1/*
2 * kvm_virtio.c - virtio for kvm on s390
3 *
4 * Copyright IBM Corp. 2008
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
9 *
10 *    Author(s): Christian Borntraeger <borntraeger@de.ibm.com>
11 */
12
13#include <linux/kernel_stat.h>
14#include <linux/init.h>
15#include <linux/bootmem.h>
16#include <linux/err.h>
17#include <linux/virtio.h>
18#include <linux/virtio_config.h>
19#include <linux/slab.h>
20#include <linux/virtio_console.h>
21#include <linux/interrupt.h>
22#include <linux/virtio_ring.h>
23#include <linux/export.h>
24#include <linux/pfn.h>
25#include <asm/io.h>
26#include <asm/kvm_para.h>
27#include <asm/kvm_virtio.h>
28#include <asm/setup.h>
29#include <asm/irq.h>
30
31#define VIRTIO_SUBCODE_64 0x0D00
32
33/*
34 * The pointer to our (page) of device descriptions.
35 */
36static void *kvm_devices;
37static struct work_struct hotplug_work;
38
39struct kvm_device {
40	struct virtio_device vdev;
41	struct kvm_device_desc *desc;
42};
43
44#define to_kvmdev(vd) container_of(vd, struct kvm_device, vdev)
45
46/*
47 * memory layout:
48 * - kvm_device_descriptor
49 *        struct kvm_device_desc
50 * - configuration
51 *        struct kvm_vqconfig
52 * - feature bits
53 * - config space
54 */
55static struct kvm_vqconfig *kvm_vq_config(const struct kvm_device_desc *desc)
56{
57	return (struct kvm_vqconfig *)(desc + 1);
58}
59
60static u8 *kvm_vq_features(const struct kvm_device_desc *desc)
61{
62	return (u8 *)(kvm_vq_config(desc) + desc->num_vq);
63}
64
65static u8 *kvm_vq_configspace(const struct kvm_device_desc *desc)
66{
67	return kvm_vq_features(desc) + desc->feature_len * 2;
68}
69
70/*
71 * The total size of the config page used by this device (incl. desc)
72 */
73static unsigned desc_size(const struct kvm_device_desc *desc)
74{
75	return sizeof(*desc)
76		+ desc->num_vq * sizeof(struct kvm_vqconfig)
77		+ desc->feature_len * 2
78		+ desc->config_len;
79}
80
81/* This gets the device's feature bits. */
82static u32 kvm_get_features(struct virtio_device *vdev)
83{
84	unsigned int i;
85	u32 features = 0;
86	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
87	u8 *in_features = kvm_vq_features(desc);
88
89	for (i = 0; i < min(desc->feature_len * 8, 32); i++)
90		if (in_features[i / 8] & (1 << (i % 8)))
91			features |= (1 << i);
92	return features;
93}
94
95static void kvm_finalize_features(struct virtio_device *vdev)
96{
97	unsigned int i, bits;
98	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
99	/* Second half of bitmap is features we accept. */
100	u8 *out_features = kvm_vq_features(desc) + desc->feature_len;
101
102	/* Give virtio_ring a chance to accept features. */
103	vring_transport_features(vdev);
104
105	memset(out_features, 0, desc->feature_len);
106	bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
107	for (i = 0; i < bits; i++) {
108		if (test_bit(i, vdev->features))
109			out_features[i / 8] |= (1 << (i % 8));
110	}
111}
112
113/*
114 * Reading and writing elements in config space
115 */
116static void kvm_get(struct virtio_device *vdev, unsigned int offset,
117		   void *buf, unsigned len)
118{
119	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
120
121	BUG_ON(offset + len > desc->config_len);
122	memcpy(buf, kvm_vq_configspace(desc) + offset, len);
123}
124
125static void kvm_set(struct virtio_device *vdev, unsigned int offset,
126		   const void *buf, unsigned len)
127{
128	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
129
130	BUG_ON(offset + len > desc->config_len);
131	memcpy(kvm_vq_configspace(desc) + offset, buf, len);
132}
133
134/*
135 * The operations to get and set the status word just access
136 * the status field of the device descriptor. set_status will also
137 * make a hypercall to the host, to tell about status changes
138 */
139static u8 kvm_get_status(struct virtio_device *vdev)
140{
141	return to_kvmdev(vdev)->desc->status;
142}
143
144static void kvm_set_status(struct virtio_device *vdev, u8 status)
145{
146	BUG_ON(!status);
147	to_kvmdev(vdev)->desc->status = status;
148	kvm_hypercall1(KVM_S390_VIRTIO_SET_STATUS,
149		       (unsigned long) to_kvmdev(vdev)->desc);
150}
151
152/*
153 * To reset the device, we use the KVM_VIRTIO_RESET hypercall, using the
154 * descriptor address. The Host will zero the status and all the
155 * features.
