Commit | Line | Data |
---|---|---|
f938d2c8 RR |
1 | /*P:200 This contains all the /dev/lguest code, whereby the userspace launcher |
2 | * controls and communicates with the Guest. For example, the first write will | |
3c6b5bfa RR |
3 | * tell us the Guest's memory layout, pagetable, entry point and kernel address |
4 | * offset. A read will run the Guest until something happens, such as a signal | |
15045275 | 5 | * or the Guest doing a NOTIFY out to the Launcher. :*/ |
d7e28ffe RR |
6 | #include <linux/uaccess.h> |
7 | #include <linux/miscdevice.h> | |
8 | #include <linux/fs.h> | |
ca94f2bd | 9 | #include <linux/sched.h> |
df60aeef RR |
10 | #include <linux/eventfd.h> |
11 | #include <linux/file.h> | |
d7e28ffe RR |
12 | #include "lg.h" |
13 | ||
df60aeef RR |
14 | bool send_notify_to_eventfd(struct lg_cpu *cpu) |
15 | { | |
16 | unsigned int i; | |
17 | struct lg_eventfd_map *map; | |
18 | ||
19 | /* lg->eventfds is RCU-protected */ | |
20 | rcu_read_lock(); | |
21 | map = rcu_dereference(cpu->lg->eventfds); | |
22 | for (i = 0; i < map->num; i++) { | |
23 | if (map->map[i].addr == cpu->pending_notify) { | |
24 | eventfd_signal(map->map[i].event, 1); | |
25 | cpu->pending_notify = 0; | |
26 | break; | |
27 | } | |
28 | } | |
29 | rcu_read_unlock(); | |
30 | return cpu->pending_notify == 0; | |
31 | } | |
32 | ||
33 | static int add_eventfd(struct lguest *lg, unsigned long addr, int fd) | |
34 | { | |
35 | struct lg_eventfd_map *new, *old = lg->eventfds; | |
36 | ||
37 | if (!addr) | |
38 | return -EINVAL; | |
39 | ||
40 | /* Replace the old array with the new one, carefully: others can | |
41 | * be accessing it at the same time */ | |
42 | new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1), | |
43 | GFP_KERNEL); | |
44 | if (!new) | |
45 | return -ENOMEM; | |
46 | ||
47 | /* First make identical copy. */ | |
48 | memcpy(new->map, old->map, sizeof(old->map[0]) * old->num); | |
49 | new->num = old->num; | |
50 | ||
51 | /* Now append new entry. */ | |
52 | new->map[new->num].addr = addr; | |
13389010 | 53 | new->map[new->num].event = eventfd_ctx_fdget(fd); |
df60aeef RR |
54 | if (IS_ERR(new->map[new->num].event)) { |
55 | kfree(new); | |
56 | return PTR_ERR(new->map[new->num].event); | |
57 | } | |
58 | new->num++; | |
59 | ||
60 | /* Now put new one in place. */ | |
61 | rcu_assign_pointer(lg->eventfds, new); | |
62 | ||
63 | /* We're not in a big hurry. Wait until noone's looking at old | |
64 | * version, then delete it. */ | |
65 | synchronize_rcu(); | |
66 | kfree(old); | |
67 | ||
68 | return 0; | |
69 | } | |
70 | ||
71 | static int attach_eventfd(struct lguest *lg, const unsigned long __user *input) | |
72 | { | |
73 | unsigned long addr, fd; | |
74 | int err; | |
75 | ||
76 | if (get_user(addr, input) != 0) | |
77 | return -EFAULT; | |
78 | input++; | |
79 | if (get_user(fd, input) != 0) | |
80 | return -EFAULT; | |
81 | ||
82 | mutex_lock(&lguest_lock); | |
83 | err = add_eventfd(lg, addr, fd); | |
84 | mutex_unlock(&lguest_lock); | |
85 | ||
86 | return 0; | |
87 | } | |
88 | ||
dde79789 RR |
89 | /*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt |
90 | * number to /dev/lguest. */ | |
177e449d | 91 | static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) |
d7e28ffe | 92 | { |
511801dc | 93 | unsigned long irq; |
d7e28ffe RR |
94 | |
95 | if (get_user(irq, input) != 0) | |
96 | return -EFAULT; | |
