]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - virt/kvm/arm/vgic/vgic-mmio.c
KVM: arm/arm64: vgic: Introduce VENG0 and VENG1 fields to vmcr struct
[mirror_ubuntu-artful-kernel.git] / virt / kvm / arm / vgic / vgic-mmio.c
1 /*
2 * VGIC MMIO handling functions
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 */
13
14 #include <linux/bitops.h>
15 #include <linux/bsearch.h>
16 #include <linux/kvm.h>
17 #include <linux/kvm_host.h>
18 #include <kvm/iodev.h>
19 #include <kvm/arm_vgic.h>
20
21 #include "vgic.h"
22 #include "vgic-mmio.h"
23
24 unsigned long vgic_mmio_read_raz(struct kvm_vcpu *vcpu,
25 gpa_t addr, unsigned int len)
26 {
27 return 0;
28 }
29
30 unsigned long vgic_mmio_read_rao(struct kvm_vcpu *vcpu,
31 gpa_t addr, unsigned int len)
32 {
33 return -1UL;
34 }
35
36 void vgic_mmio_write_wi(struct kvm_vcpu *vcpu, gpa_t addr,
37 unsigned int len, unsigned long val)
38 {
39 /* Ignore */
40 }
41
42 /*
43 * Read accesses to both GICD_ICENABLER and GICD_ISENABLER return the value
44 * of the enabled bit, so there is only one function for both here.
45 */
46 unsigned long vgic_mmio_read_enable(struct kvm_vcpu *vcpu,
47 gpa_t addr, unsigned int len)
48 {
49 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
50 u32 value = 0;
51 int i;
52
53 /* Loop over all IRQs affected by this read */
54 for (i = 0; i < len * 8; i++) {
55 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
56
57 if (irq->enabled)
58 value |= (1U << i);
59
60 vgic_put_irq(vcpu->kvm, irq);
61 }
62
63 return value;
64 }
65
66 void vgic_mmio_write_senable(struct kvm_vcpu *vcpu,
67 gpa_t addr, unsigned int len,
68 unsigned long val)
69 {
70 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
71 int i;
72
73 for_each_set_bit(i, &val, len * 8) {
74 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
75
76 spin_lock(&irq->irq_lock);
77 irq->enabled = true;
78 vgic_queue_irq_unlock(vcpu->kvm, irq);
79
80 vgic_put_irq(vcpu->kvm, irq);
81 }
82 }
83
84 void vgic_mmio_write_cenable(struct kvm_vcpu *vcpu,
85 gpa_t addr, unsigned int len,
86 unsigned long val)
87 {
88 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
89 int i;
90
91 for_each_set_bit(i, &val, len * 8) {
92 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
93
94 spin_lock(&irq->irq_lock);
95
96 irq->enabled = false;
97
98 spin_unlock(&irq->irq_lock);
99 vgic_put_irq(vcpu->kvm, irq);
100 }
101 }
102
103 unsigned long vgic_mmio_read_pending(struct kvm_vcpu *vcpu,
104 gpa_t addr, unsigned int len)
105 {
106 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
107 u32 value = 0;
108 int i;
109
110 /* Loop over all IRQs affected by this read */
111 for (i = 0; i < len * 8; i++) {
112 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
113
114 if (irq_is_pending(irq))
115 value |= (1U << i);
116
117 vgic_put_irq(vcpu->kvm, irq);
118 }
119
120 return value;
121 }
122
123 void vgic_mmio_write_spending(struct kvm_vcpu *vcpu,
124 gpa_t addr, unsigned int len,
125 unsigned long val)
126 {
127 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
128 int i;
129
130 for_each_set_bit(i, &val, len * 8) {
131 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
132
133 spin_lock(&irq->irq_lock);
134 irq->pending_latch = true;
135
136 vgic_queue_irq_unlock(vcpu->kvm, irq);
137 vgic_put_irq(vcpu->kvm, irq);
138 }
139 }
140
141 void vgic_mmio_write_cpending(struct kvm_vcpu *vcpu,
142 gpa_t addr, unsigned int len,
143 unsigned long val)
144 {
145 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
146 int i;
147
148 for_each_set_bit(i, &val, len * 8) {
149 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
150
151 spin_lock(&irq->irq_lock);
152
153 irq->pending_latch = false;
154
155 spin_unlock(&irq->irq_lock);
156 vgic_put_irq(vcpu->kvm, irq);
157 }
158 }
159
160 unsigned long vgic_mmio_read_active(struct kvm_vcpu *vcpu,
161 gpa_t addr, unsigned int len)
162 {
163 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
164 u32 value = 0;
165 int i;
166
167 /* Loop over all IRQs affected by this read */
168 for (i = 0; i < len * 8; i++) {
