]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - virt/kvm/arm/vgic-v3-emul.c
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
[karo-tx-linux.git] / virt / kvm / arm / vgic-v3-emul.c
1 /*
2  * GICv3 distributor and redistributor emulation
3  *
4  * GICv3 emulation is currently only supported on a GICv3 host (because
5  * we rely on the hardware's CPU interface virtualization support), but
6  * supports both hardware with or without the optional GICv2 backwards
7  * compatibility features.
8  *
9  * Limitations of the emulation:
10  * (RAZ/WI: read as zero, write ignore, RAO/WI: read as one, write ignore)
11  * - We do not support LPIs (yet). TYPER.LPIS is reported as 0 and is RAZ/WI.
12  * - We do not support the message based interrupts (MBIs) triggered by
13  *   writes to the GICD_{SET,CLR}SPI_* registers. TYPER.MBIS is reported as 0.
14  * - We do not support the (optional) backwards compatibility feature.
15  *   GICD_CTLR.ARE resets to 1 and is RAO/WI. If the _host_ GIC supports
16  *   the compatiblity feature, you can use a GICv2 in the guest, though.
17  * - We only support a single security state. GICD_CTLR.DS is 1 and is RAO/WI.
18  * - Priorities are not emulated (same as the GICv2 emulation). Linux
19  *   as a guest is fine with this, because it does not use priorities.
20  * - We only support Group1 interrupts. Again Linux uses only those.
21  *
22  * Copyright (C) 2014 ARM Ltd.
23  * Author: Andre Przywara <andre.przywara@arm.com>
24  *
25  * This program is free software; you can redistribute it and/or modify
26  * it under the terms of the GNU General Public License version 2 as
27  * published by the Free Software Foundation.
28  *
29  * This program is distributed in the hope that it will be useful,
30  * but WITHOUT ANY WARRANTY; without even the implied warranty of
31  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32  * GNU General Public License for more details.
33  *
34  * You should have received a copy of the GNU General Public License
35  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
36  */
37
38 #include <linux/cpu.h>
39 #include <linux/kvm.h>
40 #include <linux/kvm_host.h>
41 #include <linux/interrupt.h>
42
43 #include <linux/irqchip/arm-gic-v3.h>
44 #include <kvm/arm_vgic.h>
45
46 #include <asm/kvm_emulate.h>
47 #include <asm/kvm_arm.h>
48 #include <asm/kvm_mmu.h>
49
50 #include "vgic.h"
51
52 static bool handle_mmio_rao_wi(struct kvm_vcpu *vcpu,
53                                struct kvm_exit_mmio *mmio, phys_addr_t offset)
54 {
55         u32 reg = 0xffffffff;
56
57         vgic_reg_access(mmio, &reg, offset,
58                         ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
59
60         return false;
61 }
62
63 static bool handle_mmio_ctlr(struct kvm_vcpu *vcpu,
64                              struct kvm_exit_mmio *mmio, phys_addr_t offset)
65 {
66         u32 reg = 0;
67
68         /*
69          * Force ARE and DS to 1, the guest cannot change this.
70          * For the time being we only support Group1 interrupts.
71          */
72         if (vcpu->kvm->arch.vgic.enabled)
73                 reg = GICD_CTLR_ENABLE_SS_G1;
74         reg |= GICD_CTLR_ARE_NS | GICD_CTLR_DS;
75
76         vgic_reg_access(mmio, &reg, offset,
77                         ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
78         if (mmio->is_write) {
79                 vcpu->kvm->arch.vgic.enabled = !!(reg & GICD_CTLR_ENABLE_SS_G1);
80                 vgic_update_state(vcpu->kvm);
81                 return true;
82         }
83         return false;
84 }
85
86 /*
87  * As this implementation does not provide compatibility
88  * with GICv2 (ARE==1), we report zero CPUs in bits [5..7].
89  * Also LPIs and MBIs are not supported, so we set the respective bits to 0.
90  * Also we report at most 2**10=1024 interrupt IDs (to match 1024 SPIs).