97 | if (irq >= LGUEST_IRQS) | |
98 | return -EINVAL; | |
9f155a9b RR |
99 | |
100 | set_interrupt(cpu, irq); | |
d7e28ffe RR |
101 | return 0; |
102 | } | |
103 | ||
dde79789 RR |
104 | /*L:040 Once our Guest is initialized, the Launcher makes it run by reading |
105 | * from /dev/lguest. */ | |
d7e28ffe RR |
106 | static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) |
107 | { | |
108 | struct lguest *lg = file->private_data; | |
d0953d42 GOC |
109 | struct lg_cpu *cpu; |
110 | unsigned int cpu_id = *o; | |
d7e28ffe | 111 | |
dde79789 | 112 | /* You must write LHREQ_INITIALIZE first! */ |
d7e28ffe RR |
113 | if (!lg) |
114 | return -EINVAL; | |
115 | ||
d0953d42 GOC |
116 | /* Watch out for arbitrary vcpu indexes! */ |
117 | if (cpu_id >= lg->nr_cpus) | |
118 | return -EINVAL; | |
119 | ||
120 | cpu = &lg->cpus[cpu_id]; | |
121 | ||
e1e72965 | 122 | /* If you're not the task which owns the Guest, go away. */ |
66686c2a | 123 | if (current != cpu->tsk) |
d7e28ffe RR |
124 | return -EPERM; |
125 | ||
a6bd8e13 | 126 | /* If the Guest is already dead, we indicate why */ |
d7e28ffe RR |
127 | if (lg->dead) { |
128 | size_t len; | |
129 | ||
dde79789 | 130 | /* lg->dead either contains an error code, or a string. */ |
d7e28ffe RR |
131 | if (IS_ERR(lg->dead)) |
132 | return PTR_ERR(lg->dead); | |
133 | ||
dde79789 | 134 | /* We can only return as much as the buffer they read with. */ |
d7e28ffe RR |
135 | len = min(size, strlen(lg->dead)+1); |
136 | if (copy_to_user(user, lg->dead, len) != 0) | |
137 | return -EFAULT; | |
138 | return len; | |
139 | } | |
140 | ||
a6bd8e13 | 141 | /* If we returned from read() last time because the Guest sent I/O, |
dde79789 | 142 | * clear the flag. */ |
5e232f4f GOC |
143 | if (cpu->pending_notify) |
144 | cpu->pending_notify = 0; | |
d7e28ffe | 145 | |
dde79789 | 146 | /* Run the Guest until something interesting happens. */ |
d0953d42 | 147 | return run_guest(cpu, (unsigned long __user *)user); |
d7e28ffe RR |
148 | } |
149 | ||
a6bd8e13 RR |
150 | /*L:025 This actually initializes a CPU. For the moment, a Guest is only |
151 | * uniprocessor, so "id" is always 0. */ | |
4dcc53da GOC |
152 | static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) |
153 | { | |
a6bd8e13 | 154 | /* We have a limited number the number of CPUs in the lguest struct. */ |
24adf127 | 155 | if (id >= ARRAY_SIZE(cpu->lg->cpus)) |
4dcc53da GOC |
156 | return -EINVAL; |
157 | ||
a6bd8e13 | 158 | /* Set up this CPU's id, and pointer back to the lguest struct. */ |
4dcc53da GOC |
159 | cpu->id = id; |
160 | cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); | |
161 | cpu->lg->nr_cpus++; | |
a6bd8e13 RR |
162 | |
163 | /* Each CPU has a timer it can set. */ | |
ad8d8f3b | 164 | init_clockdev(cpu); |
4dcc53da | 165 | |
a53a35a8 GOC |
166 | /* We need a complete page for the Guest registers: they are accessible |
167 | * to the Guest and we can only grant it access to whole pages. */ | |
168 | cpu->regs_page = get_zeroed_page(GFP_KERNEL); | |
169 | if (!cpu->regs_page) | |
170 | return -ENOMEM; | |
171 | ||
172 | /* We actually put the registers at the bottom of the page. */ | |
173 | cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); | |
174 | ||
175 | /* Now we initialize the Guest's registers, handing it the start | |
176 | * address. */ | |
177 | lguest_arch_setup_regs(cpu, start_ip); | |
178 | ||
66686c2a | 179 | /* We keep a pointer to the Launcher task (ie. current task) for when |
a6bd8e13 | 180 | * other Guests want to wake this one (eg. console input). */ |
66686c2a GOC |
181 | cpu->tsk = current; |
182 | ||
183 | /* We need to keep a pointer to the Launcher's memory map, because if | |
184 | * the Launcher dies we need to clean it up. If we don't keep a | |
185 | * reference, it is destroyed before close() is called. */ | |
186 | cpu->mm = get_task_mm(cpu->tsk); | |
187 | ||
f34f8c5f GOC |
188 | /* We remember which CPU's pages this Guest used last, for optimization |
189 | * when the same Guest runs on the same CPU twice. */ | |
190 | cpu->last_pages = NULL; | |
191 | ||
a6bd8e13 | 192 | /* No error == success. */ |
4dcc53da GOC |
193 | return 0; |
194 | } | |
195 | ||
58a24566 | 196 | /*L:020 The initialization write supplies 3 pointer sized (32 or 64 bit) |
511801dc | 197 | * values (in addition to the LHREQ_INITIALIZE value). These are: |
dde79789 | 198 | * |
3c6b5bfa RR |
199 | * base: The start of the Guest-physical memory inside the Launcher memory. |
200 | * | |
dde79789 | 201 | * pfnlimit: The highest (Guest-physical) page number the Guest should be |
e1e72965 RR |
202 | * allowed to access. The Guest memory lives inside the Launcher, so it sets |
203 | * this to ensure the Guest can only reach its own memory. | |
dde79789 | 204 | * |
dde79789 | 205 | * start: The first instruction to execute ("eip" in x86-speak). |
dde79789 | 206 | */ |
511801dc | 207 | static int initialize(struct file *file, const unsigned long __user *input) |
d7e28ffe | 208 | { |
dde79789 RR |
209 | /* "struct lguest" contains everything we (the Host) know about a |
210 | * Guest. */ | |
d7e28ffe | 211 | struct lguest *lg; |
48245cc0 | 212 | int err; |
58a24566 | 213 | unsigned long args[3]; |
d7e28ffe | 214 | |
48245cc0 RR |
215 | /* We grab the Big Lguest lock, which protects against multiple |
216 | * simultaneous initializations. */ | |
d7e28ffe | 217 | mutex_lock(&lguest_lock); |
dde79789 | 218 | /* You can't initialize twice! Close the device and start again... */ |
d7e28ffe RR |
219 | if (file->private_data) { |
220 | err = -EBUSY; | |
221 | goto unlock; | |
222 | } | |
223 | ||
224 | if (copy_from_user(args, input, sizeof(args)) != 0) { | |
225 | err = -EFAULT; | |
226 | goto unlock; | |
227 | } | |
228 | ||
48245cc0 RR |
229 | lg = kzalloc(sizeof(*lg), GFP_KERNEL); |
230 | if (!lg) { | |
231 | err = -ENOMEM; | |
d7e28ffe RR |
232 | goto unlock; |
233 | } | |
dde79789 | 234 | |
df60aeef RR |
235 | lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL); |
236 | if (!lg->eventfds) { | |
237 | err = -ENOMEM; | |
238 | goto free_lg; | |
239 | } | |
240 | lg->eventfds->num = 0; | |
241 | ||
dde79789 | 242 | /* Populate the easy fields of our "struct lguest" */ |
74dbf719 | 243 | lg->mem_base = (void __user *)args[0]; |
3c6b5bfa | 244 | lg->pfn_limit = args[1]; |
dde79789 | 245 | |
58a24566 MZ |
246 | /* This is the first cpu (cpu 0) and it will start booting at args[2] */ |
247 | err = lg_cpu_start(&lg->cpus[0], 0, args[2]); | |
4dcc53da | 248 | if (err) |
df60aeef | 249 | goto free_eventfds; |
4dcc53da | 250 | |
dde79789 | 251 | /* Initialize the Guest's shadow page tables, using the toplevel |
a6bd8e13 | 252 | * address the Launcher gave us. This allocates memory, so can fail. */ |
58a24566 | 253 | err = init_guest_pagetable(lg); |
d7e28ffe RR |
254 | if (err) |
255 | goto free_regs; | |