156 */
157static void kvm_reset(struct virtio_device *vdev)
158{
159	kvm_hypercall1(KVM_S390_VIRTIO_RESET,
160		       (unsigned long) to_kvmdev(vdev)->desc);
161}
162
163/*
164 * When the virtio_ring code wants to notify the Host, it calls us here and we
165 * make a hypercall.  We hand the address  of the virtqueue so the Host
166 * knows which virtqueue we're talking about.
167 */
168static void kvm_notify(struct virtqueue *vq)
169{
170	struct kvm_vqconfig *config = vq->priv;
171
172	kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, config->address);
173}
174
175/*
176 * This routine finds the first virtqueue described in the configuration of
177 * this device and sets it up.
178 */
179static struct virtqueue *kvm_find_vq(struct virtio_device *vdev,
180				     unsigned index,
181				     void (*callback)(struct virtqueue *vq),
182				     const char *name)
183{
184	struct kvm_device *kdev = to_kvmdev(vdev);
185	struct kvm_vqconfig *config;
186	struct virtqueue *vq;
187	int err;
188
189	if (index >= kdev->desc->num_vq)
190		return ERR_PTR(-ENOENT);
191
192	config = kvm_vq_config(kdev->desc)+index;
193
194	err = vmem_add_mapping(config->address,
195			       vring_size(config->num,
196					  KVM_S390_VIRTIO_RING_ALIGN));
197	if (err)
198		goto out;
199
200	vq = vring_new_virtqueue(config->num, KVM_S390_VIRTIO_RING_ALIGN,
201				 vdev, true, (void *) config->address,
202				 kvm_notify, callback, name);
203	if (!vq) {
204		err = -ENOMEM;
205		goto unmap;
206	}
207
208	/*
209	 * register a callback token
210	 * The host will sent this via the external interrupt parameter
211	 */
212	config->token = (u64) vq;
213
214	vq->priv = config;
215	return vq;
216unmap:
217	vmem_remove_mapping(config->address,
218			    vring_size(config->num,
219				       KVM_S390_VIRTIO_RING_ALIGN));
220out:
221	return ERR_PTR(err);
222}
223
224static void kvm_del_vq(struct virtqueue *vq)
225{
226	struct kvm_vqconfig *config = vq->priv;
227
228	vring_del_virtqueue(vq);
229	vmem_remove_mapping(config->address,
230			    vring_size(config->num,
231				       KVM_S390_VIRTIO_RING_ALIGN));
232}
233
234static void kvm_del_vqs(struct virtio_device *vdev)
235{
236	struct virtqueue *vq, *n;
237
238	list_for_each_entry_safe(vq, n, &vdev->vqs, list)
239		kvm_del_vq(vq);
240}
241
242static int kvm_find_vqs(struct virtio_device *vdev, unsigned nvqs,
243			struct virtqueue *vqs[],
244			vq_callback_t *callbacks[],
245			const char *names[])
246{
247	struct kvm_device *kdev = to_kvmdev(vdev);
248	int i;
249
250	/* We must have this many virtqueues. */
251	if (nvqs > kdev->desc->num_vq)
252		return -ENOENT;
253
254	for (i = 0; i < nvqs; ++i) {
255		vqs[i] = kvm_find_vq(vdev, i, callbacks[i], names[i]);
256		if (IS_ERR(vqs[i]))
257			goto error;
258	}
259	return 0;
260
261error:
262	kvm_del_vqs(vdev);
263	return PTR_ERR(vqs[i]);
264}
265
266static const char *kvm_bus_name(struct virtio_device *vdev)
267{
268	return "";
269}
270
271/*
272 * The config ops structure as defined by virtio config
273 */
274static struct virtio_config_ops kvm_vq_configspace_ops = {
275	.get_features = kvm_get_features,
276	.finalize_features = kvm_finalize_features,
277	.get = kvm_get,
278	.set = kvm_set,
279	.get_status = kvm_get_status,
280	.set_status = kvm_set_status,
281	.reset = kvm_reset,
282	.find_vqs = kvm_find_vqs,
283	.del_vqs = kvm_del_vqs,
284	.bus_name = kvm_bus_name,
285};
286
287/*
288 * The root device for the kvm virtio devices.
289 * This makes them appear as /sys/devices/kvm_s390/0,1,2 not /sys/devices/0,1,2.
290 */
291static struct device *kvm_root;
292
293/*
294 * adds a new device and register it with virtio
295 * appropriate drivers are loaded by the device model
296 */
297static void add_kvm_device(struct kvm_device_desc *d, unsigned int offset)
298{
299	struct kvm_device *kdev;
300
301	kdev = kzalloc(sizeof(*kdev), GFP_KERNEL);
302	if (!kdev) {
303		printk(KERN_EMERG "Cannot allocate kvm dev %u type %u\n",
304		       offset, d->type);
305		return;
306	}
307
308	kdev->vdev.dev.parent = kvm_root;
309	kdev->vdev.id.device = d->type;
310	kdev->vdev.config = &kvm_vq_configspace_ops;
311	kdev->desc = d;
312
313	if (register_virtio_device(&kdev->vdev) != 0) {
314		printk(KERN_ERR "Failed to register kvm device %u type %u\n",
315		       offset, d->type);
316		kfree(kdev);
317	}
318}
319
320/*
321 * scan_devices() simply iterates through the device page.
