]>
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 | ||
406a590b | 23 | unsigned long switcher_addr; |
f1f394b1 | 24 | struct page **lg_switcher_pages; |
d7e28ffe | 25 | static struct vm_struct *switcher_vma; |
d7e28ffe | 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 | ||
93a2cdff RR |
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); | |
59 | return -EINVAL; | |
60 | } | |
61 | ||
2e04ef76 RR |
62 | /* |
63 | * We allocate an array of struct page pointers. map_vm_area() wants | |
64 | * this, rather than just an array of pages. | |
65 | */ | |
f1f394b1 RR |
66 | lg_switcher_pages = kmalloc(sizeof(lg_switcher_pages[0]) |
67 | * TOTAL_SWITCHER_PAGES, | |
68 | GFP_KERNEL); | |
69 | if (!lg_switcher_pages) { | |
d7e28ffe RR |
70 | err = -ENOMEM; |
71 | goto out; | |
72 | } | |
73 | ||
2e04ef76 RR |
74 | /* |
75 | * Now we actually allocate the pages. The Guest will see these pages, | |
76 | * so we make sure they're zeroed. | |
77 | */ | |
d7e28ffe | 78 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) { |
f1f394b1 RR |
79 | lg_switcher_pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
80 | if (!lg_switcher_pages[i]) { | |
d7e28ffe RR |
81 | err = -ENOMEM; |
82 | goto free_some_pages; | |
83 | } | |
d7e28ffe RR |
84 | } |
85 | ||
406a590b RR |
86 | switcher_addr = SWITCHER_ADDR; |
87 | ||
2e04ef76 RR |
88 | /* |
89 | * First we check that the Switcher won't overlap the fixmap area at | |
f14ae652 | 90 | * the top of memory. It's currently nowhere near, but it could have |
2e04ef76 RR |
91 | * very strange effects if it ever happened. |
92 | */ | |
406a590b | 93 | if (switcher_addr + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){ |
f14ae652 RR |
94 | err = -ENOMEM; |
95 | printk("lguest: mapping switcher would thwack fixmap\n"); | |
96 | goto free_pages; | |
97 | } | |
98 | ||
2e04ef76 | 99 | /* |
406a590b RR |
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. | |
2e04ef76 | 104 | */ |
d7e28ffe | 105 | switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE, |
406a590b | 106 | VM_ALLOC, switcher_addr, switcher_addr |
f14ae652 | 107 | + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE); |
d7e28ffe RR |
108 | if (!switcher_vma) { |
109 | err = -ENOMEM; | |
110 | printk("lguest: could not map switcher pages high\n"); | |
111 | goto free_pages; | |
112 | } | |
113 | ||
2e04ef76 RR |
114 | /* |
115 | * This code actually sets up the pages we've allocated to appear at | |
406a590b | 116 | * switcher_addr. map_vm_area() takes the vma we allocated above, the |
bff672e6 RR |
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 | |
2e04ef76 RR |
119 | * care. |
120 | */ | |
f1f394b1 | 121 | pagep = lg_switcher_pages; |
ed1dc778 | 122 | err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep); |
d7e28ffe RR |
123 | if (err) { |
124 | printk("lguest: map_vm_area failed: %i\n", err); | |
125 | goto free_vma; | |
126 | } | |
bff672e6 | 127 | |
2e04ef76 RR |
128 | /* |
129 | * Now the Switcher is mapped at the right address, we can't fail! | |
9f54288d | 130 | * Copy in the compiled-in Switcher code (from x86/switcher_32.S). |
2e04ef76 | 131 | */ |
d7e28ffe RR |
132 | memcpy(switcher_vma->addr, start_switcher_text, |
133 | end_switcher_text - start_switcher_text); | |
134 | ||
d7e28ffe RR |
135 | printk(KERN_INFO "lguest: mapped switcher at %p\n", |
136 | switcher_vma->addr); | |
bff672e6 | 137 | /* And we succeeded... */ |
d7e28ffe RR |
138 | return 0; |
139 | ||
140 | free_vma: | |
141 | vunmap(switcher_vma->addr); | |
142 | free_pages: | |
143 | i = TOTAL_SWITCHER_PAGES; | |
144 | free_some_pages: | |
145 | for (--i; i >= 0; i--) | |
f1f394b1 RR |
146 | __free_pages(lg_switcher_pages[i], 0); |
147 | kfree(lg_switcher_pages); | |
d7e28ffe RR |
148 | out: |
149 | return err; | |
150 | } | |
bff672e6 | 151 | /*:*/ |
d7e28ffe | 152 | |
2e04ef76 | 153 | /* Cleaning up the mapping when the module is unloaded is almost... too easy. */ |
d7e28ffe RR |
154 | static void unmap_switcher(void) |
155 | { | |
156 | unsigned int i; | |
157 | ||
bff672e6 | 158 | /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */ |
d7e28ffe | 159 | vunmap(switcher_vma->addr); |
