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; | |
d7e28ffe | 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 | ||
93a2cdff RR |
54 | /* We assume Switcher text fits into a single page. */ |
55 | if (end_switcher_text - start_switcher_text > PAGE_SIZE) { | |
56 | printk(KERN_ERR "lguest: switcher text too large (%zu)\n", | |
57 | end_switcher_text - start_switcher_text); | |
58 | return -EINVAL; | |
59 | } | |
60 | ||
2e04ef76 RR |
61 | /* |
62 | * We allocate an array of struct page pointers. map_vm_area() wants | |
63 | * this, rather than just an array of pages. | |
64 | */ | |
f1f394b1 RR |
65 | lg_switcher_pages = kmalloc(sizeof(lg_switcher_pages[0]) |
66 | * TOTAL_SWITCHER_PAGES, | |
67 | GFP_KERNEL); | |
68 | if (!lg_switcher_pages) { | |
d7e28ffe RR |
69 | err = -ENOMEM; |
70 | goto out; | |
71 | } | |
72 | ||
2e04ef76 RR |
73 | /* |
74 | * Now we actually allocate the pages. The Guest will see these pages, | |
75 | * so we make sure they're zeroed. | |
76 | */ | |
d7e28ffe | 77 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) { |
f1f394b1 RR |
78 | lg_switcher_pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
79 | if (!lg_switcher_pages[i]) { | |
d7e28ffe RR |
80 | err = -ENOMEM; |
81 | goto free_some_pages; | |
82 | } | |
d7e28ffe RR |
83 | } |
84 | ||
2e04ef76 | 85 | /* |
6b392717 RR |
86 | * We place the Switcher underneath the fixmap area, which is the |
87 | * highest virtual address we can get. This is important, since we | |
88 | * tell the Guest it can't access this memory, so we want its ceiling | |
89 | * as high as possible. | |
2e04ef76 | 90 | */ |
6b392717 | 91 | switcher_addr = FIXADDR_START - (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE; |
f14ae652 | 92 | |
2e04ef76 | 93 | /* |
406a590b RR |
94 | * Now we reserve the "virtual memory area" we want. We might |
95 | * not get it in theory, but in practice it's worked so far. | |
96 | * The end address needs +1 because __get_vm_area allocates an | |
97 | * extra guard page, so we need space for that. | |
2e04ef76 | 98 | */ |
d7e28ffe | 99 | switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE, |
406a590b | 100 | VM_ALLOC, switcher_addr, switcher_addr |
f14ae652 | 101 | + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE); |
d7e28ffe RR |
102 | if (!switcher_vma) { |
103 | err = -ENOMEM; | |
104 | printk("lguest: could not map switcher pages high\n"); | |
105 | goto free_pages; | |
106 | } | |
107 | ||
2e04ef76 RR |
108 | /* |
109 | * This code actually sets up the pages we've allocated to appear at | |
406a590b | 110 | * switcher_addr. map_vm_area() takes the vma we allocated above, the |
bff672e6 | 111 | * kind of pages we're mapping (kernel pages), and a pointer to our |
f6f8ed47 | 112 | * array of struct pages. |
2e04ef76 | 113 | */ |
f6f8ed47 | 114 | err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, lg_switcher_pages); |
d7e28ffe RR |
115 | if (err) { |
116 | printk("lguest: map_vm_area failed: %i\n", err); | |
117 | goto free_vma; | |
118 | } | |
bff672e6 | 119 | |
2e04ef76 RR |
120 | /* |
121 | * Now the Switcher is mapped at the right address, we can't fail! | |
9f54288d | 122 | * Copy in the compiled-in Switcher code (from x86/switcher_32.S). |
2e04ef76 | 123 | */ |
d7e28ffe RR |
124 | memcpy(switcher_vma->addr, start_switcher_text, |
