2 * This contains run_guest() which actually calls into the Host<->Guest
3 * Switcher and analyzes the return, such as determining if the Guest wants the
4 * Host to do something. This file also contains useful helper routines.
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 <linux/highmem.h>
15 #include <linux/slab.h>
16 #include <asm/paravirt.h>
17 #include <asm/pgtable.h>
18 #include <asm/uaccess.h>
20 #include <asm/asm-offsets.h>
23 unsigned long switcher_addr;
24 struct page **lg_switcher_pages;
25 static struct vm_struct *switcher_vma;
27 /* This One Big lock protects all inter-guest data structures. */
28 DEFINE_MUTEX(lguest_lock);
31 * We need to set up the Switcher at a high virtual address. Remember the
32 * Switcher is a few hundred bytes of assembler code which actually changes the
33 * CPU to run the Guest, and then changes back to the Host when a trap or
36 * The Switcher code must be at the same virtual address in the Guest as the
37 * Host since it will be running as the switchover occurs.
39 * Trying to map memory at a particular address is an unusual thing to do, so
40 * it's not a simple one-liner.
42 static __init int map_switcher(void)
48 * Map the Switcher in to high memory.
50 * It turns out that if we choose the address 0xFFC00000 (4MB under the
51 * top virtual address), it makes setting up the page tables really
55 /* We assume Switcher text fits into a single page. */
56 if (end_switcher_text - start_switcher_text > PAGE_SIZE) {
57 printk(KERN_ERR "lguest: switcher text too large (%zu)\n",
58 end_switcher_text - start_switcher_text);
63 * We allocate an array of struct page pointers. map_vm_area() wants
64 * this, rather than just an array of pages.
66 lg_switcher_pages = kmalloc(sizeof(lg_switcher_pages[0])
67 * TOTAL_SWITCHER_PAGES,
69 if (!lg_switcher_pages) {
75 * Now we actually allocate the pages. The Guest will see these pages,
76 * so we make sure they're zeroed.
78 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
79 lg_switcher_pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
80 if (!lg_switcher_pages[i]) {
86 switcher_addr = SWITCHER_ADDR;
89 * First we check that the Switcher won't overlap the fixmap area at
90 * the top of memory. It's currently nowhere near, but it could have
91 * very strange effects if it ever happened.
93 if (switcher_addr + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
95 printk("lguest: mapping switcher would thwack fixmap\n");
100 * Now we reserve the "virtual memory area" we want. We might
101 * not get it in theory, but in practice it's worked so far.
102 * The end address needs +1 because __get_vm_area allocates an
103 * extra guard page, so we need space for that.
105 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
106 VM_ALLOC, switcher_addr, switcher_addr
107 + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
110 printk("lguest: could not map switcher pages high\n");
115 * This code actually sets up the pages we've allocated to appear at
116 * switcher_addr. map_vm_area() takes the vma we allocated above, the
117 * kind of pages we're mapping (kernel pages), and a pointer to our
118 * array of struct pages. It increments that pointer, but we don't
121 pagep = lg_switcher_pages;
122 err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep);
124 printk("lguest: map_vm_area failed: %i\n", err);
129 * Now the Switcher is mapped at the right address, we can't fail!
130 * Copy in the compiled-in Switcher code (from x86/switcher_32.S).
132 memcpy(switcher_vma->addr, start_switcher_text,
133 end_switcher_text - start_switcher_text);
135 printk(KERN_INFO "lguest: mapped switcher at %p\n",
137 /* And we succeeded... */
141 vunmap(switcher_vma->addr);
143 i = TOTAL_SWITCHER_PAGES;
145 for (--i; i >= 0; i--)
146 __free_pages(lg_switcher_pages[i], 0);
147 kfree(lg_switcher_pages);
153 /* Cleaning up the mapping when the module is unloaded is almost... too easy. */
154 static void unmap_switcher(void)
158 /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
159 vunmap(switcher_vma->addr);
160 /* Now we just need to free the pages we copied the switcher into */
161 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
162 __free_pages(lg_switcher_pages[i], 0);
163 kfree(lg_switcher_pages);
167 * Dealing With Guest Memory.
169 * Before we go too much further into the Host, we need to grok the routines
170 * we use to deal with Guest memory.
172 * When the Guest gives us (what it thinks is) a physical address, we can use
173 * the normal copy_from_user() & copy_to_user() on the corresponding place in
174 * the memory region allocated by the Launcher.
