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
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
5 * or the Guest doing a NOTIFY out to the Launcher. :*/
6 #include <linux/uaccess.h>
7 #include <linux/miscdevice.h>
9 #include <linux/sched.h>
12 /*L:055 When something happens, the Waker process needs a way to stop the
13 * kernel running the Guest and return to the Launcher. So the Waker writes
14 * LHREQ_BREAK and the value "1" to /dev/lguest to do this. Once the Launcher
15 * has done whatever needs attention, it writes LHREQ_BREAK and "0" to release
17 static int break_guest_out(struct lg_cpu *cpu, const unsigned long __user*input)
21 /* Fetch whether they're turning break on or off. */
22 if (get_user(on, input) != 0)
27 /* Pop it out of the Guest (may be running on different CPU) */
28 wake_up_process(cpu->tsk);
29 /* Wait for them to reset it */
30 return wait_event_interruptible(cpu->break_wq, !cpu->break_out);
33 wake_up(&cpu->break_wq);
38 /*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
39 * number to /dev/lguest. */
40 static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
44 if (get_user(irq, input) != 0)
46 if (irq >= LGUEST_IRQS)
48 /* Next time the Guest runs, the core code will see if it can deliver
50 set_bit(irq, cpu->irqs_pending);
54 /*L:040 Once our Guest is initialized, the Launcher makes it run by reading
55 * from /dev/lguest. */
56 static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
58 struct lguest *lg = file->private_data;
60 unsigned int cpu_id = *o;
62 /* You must write LHREQ_INITIALIZE first! */
66 /* Watch out for arbitrary vcpu indexes! */
67 if (cpu_id >= lg->nr_cpus)
70 cpu = &lg->cpus[cpu_id];
72 /* If you're not the task which owns the Guest, go away. */
73 if (current != cpu->tsk)
76 /* If the guest is already dead, we indicate why */
80 /* lg->dead either contains an error code, or a string. */
82 return PTR_ERR(lg->dead);
84 /* We can only return as much as the buffer they read with. */
85 len = min(size, strlen(lg->dead)+1);
86 if (copy_to_user(user, lg->dead, len) != 0)
91 /* If we returned from read() last time because the Guest notified,
93 if (cpu->pending_notify)
94 cpu->pending_notify = 0;
96 /* Run the Guest until something interesting happens. */
97 return run_guest(cpu, (unsigned long __user *)user);
100 static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
106 cpu->lg = container_of((cpu - id), struct lguest, cpus[0]);
110 /* We need a complete page for the Guest registers: they are accessible
111 * to the Guest and we can only grant it access to whole pages. */
112 cpu->regs_page = get_zeroed_page(GFP_KERNEL);
116 /* We actually put the registers at the bottom of the page. */
117 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
119 /* Now we initialize the Guest's registers, handing it the start
121 lguest_arch_setup_regs(cpu, start_ip);
123 /* Initialize the queue for the waker to wait on */
124 init_waitqueue_head(&cpu->break_wq);
126 /* We keep a pointer to the Launcher task (ie. current task) for when
127 * other Guests want to wake this one (inter-Guest I/O). */
130 /* We need to keep a pointer to the Launcher's memory map, because if
131 * the Launcher dies we need to clean it up. If we don't keep a
132 * reference, it is destroyed before close() is called. */
133 cpu->mm = get_task_mm(cpu->tsk);
135 /* We remember which CPU's pages this Guest used last, for optimization
136 * when the same Guest runs on the same CPU twice. */
137 cpu->last_pages = NULL;
142 /*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit)
143 * values (in addition to the LHREQ_INITIALIZE value). These are:
145 * base: The start of the Guest-physical memory inside the Launcher memory.
147 * pfnlimit: The highest (Guest-physical) page number the Guest should be
148 * allowed to access. The Guest memory lives inside the Launcher, so it sets
149 * this to ensure the Guest can only reach its own memory.
151 * pgdir: The (Guest-physical) address of the top of the initial Guest
152 * pagetables (which are set up by the Launcher).
154 * start: The first instruction to execute ("eip" in x86-speak).
