Merge tag 'kvm-arm-gicv4-for-v4.15' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6-block.git] / virt / kvm / arm / vgic / vgic-its.c
1 /*
2  * GICv3 ITS emulation
3  *
4  * Copyright (C) 2015,2016 ARM Ltd.
5  * Author: Andre Przywara <andre.przywara@arm.com>
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18  */
19
20 #include <linux/cpu.h>
21 #include <linux/kvm.h>
22 #include <linux/kvm_host.h>
23 #include <linux/interrupt.h>
24 #include <linux/list.h>
25 #include <linux/uaccess.h>
26 #include <linux/list_sort.h>
27
28 #include <linux/irqchip/arm-gic-v3.h>
29
30 #include <asm/kvm_emulate.h>
31 #include <asm/kvm_arm.h>
32 #include <asm/kvm_mmu.h>
33
34 #include "vgic.h"
35 #include "vgic-mmio.h"
36
37 static int vgic_its_save_tables_v0(struct vgic_its *its);
38 static int vgic_its_restore_tables_v0(struct vgic_its *its);
39 static int vgic_its_commit_v0(struct vgic_its *its);
40 static int update_lpi_config(struct kvm *kvm, struct vgic_irq *irq,
41                              struct kvm_vcpu *filter_vcpu, bool needs_inv);
42
43 /*
44  * Creates a new (reference to a) struct vgic_irq for a given LPI.
45  * If this LPI is already mapped on another ITS, we increase its refcount
46  * and return a pointer to the existing structure.
47  * If this is a "new" LPI, we allocate and initialize a new struct vgic_irq.
48  * This function returns a pointer to the _unlocked_ structure.
49  */
50 static struct vgic_irq *vgic_add_lpi(struct kvm *kvm, u32 intid,
51                                      struct kvm_vcpu *vcpu)
52 {
53         struct vgic_dist *dist = &kvm->arch.vgic;
54         struct vgic_irq *irq = vgic_get_irq(kvm, NULL, intid), *oldirq;
55         int ret;
56
57         /* In this case there is no put, since we keep the reference. */
58         if (irq)
59                 return irq;
60
61         irq = kzalloc(sizeof(struct vgic_irq), GFP_KERNEL);
62         if (!irq)
63                 return ERR_PTR(-ENOMEM);
64
65         INIT_LIST_HEAD(&irq->lpi_list);
66         INIT_LIST_HEAD(&irq->ap_list);
67         spin_lock_init(&irq->irq_lock);
68
69         irq->config = VGIC_CONFIG_EDGE;
70         kref_init(&irq->refcount);
71         irq->intid = intid;
72         irq->target_vcpu = vcpu;
73
74         spin_lock(&dist->lpi_list_lock);
75
76         /*
77          * There could be a race with another vgic_add_lpi(), so we need to
78          * check that we don't add a second list entry with the same LPI.
79          */
80         list_for_each_entry(oldirq, &dist->lpi_list_head, lpi_list) {
81                 if (oldirq->intid != intid)
82                         continue;
83
84                 /* Someone was faster with adding this LPI, lets use that. */
85                 kfree(irq);
86                 irq = oldirq;
87
88                 /*
89                  * This increases the refcount, the caller is expected to
90                  * call vgic_put_irq() on the returned pointer once it's
91                  * finished with the IRQ.
92                  */
93                 vgic_get_irq_kref(irq);
94
95                 goto out_unlock;
96         }
97
98         list_add_tail(&irq->lpi_list, &dist->lpi_list_head);
99         dist->lpi_list_count++;
100
101 out_unlock:
102         spin_unlock(&dist->lpi_list_lock);
103
104         /*
105          * We "cache" the configuration table entries in our struct vgic_irq's.
106          * However we only have those structs for mapped IRQs, so we read in
107          * the respective config data from memory here upon mapping the LPI.
108          */
109         ret = update_lpi_config(kvm, irq, NULL, false);
110         if (ret)
111                 return ERR_PTR(ret);
112
113         ret = vgic_v3_lpi_sync_pending_status(kvm, irq);
114         if (ret)
115                 return ERR_PTR(ret);
116
117         return irq;
118 }
119
120 struct its_device {
121         struct list_head dev_list;
122
123         /* the head for the list of ITTEs */
124         struct list_head itt_head;
125         u32 num_eventid_bits;
126         gpa_t itt_addr;
127         u32 device_id;
128 };
129
130 #define COLLECTION_NOT_MAPPED ((u32)~0)
131
132 struct its_collection {
133         struct list_head coll_list;
134
135         u32 collection_id;
136         u32 target_addr;
137 };
138
139 #define its_is_collection_mapped(coll) ((coll) && \
140                                 ((coll)->target_addr != COLLECTION_NOT_MAPPED))
141
142 struct its_ite {
143         struct list_head ite_list;
144
145         struct vgic_irq *irq;
146         struct its_collection *collection;
147         u32 event_id;
148 };
149
150 /**
151  * struct vgic_its_abi - ITS abi ops and settings
152  * @cte_esz: collection table entry size
153  * @dte_esz: device table entry size
154  * @ite_esz: interrupt translation table entry size
155  * @save tables: save the ITS tables into guest RAM
156  * @restore_tables: restore the ITS internal structs from tables
157  *  stored in guest RAM
158  * @commit: initialize the registers which expose the ABI settings,
159  *  especially the entry sizes
160  */
161 struct vgic_its_abi {
162         int cte_esz;
163         int dte_esz;
164         int ite_esz;
165         int (*save_tables)(struct vgic_its *its);
166         int (*restore_tables)(struct vgic_its *its);
167         int (*commit)(struct vgic_its *its);
168 };
169
170 static const struct vgic_its_abi its_table_abi_versions[] = {
171         [0] = {.cte_esz = 8, .dte_esz = 8, .ite_esz = 8,
172          .save_tables = vgic_its_save_tables_v0,
173          .restore_tables = vgic_its_restore_tables_v0,
174          .commit = vgic_its_commit_v0,
175         },
176 };
177
178 #define NR_ITS_ABIS     ARRAY_SIZE(its_table_abi_versions)
179
180 inline const struct vgic_its_abi *vgic_its_get_abi(struct vgic_its *its)
181 {
182         return &its_table_abi_versions[its->abi_rev];
183 }
184
185 int vgic_its_set_abi(struct vgic_its *its, int rev)
186 {
187         const struct vgic_its_abi *abi;
188
189         its->abi_rev = rev;
190         abi = vgic_its_get_abi(its);
191         return abi->commit(its);
192 }
193
194 /*
195  * Find and returns a device in the device table for an ITS.
196  * Must be called with the its_lock mutex held.
197  */
198 static struct its_device *find_its_device(struct vgic_its *its, u32 device_id)
199 {
200         struct its_device *device;
201
202         list_for_each_entry(device, &its->device_list, dev_list)
203                 if (device_id == device->device_id)
204                         return device;
205
206         return NULL;
207 }
208
209 /*
210  * Find and returns an interrupt translation table entry (ITTE) for a given
211  * Device ID/Event ID pair on an ITS.
212  * Must be called with the its_lock mutex held.
213  */
214 static struct its_ite *find_ite(struct vgic_its *its, u32 device_id,
215                                   u32 event_id)
216 {
217         struct its_device *device;
218         struct its_ite *ite;
219
220         device = find_its_device(its, device_id);
221         if (device == NULL)
222                 return NULL;
223
224         list_for_each_entry(ite, &device->itt_head, ite_list)
225                 if (ite->event_id == event_id)
226                         return ite;
227
228         return NULL;
229 }
230
231 /* To be used as an iterator this macro misses the enclosing parentheses */
232 #define for_each_lpi_its(dev, ite, its) \
233         list_for_each_entry(dev, &(its)->device_list, dev_list) \
234                 list_for_each_entry(ite, &(dev)->itt_head, ite_list)
235
236 /*
237  * We only implement 48 bits of PA at the moment, although the ITS
238  * supports more. Let's be restrictive here.
239  */
240 #define BASER_ADDRESS(x)        ((x) & GENMASK_ULL(47, 16))
241 #define CBASER_ADDRESS(x)       ((x) & GENMASK_ULL(47, 12))
242
243 #define GIC_LPI_OFFSET 8192
244
245 #define VITS_TYPER_IDBITS 16
246 #define VITS_TYPER_DEVBITS 16
247 #define VITS_DTE_MAX_DEVID_OFFSET       (BIT(14) - 1)
248 #define VITS_ITE_MAX_EVENTID_OFFSET     (BIT(16) - 1)
249
250 /*
251  * Finds and returns a collection in the ITS collection table.
252  * Must be called with the its_lock mutex held.
253  */
254 static struct its_collection *find_collection(struct vgic_its *its, int coll_id)
255 {
256         struct its_collection *collection;
257
258         list_for_each_entry(collection, &its->collection_list, coll_list) {
259                 if (coll_id == collection->collection_id)
260                         return collection;
261         }
262
263         return NULL;
264 }
265
266 #define LPI_PROP_ENABLE_BIT(p)  ((p) & LPI_PROP_ENABLED)
267 #define LPI_PROP_PRIORITY(p)    ((p) & 0xfc)
268
269 /*
270  * Reads the configuration data for a given LPI from guest memory and
271  * updates the fields in struct vgic_irq.
272  * If filter_vcpu is not NULL, applies only if the IRQ is targeting this
273  * VCPU. Unconditionally applies if filter_vcpu is NULL.
274  */
275 static int update_lpi_config(struct kvm *kvm, struct vgic_irq *irq,
276                              struct kvm_vcpu *filter_vcpu, bool needs_inv)
277 {
278         u64 propbase = GICR_PROPBASER_ADDRESS(kvm->arch.vgic.propbaser);
279         u8 prop;
280         int ret;
281         unsigned long flags;
282
283         ret = kvm_read_guest(kvm, propbase + irq->intid - GIC_LPI_OFFSET,
284                              &prop, 1);
285
286         if (ret)
287                 return ret;
288
289         spin_lock_irqsave(&irq->irq_lock, flags);
290
291         if (!filter_vcpu || filter_vcpu == irq->target_vcpu) {
292                 irq->priority = LPI_PROP_PRIORITY(prop);
293                 irq->enabled = LPI_PROP_ENABLE_BIT(prop);
294
295                 if (!irq->hw) {
296                         vgic_queue_irq_unlock(kvm, irq, flags);
297                         return 0;
298                 }
299         }
300
301         spin_unlock_irqrestore(&irq->irq_lock, flags);
302
303         if (irq->hw)
304                 return its_prop_update_vlpi(irq->host_irq, prop, needs_inv);
305
306         return 0;
307 }
308
309 /*
310  * Create a snapshot of the current LPIs targeting @vcpu, so that we can
311  * enumerate those LPIs without holding any lock.
312  * Returns their number and puts the kmalloc'ed array into intid_ptr.