125 | end_switcher_text - start_switcher_text); | |
126 | ||
d7e28ffe RR |
127 | printk(KERN_INFO "lguest: mapped switcher at %p\n", |
128 | switcher_vma->addr); | |
bff672e6 | 129 | /* And we succeeded... */ |
d7e28ffe RR |
130 | return 0; |
131 | ||
132 | free_vma: | |
133 | vunmap(switcher_vma->addr); | |
134 | free_pages: | |
135 | i = TOTAL_SWITCHER_PAGES; | |
136 | free_some_pages: | |
137 | for (--i; i >= 0; i--) | |
f1f394b1 RR |
138 | __free_pages(lg_switcher_pages[i], 0); |
139 | kfree(lg_switcher_pages); | |
d7e28ffe RR |
140 | out: |
141 | return err; | |
142 | } | |
bff672e6 | 143 | /*:*/ |
d7e28ffe | 144 | |
2e04ef76 | 145 | /* Cleaning up the mapping when the module is unloaded is almost... too easy. */ |
d7e28ffe RR |
146 | static void unmap_switcher(void) |
147 | { | |
148 | unsigned int i; | |
149 | ||
bff672e6 | 150 | /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */ |
d7e28ffe | 151 | vunmap(switcher_vma->addr); |
bff672e6 | 152 | /* Now we just need to free the pages we copied the switcher into */ |
d7e28ffe | 153 | for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) |
f1f394b1 RR |
154 | __free_pages(lg_switcher_pages[i], 0); |
155 | kfree(lg_switcher_pages); | |
d7e28ffe RR |
156 | } |
157 | ||
e1e72965 | 158 | /*H:032 |
dde79789 RR |
159 | * Dealing With Guest Memory. |
160 | * | |
e1e72965 RR |
161 | * Before we go too much further into the Host, we need to grok the routines |
162 | * we use to deal with Guest memory. | |
163 | * | |
dde79789 | 164 | * When the Guest gives us (what it thinks is) a physical address, we can use |
3c6b5bfa RR |
165 | * the normal copy_from_user() & copy_to_user() on the corresponding place in |
166 | * the memory region allocated by the Launcher. | |
dde79789 RR |
167 | * |
168 | * But we can't trust the Guest: it might be trying to access the Launcher | |
169 | * code. We have to check that the range is below the pfn_limit the Launcher | |
170 | * gave us. We have to make sure that addr + len doesn't give us a false | |
2e04ef76 RR |
171 | * positive by overflowing, too. |
172 | */ | |
df1693ab MZ |
173 | bool lguest_address_ok(const struct lguest *lg, |
174 | unsigned long addr, unsigned long len) | |
d7e28ffe RR |
175 | { |
176 | return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr); | |
177 | } | |
178 | ||
2e04ef76 RR |
179 | /* |
180 | * This routine copies memory from the Guest. Here we can see how useful the | |
2d37f94a | 181 | * kill_lguest() routine we met in the Launcher can be: we return a random |
2e04ef76 RR |
182 | * value (all zeroes) instead of needing to return an error. |
183 | */ | |
382ac6b3 | 184 | void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes) |
d7e28ffe | 185 | { |
382ac6b3 GOC |
186 | if (!lguest_address_ok(cpu->lg, addr, bytes) |
187 | || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) { | |
d7e28ffe RR |
188 | /* copy_from_user should do this, but as we rely on it... */ |
189 | memset(b, 0, bytes); | |
382ac6b3 | 190 | kill_guest(cpu, "bad read address %#lx len %u", addr, bytes); |
d7e28ffe RR |
191 | } |
192 | } | |
193 | ||
a6bd8e13 | 194 | /* This is the write (copy into Guest) version. */ |
382ac6b3 | 195 | void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b, |
2d37f94a | 196 | unsigned bytes) |
d7e28ffe | 197 | { |
