2 * Lguest guests use a very simple method to describe devices. It's a
3 * series of device descriptors contained just above the top of normal Guest
6 * We use the standard "virtio" device infrastructure, which provides us with a
7 * console, a network and a block driver. Each one expects some configuration
8 * information and a "virtqueue" or two to send and receive data.
10 #include <linux/init.h>
11 #include <linux/bootmem.h>
12 #include <linux/lguest_launcher.h>
13 #include <linux/virtio.h>
14 #include <linux/virtio_config.h>
15 #include <linux/interrupt.h>
16 #include <linux/virtio_ring.h>
17 #include <linux/err.h>
19 #include <asm/paravirt.h>
20 #include <asm/lguest_hcall.h>
22 /* The pointer to our (page) of device descriptions. */
23 static void *lguest_devices
;
26 * For Guests, device memory can be used as normal memory, so we cast away the
27 * __iomem to quieten sparse.
29 static inline void *lguest_map(unsigned long phys_addr
, unsigned long pages
)
31 return (__force
void *)ioremap_cache(phys_addr
, PAGE_SIZE
*pages
);
34 static inline void lguest_unmap(void *addr
)
36 iounmap((__force
void __iomem
*)addr
);
40 * Each lguest device is just a virtio device plus a pointer to its entry
41 * in the lguest_devices page.
43 struct lguest_device
{
44 struct virtio_device vdev
;
46 /* The entry in the lguest_devices page for this device. */
47 struct lguest_device_desc
*desc
;
51 * Since the virtio infrastructure hands us a pointer to the virtio_device all
52 * the time, it helps to have a curt macro to get a pointer to the struct
53 * lguest_device it's enclosed in.
55 #define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
58 * Device configurations
60 * The configuration information for a device consists of one or more
61 * virtqueues, a feature bitmap, and some configuration bytes. The
62 * configuration bytes don't really matter to us: the Launcher sets them up, and
63 * the driver will look at them during setup.
65 * A convenient routine to return the device's virtqueue config array:
66 * immediately after the descriptor.
68 static struct lguest_vqconfig
*lg_vq(const struct lguest_device_desc
*desc
)
70 return (void *)(desc
+ 1);
73 /* The features come immediately after the virtqueues. */
74 static u8
*lg_features(const struct lguest_device_desc
*desc
)
76 return (void *)(lg_vq(desc
) + desc
->num_vq
);
79 /* The config space comes after the two feature bitmasks. */
80 static u8
*lg_config(const struct lguest_device_desc
*desc
)
82 return lg_features(desc
) + desc
->feature_len
* 2;
85 /* The total size of the config page used by this device (incl. desc) */
86 static unsigned desc_size(const struct lguest_device_desc
*desc
)
89 + desc
->num_vq
* sizeof(struct lguest_vqconfig
)
90 + desc
->feature_len
* 2
94 /* This gets the device's feature bits. */
95 static u32
lg_get_features(struct virtio_device
*vdev
)
99 struct lguest_device_desc
*desc
= to_lgdev(vdev
)->desc
;
100 u8
*in_features
= lg_features(desc
);
102 /* We do this the slow but generic way. */
103 for (i
= 0; i
< min(desc
->feature_len
* 8, 32); i
++)
104 if (in_features
[i
/ 8] & (1 << (i
% 8)))
105 features
|= (1 << i
);
111 * The virtio core takes the features the Host offers, and copies the ones
112 * supported by the driver into the vdev->features array. Once that's all
113 * sorted out, this routine is called so we can tell the Host which features we
114 * understand and accept.
116 static void lg_finalize_features(struct virtio_device
*vdev
)
118 unsigned int i
, bits
;
119 struct lguest_device_desc
*desc
= to_lgdev(vdev
)->desc
;
120 /* Second half of bitmap is features we accept. */
121 u8
*out_features
= lg_features(desc
) + desc
->feature_len
;
123 /* Give virtio_ring a chance to accept features. */
124 vring_transport_features(vdev
);
127 * The vdev->feature array is a Linux bitmask: this isn't the same as a
128 * the simple array of bits used by lguest devices for features. So we
129 * do this slow, manual conversion which is completely general.