91  */
92 #define INTERRUPT_ID_BITS 10
93 static bool handle_mmio_typer(struct kvm_vcpu *vcpu,
94                               struct kvm_exit_mmio *mmio, phys_addr_t offset)
95 {
96         u32 reg;
97
98         reg = (min(vcpu->kvm->arch.vgic.nr_irqs, 1024) >> 5) - 1;
99
100         reg |= (INTERRUPT_ID_BITS - 1) << 19;
101
102         vgic_reg_access(mmio, &reg, offset,
103                         ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
104
105         return false;
106 }
107
108 static bool handle_mmio_iidr(struct kvm_vcpu *vcpu,
109                              struct kvm_exit_mmio *mmio, phys_addr_t offset)
110 {
111         u32 reg;
112
113         reg = (PRODUCT_ID_KVM << 24) | (IMPLEMENTER_ARM << 0);
114         vgic_reg_access(mmio, &reg, offset,
115                         ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
116
117         return false;
118 }
119
120 static bool handle_mmio_set_enable_reg_dist(struct kvm_vcpu *vcpu,
121                                             struct kvm_exit_mmio *mmio,
122                                             phys_addr_t offset)
123 {
124         if (likely(offset >= VGIC_NR_PRIVATE_IRQS / 8))
125                 return vgic_handle_enable_reg(vcpu->kvm, mmio, offset,
126                                               vcpu->vcpu_id,
127                                               ACCESS_WRITE_SETBIT);
128
129         vgic_reg_access(mmio, NULL, offset,
130                         ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
131         return false;
132 }
133
134 static bool handle_mmio_clear_enable_reg_dist(struct kvm_vcpu *vcpu,
135                                               struct kvm_exit_mmio *mmio,
136                                               phys_addr_t offset)
137 {
138         if (likely(offset >= VGIC_NR_PRIVATE_IRQS / 8))
139                 return vgic_handle_enable_reg(vcpu->kvm, mmio, offset,
140                                               vcpu->vcpu_id,
141                                               ACCESS_WRITE_CLEARBIT);
142
143         vgic_reg_access(mmio, NULL, offset,
144                         ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
145         return false;
146 }
147
148 static bool handle_mmio_set_pending_reg_dist(struct kvm_vcpu *vcpu,
149                                              struct kvm_exit_mmio *mmio,
150                                              phys_addr_t offset)
151 {
152         if (likely(offset >= VGIC_NR_PRIVATE_IRQS / 8))
153                 return vgic_handle_set_pending_reg(vcpu->kvm, mmio, offset,
154                                                    vcpu->vcpu_id);
155
156         vgic_reg_access(mmio, NULL, offset,
157                         ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
158         return false;
159 }
160
161 static bool handle_mmio_clear_pending_reg_dist(struct kvm_vcpu *vcpu,
162                                                struct kvm_exit_mmio *mmio,
163                                                phys_addr_t offset)
164 {
165         if (likely(offset >= VGIC_NR_PRIVATE_IRQS / 8))
166                 return vgic_handle_clear_pending_reg(vcpu->kvm, mmio, offset,
167                                                      vcpu->vcpu_id);
168
169         vgic_reg_access(mmio, NULL, offset,
170                         ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
171         return false;
172 }
173
174 static bool handle_mmio_set_active_reg_dist(struct kvm_vcpu *vcpu,
175                                             struct kvm_exit_mmio *mmio,
176                                             phys_addr_t offset)
177 {
178         if (likely(offset >= VGIC_NR_PRIVATE_IRQS / 8))
179                 return vgic_handle_set_active_reg(vcpu->kvm, mmio, offset,
180                                                    vcpu->vcpu_id);
181
182         vgic_reg_access(mmio, NULL, offset,
183                         ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
184         return false;
185 }
186
187 static bool handle_mmio_clear_active_reg_dist(struct kvm_vcpu *vcpu,
188                                               struct kvm_exit_mmio *mmio,
189                                               phys_addr_t offset)
190 {
191         if (likely(offset >= VGIC_NR_PRIVATE_IRQS / 8))
192                 return vgic_handle_clear_active_reg(vcpu->kvm, mmio, offset,
193                                                     vcpu->vcpu_id);
194
195         vgic_reg_access(mmio, NULL, offset,
196                         ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
197         return false;
198 }
199
200 static bool handle_mmio_priority_reg_dist(struct kvm_vcpu *vcpu,
201                                           struct kvm_exit_mmio *mmio,
202                                           phys_addr_t offset)
203 {
204         u32 *reg;
205
206         if (unlikely(offset < VGIC_NR_PRIVATE_IRQS)) {
207                 vgic_reg_access(mmio, NULL, offset,
208                                 ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
209                 return false;
210         }
211
212         reg = vgic_bytemap_get_reg(&vcpu->kvm->arch.vgic.irq_priority,
213                                    vcpu->vcpu_id, offset);
214         vgic_reg_access(mmio, reg, offset,
215                 ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
216         return false;
217 }
218
219 static bool handle_mmio_cfg_reg_dist(struct kvm_vcpu *vcpu,
220                                      struct kvm_exit_mmio *mmio,
221                                      phys_addr_t offset)
222 {
223         u32 *reg;
224
225         if (unlikely(offset < VGIC_NR_PRIVATE_IRQS / 4)) {
226                 vgic_reg_access(mmio, NULL, offset,
227                                 ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
228                 return false;
229         }
230
231         reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_cfg,
232                                   vcpu->vcpu_id, offset >> 1);
233
234         return vgic_handle_cfg_reg(reg, mmio, offset);
235 }
236
237 /*
238  * We use a compressed version of the MPIDR (all 32 bits in one 32-bit word)
239  * when we store the target MPIDR written by the guest.
240  */
241 static u32 compress_mpidr(unsigned long mpidr)
242 {
243         u32 ret;
244
245         ret = MPIDR_AFFINITY_LEVEL(mpidr, 0);
246         ret |= MPIDR_AFFINITY_LEVEL(mpidr, 1) << 8;
247         ret |= MPIDR_AFFINITY_LEVEL(mpidr, 2) << 16;
248         ret |= MPIDR_AFFINITY_LEVEL(mpidr, 3) << 24;
249
250         return ret;
251 }
252
253 static unsigned long uncompress_mpidr(u32 value)
254 {
255         unsigned long mpidr;
256
257         mpidr  = ((value >>  0) & 0xFF) << MPIDR_LEVEL_SHIFT(0);
258         mpidr |= ((value >>  8) & 0xFF) << MPIDR_LEVEL_SHIFT(1);
259         mpidr |= ((value >> 16) & 0xFF) << MPIDR_LEVEL_SHIFT(2);
260         mpidr |= (u64)((value >> 24) & 0xFF) << MPIDR_LEVEL_SHIFT(3);
261
262         return mpidr;
263 }
264
265 /*
266  * Lookup the given MPIDR value to get the vcpu_id (if there is one)
267  * and store that in the irq_spi_cpu[] array.