176 * But we can't trust the Guest: it might be trying to access the Launcher
177 * code. We have to check that the range is below the pfn_limit the Launcher
178 * gave us. We have to make sure that addr + len doesn't give us a false
179 * positive by overflowing, too.
181 bool lguest_address_ok(const struct lguest *lg,
182 unsigned long addr, unsigned long len)
184 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
188 * This routine copies memory from the Guest. Here we can see how useful the
189 * kill_lguest() routine we met in the Launcher can be: we return a random
190 * value (all zeroes) instead of needing to return an error.
192 void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
194 if (!lguest_address_ok(cpu->lg, addr, bytes)
195 || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) {
196 /* copy_from_user should do this, but as we rely on it... */
198 kill_guest(cpu, "bad read address %#lx len %u", addr, bytes);
202 /* This is the write (copy into Guest) version. */
203 void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
206 if (!lguest_address_ok(cpu->lg, addr, bytes)
207 || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0)
208 kill_guest(cpu, "bad write address %#lx len %u", addr, bytes);
213 * Let's jump straight to the the main loop which runs the Guest.
214 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
215 * going around and around until something interesting happens.
217 int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
219 /* We stop running once the Guest is dead. */
220 while (!cpu->lg->dead) {
224 /* First we run any hypercalls the Guest wants done. */
229 * It's possible the Guest did a NOTIFY hypercall to the
232 if (cpu->pending_notify) {
234 * Does it just needs to write to a registered
235 * eventfd (ie. the appropriate virtqueue thread)?
237 if (!send_notify_to_eventfd(cpu)) {
238 /* OK, we tell the main Launcher. */
239 if (put_user(cpu->pending_notify, user))
241 return sizeof(cpu->pending_notify);
246 * All long-lived kernel loops need to check with this horrible
247 * thing called the freezer. If the Host is trying to suspend,
252 /* Check for signals */
253 if (signal_pending(current))
257 * Check if there are any interrupts which can be delivered now:
258 * if so, this sets up the hander to be executed when we next
261 irq = interrupt_pending(cpu, &more);
262 if (irq < LGUEST_IRQS)
263 try_deliver_interrupt(cpu, irq, more);
266 * Just make absolutely sure the Guest is still alive. One of
267 * those hypercalls could have been fatal, for example.
273 * If the Guest asked to be stopped, we sleep. The Guest's
274 * clock timer will wake us.
277 set_current_state(TASK_INTERRUPTIBLE);
279 * Just before we sleep, make sure no interrupt snuck in
280 * which we should be doing.
282 if (interrupt_pending(cpu, &more) < LGUEST_IRQS)
283 set_current_state(TASK_RUNNING);
290 * OK, now we're ready to jump into the Guest. First we put up
291 * the "Do Not Disturb" sign:
295 /* Actually run the Guest until something happens. */
296 lguest_arch_run_guest(cpu);
298 /* Now we're ready to be interrupted or moved to other CPUs */
301 /* Now we deal with whatever happened to the Guest. */
302 lguest_arch_handle_trap(cpu);
305 /* Special case: Guest is 'dead' but wants a reboot. */
306 if (cpu->lg->dead == ERR_PTR(-ERESTART))
309 /* The Guest is dead => "No such file or directory" */
314 * Welcome to the Host!
316 * By this point your brain has been tickled by the Guest code and numbed by
317 * the Launcher code; prepare for it to be stretched by the Host code. This is
318 * the heart. Let's begin at the initialization routine for the Host's lg
321 static int __init init(void)
325 /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */
326 if (get_kernel_rpl() != 0) {
327 printk("lguest is afraid of being a guest\n");
331 /* First we put the Switcher up in very high virtual memory. */
332 err = map_switcher();
336 /* We might need to reserve an interrupt vector. */
337 err = init_interrupts();
341 /* /dev/lguest needs to be registered. */
342 err = lguest_device_init();
344 goto free_interrupts;
346 /* Finally we do some architecture-specific setup. */
347 lguest_arch_host_init();
360 /* Cleaning up is just the same code, backwards. With a little French. */
361 static void __exit fini(void)
363 lguest_device_remove();
367 lguest_arch_host_fini();
372 * The Host side of lguest can be a module. This is a nice way for people to
377 MODULE_LICENSE("GPL");
378 MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>");