313  */
314 static int vgic_copy_lpi_list(struct kvm_vcpu *vcpu, u32 **intid_ptr)
315 {
316         struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
317         struct vgic_irq *irq;
318         u32 *intids;
319         int irq_count = dist->lpi_list_count, i = 0;
320
321         /*
322          * We use the current value of the list length, which may change
323          * after the kmalloc. We don't care, because the guest shouldn't
324          * change anything while the command handling is still running,
325          * and in the worst case we would miss a new IRQ, which one wouldn't
326          * expect to be covered by this command anyway.
327          */
328         intids = kmalloc_array(irq_count, sizeof(intids[0]), GFP_KERNEL);
329         if (!intids)
330                 return -ENOMEM;
331
332         spin_lock(&dist->lpi_list_lock);
333         list_for_each_entry(irq, &dist->lpi_list_head, lpi_list) {
334                 /* We don't need to "get" the IRQ, as we hold the list lock. */
335                 if (irq->target_vcpu != vcpu)
336                         continue;
337                 intids[i++] = irq->intid;
338         }
339         spin_unlock(&dist->lpi_list_lock);
340
341         *intid_ptr = intids;
342         return i;
343 }
344
345 static int update_affinity(struct vgic_irq *irq, struct kvm_vcpu *vcpu)
346 {
347         int ret = 0;
348
349         spin_lock(&irq->irq_lock);
350         irq->target_vcpu = vcpu;
351         spin_unlock(&irq->irq_lock);
352
353         if (irq->hw) {
354                 struct its_vlpi_map map;
355
356                 ret = its_get_vlpi(irq->host_irq, &map);
357                 if (ret)
358                         return ret;
359
360                 map.vpe = &vcpu->arch.vgic_cpu.vgic_v3.its_vpe;
361
362                 ret = its_map_vlpi(irq->host_irq, &map);
363         }
364
365         return ret;
366 }
367
368 /*
369  * Promotes the ITS view of affinity of an ITTE (which redistributor this LPI
370  * is targeting) to the VGIC's view, which deals with target VCPUs.
371  * Needs to be called whenever either the collection for a LPIs has
372  * changed or the collection itself got retargeted.
373  */
374 static void update_affinity_ite(struct kvm *kvm, struct its_ite *ite)
375 {
376         struct kvm_vcpu *vcpu;
377
378         if (!its_is_collection_mapped(ite->collection))
379                 return;
380
381         vcpu = kvm_get_vcpu(kvm, ite->collection->target_addr);
382         update_affinity(ite->irq, vcpu);
383 }
384
385 /*
386  * Updates the target VCPU for every LPI targeting this collection.
387  * Must be called with the its_lock mutex held.
388  */
389 static void update_affinity_collection(struct kvm *kvm, struct vgic_its *its,
390                                        struct its_collection *coll)
391 {
392         struct its_device *device;
393         struct its_ite *ite;
394
395         for_each_lpi_its(device, ite, its) {
396                 if (!ite->collection || coll != ite->collection)
397                         continue;
398
399                 update_affinity_ite(kvm, ite);
400         }
401 }
402
403 static u32 max_lpis_propbaser(u64 propbaser)
404 {
405         int nr_idbits = (propbaser & 0x1f) + 1;
406
407         return 1U << min(nr_idbits, INTERRUPT_ID_BITS_ITS);
408 }
409
410 /*
411  * Sync the pending table pending bit of LPIs targeting @vcpu
412  * with our own data structures. This relies on the LPI being
413  * mapped before.
414  */
415 static int its_sync_lpi_pending_table(struct kvm_vcpu *vcpu)
416 {
417         gpa_t pendbase = GICR_PENDBASER_ADDRESS(vcpu->arch.vgic_cpu.pendbaser);
418         struct vgic_irq *irq;
419         int last_byte_offset = -1;
420         int ret = 0;
421         u32 *intids;
422         int nr_irqs, i;
423         unsigned long flags;
424
425         nr_irqs = vgic_copy_lpi_list(vcpu, &intids);
426         if (nr_irqs < 0)
427                 return nr_irqs;
428
429         for (i = 0; i < nr_irqs; i++) {
430                 int byte_offset, bit_nr;
431                 u8 pendmask;
432
433                 byte_offset = intids[i] / BITS_PER_BYTE;
434                 bit_nr = intids[i] % BITS_PER_BYTE;
435
436                 /*
437                  * For contiguously allocated LPIs chances are we just read
438                  * this very same byte in the last iteration. Reuse that.
439                  */
440                 if (byte_offset != last_byte_offset) {
441                         ret = kvm_read_guest(vcpu->kvm, pendbase + byte_offset,
442                                              &pendmask, 1);
443                         if (ret) {
444                                 kfree(intids);
445                                 return ret;
446                         }
447                         last_byte_offset = byte_offset;
448                 }
449
450                 irq = vgic_get_irq(vcpu->kvm, NULL, intids[i]);
451                 spin_lock_irqsave(&irq->irq_lock, flags);
452                 irq->pending_latch = pendmask & (1U << bit_nr);
453                 vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
454                 vgic_put_irq(vcpu->kvm, irq);
455         }
456
457         kfree(intids);
458
459         return ret;
460 }
461
462 static unsigned long vgic_mmio_read_its_typer(struct kvm *kvm,
463                                               struct vgic_its *its,
464                                               gpa_t addr, unsigned int len)
465 {
466         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
467         u64 reg = GITS_TYPER_PLPIS;
468
469         /*
470          * We use linear CPU numbers for redistributor addressing,
471          * so GITS_TYPER.PTA is 0.
472          * Also we force all PROPBASER registers to be the same, so
473          * CommonLPIAff is 0 as well.
474          * To avoid memory waste in the guest, we keep the number of IDBits and
475          * DevBits low - as least for the time being.
476          */
477         reg |= GIC_ENCODE_SZ(VITS_TYPER_DEVBITS, 5) << GITS_TYPER_DEVBITS_SHIFT;
478         reg |= GIC_ENCODE_SZ(VITS_TYPER_IDBITS, 5) << GITS_TYPER_IDBITS_SHIFT;
479         reg |= GIC_ENCODE_SZ(abi->ite_esz, 4) << GITS_TYPER_ITT_ENTRY_SIZE_SHIFT;
480
481         return extract_bytes(reg, addr & 7, len);
482 }
483
484 static unsigned long vgic_mmio_read_its_iidr(struct kvm *kvm,
485                                              struct vgic_its *its,
486                                              gpa_t addr, unsigned int len)
487 {
488         u32 val;
489
490         val = (its->abi_rev << GITS_IIDR_REV_SHIFT) & GITS_IIDR_REV_MASK;
491         val |= (PRODUCT_ID_KVM << GITS_IIDR_PRODUCTID_SHIFT) | IMPLEMENTER_ARM;
492         return val;
493 }
494
495 static int vgic_mmio_uaccess_write_its_iidr(struct kvm *kvm,
496                                             struct vgic_its *its,
497                                             gpa_t addr, unsigned int len,
498                                             unsigned long val)
499 {
500         u32 rev = GITS_IIDR_REV(val);
501
502         if (rev >= NR_ITS_ABIS)
503                 return -EINVAL;
504         return vgic_its_set_abi(its, rev);
505 }
506
507 static unsigned long vgic_mmio_read_its_idregs(struct kvm *kvm,
508                                                struct vgic_its *its,
509                                                gpa_t addr, unsigned int len)
510 {
511         switch (addr & 0xffff) {
512         case GITS_PIDR0:
513                 return 0x92;    /* part number, bits[7:0] */
514         case GITS_PIDR1:
515                 return 0xb4;    /* part number, bits[11:8] */
516         case GITS_PIDR2:
517                 return GIC_PIDR2_ARCH_GICv3 | 0x0b;
518         case GITS_PIDR4:
519                 return 0x40;    /* This is a 64K software visible page */
520         /* The following are the ID registers for (any) GIC. */
521         case GITS_CIDR0:
522                 return 0x0d;
523         case GITS_CIDR1:
524                 return 0xf0;
525         case GITS_CIDR2:
526                 return 0x05;
527         case GITS_CIDR3:
528                 return 0xb1;
529         }
530
531         return 0;
532 }
533
534 int vgic_its_resolve_lpi(struct kvm *kvm, struct vgic_its *its,
535                          u32 devid, u32 eventid, struct vgic_irq **irq)
536 {
537         struct kvm_vcpu *vcpu;
538         struct its_ite *ite;
539
540         if (!its->enabled)
541                 return -EBUSY;
542
543         ite = find_ite(its, devid, eventid);
544         if (!ite || !its_is_collection_mapped(ite->collection))
545                 return E_ITS_INT_UNMAPPED_INTERRUPT;
546
547         vcpu = kvm_get_vcpu(kvm, ite->collection->target_addr);
548         if (!vcpu)
549                 return E_ITS_INT_UNMAPPED_INTERRUPT;
550
551         if (!vcpu->arch.vgic_cpu.lpis_enabled)
552                 return -EBUSY;
553
554         *irq = ite->irq;
555         return 0;
556 }
557
558 struct vgic_its *vgic_msi_to_its(struct kvm *kvm, struct kvm_msi *msi)
559 {
560         u64 address;
561         struct kvm_io_device *kvm_io_dev;
562         struct vgic_io_device *iodev;
563
564         if (!vgic_has_its(kvm))
565                 return ERR_PTR(-ENODEV);
566
567         if (!(msi->flags & KVM_MSI_VALID_DEVID))
568                 return ERR_PTR(-EINVAL);
569
570         address = (u64)msi->address_hi << 32 | msi->address_lo;
571
572         kvm_io_dev = kvm_io_bus_get_dev(kvm, KVM_MMIO_BUS, address);
573         if (!kvm_io_dev)
574                 return ERR_PTR(-EINVAL);
575
576         if (kvm_io_dev->ops != &kvm_io_gic_ops)
577                 return ERR_PTR(-EINVAL);
578
579         iodev = container_of(kvm_io_dev, struct vgic_io_device, dev);
580         if (iodev->iodev_type != IODEV_ITS)
581                 return ERR_PTR(-EINVAL);
582
583         return iodev->its;
584 }
585
586 /*
587  * Find the target VCPU and the LPI number for a given devid/eventid pair
588  * and make this IRQ pending, possibly injecting it.
589  * Must be called with the its_lock mutex held.
590  * Returns 0 on success, a positive error value for any ITS mapping
591  * related errors and negative error values for generic errors.