169 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
170
171 if (irq->active)
172 value |= (1U << i);
173
174 vgic_put_irq(vcpu->kvm, irq);
175 }
176
177 return value;
178 }
179
180 static void vgic_mmio_change_active(struct kvm_vcpu *vcpu, struct vgic_irq *irq,
181 bool new_active_state)
182 {
183 spin_lock(&irq->irq_lock);
184 /*
185 * If this virtual IRQ was written into a list register, we
186 * have to make sure the CPU that runs the VCPU thread has
187 * synced back LR state to the struct vgic_irq. We can only
188 * know this for sure, when either this irq is not assigned to
189 * anyone's AP list anymore, or the VCPU thread is not
190 * running on any CPUs.
191 *
192 * In the opposite case, we know the VCPU thread may be on its
193 * way back from the guest and still has to sync back this
194 * IRQ, so we release and re-acquire the spin_lock to let the
195 * other thread sync back the IRQ.
196 */
197 while (irq->vcpu && /* IRQ may have state in an LR somewhere */
198 irq->vcpu->cpu != -1) /* VCPU thread is running */
199 cond_resched_lock(&irq->irq_lock);
200
201 irq->active = new_active_state;
202 if (new_active_state)
203 vgic_queue_irq_unlock(vcpu->kvm, irq);
204 else
205 spin_unlock(&irq->irq_lock);
206 }
207
208 /*
209 * If we are fiddling with an IRQ's active state, we have to make sure the IRQ
210 * is not queued on some running VCPU's LRs, because then the change to the
211 * active state can be overwritten when the VCPU's state is synced coming back
212 * from the guest.
213 *
214 * For shared interrupts, we have to stop all the VCPUs because interrupts can
215 * be migrated while we don't hold the IRQ locks and we don't want to be
216 * chasing moving targets.
217 *
218 * For private interrupts, we only have to make sure the single and only VCPU
219 * that can potentially queue the IRQ is stopped.
220 */
221 static void vgic_change_active_prepare(struct kvm_vcpu *vcpu, u32 intid)
222 {
223 if (intid < VGIC_NR_PRIVATE_IRQS)
224 kvm_arm_halt_vcpu(vcpu);
225 else
226 kvm_arm_halt_guest(vcpu->kvm);
227 }
228
229 /* See vgic_change_active_prepare */
230 static void vgic_change_active_finish(struct kvm_vcpu *vcpu, u32 intid)
231 {
232 if (intid < VGIC_NR_PRIVATE_IRQS)
233 kvm_arm_resume_vcpu(vcpu);
234 else
235 kvm_arm_resume_guest(vcpu->kvm);
236 }
237
238 void vgic_mmio_write_cactive(struct kvm_vcpu *vcpu,
239 gpa_t addr, unsigned int len,
240 unsigned long val)
241 {
242 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
243 int i;
244
245 vgic_change_active_prepare(vcpu, intid);
246 for_each_set_bit(i, &val, len * 8) {
247 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
248 vgic_mmio_change_active(vcpu, irq, false);
249 vgic_put_irq(vcpu->kvm, irq);
250 }
251 vgic_change_active_finish(vcpu, intid);
252 }
253
254 void vgic_mmio_write_sactive(struct kvm_vcpu *vcpu,
255 gpa_t addr, unsigned int len,
256 unsigned long val)
257 {
258 u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
259 int i;
260
261 vgic_change_active_prepare(vcpu, intid);
262 for_each_set_bit(i, &val, len * 8) {
263 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
264 vgic_mmio_change_active(vcpu, irq, true);
265 vgic_put_irq(vcpu->kvm, irq);
266 }
267 vgic_change_active_finish(vcpu, intid);
268 }
269
270 unsigned long vgic_mmio_read_priority(struct kvm_vcpu *vcpu,
271 gpa_t addr, unsigned int len)
272 {
273 u32 intid = VGIC_ADDR_TO_INTID(addr, 8);
274 int i;
275 u64 val = 0;
276
277 for (i = 0; i < len; i++) {
278 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
279
280 val |= (u64)irq->priority << (i * 8);
281
282 vgic_put_irq(vcpu->kvm, irq);
283 }
284
285 return val;
286 }
287
288 /*
289 * We currently don't handle changing the priority of an interrupt that
290 * is already pending on a VCPU. If there is a need for this, we would
291 * need to make this VCPU exit and re-evaluate the priorities, potentially
292 * leading to this interrupt getting presented now to the guest (if it has
293 * been masked by the priority mask before).