131 memset(out_features
, 0, desc
->feature_len
);
132 bits
= min_t(unsigned, desc
->feature_len
, sizeof(vdev
->features
)) * 8;
133 for (i
= 0; i
< bits
; i
++) {
134 if (test_bit(i
, vdev
->features
))
135 out_features
[i
/ 8] |= (1 << (i
% 8));
139 /* Once they've found a field, getting a copy of it is easy. */
140 static void lg_get(struct virtio_device
*vdev
, unsigned int offset
,
141 void *buf
, unsigned len
)
143 struct lguest_device_desc
*desc
= to_lgdev(vdev
)->desc
;
145 /* Check they didn't ask for more than the length of the config! */
146 BUG_ON(offset
+ len
> desc
->config_len
);
147 memcpy(buf
, lg_config(desc
) + offset
, len
);
150 /* Setting the contents is also trivial. */
151 static void lg_set(struct virtio_device
*vdev
, unsigned int offset
,
152 const void *buf
, unsigned len
)
154 struct lguest_device_desc
*desc
= to_lgdev(vdev
)->desc
;
156 /* Check they didn't ask for more than the length of the config! */
157 BUG_ON(offset
+ len
> desc
->config_len
);
158 memcpy(lg_config(desc
) + offset
, buf
, len
);
162 * The operations to get and set the status word just access the status field
163 * of the device descriptor.
165 static u8
lg_get_status(struct virtio_device
*vdev
)
167 return to_lgdev(vdev
)->desc
->status
;
171 * To notify on status updates, we (ab)use the NOTIFY hypercall, with the
172 * descriptor address of the device. A zero status means "reset".
174 static void set_status(struct virtio_device
*vdev
, u8 status
)
176 unsigned long offset
= (void *)to_lgdev(vdev
)->desc
- lguest_devices
;
178 /* We set the status. */
179 to_lgdev(vdev
)->desc
->status
= status
;
180 kvm_hypercall1(LHCALL_NOTIFY
, (max_pfn
<< PAGE_SHIFT
) + offset
);
183 static void lg_set_status(struct virtio_device
*vdev
, u8 status
)
186 set_status(vdev
, status
);
189 static void lg_reset(struct virtio_device
*vdev
)
197 * The other piece of infrastructure virtio needs is a "virtqueue": a way of
198 * the Guest device registering buffers for the other side to read from or
199 * write into (ie. send and receive buffers). Each device can have multiple
200 * virtqueues: for example the console driver uses one queue for sending and
201 * another for receiving.
203 * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
204 * already exists in virtio_ring.c. We just need to connect it up.
206 * We start with the information we need to keep about each virtqueue.
209 /*D:140 This is the information we remember about each virtqueue. */
210 struct lguest_vq_info
{
211 /* A copy of the information contained in the device config. */
212 struct lguest_vqconfig config
;
214 /* The address where we mapped the virtio ring, so we can unmap it. */
219 * When the virtio_ring code wants to prod the Host, it calls us here and we
220 * make a hypercall. We hand the physical address of the virtqueue so the Host
221 * knows which virtqueue we're talking about.
223 static void lg_notify(struct virtqueue
*vq
)
226 * We store our virtqueue information in the "priv" pointer of the
227 * virtqueue structure.
229 struct lguest_vq_info
*lvq
= vq
->priv
;
231 kvm_hypercall1(LHCALL_NOTIFY
, lvq
->config
.pfn
<< PAGE_SHIFT
);
234 /* An extern declaration inside a C file is bad form. Don't do it. */
235 extern void lguest_setup_irq(unsigned int irq
);
238 * This routine finds the Nth virtqueue described in the configuration of
239 * this device and sets it up.
241 * This is kind of an ugly duckling. It'd be nicer to have a standard
242 * representation of a virtqueue in the configuration space, but it seems that
243 * everyone wants to do it differently. The KVM coders want the Guest to
244 * allocate its own pages and tell the Host where they are, but for lguest it's
245 * simpler for the Host to simply tell us where the pages are.
247 static struct virtqueue
*lg_find_vq(struct virtio_device
*vdev
,
249 void (*callback
)(struct virtqueue
*vq
),
252 struct lguest_device
*ldev
= to_lgdev(vdev
);
253 struct lguest_vq_info
*lvq
;
254 struct virtqueue
*vq
;
257 /* We must have this many virtqueues. */
258 if (index
>= ldev
->desc
->num_vq
)
259 return ERR_PTR(-ENOENT
);
261 lvq
= kmalloc(sizeof(*lvq
), GFP_KERNEL
);
263 return ERR_PTR(-ENOMEM
);
266 * Make a copy of the "struct lguest_vqconfig" entry, which sits after
267 * the descriptor. We need a copy because the config space might not
268 * be aligned correctly.
270 memcpy(&lvq
->config
, lg_vq(ldev
->desc
)+index
, sizeof(lvq
->config
));
272 printk("Mapping virtqueue %i addr %lx\n", index
,
273 (unsigned long)lvq
->config
.pfn
<< PAGE_SHIFT
);
274 /* Figure out how many pages the ring will take, and map that memory */
275 lvq
->pages
= lguest_map((unsigned long)lvq
->config
.pfn
<< PAGE_SHIFT
,
276 DIV_ROUND_UP(vring_size(lvq
->config
.num
,
285 * OK, tell virtio_ring.c to set up a virtqueue now we know its size
286 * and we've got a pointer to its pages.
288 vq
= vring_new_virtqueue(lvq
->config
.num
, LGUEST_VRING_ALIGN
,
289 vdev
, lvq
->pages
, lg_notify
, callback
, name
);
295 /* Make sure the interrupt is allocated. */
296 lguest_setup_irq(lvq
->config
.irq
);
299 * Tell the interrupt for this virtqueue to go to the virtio_ring
302 * FIXME: We used to have a flag for the Host to tell us we could use
303 * the interrupt as a source of randomness: it'd be nice to have that
306 err
= request_irq(lvq
->config
.irq
, vring_interrupt
, IRQF_SHARED
,
307 dev_name(&vdev
->dev
), vq
);
312 * Last of all we hook up our 'struct lguest_vq_info" to the
313 * virtqueue's priv pointer.
319 vring_del_virtqueue(vq
);
321 lguest_unmap(lvq
->pages
);
328 /* Cleaning up a virtqueue is easy */
329 static void lg_del_vq(struct virtqueue
*vq
)
331 struct lguest_vq_info
*lvq
= vq
->priv
;
333 /* Release the interrupt */
334 free_irq(lvq
->config
.irq
, vq
);
335 /* Tell virtio_ring.c to free the virtqueue. */
336 vring_del_virtqueue(vq
);
337 /* Unmap the pages containing the ring. */
338 lguest_unmap(lvq
->pages
);
339 /* Free our own queue information. */
343 static void lg_del_vqs(struct virtio_device
*vdev
)
345 struct virtqueue
*vq
, *n
;
347 list_for_each_entry_safe(vq
, n
, &vdev
->vqs
, list
)
351 static int lg_find_vqs(struct virtio_device
*vdev
, unsigned nvqs
,
352 struct virtqueue
*vqs
[],
353 vq_callback_t
*callbacks
[],
356 struct lguest_device
*ldev
= to_lgdev(vdev
);
359 /* We must have this many virtqueues. */
360 if (nvqs
> ldev
->desc
->num_vq
)
363 for (i
= 0; i
< nvqs
; ++i
) {
364 vqs
[i
] = lg_find_vq(vdev
, i
, callbacks
[i
], names
[i
]);
372 return PTR_ERR(vqs
[i
]);
375 /* The ops structure which hooks everything together. */
376 static struct virtio_config_ops lguest_config_ops
= {
377 .get_features
= lg_get_features
,
378 .finalize_features
= lg_finalize_features
,
381 .get_status
= lg_get_status
,
382 .set_status
= lg_set_status
,
384 .find_vqs
= lg_find_vqs
,
385 .del_vqs
= lg_del_vqs
,
389 * The root device for the lguest virtio devices. This makes them appear as
390 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2.