592  */
593 static int vgic_its_trigger_msi(struct kvm *kvm, struct vgic_its *its,
594                                 u32 devid, u32 eventid)
595 {
596         struct vgic_irq *irq = NULL;
597         unsigned long flags;
598         int err;
599
600         err = vgic_its_resolve_lpi(kvm, its, devid, eventid, &irq);
601         if (err)
602                 return err;
603
604         if (irq->hw)
605                 return irq_set_irqchip_state(irq->host_irq,
606                                              IRQCHIP_STATE_PENDING, true);
607
608         spin_lock_irqsave(&irq->irq_lock, flags);
609         irq->pending_latch = true;
610         vgic_queue_irq_unlock(kvm, irq, flags);
611
612         return 0;
613 }
614
615 /*
616  * Queries the KVM IO bus framework to get the ITS pointer from the given
617  * doorbell address.
618  * We then call vgic_its_trigger_msi() with the decoded data.
619  * According to the KVM_SIGNAL_MSI API description returns 1 on success.
620  */
621 int vgic_its_inject_msi(struct kvm *kvm, struct kvm_msi *msi)
622 {
623         struct vgic_its *its;
624         int ret;
625
626         its = vgic_msi_to_its(kvm, msi);
627         if (IS_ERR(its))
628                 return PTR_ERR(its);
629
630         mutex_lock(&its->its_lock);
631         ret = vgic_its_trigger_msi(kvm, its, msi->devid, msi->data);
632         mutex_unlock(&its->its_lock);
633
634         if (ret < 0)
635                 return ret;
636
637         /*
638          * KVM_SIGNAL_MSI demands a return value > 0 for success and 0
639          * if the guest has blocked the MSI. So we map any LPI mapping
640          * related error to that.
641          */
642         if (ret)
643                 return 0;
644         else
645                 return 1;
646 }
647
648 /* Requires the its_lock to be held. */
649 static void its_free_ite(struct kvm *kvm, struct its_ite *ite)
650 {
651         list_del(&ite->ite_list);
652
653         /* This put matches the get in vgic_add_lpi. */
654         if (ite->irq) {
655                 if (ite->irq->hw)
656                         WARN_ON(its_unmap_vlpi(ite->irq->host_irq));
657
658                 vgic_put_irq(kvm, ite->irq);
659         }
660
661         kfree(ite);
662 }
663
664 static u64 its_cmd_mask_field(u64 *its_cmd, int word, int shift, int size)
665 {
666         return (le64_to_cpu(its_cmd[word]) >> shift) & (BIT_ULL(size) - 1);
667 }
668
669 #define its_cmd_get_command(cmd)        its_cmd_mask_field(cmd, 0,  0,  8)
670 #define its_cmd_get_deviceid(cmd)       its_cmd_mask_field(cmd, 0, 32, 32)
671 #define its_cmd_get_size(cmd)           (its_cmd_mask_field(cmd, 1,  0,  5) + 1)
672 #define its_cmd_get_id(cmd)             its_cmd_mask_field(cmd, 1,  0, 32)
673 #define its_cmd_get_physical_id(cmd)    its_cmd_mask_field(cmd, 1, 32, 32)
674 #define its_cmd_get_collection(cmd)     its_cmd_mask_field(cmd, 2,  0, 16)
675 #define its_cmd_get_ittaddr(cmd)        (its_cmd_mask_field(cmd, 2,  8, 44) << 8)
676 #define its_cmd_get_target_addr(cmd)    its_cmd_mask_field(cmd, 2, 16, 32)
677 #define its_cmd_get_validbit(cmd)       its_cmd_mask_field(cmd, 2, 63,  1)
678
679 /*
680  * The DISCARD command frees an Interrupt Translation Table Entry (ITTE).
681  * Must be called with the its_lock mutex held.
682  */
683 static int vgic_its_cmd_handle_discard(struct kvm *kvm, struct vgic_its *its,
684                                        u64 *its_cmd)
685 {
686         u32 device_id = its_cmd_get_deviceid(its_cmd);
687         u32 event_id = its_cmd_get_id(its_cmd);
688         struct its_ite *ite;
689
690
691         ite = find_ite(its, device_id, event_id);
692         if (ite && ite->collection) {
693                 /*
694                  * Though the spec talks about removing the pending state, we
695                  * don't bother here since we clear the ITTE anyway and the
696                  * pending state is a property of the ITTE struct.
697                  */
698                 its_free_ite(kvm, ite);
699                 return 0;
700         }
701
702         return E_ITS_DISCARD_UNMAPPED_INTERRUPT;
703 }
704
705 /*
706  * The MOVI command moves an ITTE to a different collection.
707  * Must be called with the its_lock mutex held.
708  */
709 static int vgic_its_cmd_handle_movi(struct kvm *kvm, struct vgic_its *its,
710                                     u64 *its_cmd)
711 {
712         u32 device_id = its_cmd_get_deviceid(its_cmd);
713         u32 event_id = its_cmd_get_id(its_cmd);
714         u32 coll_id = its_cmd_get_collection(its_cmd);
715         struct kvm_vcpu *vcpu;
716         struct its_ite *ite;
717         struct its_collection *collection;
718
719         ite = find_ite(its, device_id, event_id);
720         if (!ite)
721                 return E_ITS_MOVI_UNMAPPED_INTERRUPT;
722
723         if (!its_is_collection_mapped(ite->collection))
724                 return E_ITS_MOVI_UNMAPPED_COLLECTION;
725
726         collection = find_collection(its, coll_id);
727         if (!its_is_collection_mapped(collection))
728                 return E_ITS_MOVI_UNMAPPED_COLLECTION;
729
730         ite->collection = collection;
731         vcpu = kvm_get_vcpu(kvm, collection->target_addr);
732
733         return update_affinity(ite->irq, vcpu);
734 }
735
736 /*
737  * Check whether an ID can be stored into the corresponding guest table.
738  * For a direct table this is pretty easy, but gets a bit nasty for
739  * indirect tables. We check whether the resulting guest physical address
740  * is actually valid (covered by a memslot and guest accessible).
741  * For this we have to read the respective first level entry.
742  */
743 static bool vgic_its_check_id(struct vgic_its *its, u64 baser, u32 id,
744                               gpa_t *eaddr)
745 {
746         int l1_tbl_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
747         u64 indirect_ptr, type = GITS_BASER_TYPE(baser);
748         int esz = GITS_BASER_ENTRY_SIZE(baser);
749         int index;
750         gfn_t gfn;
751
752         switch (type) {
753         case GITS_BASER_TYPE_DEVICE:
754                 if (id >= BIT_ULL(VITS_TYPER_DEVBITS))
755                         return false;
756                 break;
757         case GITS_BASER_TYPE_COLLECTION:
758                 /* as GITS_TYPER.CIL == 0, ITS supports 16-bit collection ID */
759                 if (id >= BIT_ULL(16))
760                         return false;
761                 break;
762         default:
763                 return false;
764         }
765
766         if (!(baser & GITS_BASER_INDIRECT)) {
767                 phys_addr_t addr;
768
769                 if (id >= (l1_tbl_size / esz))
770                         return false;
771
772                 addr = BASER_ADDRESS(baser) + id * esz;
773                 gfn = addr >> PAGE_SHIFT;
774
775                 if (eaddr)
776                         *eaddr = addr;
777                 return kvm_is_visible_gfn(its->dev->kvm, gfn);
778         }
779
780         /* calculate and check the index into the 1st level */
781         index = id / (SZ_64K / esz);
782         if (index >= (l1_tbl_size / sizeof(u64)))
783                 return false;
784
785         /* Each 1st level entry is represented by a 64-bit value. */
786         if (kvm_read_guest(its->dev->kvm,
787                            BASER_ADDRESS(baser) + index * sizeof(indirect_ptr),
788                            &indirect_ptr, sizeof(indirect_ptr)))
789                 return false;
790
791         indirect_ptr = le64_to_cpu(indirect_ptr);
792
793         /* check the valid bit of the first level entry */
794         if (!(indirect_ptr & BIT_ULL(63)))
795                 return false;
796
797         /*
798          * Mask the guest physical address and calculate the frame number.
799          * Any address beyond our supported 48 bits of PA will be caught
800          * by the actual check in the final step.
801          */
802         indirect_ptr &= GENMASK_ULL(51, 16);
803
804         /* Find the address of the actual entry */
805         index = id % (SZ_64K / esz);
806         indirect_ptr += index * esz;
807         gfn = indirect_ptr >> PAGE_SHIFT;
808
809         if (eaddr)
810                 *eaddr = indirect_ptr;
811         return kvm_is_visible_gfn(its->dev->kvm, gfn);
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
825         collection->collection_id = coll_id;
826         collection->target_addr = COLLECTION_NOT_MAPPED;
827
828         list_add_tail(&collection->coll_list, &its->collection_list);
829         *colp = collection;
830
831         return 0;
832 }
833
834 static void vgic_its_free_collection(struct vgic_its *its, u32 coll_id)
835 {
836         struct its_collection *collection;
837         struct its_device *device;
838         struct its_ite *ite;
839
840         /*
841          * Clearing the mapping for that collection ID removes the
842          * entry from the list. If there wasn't any before, we can
843          * go home early.
844          */
845         collection = find_collection(its, coll_id);
846         if (!collection)
847                 return;
848
849         for_each_lpi_its(device, ite, its)
850                 if (ite->collection &&
851                     ite->collection->collection_id == coll_id)
852                         ite->collection = NULL;
853
854         list_del(&collection->coll_list);
855         kfree(collection);
856 }
857
858 /* Must be called with its_lock mutex held */
859 static struct its_ite *vgic_its_alloc_ite(struct its_device *device,
860                                           struct its_collection *collection,
861                                           u32 event_id)
862 {
863         struct its_ite *ite;
864
865         ite = kzalloc(sizeof(*ite), GFP_KERNEL);
866         if (!ite)
867                 return ERR_PTR(-ENOMEM);
868
869         ite->event_id   = event_id;
870         ite->collection = collection;
871
872         list_add_tail(&ite->ite_list, &device->itt_head);
873         return ite;
874 }
875
876 /*
877  * The MAPTI and MAPI commands map LPIs to ITTEs.
878  * Must be called with its_lock mutex held.