294 */
295 void vgic_mmio_write_priority(struct kvm_vcpu *vcpu,
296 gpa_t addr, unsigned int len,
297 unsigned long val)
298 {
299 u32 intid = VGIC_ADDR_TO_INTID(addr, 8);
300 int i;
301
302 for (i = 0; i < len; i++) {
303 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
304
305 spin_lock(&irq->irq_lock);
306 /* Narrow the priority range to what we actually support */
307 irq->priority = (val >> (i * 8)) & GENMASK(7, 8 - VGIC_PRI_BITS);
308 spin_unlock(&irq->irq_lock);
309
310 vgic_put_irq(vcpu->kvm, irq);
311 }
312 }
313
314 unsigned long vgic_mmio_read_config(struct kvm_vcpu *vcpu,
315 gpa_t addr, unsigned int len)
316 {
317 u32 intid = VGIC_ADDR_TO_INTID(addr, 2);
318 u32 value = 0;
319 int i;
320
321 for (i = 0; i < len * 4; i++) {
322 struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
323
324 if (irq->config == VGIC_CONFIG_EDGE)
325 value |= (2U << (i * 2));
326
327 vgic_put_irq(vcpu->kvm, irq);
328 }
329
330 return value;
331 }
332
333 void vgic_mmio_write_config(struct kvm_vcpu *vcpu,
334 gpa_t addr, unsigned int len,
335 unsigned long val)
336 {
337 u32 intid = VGIC_ADDR_TO_INTID(addr, 2);
338 int i;
339
340 for (i = 0; i < len * 4; i++) {
341 struct vgic_irq *irq;
342
343 /*
344 * The configuration cannot be changed for SGIs in general,
345 * for PPIs this is IMPLEMENTATION DEFINED. The arch timer
346 * code relies on PPIs being level triggered, so we also
347 * make them read-only here.
348 */
349 if (intid + i < VGIC_NR_PRIVATE_IRQS)
350 continue;
351
352 irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
353 spin_lock(&irq->irq_lock);
354
355 if (test_bit(i * 2 + 1, &val))
356 irq->config = VGIC_CONFIG_EDGE;
357 else
358 irq->config = VGIC_CONFIG_LEVEL;
359
360 spin_unlock(&irq->irq_lock);
361 vgic_put_irq(vcpu->kvm, irq);
362 }
363 }
364
365 static int match_region(const void *key, const void *elt)
366 {
367 const unsigned int offset = (unsigned long)key;
368 const struct vgic_register_region *region = elt;
369
370 if (offset < region->reg_offset)
371 return -1;
372
373 if (offset >= region->reg_offset + region->len)
374 return 1;
375
376 return 0;
377 }
378
379 /* Find the proper register handler entry given a certain address offset. */
380 static const struct vgic_register_region *
381 vgic_find_mmio_region(const struct vgic_register_region *region, int nr_regions,
382 unsigned int offset)
383 {
384 return bsearch((void *)(uintptr_t)offset, region, nr_regions,
385 sizeof(region[0]), match_region);
386 }
387
388 void vgic_set_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcr)
389 {
390 if (kvm_vgic_global_state.type == VGIC_V2)
391 vgic_v2_set_vmcr(vcpu, vmcr);
392 else
393 vgic_v3_set_vmcr(vcpu, vmcr);
394 }
395
396 void vgic_get_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcr)
397 {
398 if (kvm_vgic_global_state.type == VGIC_V2)
399 vgic_v2_get_vmcr(vcpu, vmcr);
400 else
401 vgic_v3_get_vmcr(vcpu, vmcr);
402 }
403
404 /*
405 * kvm_mmio_read_buf() returns a value in a format where it can be converted
406 * to a byte array and be directly observed as the guest wanted it to appear
407 * in memory if it had done the store itself, which is LE for the GIC, as the
408 * guest knows the GIC is always LE.
