Linux-libre 5.3.12-gnu
[librecmc/linux-libre.git] / virt / kvm / arm / vgic / vgic-its.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * GICv3 ITS emulation
4  *
5  * Copyright (C) 2015,2016 ARM Ltd.
6  * Author: Andre Przywara <andre.przywara@arm.com>
7  */
8
9 #include <linux/cpu.h>
10 #include <linux/kvm.h>
11 #include <linux/kvm_host.h>
12 #include <linux/interrupt.h>
13 #include <linux/list.h>
14 #include <linux/uaccess.h>
15 #include <linux/list_sort.h>
16
17 #include <linux/irqchip/arm-gic-v3.h>
18
19 #include <asm/kvm_emulate.h>
20 #include <asm/kvm_arm.h>
21 #include <asm/kvm_mmu.h>
22
23 #include "vgic.h"
24 #include "vgic-mmio.h"
25
26 static int vgic_its_save_tables_v0(struct vgic_its *its);
27 static int vgic_its_restore_tables_v0(struct vgic_its *its);
28 static int vgic_its_commit_v0(struct vgic_its *its);
29 static int update_lpi_config(struct kvm *kvm, struct vgic_irq *irq,
30                              struct kvm_vcpu *filter_vcpu, bool needs_inv);
31
32 /*
33  * Creates a new (reference to a) struct vgic_irq for a given LPI.
34  * If this LPI is already mapped on another ITS, we increase its refcount
35  * and return a pointer to the existing structure.
36  * If this is a "new" LPI, we allocate and initialize a new struct vgic_irq.
37  * This function returns a pointer to the _unlocked_ structure.
38  */
39 static struct vgic_irq *vgic_add_lpi(struct kvm *kvm, u32 intid,
40                                      struct kvm_vcpu *vcpu)
41 {
42         struct vgic_dist *dist = &kvm->arch.vgic;
43         struct vgic_irq *irq = vgic_get_irq(kvm, NULL, intid), *oldirq;
44         unsigned long flags;
45         int ret;
46
47         /* In this case there is no put, since we keep the reference. */
48         if (irq)
49                 return irq;
50
51         irq = kzalloc(sizeof(struct vgic_irq), GFP_KERNEL);
52         if (!irq)
53                 return ERR_PTR(-ENOMEM);
54
55         INIT_LIST_HEAD(&irq->lpi_list);
56         INIT_LIST_HEAD(&irq->ap_list);
57         raw_spin_lock_init(&irq->irq_lock);
58
59         irq->config = VGIC_CONFIG_EDGE;
60         kref_init(&irq->refcount);
61         irq->intid = intid;
62         irq->target_vcpu = vcpu;
63         irq->group = 1;
64
65         raw_spin_lock_irqsave(&dist->lpi_list_lock, flags);
66
67         /*
68          * There could be a race with another vgic_add_lpi(), so we need to
69          * check that we don't add a second list entry with the same LPI.
70          */
71         list_for_each_entry(oldirq, &dist->lpi_list_head, lpi_list) {
72                 if (oldirq->intid != intid)
73                         continue;
74
75                 /* Someone was faster with adding this LPI, lets use that. */
76                 kfree(irq);
77                 irq = oldirq;
78
79                 /*
80                  * This increases the refcount, the caller is expected to
81                  * call vgic_put_irq() on the returned pointer once it's
82                  * finished with the IRQ.
83                  */
84                 vgic_get_irq_kref(irq);
85
86                 goto out_unlock;
87         }
88
89         list_add_tail(&irq->lpi_list, &dist->lpi_list_head);
90         dist->lpi_list_count++;
91
92 out_unlock:
93         raw_spin_unlock_irqrestore(&dist->lpi_list_lock, flags);
94
95         /*
96          * We "cache" the configuration table entries in our struct vgic_irq's.
97          * However we only have those structs for mapped IRQs, so we read in
98          * the respective config data from memory here upon mapping the LPI.
99          */
100         ret = update_lpi_config(kvm, irq, NULL, false);
101         if (ret)
102                 return ERR_PTR(ret);
103
104         ret = vgic_v3_lpi_sync_pending_status(kvm, irq);
105         if (ret)
106                 return ERR_PTR(ret);
107
108         return irq;
109 }
110
111 struct its_device {
112         struct list_head dev_list;
113
114         /* the head for the list of ITTEs */
115         struct list_head itt_head;
116         u32 num_eventid_bits;
117         gpa_t itt_addr;
118         u32 device_id;
119 };
120
121 #define COLLECTION_NOT_MAPPED ((u32)~0)
122
123 struct its_collection {
124         struct list_head coll_list;
125
126         u32 collection_id;
127         u32 target_addr;
128 };
129
130 #define its_is_collection_mapped(coll) ((coll) && \
131                                 ((coll)->target_addr != COLLECTION_NOT_MAPPED))
132
133 struct its_ite {
134         struct list_head ite_list;
135
136         struct vgic_irq *irq;
137         struct its_collection *collection;
138         u32 event_id;
139 };
140
141 /**
142  * struct vgic_its_abi - ITS abi ops and settings
143  * @cte_esz: collection table entry size
144  * @dte_esz: device table entry size
145  * @ite_esz: interrupt translation table entry size
146  * @save tables: save the ITS tables into guest RAM
147  * @restore_tables: restore the ITS internal structs from tables
148  *  stored in guest RAM
149  * @commit: initialize the registers which expose the ABI settings,
150  *  especially the entry sizes
151  */
152 struct vgic_its_abi {
153         int cte_esz;
154         int dte_esz;
155         int ite_esz;
156         int (*save_tables)(struct vgic_its *its);
157         int (*restore_tables)(struct vgic_its *its);
158         int (*commit)(struct vgic_its *its);
159 };
160
161 #define ABI_0_ESZ       8
162 #define ESZ_MAX         ABI_0_ESZ
163
164 static const struct vgic_its_abi its_table_abi_versions[] = {
165         [0] = {
166          .cte_esz = ABI_0_ESZ,
167          .dte_esz = ABI_0_ESZ,
168          .ite_esz = ABI_0_ESZ,
169          .save_tables = vgic_its_save_tables_v0,
170          .restore_tables = vgic_its_restore_tables_v0,
171          .commit = vgic_its_commit_v0,
172         },
173 };
174
175 #define NR_ITS_ABIS     ARRAY_SIZE(its_table_abi_versions)
176
177 inline const struct vgic_its_abi *vgic_its_get_abi(struct vgic_its *its)
178 {
179         return &its_table_abi_versions[its->abi_rev];
180 }
181
182 static int vgic_its_set_abi(struct vgic_its *its, u32 rev)
183 {
184         const struct vgic_its_abi *abi;
185
186         its->abi_rev = rev;
187         abi = vgic_its_get_abi(its);
188         return abi->commit(its);
189 }
190
191 /*
192  * Find and returns a device in the device table for an ITS.
193  * Must be called with the its_lock mutex held.
194  */
195 static struct its_device *find_its_device(struct vgic_its *its, u32 device_id)
196 {
197         struct its_device *device;
198
199         list_for_each_entry(device, &its->device_list, dev_list)
200                 if (device_id == device->device_id)
201                         return device;
202
203         return NULL;
204 }
205
206 /*
207  * Find and returns an interrupt translation table entry (ITTE) for a given
208  * Device ID/Event ID pair on an ITS.
209  * Must be called with the its_lock mutex held.
210  */
211 static struct its_ite *find_ite(struct vgic_its *its, u32 device_id,
212                                   u32 event_id)
213 {
214         struct its_device *device;
215         struct its_ite *ite;
216
217         device = find_its_device(its, device_id);
218         if (device == NULL)
219                 return NULL;
220
221         list_for_each_entry(ite, &device->itt_head, ite_list)
222                 if (ite->event_id == event_id)
223                         return ite;
224
225         return NULL;
226 }
227
228 /* To be used as an iterator this macro misses the enclosing parentheses */
229 #define for_each_lpi_its(dev, ite, its) \
230         list_for_each_entry(dev, &(its)->device_list, dev_list) \
231                 list_for_each_entry(ite, &(dev)->itt_head, ite_list)
232
233 #define GIC_LPI_OFFSET 8192
234
235 #define VITS_TYPER_IDBITS 16
236 #define VITS_TYPER_DEVBITS 16
237 #define VITS_DTE_MAX_DEVID_OFFSET       (BIT(14) - 1)
238 #define VITS_ITE_MAX_EVENTID_OFFSET     (BIT(16) - 1)
239
240 /*
241  * Finds and returns a collection in the ITS collection table.
242  * Must be called with the its_lock mutex held.
243  */
244 static struct its_collection *find_collection(struct vgic_its *its, int coll_id)
245 {
246         struct its_collection *collection;
247
248         list_for_each_entry(collection, &its->collection_list, coll_list) {
249                 if (coll_id == collection->collection_id)
250                         return collection;
251         }
252
253         return NULL;
254 }
255
256 #define LPI_PROP_ENABLE_BIT(p)  ((p) & LPI_PROP_ENABLED)
257 #define LPI_PROP_PRIORITY(p)    ((p) & 0xfc)
258
259 /*
260  * Reads the configuration data for a given LPI from guest memory and
261  * updates the fields in struct vgic_irq.
262  * If filter_vcpu is not NULL, applies only if the IRQ is targeting this
263  * VCPU. Unconditionally applies if filter_vcpu is NULL.
264  */
265 static int update_lpi_config(struct kvm *kvm, struct vgic_irq *irq,
266                              struct kvm_vcpu *filter_vcpu, bool needs_inv)
267 {
268         u64 propbase = GICR_PROPBASER_ADDRESS(kvm->arch.vgic.propbaser);
269         u8 prop;
270         int ret;
271         unsigned long flags;
272
273         ret = kvm_read_guest_lock(kvm, propbase + irq->intid - GIC_LPI_OFFSET,
274                                   &prop, 1);
275
276         if (ret)
277                 return ret;
278
279         raw_spin_lock_irqsave(&irq->irq_lock, flags);
280
281         if (!filter_vcpu || filter_vcpu == irq->target_vcpu) {
282                 irq->priority = LPI_PROP_PRIORITY(prop);
283                 irq->enabled = LPI_PROP_ENABLE_BIT(prop);
284
285                 if (!irq->hw) {
286                         vgic_queue_irq_unlock(kvm, irq, flags);
287                         return 0;
288                 }
289         }
290
291         raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
292
293         if (irq->hw)
294                 return its_prop_update_vlpi(irq->host_irq, prop, needs_inv);
295
296         return 0;
297 }
298
299 /*
300  * Create a snapshot of the current LPIs targeting @vcpu, so that we can
301  * enumerate those LPIs without holding any lock.
302  * Returns their number and puts the kmalloc'ed array into intid_ptr.
303  */
304 int vgic_copy_lpi_list(struct kvm *kvm, struct kvm_vcpu *vcpu, u32 **intid_ptr)
305 {
306         struct vgic_dist *dist = &kvm->arch.vgic;
307         struct vgic_irq *irq;
308         unsigned long flags;
309         u32 *intids;
310         int irq_count, i = 0;
311
312         /*
313          * There is an obvious race between allocating the array and LPIs
314          * being mapped/unmapped. If we ended up here as a result of a
315          * command, we're safe (locks are held, preventing another
316          * command). If coming from another path (such as enabling LPIs),
317          * we must be careful not to overrun the array.
318          */
319         irq_count = READ_ONCE(dist->lpi_list_count);
320         intids = kmalloc_array(irq_count, sizeof(intids[0]), GFP_KERNEL);
321         if (!intids)
322                 return -ENOMEM;
323
324         raw_spin_lock_irqsave(&dist->lpi_list_lock, flags);
325         list_for_each_entry(irq, &dist->lpi_list_head, lpi_list) {
326                 if (i == irq_count)
327                         break;
328                 /* We don't need to "get" the IRQ, as we hold the list lock. */
329                 if (vcpu && irq->target_vcpu != vcpu)
330                         continue;
331                 intids[i++] = irq->intid;
332         }
333         raw_spin_unlock_irqrestore(&dist->lpi_list_lock, flags);
334
335         *intid_ptr = intids;
336         return i;
337 }
338
339 static int update_affinity(struct vgic_irq *irq, struct kvm_vcpu *vcpu)
340 {
341         int ret = 0;
342         unsigned long flags;
343
344         raw_spin_lock_irqsave(&irq->irq_lock, flags);
345         irq->target_vcpu = vcpu;
346         raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
347
348         if (irq->hw) {
349                 struct its_vlpi_map map;
350
351                 ret = its_get_vlpi(irq->host_irq, &map);
352                 if (ret)
353                         return ret;
354
355                 map.vpe = &vcpu->arch.vgic_cpu.vgic_v3.its_vpe;
356
357                 ret = its_map_vlpi(irq->host_irq, &map);
358         }
359
360         return ret;
361 }
362
363 /*
364  * Promotes the ITS view of affinity of an ITTE (which redistributor this LPI
365  * is targeting) to the VGIC's view, which deals with target VCPUs.
