Commit | Line | Data |
---|---|---|
e976a2b9 CB |
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/init.h> | |
14 | #include <linux/bootmem.h> | |
15 | #include <linux/err.h> | |
16 | #include <linux/virtio.h> | |
17 | #include <linux/virtio_config.h> | |
faeba830 | 18 | #include <linux/virtio_console.h> |
e976a2b9 CB |
19 | #include <linux/interrupt.h> |
20 | #include <linux/virtio_ring.h> | |
17f34580 | 21 | #include <linux/pfn.h> |
e976a2b9 CB |
22 | #include <asm/io.h> |
23 | #include <asm/kvm_para.h> | |
24 | #include <asm/kvm_virtio.h> | |
25 | #include <asm/setup.h> | |
26 | #include <asm/s390_ext.h> | |
27 | ||
28 | #define VIRTIO_SUBCODE_64 0x0D00 | |
29 | ||
30 | /* | |
31 | * The pointer to our (page) of device descriptions. | |
32 | */ | |
33 | static void *kvm_devices; | |
34 | ||
e976a2b9 CB |
35 | struct kvm_device { |
36 | struct virtio_device vdev; | |
37 | struct kvm_device_desc *desc; | |
38 | }; | |
39 | ||
40 | #define to_kvmdev(vd) container_of(vd, struct kvm_device, vdev) | |
41 | ||
42 | /* | |
43 | * memory layout: | |
44 | * - kvm_device_descriptor | |
45 | * struct kvm_device_desc | |
46 | * - configuration | |
47 | * struct kvm_vqconfig | |
48 | * - feature bits | |
49 | * - config space | |
50 | */ | |
51 | static struct kvm_vqconfig *kvm_vq_config(const struct kvm_device_desc *desc) | |
52 | { | |
53 | return (struct kvm_vqconfig *)(desc + 1); | |
54 | } | |
55 | ||
56 | static u8 *kvm_vq_features(const struct kvm_device_desc *desc) | |
57 | { | |
58 | return (u8 *)(kvm_vq_config(desc) + desc->num_vq); | |
59 | } | |
60 | ||
61 | static u8 *kvm_vq_configspace(const struct kvm_device_desc *desc) | |
62 | { | |
63 | return kvm_vq_features(desc) + desc->feature_len * 2; | |
64 | } | |
65 | ||
66 | /* | |
67 | * The total size of the config page used by this device (incl. desc) | |
68 | */ | |
69 | static unsigned desc_size(const struct kvm_device_desc *desc) | |
70 | { | |
71 | return sizeof(*desc) | |
72 | + desc->num_vq * sizeof(struct kvm_vqconfig) | |
73 | + desc->feature_len * 2 | |
74 | + desc->config_len; | |
75 | } | |
76 | ||
5ca9fd54 HC |
77 | /* This gets the device's feature bits. */ |
78 | static u32 kvm_get_features(struct virtio_device *vdev) | |
e976a2b9 | 79 | { |
5ca9fd54 HC |
80 | unsigned int i; |
81 | u32 features = 0; | |
e976a2b9 | 82 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; |
5ca9fd54 | 83 | u8 *in_features = kvm_vq_features(desc); |
e976a2b9 | 84 | |
5ca9fd54 HC |
85 | for (i = 0; i < min(desc->feature_len * 8, 32); i++) |
86 | if (in_features[i / 8] & (1 << (i % 8))) | |
87 | features |= (1 << i); | |
88 | return features; | |
89 | } | |
e976a2b9 | 90 | |
c624896e | 91 | static void kvm_finalize_features(struct virtio_device *vdev) |
5ca9fd54 | 92 | { |
c624896e | 93 | unsigned int i, bits; |
5ca9fd54 HC |
94 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; |
95 | /* Second half of bitmap is features we accept. */ | |
96 | u8 *out_features = kvm_vq_features(desc) + desc->feature_len; | |
e976a2b9 | 97 | |
e34f8725 RR |
98 | /* Give virtio_ring a chance to accept features. */ |
99 | vring_transport_features(vdev); | |
100 | ||
5ca9fd54 | 101 | memset(out_features, 0, desc->feature_len); |