879  */
880 static int vgic_its_cmd_handle_mapi(struct kvm *kvm, struct vgic_its *its,
881                                     u64 *its_cmd)
882 {
883         u32 device_id = its_cmd_get_deviceid(its_cmd);
884         u32 event_id = its_cmd_get_id(its_cmd);
885         u32 coll_id = its_cmd_get_collection(its_cmd);
886         struct its_ite *ite;
887         struct kvm_vcpu *vcpu = NULL;
888         struct its_device *device;
889         struct its_collection *collection, *new_coll = NULL;
890         struct vgic_irq *irq;
891         int lpi_nr;
892
893         device = find_its_device(its, device_id);
894         if (!device)
895                 return E_ITS_MAPTI_UNMAPPED_DEVICE;
896
897         if (event_id >= BIT_ULL(device->num_eventid_bits))
898                 return E_ITS_MAPTI_ID_OOR;
899
900         if (its_cmd_get_command(its_cmd) == GITS_CMD_MAPTI)
901                 lpi_nr = its_cmd_get_physical_id(its_cmd);
902         else
903                 lpi_nr = event_id;
904         if (lpi_nr < GIC_LPI_OFFSET ||
905             lpi_nr >= max_lpis_propbaser(kvm->arch.vgic.propbaser))
906                 return E_ITS_MAPTI_PHYSICALID_OOR;
907
908         /* If there is an existing mapping, behavior is UNPREDICTABLE. */
909         if (find_ite(its, device_id, event_id))
910                 return 0;
911
912         collection = find_collection(its, coll_id);
913         if (!collection) {
914                 int ret = vgic_its_alloc_collection(its, &collection, coll_id);
915                 if (ret)
916                         return ret;
917                 new_coll = collection;
918         }
919
920         ite = vgic_its_alloc_ite(device, collection, event_id);
921         if (IS_ERR(ite)) {
922                 if (new_coll)
923                         vgic_its_free_collection(its, coll_id);
924                 return PTR_ERR(ite);
925         }
926
927         if (its_is_collection_mapped(collection))
928                 vcpu = kvm_get_vcpu(kvm, collection->target_addr);
929
930         irq = vgic_add_lpi(kvm, lpi_nr, vcpu);
931         if (IS_ERR(irq)) {
932                 if (new_coll)
933                         vgic_its_free_collection(its, coll_id);
934                 its_free_ite(kvm, ite);
935                 return PTR_ERR(irq);
936         }
937         ite->irq = irq;
938
939         return 0;
940 }
941
942 /* Requires the its_lock to be held. */
943 static void vgic_its_free_device(struct kvm *kvm, struct its_device *device)
944 {
945         struct its_ite *ite, *temp;
946
947         /*
948          * The spec says that unmapping a device with still valid
949          * ITTEs associated is UNPREDICTABLE. We remove all ITTEs,
950          * since we cannot leave the memory unreferenced.
951          */
952         list_for_each_entry_safe(ite, temp, &device->itt_head, ite_list)
953                 its_free_ite(kvm, ite);
954
955         list_del(&device->dev_list);
956         kfree(device);
957 }
958
959 /* its lock must be held */
960 static void vgic_its_free_device_list(struct kvm *kvm, struct vgic_its *its)
961 {
962         struct its_device *cur, *temp;
963
964         list_for_each_entry_safe(cur, temp, &its->device_list, dev_list)
965                 vgic_its_free_device(kvm, cur);
966 }
967
968 /* its lock must be held */
969 static void vgic_its_free_collection_list(struct kvm *kvm, struct vgic_its *its)
970 {
971         struct its_collection *cur, *temp;
972
973         list_for_each_entry_safe(cur, temp, &its->collection_list, coll_list)
974                 vgic_its_free_collection(its, cur->collection_id);
975 }
976
977 /* Must be called with its_lock mutex held */
978 static struct its_device *vgic_its_alloc_device(struct vgic_its *its,
979                                                 u32 device_id, gpa_t itt_addr,
980                                                 u8 num_eventid_bits)
981 {
982         struct its_device *device;
983
984         device = kzalloc(sizeof(*device), GFP_KERNEL);
985         if (!device)
986                 return ERR_PTR(-ENOMEM);
987
988         device->device_id = device_id;
989         device->itt_addr = itt_addr;
990         device->num_eventid_bits = num_eventid_bits;
991         INIT_LIST_HEAD(&device->itt_head);
992
993         list_add_tail(&device->dev_list, &its->device_list);
994         return device;
995 }
996
997 /*
998  * MAPD maps or unmaps a device ID to Interrupt Translation Tables (ITTs).
999  * Must be called with the its_lock mutex held.
1000  */
1001 static int vgic_its_cmd_handle_mapd(struct kvm *kvm, struct vgic_its *its,
1002                                     u64 *its_cmd)
1003 {
1004         u32 device_id = its_cmd_get_deviceid(its_cmd);
1005         bool valid = its_cmd_get_validbit(its_cmd);
1006         u8 num_eventid_bits = its_cmd_get_size(its_cmd);
1007         gpa_t itt_addr = its_cmd_get_ittaddr(its_cmd);
1008         struct its_device *device;
1009
1010         if (!vgic_its_check_id(its, its->baser_device_table, device_id, NULL))
1011                 return E_ITS_MAPD_DEVICE_OOR;
1012
1013         if (valid && num_eventid_bits > VITS_TYPER_IDBITS)
1014                 return E_ITS_MAPD_ITTSIZE_OOR;
1015
1016         device = find_its_device(its, device_id);
1017
1018         /*
1019          * The spec says that calling MAPD on an already mapped device
1020          * invalidates all cached data for this device. We implement this
1021          * by removing the mapping and re-establishing it.
1022          */
1023         if (device)
1024                 vgic_its_free_device(kvm, device);
1025
1026         /*
1027          * The spec does not say whether unmapping a not-mapped device
1028          * is an error, so we are done in any case.
1029          */
1030         if (!valid)
1031                 return 0;
1032
1033         device = vgic_its_alloc_device(its, device_id, itt_addr,
1034                                        num_eventid_bits);
1035         if (IS_ERR(device))
1036                 return PTR_ERR(device);
1037
1038         return 0;
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(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(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         /* Bits 15:12 contain bits 51:48 of the PA, which we don't support. */
1294         reg &= ~GENMASK_ULL(15, 12);
1295
1296         /* We support only one (ITS) page size: 64K */
1297         reg = (reg & ~GITS_BASER_PAGE_SIZE_MASK) | GITS_BASER_PAGE_SIZE_64K;
1298
1299         return reg;
1300 }
1301
1302 static u64 vgic_sanitise_its_cbaser(u64 reg)
1303 {
1304         reg = vgic_sanitise_field(reg, GITS_CBASER_SHAREABILITY_MASK,
1305                                   GITS_CBASER_SHAREABILITY_SHIFT,
1306                                   vgic_sanitise_shareability);
1307         reg = vgic_sanitise_field(reg, GITS_CBASER_INNER_CACHEABILITY_MASK,
1308                                   GITS_CBASER_INNER_CACHEABILITY_SHIFT,
1309                                   vgic_sanitise_inner_cacheability);
1310         reg = vgic_sanitise_field(reg, GITS_CBASER_OUTER_CACHEABILITY_MASK,
1311                                   GITS_CBASER_OUTER_CACHEABILITY_SHIFT,
1312                                   vgic_sanitise_outer_cacheability);
1313
1314         /*
1315          * Sanitise the physical address to be 64k aligned.
1316          * Also limit the physical addresses to 48 bits.
1317          */
1318         reg &= ~(GENMASK_ULL(51, 48) | GENMASK_ULL(15, 12));
1319
1320         return reg;
1321 }
1322
1323 static unsigned long vgic_mmio_read_its_cbaser(struct kvm *kvm,
1324                                                struct vgic_its *its,
1325                                                gpa_t addr, unsigned int len)
1326 {
1327         return extract_bytes(its->cbaser, addr & 7, len);
1328 }
1329
1330 static void vgic_mmio_write_its_cbaser(struct kvm *kvm, struct vgic_its *its,
1331                                        gpa_t addr, unsigned int len,
1332                                        unsigned long val)
1333 {
1334         /* When GITS_CTLR.Enable is 1, this register is RO. */
1335         if (its->enabled)
1336                 return;
1337
1338         mutex_lock(&its->cmd_lock);
1339         its->cbaser = update_64bit_reg(its->cbaser, addr & 7, len, val);
1340         its->cbaser = vgic_sanitise_its_cbaser(its->cbaser);
1341         its->creadr = 0;
1342         /*
1343          * CWRITER is architecturally UNKNOWN on reset, but we need to reset
1344          * it to CREADR to make sure we start with an empty command buffer.
1345          */
1346         its->cwriter = its->creadr;
1347         mutex_unlock(&its->cmd_lock);
1348 }
1349
1350 #define ITS_CMD_BUFFER_SIZE(baser)      ((((baser) & 0xff) + 1) << 12)
1351 #define ITS_CMD_SIZE                    32
1352 #define ITS_CMD_OFFSET(reg)             ((reg) & GENMASK(19, 5))
1353
1354 /* Must be called with the cmd_lock held. */
1355 static void vgic_its_process_commands(struct kvm *kvm, struct vgic_its *its)
1356 {
1357         gpa_t cbaser;
1358         u64 cmd_buf[4];
1359
1360         /* Commands are only processed when the ITS is enabled. */
1361         if (!its->enabled)
1362                 return;
1363
1364         cbaser = CBASER_ADDRESS(its->cbaser);
1365
1366         while (its->cwriter != its->creadr) {
1367                 int ret = kvm_read_guest(kvm, cbaser + its->creadr,
1368                                          cmd_buf, ITS_CMD_SIZE);
1369                 /*
1370                  * If kvm_read_guest() fails, this could be due to the guest
1371                  * programming a bogus value in CBASER or something else going
1372                  * wrong from which we cannot easily recover.
1373                  * According to section 6.3.2 in the GICv3 spec we can just
1374                  * ignore that command then.
1375                  */
1376                 if (!ret)
1377                         vgic_its_handle_command(kvm, its, cmd_buf);
1378
1379                 its->creadr += ITS_CMD_SIZE;
1380                 if (its->creadr == ITS_CMD_BUFFER_SIZE(its->cbaser))
1381                         its->creadr = 0;
1382         }
1383 }
1384
1385 /*
1386  * By writing to CWRITER the guest announces new commands to be processed.
1387  * To avoid any races in the first place, we take the its_cmd lock, which
1388  * protects our ring buffer variables, so that there is only one user
1389  * per ITS handling commands at a given time.
