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>
10 #include <linux/eventfd.h>
11 #include <linux/file.h>
14 bool send_notify_to_eventfd(struct lg_cpu
*cpu
)
17 struct lg_eventfd_map
*map
;
19 /* lg->eventfds is RCU-protected */
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;
30 return cpu
->pending_notify
== 0;
33 static int add_eventfd(struct lguest
*lg
, unsigned long addr
, int fd
)
35 struct lg_eventfd_map
*new, *old
= lg
->eventfds
;
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),
47 /* First make identical copy. */
48 memcpy(new->map
, old
->map
, sizeof(old
->map
[0]) * old
->num
);
51 /* Now append new entry. */
52 new->map
[new->num
].addr
= addr
;
53 new->map
[new->num
].event
= eventfd_fget(fd
);
54 if (IS_ERR(new->map
[new->num
].event
)) {
56 return PTR_ERR(new->map
[new->num
].event
);
60 /* Now put new one in place. */
61 rcu_assign_pointer(lg
->eventfds
, new);
63 /* We're not in a big hurry. Wait until noone's looking at old
64 * version, then delete it. */
71 static int attach_eventfd(struct lguest
*lg
, const unsigned long __user
*input
)
73 unsigned long addr
, fd
;
76 if (get_user(addr
, input
) != 0)
79 if (get_user(fd
, input
) != 0)
82 mutex_lock(&lguest_lock
);
83 err
= add_eventfd(lg
, addr
, fd
);
84 mutex_unlock(&lguest_lock
);
89 /*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
90 * number to /dev/lguest. */
91 static int user_send_irq(struct lg_cpu
*cpu
, const unsigned long __user
*input
)
95 if (get_user(irq
, input
) != 0)
97 if (irq
>= LGUEST_IRQS
)
100 set_interrupt(cpu
, irq
);
104 /*L:040 Once our Guest is initialized, the Launcher makes it run by reading
105 * from /dev/lguest. */
106 static ssize_t
read(struct file
*file
, char __user
*user
, size_t size
,loff_t
*o
)
108 struct lguest
*lg
= file
->private_data
;
110 unsigned int cpu_id
= *o
;
112 /* You must write LHREQ_INITIALIZE first! */
116 /* Watch out for arbitrary vcpu indexes! */
117 if (cpu_id
>= lg
->nr_cpus
)
120 cpu
= &lg
->cpus
[cpu_id
];
122 /* If you're not the task which owns the Guest, go away. */
123 if (current
!= cpu
->tsk
)
126 /* If the Guest is already dead, we indicate why */
130 /* lg->dead either contains an error code, or a string. */
131 if (IS_ERR(lg
->dead
))
132 return PTR_ERR(lg
->dead
);
134 /* We can only return as much as the buffer they read with. */
135 len
= min(size
, strlen(lg
->dead
)+1);
136 if (copy_to_user(user
, lg
->dead
, len
) != 0)
141 /* If we returned from read() last time because the Guest sent I/O,
143 if (cpu
->pending_notify
)
144 cpu
->pending_notify
= 0;
146 /* Run the Guest until something interesting happens. */
147 return run_guest(cpu
, (unsigned long __user
*)user
);
150 /*L:025 This actually initializes a CPU. For the moment, a Guest is only
151 * uniprocessor, so "id" is always 0. */
152 static int lg_cpu_start(struct lg_cpu
*cpu
, unsigned id
, unsigned long start_ip
)
154 /* We have a limited number the number of CPUs in the lguest struct. */
155 if (id
>= ARRAY_SIZE(cpu
->lg
->cpus
))
158 /* Set up this CPU's id, and pointer back to the lguest struct. */
160 cpu
->lg
= container_of((cpu
- id
), struct lguest
, cpus
[0]);
163 /* Each CPU has a timer it can set. */
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
);
172 /* We actually put the registers at the bottom of the page. */
173 cpu
->regs
= (void *)cpu
->regs_page
+ PAGE_SIZE
- sizeof(*cpu
->regs
);
175 /* Now we initialize the Guest's registers, handing it the start
177 lguest_arch_setup_regs(cpu
, start_ip
);
179 /* We keep a pointer to the Launcher task (ie. current task) for when
180 * other Guests want to wake this one (eg. console input). */
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
);
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
;
192 /* No error == success. */
196 /*L:020 The initialization write supplies 3 pointer sized (32 or 64 bit)
197 * values (in addition to the LHREQ_INITIALIZE value). These are:
199 * base: The start of the Guest-physical memory inside the Launcher memory.
201 * pfnlimit: The highest (Guest-physical) page number the Guest should be
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.