366  * Needs to be called whenever either the collection for a LPIs has
367  * changed or the collection itself got retargeted.
368  */
369 static void update_affinity_ite(struct kvm *kvm, struct its_ite *ite)
370 {
371         struct kvm_vcpu *vcpu;
372
373         if (!its_is_collection_mapped(ite->collection))
374                 return;
375
376         vcpu = kvm_get_vcpu(kvm, ite->collection->target_addr);
377         update_affinity(ite->irq, vcpu);
378 }
379
380 /*
381  * Updates the target VCPU for every LPI targeting this collection.
382  * Must be called with the its_lock mutex held.
383  */
384 static void update_affinity_collection(struct kvm *kvm, struct vgic_its *its,
385                                        struct its_collection *coll)
386 {
387         struct its_device *device;
388         struct its_ite *ite;
389
390         for_each_lpi_its(device, ite, its) {
391                 if (!ite->collection || coll != ite->collection)
392                         continue;
393
394                 update_affinity_ite(kvm, ite);
395         }
396 }
397
398 static u32 max_lpis_propbaser(u64 propbaser)
399 {
400         int nr_idbits = (propbaser & 0x1f) + 1;
401
402         return 1U << min(nr_idbits, INTERRUPT_ID_BITS_ITS);
403 }
404
405 /*
406  * Sync the pending table pending bit of LPIs targeting @vcpu
407  * with our own data structures. This relies on the LPI being
408  * mapped before.
409  */
410 static int its_sync_lpi_pending_table(struct kvm_vcpu *vcpu)
411 {
412         gpa_t pendbase = GICR_PENDBASER_ADDRESS(vcpu->arch.vgic_cpu.pendbaser);
413         struct vgic_irq *irq;
414         int last_byte_offset = -1;
415         int ret = 0;
416         u32 *intids;
417         int nr_irqs, i;
418         unsigned long flags;
419         u8 pendmask;
420
421         nr_irqs = vgic_copy_lpi_list(vcpu->kvm, vcpu, &intids);
422         if (nr_irqs < 0)
423                 return nr_irqs;
424
425         for (i = 0; i < nr_irqs; i++) {
426                 int byte_offset, bit_nr;
427
428                 byte_offset = intids[i] / BITS_PER_BYTE;
429                 bit_nr = intids[i] % BITS_PER_BYTE;
430
431                 /*
432                  * For contiguously allocated LPIs chances are we just read
433                  * this very same byte in the last iteration. Reuse that.
434                  */
435                 if (byte_offset != last_byte_offset) {
436                         ret = kvm_read_guest_lock(vcpu->kvm,
437                                                   pendbase + byte_offset,
438                                                   &pendmask, 1);
439                         if (ret) {
440                                 kfree(intids);
441                                 return ret;
442                         }
443                         last_byte_offset = byte_offset;
444                 }
445
446                 irq = vgic_get_irq(vcpu->kvm, NULL, intids[i]);
447                 raw_spin_lock_irqsave(&irq->irq_lock, flags);
448                 irq->pending_latch = pendmask & (1U << bit_nr);
449                 vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
450                 vgic_put_irq(vcpu->kvm, irq);
451         }
452
453         kfree(intids);
454
455         return ret;
456 }
457
458 static unsigned long vgic_mmio_read_its_typer(struct kvm *kvm,
459                                               struct vgic_its *its,
460                                               gpa_t addr, unsigned int len)
461 {
462         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
463         u64 reg = GITS_TYPER_PLPIS;
464
465         /*
466          * We use linear CPU numbers for redistributor addressing,
467          * so GITS_TYPER.PTA is 0.
468          * Also we force all PROPBASER registers to be the same, so
469          * CommonLPIAff is 0 as well.
470          * To avoid memory waste in the guest, we keep the number of IDBits and
471          * DevBits low - as least for the time being.
472          */
473         reg |= GIC_ENCODE_SZ(VITS_TYPER_DEVBITS, 5) << GITS_TYPER_DEVBITS_SHIFT;
474         reg |= GIC_ENCODE_SZ(VITS_TYPER_IDBITS, 5) << GITS_TYPER_IDBITS_SHIFT;
475         reg |= GIC_ENCODE_SZ(abi->ite_esz, 4) << GITS_TYPER_ITT_ENTRY_SIZE_SHIFT;
476
477         return extract_bytes(reg, addr & 7, len);
478 }
479
480 static unsigned long vgic_mmio_read_its_iidr(struct kvm *kvm,
481                                              struct vgic_its *its,
482                                              gpa_t addr, unsigned int len)
483 {
484         u32 val;
485
486         val = (its->abi_rev << GITS_IIDR_REV_SHIFT) & GITS_IIDR_REV_MASK;
487         val |= (PRODUCT_ID_KVM << GITS_IIDR_PRODUCTID_SHIFT) | IMPLEMENTER_ARM;
488         return val;
489 }
490
491 static int vgic_mmio_uaccess_write_its_iidr(struct kvm *kvm,
492                                             struct vgic_its *its,
493                                             gpa_t addr, unsigned int len,
494                                             unsigned long val)
495 {
496         u32 rev = GITS_IIDR_REV(val);
497
498         if (rev >= NR_ITS_ABIS)
499                 return -EINVAL;
500         return vgic_its_set_abi(its, rev);
501 }
502
503 static unsigned long vgic_mmio_read_its_idregs(struct kvm *kvm,
504                                                struct vgic_its *its,
505                                                gpa_t addr, unsigned int len)
506 {
507         switch (addr & 0xffff) {
508         case GITS_PIDR0:
509                 return 0x92;    /* part number, bits[7:0] */
510         case GITS_PIDR1:
511                 return 0xb4;    /* part number, bits[11:8] */
512         case GITS_PIDR2:
513                 return GIC_PIDR2_ARCH_GICv3 | 0x0b;
514         case GITS_PIDR4:
515                 return 0x40;    /* This is a 64K software visible page */
516         /* The following are the ID registers for (any) GIC. */
517         case GITS_CIDR0:
518                 return 0x0d;
519         case GITS_CIDR1:
520                 return 0xf0;
521         case GITS_CIDR2:
522                 return 0x05;
523         case GITS_CIDR3:
524                 return 0xb1;
525         }
526
527         return 0;
528 }
529
530 int vgic_its_resolve_lpi(struct kvm *kvm, struct vgic_its *its,
531                          u32 devid, u32 eventid, struct vgic_irq **irq)
532 {
533         struct kvm_vcpu *vcpu;
534         struct its_ite *ite;
535
536         if (!its->enabled)
537                 return -EBUSY;
538
539         ite = find_ite(its, devid, eventid);
540         if (!ite || !its_is_collection_mapped(ite->collection))
541                 return E_ITS_INT_UNMAPPED_INTERRUPT;
542
543         vcpu = kvm_get_vcpu(kvm, ite->collection->target_addr);
544         if (!vcpu)
545                 return E_ITS_INT_UNMAPPED_INTERRUPT;
546
547         if (!vcpu->arch.vgic_cpu.lpis_enabled)
548                 return -EBUSY;
549
550         *irq = ite->irq;
551         return 0;
552 }
553
554 struct vgic_its *vgic_msi_to_its(struct kvm *kvm, struct kvm_msi *msi)
555 {
556         u64 address;
557         struct kvm_io_device *kvm_io_dev;
558         struct vgic_io_device *iodev;
559
560         if (!vgic_has_its(kvm))
561                 return ERR_PTR(-ENODEV);
562
563         if (!(msi->flags & KVM_MSI_VALID_DEVID))
564                 return ERR_PTR(-EINVAL);
565
566         address = (u64)msi->address_hi << 32 | msi->address_lo;
567
568         kvm_io_dev = kvm_io_bus_get_dev(kvm, KVM_MMIO_BUS, address);
569         if (!kvm_io_dev)
570                 return ERR_PTR(-EINVAL);
571
572         if (kvm_io_dev->ops != &kvm_io_gic_ops)
573                 return ERR_PTR(-EINVAL);
574
575         iodev = container_of(kvm_io_dev, struct vgic_io_device, dev);
576         if (iodev->iodev_type != IODEV_ITS)
577                 return ERR_PTR(-EINVAL);
578
579         return iodev->its;
580 }
581
582 /*
583  * Find the target VCPU and the LPI number for a given devid/eventid pair
584  * and make this IRQ pending, possibly injecting it.
585  * Must be called with the its_lock mutex held.
586  * Returns 0 on success, a positive error value for any ITS mapping
587  * related errors and negative error values for generic errors.
588  */
589 static int vgic_its_trigger_msi(struct kvm *kvm, struct vgic_its *its,
590                                 u32 devid, u32 eventid)
591 {
592         struct vgic_irq *irq = NULL;
593         unsigned long flags;
594         int err;
595
596         err = vgic_its_resolve_lpi(kvm, its, devid, eventid, &irq);
597         if (err)
598                 return err;
599
600         if (irq->hw)
601                 return irq_set_irqchip_state(irq->host_irq,
602                                              IRQCHIP_STATE_PENDING, true);
603
604         raw_spin_lock_irqsave(&irq->irq_lock, flags);
605         irq->pending_latch = true;
606         vgic_queue_irq_unlock(kvm, irq, flags);
607
608         return 0;
609 }
610
611 /*
612  * Queries the KVM IO bus framework to get the ITS pointer from the given
613  * doorbell address.
614  * We then call vgic_its_trigger_msi() with the decoded data.
615  * According to the KVM_SIGNAL_MSI API description returns 1 on success.
616  */
617 int vgic_its_inject_msi(struct kvm *kvm, struct kvm_msi *msi)
618 {
619         struct vgic_its *its;
620         int ret;
621
622         its = vgic_msi_to_its(kvm, msi);
623         if (IS_ERR(its))
624                 return PTR_ERR(its);
625
626         mutex_lock(&its->its_lock);
627         ret = vgic_its_trigger_msi(kvm, its, msi->devid, msi->data);
628         mutex_unlock(&its->its_lock);
629
630         if (ret < 0)
631                 return ret;
632
633         /*
634          * KVM_SIGNAL_MSI demands a return value > 0 for success and 0
635          * if the guest has blocked the MSI. So we map any LPI mapping
636          * related error to that.
637          */
638         if (ret)
639                 return 0;
640         else
641                 return 1;
642 }
643
644 /* Requires the its_lock to be held. */
645 static void its_free_ite(struct kvm *kvm, struct its_ite *ite)
646 {
647         list_del(&ite->ite_list);
648
649         /* This put matches the get in vgic_add_lpi. */
650         if (ite->irq) {
651                 if (ite->irq->hw)
652                         WARN_ON(its_unmap_vlpi(ite->irq->host_irq));
653
654                 vgic_put_irq(kvm, ite->irq);
655         }
656
657         kfree(ite);
658 }
659
660 static u64 its_cmd_mask_field(u64 *its_cmd, int word, int shift, int size)
661 {
662         return (le64_to_cpu(its_cmd[word]) >> shift) & (BIT_ULL(size) - 1);
663 }
664
665 #define its_cmd_get_command(cmd)        its_cmd_mask_field(cmd, 0,  0,  8)
666 #define its_cmd_get_deviceid(cmd)       its_cmd_mask_field(cmd, 0, 32, 32)
667 #define its_cmd_get_size(cmd)           (its_cmd_mask_field(cmd, 1,  0,  5) + 1)
668 #define its_cmd_get_id(cmd)             its_cmd_mask_field(cmd, 1,  0, 32)
669 #define its_cmd_get_physical_id(cmd)    its_cmd_mask_field(cmd, 1, 32, 32)
670 #define its_cmd_get_collection(cmd)     its_cmd_mask_field(cmd, 2,  0, 16)
671 #define its_cmd_get_ittaddr(cmd)        (its_cmd_mask_field(cmd, 2,  8, 44) << 8)
672 #define its_cmd_get_target_addr(cmd)    its_cmd_mask_field(cmd, 2, 16, 32)
673 #define its_cmd_get_validbit(cmd)       its_cmd_mask_field(cmd, 2, 63,  1)
674
675 /*
676  * The DISCARD command frees an Interrupt Translation Table Entry (ITTE).