268  * This limits the number of VCPUs to 255 for now, extending the data
269  * type (or storing kvm_vcpu pointers) should lift the limit.
270  * Store the original MPIDR value in an extra array to support read-as-written.
271  * Unallocated MPIDRs are translated to a special value and caught
272  * before any array accesses.
273  */
274 static bool handle_mmio_route_reg(struct kvm_vcpu *vcpu,
275                                   struct kvm_exit_mmio *mmio,
276                                   phys_addr_t offset)
277 {
278         struct kvm *kvm = vcpu->kvm;
279         struct vgic_dist *dist = &kvm->arch.vgic;
280         int spi;
281         u32 reg;
282         int vcpu_id;
283         unsigned long *bmap, mpidr;
284
285         /*
286          * The upper 32 bits of each 64 bit register are zero,
287          * as we don't support Aff3.
288          */
289         if ((offset & 4)) {
290                 vgic_reg_access(mmio, NULL, offset,
291                                 ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
292                 return false;
293         }
294
295         /* This region only covers SPIs, so no handling of private IRQs here. */
296         spi = offset / 8;
297
298         /* get the stored MPIDR for this IRQ */
299         mpidr = uncompress_mpidr(dist->irq_spi_mpidr[spi]);
300         reg = mpidr;
301
302         vgic_reg_access(mmio, &reg, offset,
303                         ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
304
305         if (!mmio->is_write)
306                 return false;
307
308         /*
309          * Now clear the currently assigned vCPU from the map, making room
310          * for the new one to be written below
311          */
312         vcpu = kvm_mpidr_to_vcpu(kvm, mpidr);
313         if (likely(vcpu)) {
314                 vcpu_id = vcpu->vcpu_id;
315                 bmap = vgic_bitmap_get_shared_map(&dist->irq_spi_target[vcpu_id]);
316                 __clear_bit(spi, bmap);
317         }
318
319         dist->irq_spi_mpidr[spi] = compress_mpidr(reg);
320         vcpu = kvm_mpidr_to_vcpu(kvm, reg & MPIDR_HWID_BITMASK);
321
322         /*
323          * The spec says that non-existent MPIDR values should not be
324          * forwarded to any existent (v)CPU, but should be able to become
325          * pending anyway. We simply keep the irq_spi_target[] array empty, so
326          * the interrupt will never be injected.
327          * irq_spi_cpu[irq] gets a magic value in this case.
328          */
329         if (likely(vcpu)) {
330                 vcpu_id = vcpu->vcpu_id;
331                 dist->irq_spi_cpu[spi] = vcpu_id;
332                 bmap = vgic_bitmap_get_shared_map(&dist->irq_spi_target[vcpu_id]);
333                 __set_bit(spi, bmap);
334         } else {
335                 dist->irq_spi_cpu[spi] = VCPU_NOT_ALLOCATED;
336         }
337
338         vgic_update_state(kvm);
339
340         return true;
341 }
342
343 /*
344  * We should be careful about promising too much when a guest reads
345  * this register. Don't claim to be like any hardware implementation,
346  * but just report the GIC as version 3 - which is what a Linux guest
347  * would check.
348  */
349 static bool handle_mmio_idregs(struct kvm_vcpu *vcpu,
350                                struct kvm_exit_mmio *mmio,
351                                phys_addr_t offset)
352 {
353         u32 reg = 0;
354
355         switch (offset + GICD_IDREGS) {
356         case GICD_PIDR2:
357                 reg = 0x3b;
358                 break;
359         }
360
361         vgic_reg_access(mmio, &reg, offset,
362                         ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
363
364         return false;
365 }
366
367 static const struct vgic_io_range vgic_v3_dist_ranges[] = {
368         {
369                 .base           = GICD_CTLR,
370                 .len            = 0x04,
371                 .bits_per_irq   = 0,
372                 .handle_mmio    = handle_mmio_ctlr,
373         },
374         {
375                 .base           = GICD_TYPER,
376                 .len            = 0x04,
377                 .bits_per_irq   = 0,
378                 .handle_mmio    = handle_mmio_typer,
379         },
380         {
381                 .base           = GICD_IIDR,
382                 .len            = 0x04,
383                 .bits_per_irq   = 0,
384                 .handle_mmio    = handle_mmio_iidr,
385         },
386         {
387                 /* this register is optional, it is RAZ/WI if not implemented */
388                 .base           = GICD_STATUSR,
389                 .len            = 0x04,
390                 .bits_per_irq   = 0,
391                 .handle_mmio    = handle_mmio_raz_wi,
392         },
393         {
394                 /* this write only register is WI when TYPER.MBIS=0 */
395                 .base           = GICD_SETSPI_NSR,
396                 .len            = 0x04,
397                 .bits_per_irq   = 0,
398                 .handle_mmio    = handle_mmio_raz_wi,
399         },
400         {
401                 /* this write only register is WI when TYPER.MBIS=0 */