1390  */
1391 static void vgic_mmio_write_its_cwriter(struct kvm *kvm, struct vgic_its *its,
1392                                         gpa_t addr, unsigned int len,
1393                                         unsigned long val)
1394 {
1395         u64 reg;
1396
1397         if (!its)
1398                 return;
1399
1400         mutex_lock(&its->cmd_lock);
1401
1402         reg = update_64bit_reg(its->cwriter, addr & 7, len, val);
1403         reg = ITS_CMD_OFFSET(reg);
1404         if (reg >= ITS_CMD_BUFFER_SIZE(its->cbaser)) {
1405                 mutex_unlock(&its->cmd_lock);
1406                 return;
1407         }
1408         its->cwriter = reg;
1409
1410         vgic_its_process_commands(kvm, its);
1411
1412         mutex_unlock(&its->cmd_lock);
1413 }
1414
1415 static unsigned long vgic_mmio_read_its_cwriter(struct kvm *kvm,
1416                                                 struct vgic_its *its,
1417                                                 gpa_t addr, unsigned int len)
1418 {
1419         return extract_bytes(its->cwriter, addr & 0x7, len);
1420 }
1421
1422 static unsigned long vgic_mmio_read_its_creadr(struct kvm *kvm,
1423                                                struct vgic_its *its,
1424                                                gpa_t addr, unsigned int len)
1425 {
1426         return extract_bytes(its->creadr, addr & 0x7, len);
1427 }
1428
1429 static int vgic_mmio_uaccess_write_its_creadr(struct kvm *kvm,
1430                                               struct vgic_its *its,
1431                                               gpa_t addr, unsigned int len,
1432                                               unsigned long val)
1433 {
1434         u32 cmd_offset;
1435         int ret = 0;
1436
1437         mutex_lock(&its->cmd_lock);
1438
1439         if (its->enabled) {
1440                 ret = -EBUSY;
1441                 goto out;
1442         }
1443
1444         cmd_offset = ITS_CMD_OFFSET(val);
1445         if (cmd_offset >= ITS_CMD_BUFFER_SIZE(its->cbaser)) {
1446                 ret = -EINVAL;
1447                 goto out;
1448         }
1449
1450         its->creadr = cmd_offset;
1451 out:
1452         mutex_unlock(&its->cmd_lock);
1453         return ret;
1454 }
1455
1456 #define BASER_INDEX(addr) (((addr) / sizeof(u64)) & 0x7)
1457 static unsigned long vgic_mmio_read_its_baser(struct kvm *kvm,
1458                                               struct vgic_its *its,
1459                                               gpa_t addr, unsigned int len)
1460 {
1461         u64 reg;
1462
1463         switch (BASER_INDEX(addr)) {
1464         case 0:
1465                 reg = its->baser_device_table;
1466                 break;
1467         case 1:
1468                 reg = its->baser_coll_table;
1469                 break;
1470         default:
1471                 reg = 0;
1472                 break;
1473         }
1474
1475         return extract_bytes(reg, addr & 7, len);
1476 }
1477
1478 #define GITS_BASER_RO_MASK      (GENMASK_ULL(52, 48) | GENMASK_ULL(58, 56))
1479 static void vgic_mmio_write_its_baser(struct kvm *kvm,
1480                                       struct vgic_its *its,
1481                                       gpa_t addr, unsigned int len,
1482                                       unsigned long val)
1483 {
1484         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
1485         u64 entry_size, table_type;
1486         u64 reg, *regptr, clearbits = 0;
1487
1488         /* When GITS_CTLR.Enable is 1, we ignore write accesses. */
1489         if (its->enabled)
1490                 return;
1491
1492         switch (BASER_INDEX(addr)) {
1493         case 0:
1494                 regptr = &its->baser_device_table;
1495                 entry_size = abi->dte_esz;
1496                 table_type = GITS_BASER_TYPE_DEVICE;
1497                 break;
1498         case 1:
1499                 regptr = &its->baser_coll_table;
1500                 entry_size = abi->cte_esz;
1501                 table_type = GITS_BASER_TYPE_COLLECTION;
1502                 clearbits = GITS_BASER_INDIRECT;
1503                 break;
1504         default:
1505                 return;
1506         }
1507
1508         reg = update_64bit_reg(*regptr, addr & 7, len, val);
1509         reg &= ~GITS_BASER_RO_MASK;
1510         reg &= ~clearbits;
1511
1512         reg |= (entry_size - 1) << GITS_BASER_ENTRY_SIZE_SHIFT;
1513         reg |= table_type << GITS_BASER_TYPE_SHIFT;
1514         reg = vgic_sanitise_its_baser(reg);
1515
1516         *regptr = reg;
1517
1518         if (!(reg & GITS_BASER_VALID)) {
1519                 /* Take the its_lock to prevent a race with a save/restore */
1520                 mutex_lock(&its->its_lock);
1521                 switch (table_type) {
1522                 case GITS_BASER_TYPE_DEVICE:
1523                         vgic_its_free_device_list(kvm, its);
1524                         break;
1525                 case GITS_BASER_TYPE_COLLECTION:
1526                         vgic_its_free_collection_list(kvm, its);
1527                         break;
1528                 }
1529                 mutex_unlock(&its->its_lock);
1530         }
1531 }
1532
1533 static unsigned long vgic_mmio_read_its_ctlr(struct kvm *vcpu,
1534                                              struct vgic_its *its,
1535                                              gpa_t addr, unsigned int len)
1536 {
1537         u32 reg = 0;
1538
1539         mutex_lock(&its->cmd_lock);
1540         if (its->creadr == its->cwriter)
1541                 reg |= GITS_CTLR_QUIESCENT;
1542         if (its->enabled)
1543                 reg |= GITS_CTLR_ENABLE;
1544         mutex_unlock(&its->cmd_lock);
1545
1546         return reg;
1547 }
1548
1549 static void vgic_mmio_write_its_ctlr(struct kvm *kvm, struct vgic_its *its,
1550                                      gpa_t addr, unsigned int len,
1551                                      unsigned long val)
1552 {
1553         mutex_lock(&its->cmd_lock);
1554
1555         /*
1556          * It is UNPREDICTABLE to enable the ITS if any of the CBASER or
1557          * device/collection BASER are invalid
1558          */
1559         if (!its->enabled && (val & GITS_CTLR_ENABLE) &&
1560                 (!(its->baser_device_table & GITS_BASER_VALID) ||
1561                  !(its->baser_coll_table & GITS_BASER_VALID) ||
1562                  !(its->cbaser & GITS_CBASER_VALID)))
1563                 goto out;
1564
1565         its->enabled = !!(val & GITS_CTLR_ENABLE);
1566
1567         /*
1568          * Try to process any pending commands. This function bails out early
1569          * if the ITS is disabled or no commands have been queued.
1570          */
1571         vgic_its_process_commands(kvm, its);
1572
1573 out:
1574         mutex_unlock(&its->cmd_lock);
1575 }
1576
1577 #define REGISTER_ITS_DESC(off, rd, wr, length, acc)             \
1578 {                                                               \
1579         .reg_offset = off,                                      \
1580         .len = length,                                          \
1581         .access_flags = acc,                                    \
1582         .its_read = rd,                                         \
1583         .its_write = wr,                                        \
1584 }
1585
1586 #define REGISTER_ITS_DESC_UACCESS(off, rd, wr, uwr, length, acc)\
1587 {                                                               \
1588         .reg_offset = off,                                      \
1589         .len = length,                                          \
1590         .access_flags = acc,                                    \
1591         .its_read = rd,                                         \
1592         .its_write = wr,                                        \
1593         .uaccess_its_write = uwr,                               \
1594 }
1595
1596 static void its_mmio_write_wi(struct kvm *kvm, struct vgic_its *its,
1597                               gpa_t addr, unsigned int len, unsigned long val)
1598 {
1599         /* Ignore */
1600 }
1601
1602 static struct vgic_register_region its_registers[] = {
1603         REGISTER_ITS_DESC(GITS_CTLR,
1604                 vgic_mmio_read_its_ctlr, vgic_mmio_write_its_ctlr, 4,
1605                 VGIC_ACCESS_32bit),
1606         REGISTER_ITS_DESC_UACCESS(GITS_IIDR,
1607                 vgic_mmio_read_its_iidr, its_mmio_write_wi,
1608                 vgic_mmio_uaccess_write_its_iidr, 4,
1609                 VGIC_ACCESS_32bit),
1610         REGISTER_ITS_DESC(GITS_TYPER,
1611                 vgic_mmio_read_its_typer, its_mmio_write_wi, 8,
1612                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1613         REGISTER_ITS_DESC(GITS_CBASER,
1614                 vgic_mmio_read_its_cbaser, vgic_mmio_write_its_cbaser, 8,
1615                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1616         REGISTER_ITS_DESC(GITS_CWRITER,
1617                 vgic_mmio_read_its_cwriter, vgic_mmio_write_its_cwriter, 8,
1618                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1619         REGISTER_ITS_DESC_UACCESS(GITS_CREADR,
1620                 vgic_mmio_read_its_creadr, its_mmio_write_wi,
1621                 vgic_mmio_uaccess_write_its_creadr, 8,
1622                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1623         REGISTER_ITS_DESC(GITS_BASER,
1624                 vgic_mmio_read_its_baser, vgic_mmio_write_its_baser, 0x40,
1625                 VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
1626         REGISTER_ITS_DESC(GITS_IDREGS_BASE,
1627                 vgic_mmio_read_its_idregs, its_mmio_write_wi, 0x30,
1628                 VGIC_ACCESS_32bit),
1629 };
1630
1631 /* This is called on setting the LPI enable bit in the redistributor. */
1632 void vgic_enable_lpis(struct kvm_vcpu *vcpu)
1633 {
1634         if (!(vcpu->arch.vgic_cpu.pendbaser & GICR_PENDBASER_PTZ))
1635                 its_sync_lpi_pending_table(vcpu);
1636 }
1637
1638 static int vgic_register_its_iodev(struct kvm *kvm, struct vgic_its *its,
1639                                    u64 addr)
1640 {
1641         struct vgic_io_device *iodev = &its->iodev;
1642         int ret;
1643
1644         mutex_lock(&kvm->slots_lock);
1645         if (!IS_VGIC_ADDR_UNDEF(its->vgic_its_base)) {
1646                 ret = -EBUSY;
1647                 goto out;
1648         }
1649
1650         its->vgic_its_base = addr;
1651         iodev->regions = its_registers;
1652         iodev->nr_regions = ARRAY_SIZE(its_registers);
1653         kvm_iodevice_init(&iodev->dev, &kvm_io_gic_ops);
1654
1655         iodev->base_addr = its->vgic_its_base;
1656         iodev->iodev_type = IODEV_ITS;
1657         iodev->its = its;
1658         ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, iodev->base_addr,
1659                                       KVM_VGIC_V3_ITS_SIZE, &iodev->dev);
1660 out:
1661         mutex_unlock(&kvm->slots_lock);
1662
1663         return ret;
1664 }
1665
1666 #define INITIAL_BASER_VALUE                                               \
1667         (GIC_BASER_CACHEABILITY(GITS_BASER, INNER, RaWb)                | \
1668          GIC_BASER_CACHEABILITY(GITS_BASER, OUTER, SameAsInner)         | \
1669          GIC_BASER_SHAREABILITY(GITS_BASER, InnerShareable)             | \
1670          GITS_BASER_PAGE_SIZE_64K)
1671
1672 #define INITIAL_PROPBASER_VALUE                                           \
1673         (GIC_BASER_CACHEABILITY(GICR_PROPBASER, INNER, RaWb)            | \
1674          GIC_BASER_CACHEABILITY(GICR_PROPBASER, OUTER, SameAsInner)     | \
1675          GIC_BASER_SHAREABILITY(GICR_PROPBASER, InnerShareable))
1676
1677 static int vgic_its_create(struct kvm_device *dev, u32 type)