677  * Must be called with the its_lock mutex held.
678  */
679 static int vgic_its_cmd_handle_discard(struct kvm *kvm, struct vgic_its *its,
680                                        u64 *its_cmd)
681 {
682         u32 device_id = its_cmd_get_deviceid(its_cmd);
683         u32 event_id = its_cmd_get_id(its_cmd);
684         struct its_ite *ite;
685
686
687         ite = find_ite(its, device_id, event_id);
688         if (ite && ite->collection) {
689                 /*
690                  * Though the spec talks about removing the pending state, we
691                  * don't bother here since we clear the ITTE anyway and the
692                  * pending state is a property of the ITTE struct.
693                  */
694                 its_free_ite(kvm, ite);
695                 return 0;
696         }
697
698         return E_ITS_DISCARD_UNMAPPED_INTERRUPT;
699 }
700
701 /*
702  * The MOVI command moves an ITTE to a different collection.
703  * Must be called with the its_lock mutex held.
704  */
705 static int vgic_its_cmd_handle_movi(struct kvm *kvm, struct vgic_its *its,
706                                     u64 *its_cmd)
707 {
708         u32 device_id = its_cmd_get_deviceid(its_cmd);
709         u32 event_id = its_cmd_get_id(its_cmd);
710         u32 coll_id = its_cmd_get_collection(its_cmd);
711         struct kvm_vcpu *vcpu;
712         struct its_ite *ite;
713         struct its_collection *collection;
714
715         ite = find_ite(its, device_id, event_id);
716         if (!ite)
717                 return E_ITS_MOVI_UNMAPPED_INTERRUPT;
718
719         if (!its_is_collection_mapped(ite->collection))
720                 return E_ITS_MOVI_UNMAPPED_COLLECTION;
721
722         collection = find_collection(its, coll_id);
723         if (!its_is_collection_mapped(collection))
724                 return E_ITS_MOVI_UNMAPPED_COLLECTION;
725
726         ite->collection = collection;
727         vcpu = kvm_get_vcpu(kvm, collection->target_addr);
728
729         return update_affinity(ite->irq, vcpu);
730 }
731
732 /*
733  * Check whether an ID can be stored into the corresponding guest table.
734  * For a direct table this is pretty easy, but gets a bit nasty for
735  * indirect tables. We check whether the resulting guest physical address
736  * is actually valid (covered by a memslot and guest accessible).
737  * For this we have to read the respective first level entry.
738  */
739 static bool vgic_its_check_id(struct vgic_its *its, u64 baser, u32 id,
740                               gpa_t *eaddr)
741 {
742         int l1_tbl_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
743         u64 indirect_ptr, type = GITS_BASER_TYPE(baser);
744         phys_addr_t base = GITS_BASER_ADDR_48_to_52(baser);
745         int esz = GITS_BASER_ENTRY_SIZE(baser);
746         int index, idx;
747         gfn_t gfn;
748         bool ret;
749
750         switch (type) {
751         case GITS_BASER_TYPE_DEVICE:
752                 if (id >= BIT_ULL(VITS_TYPER_DEVBITS))
753                         return false;
754                 break;
755         case GITS_BASER_TYPE_COLLECTION:
756                 /* as GITS_TYPER.CIL == 0, ITS supports 16-bit collection ID */
757                 if (id >= BIT_ULL(16))
758                         return false;
759                 break;
760         default:
761                 return false;
762         }
763
764         if (!(baser & GITS_BASER_INDIRECT)) {
765                 phys_addr_t addr;
766
767                 if (id >= (l1_tbl_size / esz))
768                         return false;
769
770                 addr = base + id * esz;
771                 gfn = addr >> PAGE_SHIFT;
772
773                 if (eaddr)
774                         *eaddr = addr;
775
776                 goto out;
777         }
778
779         /* calculate and check the index into the 1st level */
780         index = id / (SZ_64K / esz);
781         if (index >= (l1_tbl_size / sizeof(u64)))
782                 return false;
783
784         /* Each 1st level entry is represented by a 64-bit value. */
785         if (kvm_read_guest_lock(its->dev->kvm,
786                            base + index * sizeof(indirect_ptr),
787                            &indirect_ptr, sizeof(indirect_ptr)))
788                 return false;
789
790         indirect_ptr = le64_to_cpu(indirect_ptr);
791
792         /* check the valid bit of the first level entry */
793         if (!(indirect_ptr & BIT_ULL(63)))
794                 return false;
795
796         /* Mask the guest physical address and calculate the frame number. */
797         indirect_ptr &= GENMASK_ULL(51, 16);
798
799         /* Find the address of the actual entry */
800         index = id % (SZ_64K / esz);
801         indirect_ptr += index * esz;
802         gfn = indirect_ptr >> PAGE_SHIFT;
803
804         if (eaddr)
805                 *eaddr = indirect_ptr;
806
807 out:
808         idx = srcu_read_lock(&its->dev->kvm->srcu);
809         ret = kvm_is_visible_gfn(its->dev->kvm, gfn);
810         srcu_read_unlock(&its->dev->kvm->srcu, idx);
811         return ret;
812 }
813
814 static int vgic_its_alloc_collection(struct vgic_its *its,
815                                      struct its_collection **colp,
816                                      u32 coll_id)
817 {
818         struct its_collection *collection;
819
820         if (!vgic_its_check_id(its, its->baser_coll_table, coll_id, NULL))
821                 return E_ITS_MAPC_COLLECTION_OOR;
822
823         collection = kzalloc(sizeof(*collection), GFP_KERNEL);
824         if (!collection)
825                 return -ENOMEM;
826
827         collection->collection_id = coll_id;
828         collection->target_addr = COLLECTION_NOT_MAPPED;
829
830         list_add_tail(&collection->coll_list, &its->collection_list);
831         *colp = collection;
832
833         return 0;
834 }
835
836 static void vgic_its_free_collection(struct vgic_its *its, u32 coll_id)
837 {
838         struct its_collection *collection;
839         struct its_device *device;
840         struct its_ite *ite;
841
842         /*
843          * Clearing the mapping for that collection ID removes the
844          * entry from the list. If there wasn't any before, we can
845          * go home early.
846          */
847         collection = find_collection(its, coll_id);
848         if (!collection)
849                 return;
850
851         for_each_lpi_its(device, ite, its)
852                 if (ite->collection &&
853                     ite->collection->collection_id == coll_id)
854                         ite->collection = NULL;
855
856         list_del(&collection->coll_list);
857         kfree(collection);
858 }
859
860 /* Must be called with its_lock mutex held */
861 static struct its_ite *vgic_its_alloc_ite(struct its_device *device,
862                                           struct its_collection *collection,
863                                           u32 event_id)
864 {
865         struct its_ite *ite;
866
867         ite = kzalloc(sizeof(*ite), GFP_KERNEL);
868         if (!ite)
869                 return ERR_PTR(-ENOMEM);
870
871         ite->event_id   = event_id;
872         ite->collection = collection;
873
874         list_add_tail(&ite->ite_list, &device->itt_head);
875         return ite;
876 }
877
878 /*
879  * The MAPTI and MAPI commands map LPIs to ITTEs.
880  * Must be called with its_lock mutex held.
881  */
882 static int vgic_its_cmd_handle_mapi(struct kvm *kvm, struct vgic_its *its,
883                                     u64 *its_cmd)
884 {
885         u32 device_id = its_cmd_get_deviceid(its_cmd);
886         u32 event_id = its_cmd_get_id(its_cmd);
887         u32 coll_id = its_cmd_get_collection(its_cmd);
888         struct its_ite *ite;
889         struct kvm_vcpu *vcpu = NULL;
890         struct its_device *device;
891         struct its_collection *collection, *new_coll = NULL;
892         struct vgic_irq *irq;
893         int lpi_nr;
894
895         device = find_its_device(its, device_id);
896         if (!device)
897                 return E_ITS_MAPTI_UNMAPPED_DEVICE;
898
899         if (event_id >= BIT_ULL(device->num_eventid_bits))
900                 return E_ITS_MAPTI_ID_OOR;
901
902         if (its_cmd_get_command(its_cmd) == GITS_CMD_MAPTI)
903                 lpi_nr = its_cmd_get_physical_id(its_cmd);
904         else
905                 lpi_nr = event_id;
906         if (lpi_nr < GIC_LPI_OFFSET ||
907             lpi_nr >= max_lpis_propbaser(kvm->arch.vgic.propbaser))
908                 return E_ITS_MAPTI_PHYSICALID_OOR;
909
910         /* If there is an existing mapping, behavior is UNPREDICTABLE. */
911         if (find_ite(its, device_id, event_id))
912                 return 0;
913
914         collection = find_collection(its, coll_id);
915         if (!collection) {
916                 int ret = vgic_its_alloc_collection(its, &collection, coll_id);
917                 if (ret)
918                         return ret;
919                 new_coll = collection;
920         }
921
922         ite = vgic_its_alloc_ite(device, collection, event_id);
923         if (IS_ERR(ite)) {
924                 if (new_coll)
925                         vgic_its_free_collection(its, coll_id);
926                 return PTR_ERR(ite);
927         }
928
929         if (its_is_collection_mapped(collection))
930                 vcpu = kvm_get_vcpu(kvm, collection->target_addr);
931
932         irq = vgic_add_lpi(kvm, lpi_nr, vcpu);
933         if (IS_ERR(irq)) {
934                 if (new_coll)
935                         vgic_its_free_collection(its, coll_id);
936                 its_free_ite(kvm, ite);
937                 return PTR_ERR(irq);
938         }
939         ite->irq = irq;
940
941         return 0;
942 }
943
944 /* Requires the its_lock to be held. */
945 static void vgic_its_free_device(struct kvm *kvm, struct its_device *device)
946 {
947         struct its_ite *ite, *temp;
948
949         /*
950          * The spec says that unmapping a device with still valid
951          * ITTEs associated is UNPREDICTABLE. We remove all ITTEs,
952          * since we cannot leave the memory unreferenced.
953          */
954         list_for_each_entry_safe(ite, temp, &device->itt_head, ite_list)
955                 its_free_ite(kvm, ite);
956
957         list_del(&device->dev_list);
958         kfree(device);
959 }
960
961 /* its lock must be held */
962 static void vgic_its_free_device_list(struct kvm *kvm, struct vgic_its *its)
963 {
964         struct its_device *cur, *temp;
965
966         list_for_each_entry_safe(cur, temp, &its->device_list, dev_list)
967                 vgic_its_free_device(kvm, cur);
968 }
969
970 /* its lock must be held */
971 static void vgic_its_free_collection_list(struct kvm *kvm, struct vgic_its *its)
972 {
973         struct its_collection *cur, *temp;
974
975         list_for_each_entry_safe(cur, temp, &its->collection_list, coll_list)
976                 vgic_its_free_collection(its, cur->collection_id);
977 }
978
979 /* Must be called with its_lock mutex held */
980 static struct its_device *vgic_its_alloc_device(struct vgic_its *its,
981                                                 u32 device_id, gpa_t itt_addr,
982                                                 u8 num_eventid_bits)
983 {
984         struct its_device *device;
985
986         device = kzalloc(sizeof(*device), GFP_KERNEL);
987         if (!device)
988                 return ERR_PTR(-ENOMEM);
989
990         device->device_id = device_id;
991         device->itt_addr = itt_addr;
992         device->num_eventid_bits = num_eventid_bits;
993         INIT_LIST_HEAD(&device->itt_head);
994
995         list_add_tail(&device->dev_list, &its->device_list);
996         return device;
997 }
998
999 /*
1000  * MAPD maps or unmaps a device ID to Interrupt Translation Tables (ITTs).
