Commit | Line | Data |
---|---|---|
2e04ef76 RR |
1 | /*P:400 |
2 | * This contains run_guest() which actually calls into the Host<->Guest | |
f938d2c8 | 3 | * Switcher and analyzes the return, such as determining if the Guest wants the |
2e04ef76 RR |
4 | * Host to do something. This file also contains useful helper routines. |
5 | :*/ | |
d7e28ffe RR |
6 | #include <linux/module.h> |
7 | #include <linux/stringify.h> | |
8 | #include <linux/stddef.h> | |
9 | #include <linux/io.h> | |
10 | #include <linux/mm.h> | |
11 | #include <linux/vmalloc.h> | |
12 | #include <linux/cpu.h> | |
13 | #include <linux/freezer.h> | |
625efab1 | 14 | #include <linux/highmem.h> |
5a0e3ad6 | 15 | #include <linux/slab.h> |
d7e28ffe | 16 | #include <asm/paravirt.h> |
d7e28ffe RR |
17 | #include <asm/pgtable.h> |
18 | #include <asm/uaccess.h> | |
19 | #include <asm/poll.h> | |
d7e28ffe | 20 | #include <asm/asm-offsets.h> |
d7e28ffe RR |
21 | #include "lg.h" |
22 | ||
d7e28ffe RR |
23 | |
24 | static struct vm_struct *switcher_vma; | |
25 | static struct page **switcher_page; | |
26 | ||
d7e28ffe RR |
27 | /* This One Big lock protects all inter-guest data structures. */ |
28 | DEFINE_MUTEX(lguest_lock); | |
d7e28ffe | 29 | |
2e04ef76 RR |
30 | /*H:010 |
31 | * We need to set up the Switcher at a high virtual address. Remember the | |
bff672e6 RR |
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 | |
34 | * interrupt happens. | |
35 | * | |
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. | |
38 | * | |
39 | * Trying to map memory at a particular address is an unusual thing to do, so | |
2e04ef76 RR |
40 | * it's not a simple one-liner. |
41 | */ | |
d7e28ffe RR |
42 | static __init int map_switcher(void) |
43 | { | |
44 | int i, err; | |
45 | struct page **pagep; | |
46 | ||
bff672e6 RR |
47 | /* |
48 | * Map the Switcher in to high memory. | |
49 | * | |
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 | |
52 | * easy. | |
53 | */ | |
54 | ||
2e04ef76 RR |
55 | /* |
56 | * We allocate an array of struct page pointers. map_vm_area() wants | |
57 | * this, rather than just an array of pages. | |
58 | */ | |
d7e28ffe RR |
59 | switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES, |
60 | GFP_KERNEL); | |
61 | if (!switcher_page) { | |
62 | err = -ENOMEM; | |
63 | goto out; | |
64 | } | |
65 | ||
2e04ef76 RR |
66 | /* |
67 | * Now we actually allocate the pages. The Guest will see these pages, | |
68 | * so we make sure they're zeroed. | |
69 | */ | |
d7e28ffe | 70 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) { |
6c189d83 XG |
71 | switcher_page[i] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
72 | if (!switcher_page[i]) { | |
d7e28ffe RR |
73 | err = -ENOMEM; |
74 | goto free_some_pages; | |
75 | } | |
d7e28ffe RR |
76 | } |
77 | ||
2e04ef76 RR |
78 | /* |
79 | * First we check that the Switcher won't overlap the fixmap area at | |
f14ae652 | 80 | * the top of memory. It's currently nowhere near, but it could have |
2e04ef76 RR |
81 | * very strange effects if it ever happened. |
82 | */ | |
f14ae652 RR |
83 | if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){ |
84 | err = -ENOMEM; | |
85 | printk("lguest: mapping switcher would thwack fixmap\n"); | |
86 | goto free_pages; | |
87 | } | |
88 | ||
2e04ef76 RR |
89 | /* |
90 | * Now we reserve the "virtual memory area" we want: 0xFFC00000 | |
bff672e6 | 91 | * (SWITCHER_ADDR). We might not get it in theory, but in practice |
f14ae652 | 92 | * it's worked so far. The end address needs +1 because __get_vm_area |
2e04ef76 RR |
93 | * allocates an extra guard page, so we need space for that. |
94 | */ | |
d7e28ffe | 95 | switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE, |
f14ae652 RR |
96 | VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR |
97 | + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE); | |
d7e28ffe RR |
98 | if (!switcher_vma) { |
99 | err = -ENOMEM; | |
100 | printk("lguest: could not map switcher pages high\n"); | |
101 | goto free_pages; | |
102 | } | |
103 | ||
2e04ef76 RR |
104 | /* |
105 | * This code actually sets up the pages we've allocated to appear at | |
bff672e6 RR |
106 | * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the |
107 | * kind of pages we're mapping (kernel pages), and a pointer to our | |
108 | * array of struct pages. It increments that pointer, but we don't | |
2e04ef76 RR |
109 | * care. |
110 | */ | |
d7e28ffe | 111 | pagep = switcher_page; |
ed1dc778 | 112 | err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep); |
d7e28ffe RR |
113 | if (err) { |
114 | printk("lguest: map_vm_area failed: %i\n", err); | |
115 | goto free_vma; | |
116 | } | |
bff672e6 | 117 | |
2e04ef76 RR |
118 | /* |
119 | * Now the Switcher is mapped at the right address, we can't fail! | |
120 | * Copy in the compiled-in Switcher code (from <arch>_switcher.S). | |
121 | */ | |
d7e28ffe RR |
122 | memcpy(switcher_vma->addr, start_switcher_text, |
123 | end_switcher_text - start_switcher_text); | |
124 | ||
d7e28ffe RR |
125 | printk(KERN_INFO "lguest: mapped switcher at %p\n", |
126 | switcher_vma->addr); | |
bff672e6 | 127 | /* And we succeeded... */ |
d7e28ffe RR |
128 | return 0; |
129 | ||
130 | free_vma: | |
131 | vunmap(switcher_vma->addr); | |
132 | free_pages: | |
133 | i = TOTAL_SWITCHER_PAGES; | |
134 | free_some_pages: | |
135 | for (--i; i >= 0; i--) | |
136 | __free_pages(switcher_page[i], 0); | |
137 | kfree(switcher_page); | |
138 | out: | |
139 | return err; | |
140 | } | |
bff672e6 | 141 | /*:*/ |
d7e28ffe | 142 | |
2e04ef76 | 143 | /* Cleaning up the mapping when the module is unloaded is almost... too easy. */ |
d7e28ffe RR |
144 | static void unmap_switcher(void) |
145 | { | |
146 | unsigned int i; | |
147 | ||
bff672e6 | 148 | /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */ |
d7e28ffe | 149 | vunmap(switcher_vma->addr); |
bff672e6 | 150 | /* Now we just need to free the pages we copied the switcher into */ |
d7e28ffe RR |
151 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) |
152 | __free_pages(switcher_page[i], 0); | |
0a707210 | 153 | kfree(switcher_page); |
d7e28ffe RR |
154 | } |
155 | ||
e1e72965 | 156 | /*H:032 |
dde79789 RR |
157 | * Dealing With Guest Memory. |
158 | * | |
e1e72965 RR |
159 | * Before we go too much further into the Host, we need to grok the routines |
160 | * we use to deal with Guest memory. | |
161 | * | |
dde79789 | 162 | * When the Guest gives us (what it thinks is) a physical address, we can use |
3c6b5bfa RR |
163 | * the normal copy_from_user() & copy_to_user() on the corresponding place in |
164 | * the memory region allocated by the Launcher. | |
dde79789 RR |
165 | * |
166 | * But we can't trust the Guest: it might be trying to access the Launcher | |
167 | * code. We have to check that the range is below the pfn_limit the Launcher | |
168 | * gave us. We have to make sure that addr + len doesn't give us a false | |
2e04ef76 RR |
169 | * positive by overflowing, too. |
170 | */ | |
df1693ab MZ |
171 | bool lguest_address_ok(const struct lguest *lg, |
172 | unsigned long addr, unsigned long len) | |
d7e28ffe RR |
173 | { |
174 | return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr); | |
175 | } | |
176 | ||
2e04ef76 RR |
177 | /* |
178 | * This routine copies memory from the Guest. Here we can see how useful the | |
2d37f94a | 179 | * kill_lguest() routine we met in the Launcher can be: we return a random |
2e04ef76 RR |
180 | * value (all zeroes) instead of needing to return an error. |
181 | */ | |
382ac6b3 | 182 | void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes) |