409 *
410 * We convert this value to the CPUs native format to deal with it as a data
411 * value.
412 */
413 unsigned long vgic_data_mmio_bus_to_host(const void *val, unsigned int len)
414 {
415 unsigned long data = kvm_mmio_read_buf(val, len);
416
417 switch (len) {
418 case 1:
419 return data;
420 case 2:
421 return le16_to_cpu(data);
422 case 4:
423 return le32_to_cpu(data);
424 default:
425 return le64_to_cpu(data);
426 }
427 }
428
429 /*
430 * kvm_mmio_write_buf() expects a value in a format such that if converted to
431 * a byte array it is observed as the guest would see it if it could perform
432 * the load directly. Since the GIC is LE, and the guest knows this, the
433 * guest expects a value in little endian format.
434 *
435 * We convert the data value from the CPUs native format to LE so that the
436 * value is returned in the proper format.
437 */
438 void vgic_data_host_to_mmio_bus(void *buf, unsigned int len,
439 unsigned long data)
440 {
441 switch (len) {
442 case 1:
443 break;
444 case 2:
445 data = cpu_to_le16(data);
446 break;
447 case 4:
448 data = cpu_to_le32(data);
449 break;
450 default:
451 data = cpu_to_le64(data);
452 }
453
454 kvm_mmio_write_buf(buf, len, data);
455 }
456
457 static
458 struct vgic_io_device *kvm_to_vgic_iodev(const struct kvm_io_device *dev)
459 {
460 return container_of(dev, struct vgic_io_device, dev);
461 }
462
463 static bool check_region(const struct kvm *kvm,
464 const struct vgic_register_region *region,
465 gpa_t addr, int len)
466 {
467 int flags, nr_irqs = kvm->arch.vgic.nr_spis + VGIC_NR_PRIVATE_IRQS;
468
469 switch (len) {
470 case sizeof(u8):
471 flags = VGIC_ACCESS_8bit;
472 break;
473 case sizeof(u32):
474 flags = VGIC_ACCESS_32bit;
475 break;
476 case sizeof(u64):
477 flags = VGIC_ACCESS_64bit;
478 break;
479 default:
480 return false;
481 }
482
483 if ((region->access_flags & flags) && IS_ALIGNED(addr, len)) {
484 if (!region->bits_per_irq)
485 return true;
486
487 /* Do we access a non-allocated IRQ? */
488 return VGIC_ADDR_TO_INTID(addr, region->bits_per_irq) < nr_irqs;
489 }
490
491 return false;
492 }
493
494 const struct vgic_register_region *
495 vgic_get_mmio_region(struct kvm_vcpu *vcpu, struct vgic_io_device *iodev,
496 gpa_t addr, int len)
497 {
498 const struct vgic_register_region *region;
499
500 region = vgic_find_mmio_region(iodev->regions, iodev->nr_regions,
501 addr - iodev->base_addr);
502 if (!region || !check_region(vcpu->kvm, region, addr, len))
503 return NULL;
504
505 return region;
506 }
507
508 static int vgic_uaccess_read(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
509 gpa_t addr, u32 *val)
510 {
511 struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
512 const struct vgic_register_region *region;
513 struct kvm_vcpu *r_vcpu;
514
515 region = vgic_get_mmio_region(vcpu, iodev, addr, sizeof(u32));
516 if (!region) {
517 *val = 0;
518 return 0;
519 }
520
521 r_vcpu = iodev->redist_vcpu ? iodev->redist_vcpu : vcpu;
522 if (region->uaccess_read)
523 *val = region->uaccess_read(r_vcpu, addr, sizeof(u32));
524 else
525 *val = region->read(r_vcpu, addr, sizeof(u32));
526
527 return 0;
528 }
529
530 static int vgic_uaccess_write(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
531 gpa_t addr, const u32 *val)
532 {
533 struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
534 const struct vgic_register_region *region;
535 struct kvm_vcpu *r_vcpu;
536
537 region = vgic_get_mmio_region(vcpu, iodev, addr, sizeof(u32));
538 if (!region)
539 return 0;
540
541 r_vcpu = iodev->redist_vcpu ? iodev->redist_vcpu : vcpu;
542 if (region->uaccess_write)
543 region->uaccess_write(r_vcpu, addr, sizeof(u32), *val);
544 else
545 region->write(r_vcpu, addr, sizeof(u32), *val);
546
547 return 0;
548 }
549
550 /*
551 * Userland access to VGIC registers.