382ac6b3 GOC |
198 | if (!lguest_address_ok(cpu->lg, addr, bytes) |
199 | || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0) | |
200 | kill_guest(cpu, "bad write address %#lx len %u", addr, bytes); | |
d7e28ffe | 201 | } |
2d37f94a | 202 | /*:*/ |
d7e28ffe | 203 | |
2e04ef76 RR |
204 | /*H:030 |
205 | * Let's jump straight to the the main loop which runs the Guest. | |
bff672e6 | 206 | * Remember, this is called by the Launcher reading /dev/lguest, and we keep |
2e04ef76 RR |
207 | * going around and around until something interesting happens. |
208 | */ | |
d0953d42 | 209 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user) |
d7e28ffe | 210 | { |
bff672e6 | 211 | /* We stop running once the Guest is dead. */ |
382ac6b3 | 212 | while (!cpu->lg->dead) { |
abd41f03 | 213 | unsigned int irq; |
a32a8813 | 214 | bool more; |
abd41f03 | 215 | |
cc6d4fbc | 216 | /* First we run any hypercalls the Guest wants done. */ |
73044f05 GOC |
217 | if (cpu->hcall) |
218 | do_hypercalls(cpu); | |
cc6d4fbc | 219 | |
2e04ef76 RR |
220 | /* |
221 | * It's possible the Guest did a NOTIFY hypercall to the | |
a91d74a3 | 222 | * Launcher. |
2e04ef76 | 223 | */ |
5e232f4f | 224 | if (cpu->pending_notify) { |
a91d74a3 RR |
225 | /* |
226 | * Does it just needs to write to a registered | |
227 | * eventfd (ie. the appropriate virtqueue thread)? | |
228 | */ | |
df60aeef | 229 | if (!send_notify_to_eventfd(cpu)) { |
681f2066 | 230 | /* OK, we tell the main Launcher. */ |
df60aeef RR |
231 | if (put_user(cpu->pending_notify, user)) |
232 | return -EFAULT; | |
233 | return sizeof(cpu->pending_notify); | |
234 | } | |
d7e28ffe RR |
235 | } |
236 | ||
0acf0001 MH |
237 | /* |
238 | * All long-lived kernel loops need to check with this horrible | |
239 | * thing called the freezer. If the Host is trying to suspend, | |
240 | * it stops us. | |
241 | */ | |
242 | try_to_freeze(); | |
243 | ||
bff672e6 | 244 | /* Check for signals */ |
d7e28ffe RR |
245 | if (signal_pending(current)) |
246 | return -ERESTARTSYS; | |
247 | ||
2e04ef76 RR |
248 | /* |
249 | * Check if there are any interrupts which can be delivered now: | |
a6bd8e13 | 250 | * if so, this sets up the hander to be executed when we next |
2e04ef76 RR |
251 | * run the Guest. |
252 | */ | |
a32a8813 | 253 | irq = interrupt_pending(cpu, &more); |
abd41f03 | 254 | if (irq < LGUEST_IRQS) |
a32a8813 | 255 | try_deliver_interrupt(cpu, irq, more); |
d7e28ffe | 256 | |
2e04ef76 RR |
257 | /* |
258 | * Just make absolutely sure the Guest is still alive. One of | |
259 | * those hypercalls could have been fatal, for example. | |
260 | */ | |
382ac6b3 | 261 | if (cpu->lg->dead) |
d7e28ffe RR |
262 | break; |
263 | ||
2e04ef76 RR |
264 | /* |
265 | * If the Guest asked to be stopped, we sleep. The Guest's | |
266 | * clock timer will wake us. | |
267 | */ | |
66686c2a | 268 | if (cpu->halted) { |
d7e28ffe | 269 | set_current_state(TASK_INTERRUPTIBLE); |
2e04ef76 RR |
270 | /* |
271 | * Just before we sleep, make sure no interrupt snuck in | |
272 | * which we should be doing. | |
273 | */ | |
5dac051b | 274 | if (interrupt_pending(cpu, &more) < LGUEST_IRQS) |
abd41f03 RR |
275 | set_current_state(TASK_RUNNING); |
276 | else | |
277 | schedule(); | |
d7e28ffe RR |
278 | continue; |
279 | } | |
280 | ||
2e04ef76 RR |