d7e28ffe | 183 | { |
382ac6b3 GOC |
184 | if (!lguest_address_ok(cpu->lg, addr, bytes) |
185 | || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) { | |
d7e28ffe RR |
186 | /* copy_from_user should do this, but as we rely on it... */ |
187 | memset(b, 0, bytes); | |
382ac6b3 | 188 | kill_guest(cpu, "bad read address %#lx len %u", addr, bytes); |
d7e28ffe RR |
189 | } |
190 | } | |
191 | ||
a6bd8e13 | 192 | /* This is the write (copy into Guest) version. */ |
382ac6b3 | 193 | void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b, |
2d37f94a | 194 | unsigned bytes) |
d7e28ffe | 195 | { |
382ac6b3 GOC |
196 | if (!lguest_address_ok(cpu->lg, addr, bytes) |
197 | || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0) | |
198 | kill_guest(cpu, "bad write address %#lx len %u", addr, bytes); | |
d7e28ffe | 199 | } |
2d37f94a | 200 | /*:*/ |
d7e28ffe | 201 | |
2e04ef76 RR |
202 | /*H:030 |
203 | * Let's jump straight to the the main loop which runs the Guest. | |
bff672e6 | 204 | * Remember, this is called by the Launcher reading /dev/lguest, and we keep |
2e04ef76 RR |
205 | * going around and around until something interesting happens. |
206 | */ | |
d0953d42 | 207 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user) |
d7e28ffe | 208 | { |
bff672e6 | 209 | /* We stop running once the Guest is dead. */ |
382ac6b3 | 210 | while (!cpu->lg->dead) { |
abd41f03 | 211 | unsigned int irq; |
a32a8813 | 212 | bool more; |
abd41f03 | 213 | |
cc6d4fbc | 214 | /* First we run any hypercalls the Guest wants done. */ |
73044f05 GOC |
215 | if (cpu->hcall) |
216 | do_hypercalls(cpu); | |
cc6d4fbc | 217 | |
2e04ef76 RR |
218 | /* |
219 | * It's possible the Guest did a NOTIFY hypercall to the | |
a91d74a3 | 220 | * Launcher. |
2e04ef76 | 221 | */ |
5e232f4f | 222 | if (cpu->pending_notify) { |
a91d74a3 RR |
223 | /* |
224 | * Does it just needs to write to a registered | |
225 | * eventfd (ie. the appropriate virtqueue thread)? | |
226 | */ | |
df60aeef | 227 | if (!send_notify_to_eventfd(cpu)) { |
a91d74a3 | 228 | /* OK, we tell the main Laucher. */ |
df60aeef RR |
229 | if (put_user(cpu->pending_notify, user)) |
230 | return -EFAULT; | |
231 | return sizeof(cpu->pending_notify); | |
232 | } | |
d7e28ffe RR |
233 | } |
234 | ||
bff672e6 | 235 | /* Check for signals */ |
d7e28ffe RR |
236 | if (signal_pending(current)) |
237 | return -ERESTARTSYS; | |
238 | ||
2e04ef76 RR |
239 | /* |
240 | * Check if there are any interrupts which can be delivered now: | |
a6bd8e13 | 241 | * if so, this sets up the hander to be executed when we next |
2e04ef76 RR |
242 | * run the Guest. |
243 | */ | |
a32a8813 | 244 | irq = interrupt_pending(cpu, &more); |
abd41f03 | 245 | if (irq < LGUEST_IRQS) |
a32a8813 | 246 | try_deliver_interrupt(cpu, irq, more); |
d7e28ffe | 247 | |
2e04ef76 RR |
248 | /* |
249 | * All long-lived kernel loops need to check with this horrible | |
bff672e6 | 250 | * thing called the freezer. If the Host is trying to suspend, |
2e04ef76 RR |
251 | * it stops us. |
252 | */ | |
d7e28ffe RR |
253 | try_to_freeze(); |
254 | ||
2e04ef76 RR |
255 | /* |
256 | * Just make absolutely sure the Guest is still alive. One of | |
257 | * those hypercalls could have been fatal, for example. | |
258 | */ | |
382ac6b3 | 259 | if (cpu->lg->dead) |
d7e28ffe RR |
260 | break; |
261 | ||
2e04ef76 RR |
262 | /* |
263 | * If the Guest asked to be stopped, we sleep. The Guest's | |
264 | * clock timer will wake us. | |
265 | */ | |
66686c2a | 266 | if (cpu->halted) { |
d7e28ffe | 267 | set_current_state(TASK_INTERRUPTIBLE); |
2e04ef76 RR |
268 | /* |
269 | * Just before we sleep, make sure no interrupt snuck in | |
270 | * which we should be doing. | |
271 | */ | |
5dac051b | 272 | if (interrupt_pending(cpu, &more) < LGUEST_IRQS) |
