path: root/drivers/lguest/core.c

/*P:400 This contains run_guest() which actually calls into the Host<->Guest
 * Switcher and analyzes the return, such as determining if the Guest wants the
 * Host to do something.  This file also contains useful helper routines, and a
 * couple of non-obvious setup and teardown pieces which were implemented after
 * days of debugging pain. :*/
#include <linux/module.h>
#include <linux/stringify.h>
#include <linux/stddef.h>
#include <linux/io.h>
#include <linux/mm.h>
#include <linux/vmalloc.h>
#include <linux/cpu.h>
#include <linux/freezer.h>
#include <asm/paravirt.h>
#include <asm/desc.h>
#include <asm/pgtable.h>
#include <asm/uaccess.h>
#include <asm/poll.h>
#include <asm/highmem.h>
#include <asm/asm-offsets.h>
#include <asm/i387.h>
#include "lg.h"

/* Found in switcher.S */
extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
extern unsigned long default_idt_entries[];

/* Every guest maps the core switcher code. */
#define SHARED_SWITCHER_PAGES \
	DIV_ROUND_UP(end_switcher_text - start_switcher_text, PAGE_SIZE)
/* Pages for switcher itself, then two pages per cpu */
#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * NR_CPUS)

/* We map at -4M for ease of mapping into the guest (one PTE page). */
#define SWITCHER_ADDR 0xFFC00000

static struct vm_struct *switcher_vma;
static struct page **switcher_page;

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

/* This One Big lock protects all inter-guest data structures. */
DEFINE_MUTEX(lguest_lock);
static DEFINE_PER_CPU(struct lguest *, last_guest);

/* FIXME: Make dynamic. */
#define MAX_LGUEST_GUESTS 16
struct lguest lguests[MAX_LGUEST_GUESTS];

/* 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]);
}

/*H:010 We need to set up the Switcher at a high virtual address.  Remember the
 * Switcher is a few hundred bytes of assembler code which actually changes the
 * CPU to run the Guest, and then changes back to the Host when a trap or
 * interrupt happens.
 *
 * The Switcher code must be at the same virtual address in the Guest as the
 * Host since it will be running as the switchover occurs.
 *
 * Trying to map memory at a particular address is an unusual thing to do, so
 * it's not a simple one-liner.  We also set up the per-cpu parts of the
 * Switcher here.
 */
static __init int map_switcher(void)
{
	int i, err;
	struct page **pagep;

	/*
	 * Map the Switcher in to high memory.
	 *
	 * It turns out that if we choose the address 0xFFC00000 (4MB under the
	 * top virtual address), it makes setting up the page tables really
	 * easy.
	 */

	/* We allocate an array of "struct page"s.  map_vm_area() wants the
	 * pages in this form, rather than just an array of pointers. */
	switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
				GFP_KERNEL);
	if (!switcher_page) {
		err = -ENOMEM;
		goto out;
	}

	/* Now we actually allocate the pages.  The Guest will see these pages,
	 * so we make sure they're zeroed. */
	for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
		unsigned long addr = get_zeroed_page(GFP_KERNEL);
		if (!addr) {
			err = -ENOMEM;
			goto free_some_pages;
		}
		switcher_page[i] = virt_to_page(addr);
	}

	/* Now we reserve the "virtual memory area" we want: 0xFFC00000
	 * (SWITCHER_ADDR).  We might not get it in theory, but in practice
	 * it's worked so far. */
	switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
				       VM_ALLOC, SWITCHER_ADDR, VMALLOC_END);
	if (!switcher_vma) {
		err = -ENOMEM;
		printk("lguest: could not map switcher pages high\n");
		goto free_pages;
	}

	/* This code actually sets up the pages we've allocated to appear at
	 * SWITCHER_ADDR.  map_vm_area() takes the vma we allocated above, the
	 * kind of pages we're mapping (kernel pages), and a pointer to our
	 * array of struct pages.  It increments that pointer, but we don't
	 * care. */
	pagep = switcher_page;
	err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
	if (err) {
		printk("lguest: map_vm_area failed: %i\n", err);
		goto free_vma;
	}

	/* Now the switcher is mapped at the right address, we can't fail!
	 * Copy in the compiled-in Switcher code (from switcher.S). */
	memcpy(switcher_vma->addr, start_switcher_text,
	       end_switcher_text - start_switcher_text);

	/* Most of the switcher.S doesn't care that it's been moved; on 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
	 * convenience function which returns the distance between the builtin
	 * switcher code and the high-mapped copy we just made. */
	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);
		/* This is a convenience pointer to make the code fit one
		 * statement to a line. */
		struct lguest_ro_state *state = &pages->state;

		/* The Global Descriptor Table: the Host has a different one
		 * 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
		 * byte, not the size, hence the "-1"). */
		state->host_gdt_desc.size = GDT_SIZE-1;
		state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);

		/* All CPUs on the Host use the same Interrupt Descriptor
		 * Table, so we just use store_idt(), which gets this CPU's IDT
		 * descriptor. */
		store_idt(&state->host_idt_desc);

		/* The descriptors for the Guest's GDT and IDT can be filled
		 * out now, too.  We copy the GDT & IDT into ->guest_gdt and
		 * ->guest_idt before actually running the Guest. */
		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;

		/* We know where we want the stack to be when the Guest enters
		 * the switcher: in pages->regs.  The stack grows upwards, so
		 * we start it at the end of that structure. */
		state->guest_tss.esp0 = (long)(&pages->regs + 1);
		/* And this is the GDT entry to use for the stack: we keep a
		 * couple of special LGUEST entries. */
		state->guest_tss.ss0 = LGUEST_DS;

		/* x86 can have a finegrained bitmap which indicates what I/O
		 * ports the process can use.  We set it to the end of our
		 * structure, meaning "none". */
		state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);

		/* Some GDT entries are the same across all Guests, so we can
		 * set them up now. */
		setup_default_gdt_entries(state);
		/* Most IDT entries are the same for all Guests, too.*/
		setup_default_idt_entries(state, default_idt_entries);

		/* The Host needs to be able to use the LGUEST segments on this
		 * CPU, too, so put them in the Host GDT. */
		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
	}

	/* In the Switcher, we want the %cs segment register to use the
	 * 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
	 * need this structure to feed to Intel's "lcall" instruction. */
	lguest_entry.offset = (long)switch_to_guest + switcher_offset();
	lguest_entry.segment = LGUEST_CS;

	printk(KERN_INFO "lguest: mapped switcher at %p\n",
	       switcher_vma->addr);
	/* And we succeeded... */
	return 0;

free_vma:
	vunmap(switcher_vma->addr);
free_pages:
	i = TOTAL_SWITCHER_PAGES;
free_some_pages:
	for (--i; i >= 0; i--)
		__free_pages(switcher_page[i], 0);
	kfree(switcher_page);
out: