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> |
d7e28ffe | 15 | #include <asm/paravirt.h> |
d7e28ffe RR |
16 | #include <asm/pgtable.h> |
17 | #include <asm/uaccess.h> | |
18 | #include <asm/poll.h> | |
d7e28ffe | 19 | #include <asm/asm-offsets.h> |
d7e28ffe RR |
20 | #include "lg.h" |
21 | ||
d7e28ffe RR |
22 | |
23 | static struct vm_struct *switcher_vma; | |
24 | static struct page **switcher_page; | |
25 | ||
d7e28ffe RR |
26 | /* This One Big lock protects all inter-guest data structures. */ |
27 | DEFINE_MUTEX(lguest_lock); | |
d7e28ffe | 28 | |
2e04ef76 RR |
29 | /*H:010 |
30 | * We need to set up the Switcher at a high virtual address. Remember the | |
bff672e6 RR |
31 | * Switcher is a few hundred bytes of assembler code which actually changes the |
32 | * CPU to run the Guest, and then changes back to the Host when a trap or | |
33 | * interrupt happens. | |
34 | * | |
35 | * The Switcher code must be at the same virtual address in the Guest as the | |
36 | * Host since it will be running as the switchover occurs. | |
37 | * | |
38 | * Trying to map memory at a particular address is an unusual thing to do, so | |
2e04ef76 RR |
39 | * it's not a simple one-liner. |
40 | */ | |
d7e28ffe RR |
41 | static __init int map_switcher(void) |
42 | { | |
43 | int i, err; | |
44 | struct page **pagep; | |
45 | ||
bff672e6 RR |
46 | /* |
47 | * Map the Switcher in to high memory. | |
48 | * | |
49 | * It turns out that if we choose the address 0xFFC00000 (4MB under the | |
50 | * top virtual address), it makes setting up the page tables really | |
51 | * easy. | |
52 | */ | |
53 | ||
2e04ef76 RR |
54 | /* |
55 | * We allocate an array of struct page pointers. map_vm_area() wants | |
56 | * this, rather than just an array of pages. | |
57 | */ | |
d7e28ffe RR |
58 | switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES, |
59 | GFP_KERNEL); | |
60 | if (!switcher_page) { | |
61 | err = -ENOMEM; | |
62 | goto out; | |
63 | } | |
64 | ||
2e04ef76 RR |
65 | /* |
66 | * Now we actually allocate the pages. The Guest will see these pages, | |
67 | * so we make sure they're zeroed. | |
68 | */ | |
d7e28ffe | 69 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) { |
6c189d83 XG |
70 | switcher_page[i] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
71 | if (!switcher_page[i]) { | |
d7e28ffe RR |
72 | err = -ENOMEM; |
73 | goto free_some_pages; | |
74 | } | |
d7e28ffe RR |
75 | } |
76 | ||
2e04ef76 RR |
77 | /* |
78 | * First we check that the Switcher won't overlap the fixmap area at | |
f14ae652 | 79 | * the top of memory. It's currently nowhere near, but it could have |
2e04ef76 RR |
80 | * very strange effects if it ever happened. |
81 | */ | |
f14ae652 RR |
82 | if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){ |
83 | err = -ENOMEM; | |
84 | printk("lguest: mapping switcher would thwack fixmap\n"); | |
85 | goto free_pages; | |
86 | } | |
87 | ||
2e04ef76 RR |
88 | /* |
89 | * Now we reserve the "virtual memory area" we want: 0xFFC00000 | |
bff672e6 | 90 | * (SWITCHER_ADDR). We might not get it in theory, but in practice |
f14ae652 | 91 | * it's worked so far. The end address needs +1 because __get_vm_area |
2e04ef76 RR |
92 | * allocates an extra guard page, so we need space for that. |
93 | */ | |
d7e28ffe | 94 | switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE, |
f14ae652 RR |
95 | VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR |
96 | + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE); | |
d7e28ffe RR |
97 | if (!switcher_vma) { |
98 | err = -ENOMEM; | |
99 | printk("lguest: could not map switcher pages high\n"); | |
100 | goto free_pages; | |
101 | } | |
102 | ||
2e04ef76 RR |
103 | /* |
104 | * This code actually sets up the pages we've allocated to appear at | |
bff672e6 RR |
105 | * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the |
106 | * kind of pages we're mapping (kernel pages), and a pointer to our | |
107 | * array of struct pages. It increments that pointer, but we don't | |
2e04ef76 RR |
108 | * care. |
109 | */ | |
d7e28ffe | 110 | pagep = switcher_page; |
ed1dc778 | 111 | err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep); |
d7e28ffe RR |
112 | if (err) { |
113 | printk("lguest: map_vm_area failed: %i\n", err); | |
114 | goto free_vma; | |
115 | } | |
bff672e6 | 116 | |
2e04ef76 RR |
117 | /* |
118 | * Now the Switcher is mapped at the right address, we can't fail! | |
119 | * Copy in the compiled-in Switcher code (from <arch>_switcher.S). | |
120 | */ | |
d7e28ffe RR |
121 | memcpy(switcher_vma->addr, start_switcher_text, |
122 | end_switcher_text - start_switcher_text); | |
123 | ||
d7e28ffe RR |
124 | printk(KERN_INFO "lguest: mapped switcher at %p\n", |
125 | switcher_vma->addr); | |
bff672e6 | 126 | /* And we succeeded... */ |
d7e28ffe RR |
127 | return 0; |
128 | ||
129 | free_vma: | |
130 | vunmap(switcher_vma->addr); | |
131 | free_pages: | |
132 | i = TOTAL_SWITCHER_PAGES; | |
133 | free_some_pages: | |
134 | for (--i; i >= 0; i--) | |
135 | __free_pages(switcher_page[i], 0); | |
136 | kfree(switcher_page); | |
137 | out: | |
138 | return err; | |
139 | } | |
bff672e6 | 140 | /*:*/ |
d7e28ffe | 141 | |
2e04ef76 | 142 | /* Cleaning up the mapping when the module is unloaded is almost... too easy. */ |
d7e28ffe RR |
143 | static void unmap_switcher(void) |
144 | { | |
145 | unsigned int i; | |
146 | ||
bff672e6 | 147 | /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */ |
d7e28ffe | 148 | vunmap(switcher_vma->addr); |
bff672e6 | 149 | /* Now we just need to free the pages we copied the switcher into */ |
d7e28ffe RR |
150 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) |
151 | __free_pages(switcher_page[i], 0); | |
0a707210 | 152 | kfree(switcher_page); |
d7e28ffe RR |
153 | } |
154 | ||
e1e72965 | 155 | /*H:032 |
dde79789 RR |
156 | * Dealing With Guest Memory. |
157 | * | |
e1e72965 RR |
158 | * Before we go too much further into the Host, we need to grok the routines |
159 | * we use to deal with Guest memory. | |
160 | * | |
dde79789 | 161 | * When the Guest gives us (what it thinks is) a physical address, we can use |
3c6b5bfa RR |
162 | * the normal copy_from_user() & copy_to_user() on the corresponding place in |
163 | * the memory region allocated by the Launcher. | |
dde79789 RR |
164 | * |
165 | * But we can't trust the Guest: it might be trying to access the Launcher | |
166 | * code. We have to check that the range is below the pfn_limit the Launcher | |
167 | * gave us. We have to make sure that addr + len doesn't give us a false | |
2e04ef76 RR |
168 | * positive by overflowing, too. |
169 | */ | |
df1693ab MZ |
170 | bool lguest_address_ok(const struct lguest *lg, |
171 | unsigned long addr, unsigned long len) | |
d7e28ffe RR |
172 | { |
173 | return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr); | |
174 | } | |
175 | ||
2e04ef76 RR |
176 | /* |
177 | * This routine copies memory from the Guest. Here we can see how useful the | |
2d37f94a | 178 | * kill_lguest() routine we met in the Launcher can be: we return a random |
2e04ef76 RR |
179 | * value (all zeroes) instead of needing to return an error. |
180 | */ | |
382ac6b3 | 181 | void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes) |
d7e28ffe | 182 | { |
382ac6b3 GOC |
183 | if (!lguest_address_ok(cpu->lg, addr, bytes) |
184 | || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) { | |
d7e28ffe RR |
185 | /* copy_from_user should do this, but as we rely on it... */ |
186 | memset(b, 0, bytes); | |
382ac6b3 | 187 | kill_guest(cpu, "bad read address %#lx len %u", addr, bytes); |
d7e28ffe RR |
188 | } |
189 | } | |
190 | ||
a6bd8e13 | 191 | /* This is the write (copy into Guest) version. */ |
382ac6b3 | 192 | void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b, |
2d37f94a | 193 | unsigned bytes) |
d7e28ffe | 194 | { |
382ac6b3 GOC |
195 | if (!lguest_address_ok(cpu->lg, addr, bytes) |
196 | || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0) | |
197 | kill_guest(cpu, "bad write address %#lx len %u", addr, bytes); | |
d7e28ffe | 198 | } |
2d37f94a | 199 | /*:*/ |
d7e28ffe | 200 | |
2e04ef76 RR |
201 | /*H:030 |
202 | * Let's jump straight to the the main loop which runs the Guest. | |
bff672e6 | 203 | * Remember, this is called by the Launcher reading /dev/lguest, and we keep |
2e04ef76 RR |
204 | * going around and around until something interesting happens. |
205 | */ | |
d0953d42 | 206 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user) |
d7e28ffe | 207 | { |
bff672e6 | 208 | /* We stop running once the Guest is dead. */ |
382ac6b3 | 209 | while (!cpu->lg->dead) { |
abd41f03 | 210 | unsigned int irq; |
a32a8813 | 211 | bool more; |
abd41f03 | 212 | |
cc6d4fbc | 213 | /* First we run any hypercalls the Guest wants done. */ |
73044f05 GOC |
214 | if (cpu->hcall) |
215 | do_hypercalls(cpu); | |
cc6d4fbc | 216 | |
2e04ef76 RR |
217 | /* |
218 | * It's possible the Guest did a NOTIFY hypercall to the | |
a91d74a3 | 219 | * Launcher. |
2e04ef76 | 220 | */ |
5e232f4f | 221 | if (cpu->pending_notify) { |
a91d74a3 RR |
222 | /* |
223 | * Does it just needs to write to a registered | |
224 | * eventfd (ie. the appropriate virtqueue thread)? | |
225 | */ | |
df60aeef | 226 | if (!send_notify_to_eventfd(cpu)) { |
a91d74a3 | 227 | /* OK, we tell the main Laucher. */ |
df60aeef RR |
228 | if (put_user(cpu->pending_notify, user)) |
229 | return -EFAULT; | |
230 | return sizeof(cpu->pending_notify); | |
231 | } | |
d7e28ffe RR |
232 | } |
233 | ||
bff672e6 | 234 | /* Check for signals */ |
d7e28ffe RR |
235 | if (signal_pending(current)) |
236 | return -ERESTARTSYS; | |
237 | ||
2e04ef76 RR |
238 | /* |
239 | * Check if there are any interrupts which can be delivered now: | |
a6bd8e13 | 240 | * if so, this sets up the hander to be executed when we next |
2e04ef76 RR |
241 | * run the Guest. |
242 | */ | |
a32a8813 | 243 | irq = interrupt_pending(cpu, &more); |
abd41f03 | 244 | if (irq < LGUEST_IRQS) |
a32a8813 | 245 | try_deliver_interrupt(cpu, irq, more); |
d7e28ffe | 246 | |
2e04ef76 RR |
247 | /* |
248 | * All long-lived kernel loops need to check with this horrible | |
bff672e6 | 249 | * thing called the freezer. If the Host is trying to suspend, |
2e04ef76 RR |
250 | * it stops us. |
251 | */ | |
d7e28ffe RR |
252 | try_to_freeze(); |
253 | ||
2e04ef76 RR |
254 | /* |
255 | * Just make absolutely sure the Guest is still alive. One of | |
256 | * those hypercalls could have been fatal, for example. | |
257 | */ | |
382ac6b3 | 258 | if (cpu->lg->dead) |
d7e28ffe RR |
259 | break; |
260 | ||
2e04ef76 RR |
261 | /* |
262 | * If the Guest asked to be stopped, we sleep. The Guest's | |
263 | * clock timer will wake us. | |
264 | */ | |
66686c2a | 265 | if (cpu->halted) { |
d7e28ffe | 266 | set_current_state(TASK_INTERRUPTIBLE); |
2e04ef76 RR |
267 | /* |
268 | * Just before we sleep, make sure no interrupt snuck in | |
269 | * which we should be doing. | |
270 | */ | |
5dac051b | 271 | if (interrupt_pending(cpu, &more) < LGUEST_IRQS) |
abd41f03 RR |
272 | set_current_state(TASK_RUNNING); |
273 | else | |
