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 <linux/highmem.h>
15 #include <asm/paravirt.h>
16 #include <asm/pgtable.h>
17 #include <asm/uaccess.h>
19 #include <asm/asm-offsets.h>
23 static struct vm_struct *switcher_vma;
24 static struct page **switcher_page;
26 /* This One Big lock protects all inter-guest data structures. */
27 DEFINE_MUTEX(lguest_lock);
29 /*H:010 We need to set up the Switcher at a high virtual address. Remember the
30 * Switcher is a few hundred bytes of assembler code which actually changes the
31 * CPU to run the Guest, and then changes back to the Host when a trap or
34 * The Switcher code must be at the same virtual address in the Guest as the
35 * Host since it will be running as the switchover occurs.
37 * Trying to map memory at a particular address is an unusual thing to do, so
38 * it's not a simple one-liner. */
39 static __init int map_switcher(void)
45 * Map the Switcher in to high memory.
47 * It turns out that if we choose the address 0xFFC00000 (4MB under the
48 * top virtual address), it makes setting up the page tables really
52 /* We allocate an array of "struct page"s. map_vm_area() wants the
53 * pages in this form, rather than just an array of pointers. */
54 switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
61 /* Now we actually allocate the pages. The Guest will see these pages,
62 * so we make sure they're zeroed. */
63 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
64 unsigned long addr = get_zeroed_page(GFP_KERNEL);
69 switcher_page[i] = virt_to_page(addr);
72 /* Now we reserve the "virtual memory area" we want: 0xFFC00000
73 * (SWITCHER_ADDR). We might not get it in theory, but in practice
74 * it's worked so far. */
75 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
76 VM_ALLOC, SWITCHER_ADDR, VMALLOC_END);
79 printk("lguest: could not map switcher pages high\n");
83 /* This code actually sets up the pages we've allocated to appear at
84 * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
85 * kind of pages we're mapping (kernel pages), and a pointer to our
86 * array of struct pages. It increments that pointer, but we don't
88 pagep = switcher_page;
89 err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep);
91 printk("lguest: map_vm_area failed: %i\n", err);
95 /* Now the Switcher is mapped at the right address, we can't fail!
96 * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
97 memcpy(switcher_vma->addr, start_switcher_text,
98 end_switcher_text - start_switcher_text);
100 printk(KERN_INFO "lguest: mapped switcher at %p\n",
102 /* And we succeeded... */
106 vunmap(switcher_vma->addr);
108 i = TOTAL_SWITCHER_PAGES;
110 for (--i; i >= 0; i--)
111 __free_pages(switcher_page[i], 0);
112 kfree(switcher_page);
118 /* Cleaning up the mapping when the module is unloaded is almost...
120 static void unmap_switcher(void)
124 /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
125 vunmap(switcher_vma->addr);
126 /* Now we just need to free the pages we copied the switcher into */
127 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
128 __free_pages(switcher_page[i], 0);
132 * Dealing With Guest Memory.
134 * Before we go too much further into the Host, we need to grok the routines
135 * we use to deal with Guest memory.
137 * When the Guest gives us (what it thinks is) a physical address, we can use
138 * the normal copy_from_user() & copy_to_user() on the corresponding place in
139 * the memory region allocated by the Launcher.
141 * But we can't trust the Guest: it might be trying to access the Launcher
142 * code. We have to check that the range is below the pfn_limit the Launcher
143 * gave us. We have to make sure that addr + len doesn't give us a false
144 * positive by overflowing, too. */
145 int lguest_address_ok(const struct lguest *lg,
146 unsigned long addr, unsigned long len)
148 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
151 /* This routine copies memory from the Guest. Here we can see how useful the
152 * kill_lguest() routine we met in the Launcher can be: we return a random
153 * value (all zeroes) instead of needing to return an error. */
154 void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
156 if (!lguest_address_ok(cpu->lg, addr, bytes)
157 || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) {
158 /* copy_from_user should do this, but as we rely on it... */
160 kill_guest(cpu, "bad read address %#lx len %u", addr, bytes);
164 /* This is the write (copy into guest) version. */
165 void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
168 if (!lguest_address_ok(cpu->lg, addr, bytes)
169 || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0)
170 kill_guest(cpu, "bad write address %#lx len %u", addr, bytes);
174 /*H:030 Let's jump straight to the the main loop which runs the Guest.
175 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
176 * going around and around until something interesting happens. */
177 int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
179 /* We stop running once the Guest is dead. */
180 while (!cpu->lg->dead) {
181 /* First we run any hypercalls the Guest wants done. */
185 /* It's possible the Guest did a NOTIFY hypercall to the
186 * Launcher, in which case we return from the read() now. */
187 if (cpu->pending_notify) {
188 if (put_user(cpu->pending_notify, user))
190 return sizeof(cpu->pending_notify);
193 /* Check for signals */
194 if (signal_pending(current))
197 /* If Waker set break_out, return to Launcher. */
201 /* Check if there are any interrupts which can be delivered
202 * now: if so, this sets up the hander to be executed when we
203 * next run the Guest. */
204 maybe_do_interrupt(cpu);
206 /* All long-lived kernel loops need to check with this horrible
207 * thing called the freezer. If the Host is trying to suspend,
211 /* Just make absolutely sure the Guest is still alive. One of
212 * those hypercalls could have been fatal, for example. */
216 /* If the Guest asked to be stopped, we sleep. The Guest's
217 * clock timer or LHCALL_BREAK from the Waker will wake us. */
219 set_current_state(TASK_INTERRUPTIBLE);
224 /* OK, now we're ready to jump into the Guest. First we put up
225 * the "Do Not Disturb" sign: */
228 /* Actually run the Guest until something happens. */
229 lguest_arch_run_guest(cpu);
231 /* Now we're ready to be interrupted or moved to other CPUs */
234 /* Now we deal with whatever happened to the Guest. */
235 lguest_arch_handle_trap(cpu);
238 if (cpu->lg->dead == ERR_PTR(-ERESTART))
240 /* The Guest is dead => "No such file or directory" */
245 * Welcome to the Host!
247 * By this point your brain has been tickled by the Guest code and numbed by
248 * the Launcher code; prepare for it to be stretched by the Host code. This is
249 * the heart. Let's begin at the initialization routine for the Host's lg
252 static int __init init(void)
256 /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */
257 if (paravirt_enabled()) {
258 printk("lguest is afraid of being a guest\n");
262 /* First we put the Switcher up in very high virtual memory. */
263 err = map_switcher();
267 /* Now we set up the pagetable implementation for the Guests. */
268 err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
272 /* We might need to reserve an interrupt vector. */
273 err = init_interrupts();
277 /* /dev/lguest needs to be registered. */
278 err = lguest_device_init();
280 goto free_interrupts;
282 /* Finally we do some architecture-specific setup. */
283 lguest_arch_host_init();
298 /* Cleaning up is just the same code, backwards. With a little French. */
299 static void __exit fini(void)
301 lguest_device_remove();
306 lguest_arch_host_fini();
310 /* The Host side of lguest can be a module. This is a nice way for people to
314 MODULE_LICENSE("GPL");
315 MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>");