402                 .base           = GICD_CLRSPI_NSR,
403                 .len            = 0x04,
404                 .bits_per_irq   = 0,
405                 .handle_mmio    = handle_mmio_raz_wi,
406         },
407         {
408                 /* this is RAZ/WI when DS=1 */
409                 .base           = GICD_SETSPI_SR,
410                 .len            = 0x04,
411                 .bits_per_irq   = 0,
412                 .handle_mmio    = handle_mmio_raz_wi,
413         },
414         {
415                 /* this is RAZ/WI when DS=1 */
416                 .base           = GICD_CLRSPI_SR,
417                 .len            = 0x04,
418                 .bits_per_irq   = 0,
419                 .handle_mmio    = handle_mmio_raz_wi,
420         },
421         {
422                 .base           = GICD_IGROUPR,
423                 .len            = 0x80,
424                 .bits_per_irq   = 1,
425                 .handle_mmio    = handle_mmio_rao_wi,
426         },
427         {
428                 .base           = GICD_ISENABLER,
429                 .len            = 0x80,
430                 .bits_per_irq   = 1,
431                 .handle_mmio    = handle_mmio_set_enable_reg_dist,
432         },
433         {
434                 .base           = GICD_ICENABLER,
435                 .len            = 0x80,
436                 .bits_per_irq   = 1,
437                 .handle_mmio    = handle_mmio_clear_enable_reg_dist,
438         },
439         {
440                 .base           = GICD_ISPENDR,
441                 .len            = 0x80,
442                 .bits_per_irq   = 1,
443                 .handle_mmio    = handle_mmio_set_pending_reg_dist,
444         },
445         {
446                 .base           = GICD_ICPENDR,
447                 .len            = 0x80,
448                 .bits_per_irq   = 1,
449                 .handle_mmio    = handle_mmio_clear_pending_reg_dist,
450         },
451         {
452                 .base           = GICD_ISACTIVER,
453                 .len            = 0x80,
454                 .bits_per_irq   = 1,
455                 .handle_mmio    = handle_mmio_set_active_reg_dist,
456         },
457         {
458                 .base           = GICD_ICACTIVER,
459                 .len            = 0x80,
460                 .bits_per_irq   = 1,
461                 .handle_mmio    = handle_mmio_clear_active_reg_dist,
462         },
463         {
464                 .base           = GICD_IPRIORITYR,
465                 .len            = 0x400,
466                 .bits_per_irq   = 8,
467                 .handle_mmio    = handle_mmio_priority_reg_dist,
468         },
469         {
470                 /* TARGETSRn is RES0 when ARE=1 */
471                 .base           = GICD_ITARGETSR,
472                 .len            = 0x400,
473                 .bits_per_irq   = 8,
474                 .handle_mmio    = handle_mmio_raz_wi,
475         },
476         {
477                 .base           = GICD_ICFGR,
478                 .len            = 0x100,
479                 .bits_per_irq   = 2,
480                 .handle_mmio    = handle_mmio_cfg_reg_dist,
481         },
482         {
483                 /* this is RAZ/WI when DS=1 */
484                 .base           = GICD_IGRPMODR,
485                 .len            = 0x80,
486                 .bits_per_irq   = 1,
487                 .handle_mmio    = handle_mmio_raz_wi,
488         },
489         {
490                 /* this is RAZ/WI when DS=1 */
491                 .base           = GICD_NSACR,
492                 .len            = 0x100,
493                 .bits_per_irq   = 2,
494                 .handle_mmio    = handle_mmio_raz_wi,
495         },
496         {
497                 /* this is RAZ/WI when ARE=1 */
498                 .base           = GICD_SGIR,
499                 .len            = 0x04,
500                 .handle_mmio    = handle_mmio_raz_wi,
501         },
502         {
503                 /* this is RAZ/WI when ARE=1 */
504                 .base           = GICD_CPENDSGIR,
505                 .len            = 0x10,
506                 .handle_mmio    = handle_mmio_raz_wi,
507         },
508         {
509                 /* this is RAZ/WI when ARE=1 */
510                 .base           = GICD_SPENDSGIR,
511                 .len            = 0x10,
512                 .handle_mmio    = handle_mmio_raz_wi,
513         },
514         {
515                 .base           = GICD_IROUTER + 0x100,
516                 .len            = 0x1ee0,
517                 .bits_per_irq   = 64,
518                 .handle_mmio    = handle_mmio_route_reg,
519         },
520         {
521                 .base           = GICD_IDREGS,
522                 .len            = 0x30,
523                 .bits_per_irq   = 0,
524                 .handle_mmio    = handle_mmio_idregs,
525         },
526         {},
527 };
528
529 static bool handle_mmio_ctlr_redist(struct kvm_vcpu *vcpu,
530                                     struct kvm_exit_mmio *mmio,
531                                     phys_addr_t offset)
532 {
533         /* since we don't support LPIs, this register is zero for now */