1001  * Must be called with the its_lock mutex held.
1002  */
1003 static int vgic_its_cmd_handle_mapd(struct kvm *kvm, struct vgic_its *its,
1004                                     u64 *its_cmd)
1005 {
1006         u32 device_id = its_cmd_get_deviceid(its_cmd);
1007         bool valid = its_cmd_get_validbit(its_cmd);
1008         u8 num_eventid_bits = its_cmd_get_size(its_cmd);
1009         gpa_t itt_addr = its_cmd_get_ittaddr(its_cmd);
1010         struct its_device *device;
1011
1012         if (!vgic_its_check_id(its, its->baser_device_table, device_id, NULL))
1013                 return E_ITS_MAPD_DEVICE_OOR;
1014
1015         if (valid && num_eventid_bits > VITS_TYPER_IDBITS)
1016                 return E_ITS_MAPD_ITTSIZE_OOR;
1017
1018         device = find_its_device(its, device_id);
1019
1020         /*
1021          * The spec says that calling MAPD on an already mapped device
1022          * invalidates all cached data for this device. We implement this
1023          * by removing the mapping and re-establishing it.
1024          */
1025         if (device)
1026                 vgic_its_free_device(kvm, device);
1027
1028         /*
1029          * The spec does not say whether unmapping a not-mapped device
1030          * is an error, so we are done in any case.
1031          */
1032         if (!valid)
1033                 return 0;
1034
1035         device = vgic_its_alloc_device(its, device_id, itt_addr,
1036                                        num_eventid_bits);
1037
1038         return PTR_ERR_OR_ZERO(device);
1039 }
1040
1041 /*
1042  * The MAPC command maps collection IDs to redistributors.
1043  * Must be called with the its_lock mutex held.
1044  */
1045 static int vgic_its_cmd_handle_mapc(struct kvm *kvm, struct vgic_its *its,
1046                                     u64 *its_cmd)
1047 {
1048         u16 coll_id;
1049         u32 target_addr;
1050         struct its_collection *collection;
1051         bool valid;
1052
1053         valid = its_cmd_get_validbit(its_cmd);
1054         coll_id = its_cmd_get_collection(its_cmd);
1055         target_addr = its_cmd_get_target_addr(its_cmd);
1056
1057         if (target_addr >= atomic_read(&kvm->online_vcpus))
1058                 return E_ITS_MAPC_PROCNUM_OOR;
1059
1060         if (!valid) {
1061                 vgic_its_free_collection(its, coll_id);
1062         } else {
1063                 collection = find_collection(its, coll_id);
1064
1065                 if (!collection) {
1066                         int ret;
1067
1068                         ret = vgic_its_alloc_collection(its, &collection,
1069                                                         coll_id);
1070                         if (ret)
1071                                 return ret;
1072                         collection->target_addr = target_addr;
1073                 } else {
1074                         collection->target_addr = target_addr;
1075                         update_affinity_collection(kvm, its, collection);
1076                 }
1077         }
1078
1079         return 0;
1080 }
1081
1082 /*
1083  * The CLEAR command removes the pending state for a particular LPI.
1084  * Must be called with the its_lock mutex held.
1085  */
1086 static int vgic_its_cmd_handle_clear(struct kvm *kvm, struct vgic_its *its,
1087                                      u64 *its_cmd)
1088 {
1089         u32 device_id = its_cmd_get_deviceid(its_cmd);
1090         u32 event_id = its_cmd_get_id(its_cmd);
1091         struct its_ite *ite;
1092
1093
1094         ite = find_ite(its, device_id, event_id);
1095         if (!ite)
1096                 return E_ITS_CLEAR_UNMAPPED_INTERRUPT;
1097
1098         ite->irq->pending_latch = false;
1099
1100         if (ite->irq->hw)
1101                 return irq_set_irqchip_state(ite->irq->host_irq,
1102                                              IRQCHIP_STATE_PENDING, false);
1103
1104         return 0;
1105 }
1106
1107 /*
1108  * The INV command syncs the configuration bits from the memory table.
1109  * Must be called with the its_lock mutex held.
1110  */
1111 static int vgic_its_cmd_handle_inv(struct kvm *kvm, struct vgic_its *its,
1112                                    u64 *its_cmd)
1113 {
1114         u32 device_id = its_cmd_get_deviceid(its_cmd);
1115         u32 event_id = its_cmd_get_id(its_cmd);
1116         struct its_ite *ite;
1117
1118
1119         ite = find_ite(its, device_id, event_id);
1120         if (!ite)
1121                 return E_ITS_INV_UNMAPPED_INTERRUPT;
1122
1123         return update_lpi_config(kvm, ite->irq, NULL, true);
1124 }
1125
1126 /*
1127  * The INVALL command requests flushing of all IRQ data in this collection.
1128  * Find the VCPU mapped to that collection, then iterate over the VM's list
1129  * of mapped LPIs and update the configuration for each IRQ which targets
1130  * the specified vcpu. The configuration will be read from the in-memory
1131  * configuration table.
1132  * Must be called with the its_lock mutex held.
1133  */
1134 static int vgic_its_cmd_handle_invall(struct kvm *kvm, struct vgic_its *its,
1135                                       u64 *its_cmd)
1136 {
1137         u32 coll_id = its_cmd_get_collection(its_cmd);
1138         struct its_collection *collection;
1139         struct kvm_vcpu *vcpu;
1140         struct vgic_irq *irq;
1141         u32 *intids;
1142         int irq_count, i;
1143
1144         collection = find_collection(its, coll_id);
1145         if (!its_is_collection_mapped(collection))
1146                 return E_ITS_INVALL_UNMAPPED_COLLECTION;
1147
1148         vcpu = kvm_get_vcpu(kvm, collection->target_addr);
1149
1150         irq_count = vgic_copy_lpi_list(kvm, vcpu, &intids);
1151         if (irq_count < 0)
1152                 return irq_count;
1153
1154         for (i = 0; i < irq_count; i++) {
1155                 irq = vgic_get_irq(kvm, NULL, intids[i]);
1156                 if (!irq)
1157                         continue;
1158                 update_lpi_config(kvm, irq, vcpu, false);
1159                 vgic_put_irq(kvm, irq);
1160         }
1161
1162         kfree(intids);
1163
1164         if (vcpu->arch.vgic_cpu.vgic_v3.its_vpe.its_vm)
1165                 its_invall_vpe(&vcpu->arch.vgic_cpu.vgic_v3.its_vpe);
1166
1167         return 0;
1168 }
1169
1170 /*
1171  * The MOVALL command moves the pending state of all IRQs targeting one
1172  * redistributor to another. We don't hold the pending state in the VCPUs,
1173  * but in the IRQs instead, so there is really not much to do for us here.
1174  * However the spec says that no IRQ must target the old redistributor
1175  * afterwards, so we make sure that no LPI is using the associated target_vcpu.
1176  * This command affects all LPIs in the system that target that redistributor.
1177  */
1178 static int vgic_its_cmd_handle_movall(struct kvm *kvm, struct vgic_its *its,
1179                                       u64 *its_cmd)
1180 {
1181         u32 target1_addr = its_cmd_get_target_addr(its_cmd);
1182         u32 target2_addr = its_cmd_mask_field(its_cmd, 3, 16, 32);
1183         struct kvm_vcpu *vcpu1, *vcpu2;
1184         struct vgic_irq *irq;
1185         u32 *intids;
1186         int irq_count, i;
1187
1188         if (target1_addr >= atomic_read(&kvm->online_vcpus) ||
1189             target2_addr >= atomic_read(&kvm->online_vcpus))
1190                 return E_ITS_MOVALL_PROCNUM_OOR;
1191
1192         if (target1_addr == target2_addr)
1193                 return 0;
1194
1195         vcpu1 = kvm_get_vcpu(kvm, target1_addr);
1196         vcpu2 = kvm_get_vcpu(kvm, target2_addr);
1197
1198         irq_count = vgic_copy_lpi_list(kvm, vcpu1, &intids);
1199         if (irq_count < 0)
1200                 return irq_count;
1201
1202         for (i = 0; i < irq_count; i++) {
1203                 irq = vgic_get_irq(kvm, NULL, intids[i]);
1204
1205                 update_affinity(irq, vcpu2);
1206
1207                 vgic_put_irq(kvm, irq);
1208         }
1209
1210         kfree(intids);
1211         return 0;
1212 }
1213
1214 /*
1215  * The INT command injects the LPI associated with that DevID/EvID pair.
1216  * Must be called with the its_lock mutex held.
1217  */
1218 static int vgic_its_cmd_handle_int(struct kvm *kvm, struct vgic_its *its,
1219                                    u64 *its_cmd)
1220 {
1221         u32 msi_data = its_cmd_get_id(its_cmd);
1222         u64 msi_devid = its_cmd_get_deviceid(its_cmd);
1223
1224         return vgic_its_trigger_msi(kvm, its, msi_devid, msi_data);
1225 }
1226
1227 /*
1228  * This function is called with the its_cmd lock held, but the ITS data
1229  * structure lock dropped.
1230  */
1231 static int vgic_its_handle_command(struct kvm *kvm, struct vgic_its *its,
1232                                    u64 *its_cmd)
1233 {
1234         int ret = -ENODEV;
1235
1236         mutex_lock(&its->its_lock);
1237         switch (its_cmd_get_command(its_cmd)) {
1238         case GITS_CMD_MAPD:
1239                 ret = vgic_its_cmd_handle_mapd(kvm, its, its_cmd);
1240                 break;
1241         case GITS_CMD_MAPC:
1242                 ret = vgic_its_cmd_handle_mapc(kvm, its, its_cmd);
1243                 break;
1244         case GITS_CMD_MAPI:
1245                 ret = vgic_its_cmd_handle_mapi(kvm, its, its_cmd);
1246                 break;
1247         case GITS_CMD_MAPTI:
1248                 ret = vgic_its_cmd_handle_mapi(kvm, its, its_cmd);
1249                 break;
1250         case GITS_CMD_MOVI:
1251                 ret = vgic_its_cmd_handle_movi(kvm, its, its_cmd);
1252                 break;
1253         case GITS_CMD_DISCARD:
1254                 ret = vgic_its_cmd_handle_discard(kvm, its, its_cmd);
1255                 break;
1256         case GITS_CMD_CLEAR:
1257                 ret = vgic_its_cmd_handle_clear(kvm, its, its_cmd);
1258                 break;
1259         case GITS_CMD_MOVALL:
1260                 ret = vgic_its_cmd_handle_movall(kvm, its, its_cmd);
1261                 break;
1262         case GITS_CMD_INT:
1263                 ret = vgic_its_cmd_handle_int(kvm, its, its_cmd);
1264                 break;
1265         case GITS_CMD_INV:
1266                 ret = vgic_its_cmd_handle_inv(kvm, its, its_cmd);
1267                 break;
1268         case GITS_CMD_INVALL:
1269                 ret = vgic_its_cmd_handle_invall(kvm, its, its_cmd);
1270                 break;
1271         case GITS_CMD_SYNC:
1272                 /* we ignore this command: we are in sync all of the time */
1273                 ret = 0;
1274                 break;
1275         }
1276         mutex_unlock(&its->its_lock);
1277
1278         return ret;
1279 }
1280
1281 static u64 vgic_sanitise_its_baser(u64 reg)
1282 {
1283         reg = vgic_sanitise_field(reg, GITS_BASER_SHAREABILITY_MASK,
1284                                   GITS_BASER_SHAREABILITY_SHIFT,
1285                                   vgic_sanitise_shareability);
1286         reg = vgic_sanitise_field(reg, GITS_BASER_INNER_CACHEABILITY_MASK,
1287                                   GITS_BASER_INNER_CACHEABILITY_SHIFT,
1288                                   vgic_sanitise_inner_cacheability);
1289         reg = vgic_sanitise_field(reg, GITS_BASER_OUTER_CACHEABILITY_MASK,
1290                                   GITS_BASER_OUTER_CACHEABILITY_SHIFT,
1291                                   vgic_sanitise_outer_cacheability);
1292
1293         /* We support only one (ITS) page size: 64K */
1294         reg = (reg & ~GITS_BASER_PAGE_SIZE_MASK) | GITS_BASER_PAGE_SIZE_64K;
1295
1296         return reg;
1297 }
1298
1299 static u64 vgic_sanitise_its_cbaser(u64 reg)
1300 {
1301         reg = vgic_sanitise_field(reg, GITS_CBASER_SHAREABILITY_MASK,
1302                                   GITS_CBASER_SHAREABILITY_SHIFT,
1303                                   vgic_sanitise_shareability);
1304         reg = vgic_sanitise_field(reg, GITS_CBASER_INNER_CACHEABILITY_MASK,
1305                                   GITS_CBASER_INNER_CACHEABILITY_SHIFT,
1306                                   vgic_sanitise_inner_cacheability);
1307         reg = vgic_sanitise_field(reg, GITS_CBASER_OUTER_CACHEABILITY_MASK,
1308                                   GITS_CBASER_OUTER_CACHEABILITY_SHIFT,
1309                                   vgic_sanitise_outer_cacheability);
1310
1311         /* Sanitise the physical address to be 64k aligned. */
1312         reg &= ~GENMASK_ULL(15, 12);
1313
1314         return reg;
1315 }
1316
1317 static unsigned long vgic_mmio_read_its_cbaser(struct kvm *kvm,
1318                                                struct vgic_its *its,
1319                                                gpa_t addr, unsigned int len)
1320 {
1321         return extract_bytes(its->cbaser, addr & 7, len);
1322 }
1323
1324 static void vgic_mmio_write_its_cbaser(struct kvm *kvm, struct vgic_its *its,
1325                                        gpa_t addr, unsigned int len,
1326                                        unsigned long val)
1327 {
1328         /* When GITS_CTLR.Enable is 1, this register is RO. */
1329         if (its->enabled)
1330                 return;
1331
1332         mutex_lock(&its->cmd_lock);
1333         its->cbaser = update_64bit_reg(its->cbaser, addr & 7, len, val);
1334         its->cbaser = vgic_sanitise_its_cbaser(its->cbaser);
1335         its->creadr = 0;
1336         /*
1337          * CWRITER is architecturally UNKNOWN on reset, but we need to reset
1338          * it to CREADR to make sure we start with an empty command buffer.