256 | ||
dde79789 | 257 | /* We keep our "struct lguest" in the file's private_data. */ |
d7e28ffe RR |
258 | file->private_data = lg; |
259 | ||
260 | mutex_unlock(&lguest_lock); | |
261 | ||
dde79789 | 262 | /* And because this is a write() call, we return the length used. */ |
d7e28ffe RR |
263 | return sizeof(args); |
264 | ||
265 | free_regs: | |
a53a35a8 GOC |
266 | /* FIXME: This should be in free_vcpu */ |
267 | free_page(lg->cpus[0].regs_page); | |
df60aeef RR |
268 | free_eventfds: |
269 | kfree(lg->eventfds); | |
270 | free_lg: | |
43054412 | 271 | kfree(lg); |
d7e28ffe RR |
272 | unlock: |
273 | mutex_unlock(&lguest_lock); | |
274 | return err; | |
275 | } | |
276 | ||
dde79789 | 277 | /*L:010 The first operation the Launcher does must be a write. All writes |
e1e72965 | 278 | * start with an unsigned long number: for the first write this must be |
dde79789 | 279 | * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use |
a6bd8e13 RR |
280 | * writes of other values to send interrupts. |
281 | * | |
282 | * Note that we overload the "offset" in the /dev/lguest file to indicate what | |
283 | * CPU number we're dealing with. Currently this is always 0, since we only | |
284 | * support uniprocessor Guests, but you can see the beginnings of SMP support | |
285 | * here. */ | |
511801dc | 286 | static ssize_t write(struct file *file, const char __user *in, |
d7e28ffe RR |
287 | size_t size, loff_t *off) |
288 | { | |
a6bd8e13 | 289 | /* Once the Guest is initialized, we hold the "struct lguest" in the |
dde79789 | 290 | * file private data. */ |
d7e28ffe | 291 | struct lguest *lg = file->private_data; |
511801dc JS |
292 | const unsigned long __user *input = (const unsigned long __user *)in; |
293 | unsigned long req; | |
177e449d | 294 | struct lg_cpu *uninitialized_var(cpu); |
7ea07a15 | 295 | unsigned int cpu_id = *off; |
d7e28ffe | 296 | |
a6bd8e13 | 297 | /* The first value tells us what this request is. */ |
d7e28ffe RR |
298 | if (get_user(req, input) != 0) |
299 | return -EFAULT; | |
511801dc | 300 | input++; |
d7e28ffe | 301 | |
dde79789 | 302 | /* If you haven't initialized, you must do that first. */ |
7ea07a15 GOC |
303 | if (req != LHREQ_INITIALIZE) { |
304 | if (!lg || (cpu_id >= lg->nr_cpus)) | |
305 | return -EINVAL; | |
306 | cpu = &lg->cpus[cpu_id]; | |
dde79789 | 307 | |
f73d1e6c ET |
308 | /* Once the Guest is dead, you can only read() why it died. */ |
309 | if (lg->dead) | |
310 | return -ENOENT; | |
f73d1e6c | 311 | } |
d7e28ffe RR |
312 | |
313 | switch (req) { | |
314 | case LHREQ_INITIALIZE: | |
511801dc | 315 | return initialize(file, input); |
d7e28ffe | 316 | case LHREQ_IRQ: |
177e449d | 317 | return user_send_irq(cpu, input); |
df60aeef RR |
318 | case LHREQ_EVENTFD: |
319 | return attach_eventfd(lg, input); | |
d7e28ffe RR |
320 | default: |
321 | return -EINVAL; | |
322 | } | |
323 | } | |
324 | ||
dde79789 RR |
325 | /*L:060 The final piece of interface code is the close() routine. It reverses |
326 | * everything done in initialize(). This is usually called because the | |
327 | * Launcher exited. | |
328 | * | |
329 | * Note that the close routine returns 0 or a negative error number: it can't | |
330 | * really fail, but it can whine. I blame Sun for this wart, and K&R C for | |
331 | * letting them do it. :*/ | |
d7e28ffe RR |
332 | static int close(struct inode *inode, struct file *file) |