c624896e RR |
102 | bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8; |
103 | for (i = 0; i < bits; i++) { | |
104 | if (test_bit(i, vdev->features)) | |
5ca9fd54 HC |
105 | out_features[i / 8] |= (1 << (i % 8)); |
106 | } | |
e976a2b9 CB |
107 | } |
108 | ||
109 | /* | |
110 | * Reading and writing elements in config space | |
111 | */ | |
112 | static void kvm_get(struct virtio_device *vdev, unsigned int offset, | |
113 | void *buf, unsigned len) | |
114 | { | |
115 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; | |
116 | ||
117 | BUG_ON(offset + len > desc->config_len); | |
118 | memcpy(buf, kvm_vq_configspace(desc) + offset, len); | |
119 | } | |
120 | ||
121 | static void kvm_set(struct virtio_device *vdev, unsigned int offset, | |
122 | const void *buf, unsigned len) | |
123 | { | |
124 | struct kvm_device_desc *desc = to_kvmdev(vdev)->desc; | |
125 | ||
126 | BUG_ON(offset + len > desc->config_len); | |
127 | memcpy(kvm_vq_configspace(desc) + offset, buf, len); | |
128 | } | |
129 | ||
130 | /* | |
131 | * The operations to get and set the status word just access | |
132 | * the status field of the device descriptor. set_status will also | |
133 | * make a hypercall to the host, to tell about status changes | |
134 | */ | |
135 | static u8 kvm_get_status(struct virtio_device *vdev) | |
136 | { | |
137 | return to_kvmdev(vdev)->desc->status; | |
138 | } | |
139 | ||
140 | static void kvm_set_status(struct virtio_device *vdev, u8 status) | |
141 | { | |
142 | BUG_ON(!status); | |
143 | to_kvmdev(vdev)->desc->status = status; | |
144 | kvm_hypercall1(KVM_S390_VIRTIO_SET_STATUS, | |
145 | (unsigned long) to_kvmdev(vdev)->desc); | |
146 | } | |
147 | ||
148 | /* | |
149 | * To reset the device, we use the KVM_VIRTIO_RESET hypercall, using the | |
150 | * descriptor address. The Host will zero the status and all the | |
151 | * features. | |
152 | */ | |
153 | static void kvm_reset(struct virtio_device *vdev) | |
154 | { | |
155 | kvm_hypercall1(KVM_S390_VIRTIO_RESET, | |
156 | (unsigned long) to_kvmdev(vdev)->desc); | |
157 | } | |
158 | ||
159 | /* | |
160 | * When the virtio_ring code wants to notify the Host, it calls us here and we | |
161 | * make a hypercall. We hand the address of the virtqueue so the Host | |
162 | * knows which virtqueue we're talking about. | |
163 | */ | |
164 | static void kvm_notify(struct virtqueue *vq) | |
165 | { | |
166 | struct kvm_vqconfig *config = vq->priv; | |
167 | ||
168 | kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, config->address); | |
169 | } | |
170 | ||
171 | /* | |
172 | * This routine finds the first virtqueue described in the configuration of | |
173 | * this device and sets it up. | |
174 | */ | |
175 | static struct virtqueue *kvm_find_vq(struct virtio_device *vdev, | |
176 | unsigned index, | |
177 | void (*callback)(struct virtqueue *vq)) | |
178 | { | |
179 | struct kvm_device *kdev = to_kvmdev(vdev); | |
180 | struct kvm_vqconfig *config; | |
181 | struct virtqueue *vq; | |
182 | int err; | |
183 | ||
184 | if (index >= kdev->desc->num_vq) | |
185 | return ERR_PTR(-ENOENT); | |
186 | ||
187 | config = kvm_vq_config(kdev->desc)+index; | |
188 | ||
17f34580 | 189 | err = vmem_add_mapping(config->address, |