534         vgic_reg_access(mmio, NULL, offset,
535                         ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
536         return false;
537 }
538
539 static bool handle_mmio_typer_redist(struct kvm_vcpu *vcpu,
540                                      struct kvm_exit_mmio *mmio,
541                                      phys_addr_t offset)
542 {
543         u32 reg;
544         u64 mpidr;
545         struct kvm_vcpu *redist_vcpu = mmio->private;
546         int target_vcpu_id = redist_vcpu->vcpu_id;
547
548         /* the upper 32 bits contain the affinity value */
549         if ((offset & ~3) == 4) {
550                 mpidr = kvm_vcpu_get_mpidr_aff(redist_vcpu);
551                 reg = compress_mpidr(mpidr);
552
553                 vgic_reg_access(mmio, &reg, offset,
554                                 ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
555                 return false;
556         }
557
558         reg = redist_vcpu->vcpu_id << 8;
559         if (target_vcpu_id == atomic_read(&vcpu->kvm->online_vcpus) - 1)
560                 reg |= GICR_TYPER_LAST;
561         vgic_reg_access(mmio, &reg, offset,
562                         ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
563         return false;
564 }
565
566 static bool handle_mmio_set_enable_reg_redist(struct kvm_vcpu *vcpu,
567                                               struct kvm_exit_mmio *mmio,
568                                               phys_addr_t offset)
569 {
570         struct kvm_vcpu *redist_vcpu = mmio->private;
571
572         return vgic_handle_enable_reg(vcpu->kvm, mmio, offset,
573                                       redist_vcpu->vcpu_id,
574                                       ACCESS_WRITE_SETBIT);
575 }
576
577 static bool handle_mmio_clear_enable_reg_redist(struct kvm_vcpu *vcpu,
578                                                 struct kvm_exit_mmio *mmio,
579                                                 phys_addr_t offset)
580 {
581         struct kvm_vcpu *redist_vcpu = mmio->private;
582
583         return vgic_handle_enable_reg(vcpu->kvm, mmio, offset,
584                                       redist_vcpu->vcpu_id,
585                                       ACCESS_WRITE_CLEARBIT);
586 }
587
588 static bool handle_mmio_set_active_reg_redist(struct kvm_vcpu *vcpu,
589                                               struct kvm_exit_mmio *mmio,
590                                               phys_addr_t offset)
591 {
592         struct kvm_vcpu *redist_vcpu = mmio->private;
593
594         return vgic_handle_set_active_reg(vcpu->kvm, mmio, offset,
595                                           redist_vcpu->vcpu_id);
596 }
597
598 static bool handle_mmio_clear_active_reg_redist(struct kvm_vcpu *vcpu,
599                                                 struct kvm_exit_mmio *mmio,
600                                                 phys_addr_t offset)
601 {
602         struct kvm_vcpu *redist_vcpu = mmio->private;
603
604         return vgic_handle_clear_active_reg(vcpu->kvm, mmio, offset,
605                                              redist_vcpu->vcpu_id);
606 }
607
608 static bool handle_mmio_set_pending_reg_redist(struct kvm_vcpu *vcpu,
609                                                struct kvm_exit_mmio *mmio,
610                                                phys_addr_t offset)
611 {
612         struct kvm_vcpu *redist_vcpu = mmio->private;
613
614         return vgic_handle_set_pending_reg(vcpu->kvm, mmio, offset,
615                                            redist_vcpu->vcpu_id);
616 }
617
618 static bool handle_mmio_clear_pending_reg_redist(struct kvm_vcpu *vcpu,
619                                                  struct kvm_exit_mmio *mmio,
620                                                  phys_addr_t offset)
621 {
622         struct kvm_vcpu *redist_vcpu = mmio->private;
623
624         return vgic_handle_clear_pending_reg(vcpu->kvm, mmio, offset,
625                                              redist_vcpu->vcpu_id);
626 }
627
628 static bool handle_mmio_priority_reg_redist(struct kvm_vcpu *vcpu,
629                                             struct kvm_exit_mmio *mmio,
630                                             phys_addr_t offset)
631 {
632         struct kvm_vcpu *redist_vcpu = mmio->private;
633         u32 *reg;
634
635         reg = vgic_bytemap_get_reg(&vcpu->kvm->arch.vgic.irq_priority,
636                                    redist_vcpu->vcpu_id, offset);
637         vgic_reg_access(mmio, reg, offset,
638                         ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
639         return false;
640 }
641
642 static bool handle_mmio_cfg_reg_redist(struct kvm_vcpu *vcpu,
643                                        struct kvm_exit_mmio *mmio,
644                                        phys_addr_t offset)
645 {
646         struct kvm_vcpu *redist_vcpu = mmio->private;
647
648         u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_cfg,
649                                        redist_vcpu->vcpu_id, offset >> 1);
650
651         return vgic_handle_cfg_reg(reg, mmio, offset);
652 }
653
