1/*
2 * 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/sclp.h>
29#include <asm/setup.h>
30#include <asm/irq.h>
31
32#define VIRTIO_SUBCODE_64 0x0D00
33
34/*
35 * The pointer to our (page) of device descriptions.
36 */
37static void *kvm_devices;
38static struct work_struct hotplug_work;
39
40struct kvm_device {
41	struct virtio_device vdev;
42	struct kvm_device_desc *desc;
43};
44
45#define to_kvmdev(vd) container_of(vd, struct kvm_device, vdev)
46
47/*
48 * memory layout:
49 * - kvm_device_descriptor
50 *        struct kvm_device_desc
51 * - configuration
52 *        struct kvm_vqconfig
53 * - feature bits
54 * - config space
55 */
56static struct kvm_vqconfig *kvm_vq_config(const struct kvm_device_desc *desc)
57{
58	return (struct kvm_vqconfig *)(desc + 1);
59}
60
61static u8 *kvm_vq_features(const struct kvm_device_desc *desc)
62{
63	return (u8 *)(kvm_vq_config(desc) + desc->num_vq);
64}
65
66static u8 *kvm_vq_configspace(const struct kvm_device_desc *desc)
67{
68	return kvm_vq_features(desc) + desc->feature_len * 2;
69}
70
71/*
72 * The total size of the config page used by this device (incl. desc)
73 */
74static unsigned desc_size(const struct kvm_device_desc *desc)
75{
76	return sizeof(*desc)
77		+ desc->num_vq * sizeof(struct kvm_vqconfig)
78		+ desc->feature_len * 2
79		+ desc->config_len;
80}
81
82/* This gets the device's feature bits. */
83static u64 kvm_get_features(struct virtio_device *vdev)
84{
85	unsigned int i;
86	u32 features = 0;
87	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
88	u8 *in_features = kvm_vq_features(desc);
89
90	for (i = 0; i < min(desc->feature_len * 8, 32); i++)
91		if (in_features[i / 8] & (1 << (i % 8)))
92			features |= (1 << i);
93	return features;
94}
95
96static int kvm_finalize_features(struct virtio_device *vdev)
97{
98	unsigned int i, bits;
99	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
100	/* Second half of bitmap is features we accept. */
101	u8 *out_features = kvm_vq_features(desc) + desc->feature_len;
102
103	/* Give virtio_ring a chance to accept features. */
104	vring_transport_features(vdev);
105
106	/* Make sure we don't have any features > 32 bits! */
107	BUG_ON((u32)vdev->features != vdev->features);
108
109	memset(out_features, 0, desc->feature_len);
110	bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
111	for (i = 0; i < bits; i++) {
112		if (__virtio_test_bit(vdev, i))
113			out_features[i / 8] |= (1 << (i % 8));
114	}
115
116	return 0;
117}
118
119/*
120 * Reading and writing elements in config space
121 */
122static void kvm_get(struct virtio_device *vdev, unsigned int offset,
123		   void *buf, unsigned len)
124{
125	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
126
127	BUG_ON(offset + len > desc->config_len);
128	memcpy(buf, kvm_vq_configspace(desc) + offset, len);
129}
130
131static void kvm_set(struct virtio_device *vdev, unsigned int offset,
132		   const void *buf, unsigned len)
133{
134	struct kvm_device_desc *desc = to_kvmdev(vdev)->desc;
135
136	BUG_ON(offset + len > desc->config_len);
137	memcpy(kvm_vq_configspace(desc) + offset, buf, len);
138}
139
140/*
141 * The operations to get and set the status word just access
142 * the status field of the device descriptor. set_status will also
143 * make a hypercall to the host, to tell about status changes
144 */
145static u8 kvm_get_status(struct virtio_device *vdev)
146{
147	return to_kvmdev(vdev)->desc->status;
148}
149
150static void kvm_set_status(struct virtio_device *vdev, u8 status)
151{
152	BUG_ON(!status);
153	to_kvmdev(vdev)->desc->status = status;
154	kvm_hypercall1(KVM_S390_VIRTIO_SET_STATUS,
155		       (unsigned long) to_kvmdev(vdev)->desc);
156}
157
158/*
159 * To reset the device, we use the KVM_VIRTIO_RESET hypercall, using the
160 * descriptor address. The Host will zero the status and all the
161 * features.
162 */
163static void kvm_reset(struct virtio_device *vdev)
164{
165	kvm_hypercall1(KVM_S390_VIRTIO_RESET,
166		       (unsigned long) to_kvmdev(vdev)->desc);
167}
168
169/*
170 * When the virtio_ring code wants to notify the Host, it calls us here and we
171 * make a hypercall.  We hand the address  of the virtqueue so the Host
172 * knows which virtqueue we're talking about.
173 */
174static bool kvm_notify(struct virtqueue *vq)
175{
176	long rc;
177	struct kvm_vqconfig *config = vq->priv;
178
179	rc = kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, config->address);
180	if (rc < 0)
181		return false;
182	return true;
183}
184
185/*
186 * This routine finds the first virtqueue described in the configuration of
187 * this device and sets it up.
