to the Guest (bad segment registers or bug), this will
454 * cause us to abort the Guest. */
455 lg->regs->trapnum = 256;
457 /* Now: we push the "eflags" register on the stack, then do an "lcall".
458 * This is how we change from using the kernel code segment to using
459 * the dedicated lguest code segment, as well as jumping into the
462 * The lcall also pushes the old code segment (KERNEL_CS) onto the
463 * stack, then the address of this call. This stack layout happens to
464 * exactly match the stack of an interrupt... */
465 asm volatile("pushf; lcall *lguest_entry"
466 /* This is how we tell GCC that %eax ("a") and %ebx ("b")
467 * are changed by this routine. The "=" means output. */
468 : "=a"(clobber), "=b"(clobber)
469 /* %eax contains the pages pointer. ("0" refers to the
470 * 0-th argument above, ie "a"). %ebx contains the
471 * physical address of the Guest's top-level page
473 : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
474 /* We tell gcc that all these registers could change,
475 * which means we don't have to save and restore them in
477 : "memory", "%edx", "%ecx", "%edi", "%esi");
481 /*H:030 Let's jump straight to the the main loop which runs the Guest.
482 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
483 * going around and around until something interesting happens. */
484 int run_guest(struct lguest *lg, unsigned long __user *user)
486 /* We stop running once the Guest is dead. */
488 /* We need to initialize this, otherwise gcc complains. It's
489 * not (yet) clever enough to see that it's initialized when we
491 unsigned int cr2 = 0; /* Damn gcc */
493 /* First we run any hypercalls the Guest wants done: either in
494 * the hypercall ring in "struct lguest_data", or directly by
495 * using int 31 (LGUEST_TRAP_ENTRY). */
497 /* It's possible the Guest did a SEND_DMA hypercall to the
498 * Launcher, in which case we return from the read() now. */
499 if (lg->dma_is_pending) {
500 if (put_user(lg->pending_dma, user) ||
501 put_user(lg->pending_key, user+1))
503 return sizeof(unsigned long)*2;
506 /* Check for signals */
507 if (signal_pending(current))
510 /* If Waker set break_out, return to Launcher. */
514 /* Check if there are any interrupts which can be delivered
515 * now: if so, this sets up the hander to be executed when we
516 * next run the Guest. */
517 maybe_do_interrupt(lg);
519 /* All long-lived kernel loops need to check with this horrible
520 * thing called the freezer. If the Host is trying to suspend,
524 /* Just make absolutely sure the Guest is still alive. One of
525 * those hypercalls could have been fatal, for example. */
529 /* If the Guest asked to be stopped, we sleep. The Guest's
530 * clock timer or LHCALL_BREAK from the Waker will wake us. */
532 set_current_state(TASK_INTERRUPTIBLE);
537 /* OK, now we're ready to jump into the Guest. First we put up
538 * the "Do Not Disturb" sign: */
541 /* Remember the awfully-named TS bit? If the Guest has asked
542 * to set it we set it now, so we can trap and pass that trap
543 * to the Guest if it uses the FPU. */
547 /* SYSENTER is an optimized way of doing system calls. We
548 * can't allow it because it always jumps to privilege level 0.
549 * A normal Guest won't try it because we don't advertise it in
550 * CPUID, but a malicious Guest (or malicious Guest userspace
551 * program) could, so we tell the CPU to disable it before
552 * running the Guest. */
553 if (boot_cpu_has(X86_FEATURE_SEP))
554 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
556 /* Now we actually run the Guest. It will pop back out when
557 * something interesting happens, and we can examine its
558 * registers to see what it was doing. */
559 run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
561 /* The "regs" pointer contains two extra entries which are not
562 * really registers: a trap number which says what interrupt or
563 * trap made the switcher code come back, and an error code
564 * which some traps set. */
566 /* If the Guest page faulted, then the cr2 register will tell
567 * us the bad virtual address. We have to grab this now,
568 * because once we re-enable interrupts an interrupt could
569 * fault and thus overwrite cr2, or we could even move off to a
571 if (lg->regs->trapnum == 14)
573 /* Similarly, if we took a trap because the Guest used the FPU,
574 * we have to restore the FPU it expects to see. */
575 else if (lg->regs->trapnum == 7)
576 math_state_restore();
578 /* Restore SYSENTER if it's supposed to be on. */
579 if (boot_cpu_has(X86_FEATURE_SEP))
580 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
582 /* Now we're ready to be interrupted or moved to other CPUs */
585 /* OK, so what happened? */
586 switch (lg->regs->trapnum) {
587 case 13: /* We've intercepted a GPF. */
588 /* Check if this was one of those annoying IN or OUT
589 * instructions which we need to emulate. If so, we
590 * just go back into the Guest after we've done it. */
591 if (lg->regs->errcode == 0) {
592 if (emulate_insn(lg))
596 case 14: /* We've intercepted a page fault. */
597 /* The Guest accessed a virtual address that wasn't
598 * mapped. This happens a lot: we don't actually set
599 * up most of the page tables for the Guest at all when
600 * we start: as it runs it asks for more and more, and
601 * we set them up as required. In this case, we don't
602 * even tell the Guest that the fault happened.