274 | schedule(); | |
d7e28ffe RR |
275 | continue; |
276 | } | |
277 | ||
2e04ef76 RR |
278 | /* |
279 | * OK, now we're ready to jump into the Guest. First we put up | |
280 | * the "Do Not Disturb" sign: | |
281 | */ | |
d7e28ffe RR |
282 | local_irq_disable(); |
283 | ||
625efab1 | 284 | /* Actually run the Guest until something happens. */ |
d0953d42 | 285 | lguest_arch_run_guest(cpu); |
bff672e6 RR |
286 | |
287 | /* Now we're ready to be interrupted or moved to other CPUs */ | |
d7e28ffe RR |
288 | local_irq_enable(); |
289 | ||
625efab1 | 290 | /* Now we deal with whatever happened to the Guest. */ |
73044f05 | 291 | lguest_arch_handle_trap(cpu); |
d7e28ffe | 292 | } |
625efab1 | 293 | |
a6bd8e13 | 294 | /* Special case: Guest is 'dead' but wants a reboot. */ |
382ac6b3 | 295 | if (cpu->lg->dead == ERR_PTR(-ERESTART)) |
ec04b13f | 296 | return -ERESTART; |
a6bd8e13 | 297 | |
bff672e6 | 298 | /* The Guest is dead => "No such file or directory" */ |
d7e28ffe RR |
299 | return -ENOENT; |
300 | } | |
301 | ||
bff672e6 RR |
302 | /*H:000 |
303 | * Welcome to the Host! | |
304 | * | |
305 | * By this point your brain has been tickled by the Guest code and numbed by | |
306 | * the Launcher code; prepare for it to be stretched by the Host code. This is | |
307 | * the heart. Let's begin at the initialization routine for the Host's lg | |
308 | * module. | |
309 | */ | |
d7e28ffe RR |
310 | static int __init init(void) |
311 | { | |
312 | int err; | |
313 | ||
bff672e6 | 314 | /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */ |
d7e28ffe | 315 | if (paravirt_enabled()) { |
5c55841d | 316 | printk("lguest is afraid of being a guest\n"); |
d7e28ffe RR |
317 | return -EPERM; |
318 | } | |
319 | ||
bff672e6 | 320 | /* First we put the Switcher up in very high virtual memory. */ |
d7e28ffe RR |
321 | err = map_switcher(); |
322 | if (err) | |
c18acd73 | 323 | goto out; |
d7e28ffe | 324 | |
bff672e6 | 325 | /* Now we set up the pagetable implementation for the Guests. */ |
d7e28ffe | 326 | err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES); |
c18acd73 RR |
327 | if (err) |
328 | goto unmap; | |
bff672e6 | 329 | |
c18acd73 RR |
330 | /* We might need to reserve an interrupt vector. */ |
331 | err = init_interrupts(); | |
332 | if (err) | |
333 | goto free_pgtables; | |
334 | ||
bff672e6 | 335 | /* /dev/lguest needs to be registered. */ |
d7e28ffe | 336 | err = lguest_device_init(); |
c18acd73 RR |
337 | if (err) |
338 | goto free_interrupts; | |
bff672e6 | 339 | |
625efab1 JS |
340 | /* Finally we do some architecture-specific setup. */ |
341 | lguest_arch_host_init(); | |
bff672e6 RR |
342 | |
343 | /* All good! */ | |
d7e28ffe | 344 | return 0; |
c18acd73 RR |
345 | |
346 | free_interrupts: | |
347 | free_interrupts(); | |
348 | free_pgtables: | |
349 | free_pagetables(); | |
350 | unmap: | |
351 | unmap_switcher(); | |
352 | out: | |
353 | return err; | |
d7e28ffe RR |
354 | } |
355 | ||
bff672e6 | 356 | /* Cleaning up is just the same code, backwards. With a little French. */ |
d7e28ffe RR |
357 | static void __exit fini(void) |
358 | { | |
359 | lguest_device_remove(); | |
c18acd73 | 360 | free_interrupts(); |
d7e28ffe RR |
361 | free_pagetables(); |
362 | unmap_switcher(); | |
bff672e6 | 363 | |
625efab1 | 364 | lguest_arch_host_fini(); |
d7e28ffe | 365 | } |
625efab1 | 366 | /*:*/ |
d7e28ffe | 367 | |
2e04ef76 RR |
368 | /* |
369 | * The Host side of lguest can be a module. This is a nice way for people to | |
370 | * play with it. | |
371 | */ | |
d7e28ffe RR |
372 | module_init(init); |
373 | module_exit(fini); | |
374 | MODULE_LICENSE("GPL"); | |
375 | MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>"); |