333 | { | |
334 | struct lguest *lg = file->private_data; | |
ad8d8f3b | 335 | unsigned int i; |
d7e28ffe | 336 | |
dde79789 | 337 | /* If we never successfully initialized, there's nothing to clean up */ |
d7e28ffe RR |
338 | if (!lg) |
339 | return 0; | |
340 | ||
dde79789 RR |
341 | /* We need the big lock, to protect from inter-guest I/O and other |
342 | * Launchers initializing guests. */ | |
d7e28ffe | 343 | mutex_lock(&lguest_lock); |
66686c2a GOC |
344 | |
345 | /* Free up the shadow page tables for the Guest. */ | |
346 | free_guest_pagetable(lg); | |
347 | ||
a53a35a8 | 348 | for (i = 0; i < lg->nr_cpus; i++) { |
ad8d8f3b GOC |
349 | /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ |
350 | hrtimer_cancel(&lg->cpus[i].hrt); | |
a53a35a8 GOC |
351 | /* We can free up the register page we allocated. */ |
352 | free_page(lg->cpus[i].regs_page); | |
66686c2a GOC |
353 | /* Now all the memory cleanups are done, it's safe to release |
354 | * the Launcher's memory management structure. */ | |
355 | mmput(lg->cpus[i].mm); | |
a53a35a8 | 356 | } |
df60aeef RR |
357 | |
358 | /* Release any eventfds they registered. */ | |
359 | for (i = 0; i < lg->eventfds->num; i++) | |
13389010 | 360 | eventfd_ctx_put(lg->eventfds->map[i].event); |
df60aeef RR |
361 | kfree(lg->eventfds); |
362 | ||
dde79789 RR |
363 | /* If lg->dead doesn't contain an error code it will be NULL or a |
364 | * kmalloc()ed string, either of which is ok to hand to kfree(). */ | |
d7e28ffe RR |
365 | if (!IS_ERR(lg->dead)) |
366 | kfree(lg->dead); | |
05dfdbbd MW |
367 | /* Free the memory allocated to the lguest_struct */ |
368 | kfree(lg); | |
dde79789 | 369 | /* Release lock and exit. */ |
d7e28ffe | 370 | mutex_unlock(&lguest_lock); |
dde79789 | 371 | |
d7e28ffe RR |
372 | return 0; |
373 | } | |
374 | ||
dde79789 RR |
375 | /*L:000 |
376 | * Welcome to our journey through the Launcher! | |
377 | * | |
378 | * The Launcher is the Host userspace program which sets up, runs and services | |
379 | * the Guest. In fact, many comments in the Drivers which refer to "the Host" | |
380 | * doing things are inaccurate: the Launcher does all the device handling for | |
e1e72965 | 381 | * the Guest, but the Guest can't know that. |
dde79789 RR |
382 | * |
383 | * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we | |
384 | * shall see more of that later. | |
385 | * | |
386 | * We begin our understanding with the Host kernel interface which the Launcher | |
387 | * uses: reading and writing a character device called /dev/lguest. All the | |
388 | * work happens in the read(), write() and close() routines: */ | |
d7e28ffe RR |
389 | static struct file_operations lguest_fops = { |
390 | .owner = THIS_MODULE, | |
391 | .release = close, | |
392 | .write = write, | |
393 | .read = read, | |
394 | }; | |
dde79789 RR |
395 | |
396 | /* This is a textbook example of a "misc" character device. Populate a "struct | |
397 | * miscdevice" and register it with misc_register(). */ | |
d7e28ffe RR |
398 | static struct miscdevice lguest_dev = { |
399 | .minor = MISC_DYNAMIC_MINOR, | |
400 | .name = "lguest", | |
401 | .fops = &lguest_fops, | |
402 | }; | |
403 | ||
404 | int __init lguest_device_init(void) | |
405 | { | |
406 | return misc_register(&lguest_dev); | |
407 | } | |
408 | ||
409 | void __exit lguest_device_remove(void) | |
410 | { | |
411 | misc_deregister(&lguest_dev); | |
412 | } |