552 */
553 int vgic_uaccess(struct kvm_vcpu *vcpu, struct vgic_io_device *dev,
554 bool is_write, int offset, u32 *val)
555 {
556 if (is_write)
557 return vgic_uaccess_write(vcpu, &dev->dev, offset, val);
558 else
559 return vgic_uaccess_read(vcpu, &dev->dev, offset, val);
560 }
561
562 static int dispatch_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
563 gpa_t addr, int len, void *val)
564 {
565 struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
566 const struct vgic_register_region *region;
567 unsigned long data = 0;
568
569 region = vgic_get_mmio_region(vcpu, iodev, addr, len);
570 if (!region) {
571 memset(val, 0, len);
572 return 0;
573 }
574
575 switch (iodev->iodev_type) {
576 case IODEV_CPUIF:
577 data = region->read(vcpu, addr, len);
578 break;
579 case IODEV_DIST:
580 data = region->read(vcpu, addr, len);
581 break;
582 case IODEV_REDIST:
583 data = region->read(iodev->redist_vcpu, addr, len);
584 break;
585 case IODEV_ITS:
586 data = region->its_read(vcpu->kvm, iodev->its, addr, len);
587 break;
588 }
589
590 vgic_data_host_to_mmio_bus(val, len, data);
591 return 0;
592 }
593
594 static int dispatch_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
595 gpa_t addr, int len, const void *val)
596 {
597 struct vgic_io_device *iodev = kvm_to_vgic_iodev(dev);
598 const struct vgic_register_region *region;
599 unsigned long data = vgic_data_mmio_bus_to_host(val, len);
600
601 region = vgic_get_mmio_region(vcpu, iodev, addr, len);
602 if (!region)
603 return 0;
604
605 switch (iodev->iodev_type) {
606 case IODEV_CPUIF:
607 region->write(vcpu, addr, len, data);
608 break;
609 case IODEV_DIST:
610 region->write(vcpu, addr, len, data);
611 break;
612 case IODEV_REDIST:
613 region->write(iodev->redist_vcpu, addr, len, data);
614 break;
615 case IODEV_ITS:
616 region->its_write(vcpu->kvm, iodev->its, addr, len, data);
617 break;
618 }
619
620 return 0;
621 }
622
623 struct kvm_io_device_ops kvm_io_gic_ops = {
624 .read = dispatch_mmio_read,
625 .write = dispatch_mmio_write,
626 };
627
628 int vgic_register_dist_iodev(struct kvm *kvm, gpa_t dist_base_address,
629 enum vgic_type type)
630 {
631 struct vgic_io_device *io_device = &kvm->arch.vgic.dist_iodev;
632 int ret = 0;
633 unsigned int len;
634
635 switch (type) {
636 case VGIC_V2:
637 len = vgic_v2_init_dist_iodev(io_device);
638 break;
639 case VGIC_V3:
640 len = vgic_v3_init_dist_iodev(io_device);
641 break;
642 default:
643 BUG_ON(1);
644 }
645
646 io_device->base_addr = dist_base_address;
647 io_device->iodev_type = IODEV_DIST;
648 io_device->redist_vcpu = NULL;
649
650 mutex_lock(&kvm->slots_lock);
651 ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, dist_base_address,
652 len, &io_device->dev);
653 mutex_unlock(&kvm->slots_lock);
654
655 return ret;
656 }