1339          */
1340         its->cwriter = its->creadr;
1341         mutex_unlock(&its->cmd_lock);
1342 }
1343
1344 #define ITS_CMD_BUFFER_SIZE(baser)      ((((baser) & 0xff) + 1) << 12)
1345 #define ITS_CMD_SIZE                    32
1346 #define ITS_CMD_OFFSET(reg)             ((reg) & GENMASK(19, 5))
1347
1348 /* Must be called with the cmd_lock held. */
1349 static void vgic_its_process_commands(struct kvm *kvm, struct vgic_its *its)
1350 {
1351         gpa_t cbaser;
1352         u64 cmd_buf[4];
1353
1354         /* Commands are only processed when the ITS is enabled. */
1355         if (!its->enabled)
1356                 return;
1357
1358         cbaser = GITS_CBASER_ADDRESS(its->cbaser);
1359
1360         while (its->cwriter != its->creadr) {
1361                 int ret = kvm_read_guest_lock(kvm, cbaser + its->creadr,
1362                                               cmd_buf, ITS_CMD_SIZE);
1363                 /*
1364                  * If kvm_read_guest() fails, this could be due to the guest
1365                  * programming a bogus value in CBASER or something else going
1366                  * wrong from which we cannot easily recover.
1367                  * According to section 6.3.2 in the GICv3 spec we can just
1368                  * ignore that command then.
1369                  */
1370                 if (!ret)
1371                         vgic_its_handle_command(kvm, its, cmd_buf);
1372
1373                 its->creadr += ITS_CMD_SIZE;
1374                 if (its->creadr == ITS_CMD_BUFFER_SIZE(its->cbaser))
1375                         its->creadr = 0;
1376         }
1377 }
1378
1379 /*
1380  * By writing to CWRITER the guest announces new commands to be processed.
1381  * To avoid any races in the first place, we take the its_cmd lock, which
1382  * protects our ring buffer variables, so that there is only one user
1383  * per ITS handling commands at a given time.
1384  */
1385 static void vgic_mmio_write_its_cwriter(struct kvm *kvm, struct vgic_its *its,
1386                                         gpa_t addr, unsigned int len,
1387                                         unsigned long val)
1388 {
1389         u64 reg;
1390
1391         if (!its)
1392                 return;
1393
1394         mutex_lock(&its->cmd_lock);
1395
1396         reg = update_64bit_reg(its->cwriter, addr & 7, len, val);
1397         reg = ITS_CMD_OFFSET(reg);
1398         if (reg >= ITS_CMD_BUFFER_SIZE(its->cbaser)) {
1399                 mutex_unlock(&its->cmd_lock);
1400                 return;
1401         }
1402         its->cwriter = reg;
1403
1404         vgic_its_process_commands(kvm, its);
1405
1406         mutex_unlock(&its->cmd_lock);
1407 }
1408
1409 static unsigned long vgic_mmio_read_its_cwriter(struct kvm *kvm,
1410                                                 struct vgic_its *its,
1411                                                 gpa_t addr, unsigned int len)
1412 {
1413         return extract_bytes(its->cwriter, addr & 0x7, len);
1414 }
1415
1416 static unsigned long vgic_mmio_read_its_creadr(struct kvm *kvm,
1417                                                struct vgic_its *its,
1418                                                gpa_t addr, unsigned int len)
1419 {
1420         return extract_bytes(its->creadr, addr & 0x7, len);
1421 }
1422
1423 static int vgic_mmio_uaccess_write_its_creadr(struct kvm *kvm,
1424                                               struct vgic_its *its,
1425                                               gpa_t addr, unsigned int len,
1426                                               unsigned long val)
1427 {
1428         u32 cmd_offset;
1429         int ret = 0;
1430
1431         mutex_lock(&its->cmd_lock);
1432
1433         if (its->enabled) {
1434                 ret = -EBUSY;
1435                 goto out;
1436         }
1437
1438         cmd_offset = ITS_CMD_OFFSET(val);
1439         if (cmd_offset >= ITS_CMD_BUFFER_SIZE(its->cbaser)) {
1440                 ret = -EINVAL;
1441                 goto out;
1442         }
1443
1444         its->creadr = cmd_offset;
1445 out:
1446         mutex_unlock(&its->cmd_lock);
1447         return ret;
1448 }
1449
1450 #define BASER_INDEX(addr) (((addr) / sizeof(u64)) & 0x7)
1451 static unsigned long vgic_mmio_read_its_baser(struct kvm *kvm,
1452                                               struct vgic_its *its,
1453                                               gpa_t addr, unsigned int len)
1454 {
1455         u64 reg;
1456
1457         switch (BASER_INDEX(addr)) {
1458         case 0:
1459                 reg = its->baser_device_table;
1460                 break;
1461         case 1:
1462                 reg = its->baser_coll_table;
1463                 break;
1464         default:
1465                 reg = 0;
1466                 break;
1467         }
1468
1469         return extract_bytes(reg, addr & 7, len);
1470 }
1471
1472 #define GITS_BASER_RO_MASK      (GENMASK_ULL(52, 48) | GENMASK_ULL(58, 56))
1473 static void vgic_mmio_write_its_baser(struct kvm *kvm,
1474                                       struct vgic_its *its,
1475                                       gpa_t addr, unsigned int len,
1476                                       unsigned long val)
1477 {
1478         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
1479         u64 entry_size, table_type;
1480         u64 reg, *regptr, clearbits = 0;
1481
1482         /* When GITS_CTLR.Enable is 1, we ignore write accesses. */
1483         if (its->enabled)
1484                 return;
1485
1486         switch (BASER_INDEX(addr)) {
1487         case 0:
1488                 regptr = &its->baser_device_table;
1489                 entry_size = abi->dte_esz;
1490                 table_type = GITS_BASER_TYPE_DEVICE;
1491                 break;
1492         case 1:
1493                 regptr = &its->baser_coll_table;
1494                 entry_size = abi->cte_esz;
1495                 table_type = GITS_BASER_TYPE_COLLECTION;
1496                 clearbits = GITS_BASER_INDIRECT;
1497                 break;
1498         default:
1499                 return;
1500         }
1501
1502         reg = update_64bit_reg(*regptr, addr & 7, len, val);
1503         reg &= ~GITS_BASER_RO_MASK;
1504         reg &= ~clearbits;
1505
1506         reg |= (entry_size - 1) << GITS_BASER_ENTRY_SIZE_SHIFT;
1507         reg |= table_type << GITS_BASER_TYPE_SHIFT;
1508         reg = vgic_sanitise_its_baser(reg);
1509
1510         *regptr = reg;
1511
1512         if (!(reg & GITS_BASER_VALID)) {
1513                 /* Take the its_lock to prevent a race with a save/restore */
1514                 mutex_lock(&its->its_lock);
1515                 switch (table_type) {
1516                 case GITS_BASER_TYPE_DEVICE:
1517                         vgic_its_free_device_list(kvm, its);
1518                         break;
1519                 case GITS_BASER_TYPE_COLLECTION:
1520                         vgic_its_free_collection_list(kvm, its);
1521                         break;
1522                 }
1523                 mutex_unlock(&its->its_lock);
1524         }
1525 }
1526
1527 static unsigned long vgic_mmio_read_its_ctlr(struct kvm *vcpu,
1528                                              struct vgic_its *its,
1529                                              gpa_t addr, unsigned int len)
1530 {
1531         u32 reg = 0;
1532
1533         mutex_lock(&its->cmd_lock);
1534         if (its->creadr == its->cwriter)
1535                 reg |= GITS_CTLR_QUIESCENT;
1536         if (its->enabled)
1537                 reg |= GITS_CTLR_ENABLE;
1538         mutex_unlock(&its->cmd_lock);
1539
1540         return reg;
1541 }
1542
1543 static void vgic_mmio_write_its_ctlr(struct kvm *kvm, struct vgic_its *its,
1544                                      gpa_t addr, unsigned int len,
1545                                      unsigned long val)
1546 {
1547         mutex_lock(&its->cmd_lock);
1548
1549         /*
1550          * It is UNPREDICTABLE to enable the ITS if any of the CBASER or
1551          * device/collection BASER are invalid
1552          */
1553         if (!its->enabled && (val & GITS_CTLR_ENABLE) &&
1554                 (!(its->baser_device_table & GITS_BASER_VALID) ||
1555                  !(its->baser_coll_table & GITS_BASER_VALID) ||
1556                  !(its->cbaser & GITS_CBASER_VALID)))
1557                 goto out;
1558
1559         its->enabled = !!(val & GITS_CTLR_ENABLE);
1560
1561         /*
1562          * Try to process any pending commands. This function bails out early
1563          * if the ITS is disabled or no commands have been queued.