654 #define SGI_base(x) ((x) + SZ_64K)
655
656 static const struct vgic_io_range vgic_redist_ranges[] = {
657         {
658                 .base           = GICR_CTLR,
659                 .len            = 0x04,
660                 .bits_per_irq   = 0,
661                 .handle_mmio    = handle_mmio_ctlr_redist,
662         },
663         {
664                 .base           = GICR_TYPER,
665                 .len            = 0x08,
666                 .bits_per_irq   = 0,
667                 .handle_mmio    = handle_mmio_typer_redist,
668         },
669         {
670                 .base           = GICR_IIDR,
671                 .len            = 0x04,
672                 .bits_per_irq   = 0,
673                 .handle_mmio    = handle_mmio_iidr,
674         },
675         {
676                 .base           = GICR_WAKER,
677                 .len            = 0x04,
678                 .bits_per_irq   = 0,
679                 .handle_mmio    = handle_mmio_raz_wi,
680         },
681         {
682                 .base           = GICR_IDREGS,
683                 .len            = 0x30,
684                 .bits_per_irq   = 0,
685                 .handle_mmio    = handle_mmio_idregs,
686         },
687         {
688                 .base           = SGI_base(GICR_IGROUPR0),
689                 .len            = 0x04,
690                 .bits_per_irq   = 1,
691                 .handle_mmio    = handle_mmio_rao_wi,
692         },
693         {
694                 .base           = SGI_base(GICR_ISENABLER0),
695                 .len            = 0x04,
696                 .bits_per_irq   = 1,
697                 .handle_mmio    = handle_mmio_set_enable_reg_redist,
698         },
699         {
700                 .base           = SGI_base(GICR_ICENABLER0),
701                 .len            = 0x04,
702                 .bits_per_irq   = 1,
703                 .handle_mmio    = handle_mmio_clear_enable_reg_redist,
704         },
705         {
706                 .base           = SGI_base(GICR_ISPENDR0),
707                 .len            = 0x04,
708                 .bits_per_irq   = 1,
709                 .handle_mmio    = handle_mmio_set_pending_reg_redist,
710         },
711         {
712                 .base           = SGI_base(GICR_ICPENDR0),
713                 .len            = 0x04,
714                 .bits_per_irq   = 1,
715                 .handle_mmio    = handle_mmio_clear_pending_reg_redist,
716         },
717         {
718                 .base           = SGI_base(GICR_ISACTIVER0),
719                 .len            = 0x04,
720                 .bits_per_irq   = 1,
721                 .handle_mmio    = handle_mmio_set_active_reg_redist,
722         },
723         {
724                 .base           = SGI_base(GICR_ICACTIVER0),
725                 .len            = 0x04,
726                 .bits_per_irq   = 1,
727                 .handle_mmio    = handle_mmio_clear_active_reg_redist,
728         },
729         {
730                 .base           = SGI_base(GICR_IPRIORITYR0),
731                 .len            = 0x20,
732                 .bits_per_irq   = 8,
733                 .handle_mmio    = handle_mmio_priority_reg_redist,
734         },
735         {
736                 .base           = SGI_base(GICR_ICFGR0),
737                 .len            = 0x08,
738                 .bits_per_irq   = 2,
739                 .handle_mmio    = handle_mmio_cfg_reg_redist,
740         },
741         {
742                 .base           = SGI_base(GICR_IGRPMODR0),
743                 .len            = 0x04,
744                 .bits_per_irq   = 1,
745                 .handle_mmio    = handle_mmio_raz_wi,
746         },
747         {
748                 .base           = SGI_base(GICR_NSACR),
749                 .len            = 0x04,
750                 .handle_mmio    = handle_mmio_raz_wi,
751         },
752         {},
753 };
754
755 static bool vgic_v3_queue_sgi(struct kvm_vcpu *vcpu, int irq)
756 {
757         if (vgic_queue_irq(vcpu, 0, irq)) {
758                 vgic_dist_irq_clear_pending(vcpu, irq);
759                 vgic_cpu_irq_clear(vcpu, irq);
760                 return true;
761         }
762
763         return false;
764 }
765
766 static int vgic_v3_map_resources(struct kvm *kvm,
767                                  const struct vgic_params *params)
768 {
769         int ret = 0;
770         struct vgic_dist *dist = &kvm->arch.vgic;
771         gpa_t rdbase = dist->vgic_redist_base;
772         struct vgic_io_device *iodevs = NULL;
773         int i;
774
775         if (!irqchip_in_kernel(kvm))
776                 return 0;
777
778         mutex_lock(&kvm->lock);
779
780         if (vgic_ready(kvm))
781                 goto out;
782
783         if (IS_VGIC_ADDR_UNDEF(dist->vgic_dist_base) ||
784             IS_VGIC_ADDR_UNDEF(dist->vgic_redist_base)) {
785                 kvm_err("Need to set vgic distributor addresses first\n");
786                 ret = -ENXIO;
787                 goto out;
788         }
789
790         /*
791          * For a VGICv3 we require the userland to explicitly initialize
792          * the VGIC before we need to use it.
