1/*P:050
  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
  4 * memory.
  5 *
  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.
  9:*/
 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>
 18#include <linux/slab.h>
 19#include <asm/io.h>
 20#include <asm/paravirt.h>
 21#include <asm/lguest_hcall.h>
 22
 23/* The pointer to our (page) of device descriptions. */
 24static void *lguest_devices;
 25
 26/*
 27 * For Guests, device memory can be used as normal memory, so we cast away the
 28 * __iomem to quieten sparse.
 29 */
 30static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
 31{
 32	return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages);
 33}
 34
 35static inline void lguest_unmap(void *addr)
 36{
 37	iounmap((__force void __iomem *)addr);
 38}
 39
 40/*D:100
 41 * Each lguest device is just a virtio device plus a pointer to its entry
 42 * in the lguest_devices page.
 43 */
 44struct lguest_device {
 45	struct virtio_device vdev;
 46
 47	/* The entry in the lguest_devices page for this device. */
 48	struct lguest_device_desc *desc;
 49};
 50
 51/*
 52 * Since the virtio infrastructure hands us a pointer to the virtio_device all
 53 * the time, it helps to have a curt macro to get a pointer to the struct
 54 * lguest_device it's enclosed in.
 55 */
 56#define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
 57
 58/*D:130
 59 * Device configurations
 60 *
 61 * The configuration information for a device consists of one or more
 62 * virtqueues, a feature bitmap, and some configuration bytes.  The
 63 * configuration bytes don't really matter to us: the Launcher sets them up, and
 64 * the driver will look at them during setup.
 65 *
 66 * A convenient routine to return the device's virtqueue config array:
 67 * immediately after the descriptor.
 68 */
 69static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc)
 70{
 71	return (void *)(desc + 1);
 72}
 73
 74/* The features come immediately after the virtqueues. */
 75static u8 *lg_features(const struct lguest_device_desc *desc)
 76{
 77	return (void *)(lg_vq(desc) + desc->num_vq);
 78}
 79
 80/* The config space comes after the two feature bitmasks. */
 81static u8 *lg_config(const struct lguest_device_desc *desc)
 82{
 83	return lg_features(desc) + desc->feature_len * 2;
 84}
 85
 86/* The total size of the config page used by this device (incl. desc) */
 87static unsigned desc_size(const struct lguest_device_desc *desc)
 88{
 89	return sizeof(*desc)
 90		+ desc->num_vq * sizeof(struct lguest_vqconfig)
 91		+ desc->feature_len * 2
 92		+ desc->config_len;
 93}
 94
 95/* This gets the device's feature bits. */
 96static u32 lg_get_features(struct virtio_device *vdev)
 97{
 98	unsigned int i;
 99	u32 features = 0;
100	struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
101	u8 *in_features = lg_features(desc);
102
103	/* We do this the slow but generic way. */
104	for (i = 0; i < min(desc->feature_len * 8, 32); i++)
105		if (in_features[i / 8] & (1 << (i % 8)))
106			features |= (1 << i);
107
108	return features;
109}
110
111/*
112 * To notify on reset or feature finalization, we (ab)use the NOTIFY
113 * hypercall, with the descriptor address of the device.
114 */
115static void status_notify(struct virtio_device *vdev)
116{
117	unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
118
119	hcall(LHCALL_NOTIFY, (max_pfn << PAGE_SHIFT) + offset, 0, 0, 0);
120}
121
122/*
123 * The virtio core takes the features the Host offers, and copies the ones
124 * supported by the driver into the vdev->features array.  Once that's all
125 * sorted out, this routine is called so we can tell the Host which features we
126 * understand and accept.
127 */
128static void lg_finalize_features(struct virtio_device *vdev)
129{
130	unsigned int i, bits;
131	struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
132	/* Second half of bitmap is features we accept. */
133	u8 *out_features = lg_features(desc) + desc->feature_len;
134
135	/* Give virtio_ring a chance to accept features. */
136	vring_transport_features(vdev);
137
138	/*
139	 * The vdev->feature array is a Linux bitmask: this isn't the same as a
140	 * the simple array of bits used by lguest devices for features.  So we
141	 * do this slow, manual conversion which is completely general.
142	 */
143	memset(out_features, 0, desc->feature_len);
144	bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
145	for (i = 0; i < bits; i++) {
146		if (test_bit(i, vdev->features))
147			out_features[i / 8] |= (1 << (i % 8));
148	}
149
150	/* Tell Host we've finished with this device's feature negotiation */
151	status_notify(vdev);
152}
153
154/* Once they've found a field, getting a copy of it is easy. */
155static void lg_get(struct virtio_device *vdev, unsigned int offset,
156		   void *buf, unsigned len)
157{
158	struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
159
160	/* Check they didn't ask for more than the length of the config! */
161	BUG_ON(offset + len > desc->config_len);
162	memcpy(buf, lg_config(desc) + offset, len);
163}
164
165/* Setting the contents is also trivial. */
166static void lg_set(struct virtio_device *vdev, unsigned int offset,
167		   const void *buf, unsigned len)
168{
169	struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
170
171	/* Check they didn't ask for more than the length of the config! */
172	BUG_ON(offset + len > desc->config_len);
173	memcpy(lg_config(desc) + offset, buf, len);
174}
175
176/*
177 * The operations to get and set the status word just access the status field
178 * of the device descriptor.
179 */
180static u8 lg_get_status(struct virtio_device *vdev)
181{
182	return to_lgdev(vdev)->desc->status;
183}
184
185static void lg_set_status(struct virtio_device *vdev, u8 status)
186{
187	BUG_ON(!status);
188	to_lgdev(vdev)->desc->status = status;
189
190	/* Tell Host immediately if we failed. */
191	if (status & VIRTIO_CONFIG_S_FAILED)
192		status_notify(vdev);
193}
194
195static void lg_reset(struct virtio_device *vdev)
196{
197	/* 0 status means "reset" */
198	to_lgdev(vdev)->desc->status = 0;
199	status_notify(vdev);
200}
201
202/*
203 * Virtqueues
204 *
205 * The other piece of infrastructure virtio needs is a "virtqueue": a way of
206 * the Guest device registering buffers for the other side to read from or
207 * write into (ie. send and receive buffers).  Each device can have multiple
208 * virtqueues: for example the console driver uses one queue for sending and
209 * another for receiving.
210 *
211 * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
212 * already exists in virtio_ring.c.  We just need to connect it up.
213 *
214 * We start with the information we need to keep about each virtqueue.
215 */
216
217/*D:140 This is the information we remember about each virtqueue. */
218struct lguest_vq_info {
219	/* A copy of the information contained in the device config. */
220	struct lguest_vqconfig config;
221
222	/* The address where we mapped the virtio ring, so we can unmap it. */
223	void *pages;
224};
225
226/*
227 * When the virtio_ring code wants to prod the Host, it calls us here and we
228 * make a hypercall.  We hand the physical address of the virtqueue so the Host
229 * knows which virtqueue we're talking about.
230 */
231static void lg_notify(struct virtqueue *vq)
232{
233	/*
234	 * We store our virtqueue information in the "priv" pointer of the
235	 * virtqueue structure.
236	 */
237	struct lguest_vq_info *lvq = vq->priv;
238
239	hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0, 0);
240}
241
242/* An extern declaration inside a C file is bad form.  Don't do it. */
243extern void lguest_setup_irq(unsigned int irq);
244
245/*
246 * This routine finds the Nth virtqueue described in the configuration of
247 * this device and sets it up.
248 *
249 * This is kind of an ugly duckling.  It'd be nicer to have a standard
250 * representation of a virtqueue in the configuration space, but it seems that
251 * everyone wants to do it differently.  The KVM coders want the Guest to
252 * allocate its own pages and tell the Host where they are, but for lguest it's
253 * simpler for the Host to simply tell us where the pages are.
254 */
255static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
256				    unsigned index,
257				    void (*callback)(struct virtqueue *vq),
258				    const char *name)
259{
260	struct lguest_device *ldev = to_lgdev(vdev);
261	struct lguest_vq_info *lvq;
262	struct virtqueue *vq;
263	int err;
264
265	/* We must have this many virtqueues. */
266	if (index >= ldev->desc->num_vq)
267		return ERR_PTR(-ENOENT);
268
269	lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
270	if (!lvq)
271		return ERR_PTR(-ENOMEM);
272
273	/*
274	 * Make a copy of the "struct lguest_vqconfig" entry, which sits after
275	 * the descriptor.  We need a copy because the config space might not
276	 * be aligned correctly.
277	 */
278	memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
279
280	printk("Mapping virtqueue %i addr %lx\n", index,
281	       (unsigned long)lvq->config.pfn << PAGE_SHIFT);
282	/* Figure out how many pages the ring will take, and map that memory */
283	lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
284				DIV_ROUND_UP(vring_size(lvq->config.num,
285							LGUEST_VRING_ALIGN),
286					     PAGE_SIZE));
287	if (!lvq->pages) {
288		err = -ENOMEM;
289		goto free_lvq;
290	}
291
292	/*
293	 * OK, tell virtio_ring.c to set up a virtqueue now we know its size
294	 * and we've got a pointer to its pages.
295	 */
296	vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN,
297				 vdev, lvq->pages, lg_notify, callback, name);
298	if (!vq) {
299		err = -ENOMEM;
300		goto unmap;
301	}
302
303	/* Make sure the interrupt is allocated. */
304	lguest_setup_irq(lvq->config.irq);
305
306	/*
307	 * Tell the interrupt for this virtqueue to go to the virtio_ring
308	 * interrupt handler.
309	 *
310	 * FIXME: We used to have a flag for the Host to tell us we could use
311	 * the interrupt as a source of randomness: it'd be nice to have that
312	 * back.
313	 */
314	err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
315			  dev_name(&vdev->dev), vq);
316	if (err)
317		goto destroy_vring;
318
319	/*
320	 * Last of all we hook up our 'struct lguest_vq_info" to the
321	 * virtqueue's priv pointer.
322	 */
323	vq->priv = lvq;
324	return vq;
325
326destroy_vring:
327	vring_del_virtqueue(vq);
328unmap:
329	lguest_unmap(lvq->pages);
330free_lvq:
331	kfree(lvq);
332	return ERR_PTR(err);
333}
334/*:*/
335
336/* Cleaning up a virtqueue is easy */
337static void lg_del_vq(struct virtqueue *vq)
338{
339	struct lguest_vq_info *lvq = vq->priv;
340
341	/* Release the interrupt */
342	free_irq(lvq->config.irq, vq);
343	/* Tell virtio_ring.c to free the virtqueue. */
344	vring_del_virtqueue(vq);
345	/* Unmap the pages containing the ring. */
346	lguest_unmap(lvq->pages);
347	/* Free our own queue information. */
348	kfree(lvq);
349}
350
351static void lg_del_vqs(struct virtio_device *vdev)
352{
353	struct virtqueue *vq, *n;
354
355	list_for_each_entry_safe(vq, n, &vdev->vqs, list)
356		lg_del_vq(vq);
357}
358
359static int lg_find_vqs(struct virtio_device *vdev, unsigned nvqs,
360		       struct virtqueue *vqs[],
361		       vq_callback_t *callbacks[],
362		       const char *names[])
363{
364	struct lguest_device *ldev = to_lgdev(vdev);
365	int i;
366
367	/* We must have this many virtqueues. */
368	if (nvqs > ldev->desc->num_vq)
369		return -ENOENT;
370
371	for (i = 0; i < nvqs; ++i) {
372		vqs[i] = lg_find_vq(vdev, i, callbacks[i], names[i]);
373		if (IS_ERR(vqs[i]))
374			goto error;
375	}
376	return 0;
377
378error:
379	lg_del_vqs(vdev);
380	return PTR_ERR(vqs[i]);
381}
382
383/* The ops structure which hooks everything together. */
384static struct virtio_config_ops lguest_config_ops = {
385	.get_features = lg_get_features,
386	.finalize_features = lg_finalize_features,
387	.get = lg_get,
388	.set = lg_set,
389	.get_status = lg_get_status,
390	.set_status = lg_set_status,
391	.reset = lg_reset,
392	.find_vqs = lg_find_vqs,
393	.del_vqs = lg_del_vqs,
394};
395
396/*
397 * The root device for the lguest virtio devices.  This makes them appear as
398 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2.
399 */
400static struct device *lguest_root;
401
402/*D:120
403 * This is the core of the lguest bus: actually adding a new device.
404 * It's a separate function because it's neater that way, and because an
405 * earlier version of the code supported hotplug and unplug.  They were removed
406 * early on because they were never used.
407 *
408 * As Andrew Tridgell says, "Untested code is buggy code".
409 *
410 * It's worth reading this carefully: we start with a pointer to the new device
411 * descriptor in the "lguest_devices" page, and the offset into the device
412 * descriptor page so we can uniquely identify it if things go badly wrong.
413 */
414static void add_lguest_device(struct lguest_device_desc *d,
415			      unsigned int offset)
416{
417	struct lguest_device *ldev;
418
419	/* Start with zeroed memory; Linux's device layer counts on it. */
420	ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
421	if (!ldev) {
422		printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
423		       offset, d->type);
424		return;
425	}
426
427	/* This devices' parent is the lguest/ dir. */
428	ldev->vdev.dev.parent = lguest_root;
429	/*
430	 * The device type comes straight from the descriptor.  There's also a
431	 * device vendor field in the virtio_device struct, which we leave as
432	 * 0.
433	 */
434	ldev->vdev.id.device = d->type;
435	/*
436	 * We have a simple set of routines for querying the device's
437	 * configuration information and setting its status.
438	 */
439	ldev->vdev.config = &lguest_config_ops;
440	/* And we remember the device's descriptor for lguest_config_ops. */
441	ldev->desc = d;
442
443	/*
444	 * register_virtio_device() sets up the generic fields for the struct
445	 * virtio_device and calls device_register().  This makes the bus
446	 * infrastructure look for a matching driver.
447	 */
448	if (register_virtio_device(&ldev->vdev) != 0) {
449		printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
450		       offset, d->type);
451		kfree(ldev);
452	}
453}
454
455/*D:110
456 * scan_devices() simply iterates through the device page.  The type 0 is
457 * reserved to mean "end of devices".
458 */
459static void scan_devices(void)
460{
461	unsigned int i;
462	struct lguest_device_desc *d;
463
464	/* We start at the page beginning, and skip over each entry. */
465	for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
466		d = lguest_devices + i;
467
468		/* Once we hit a zero, stop. */
469		if (d->type == 0)
470			break;
471
472		printk("Device at %i has size %u\n", i, desc_size(d));
473		add_lguest_device(d, i);
474	}
475}
476
477/*D:105
478 * Fairly early in boot, lguest_devices_init() is called to set up the
479 * lguest device infrastructure.  We check that we are a Guest by checking
480 * pv_info.name: there are other ways of checking, but this seems most
481 * obvious to me.
482 *
483 * So we can access the "struct lguest_device_desc"s easily, we map that memory
484 * and store the pointer in the global "lguest_devices".  Then we register a
485 * root device from which all our devices will hang (this seems to be the
486 * correct sysfs incantation).
487 *
488 * Finally we call scan_devices() which adds all the devices found in the
489 * lguest_devices page.
490 */
491static int __init lguest_devices_init(void)
492{
493	if (strcmp(pv_info.name, "lguest") != 0)
494		return 0;
495
496	lguest_root = root_device_register("lguest");
497	if (IS_ERR(lguest_root))
498		panic("Could not register lguest root");
499
500	/* Devices are in a single page above top of "normal" mem */
501	lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
502
503	scan_devices();
504	return 0;
505}
506/* We do this after core stuff, but before the drivers. */
507postcore_initcall(lguest_devices_init);
508
509/*D:150
510 * At this point in the journey we used to now wade through the lguest
511 * devices themselves: net, block and console.  Since they're all now virtio
512 * devices rather than lguest-specific, I've decided to ignore them.  Mostly,
513 * they're kind of boring.  But this does mean you'll never experience the
514 * thrill of reading the forbidden love scene buried deep in the block driver.
515 *
516 * "make Launcher" beckons, where we answer questions like "Where do Guests
517 * come from?", and "What do you do when someone asks for optimization?".
518 */