1564          */
1565         vgic_its_process_commands(kvm, its);
1566
1567 out:
1568         mutex_unlock(&its->cmd_lock);
1569 }
1570
1571 #define REGISTER_ITS_DESC(off, rd, wr, length, acc)             \
1572 {                                                               \
1573         .reg_offset = off,                                      \
1574         .len = length,                                          \
1575         .access_flags = acc,                                    \
1576         .its_read = rd,                                         \
1577         .its_write = wr,                                        \
1578 }
1579
1580 #define REGISTER_ITS_DESC_UACCESS(off, rd, wr, uwr, length, acc)\
1581 {                                                               \
1582         .reg_offset = off,                                      \
1583         .len = length,                                          \
1584         .access_flags = acc,                                    \
1585         .its_read = rd,                                         \
1586         .its_write = wr,                                        \
1587         .uaccess_its_write = uwr,                               \
1588 }
1589
1590 static void its_mmio_write_wi(struct kvm *kvm, struct vgic_its *its,
1591                               gpa_t addr, unsigned int len, unsigned long val)
1592 {
1593         /* Ignore */
1594 }
1595
1596 static struct vgic_register_region its_registers[] = {
1597         REGISTER_ITS_DESC(GITS_CTLR,
1598                 vgic_mmio_read_its_ctlr, vgic_mmio_write_its_ctlr, 4,
1599                 VGIC_ACCESS_32bit),
1600         REGISTER_ITS_DESC_UACCESS(GITS_IIDR,
1601                 vgic_mmio_read_its_iidr, its_mmio_write_wi,
1602                 vgic_mmio_uaccess_write_its_iidr, 4,
1603                 VGIC_ACCESS_32bit),
1604         REGISTER_ITS_DESC(GITS_TYPER,
1605                 vgic_mmio_read_its_typer, its_mmio_write_wi, 8,
1606                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1607         REGISTER_ITS_DESC(GITS_CBASER,
1608                 vgic_mmio_read_its_cbaser, vgic_mmio_write_its_cbaser, 8,
1609                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1610         REGISTER_ITS_DESC(GITS_CWRITER,
1611                 vgic_mmio_read_its_cwriter, vgic_mmio_write_its_cwriter, 8,
1612                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1613         REGISTER_ITS_DESC_UACCESS(GITS_CREADR,
1614                 vgic_mmio_read_its_creadr, its_mmio_write_wi,
1615                 vgic_mmio_uaccess_write_its_creadr, 8,
1616                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1617         REGISTER_ITS_DESC(GITS_BASER,
1618                 vgic_mmio_read_its_baser, vgic_mmio_write_its_baser, 0x40,
1619                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1620         REGISTER_ITS_DESC(GITS_IDREGS_BASE,
1621                 vgic_mmio_read_its_idregs, its_mmio_write_wi, 0x30,
1622                 VGIC_ACCESS_32bit),
1623 };
1624
1625 /* This is called on setting the LPI enable bit in the redistributor. */
1626 void vgic_enable_lpis(struct kvm_vcpu *vcpu)
1627 {
1628         if (!(vcpu->arch.vgic_cpu.pendbaser & GICR_PENDBASER_PTZ))
1629                 its_sync_lpi_pending_table(vcpu);
1630 }
1631
1632 static int vgic_register_its_iodev(struct kvm *kvm, struct vgic_its *its,
1633                                    u64 addr)
1634 {
1635         struct vgic_io_device *iodev = &its->iodev;
1636         int ret;
1637
1638         mutex_lock(&kvm->slots_lock);
1639         if (!IS_VGIC_ADDR_UNDEF(its->vgic_its_base)) {
1640                 ret = -EBUSY;
1641                 goto out;
1642         }
1643
1644         its->vgic_its_base = addr;
1645         iodev->regions = its_registers;
1646         iodev->nr_regions = ARRAY_SIZE(its_registers);
1647         kvm_iodevice_init(&iodev->dev, &kvm_io_gic_ops);
1648
1649         iodev->base_addr = its->vgic_its_base;
1650         iodev->iodev_type = IODEV_ITS;
1651         iodev->its = its;
1652         ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, iodev->base_addr,
1653                                       KVM_VGIC_V3_ITS_SIZE, &iodev->dev);
1654 out:
1655         mutex_unlock(&kvm->slots_lock);
1656
1657         return ret;
1658 }
1659
1660 #define INITIAL_BASER_VALUE                                               \
1661         (GIC_BASER_CACHEABILITY(GITS_BASER, INNER, RaWb)                | \
1662          GIC_BASER_CACHEABILITY(GITS_BASER, OUTER, SameAsInner)         | \
1663          GIC_BASER_SHAREABILITY(GITS_BASER, InnerShareable)             | \
1664          GITS_BASER_PAGE_SIZE_64K)
1665
1666 #define INITIAL_PROPBASER_VALUE                                           \
1667         (GIC_BASER_CACHEABILITY(GICR_PROPBASER, INNER, RaWb)            | \
1668          GIC_BASER_CACHEABILITY(GICR_PROPBASER, OUTER, SameAsInner)     | \
1669          GIC_BASER_SHAREABILITY(GICR_PROPBASER, InnerShareable))
1670
1671 static int vgic_its_create(struct kvm_device *dev, u32 type)
1672 {
1673         struct vgic_its *its;
1674
1675         if (type != KVM_DEV_TYPE_ARM_VGIC_ITS)
1676                 return -ENODEV;
1677
1678         its = kzalloc(sizeof(struct vgic_its), GFP_KERNEL);
1679         if (!its)
1680                 return -ENOMEM;
1681
1682         if (vgic_initialized(dev->kvm)) {
1683                 int ret = vgic_v4_init(dev->kvm);
1684                 if (ret < 0) {
1685                         kfree(its);
1686                         return ret;
1687                 }
1688         }
1689
1690         mutex_init(&its->its_lock);
1691         mutex_init(&its->cmd_lock);
1692
1693         its->vgic_its_base = VGIC_ADDR_UNDEF;
1694
1695         INIT_LIST_HEAD(&its->device_list);
1696         INIT_LIST_HEAD(&its->collection_list);
1697
1698         dev->kvm->arch.vgic.msis_require_devid = true;
1699         dev->kvm->arch.vgic.has_its = true;
1700         its->enabled = false;
1701         its->dev = dev;
1702
1703         its->baser_device_table = INITIAL_BASER_VALUE                   |
1704                 ((u64)GITS_BASER_TYPE_DEVICE << GITS_BASER_TYPE_SHIFT);
1705         its->baser_coll_table = INITIAL_BASER_VALUE |
1706                 ((u64)GITS_BASER_TYPE_COLLECTION << GITS_BASER_TYPE_SHIFT);
1707         dev->kvm->arch.vgic.propbaser = INITIAL_PROPBASER_VALUE;
1708
1709         dev->private = its;
1710
1711         return vgic_its_set_abi(its, NR_ITS_ABIS - 1);
1712 }
1713
1714 static void vgic_its_destroy(struct kvm_device *kvm_dev)
1715 {
1716         struct kvm *kvm = kvm_dev->kvm;
1717         struct vgic_its *its = kvm_dev->private;
1718
1719         mutex_lock(&its->its_lock);
1720
1721         vgic_its_free_device_list(kvm, its);
1722         vgic_its_free_collection_list(kvm, its);
1723
1724         mutex_unlock(&its->its_lock);
1725         kfree(its);
1726         kfree(kvm_dev);/* alloc by kvm_ioctl_create_device, free by .destroy */
1727 }
1728
1729 static int vgic_its_has_attr_regs(struct kvm_device *dev,
1730                                   struct kvm_device_attr *attr)
1731 {
1732         const struct vgic_register_region *region;
1733         gpa_t offset = attr->attr;
1734         int align;
1735
1736         align = (offset < GITS_TYPER) || (offset >= GITS_PIDR4) ? 0x3 : 0x7;
1737
1738         if (offset & align)
1739                 return -EINVAL;
1740
1741         region = vgic_find_mmio_region(its_registers,
1742                                        ARRAY_SIZE(its_registers),
1743                                        offset);
1744         if (!region)
1745                 return -ENXIO;
1746
1747         return 0;
1748 }
1749
1750 static int vgic_its_attr_regs_access(struct kvm_device *dev,
1751                                      struct kvm_device_attr *attr,
1752                                      u64 *reg, bool is_write)
1753 {
1754         const struct vgic_register_region *region;
1755         struct vgic_its *its;
1756         gpa_t addr, offset;
1757         unsigned int len;
1758         int align, ret = 0;
1759
1760         its = dev->private;
1761         offset = attr->attr;
1762
1763         /*
1764          * Although the spec supports upper/lower 32-bit accesses to
1765          * 64-bit ITS registers, the userspace ABI requires 64-bit
1766          * accesses to all 64-bit wide registers. We therefore only
1767          * support 32-bit accesses to GITS_CTLR, GITS_IIDR and GITS ID
1768          * registers
1769          */
1770         if ((offset < GITS_TYPER) || (offset >= GITS_PIDR4))
1771                 align = 0x3;
1772         else
1773                 align = 0x7;
1774
1775         if (offset & align)
1776                 return -EINVAL;
1777
1778         mutex_lock(&dev->kvm->lock);
1779
1780         if (IS_VGIC_ADDR_UNDEF(its->vgic_its_base)) {
1781                 ret = -ENXIO;
1782                 goto out;
1783         }
1784
1785         region = vgic_find_mmio_region(its_registers,
1786                                        ARRAY_SIZE(its_registers),
1787                                        offset);
1788         if (!region) {
1789                 ret = -ENXIO;
1790                 goto out;
1791         }
1792
1793         if (!lock_all_vcpus(dev->kvm)) {
1794                 ret = -EBUSY;
1795                 goto out;
1796         }
1797
1798         addr = its->vgic_its_base + offset;
1799
1800         len = region->access_flags & VGIC_ACCESS_64bit ? 8 : 4;
1801
1802         if (is_write) {
1803                 if (region->uaccess_its_write)
1804                         ret = region->uaccess_its_write(dev->kvm, its, addr,
1805                                                         len, *reg);
1806                 else
1807                         region->its_write(dev->kvm, its, addr, len, *reg);
1808         } else {
1809                 *reg = region->its_read(dev->kvm, its, addr, len);
1810         }
1811         unlock_all_vcpus(dev->kvm);
1812 out:
1813         mutex_unlock(&dev->kvm->lock);
1814         return ret;
1815 }
1816
1817 static u32 compute_next_devid_offset(struct list_head *h,
1818                                      struct its_device *dev)
1819 {
1820         struct its_device *next;
1821         u32 next_offset;
1822
1823         if (list_is_last(&dev->dev_list, h))
1824                 return 0;
1825         next = list_next_entry(dev, dev_list);
1826         next_offset = next->device_id - dev->device_id;
1827
1828         return min_t(u32, next_offset, VITS_DTE_MAX_DEVID_OFFSET);
1829 }
1830
1831 static u32 compute_next_eventid_offset(struct list_head *h, struct its_ite *ite)
1832 {
1833         struct its_ite *next;
1834         u32 next_offset;
1835
1836         if (list_is_last(&ite->ite_list, h))
1837                 return 0;
1838         next = list_next_entry(ite, ite_list);
1839         next_offset = next->event_id - ite->event_id;
1840
1841         return min_t(u32, next_offset, VITS_ITE_MAX_EVENTID_OFFSET);
1842 }
1843
1844 /**
1845  * entry_fn_t - Callback called on a table entry restore path
1846  * @its: its handle
1847  * @id: id of the entry
1848  * @entry: pointer to the entry
1849  * @opaque: pointer to an opaque data
1850  *
1851  * Return: < 0 on error, 0 if last element was identified, id offset to next
1852  * element otherwise
1853  */
1854 typedef int (*entry_fn_t)(struct vgic_its *its, u32 id, void *entry,
1855                           void *opaque);
1856
1857 /**
1858  * scan_its_table - Scan a contiguous table in guest RAM and applies a function
1859  * to each entry
1860  *
1861  * @its: its handle
1862  * @base: base gpa of the table
1863  * @size: size of the table in bytes
1864  * @esz: entry size in bytes
1865  * @start_id: the ID of the first entry in the table
1866  * (non zero for 2d level tables)
1867  * @fn: function to apply on each entry
1868  *
1869  * Return: < 0 on error, 0 if last element was identified, 1 otherwise
1870  * (the last element may not be found on second level tables)
1871  */
1872 static int scan_its_table(struct vgic_its *its, gpa_t base, int size, u32 esz,
1873                           int start_id, entry_fn_t fn, void *opaque)
1874 {
1875         struct kvm *kvm = its->dev->kvm;
1876         unsigned long len = size;
1877         int id = start_id;
1878         gpa_t gpa = base;
1879         char entry[ESZ_MAX];
1880         int ret;
1881
1882         memset(entry, 0, esz);
1883
1884         while (len > 0) {
1885                 int next_offset;
1886                 size_t byte_offset;
1887
1888                 ret = kvm_read_guest_lock(kvm, gpa, entry, esz);
1889                 if (ret)
1890                         return ret;
1891
1892                 next_offset = fn(its, id, entry, opaque);
1893                 if (next_offset <= 0)
1894                         return next_offset;
1895
1896                 byte_offset = next_offset * esz;
1897                 id += next_offset;
1898                 gpa += byte_offset;
1899                 len -= byte_offset;
1900         }
1901         return 1;
1902 }
1903
1904 /**
1905  * vgic_its_save_ite - Save an interrupt translation entry at @gpa
1906  */
1907 static int vgic_its_save_ite(struct vgic_its *its, struct its_device *dev,
1908                               struct its_ite *ite, gpa_t gpa, int ite_esz)
1909 {
1910         struct kvm *kvm = its->dev->kvm;
1911         u32 next_offset;
1912         u64 val;
1913
1914         next_offset = compute_next_eventid_offset(&dev->itt_head, ite);
1915         val = ((u64)next_offset << KVM_ITS_ITE_NEXT_SHIFT) |
1916                ((u64)ite->irq->intid << KVM_ITS_ITE_PINTID_SHIFT) |
1917                 ite->collection->collection_id;
1918         val = cpu_to_le64(val);
1919         return kvm_write_guest_lock(kvm, gpa, &val, ite_esz);
1920 }
1921
1922 /**
1923  * vgic_its_restore_ite - restore an interrupt translation entry
1924  * @event_id: id used for indexing
1925  * @ptr: pointer to the ITE entry