db405988 RR |
190 | vring_size(config->num, |
191 | KVM_S390_VIRTIO_RING_ALIGN)); | |
17f34580 | 192 | if (err) |
e976a2b9 | 193 | goto out; |
e976a2b9 | 194 | |
87c7d57c RR |
195 | vq = vring_new_virtqueue(config->num, KVM_S390_VIRTIO_RING_ALIGN, |
196 | vdev, (void *) config->address, | |
e976a2b9 CB |
197 | kvm_notify, callback); |
198 | if (!vq) { | |
199 | err = -ENOMEM; | |
200 | goto unmap; | |
201 | } | |
202 | ||
203 | /* | |
204 | * register a callback token | |
205 | * The host will sent this via the external interrupt parameter | |
206 | */ | |
207 | config->token = (u64) vq; | |
208 | ||
209 | vq->priv = config; | |
210 | return vq; | |
211 | unmap: | |
17f34580 | 212 | vmem_remove_mapping(config->address, |
db405988 RR |
213 | vring_size(config->num, |
214 | KVM_S390_VIRTIO_RING_ALIGN)); | |
e976a2b9 CB |
215 | out: |
216 | return ERR_PTR(err); | |
217 | } | |
218 | ||
219 | static void kvm_del_vq(struct virtqueue *vq) | |
220 | { | |
221 | struct kvm_vqconfig *config = vq->priv; | |
222 | ||
223 | vring_del_virtqueue(vq); | |
17f34580 | 224 | vmem_remove_mapping(config->address, |
db405988 RR |
225 | vring_size(config->num, |
226 | KVM_S390_VIRTIO_RING_ALIGN)); | |
e976a2b9 CB |
227 | } |
228 | ||
229 | /* | |
230 | * The config ops structure as defined by virtio config | |
231 | */ | |
232 | static struct virtio_config_ops kvm_vq_configspace_ops = { | |
5ca9fd54 | 233 | .get_features = kvm_get_features, |
c624896e | 234 | .finalize_features = kvm_finalize_features, |
e976a2b9 CB |
235 | .get = kvm_get, |
236 | .set = kvm_set, | |
237 | .get_status = kvm_get_status, | |
238 | .set_status = kvm_set_status, | |
239 | .reset = kvm_reset, | |
240 | .find_vq = kvm_find_vq, | |
241 | .del_vq = kvm_del_vq, | |
242 | }; | |
243 | ||
244 | /* | |
245 | * The root device for the kvm virtio devices. | |
246 | * This makes them appear as /sys/devices/kvm_s390/0,1,2 not /sys/devices/0,1,2. | |
247 | */ | |
37f1c012 | 248 | static struct device *kvm_root; |
e976a2b9 CB |
249 | |
250 | /* | |
251 | * adds a new device and register it with virtio | |
252 | * appropriate drivers are loaded by the device model | |
253 | */ | |
b769f579 | 254 | static void add_kvm_device(struct kvm_device_desc *d, unsigned int offset) |
e976a2b9 CB |
255 | { |
256 | struct kvm_device *kdev; | |
257 | ||
258 | kdev = kzalloc(sizeof(*kdev), GFP_KERNEL); | |
259 | if (!kdev) { | |
b769f579 RR |
260 | printk(KERN_EMERG "Cannot allocate kvm dev %u type %u\n", |
261 | offset, d->type); | |
e976a2b9 CB |
262 | return; |
263 | } | |
264 | ||
37f1c012 | 265 | kdev->vdev.dev.parent = kvm_root; |
e976a2b9 CB |
266 | kdev->vdev.id.device = d->type; |
267 | kdev->vdev.config = &kvm_vq_configspace_ops; | |
268 | kdev->desc = d; | |
269 | ||
270 | if (register_virtio_device(&kdev->vdev) != 0) { | |
b769f579 RR |
271 | printk(KERN_ERR "Failed to register kvm device %u type %u\n", |
272 | offset, d->type); | |
e976a2b9 CB |
273 | kfree(kdev); |
274 | } | |
275 | } | |
276 | ||
277 | /* | |
278 | * scan_devices() simply iterates through the device page. | |
279 | * The type 0 is reserved to mean "end of devices". | |
280 | */ | |
281 | static void scan_devices(void) | |
282 | { | |
283 | unsigned int i; | |