793          */
794         if (!vgic_initialized(kvm)) {
795                 ret = -EBUSY;
796                 goto out;
797         }
798
799         ret = vgic_register_kvm_io_dev(kvm, dist->vgic_dist_base,
800                                        GIC_V3_DIST_SIZE, vgic_v3_dist_ranges,
801                                        -1, &dist->dist_iodev);
802         if (ret)
803                 goto out;
804
805         iodevs = kcalloc(dist->nr_cpus, sizeof(iodevs[0]), GFP_KERNEL);
806         if (!iodevs) {
807                 ret = -ENOMEM;
808                 goto out_unregister;
809         }
810
811         for (i = 0; i < dist->nr_cpus; i++) {
812                 ret = vgic_register_kvm_io_dev(kvm, rdbase,
813                                                SZ_128K, vgic_redist_ranges,
814                                                i, &iodevs[i]);
815                 if (ret)
816                         goto out_unregister;
817                 rdbase += GIC_V3_REDIST_SIZE;
818         }
819
820         dist->redist_iodevs = iodevs;
821         dist->ready = true;
822         goto out;
823
824 out_unregister:
825         kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &dist->dist_iodev.dev);
826         if (iodevs) {
827                 for (i = 0; i < dist->nr_cpus; i++) {
828                         if (iodevs[i].dev.ops)
829                                 kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS,
830                                                           &iodevs[i].dev);
831                 }
832         }
833
834 out:
835         if (ret)
836                 kvm_vgic_destroy(kvm);
837         mutex_unlock(&kvm->lock);
838         return ret;
839 }
840
841 static int vgic_v3_init_model(struct kvm *kvm)
842 {
843         int i;
844         u32 mpidr;
845         struct vgic_dist *dist = &kvm->arch.vgic;
846         int nr_spis = dist->nr_irqs - VGIC_NR_PRIVATE_IRQS;
847
848         dist->irq_spi_mpidr = kcalloc(nr_spis, sizeof(dist->irq_spi_mpidr[0]),
849                                       GFP_KERNEL);
850
851         if (!dist->irq_spi_mpidr)
852                 return -ENOMEM;
853
854         /* Initialize the target VCPUs for each IRQ to VCPU 0 */
855         mpidr = compress_mpidr(kvm_vcpu_get_mpidr_aff(kvm_get_vcpu(kvm, 0)));
856         for (i = VGIC_NR_PRIVATE_IRQS; i < dist->nr_irqs; i++) {
857                 dist->irq_spi_cpu[i - VGIC_NR_PRIVATE_IRQS] = 0;
858                 dist->irq_spi_mpidr[i - VGIC_NR_PRIVATE_IRQS] = mpidr;
859                 vgic_bitmap_set_irq_val(dist->irq_spi_target, 0, i, 1);
860         }
861
862         return 0;
863 }
864
865 /* GICv3 does not keep track of SGI sources anymore. */
866 static void vgic_v3_add_sgi_source(struct kvm_vcpu *vcpu, int irq, int source)
867 {
868 }
869
870 void vgic_v3_init_emulation(struct kvm *kvm)
871 {
872         struct vgic_dist *dist = &kvm->arch.vgic;
873
874         dist->vm_ops.queue_sgi = vgic_v3_queue_sgi;
875         dist->vm_ops.add_sgi_source = vgic_v3_add_sgi_source;
876         dist->vm_ops.init_model = vgic_v3_init_model;
877         dist->vm_ops.map_resources = vgic_v3_map_resources;
878
879         kvm->arch.max_vcpus = KVM_MAX_VCPUS;
880 }
881
882 /*
883  * Compare a given affinity (level 1-3 and a level 0 mask, from the SGI
884  * generation register ICC_SGI1R_EL1) with a given VCPU.
885  * If the VCPU's MPIDR matches, return the level0 affinity, otherwise
886  * return -1.
887  */
888 static int match_mpidr(u64 sgi_aff, u16 sgi_cpu_mask, struct kvm_vcpu *vcpu)
889 {
890         unsigned long affinity;
891         int level0;
892
893         /*
894          * Split the current VCPU's MPIDR into affinity level 0 and the
895          * rest as this is what we have to compare against.
896          */
897         affinity = kvm_vcpu_get_mpidr_aff(vcpu);
898         level0 = MPIDR_AFFINITY_LEVEL(affinity, 0);
899         affinity &= ~MPIDR_LEVEL_MASK;
900
901         /* bail out if the upper three levels don't match */
902         if (sgi_aff != affinity)
903                 return -1;
904
905         /* Is this VCPU's bit set in the mask ? */
906         if (!(sgi_cpu_mask & BIT(level0)))
907                 return -1;
908
909         return level0;
910 }
911
912 #define SGI_AFFINITY_LEVEL(reg, level) \
913         ((((reg) & ICC_SGI1R_AFFINITY_## level ##_MASK) \
914         >> ICC_SGI1R_AFFINITY_## level ##_SHIFT) << MPIDR_LEVEL_SHIFT(level))
915
916 /**
917  * vgic_v3_dispatch_sgi - handle SGI requests from VCPUs
918  * @vcpu: The VCPU requesting a SGI
919  * @reg: The value written into the ICC_SGI1R_EL1 register by that VCPU
920  *
921  * With GICv3 (and ARE=1) CPUs trigger SGIs by writing to a system register.
922  * This will trap in sys_regs.c and call this function.
923  * This ICC_SGI1R_EL1 register contains the upper three affinity levels of the
924  * target processors as well as a bitmask of 16 Aff0 CPUs.
925  * If the interrupt routing mode bit is not set, we iterate over all VCPUs to
926  * check for matching ones. If this bit is set, we signal all, but not the
927  * calling VCPU.