1926  * @opaque: pointer to the its_device
1927  */
1928 static int vgic_its_restore_ite(struct vgic_its *its, u32 event_id,
1929                                 void *ptr, void *opaque)
1930 {
1931         struct its_device *dev = (struct its_device *)opaque;
1932         struct its_collection *collection;
1933         struct kvm *kvm = its->dev->kvm;
1934         struct kvm_vcpu *vcpu = NULL;
1935         u64 val;
1936         u64 *p = (u64 *)ptr;
1937         struct vgic_irq *irq;
1938         u32 coll_id, lpi_id;
1939         struct its_ite *ite;
1940         u32 offset;
1941
1942         val = *p;
1943
1944         val = le64_to_cpu(val);
1945
1946         coll_id = val & KVM_ITS_ITE_ICID_MASK;
1947         lpi_id = (val & KVM_ITS_ITE_PINTID_MASK) >> KVM_ITS_ITE_PINTID_SHIFT;
1948
1949         if (!lpi_id)
1950                 return 1; /* invalid entry, no choice but to scan next entry */
1951
1952         if (lpi_id < VGIC_MIN_LPI)
1953                 return -EINVAL;
1954
1955         offset = val >> KVM_ITS_ITE_NEXT_SHIFT;
1956         if (event_id + offset >= BIT_ULL(dev->num_eventid_bits))
1957                 return -EINVAL;
1958
1959         collection = find_collection(its, coll_id);
1960         if (!collection)
1961                 return -EINVAL;
1962
1963         ite = vgic_its_alloc_ite(dev, collection, event_id);
1964         if (IS_ERR(ite))
1965                 return PTR_ERR(ite);
1966
1967         if (its_is_collection_mapped(collection))
1968                 vcpu = kvm_get_vcpu(kvm, collection->target_addr);
1969
1970         irq = vgic_add_lpi(kvm, lpi_id, vcpu);
1971         if (IS_ERR(irq))
1972                 return PTR_ERR(irq);
1973         ite->irq = irq;
1974
1975         return offset;
1976 }
1977
1978 static int vgic_its_ite_cmp(void *priv, struct list_head *a,
1979                             struct list_head *b)
1980 {
1981         struct its_ite *itea = container_of(a, struct its_ite, ite_list);
1982         struct its_ite *iteb = container_of(b, struct its_ite, ite_list);
1983
1984         if (itea->event_id < iteb->event_id)
1985                 return -1;
1986         else
1987                 return 1;
1988 }
1989
1990 static int vgic_its_save_itt(struct vgic_its *its, struct its_device *device)
1991 {
1992         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
1993         gpa_t base = device->itt_addr;
1994         struct its_ite *ite;
1995         int ret;
1996         int ite_esz = abi->ite_esz;
1997
1998         list_sort(NULL, &device->itt_head, vgic_its_ite_cmp);
1999
2000         list_for_each_entry(ite, &device->itt_head, ite_list) {
2001                 gpa_t gpa = base + ite->event_id * ite_esz;
2002
2003                 /*
2004                  * If an LPI carries the HW bit, this means that this
2005                  * interrupt is controlled by GICv4, and we do not
2006                  * have direct access to that state. Let's simply fail
2007                  * the save operation...
2008                  */
2009                 if (ite->irq->hw)
2010                         return -EACCES;
2011
2012                 ret = vgic_its_save_ite(its, device, ite, gpa, ite_esz);
2013                 if (ret)
2014                         return ret;
2015         }
2016         return 0;
2017 }
2018
2019 /**
2020  * vgic_its_restore_itt - restore the ITT of a device
2021  *
2022  * @its: its handle
2023  * @dev: device handle
2024  *
2025  * Return 0 on success, < 0 on error
2026  */
2027 static int vgic_its_restore_itt(struct vgic_its *its, struct its_device *dev)
2028 {
2029         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2030         gpa_t base = dev->itt_addr;
2031         int ret;
2032         int ite_esz = abi->ite_esz;
2033         size_t max_size = BIT_ULL(dev->num_eventid_bits) * ite_esz;
2034
2035         ret = scan_its_table(its, base, max_size, ite_esz, 0,
2036                              vgic_its_restore_ite, dev);
2037
2038         /* scan_its_table returns +1 if all ITEs are invalid */
2039         if (ret > 0)
2040                 ret = 0;
2041
2042         return ret;
2043 }
2044
2045 /**
2046  * vgic_its_save_dte - Save a device table entry at a given GPA
2047  *
2048  * @its: ITS handle
2049  * @dev: ITS device
2050  * @ptr: GPA
2051  */
2052 static int vgic_its_save_dte(struct vgic_its *its, struct its_device *dev,
2053                              gpa_t ptr, int dte_esz)
2054 {
2055         struct kvm *kvm = its->dev->kvm;
2056         u64 val, itt_addr_field;
2057         u32 next_offset;
2058
2059         itt_addr_field = dev->itt_addr >> 8;
2060         next_offset = compute_next_devid_offset(&its->device_list, dev);
2061         val = (1ULL << KVM_ITS_DTE_VALID_SHIFT |
2062                ((u64)next_offset << KVM_ITS_DTE_NEXT_SHIFT) |
2063                (itt_addr_field << KVM_ITS_DTE_ITTADDR_SHIFT) |
2064                 (dev->num_eventid_bits - 1));
2065         val = cpu_to_le64(val);
2066         return kvm_write_guest_lock(kvm, ptr, &val, dte_esz);
2067 }
2068
2069 /**
2070  * vgic_its_restore_dte - restore a device table entry
2071  *
2072  * @its: its handle
2073  * @id: device id the DTE corresponds to
2074  * @ptr: kernel VA where the 8 byte DTE is located
2075  * @opaque: unused
2076  *
2077  * Return: < 0 on error, 0 if the dte is the last one, id offset to the
2078  * next dte otherwise
2079  */
2080 static int vgic_its_restore_dte(struct vgic_its *its, u32 id,
2081                                 void *ptr, void *opaque)
2082 {
2083         struct its_device *dev;
2084         gpa_t itt_addr;
2085         u8 num_eventid_bits;
2086         u64 entry = *(u64 *)ptr;
2087         bool valid;
2088         u32 offset;
2089         int ret;
2090
2091         entry = le64_to_cpu(entry);
2092
2093         valid = entry >> KVM_ITS_DTE_VALID_SHIFT;
2094         num_eventid_bits = (entry & KVM_ITS_DTE_SIZE_MASK) + 1;
2095         itt_addr = ((entry & KVM_ITS_DTE_ITTADDR_MASK)
2096                         >> KVM_ITS_DTE_ITTADDR_SHIFT) << 8;
2097
2098         if (!valid)
2099                 return 1;
2100
2101         /* dte entry is valid */
2102         offset = (entry & KVM_ITS_DTE_NEXT_MASK) >> KVM_ITS_DTE_NEXT_SHIFT;
2103
2104         dev = vgic_its_alloc_device(its, id, itt_addr, num_eventid_bits);
2105         if (IS_ERR(dev))
2106                 return PTR_ERR(dev);
2107
2108         ret = vgic_its_restore_itt(its, dev);
2109         if (ret) {
2110                 vgic_its_free_device(its->dev->kvm, dev);
2111                 return ret;
2112         }
2113
2114         return offset;
2115 }
2116
2117 static int vgic_its_device_cmp(void *priv, struct list_head *a,
2118                                struct list_head *b)
2119 {
2120         struct its_device *deva = container_of(a, struct its_device, dev_list);
2121         struct its_device *devb = container_of(b, struct its_device, dev_list);
2122
2123         if (deva->device_id < devb->device_id)
2124                 return -1;
2125         else
2126                 return 1;
2127 }
2128
2129 /**
2130  * vgic_its_save_device_tables - Save the device table and all ITT
2131  * into guest RAM
2132  *
2133  * L1/L2 handling is hidden by vgic_its_check_id() helper which directly
2134  * returns the GPA of the device entry
2135  */
2136 static int vgic_its_save_device_tables(struct vgic_its *its)
2137 {
2138         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2139         u64 baser = its->baser_device_table;
2140         struct its_device *dev;
2141         int dte_esz = abi->dte_esz;
2142
2143         if (!(baser & GITS_BASER_VALID))
2144                 return 0;
2145
2146         list_sort(NULL, &its->device_list, vgic_its_device_cmp);
2147
2148         list_for_each_entry(dev, &its->device_list, dev_list) {
2149                 int ret;
2150                 gpa_t eaddr;
2151
2152                 if (!vgic_its_check_id(its, baser,
2153                                        dev->device_id, &eaddr))
2154                         return -EINVAL;
2155
2156                 ret = vgic_its_save_itt(its, dev);
2157                 if (ret)
2158                         return ret;
2159
2160                 ret = vgic_its_save_dte(its, dev, eaddr, dte_esz);
2161                 if (ret)
2162                         return ret;
2163         }
2164         return 0;
2165 }
2166
2167 /**
2168  * handle_l1_dte - callback used for L1 device table entries (2 stage case)
2169  *
2170  * @its: its handle
2171  * @id: index of the entry in the L1 table
2172  * @addr: kernel VA
2173  * @opaque: unused
2174  *
2175  * L1 table entries are scanned by steps of 1 entry
2176  * Return < 0 if error, 0 if last dte was found when scanning the L2
2177  * table, +1 otherwise (meaning next L1 entry must be scanned)
2178  */
2179 static int handle_l1_dte(struct vgic_its *its, u32 id, void *addr,
2180                          void *opaque)
2181 {
2182         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2183         int l2_start_id = id * (SZ_64K / abi->dte_esz);
2184         u64 entry = *(u64 *)addr;
2185         int dte_esz = abi->dte_esz;
2186         gpa_t gpa;
2187         int ret;
2188
2189         entry = le64_to_cpu(entry);
2190
2191         if (!(entry & KVM_ITS_L1E_VALID_MASK))
2192                 return 1;
2193
2194         gpa = entry & KVM_ITS_L1E_ADDR_MASK;
2195
2196         ret = scan_its_table(its, gpa, SZ_64K, dte_esz,
2197                              l2_start_id, vgic_its_restore_dte, NULL);
2198
2199         return ret;
2200 }
2201
2202 /**
2203  * vgic_its_restore_device_tables - Restore the device table and all ITT
2204  * from guest RAM to internal data structs
2205  */
2206 static int vgic_its_restore_device_tables(struct vgic_its *its)
2207 {
2208         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2209         u64 baser = its->baser_device_table;
2210         int l1_esz, ret;
2211         int l1_tbl_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
2212         gpa_t l1_gpa;
2213
2214         if (!(baser & GITS_BASER_VALID))
2215                 return 0;
2216
2217         l1_gpa = GITS_BASER_ADDR_48_to_52(baser);
2218
2219         if (baser & GITS_BASER_INDIRECT) {
2220                 l1_esz = GITS_LVL1_ENTRY_SIZE;
2221                 ret = scan_its_table(its, l1_gpa, l1_tbl_size, l1_esz, 0,
2222                                      handle_l1_dte, NULL);
2223         } else {
2224                 l1_esz = abi->dte_esz;
2225                 ret = scan_its_table(its, l1_gpa, l1_tbl_size, l1_esz, 0,
2226                                      vgic_its_restore_dte, NULL);
2227         }
2228
2229         /* scan_its_table returns +1 if all entries are invalid */
2230         if (ret > 0)
2231                 ret = 0;
2232
2233         return ret;
2234 }
2235
2236 static int vgic_its_save_cte(struct vgic_its *its,
2237                              struct its_collection *collection,
2238                              gpa_t gpa, int esz)
2239 {
2240         u64 val;
2241
2242         val = (1ULL << KVM_ITS_CTE_VALID_SHIFT |
2243                ((u64)collection->target_addr << KVM_ITS_CTE_RDBASE_SHIFT) |
2244                collection->collection_id);
2245         val = cpu_to_le64(val);
2246         return kvm_write_guest_lock(its->dev->kvm, gpa, &val, esz);
2247 }
2248
2249 static int vgic_its_restore_cte(struct vgic_its *its, gpa_t gpa, int esz)
2250 {
2251         struct its_collection *collection;
2252         struct kvm *kvm = its->dev->kvm;
2253         u32 target_addr, coll_id;
2254         u64 val;
2255         int ret;
2256
2257         BUG_ON(esz > sizeof(val));
2258         ret = kvm_read_guest_lock(kvm, gpa, &val, esz);
2259         if (ret)
2260                 return ret;
2261         val = le64_to_cpu(val);
2262         if (!(val & KVM_ITS_CTE_VALID_MASK))
2263                 return 0;
2264
2265         target_addr = (u32)(val >> KVM_ITS_CTE_RDBASE_SHIFT);
2266         coll_id = val & KVM_ITS_CTE_ICID_MASK;
2267
2268         if (target_addr >= atomic_read(&kvm->online_vcpus))
2269                 return -EINVAL;
2270
2271         collection = find_collection(its, coll_id);
2272         if (collection)
2273                 return -EEXIST;
2274         ret = vgic_its_alloc_collection(its, &collection, coll_id);
2275         if (ret)
2276                 return ret;
2277         collection->target_addr = target_addr;
2278         return 1;
2279 }
2280
2281 /**
2282  * vgic_its_save_collection_table - Save the collection table into
2283  * guest RAM
2284  */
2285 static int vgic_its_save_collection_table(struct vgic_its *its)
2286 {
2287         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2288         u64 baser = its->baser_coll_table;
2289         gpa_t gpa = GITS_BASER_ADDR_48_to_52(baser);
2290         struct its_collection *collection;
2291         u64 val;
2292         size_t max_size, filled = 0;
2293         int ret, cte_esz = abi->cte_esz;
2294
2295         if (!(baser & GITS_BASER_VALID))
2296                 return 0;
2297
2298         max_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
2299
2300         list_for_each_entry(collection, &its->collection_list, coll_list) {
2301                 ret = vgic_its_save_cte(its, collection, gpa, cte_esz);
2302                 if (ret)
2303                         return ret;
2304                 gpa += cte_esz;
2305                 filled += cte_esz;
2306         }
2307
2308         if (filled == max_size)
2309                 return 0;
2310
2311         /*
2312          * table is not fully filled, add a last dummy element
2313          * with valid bit unset
2314          */
2315         val = 0;
2316         BUG_ON(cte_esz > sizeof(val));
2317         ret = kvm_write_guest_lock(its->dev->kvm, gpa, &val, cte_esz);
2318         return ret;
2319 }
2320
2321 /**
2322  * vgic_its_restore_collection_table - reads the collection table
2323  * in guest memory and restores the ITS internal state. Requires the
2324  * BASER registers to be restored before.