188 */
189static struct virtqueue *kvm_find_vq(struct virtio_device *vdev,
190				     unsigned index,
191				     void (*callback)(struct virtqueue *vq),
192				     const char *name)
193{
194	struct kvm_device *kdev = to_kvmdev(vdev);
195	struct kvm_vqconfig *config;
196	struct virtqueue *vq;
197	int err;
198
199	if (index >= kdev->desc->num_vq)
200		return ERR_PTR(-ENOENT);
201
202	if (!name)
203		return NULL;
204
205	config = kvm_vq_config(kdev->desc)+index;
206
207	err = vmem_add_mapping(config->address,
208			       vring_size(config->num,
209					  KVM_S390_VIRTIO_RING_ALIGN));
210	if (err)
211		goto out;
212
213	vq = vring_new_virtqueue(index, config->num, KVM_S390_VIRTIO_RING_ALIGN,
214				 vdev, true, (void *) config->address,
215				 kvm_notify, callback, name);
216	if (!vq) {
217		err = -ENOMEM;
218		goto unmap;
219	}
220
221	/*
222	 * register a callback token
223	 * The host will sent this via the external interrupt parameter
224	 */
225	config->token = (u64) vq;
226
227	vq->priv = config;
228	return vq;
229unmap:
230	vmem_remove_mapping(config->address,
231			    vring_size(config->num,
232				       KVM_S390_VIRTIO_RING_ALIGN));
233out:
234	return ERR_PTR(err);
235}
236
237static void kvm_del_vq(struct virtqueue *vq)
238{
239	struct kvm_vqconfig *config = vq->priv;
240
241	vring_del_virtqueue(vq);
242	vmem_remove_mapping(config->address,
243			    vring_size(config->num,
244				       KVM_S390_VIRTIO_RING_ALIGN));
245}
246
247static void kvm_del_vqs(struct virtio_device *vdev)
248{
249	struct virtqueue *vq, *n;
250
251	list_for_each_entry_safe(vq, n, &vdev->vqs, list)
252		kvm_del_vq(vq);
253}
254
255static int kvm_find_vqs(struct virtio_device *vdev, unsigned nvqs,
256			struct virtqueue *vqs[],
257			vq_callback_t *callbacks[],
258			const char *names[])
259{
260	struct kvm_device *kdev = to_kvmdev(vdev);
261	int i;
262
263	/* We must have this many virtqueues. */
264	if (nvqs > kdev->desc->num_vq)
265		return -ENOENT;
266
267	for (i = 0; i < nvqs; ++i) {
268		vqs[i] = kvm_find_vq(vdev, i, callbacks[i], names[i]);
269		if (IS_ERR(vqs[i]))
270			goto error;
271	}
272	return 0;
273
274error:
275	kvm_del_vqs(vdev);
276	return PTR_ERR(vqs[i]);
277}
278
279static const char *kvm_bus_name(struct virtio_device *vdev)
280{
281	return "";
282}
283
284/*
285 * The config ops structure as defined by virtio config
286 */
287static const struct virtio_config_ops kvm_vq_configspace_ops = {
288	.get_features = kvm_get_features,
289	.finalize_features = kvm_finalize_features,
290	.get = kvm_get,
291	.set = kvm_set,
292	.get_status = kvm_get_status,
293	.set_status = kvm_set_status,
294	.reset = kvm_reset,
295	.find_vqs = kvm_find_vqs,
296	.del_vqs = kvm_del_vqs,
297	.bus_name = kvm_bus_name,
298};
299
300/*
301 * The root device for the kvm virtio devices.
302 * This makes them appear as /sys/devices/kvm_s390/0,1,2 not /sys/devices/0,1,2.
303 */
304static struct device *kvm_root;
305
306/*
307 * adds a new device and register it with virtio
308 * appropriate drivers are loaded by the device model
309 */
310static void add_kvm_device(struct kvm_device_desc *d, unsigned int offset)
311{
312	struct kvm_device *kdev;
313
314	kdev = kzalloc(sizeof(*kdev), GFP_KERNEL);
315	if (!kdev) {
316		printk(KERN_EMERG "Cannot allocate kvm dev %u type %u\n",
317		       offset, d->type);
318		return;
319	}
320
321	kdev->vdev.dev.parent = kvm_root;
322	kdev->vdev.id.device = d->type;
323	kdev->vdev.config = &kvm_vq_configspace_ops;
324	kdev->desc = d;
325
326	if (register_virtio_device(&kdev->vdev) != 0) {
327		printk(KERN_ERR "Failed to register kvm device %u type %u\n",
328		       offset, d->type);
329		kfree(kdev);
330	}
331}
332
333/*
334 * scan_devices() simply iterates through the device page.
335 * The type 0 is reserved to mean "end of devices".