156 static int initialize(struct file *file, const unsigned long __user *input)
158 /* "struct lguest" contains everything we (the Host) know about a
162 unsigned long args[4];
164 /* We grab the Big Lguest lock, which protects against multiple
165 * simultaneous initializations. */
166 mutex_lock(&lguest_lock);
167 /* You can't initialize twice! Close the device and start again... */
168 if (file->private_data) {
173 if (copy_from_user(args, input, sizeof(args)) != 0) {
178 lg = kzalloc(sizeof(*lg), GFP_KERNEL);
184 /* Populate the easy fields of our "struct lguest" */
185 lg->mem_base = (void __user *)(long)args[0];
186 lg->pfn_limit = args[1];
188 /* This is the first cpu */
189 err = lg_cpu_start(&lg->cpus[0], 0, args[3]);
193 /* Initialize the Guest's shadow page tables, using the toplevel
194 * address the Launcher gave us. This allocates memory, so can
196 err = init_guest_pagetable(lg, args[2]);
200 /* We keep our "struct lguest" in the file's private_data. */
201 file->private_data = lg;
203 mutex_unlock(&lguest_lock);
205 /* And because this is a write() call, we return the length used. */
209 /* FIXME: This should be in free_vcpu */
210 free_page(lg->cpus[0].regs_page);
214 mutex_unlock(&lguest_lock);
218 /*L:010 The first operation the Launcher does must be a write. All writes
219 * start with an unsigned long number: for the first write this must be
220 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use
221 * writes of other values to send interrupts. */
222 static ssize_t write(struct file *file, const char __user *in,
223 size_t size, loff_t *off)
225 /* Once the guest is initialized, we hold the "struct lguest" in the
226 * file private data. */
227 struct lguest *lg = file->private_data;
228 const unsigned long __user *input = (const unsigned long __user *)in;
230 struct lg_cpu *uninitialized_var(cpu);
231 unsigned int cpu_id = *off;
233 if (get_user(req, input) != 0)
237 /* If you haven't initialized, you must do that first. */
238 if (req != LHREQ_INITIALIZE) {
239 if (!lg || (cpu_id >= lg->nr_cpus))
241 cpu = &lg->cpus[cpu_id];
245 /* Once the Guest is dead, you can only read() why it died. */
249 /* If you're not the task which owns the Guest, all you can do
250 * is break the Launcher out of running the Guest. */
251 if (current != cpu->tsk && req != LHREQ_BREAK)
256 case LHREQ_INITIALIZE:
257 return initialize(file, input);
259 return user_send_irq(cpu, input);
261 return break_guest_out(cpu, input);
267 /*L:060 The final piece of interface code is the close() routine. It reverses
268 * everything done in initialize(). This is usually called because the
271 * Note that the close routine returns 0 or a negative error number: it can't
272 * really fail, but it can whine. I blame Sun for this wart, and K&R C for
273 * letting them do it. :*/
274 static int close(struct inode *inode, struct file *file)
276 struct lguest *lg = file->private_data;
279 /* If we never successfully initialized, there's nothing to clean up */
283 /* We need the big lock, to protect from inter-guest I/O and other
284 * Launchers initializing guests. */
285 mutex_lock(&lguest_lock);
287 /* Free up the shadow page tables for the Guest. */
288 free_guest_pagetable(lg);
290 for (i = 0; i < lg->nr_cpus; i++) {
291 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
292 hrtimer_cancel(&lg->cpus[i].hrt);
293 /* We can free up the register page we allocated. */
294 free_page(lg->cpus[i].regs_page);
295 /* Now all the memory cleanups are done, it's safe to release
296 * the Launcher's memory management structure. */
297 mmput(lg->cpus[i].mm);
299 /* If lg->dead doesn't contain an error code it will be NULL or a
300 * kmalloc()ed string, either of which is ok to hand to kfree(). */
301 if (!IS_ERR(lg->dead))
303 /* We clear the entire structure, which also marks it as free for the
305 memset(lg, 0, sizeof(*lg));
306 /* Release lock and exit. */
307 mutex_unlock(&lguest_lock);
313 * Welcome to our journey through the Launcher!
315 * The Launcher is the Host userspace program which sets up, runs and services
316 * the Guest. In fact, many comments in the Drivers which refer to "the Host"
317 * doing things are inaccurate: the Launcher does all the device handling for
318 * the Guest, but the Guest can't know that.
320 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
321 * shall see more of that later.
323 * We begin our understanding with the Host kernel interface which the Launcher
324 * uses: reading and writing a character device called /dev/lguest. All the
325 * work happens in the read(), write() and close() routines: */
326 static struct file_operations lguest_fops = {
327 .owner = THIS_MODULE,
333 /* This is a textbook example of a "misc" character device. Populate a "struct
334 * miscdevice" and register it with misc_register(). */
335 static struct miscdevice lguest_dev = {
336 .minor = MISC_DYNAMIC_MINOR,
338 .fops = &lguest_fops,
341 int __init lguest_device_init(void)
343 return misc_register(&lguest_dev);
346 void __exit lguest_device_remove(void)
348 misc_deregister(&lguest_dev);