1 /*P:200 This contains all the /dev/lguest code, whereby the userspace
2 * launcher controls and communicates with the Guest. For example,
3 * the first write will tell us the Guest's memory layout and entry
4 * point. A read will run the Guest until something happens, such as
5 * a signal or the Guest doing a NOTIFY out to the Launcher. There is
6 * also a way for the Launcher to attach eventfds to particular NOTIFY
7 * values instead of returning from the read() call.
9 #include <linux/uaccess.h>
10 #include <linux/miscdevice.h>
12 #include <linux/sched.h>
13 #include <linux/eventfd.h>
14 #include <linux/file.h>
15 #include <linux/slab.h>
19 * Before we move on, let's jump ahead and look at what the kernel does when
20 * it needs to look up the eventfds. That will complete our picture of how we
23 * The notification value is in cpu->pending_notify: we return true if it went
26 bool send_notify_to_eventfd(struct lg_cpu *cpu)
29 struct lg_eventfd_map *map;
32 * This "rcu_read_lock()" helps track when someone is still looking at
33 * the (RCU-using) eventfds array. It's not actually a lock at all;
34 * indeed it's a noop in many configurations. (You didn't expect me to
35 * explain all the RCU secrets here, did you?)
39 * rcu_dereference is the counter-side of rcu_assign_pointer(); it
40 * makes sure we don't access the memory pointed to by
41 * cpu->lg->eventfds before cpu->lg->eventfds is set. Sounds crazy,
42 * but Alpha allows this! Paul McKenney points out that a really
43 * aggressive compiler could have the same effect:
44 * http://lists.ozlabs.org/pipermail/lguest/2009-July/001560.html
46 * So play safe, use rcu_dereference to get the rcu-protected pointer:
48 map = rcu_dereference(cpu->lg->eventfds);
50 * Simple array search: even if they add an eventfd while we do this,
51 * we'll continue to use the old array and just won't see the new one.
53 for (i = 0; i < map->num; i++) {
54 if (map->map[i].addr == cpu->pending_notify) {
55 eventfd_signal(map->map[i].event, 1);
56 cpu->pending_notify = 0;
60 /* We're done with the rcu-protected variable cpu->lg->eventfds. */
63 /* If we cleared the notification, it's because we found a match. */
64 return cpu->pending_notify == 0;
68 * One of the more tricksy tricks in the Linux Kernel is a technique called
69 * Read Copy Update. Since one point of lguest is to teach lguest journeyers
70 * about kernel coding, I use it here. (In case you're curious, other purposes
71 * include learning about virtualization and instilling a deep appreciation for
72 * simplicity and puppies).
74 * We keep a simple array which maps LHCALL_NOTIFY values to eventfds, but we
75 * add new eventfds without ever blocking readers from accessing the array.
76 * The current Launcher only does this during boot, so that never happens. But
77 * Read Copy Update is cool, and adding a lock risks damaging even more puppies
78 * than this code does.
80 * We allocate a brand new one-larger array, copy the old one and add our new
81 * element. Then we make the lg eventfd pointer point to the new array.
82 * That's the easy part: now we need to free the old one, but we need to make
83 * sure no slow CPU somewhere is still looking at it. That's what
84 * synchronize_rcu does for us: waits until every CPU has indicated that it has
85 * moved on to know it's no longer using the old one.
87 * If that's unclear, see http://en.wikipedia.org/wiki/Read-copy-update.
89 static int add_eventfd(struct lguest *lg, unsigned long addr, int fd)
91 struct lg_eventfd_map *new, *old = lg->eventfds;
94 * We don't allow notifications on value 0 anyway (pending_notify of
95 * 0 means "nothing pending").
101 * Replace the old array with the new one, carefully: others can
102 * be accessing it at the same time.
104 new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1),
109 /* First make identical copy. */
110 memcpy(new->map, old->map, sizeof(old->map[0]) * old->num);
113 /* Now append new entry. */
114 new->map[new->num].addr = addr;
115 new->map[new->num].event = eventfd_ctx_fdget(fd);
116 if (IS_ERR(new->map[new->num].event)) {
117 int err = PTR_ERR(new->map[new->num].event);
124 * Now put new one in place: rcu_assign_pointer() is a fancy way of
125 * doing "lg->eventfds = new", but it uses memory barriers to make
126 * absolutely sure that the contents of "new" written above is nailed
127 * down before we actually do the assignment.
129 * We have to think about these kinds of things when we're operating on
130 * live data without locks.
132 rcu_assign_pointer(lg->eventfds, new);
135 * We're not in a big hurry. Wait until no one's looking at old
136 * version, then free it.