bff672e6 | 160 | /* Now we just need to free the pages we copied the switcher into */ |
d7e28ffe | 161 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) |
f1f394b1 RR |
162 | __free_pages(lg_switcher_pages[i], 0); |
163 | kfree(lg_switcher_pages); | |
d7e28ffe RR |
164 | } |
165 | ||
e1e72965 | 166 | /*H:032 |
dde79789 RR |
167 | * Dealing With Guest Memory. |
168 | * | |
e1e72965 RR |
169 | * Before we go too much further into the Host, we need to grok the routines |
170 | * we use to deal with Guest memory. | |
171 | * | |
dde79789 | 172 | * When the Guest gives us (what it thinks is) a physical address, we can use |
3c6b5bfa RR |
173 | * the normal copy_from_user() & copy_to_user() on the corresponding place in |
174 | * the memory region allocated by the Launcher. | |
dde79789 RR |
175 | * |
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 | |
2e04ef76 RR |
179 | * positive by overflowing, too. |
180 | */ | |
df1693ab MZ |
181 | bool lguest_address_ok(const struct lguest *lg, |
182 | unsigned long addr, unsigned long len) | |
d7e28ffe RR |
183 | { |
184 | return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr); | |
185 | } | |
186 | ||
2e04ef76 RR |
187 | /* |
188 | * This routine copies memory from the Guest. Here we can see how useful the | |
2d37f94a | 189 | * kill_lguest() routine we met in the Launcher can be: we return a random |
2e04ef76 RR |
190 | * value (all zeroes) instead of needing to return an error. |
191 | */ | |
382ac6b3 | 192 | void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes) |
d7e28ffe | 193 | { |
382ac6b3 GOC |
194 | if (!lguest_address_ok(cpu->lg, addr, bytes) |
195 | || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) { | |
d7e28ffe RR |
196 | /* copy_from_user should do this, but as we rely on it... */ |
197 | memset(b, 0, bytes); | |
382ac6b3 | 198 | kill_guest(cpu, "bad read address %#lx len %u", addr, bytes); |
d7e28ffe RR |
199 | } |
200 | } | |
201 | ||
a6bd8e13 | 202 | /* This is the write (copy into Guest) version. */ |
382ac6b3 | 203 | void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b, |
2d37f94a | 204 | unsigned bytes) |
d7e28ffe | 205 | { |
382ac6b3 GOC |
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); | |
d7e28ffe | 209 | } |
2d37f94a | 210 | /*:*/ |
d7e28ffe | 211 | |
2e04ef76 RR |
212 | /*H:030 |
213 | * Let's jump straight to the the main loop which runs the Guest. | |
bff672e6 | 214 | * Remember, this is called by the Launcher reading /dev/lguest, and we keep |
2e04ef76 RR |
215 | * going around and around until something interesting happens. |
216 | */ | |
d0953d42 | 217 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user) |
d7e28ffe | 218 | { |
bff672e6 | 219 | /* We stop running once the Guest is dead. */ |
382ac6b3 | 220 | while (!cpu->lg->dead) { |
abd41f03 | 221 | unsigned int irq; |
a32a8813 | 222 | bool more; |
abd41f03 | 223 | |
cc6d4fbc | 224 | /* First we run any hypercalls the Guest wants done. */ |
73044f05 GOC |
225 | if (cpu->hcall) |
226 | do_hypercalls(cpu); | |
cc6d4fbc | 227 | |
2e04ef76 RR |
228 | /* |
229 | * It's possible the Guest did a NOTIFY hypercall to the | |
a91d74a3 | 230 | * Launcher. |
2e04ef76 | 231 | */ |
5e232f4f | 232 | if (cpu->pending_notify) { |
a91d74a3 RR |
233 | /* |
234 | * Does it just needs to write to a registered | |
235 | * eventfd (ie. the appropriate virtqueue thread)? | |
236 | */ | |
df60aeef | 237 | if (!send_notify_to_eventfd(cpu)) { |
681f2066 | 238 | /* OK, we tell the main Launcher. */ |
df60aeef RR |
239 | if (put_user(cpu->pending_notify, user)) |
240 | return -EFAULT; | |
241 | return sizeof(cpu->pending_notify); | |
242 | } | |
d7e28ffe RR |
243 | } |
244 | ||
0acf0001 MH |
245 | /* |
246 | * All long-lived kernel loops need to check with this horrible | |
247 | * thing called the freezer. If the Host is trying to suspend, | |
248 | * it stops us. | |
249 | */ | |
250 | try_to_freeze(); | |
251 | ||
bff672e6 | 252 | /* Check for signals */ |
d7e28ffe RR |
253 | if (signal_pending(current)) |
254 | return -ERESTARTSYS; | |
255 | ||
2e04ef76 RR |
256 | /* |
257 | * Check if there are any interrupts which can be delivered now: | |
a6bd8e13 | 258 | * if so, this sets up the hander to be executed when we next |