392 static struct device
*lguest_root
;
395 * This is the core of the lguest bus: actually adding a new device.
396 * It's a separate function because it's neater that way, and because an
397 * earlier version of the code supported hotplug and unplug. They were removed
398 * early on because they were never used.
400 * As Andrew Tridgell says, "Untested code is buggy code".
402 * It's worth reading this carefully: we start with a pointer to the new device
403 * descriptor in the "lguest_devices" page, and the offset into the device
404 * descriptor page so we can uniquely identify it if things go badly wrong.
406 static void add_lguest_device(struct lguest_device_desc
*d
,
409 struct lguest_device
*ldev
;
411 /* Start with zeroed memory; Linux's device layer counts on it. */
412 ldev
= kzalloc(sizeof(*ldev
), GFP_KERNEL
);
414 printk(KERN_EMERG
"Cannot allocate lguest dev %u type %u\n",
419 /* This devices' parent is the lguest/ dir. */
420 ldev
->vdev
.dev
.parent
= lguest_root
;
422 * The device type comes straight from the descriptor. There's also a
423 * device vendor field in the virtio_device struct, which we leave as
426 ldev
->vdev
.id
.device
= d
->type
;
428 * We have a simple set of routines for querying the device's
429 * configuration information and setting its status.
431 ldev
->vdev
.config
= &lguest_config_ops
;
432 /* And we remember the device's descriptor for lguest_config_ops. */
436 * register_virtio_device() sets up the generic fields for the struct
437 * virtio_device and calls device_register(). This makes the bus
438 * infrastructure look for a matching driver.
440 if (register_virtio_device(&ldev
->vdev
) != 0) {
441 printk(KERN_ERR
"Failed to register lguest dev %u type %u\n",
448 * scan_devices() simply iterates through the device page. The type 0 is
449 * reserved to mean "end of devices".
451 static void scan_devices(void)
454 struct lguest_device_desc
*d
;
456 /* We start at the page beginning, and skip over each entry. */
457 for (i
= 0; i
< PAGE_SIZE
; i
+= desc_size(d
)) {
458 d
= lguest_devices
+ i
;
460 /* Once we hit a zero, stop. */
464 printk("Device at %i has size %u\n", i
, desc_size(d
));
465 add_lguest_device(d
, i
);
470 * Fairly early in boot, lguest_devices_init() is called to set up the
471 * lguest device infrastructure. We check that we are a Guest by checking
472 * pv_info.name: there are other ways of checking, but this seems most
475 * So we can access the "struct lguest_device_desc"s easily, we map that memory
476 * and store the pointer in the global "lguest_devices". Then we register a
477 * root device from which all our devices will hang (this seems to be the
478 * correct sysfs incantation).
480 * Finally we call scan_devices() which adds all the devices found in the
481 * lguest_devices page.
483 static int __init
lguest_devices_init(void)
485 if (strcmp(pv_info
.name
, "lguest") != 0)
488 lguest_root
= root_device_register("lguest");
489 if (IS_ERR(lguest_root
))
490 panic("Could not register lguest root");
492 /* Devices are in a single page above top of "normal" mem */
493 lguest_devices
= lguest_map(max_pfn
<<PAGE_SHIFT
, 1);
498 /* We do this after core stuff, but before the drivers. */
499 postcore_initcall(lguest_devices_init
);
502 * At this point in the journey we used to now wade through the lguest
503 * devices themselves: net, block and console. Since they're all now virtio
504 * devices rather than lguest-specific, I've decided to ignore them. Mostly,
505 * they're kind of boring. But this does mean you'll never experience the
506 * thrill of reading the forbidden love scene buried deep in the block driver.
508 * "make Launcher" beckons, where we answer questions like "Where do Guests
509 * come from?", and "What do you do when someone asks for optimization?".