284 | struct kvm_device_desc *d; | |
285 | ||
286 | for (i = 0; i < PAGE_SIZE; i += desc_size(d)) { | |
287 | d = kvm_devices + i; | |
288 | ||
289 | if (d->type == 0) | |
290 | break; | |
291 | ||
b769f579 | 292 | add_kvm_device(d, i); |
e976a2b9 CB |
293 | } |
294 | } | |
295 | ||
296 | /* | |
297 | * we emulate the request_irq behaviour on top of s390 extints | |
298 | */ | |
299 | static void kvm_extint_handler(u16 code) | |
300 | { | |
be3c5832 CB |
301 | struct virtqueue *vq; |
302 | u16 subcode; | |
303 | int config_changed; | |
e976a2b9 | 304 | |
be3c5832 | 305 | subcode = S390_lowcore.cpu_addr; |
e976a2b9 CB |
306 | if ((subcode & 0xff00) != VIRTIO_SUBCODE_64) |
307 | return; | |
308 | ||
be3c5832 CB |
309 | /* The LSB might be overloaded, we have to mask it */ |
310 | vq = (struct virtqueue *) ((*(long *) __LC_PFAULT_INTPARM) & ~1UL); | |
311 | ||
312 | /* We use the LSB of extparam, to decide, if this interrupt is a config | |
313 | * change or a "standard" interrupt */ | |
314 | config_changed = (*(int *) __LC_EXT_PARAMS & 1); | |
315 | ||
316 | if (config_changed) { | |
317 | struct virtio_driver *drv; | |
318 | drv = container_of(vq->vdev->dev.driver, | |
319 | struct virtio_driver, driver); | |
320 | if (drv->config_changed) | |
321 | drv->config_changed(vq->vdev); | |
322 | } else | |
323 | vring_interrupt(0, vq); | |
e976a2b9 CB |
324 | } |
325 | ||
326 | /* | |
327 | * Init function for virtio | |
328 | * devices are in a single page above top of "normal" mem | |
329 | */ | |
330 | static int __init kvm_devices_init(void) | |
331 | { | |
332 | int rc; | |
333 | ||
334 | if (!MACHINE_IS_KVM) | |
335 | return -ENODEV; | |
336 | ||
035da16f | 337 | kvm_root = root_device_register("kvm_s390"); |
37f1c012 CH |
338 | if (IS_ERR(kvm_root)) { |
339 | rc = PTR_ERR(kvm_root); | |
e976a2b9 CB |
340 | printk(KERN_ERR "Could not register kvm_s390 root device"); |
341 | return rc; | |
342 | } | |
343 | ||
cc835f78 | 344 | rc = vmem_add_mapping(real_memory_size, PAGE_SIZE); |
17f34580 | 345 | if (rc) { |
035da16f | 346 | root_device_unregister(kvm_root); |
17f34580 | 347 | return rc; |
e976a2b9 CB |
348 | } |
349 | ||
cc835f78 | 350 | kvm_devices = (void *) real_memory_size; |
e976a2b9 CB |
351 | |
352 | ctl_set_bit(0, 9); | |
353 | register_external_interrupt(0x2603, kvm_extint_handler); | |
354 | ||
355 | scan_devices(); | |
356 | return 0; | |
357 | } | |
358 | ||
faeba830 CB |
359 | /* code for early console output with virtio_console */ |
360 | static __init int early_put_chars(u32 vtermno, const char *buf, int count) | |
361 | { | |
362 | char scratch[17]; | |
363 | unsigned int len = count; | |
364 | ||
365 | if (len > sizeof(scratch) - 1) | |
366 | len = sizeof(scratch) - 1; | |
367 | scratch[len] = '\0'; | |
368 | memcpy(scratch, buf, len); | |
369 | kvm_hypercall1(KVM_S390_VIRTIO_NOTIFY, __pa(scratch)); | |
370 | return len; | |
371 | } | |
372 | ||
26f746f3 | 373 | void __init s390_virtio_console_init(void) |
faeba830 CB |
374 | { |
375 | virtio_cons_early_init(early_put_chars); | |
376 | } | |
377 | ||
e976a2b9 CB |
378 | /* |
379 | * We do this after core stuff, but before the drivers. | |
380 | */ | |
381 | postcore_initcall(kvm_devices_init); |