205 * start: The first instruction to execute ("eip" in x86-speak).
207 static int initialize(struct file
*file
, const unsigned long __user
*input
)
209 /* "struct lguest" contains everything we (the Host) know about a
213 unsigned long args
[3];
215 /* We grab the Big Lguest lock, which protects against multiple
216 * simultaneous initializations. */
217 mutex_lock(&lguest_lock
);
218 /* You can't initialize twice! Close the device and start again... */
219 if (file
->private_data
) {
224 if (copy_from_user(args
, input
, sizeof(args
)) != 0) {
229 lg
= kzalloc(sizeof(*lg
), GFP_KERNEL
);
235 lg
->eventfds
= kmalloc(sizeof(*lg
->eventfds
), GFP_KERNEL
);
240 lg
->eventfds
->num
= 0;
242 /* Populate the easy fields of our "struct lguest" */
243 lg
->mem_base
= (void __user
*)args
[0];
244 lg
->pfn_limit
= args
[1];
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]);
251 /* Initialize the Guest's shadow page tables, using the toplevel
252 * address the Launcher gave us. This allocates memory, so can fail. */
253 err
= init_guest_pagetable(lg
);
257 /* We keep our "struct lguest" in the file's private_data. */
258 file
->private_data
= lg
;
260 mutex_unlock(&lguest_lock
);
262 /* And because this is a write() call, we return the length used. */
266 /* FIXME: This should be in free_vcpu */
267 free_page(lg
->cpus
[0].regs_page
);
273 mutex_unlock(&lguest_lock
);
277 /*L:010 The first operation the Launcher does must be a write. All writes
278 * start with an unsigned long number: for the first write this must be
279 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use
280 * writes of other values to send interrupts.
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
286 static ssize_t
write(struct file
*file
, const char __user
*in
,
287 size_t size
, loff_t
*off
)
289 /* Once the Guest is initialized, we hold the "struct lguest" in the
290 * file private data. */
291 struct lguest
*lg
= file
->private_data
;
292 const unsigned long __user
*input
= (const unsigned long __user
*)in
;
294 struct lg_cpu
*uninitialized_var(cpu
);
295 unsigned int cpu_id
= *off
;
297 /* The first value tells us what this request is. */
298 if (get_user(req
, input
) != 0)
302 /* If you haven't initialized, you must do that first. */
303 if (req
!= LHREQ_INITIALIZE
) {
304 if (!lg
|| (cpu_id
>= lg
->nr_cpus
))
306 cpu
= &lg
->cpus
[cpu_id
];
308 /* Once the Guest is dead, you can only read() why it died. */
314 case LHREQ_INITIALIZE
:
315 return initialize(file
, input
);
317 return user_send_irq(cpu
, input
);
319 return attach_eventfd(lg
, input
);
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
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. :*/
332 static int close(struct inode
*inode
, struct file
*file
)
334 struct lguest
*lg
= file
->private_data
;
337 /* If we never successfully initialized, there's nothing to clean up */
341 /* We need the big lock, to protect from inter-guest I/O and other
342 * Launchers initializing guests. */
343 mutex_lock(&lguest_lock
);
345 /* Free up the shadow page tables for the Guest. */
346 free_guest_pagetable(lg
);
348 for (i
= 0; i
< lg
->nr_cpus
; i
++) {
349 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
350 hrtimer_cancel(&lg
->cpus
[i
].hrt
);
351 /* We can free up the register page we allocated. */
352 free_page(lg
->cpus
[i
].regs_page
);
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
);
358 /* Release any eventfds they registered. */
359 for (i
= 0; i
< lg
->eventfds
->num
; i
++)
360 fput(lg
->eventfds
->map
[i
].event
);
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(). */
365 if (!IS_ERR(lg
->dead
))
367 /* Free the memory allocated to the lguest_struct */
369 /* Release lock and exit. */
370 mutex_unlock(&lguest_lock
);
376 * Welcome to our journey through the Launcher!
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
381 * the Guest, but the Guest can't know that.
383 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
384 * shall see more of that later.
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: */
389 static struct file_operations lguest_fops
= {
390 .owner
= THIS_MODULE
,
396 /* This is a textbook example of a "misc" character device. Populate a "struct
397 * miscdevice" and register it with misc_register(). */
398 static struct miscdevice lguest_dev
= {
399 .minor
= MISC_DYNAMIC_MINOR
,
401 .fops
= &lguest_fops
,
404 int __init
lguest_device_init(void)
406 return misc_register(&lguest_dev
);
409 void __exit
lguest_device_remove(void)
411 misc_deregister(&lguest_dev
);