604 * The errcode tells whether this was a read or a
605 * write, and whether kernel or userspace code. */
606 if (demand_page(lg, cr2, lg->regs->errcode))
609 /* OK, it's really not there (or not OK): the Guest
610 * needs to know. We write out the cr2 value so it
611 * knows where the fault occurred.
613 * Note that if the Guest were really messed up, this
614 * could happen before it's done the INITIALIZE
615 * hypercall, so lg->lguest_data will be NULL */
617 && put_user(cr2, &lg->lguest_data->cr2))
618 kill_guest(lg, "Writing cr2");
620 case 7: /* We've intercepted a Device Not Available fault. */
621 /* If the Guest doesn't want to know, we already
622 * restored the Floating Point Unit, so we just
623 * continue without telling it. */
628 /* These values mean a real interrupt occurred, in
629 * which case the Host handler has already been run.
630 * We just do a friendly check if another process
631 * should now be run, then fall through to loop
634 case LGUEST_TRAP_ENTRY: /* Handled at top of loop */
638 /* If we get here, it's a trap the Guest wants to know
640 if (deliver_trap(lg, lg->regs->trapnum))
643 /* If the Guest doesn't have a handler (either it hasn't
644 * registered any yet, or it's one of the faults we don't let
645 * it handle), it dies with a cryptic error message. */
646 kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
647 lg->regs->trapnum, lg->regs->eip,
648 lg->regs->trapnum == 14 ? cr2 : lg->regs->errcode);
650 /* The Guest is dead => "No such file or directory" */
654 /* Now we can look at each of the routines this calls, in increasing order of
655 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
656 * deliver_trap() and demand_page(). After all those, we'll be ready to
657 * examine the Switcher, and our philosophical understanding of the Host/Guest
658 * duality will be complete. :*/
659 static void adjust_pge(void *on)
662 write_cr4(read_cr4() | X86_CR4_PGE);
664 write_cr4(read_cr4() & ~X86_CR4_PGE);
668 * Welcome to the Host!
670 * By this point your brain has been tickled by the Guest code and numbed by
671 * the Launcher code; prepare for it to be stretched by the Host code. This is
672 * the heart. Let's begin at the initialization routine for the Host's lg
675 static int __init init(void)
679 /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */
680 if (paravirt_enabled()) {
681 printk("lguest is afraid of %s\n", pv_info.name);
685 /* First we put the Switcher up in very high virtual memory. */
686 err = map_switcher();
690 /* Now we set up the pagetable implementation for the Guests. */
691 err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
697 /* The I/O subsystem needs some things initialized. */
700 /* /dev/lguest needs to be registered. */
701 err = lguest_device_init();
708 /* Finally, we need to turn off "Page Global Enable". PGE is an
709 * optimization where page table entries are specially marked to show
710 * they never change. The Host kernel marks all the kernel pages this
711 * way because it's always present, even when userspace is running.
713 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
714 * switch to the Guest kernel. If you don't disable this on all CPUs,
715 * you'll get really weird bugs that you'll chase for two days.
717 * I used to turn PGE off every time we switched to the Guest and back
718 * on when we return, but that slowed the Switcher down noticibly. */
720 /* We don't need the complexity of CPUs coming and going while we're
723 if (cpu_has_pge) { /* We have a broader idea of "global". */
724 /* Remember that this was originally set (for cleanup). */
726 /* adjust_pge is a helper function which sets or unsets the PGE
727 * bit on its CPU, depending on the argument (0 == unset). */
728 on_each_cpu(adjust_pge, (void *)0, 0, 1);
729 /* Turn off the feature in the global feature set. */
730 clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
732 unlock_cpu_hotplug();
738 /* Cleaning up is just the same code, backwards. With a little French. */
739 static void __exit fini(void)
741 lguest_device_remove();
745 /* If we had PGE before we started, turn it back on now. */
748 set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
749 /* adjust_pge's argument "1" means set PGE. */
750 on_each_cpu(adjust_pge, (void *)1, 0, 1);
752 unlock_cpu_hotplug();
755 /* The Host side of lguest can be a module. This is a nice way for people to
759 MODULE_LICENSE("GPL");
760 MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>");