281 | /* |
282 | * OK, now we're ready to jump into the Guest. First we put up | |
283 | * the "Do Not Disturb" sign: | |
284 | */ | |
d7e28ffe RR |
285 | local_irq_disable(); |
286 | ||
625efab1 | 287 | /* Actually run the Guest until something happens. */ |
d0953d42 | 288 | lguest_arch_run_guest(cpu); |
bff672e6 RR |
289 | |
290 | /* Now we're ready to be interrupted or moved to other CPUs */ | |
d7e28ffe RR |
291 | local_irq_enable(); |
292 | ||
625efab1 | 293 | /* Now we deal with whatever happened to the Guest. */ |
73044f05 | 294 | lguest_arch_handle_trap(cpu); |
d7e28ffe | 295 | } |
625efab1 | 296 | |
a6bd8e13 | 297 | /* Special case: Guest is 'dead' but wants a reboot. */ |
382ac6b3 | 298 | if (cpu->lg->dead == ERR_PTR(-ERESTART)) |
ec04b13f | 299 | return -ERESTART; |
a6bd8e13 | 300 | |
bff672e6 | 301 | /* The Guest is dead => "No such file or directory" */ |
d7e28ffe RR |
302 | return -ENOENT; |
303 | } | |
304 | ||
bff672e6 RR |
305 | /*H:000 |
306 | * Welcome to the Host! | |
307 | * | |
308 | * By this point your brain has been tickled by the Guest code and numbed by | |
309 | * the Launcher code; prepare for it to be stretched by the Host code. This is | |
310 | * the heart. Let's begin at the initialization routine for the Host's lg | |
311 | * module. | |
312 | */ | |
d7e28ffe RR |
313 | static int __init init(void) |
314 | { | |
315 | int err; | |
316 | ||
bff672e6 | 317 | /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */ |
b56e3215 | 318 | if (get_kernel_rpl() != 0) { |
5c55841d | 319 | printk("lguest is afraid of being a guest\n"); |
d7e28ffe RR |
320 | return -EPERM; |
321 | } | |
322 | ||
bff672e6 | 323 | /* First we put the Switcher up in very high virtual memory. */ |
d7e28ffe RR |
324 | err = map_switcher(); |
325 | if (err) | |
c18acd73 | 326 | goto out; |
d7e28ffe | 327 | |
c18acd73 RR |
328 | /* We might need to reserve an interrupt vector. */ |
329 | err = init_interrupts(); | |
330 | if (err) | |
3412b6ae | 331 | goto unmap; |
c18acd73 | 332 | |
bff672e6 | 333 | /* /dev/lguest needs to be registered. */ |
d7e28ffe | 334 | err = lguest_device_init(); |
c18acd73 RR |
335 | if (err) |
336 | goto free_interrupts; | |
bff672e6 | 337 | |
625efab1 JS |
338 | /* Finally we do some architecture-specific setup. */ |
339 | lguest_arch_host_init(); | |
bff672e6 RR |
340 | |
341 | /* All good! */ | |
d7e28ffe | 342 | return 0; |
c18acd73 RR |
343 | |
344 | free_interrupts: | |
345 | free_interrupts(); | |
c18acd73 RR |
346 | unmap: |
347 | unmap_switcher(); | |
348 | out: | |
349 | return err; | |
d7e28ffe RR |
350 | } |
351 | ||
bff672e6 | 352 | /* Cleaning up is just the same code, backwards. With a little French. */ |
d7e28ffe RR |
353 | static void __exit fini(void) |
354 | { | |
355 | lguest_device_remove(); | |
c18acd73 | 356 | free_interrupts(); |
d7e28ffe | 357 | unmap_switcher(); |
bff672e6 | 358 | |
625efab1 | 359 | lguest_arch_host_fini(); |
d7e28ffe | 360 | } |
625efab1 | 361 | /*:*/ |
d7e28ffe | 362 | |
2e04ef76 RR |
363 | /* |
364 | * The Host side of lguest can be a module. This is a nice way for people to | |
365 | * play with it. | |
366 | */ | |
d7e28ffe RR |
367 | module_init(init); |
368 | module_exit(fini); | |
369 | MODULE_LICENSE("GPL"); | |
370 | MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>"); |