abd41f03 RR |
273 | set_current_state(TASK_RUNNING); |
274 | else | |
275 | schedule(); | |
d7e28ffe RR |
276 | continue; |
277 | } | |
278 | ||
2e04ef76 RR |
279 | /* |
280 | * OK, now we're ready to jump into the Guest. First we put up | |
281 | * the "Do Not Disturb" sign: | |
282 | */ | |
d7e28ffe RR |
283 | local_irq_disable(); |
284 | ||
625efab1 | 285 | /* Actually run the Guest until something happens. */ |
d0953d42 | 286 | lguest_arch_run_guest(cpu); |
bff672e6 RR |
287 | |
288 | /* Now we're ready to be interrupted or moved to other CPUs */ | |
d7e28ffe RR |
289 | local_irq_enable(); |
290 | ||
625efab1 | 291 | /* Now we deal with whatever happened to the Guest. */ |
73044f05 | 292 | lguest_arch_handle_trap(cpu); |
d7e28ffe | 293 | } |
625efab1 | 294 | |
a6bd8e13 | 295 | /* Special case: Guest is 'dead' but wants a reboot. */ |
382ac6b3 | 296 | if (cpu->lg->dead == ERR_PTR(-ERESTART)) |
ec04b13f | 297 | return -ERESTART; |
a6bd8e13 | 298 | |
bff672e6 | 299 | /* The Guest is dead => "No such file or directory" */ |
d7e28ffe RR |
300 | return -ENOENT; |
301 | } | |
302 | ||
bff672e6 RR |
303 | /*H:000 |
304 | * Welcome to the Host! | |
305 | * | |
306 | * By this point your brain has been tickled by the Guest code and numbed by | |
307 | * the Launcher code; prepare for it to be stretched by the Host code. This is | |
308 | * the heart. Let's begin at the initialization routine for the Host's lg | |
309 | * module. | |
310 | */ | |
d7e28ffe RR |
311 | static int __init init(void) |
312 | { | |
313 | int err; | |
314 | ||
bff672e6 | 315 | /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */ |
d7e28ffe | 316 | if (paravirt_enabled()) { |
5c55841d | 317 | printk("lguest is afraid of being a guest\n"); |
d7e28ffe RR |
318 | return -EPERM; |
319 | } | |
320 | ||
bff672e6 | 321 | /* First we put the Switcher up in very high virtual memory. */ |
d7e28ffe RR |
322 | err = map_switcher(); |
323 | if (err) | |
c18acd73 | 324 | goto out; |
d7e28ffe | 325 | |
bff672e6 | 326 | /* Now we set up the pagetable implementation for the Guests. */ |
d7e28ffe | 327 | err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES); |
c18acd73 RR |
328 | if (err) |
329 | goto unmap; | |
bff672e6 | 330 | |
c18acd73 RR |
331 | /* We might need to reserve an interrupt vector. */ |
332 | err = init_interrupts(); | |
333 | if (err) | |
334 | goto free_pgtables; | |
335 | ||
bff672e6 | 336 | /* /dev/lguest needs to be registered. */ |
d7e28ffe | 337 | err = lguest_device_init(); |
c18acd73 RR |
338 | if (err) |
339 | goto free_interrupts; | |
bff672e6 | 340 | |
625efab1 JS |
341 | /* Finally we do some architecture-specific setup. */ |
342 | lguest_arch_host_init(); | |
bff672e6 RR |
343 | |
344 | /* All good! */ | |
d7e28ffe | 345 | return 0; |
c18acd73 RR |
346 | |
347 | free_interrupts: | |
348 | free_interrupts(); | |
349 | free_pgtables: | |
350 | free_pagetables(); | |
351 | unmap: | |
352 | unmap_switcher(); | |
353 | out: | |
354 | return err; | |
d7e28ffe RR |
355 | } |
356 | ||
bff672e6 | 357 | /* Cleaning up is just the same code, backwards. With a little French. */ |
d7e28ffe RR |
358 | static void __exit fini(void) |
359 | { | |
360 | lguest_device_remove(); | |
c18acd73 | 361 | free_interrupts(); |
d7e28ffe RR |
362 | free_pagetables(); |
363 | unmap_switcher(); | |
bff672e6 | 364 | |
625efab1 | 365 | lguest_arch_host_fini(); |
d7e28ffe | 366 | } |
625efab1 | 367 | /*:*/ |
d7e28ffe | 368 | |
2e04ef76 RR |
369 | /* |
370 | * The Host side of lguest can be a module. This is a nice way for people to | |
371 | * play with it. | |
372 | */ | |
d7e28ffe RR |
373 | module_init(init); |
374 | module_exit(fini); | |
375 | MODULE_LICENSE("GPL"); | |
376 | MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>"); |