336 */
337static void scan_devices(void)
338{
339	unsigned int i;
340	struct kvm_device_desc *d;
341
342	for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
343		d = kvm_devices + i;
344
345		if (d->type == 0)
346			break;
347
348		add_kvm_device(d, i);
349	}
350}
351
352/*
353 * match for a kvm device with a specific desc pointer
354 */
355static int match_desc(struct device *dev, void *data)
356{
357	struct virtio_device *vdev = dev_to_virtio(dev);
358	struct kvm_device *kdev = to_kvmdev(vdev);
359
360	return kdev->desc == data;
361}
362
363/*
364 * hotplug_device tries to find changes in the device page.
365 */
366static void hotplug_devices(struct work_struct *dummy)
367{
368	unsigned int i;
369	struct kvm_device_desc *d;
370	struct device *dev;
371
372	for (i = 0; i < PAGE_SIZE; i += desc_size(d)) {
373		d = kvm_devices + i;
374
375		/* end of list */
376		if (d->type == 0)
377			break;
378
379		/* device already exists */
380		dev = device_find_child(kvm_root, d, match_desc);
381		if (dev) {
382			/* XXX check for hotplug remove */
383			put_device(dev);
384			continue;
385		}
386
387		/* new device */
388		printk(KERN_INFO "Adding new virtio device %p\n", d);
389		add_kvm_device(d, i);
390	}
391}
392
393/*
394 * we emulate the request_irq behaviour on top of s390 extints
395 */
396static void kvm_extint_handler(struct ext_code ext_code,
397			       unsigned int param32, unsigned long param64)
398{
399	struct virtqueue *vq;
400	u32 param;
401
402	if ((ext_code.subcode & 0xff00) != VIRTIO_SUBCODE_64)
403		return;
404	inc_irq_stat(IRQEXT_VRT);
405
406	/* The LSB might be overloaded, we have to mask it */
407	vq = (struct virtqueue *)(param64 & ~1UL);
408
409	/* We use ext_params to decide what this interrupt means */
410	param = param32 & VIRTIO_PARAM_MASK;
411
412	switch (param) {
413	case VIRTIO_PARAM_CONFIG_CHANGED:
414		virtio_config_changed(vq->vdev);
415		break;
416	case VIRTIO_PARAM_DEV_ADD:
417		schedule_work(&hotplug_work);
418		break;
419	case VIRTIO_PARAM_VRING_INTERRUPT:
420	default:
421		vring_interrupt(0, vq);
422		break;
423	}
424}
425
426/*
427 * For s390-virtio, we expect a page above main storage containing
428 * the virtio configuration. Try to actually load from this area
429 * in order to figure out if the host provides this page.
430 */
431static int __init test_devices_support(unsigned long addr)
432{
433	int ret = -EIO;
434
435	asm volatile(
436		"0:	lura	0,%1\n"
437		"1:	xgr	%0,%0\n"
438		"2:\n"
439		EX_TABLE(0b,2b)
440		EX_TABLE(1b,2b)
441		: "+d" (ret)
442		: "a" (addr)
443		: "0", "cc");
444	return ret;
445}
446/*
447 * Init function for virtio
448 * devices are in a single page above top of "normal" + standby mem
449 */
450static int __init kvm_devices_init(void)
451{
452	int rc;
453	unsigned long total_memory_size = sclp_get_rzm() * sclp_get_rnmax();
454
455	if (!MACHINE_IS_KVM)
456		return -ENODEV;
457
458	if (test_devices_support(total_memory_size) < 0)
459		return -ENODEV;
460
461	rc = vmem_add_mapping(total_memory_size, PAGE_SIZE);
462	if (rc)
463		return rc;
464
465	kvm_devices = (void *) total_memory_size;
466
467	kvm_root = root_device_register("kvm_s390");
468	if (IS_ERR(kvm_root)) {
469		rc = PTR_ERR(kvm_root);
470		printk(KERN_ERR "Could not register kvm_s390 root device");
471		vmem_remove_mapping(total_memory_size, PAGE_SIZE);
472		return rc;
473	}
474
475	INIT_WORK(&hotplug_work, hotplug_devices);
476
477	irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL);
478	register_external_irq(EXT_IRQ_CP_SERVICE, kvm_extint_handler);
479
480	scan_devices();
481	return 0;
482}
483
484/* code for early console output with virtio_console */
485static __init int early_put_chars(u32 vtermno, const char *buf, int count)
486{
487	char scratch[17];
488	unsigned int len = count;
489
490	if (len > sizeof(scratch) - 1)
491		len = sizeof(scratch) - 1;
492	scratch[len] = '\0';
493	memcpy(scratch, buf, len);
494	kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, __pa(scratch));
495	return len;
496}
497
498static int __init s390_virtio_console_init(void)
499{
500	if (sclp_has_vt220() || sclp_has_linemode())
501		return -ENODEV;
502	return virtio_cons_early_init(early_put_chars);
503}
504console_initcall(s390_virtio_console_init);
505
506
507/*
508 * We do this after core stuff, but before the drivers.
509 */
510postcore_initcall(kvm_devices_init);
511