928  */
929 void vgic_v3_dispatch_sgi(struct kvm_vcpu *vcpu, u64 reg)
930 {
931         struct kvm *kvm = vcpu->kvm;
932         struct kvm_vcpu *c_vcpu;
933         struct vgic_dist *dist = &kvm->arch.vgic;
934         u16 target_cpus;
935         u64 mpidr;
936         int sgi, c;
937         int vcpu_id = vcpu->vcpu_id;
938         bool broadcast;
939         int updated = 0;
940
941         sgi = (reg & ICC_SGI1R_SGI_ID_MASK) >> ICC_SGI1R_SGI_ID_SHIFT;
942         broadcast = reg & BIT(ICC_SGI1R_IRQ_ROUTING_MODE_BIT);
943         target_cpus = (reg & ICC_SGI1R_TARGET_LIST_MASK) >> ICC_SGI1R_TARGET_LIST_SHIFT;
944         mpidr = SGI_AFFINITY_LEVEL(reg, 3);
945         mpidr |= SGI_AFFINITY_LEVEL(reg, 2);
946         mpidr |= SGI_AFFINITY_LEVEL(reg, 1);
947
948         /*
949          * We take the dist lock here, because we come from the sysregs
950          * code path and not from the MMIO one (which already takes the lock).
951          */
952         spin_lock(&dist->lock);
953
954         /*
955          * We iterate over all VCPUs to find the MPIDRs matching the request.
956          * If we have handled one CPU, we clear it's bit to detect early
957          * if we are already finished. This avoids iterating through all
958          * VCPUs when most of the times we just signal a single VCPU.
959          */
960         kvm_for_each_vcpu(c, c_vcpu, kvm) {
961
962                 /* Exit early if we have dealt with all requested CPUs */
963                 if (!broadcast && target_cpus == 0)
964                         break;
965
966                  /* Don't signal the calling VCPU */
967                 if (broadcast && c == vcpu_id)
968                         continue;
969
970                 if (!broadcast) {
971                         int level0;
972
973                         level0 = match_mpidr(mpidr, target_cpus, c_vcpu);
974                         if (level0 == -1)
975                                 continue;
976
977                         /* remove this matching VCPU from the mask */
978                         target_cpus &= ~BIT(level0);
979                 }
980
981                 /* Flag the SGI as pending */
982                 vgic_dist_irq_set_pending(c_vcpu, sgi);
983                 updated = 1;
984                 kvm_debug("SGI%d from CPU%d to CPU%d\n", sgi, vcpu_id, c);
985         }
986         if (updated)
987                 vgic_update_state(vcpu->kvm);
988         spin_unlock(&dist->lock);
989         if (updated)
990                 vgic_kick_vcpus(vcpu->kvm);
991 }
992
993 static int vgic_v3_create(struct kvm_device *dev, u32 type)
994 {
995         return kvm_vgic_create(dev->kvm, type);
996 }
997
998 static void vgic_v3_destroy(struct kvm_device *dev)
999 {
1000         kfree(dev);
1001 }
1002
1003 static int vgic_v3_set_attr(struct kvm_device *dev,
1004                             struct kvm_device_attr *attr)
1005 {
1006         int ret;
1007
1008         ret = vgic_set_common_attr(dev, attr);
1009         if (ret != -ENXIO)
1010                 return ret;
1011
1012         switch (attr->group) {
1013         case KVM_DEV_ARM_VGIC_GRP_DIST_REGS:
1014         case KVM_DEV_ARM_VGIC_GRP_CPU_REGS:
1015                 return -ENXIO;
1016         }
1017
1018         return -ENXIO;
1019 }
1020
1021 static int vgic_v3_get_attr(struct kvm_device *dev,
1022                             struct kvm_device_attr *attr)
1023 {
1024         int ret;
1025
1026         ret = vgic_get_common_attr(dev, attr);
1027         if (ret != -ENXIO)
1028                 return ret;
1029
1030         switch (attr->group) {
1031         case KVM_DEV_ARM_VGIC_GRP_DIST_REGS:
1032         case KVM_DEV_ARM_VGIC_GRP_CPU_REGS:
1033                 return -ENXIO;
1034         }
1035
1036         return -ENXIO;
1037 }
1038
1039 static int vgic_v3_has_attr(struct kvm_device *dev,
1040                             struct kvm_device_attr *attr)
1041 {
1042         switch (attr->group) {
1043         case KVM_DEV_ARM_VGIC_GRP_ADDR:
1044                 switch (attr->attr) {
1045                 case KVM_VGIC_V2_ADDR_TYPE_DIST:
1046                 case KVM_VGIC_V2_ADDR_TYPE_CPU:
1047                         return -ENXIO;
1048                 case KVM_VGIC_V3_ADDR_TYPE_DIST:
1049                 case KVM_VGIC_V3_ADDR_TYPE_REDIST:
1050                         return 0;
1051                 }
1052                 break;
1053         case KVM_DEV_ARM_VGIC_GRP_DIST_REGS:
1054         case KVM_DEV_ARM_VGIC_GRP_CPU_REGS:
1055                 return -ENXIO;
1056         case KVM_DEV_ARM_VGIC_GRP_NR_IRQS:
1057                 return 0;
1058         case KVM_DEV_ARM_VGIC_GRP_CTRL:
1059                 switch (attr->attr) {
1060                 case KVM_DEV_ARM_VGIC_CTRL_INIT:
1061                         return 0;
1062                 }
1063         }
1064         return -ENXIO;
1065 }
1066
1067 struct kvm_device_ops kvm_arm_vgic_v3_ops = {
1068         .name = "kvm-arm-vgic-v3",
1069         .create = vgic_v3_create,
1070         .destroy = vgic_v3_destroy,
1071         .set_attr = vgic_v3_set_attr,
1072         .get_attr = vgic_v3_get_attr,
1073         .has_attr = vgic_v3_has_attr,
1074 };