322 * The type 0 is reserved to mean "end of devices".
323 */
324static void scan_devices(void)
325{
326	unsigned int i;
327	struct kvm_device_desc *d;
328
329	for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
330		d = kvm_devices + i;
331
332		if (d->type == 0)
333			break;
334
335		add_kvm_device(d, i);
336	}
337}
338
339/*
340 * match for a kvm device with a specific desc pointer
341 */
342static int match_desc(struct device *dev, void *data)
343{
344	struct virtio_device *vdev = dev_to_virtio(dev);
345	struct kvm_device *kdev = to_kvmdev(vdev);
346
347	return kdev->desc == data;
348}
349
350/*
351 * hotplug_device tries to find changes in the device page.
352 */
353static void hotplug_devices(struct work_struct *dummy)
354{
355	unsigned int i;
356	struct kvm_device_desc *d;
357	struct device *dev;
358
359	for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
360		d = kvm_devices + i;
361
362		/* end of list */
363		if (d->type == 0)
364			break;
365
366		/* device already exists */
367		dev = device_find_child(kvm_root, d, match_desc);
368		if (dev) {
369			/* XXX check for hotplug remove */
370			put_device(dev);
371			continue;
372		}
373
374		/* new device */
375		printk(KERN_INFO "Adding new virtio device %p\n", d);
376		add_kvm_device(d, i);
377	}
378}
379
380/*
381 * we emulate the request_irq behaviour on top of s390 extints
382 */
383static void kvm_extint_handler(unsigned int ext_int_code,
384			       unsigned int param32, unsigned long param64)
385{
386	struct virtqueue *vq;
387	u16 subcode;
388	u32 param;
389
390	subcode = ext_int_code >> 16;
391	if ((subcode & 0xff00) != VIRTIO_SUBCODE_64)
392		return;
393	kstat_cpu(smp_processor_id()).irqs[EXTINT_VRT]++;
394
395	/* The LSB might be overloaded, we have to mask it */
396	vq = (struct virtqueue *)(param64 & ~1UL);
397
398	/* We use ext_params to decide what this interrupt means */
399	param = param32 & VIRTIO_PARAM_MASK;
400
401	switch (param) {
402	case VIRTIO_PARAM_CONFIG_CHANGED:
403	{
404		struct virtio_driver *drv;
405		drv = container_of(vq->vdev->dev.driver,
406				   struct virtio_driver, driver);
407		if (drv->config_changed)
408			drv->config_changed(vq->vdev);
409
410		break;
411	}
412	case VIRTIO_PARAM_DEV_ADD:
413		schedule_work(&hotplug_work);
414		break;
415	case VIRTIO_PARAM_VRING_INTERRUPT:
416	default:
417		vring_interrupt(0, vq);
418		break;
419	}
420}
421
422/*
423 * Init function for virtio
424 * devices are in a single page above top of "normal" mem
425 */
426static int __init kvm_devices_init(void)
427{
428	int rc;
429
430	if (!MACHINE_IS_KVM)
431		return -ENODEV;
432
433	kvm_root = root_device_register("kvm_s390");
434	if (IS_ERR(kvm_root)) {
435		rc = PTR_ERR(kvm_root);
436		printk(KERN_ERR "Could not register kvm_s390 root device");
437		return rc;
438	}
439
440	rc = vmem_add_mapping(real_memory_size, PAGE_SIZE);
441	if (rc) {
442		root_device_unregister(kvm_root);
443		return rc;
444	}
445
446	kvm_devices = (void *) real_memory_size;
447
448	INIT_WORK(&hotplug_work, hotplug_devices);
449
450	service_subclass_irq_register();
451	register_external_interrupt(0x2603, kvm_extint_handler);
452
453	scan_devices();
454	return 0;
455}
456
457/* code for early console output with virtio_console */
458static __init int early_put_chars(u32 vtermno, const char *buf, int count)
459{
460	char scratch[17];
461	unsigned int len = count;
462
463	if (len > sizeof(scratch) - 1)
464		len = sizeof(scratch) - 1;
465	scratch[len] = '\0';
466	memcpy(scratch, buf, len);
467	kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, __pa(scratch));
468	return len;
469}
470
471static int __init s390_virtio_console_init(void)
472{
473	if (!MACHINE_IS_KVM)
474		return -ENODEV;
475	return virtio_cons_early_init(early_put_chars);
476}
477console_initcall(s390_virtio_console_init);
478
479
480/*
481 * We do this after core stuff, but before the drivers.
482 */
483postcore_initcall(kvm_devices_init);
484