145 * Receiving notifications from the Guest is usually done by attaching a
146 * particular LHCALL_NOTIFY value to an event filedescriptor. The eventfd will
147 * become readable when the Guest does an LHCALL_NOTIFY with that value.
149 * This is really convenient for processing each virtqueue in a separate
152 static int attach_eventfd(struct lguest *lg, const unsigned long __user *input)
154 unsigned long addr, fd;
157 if (get_user(addr, input) != 0)
160 if (get_user(fd, input) != 0)
164 * Just make sure two callers don't add eventfds at once. We really
165 * only need to lock against callers adding to the same Guest, so using
166 * the Big Lguest Lock is overkill. But this is setup, not a fast path.
168 mutex_lock(&lguest_lock);
169 err = add_eventfd(lg, addr, fd);
170 mutex_unlock(&lguest_lock);
176 * Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
177 * number to /dev/lguest.
179 static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
183 if (get_user(irq, input) != 0)
185 if (irq >= LGUEST_IRQS)
189 * Next time the Guest runs, the core code will see if it can deliver
192 set_interrupt(cpu, irq);
197 * Once our Guest is initialized, the Launcher makes it run by reading
200 static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
202 struct lguest *lg = file->private_data;
204 unsigned int cpu_id = *o;
206 /* You must write LHREQ_INITIALIZE first! */
210 /* Watch out for arbitrary vcpu indexes! */
211 if (cpu_id >= lg->nr_cpus)
214 cpu = &lg->cpus[cpu_id];
216 /* If you're not the task which owns the Guest, go away. */
217 if (current != cpu->tsk)
220 /* If the Guest is already dead, we indicate why */
224 /* lg->dead either contains an error code, or a string. */
225 if (IS_ERR(lg->dead))
226 return PTR_ERR(lg->dead);
228 /* We can only return as much as the buffer they read with. */
229 len = min(size, strlen(lg->dead)+1);
230 if (copy_to_user(user, lg->dead, len) != 0)
236 * If we returned from read() last time because the Guest sent I/O,
239 if (cpu->pending_notify)
240 cpu->pending_notify = 0;
242 /* Run the Guest until something interesting happens. */
243 return run_guest(cpu, (unsigned long __user *)user);
247 * This actually initializes a CPU. For the moment, a Guest is only
248 * uniprocessor, so "id" is always 0.
250 static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
252 /* We have a limited number the number of CPUs in the lguest struct. */
253 if (id >= ARRAY_SIZE(cpu->lg->cpus))
256 /* Set up this CPU's id, and pointer back to the lguest struct. */
258 cpu->lg = container_of((cpu - id), struct lguest, cpus[0]);
261 /* Each CPU has a timer it can set. */
265 * We need a complete page for the Guest registers: they are accessible
266 * to the Guest and we can only grant it access to whole pages.
268 cpu->regs_page = get_zeroed_page(GFP_KERNEL);
272 /* We actually put the registers at the bottom of the page. */
273 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
276 * Now we initialize the Guest's registers, handing it the start
279 lguest_arch_setup_regs(cpu, start_ip);
282 * We keep a pointer to the Launcher task (ie. current task) for when
283 * other Guests want to wake this one (eg. console input).
288 * We need to keep a pointer to the Launcher's memory map, because if
289 * the Launcher dies we need to clean it up. If we don't keep a
290 * reference, it is destroyed before close() is called.
292 cpu->mm = get_task_mm(cpu->tsk);
295 * We remember which CPU's pages this Guest used last, for optimization
296 * when the same Guest runs on the same CPU twice.
298 cpu->last_pages = NULL;
300 /* No error == success. */
305 * The initialization write supplies 3 pointer sized (32 or 64 bit) values (in
306 * addition to the LHREQ_INITIALIZE value). These are:
308 * base: The start of the Guest-physical memory inside the Launcher memory.
310 * pfnlimit: The highest (Guest-physical) page number the Guest should be
311 * allowed to access. The Guest memory lives inside the Launcher, so it sets
312 * this to ensure the Guest can only reach its own memory.
314 * start: The first instruction to execute ("eip" in x86-speak).
316 static int initialize(struct file *file, const unsigned long __user *input)
318 /* "struct lguest" contains all we (the Host) know about a Guest. */
321 unsigned long args[3];
324 * We grab the Big Lguest lock, which protects against multiple
325 * simultaneous initializations.
327 mutex_lock(&lguest_lock);
328 /* You can't initialize twice! Close the device and start again... */
329 if (file->private_data) {
334 if (copy_from_user(args, input, sizeof(args)) != 0) {
339 lg = kzalloc(sizeof(*lg), GFP_KERNEL);
345 lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL);
350 lg->eventfds->num = 0;
352 /* Populate the easy fields of our "struct lguest" */
353 lg->mem_base = (void __user *)args[0];
354 lg->pfn_limit = args[1];
356 /* This is the first cpu (cpu 0) and it will start booting at args[2] */
357 err = lg_cpu_start(&lg->cpus[0], 0, args[2]);
362 * Initialize the Guest's shadow page tables. This allocates
363 * memory, so can fail.