1678 {
1679         struct vgic_its *its;
1680
1681         if (type != KVM_DEV_TYPE_ARM_VGIC_ITS)
1682                 return -ENODEV;
1683
1684         its = kzalloc(sizeof(struct vgic_its), GFP_KERNEL);
1685         if (!its)
1686                 return -ENOMEM;
1687
1688         if (vgic_initialized(dev->kvm)) {
1689                 int ret = vgic_v4_init(dev->kvm);
1690                 if (ret < 0) {
1691                         kfree(its);
1692                         return ret;
1693                 }
1694         }
1695
1696         mutex_init(&its->its_lock);
1697         mutex_init(&its->cmd_lock);
1698
1699         its->vgic_its_base = VGIC_ADDR_UNDEF;
1700
1701         INIT_LIST_HEAD(&its->device_list);
1702         INIT_LIST_HEAD(&its->collection_list);
1703
1704         dev->kvm->arch.vgic.msis_require_devid = true;
1705         dev->kvm->arch.vgic.has_its = true;
1706         its->enabled = false;
1707         its->dev = dev;
1708
1709         its->baser_device_table = INITIAL_BASER_VALUE                   |
1710                 ((u64)GITS_BASER_TYPE_DEVICE << GITS_BASER_TYPE_SHIFT);
1711         its->baser_coll_table = INITIAL_BASER_VALUE |
1712                 ((u64)GITS_BASER_TYPE_COLLECTION << GITS_BASER_TYPE_SHIFT);
1713         dev->kvm->arch.vgic.propbaser = INITIAL_PROPBASER_VALUE;
1714
1715         dev->private = its;
1716
1717         return vgic_its_set_abi(its, NR_ITS_ABIS - 1);
1718 }
1719
1720 static void vgic_its_destroy(struct kvm_device *kvm_dev)
1721 {
1722         struct kvm *kvm = kvm_dev->kvm;
1723         struct vgic_its *its = kvm_dev->private;
1724
1725         mutex_lock(&its->its_lock);
1726
1727         vgic_its_free_device_list(kvm, its);
1728         vgic_its_free_collection_list(kvm, its);
1729
1730         mutex_unlock(&its->its_lock);
1731         kfree(its);
1732 }
1733
1734 int vgic_its_has_attr_regs(struct kvm_device *dev,
1735                            struct kvm_device_attr *attr)
1736 {
1737         const struct vgic_register_region *region;
1738         gpa_t offset = attr->attr;
1739         int align;
1740
1741         align = (offset < GITS_TYPER) || (offset >= GITS_PIDR4) ? 0x3 : 0x7;
1742
1743         if (offset & align)
1744                 return -EINVAL;
1745
1746         region = vgic_find_mmio_region(its_registers,
1747                                        ARRAY_SIZE(its_registers),
1748                                        offset);
1749         if (!region)
1750                 return -ENXIO;
1751
1752         return 0;
1753 }
1754
1755 int vgic_its_attr_regs_access(struct kvm_device *dev,
1756                               struct kvm_device_attr *attr,
1757                               u64 *reg, bool is_write)
1758 {
1759         const struct vgic_register_region *region;
1760         struct vgic_its *its;
1761         gpa_t addr, offset;
1762         unsigned int len;
1763         int align, ret = 0;
1764
1765         its = dev->private;
1766         offset = attr->attr;
1767
1768         /*
1769          * Although the spec supports upper/lower 32-bit accesses to
1770          * 64-bit ITS registers, the userspace ABI requires 64-bit
1771          * accesses to all 64-bit wide registers. We therefore only
1772          * support 32-bit accesses to GITS_CTLR, GITS_IIDR and GITS ID
1773          * registers
1774          */
1775         if ((offset < GITS_TYPER) || (offset >= GITS_PIDR4))
1776                 align = 0x3;
1777         else
1778                 align = 0x7;
1779
1780         if (offset & align)
1781                 return -EINVAL;
1782
1783         mutex_lock(&dev->kvm->lock);
1784
1785         if (IS_VGIC_ADDR_UNDEF(its->vgic_its_base)) {
1786                 ret = -ENXIO;
1787                 goto out;
1788         }
1789
1790         region = vgic_find_mmio_region(its_registers,
1791                                        ARRAY_SIZE(its_registers),
1792                                        offset);
1793         if (!region) {
1794                 ret = -ENXIO;
1795                 goto out;
1796         }
1797
1798         if (!lock_all_vcpus(dev->kvm)) {
1799                 ret = -EBUSY;
1800                 goto out;
1801         }
1802
1803         addr = its->vgic_its_base + offset;
1804
1805         len = region->access_flags & VGIC_ACCESS_64bit ? 8 : 4;
1806
1807         if (is_write) {
1808                 if (region->uaccess_its_write)
1809                         ret = region->uaccess_its_write(dev->kvm, its, addr,
1810                                                         len, *reg);
1811                 else
1812                         region->its_write(dev->kvm, its, addr, len, *reg);
1813         } else {
1814                 *reg = region->its_read(dev->kvm, its, addr, len);
1815         }
1816         unlock_all_vcpus(dev->kvm);
1817 out:
1818         mutex_unlock(&dev->kvm->lock);
1819         return ret;
1820 }
1821
1822 static u32 compute_next_devid_offset(struct list_head *h,
1823                                      struct its_device *dev)
1824 {
1825         struct its_device *next;
1826         u32 next_offset;
1827
1828         if (list_is_last(&dev->dev_list, h))
1829                 return 0;
1830         next = list_next_entry(dev, dev_list);
1831         next_offset = next->device_id - dev->device_id;
1832
1833         return min_t(u32, next_offset, VITS_DTE_MAX_DEVID_OFFSET);
1834 }
1835
1836 static u32 compute_next_eventid_offset(struct list_head *h, struct its_ite *ite)
1837 {
1838         struct its_ite *next;
1839         u32 next_offset;
1840
1841         if (list_is_last(&ite->ite_list, h))
1842                 return 0;
1843         next = list_next_entry(ite, ite_list);
1844         next_offset = next->event_id - ite->event_id;
1845
1846         return min_t(u32, next_offset, VITS_ITE_MAX_EVENTID_OFFSET);
1847 }
1848
1849 /**
1850  * entry_fn_t - Callback called on a table entry restore path
1851  * @its: its handle
1852  * @id: id of the entry
1853  * @entry: pointer to the entry
1854  * @opaque: pointer to an opaque data
1855  *
1856  * Return: < 0 on error, 0 if last element was identified, id offset to next
1857  * element otherwise
1858  */
1859 typedef int (*entry_fn_t)(struct vgic_its *its, u32 id, void *entry,
1860                           void *opaque);
1861
1862 /**
1863  * scan_its_table - Scan a contiguous table in guest RAM and applies a function
1864  * to each entry
1865  *
1866  * @its: its handle
1867  * @base: base gpa of the table
1868  * @size: size of the table in bytes
1869  * @esz: entry size in bytes
1870  * @start_id: the ID of the first entry in the table
1871  * (non zero for 2d level tables)
1872  * @fn: function to apply on each entry
1873  *
1874  * Return: < 0 on error, 0 if last element was identified, 1 otherwise
1875  * (the last element may not be found on second level tables)
1876  */
1877 static int scan_its_table(struct vgic_its *its, gpa_t base, int size, int esz,
1878                           int start_id, entry_fn_t fn, void *opaque)
1879 {
1880         struct kvm *kvm = its->dev->kvm;
1881         unsigned long len = size;
1882         int id = start_id;
1883         gpa_t gpa = base;
1884         char entry[esz];
1885         int ret;
1886
1887         memset(entry, 0, esz);
1888
1889         while (len > 0) {
1890                 int next_offset;
1891                 size_t byte_offset;
1892
1893                 ret = kvm_read_guest(kvm, gpa, entry, esz);
1894                 if (ret)
1895                         return ret;
1896
1897                 next_offset = fn(its, id, entry, opaque);
1898                 if (next_offset <= 0)
1899                         return next_offset;
1900
1901                 byte_offset = next_offset * esz;
1902                 id += next_offset;
1903                 gpa += byte_offset;
1904                 len -= byte_offset;
1905         }
1906         return 1;
1907 }
1908
1909 /**
1910  * vgic_its_save_ite - Save an interrupt translation entry at @gpa
1911  */
1912 static int vgic_its_save_ite(struct vgic_its *its, struct its_device *dev,
1913                               struct its_ite *ite, gpa_t gpa, int ite_esz)
1914 {
1915         struct kvm *kvm = its->dev->kvm;
1916         u32 next_offset;
1917         u64 val;
1918
1919         next_offset = compute_next_eventid_offset(&dev->itt_head, ite);
1920         val = ((u64)next_offset << KVM_ITS_ITE_NEXT_SHIFT) |
1921                ((u64)ite->irq->intid << KVM_ITS_ITE_PINTID_SHIFT) |
1922                 ite->collection->collection_id;
1923         val = cpu_to_le64(val);
1924         return kvm_write_guest(kvm, gpa, &val, ite_esz);
1925 }
1926
1927 /**
1928  * vgic_its_restore_ite - restore an interrupt translation entry
1929  * @event_id: id used for indexing
1930  * @ptr: pointer to the ITE entry
1931  * @opaque: pointer to the its_device
1932  */
1933 static int vgic_its_restore_ite(struct vgic_its *its, u32 event_id,
1934                                 void *ptr, void *opaque)
1935 {
1936         struct its_device *dev = (struct its_device *)opaque;
1937         struct its_collection *collection;
1938         struct kvm *kvm = its->dev->kvm;
1939         struct kvm_vcpu *vcpu = NULL;
1940         u64 val;
1941         u64 *p = (u64 *)ptr;
1942         struct vgic_irq *irq;
1943         u32 coll_id, lpi_id;
1944         struct its_ite *ite;
1945         u32 offset;
1946
1947         val = *p;
1948
1949         val = le64_to_cpu(val);
1950
1951         coll_id = val & KVM_ITS_ITE_ICID_MASK;
1952         lpi_id = (val & KVM_ITS_ITE_PINTID_MASK) >> KVM_ITS_ITE_PINTID_SHIFT;
1953
1954         if (!lpi_id)
1955                 return 1; /* invalid entry, no choice but to scan next entry */
1956
1957         if (lpi_id < VGIC_MIN_LPI)
1958                 return -EINVAL;
1959
1960         offset = val >> KVM_ITS_ITE_NEXT_SHIFT;
1961         if (event_id + offset >= BIT_ULL(dev->num_eventid_bits))
1962                 return -EINVAL;
1963
1964         collection = find_collection(its, coll_id);
1965         if (!collection)
1966                 return -EINVAL;
1967
1968         ite = vgic_its_alloc_ite(dev, collection, event_id);
1969         if (IS_ERR(ite))
1970                 return PTR_ERR(ite);
1971
1972         if (its_is_collection_mapped(collection))
1973                 vcpu = kvm_get_vcpu(kvm, collection->target_addr);
1974
1975         irq = vgic_add_lpi(kvm, lpi_id, vcpu);
1976         if (IS_ERR(irq))
1977                 return PTR_ERR(irq);
1978         ite->irq = irq;
1979
1980         return offset;
1981 }
1982
1983 static int vgic_its_ite_cmp(void *priv, struct list_head *a,
1984                             struct list_head *b)
1985 {
1986         struct its_ite *itea = container_of(a, struct its_ite, ite_list);
1987         struct its_ite *iteb = container_of(b, struct its_ite, ite_list);
1988
1989         if (itea->event_id < iteb->event_id)
1990                 return -1;
1991         else
1992                 return 1;
1993 }
1994
1995 static int vgic_its_save_itt(struct vgic_its *its, struct its_device *device)
1996 {
1997         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
1998         gpa_t base = device->itt_addr;
1999         struct its_ite *ite;
2000         int ret;
2001         int ite_esz = abi->ite_esz;
2002
2003         list_sort(NULL, &device->itt_head, vgic_its_ite_cmp);
2004
2005         list_for_each_entry(ite, &device->itt_head, ite_list) {
2006                 gpa_t gpa = base + ite->event_id * ite_esz;
2007
2008                 /*
2009                  * If an LPI carries the HW bit, this means that this
2010                  * interrupt is controlled by GICv4, and we do not
2011                  * have direct access to that state. Let's simply fail
2012                  * the save operation...