2e04ef76 RR |
259 | * run the Guest. |
260 | */ | |
a32a8813 | 261 | irq = interrupt_pending(cpu, &more); |
abd41f03 | 262 | if (irq < LGUEST_IRQS) |
a32a8813 | 263 | try_deliver_interrupt(cpu, irq, more); |
d7e28ffe | 264 | |
2e04ef76 RR |
265 | /* |
266 | * Just make absolutely sure the Guest is still alive. One of | |
267 | * those hypercalls could have been fatal, for example. | |
268 | */ | |
382ac6b3 | 269 | if (cpu->lg->dead) |
d7e28ffe RR |
270 | break; |
271 | ||
2e04ef76 RR |
272 | /* |
273 | * If the Guest asked to be stopped, we sleep. The Guest's | |
274 | * clock timer will wake us. | |
275 | */ | |
66686c2a | 276 | if (cpu->halted) { |
d7e28ffe | 277 | set_current_state(TASK_INTERRUPTIBLE); |
2e04ef76 RR |
278 | /* |
279 | * Just before we sleep, make sure no interrupt snuck in | |
280 | * which we should be doing. | |
281 | */ | |
5dac051b | 282 | if (interrupt_pending(cpu, &more) < LGUEST_IRQS) |
abd41f03 RR |
283 | set_current_state(TASK_RUNNING); |
284 | else | |
285 | schedule(); | |
d7e28ffe RR |
286 | continue; |
287 | } | |
288 | ||
2e04ef76 RR |
289 | /* |
290 | * OK, now we're ready to jump into the Guest. First we put up | |
291 | * the "Do Not Disturb" sign: | |
292 | */ | |
d7e28ffe RR |
293 | local_irq_disable(); |
294 | ||
625efab1 | 295 | /* Actually run the Guest until something happens. */ |
d0953d42 | 296 | lguest_arch_run_guest(cpu); |
bff672e6 RR |
297 | |
298 | /* Now we're ready to be interrupted or moved to other CPUs */ | |
d7e28ffe RR |
299 | local_irq_enable(); |
300 | ||
625efab1 | 301 | /* Now we deal with whatever happened to the Guest. */ |
73044f05 | 302 | lguest_arch_handle_trap(cpu); |
d7e28ffe | 303 | } |
625efab1 | 304 | |
a6bd8e13 | 305 | /* Special case: Guest is 'dead' but wants a reboot. */ |
382ac6b3 | 306 | if (cpu->lg->dead == ERR_PTR(-ERESTART)) |
ec04b13f | 307 | return -ERESTART; |
a6bd8e13 | 308 | |
bff672e6 | 309 | /* The Guest is dead => "No such file or directory" */ |
d7e28ffe RR |
310 | return -ENOENT; |
311 | } | |
312 | ||
bff672e6 RR |
313 | /*H:000 |
314 | * Welcome to the Host! | |
315 | * | |
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 | |
319 | * module. | |
320 | */ | |
d7e28ffe RR |
321 | static int __init init(void) |
322 | { | |
323 | int err; | |
324 | ||
bff672e6 | 325 | /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */ |
b56e3215 | 326 | if (get_kernel_rpl() != 0) { |
5c55841d | 327 | printk("lguest is afraid of being a guest\n"); |
d7e28ffe RR |
328 | return -EPERM; |
329 | } | |
330 | ||
bff672e6 | 331 | /* First we put the Switcher up in very high virtual memory. */ |
d7e28ffe RR |
332 | err = map_switcher(); |
333 | if (err) | |
c18acd73 | 334 | goto out; |
d7e28ffe | 335 | |
c18acd73 RR |
336 | /* We might need to reserve an interrupt vector. */ |
337 | err = init_interrupts(); | |
338 | if (err) | |
3412b6ae | 339 | goto unmap; |
c18acd73 | 340 | |
bff672e6 | 341 | /* /dev/lguest needs to be registered. */ |
d7e28ffe | 342 | err = lguest_device_init(); |
c18acd73 RR |
343 | if (err) |
344 | goto free_interrupts; | |
bff672e6 | 345 | |
625efab1 JS |
346 | /* Finally we do some architecture-specific setup. */ |
347 | lguest_arch_host_init(); | |
bff672e6 RR |
348 | |
349 | /* All good! */ | |
d7e28ffe | 350 | return 0; |
c18acd73 RR |
351 | |
352 | free_interrupts: | |
353 | free_interrupts(); | |
c18acd73 RR |
354 | unmap: |
355 | unmap_switcher(); | |
356 | out: | |
357 | return err; | |
d7e28ffe RR |
358 | } |
359 | ||
bff672e6 | 360 | /* Cleaning up is just the same code, backwards. With a little French. */ |
d7e28ffe RR |
361 | static void __exit fini(void) |
362 | { | |
363 | lguest_device_remove(); | |
c18acd73 | 364 | free_interrupts(); |
d7e28ffe | 365 | unmap_switcher(); |
bff672e6 | 366 | |
625efab1 | 367 | lguest_arch_host_fini(); |
d7e28ffe | 368 | } |
625efab1 | 369 | /*:*/ |
d7e28ffe | 370 | |
2e04ef76 RR |
371 | /* |
372 | * The Host side of lguest can be a module. This is a nice way for people to | |
373 | * play with it. | |
374 | */ | |
d7e28ffe RR |
375 | module_init(init); |
376 | module_exit(fini); | |
377 | MODULE_LICENSE("GPL"); | |
378 | MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>"); |