365 err = init_guest_pagetable(lg);
369 /* We keep our "struct lguest" in the file's private_data. */
370 file->private_data = lg;
372 mutex_unlock(&lguest_lock);
374 /* And because this is a write() call, we return the length used. */
378 /* FIXME: This should be in free_vcpu */
379 free_page(lg->cpus[0].regs_page);
385 mutex_unlock(&lguest_lock);
390 * The first operation the Launcher does must be a write. All writes
391 * start with an unsigned long number: for the first write this must be
392 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use
393 * writes of other values to send interrupts or set up receipt of notifications.
395 * Note that we overload the "offset" in the /dev/lguest file to indicate what
396 * CPU number we're dealing with. Currently this is always 0 since we only
397 * support uniprocessor Guests, but you can see the beginnings of SMP support
400 static ssize_t write(struct file *file, const char __user *in,
401 size_t size, loff_t *off)
404 * Once the Guest is initialized, we hold the "struct lguest" in the
407 struct lguest *lg = file->private_data;
408 const unsigned long __user *input = (const unsigned long __user *)in;
410 struct lg_cpu *uninitialized_var(cpu);
411 unsigned int cpu_id = *off;
413 /* The first value tells us what this request is. */
414 if (get_user(req, input) != 0)
418 /* If you haven't initialized, you must do that first. */
419 if (req != LHREQ_INITIALIZE) {
420 if (!lg || (cpu_id >= lg->nr_cpus))
422 cpu = &lg->cpus[cpu_id];
424 /* Once the Guest is dead, you can only read() why it died. */
430 case LHREQ_INITIALIZE:
431 return initialize(file, input);
433 return user_send_irq(cpu, input);
435 return attach_eventfd(lg, input);
442 * The final piece of interface code is the close() routine. It reverses
443 * everything done in initialize(). This is usually called because the
446 * Note that the close routine returns 0 or a negative error number: it can't
447 * really fail, but it can whine. I blame Sun for this wart, and K&R C for
448 * letting them do it.
450 static int close(struct inode *inode, struct file *file)
452 struct lguest *lg = file->private_data;
455 /* If we never successfully initialized, there's nothing to clean up */
460 * We need the big lock, to protect from inter-guest I/O and other
461 * Launchers initializing guests.
463 mutex_lock(&lguest_lock);
465 /* Free up the shadow page tables for the Guest. */
466 free_guest_pagetable(lg);
468 for (i = 0; i < lg->nr_cpus; i++) {
469 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
470 hrtimer_cancel(&lg->cpus[i].hrt);
471 /* We can free up the register page we allocated. */
472 free_page(lg->cpus[i].regs_page);
474 * Now all the memory cleanups are done, it's safe to release
475 * the Launcher's memory management structure.
477 mmput(lg->cpus[i].mm);
480 /* Release any eventfds they registered. */
481 for (i = 0; i < lg->eventfds->num; i++)
482 eventfd_ctx_put(lg->eventfds->map[i].event);
486 * If lg->dead doesn't contain an error code it will be NULL or a
487 * kmalloc()ed string, either of which is ok to hand to kfree().
489 if (!IS_ERR(lg->dead))
491 /* Free the memory allocated to the lguest_struct */
493 /* Release lock and exit. */
494 mutex_unlock(&lguest_lock);
500 * Welcome to our journey through the Launcher!
502 * The Launcher is the Host userspace program which sets up, runs and services
503 * the Guest. In fact, many comments in the Drivers which refer to "the Host"
504 * doing things are inaccurate: the Launcher does all the device handling for
505 * the Guest, but the Guest can't know that.
507 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
508 * shall see more of that later.
510 * We begin our understanding with the Host kernel interface which the Launcher
511 * uses: reading and writing a character device called /dev/lguest. All the
512 * work happens in the read(), write() and close() routines:
514 static const struct file_operations lguest_fops = {
515 .owner = THIS_MODULE,
519 .llseek = default_llseek,
524 * This is a textbook example of a "misc" character device. Populate a "struct
525 * miscdevice" and register it with misc_register().
527 static struct miscdevice lguest_dev = {
528 .minor = MISC_DYNAMIC_MINOR,
530 .fops = &lguest_fops,
533 int __init lguest_device_init(void)
535 return misc_register(&lguest_dev);
538 void __exit lguest_device_remove(void)
540 misc_deregister(&lguest_dev);