2013                  */
2014                 if (ite->irq->hw)
2015                         return -EACCES;
2016
2017                 ret = vgic_its_save_ite(its, device, ite, gpa, ite_esz);
2018                 if (ret)
2019                         return ret;
2020         }
2021         return 0;
2022 }
2023
2024 /**
2025  * vgic_its_restore_itt - restore the ITT of a device
2026  *
2027  * @its: its handle
2028  * @dev: device handle
2029  *
2030  * Return 0 on success, < 0 on error
2031  */
2032 static int vgic_its_restore_itt(struct vgic_its *its, struct its_device *dev)
2033 {
2034         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2035         gpa_t base = dev->itt_addr;
2036         int ret;
2037         int ite_esz = abi->ite_esz;
2038         size_t max_size = BIT_ULL(dev->num_eventid_bits) * ite_esz;
2039
2040         ret = scan_its_table(its, base, max_size, ite_esz, 0,
2041                              vgic_its_restore_ite, dev);
2042
2043         /* scan_its_table returns +1 if all ITEs are invalid */
2044         if (ret > 0)
2045                 ret = 0;
2046
2047         return ret;
2048 }
2049
2050 /**
2051  * vgic_its_save_dte - Save a device table entry at a given GPA
2052  *
2053  * @its: ITS handle
2054  * @dev: ITS device
2055  * @ptr: GPA
2056  */
2057 static int vgic_its_save_dte(struct vgic_its *its, struct its_device *dev,
2058                              gpa_t ptr, int dte_esz)
2059 {
2060         struct kvm *kvm = its->dev->kvm;
2061         u64 val, itt_addr_field;
2062         u32 next_offset;
2063
2064         itt_addr_field = dev->itt_addr >> 8;
2065         next_offset = compute_next_devid_offset(&its->device_list, dev);
2066         val = (1ULL << KVM_ITS_DTE_VALID_SHIFT |
2067                ((u64)next_offset << KVM_ITS_DTE_NEXT_SHIFT) |
2068                (itt_addr_field << KVM_ITS_DTE_ITTADDR_SHIFT) |
2069                 (dev->num_eventid_bits - 1));
2070         val = cpu_to_le64(val);
2071         return kvm_write_guest(kvm, ptr, &val, dte_esz);
2072 }
2073
2074 /**
2075  * vgic_its_restore_dte - restore a device table entry
2076  *
2077  * @its: its handle
2078  * @id: device id the DTE corresponds to
2079  * @ptr: kernel VA where the 8 byte DTE is located
2080  * @opaque: unused
2081  *
2082  * Return: < 0 on error, 0 if the dte is the last one, id offset to the
2083  * next dte otherwise
2084  */
2085 static int vgic_its_restore_dte(struct vgic_its *its, u32 id,
2086                                 void *ptr, void *opaque)
2087 {
2088         struct its_device *dev;
2089         gpa_t itt_addr;
2090         u8 num_eventid_bits;
2091         u64 entry = *(u64 *)ptr;
2092         bool valid;
2093         u32 offset;
2094         int ret;
2095
2096         entry = le64_to_cpu(entry);
2097
2098         valid = entry >> KVM_ITS_DTE_VALID_SHIFT;
2099         num_eventid_bits = (entry & KVM_ITS_DTE_SIZE_MASK) + 1;
2100         itt_addr = ((entry & KVM_ITS_DTE_ITTADDR_MASK)
2101                         >> KVM_ITS_DTE_ITTADDR_SHIFT) << 8;
2102
2103         if (!valid)
2104                 return 1;
2105
2106         /* dte entry is valid */
2107         offset = (entry & KVM_ITS_DTE_NEXT_MASK) >> KVM_ITS_DTE_NEXT_SHIFT;
2108
2109         dev = vgic_its_alloc_device(its, id, itt_addr, num_eventid_bits);
2110         if (IS_ERR(dev))
2111                 return PTR_ERR(dev);
2112
2113         ret = vgic_its_restore_itt(its, dev);
2114         if (ret) {
2115                 vgic_its_free_device(its->dev->kvm, dev);
2116                 return ret;
2117         }
2118
2119         return offset;
2120 }
2121
2122 static int vgic_its_device_cmp(void *priv, struct list_head *a,
2123                                struct list_head *b)
2124 {
2125         struct its_device *deva = container_of(a, struct its_device, dev_list);
2126         struct its_device *devb = container_of(b, struct its_device, dev_list);
2127
2128         if (deva->device_id < devb->device_id)
2129                 return -1;
2130         else
2131                 return 1;
2132 }
2133
2134 /**
2135  * vgic_its_save_device_tables - Save the device table and all ITT
2136  * into guest RAM
2137  *
2138  * L1/L2 handling is hidden by vgic_its_check_id() helper which directly
2139  * returns the GPA of the device entry
2140  */
2141 static int vgic_its_save_device_tables(struct vgic_its *its)
2142 {
2143         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2144         u64 baser = its->baser_device_table;
2145         struct its_device *dev;
2146         int dte_esz = abi->dte_esz;
2147
2148         if (!(baser & GITS_BASER_VALID))
2149                 return 0;
2150
2151         list_sort(NULL, &its->device_list, vgic_its_device_cmp);
2152
2153         list_for_each_entry(dev, &its->device_list, dev_list) {
2154                 int ret;
2155                 gpa_t eaddr;
2156
2157                 if (!vgic_its_check_id(its, baser,
2158                                        dev->device_id, &eaddr))
2159                         return -EINVAL;
2160
2161                 ret = vgic_its_save_itt(its, dev);
2162                 if (ret)
2163                         return ret;
2164
2165                 ret = vgic_its_save_dte(its, dev, eaddr, dte_esz);
2166                 if (ret)
2167                         return ret;
2168         }
2169         return 0;
2170 }
2171
2172 /**
2173  * handle_l1_dte - callback used for L1 device table entries (2 stage case)
2174  *
2175  * @its: its handle
2176  * @id: index of the entry in the L1 table
2177  * @addr: kernel VA
2178  * @opaque: unused
2179  *
2180  * L1 table entries are scanned by steps of 1 entry
2181  * Return < 0 if error, 0 if last dte was found when scanning the L2
2182  * table, +1 otherwise (meaning next L1 entry must be scanned)
2183  */
2184 static int handle_l1_dte(struct vgic_its *its, u32 id, void *addr,
2185                          void *opaque)
2186 {
2187         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2188         int l2_start_id = id * (SZ_64K / abi->dte_esz);
2189         u64 entry = *(u64 *)addr;
2190         int dte_esz = abi->dte_esz;
2191         gpa_t gpa;
2192         int ret;
2193
2194         entry = le64_to_cpu(entry);
2195
2196         if (!(entry & KVM_ITS_L1E_VALID_MASK))
2197                 return 1;
2198
2199         gpa = entry & KVM_ITS_L1E_ADDR_MASK;
2200
2201         ret = scan_its_table(its, gpa, SZ_64K, dte_esz,
2202                              l2_start_id, vgic_its_restore_dte, NULL);
2203
2204         return ret;
2205 }
2206
2207 /**
2208  * vgic_its_restore_device_tables - Restore the device table and all ITT
2209  * from guest RAM to internal data structs
2210  */
2211 static int vgic_its_restore_device_tables(struct vgic_its *its)
2212 {
2213         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2214         u64 baser = its->baser_device_table;
2215         int l1_esz, ret;
2216         int l1_tbl_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
2217         gpa_t l1_gpa;
2218
2219         if (!(baser & GITS_BASER_VALID))
2220                 return 0;
2221
2222         l1_gpa = BASER_ADDRESS(baser);
2223
2224         if (baser & GITS_BASER_INDIRECT) {
2225                 l1_esz = GITS_LVL1_ENTRY_SIZE;
2226                 ret = scan_its_table(its, l1_gpa, l1_tbl_size, l1_esz, 0,
2227                                      handle_l1_dte, NULL);
2228         } else {
2229                 l1_esz = abi->dte_esz;
2230                 ret = scan_its_table(its, l1_gpa, l1_tbl_size, l1_esz, 0,
2231                                      vgic_its_restore_dte, NULL);
2232         }
2233
2234         /* scan_its_table returns +1 if all entries are invalid */
2235         if (ret > 0)
2236                 ret = 0;
2237
2238         return ret;
2239 }
2240
2241 static int vgic_its_save_cte(struct vgic_its *its,
2242                              struct its_collection *collection,
2243                              gpa_t gpa, int esz)
2244 {
2245         u64 val;
2246
2247         val = (1ULL << KVM_ITS_CTE_VALID_SHIFT |
2248                ((u64)collection->target_addr << KVM_ITS_CTE_RDBASE_SHIFT) |
2249                collection->collection_id);
2250         val = cpu_to_le64(val);
2251         return kvm_write_guest(its->dev->kvm, gpa, &val, esz);
2252 }
2253
2254 static int vgic_its_restore_cte(struct vgic_its *its, gpa_t gpa, int esz)
2255 {
2256         struct its_collection *collection;
2257         struct kvm *kvm = its->dev->kvm;
2258         u32 target_addr, coll_id;
2259         u64 val;
2260         int ret;
2261
2262         BUG_ON(esz > sizeof(val));
2263         ret = kvm_read_guest(kvm, gpa, &val, esz);
2264         if (ret)
2265                 return ret;
2266         val = le64_to_cpu(val);
2267         if (!(val & KVM_ITS_CTE_VALID_MASK))
2268                 return 0;
2269
2270         target_addr = (u32)(val >> KVM_ITS_CTE_RDBASE_SHIFT);
2271         coll_id = val & KVM_ITS_CTE_ICID_MASK;
2272
2273         if (target_addr >= atomic_read(&kvm->online_vcpus))
2274                 return -EINVAL;
2275
2276         collection = find_collection(its, coll_id);
2277         if (collection)
2278                 return -EEXIST;
2279         ret = vgic_its_alloc_collection(its, &collection, coll_id);
2280         if (ret)
2281                 return ret;
2282         collection->target_addr = target_addr;
2283         return 1;
2284 }
2285
2286 /**
2287  * vgic_its_save_collection_table - Save the collection table into
2288  * guest RAM
2289  */
2290 static int vgic_its_save_collection_table(struct vgic_its *its)
2291 {
2292         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2293         u64 baser = its->baser_coll_table;
2294         gpa_t gpa = BASER_ADDRESS(baser);
2295         struct its_collection *collection;
2296         u64 val;
2297         size_t max_size, filled = 0;
2298         int ret, cte_esz = abi->cte_esz;
2299
2300         if (!(baser & GITS_BASER_VALID))
2301                 return 0;
2302
2303         max_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
2304
2305         list_for_each_entry(collection, &its->collection_list, coll_list) {
2306                 ret = vgic_its_save_cte(its, collection, gpa, cte_esz);
2307                 if (ret)
2308                         return ret;
2309                 gpa += cte_esz;
2310                 filled += cte_esz;
2311         }
2312
2313         if (filled == max_size)
2314                 return 0;
2315
2316         /*
2317          * table is not fully filled, add a last dummy element
2318          * with valid bit unset
2319          */
2320         val = 0;
2321         BUG_ON(cte_esz > sizeof(val));
2322         ret = kvm_write_guest(its->dev->kvm, gpa, &val, cte_esz);
2323         return ret;
2324 }
2325
2326 /**
2327  * vgic_its_restore_collection_table - reads the collection table
2328  * in guest memory and restores the ITS internal state. Requires the
2329  * BASER registers to be restored before.