2325  */
2326 static int vgic_its_restore_collection_table(struct vgic_its *its)
2327 {
2328         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2329         u64 baser = its->baser_coll_table;
2330         int cte_esz = abi->cte_esz;
2331         size_t max_size, read = 0;
2332         gpa_t gpa;
2333         int ret;
2334
2335         if (!(baser & GITS_BASER_VALID))
2336                 return 0;
2337
2338         gpa = GITS_BASER_ADDR_48_to_52(baser);
2339
2340         max_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
2341
2342         while (read < max_size) {
2343                 ret = vgic_its_restore_cte(its, gpa, cte_esz);
2344                 if (ret <= 0)
2345                         break;
2346                 gpa += cte_esz;
2347                 read += cte_esz;
2348         }
2349
2350         if (ret > 0)
2351                 return 0;
2352
2353         return ret;
2354 }
2355
2356 /**
2357  * vgic_its_save_tables_v0 - Save the ITS tables into guest ARM
2358  * according to v0 ABI
2359  */
2360 static int vgic_its_save_tables_v0(struct vgic_its *its)
2361 {
2362         int ret;
2363
2364         ret = vgic_its_save_device_tables(its);
2365         if (ret)
2366                 return ret;
2367
2368         return vgic_its_save_collection_table(its);
2369 }
2370
2371 /**
2372  * vgic_its_restore_tables_v0 - Restore the ITS tables from guest RAM
2373  * to internal data structs according to V0 ABI
2374  *
2375  */
2376 static int vgic_its_restore_tables_v0(struct vgic_its *its)
2377 {
2378         int ret;
2379
2380         ret = vgic_its_restore_collection_table(its);
2381         if (ret)
2382                 return ret;
2383
2384         return vgic_its_restore_device_tables(its);
2385 }
2386
2387 static int vgic_its_commit_v0(struct vgic_its *its)
2388 {
2389         const struct vgic_its_abi *abi;
2390
2391         abi = vgic_its_get_abi(its);
2392         its->baser_coll_table &= ~GITS_BASER_ENTRY_SIZE_MASK;
2393         its->baser_device_table &= ~GITS_BASER_ENTRY_SIZE_MASK;
2394
2395         its->baser_coll_table |= (GIC_ENCODE_SZ(abi->cte_esz, 5)
2396                                         << GITS_BASER_ENTRY_SIZE_SHIFT);
2397
2398         its->baser_device_table |= (GIC_ENCODE_SZ(abi->dte_esz, 5)
2399                                         << GITS_BASER_ENTRY_SIZE_SHIFT);
2400         return 0;
2401 }
2402
2403 static void vgic_its_reset(struct kvm *kvm, struct vgic_its *its)
2404 {
2405         /* We need to keep the ABI specific field values */
2406         its->baser_coll_table &= ~GITS_BASER_VALID;
2407         its->baser_device_table &= ~GITS_BASER_VALID;
2408         its->cbaser = 0;
2409         its->creadr = 0;
2410         its->cwriter = 0;
2411         its->enabled = 0;
2412         vgic_its_free_device_list(kvm, its);
2413         vgic_its_free_collection_list(kvm, its);
2414 }
2415
2416 static int vgic_its_has_attr(struct kvm_device *dev,
2417                              struct kvm_device_attr *attr)
2418 {
2419         switch (attr->group) {
2420         case KVM_DEV_ARM_VGIC_GRP_ADDR:
2421                 switch (attr->attr) {
2422                 case KVM_VGIC_ITS_ADDR_TYPE:
2423                         return 0;
2424                 }
2425                 break;
2426         case KVM_DEV_ARM_VGIC_GRP_CTRL:
2427                 switch (attr->attr) {
2428                 case KVM_DEV_ARM_VGIC_CTRL_INIT:
2429                         return 0;
2430                 case KVM_DEV_ARM_ITS_CTRL_RESET:
2431                         return 0;
2432                 case KVM_DEV_ARM_ITS_SAVE_TABLES:
2433                         return 0;
2434                 case KVM_DEV_ARM_ITS_RESTORE_TABLES:
2435                         return 0;
2436                 }
2437                 break;
2438         case KVM_DEV_ARM_VGIC_GRP_ITS_REGS:
2439                 return vgic_its_has_attr_regs(dev, attr);
2440         }
2441         return -ENXIO;
2442 }
2443
2444 static int vgic_its_ctrl(struct kvm *kvm, struct vgic_its *its, u64 attr)
2445 {
2446         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2447         int ret = 0;
2448
2449         if (attr == KVM_DEV_ARM_VGIC_CTRL_INIT) /* Nothing to do */
2450                 return 0;
2451
2452         mutex_lock(&kvm->lock);
2453         mutex_lock(&its->its_lock);
2454
2455         if (!lock_all_vcpus(kvm)) {
2456                 mutex_unlock(&its->its_lock);
2457                 mutex_unlock(&kvm->lock);
2458                 return -EBUSY;
2459         }
2460
2461         switch (attr) {
2462         case KVM_DEV_ARM_ITS_CTRL_RESET:
2463                 vgic_its_reset(kvm, its);
2464                 break;
2465         case KVM_DEV_ARM_ITS_SAVE_TABLES:
2466                 ret = abi->save_tables(its);
2467                 break;
2468         case KVM_DEV_ARM_ITS_RESTORE_TABLES:
2469                 ret = abi->restore_tables(its);
2470                 break;
2471         }
2472
2473         unlock_all_vcpus(kvm);
2474         mutex_unlock(&its->its_lock);
2475         mutex_unlock(&kvm->lock);
2476         return ret;
2477 }
2478
2479 static int vgic_its_set_attr(struct kvm_device *dev,
2480                              struct kvm_device_attr *attr)
2481 {
2482         struct vgic_its *its = dev->private;
2483         int ret;
2484
2485         switch (attr->group) {
2486         case KVM_DEV_ARM_VGIC_GRP_ADDR: {
2487                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2488                 unsigned long type = (unsigned long)attr->attr;
2489                 u64 addr;
2490
2491                 if (type != KVM_VGIC_ITS_ADDR_TYPE)
2492                         return -ENODEV;
2493
2494                 if (copy_from_user(&addr, uaddr, sizeof(addr)))
2495                         return -EFAULT;
2496
2497                 ret = vgic_check_ioaddr(dev->kvm, &its->vgic_its_base,
2498                                         addr, SZ_64K);
2499                 if (ret)
2500                         return ret;
2501
2502                 return vgic_register_its_iodev(dev->kvm, its, addr);
2503         }
2504         case KVM_DEV_ARM_VGIC_GRP_CTRL:
2505                 return vgic_its_ctrl(dev->kvm, its, attr->attr);
2506         case KVM_DEV_ARM_VGIC_GRP_ITS_REGS: {
2507                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2508                 u64 reg;
2509
2510                 if (get_user(reg, uaddr))
2511                         return -EFAULT;
2512
2513                 return vgic_its_attr_regs_access(dev, attr, &reg, true);
2514         }
2515         }
2516         return -ENXIO;
2517 }
2518
2519 static int vgic_its_get_attr(struct kvm_device *dev,
2520                              struct kvm_device_attr *attr)
2521 {
2522         switch (attr->group) {
2523         case KVM_DEV_ARM_VGIC_GRP_ADDR: {
2524                 struct vgic_its *its = dev->private;
2525                 u64 addr = its->vgic_its_base;
2526                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2527                 unsigned long type = (unsigned long)attr->attr;
2528
2529                 if (type != KVM_VGIC_ITS_ADDR_TYPE)
2530                         return -ENODEV;
2531
2532                 if (copy_to_user(uaddr, &addr, sizeof(addr)))
2533                         return -EFAULT;
2534                 break;
2535         }
2536         case KVM_DEV_ARM_VGIC_GRP_ITS_REGS: {
2537                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2538                 u64 reg;
2539                 int ret;
2540
2541                 ret = vgic_its_attr_regs_access(dev, attr, &reg, false);
2542                 if (ret)
2543                         return ret;
2544                 return put_user(reg, uaddr);
2545         }
2546         default:
2547                 return -ENXIO;
2548         }
2549
2550         return 0;
2551 }
2552
2553 static struct kvm_device_ops kvm_arm_vgic_its_ops = {
2554         .name = "kvm-arm-vgic-its",
2555         .create = vgic_its_create,
2556         .destroy = vgic_its_destroy,
2557         .set_attr = vgic_its_set_attr,
2558         .get_attr = vgic_its_get_attr,
2559         .has_attr = vgic_its_has_attr,
2560 };
2561
2562 int kvm_vgic_register_its_device(void)
2563 {
2564         return kvm_register_device_ops(&kvm_arm_vgic_its_ops,
2565                                        KVM_DEV_TYPE_ARM_VGIC_ITS);
2566 }