2330  */
2331 static int vgic_its_restore_collection_table(struct vgic_its *its)
2332 {
2333         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2334         u64 baser = its->baser_coll_table;
2335         int cte_esz = abi->cte_esz;
2336         size_t max_size, read = 0;
2337         gpa_t gpa;
2338         int ret;
2339
2340         if (!(baser & GITS_BASER_VALID))
2341                 return 0;
2342
2343         gpa = BASER_ADDRESS(baser);
2344
2345         max_size = GITS_BASER_NR_PAGES(baser) * SZ_64K;
2346
2347         while (read < max_size) {
2348                 ret = vgic_its_restore_cte(its, gpa, cte_esz);
2349                 if (ret <= 0)
2350                         break;
2351                 gpa += cte_esz;
2352                 read += cte_esz;
2353         }
2354
2355         if (ret > 0)
2356                 return 0;
2357
2358         return ret;
2359 }
2360
2361 /**
2362  * vgic_its_save_tables_v0 - Save the ITS tables into guest ARM
2363  * according to v0 ABI
2364  */
2365 static int vgic_its_save_tables_v0(struct vgic_its *its)
2366 {
2367         int ret;
2368
2369         ret = vgic_its_save_device_tables(its);
2370         if (ret)
2371                 return ret;
2372
2373         return vgic_its_save_collection_table(its);
2374 }
2375
2376 /**
2377  * vgic_its_restore_tables_v0 - Restore the ITS tables from guest RAM
2378  * to internal data structs according to V0 ABI
2379  *
2380  */
2381 static int vgic_its_restore_tables_v0(struct vgic_its *its)
2382 {
2383         int ret;
2384
2385         ret = vgic_its_restore_collection_table(its);
2386         if (ret)
2387                 return ret;
2388
2389         return vgic_its_restore_device_tables(its);
2390 }
2391
2392 static int vgic_its_commit_v0(struct vgic_its *its)
2393 {
2394         const struct vgic_its_abi *abi;
2395
2396         abi = vgic_its_get_abi(its);
2397         its->baser_coll_table &= ~GITS_BASER_ENTRY_SIZE_MASK;
2398         its->baser_device_table &= ~GITS_BASER_ENTRY_SIZE_MASK;
2399
2400         its->baser_coll_table |= (GIC_ENCODE_SZ(abi->cte_esz, 5)
2401                                         << GITS_BASER_ENTRY_SIZE_SHIFT);
2402
2403         its->baser_device_table |= (GIC_ENCODE_SZ(abi->dte_esz, 5)
2404                                         << GITS_BASER_ENTRY_SIZE_SHIFT);
2405         return 0;
2406 }
2407
2408 static void vgic_its_reset(struct kvm *kvm, struct vgic_its *its)
2409 {
2410         /* We need to keep the ABI specific field values */
2411         its->baser_coll_table &= ~GITS_BASER_VALID;
2412         its->baser_device_table &= ~GITS_BASER_VALID;
2413         its->cbaser = 0;
2414         its->creadr = 0;
2415         its->cwriter = 0;
2416         its->enabled = 0;
2417         vgic_its_free_device_list(kvm, its);
2418         vgic_its_free_collection_list(kvm, its);
2419 }
2420
2421 static int vgic_its_has_attr(struct kvm_device *dev,
2422                              struct kvm_device_attr *attr)
2423 {
2424         switch (attr->group) {
2425         case KVM_DEV_ARM_VGIC_GRP_ADDR:
2426                 switch (attr->attr) {
2427                 case KVM_VGIC_ITS_ADDR_TYPE:
2428                         return 0;
2429                 }
2430                 break;
2431         case KVM_DEV_ARM_VGIC_GRP_CTRL:
2432                 switch (attr->attr) {
2433                 case KVM_DEV_ARM_VGIC_CTRL_INIT:
2434                         return 0;
2435                 case KVM_DEV_ARM_ITS_CTRL_RESET:
2436                         return 0;
2437                 case KVM_DEV_ARM_ITS_SAVE_TABLES:
2438                         return 0;
2439                 case KVM_DEV_ARM_ITS_RESTORE_TABLES:
2440                         return 0;
2441                 }
2442                 break;
2443         case KVM_DEV_ARM_VGIC_GRP_ITS_REGS:
2444                 return vgic_its_has_attr_regs(dev, attr);
2445         }
2446         return -ENXIO;
2447 }
2448
2449 static int vgic_its_ctrl(struct kvm *kvm, struct vgic_its *its, u64 attr)
2450 {
2451         const struct vgic_its_abi *abi = vgic_its_get_abi(its);
2452         int ret = 0;
2453
2454         if (attr == KVM_DEV_ARM_VGIC_CTRL_INIT) /* Nothing to do */
2455                 return 0;
2456
2457         mutex_lock(&kvm->lock);
2458         mutex_lock(&its->its_lock);
2459
2460         if (!lock_all_vcpus(kvm)) {
2461                 mutex_unlock(&its->its_lock);
2462                 mutex_unlock(&kvm->lock);
2463                 return -EBUSY;
2464         }
2465
2466         switch (attr) {
2467         case KVM_DEV_ARM_ITS_CTRL_RESET:
2468                 vgic_its_reset(kvm, its);
2469                 break;
2470         case KVM_DEV_ARM_ITS_SAVE_TABLES:
2471                 ret = abi->save_tables(its);
2472                 break;
2473         case KVM_DEV_ARM_ITS_RESTORE_TABLES:
2474                 ret = abi->restore_tables(its);
2475                 break;
2476         }
2477
2478         unlock_all_vcpus(kvm);
2479         mutex_unlock(&its->its_lock);
2480         mutex_unlock(&kvm->lock);
2481         return ret;
2482 }
2483
2484 static int vgic_its_set_attr(struct kvm_device *dev,
2485                              struct kvm_device_attr *attr)
2486 {
2487         struct vgic_its *its = dev->private;
2488         int ret;
2489
2490         switch (attr->group) {
2491         case KVM_DEV_ARM_VGIC_GRP_ADDR: {
2492                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2493                 unsigned long type = (unsigned long)attr->attr;
2494                 u64 addr;
2495
2496                 if (type != KVM_VGIC_ITS_ADDR_TYPE)
2497                         return -ENODEV;
2498
2499                 if (copy_from_user(&addr, uaddr, sizeof(addr)))
2500                         return -EFAULT;
2501
2502                 ret = vgic_check_ioaddr(dev->kvm, &its->vgic_its_base,
2503                                         addr, SZ_64K);
2504                 if (ret)
2505                         return ret;
2506
2507                 return vgic_register_its_iodev(dev->kvm, its, addr);
2508         }
2509         case KVM_DEV_ARM_VGIC_GRP_CTRL:
2510                 return vgic_its_ctrl(dev->kvm, its, attr->attr);
2511         case KVM_DEV_ARM_VGIC_GRP_ITS_REGS: {
2512                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2513                 u64 reg;
2514
2515                 if (get_user(reg, uaddr))
2516                         return -EFAULT;
2517
2518                 return vgic_its_attr_regs_access(dev, attr, &reg, true);
2519         }
2520         }
2521         return -ENXIO;
2522 }
2523
2524 static int vgic_its_get_attr(struct kvm_device *dev,
2525                              struct kvm_device_attr *attr)
2526 {
2527         switch (attr->group) {
2528         case KVM_DEV_ARM_VGIC_GRP_ADDR: {
2529                 struct vgic_its *its = dev->private;
2530                 u64 addr = its->vgic_its_base;
2531                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2532                 unsigned long type = (unsigned long)attr->attr;
2533
2534                 if (type != KVM_VGIC_ITS_ADDR_TYPE)
2535                         return -ENODEV;
2536
2537                 if (copy_to_user(uaddr, &addr, sizeof(addr)))
2538                         return -EFAULT;
2539                 break;
2540         }
2541         case KVM_DEV_ARM_VGIC_GRP_ITS_REGS: {
2542                 u64 __user *uaddr = (u64 __user *)(long)attr->addr;
2543                 u64 reg;
2544                 int ret;
2545
2546                 ret = vgic_its_attr_regs_access(dev, attr, &reg, false);
2547                 if (ret)
2548                         return ret;
2549                 return put_user(reg, uaddr);
2550         }
2551         default:
2552                 return -ENXIO;
2553         }
2554
2555         return 0;
2556 }
2557
2558 static struct kvm_device_ops kvm_arm_vgic_its_ops = {
2559         .name = "kvm-arm-vgic-its",
2560         .create = vgic_its_create,
2561         .destroy = vgic_its_destroy,
2562         .set_attr = vgic_its_set_attr,
2563         .get_attr = vgic_its_get_attr,
2564         .has_attr = vgic_its_has_attr,
2565 };
2566
2567 int kvm_vgic_register_its_device(void)
2568 {
2569         return kvm_register_device_ops(&kvm_arm_vgic_its_ops,
2570                                        KVM_DEV_TYPE_ARM_VGIC_ITS);
2571 }