xref: /openbmc/linux/arch/arm64/kvm/vgic/vgic-mmio-v3.c (revision f97769fd)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VGICv3 MMIO handling functions
4  */
5 
6 #include <linux/bitfield.h>
7 #include <linux/irqchip/arm-gic-v3.h>
8 #include <linux/kvm.h>
9 #include <linux/kvm_host.h>
10 #include <linux/interrupt.h>
11 #include <kvm/iodev.h>
12 #include <kvm/arm_vgic.h>
13 
14 #include <asm/kvm_emulate.h>
15 #include <asm/kvm_arm.h>
16 #include <asm/kvm_mmu.h>
17 
18 #include "vgic.h"
19 #include "vgic-mmio.h"
20 
21 /* extract @num bytes at @offset bytes offset in data */
22 unsigned long extract_bytes(u64 data, unsigned int offset,
23 			    unsigned int num)
24 {
25 	return (data >> (offset * 8)) & GENMASK_ULL(num * 8 - 1, 0);
26 }
27 
28 /* allows updates of any half of a 64-bit register (or the whole thing) */
29 u64 update_64bit_reg(u64 reg, unsigned int offset, unsigned int len,
30 		     unsigned long val)
31 {
32 	int lower = (offset & 4) * 8;
33 	int upper = lower + 8 * len - 1;
34 
35 	reg &= ~GENMASK_ULL(upper, lower);
36 	val &= GENMASK_ULL(len * 8 - 1, 0);
37 
38 	return reg | ((u64)val << lower);
39 }
40 
41 bool vgic_has_its(struct kvm *kvm)
42 {
43 	struct vgic_dist *dist = &kvm->arch.vgic;
44 
45 	if (dist->vgic_model != KVM_DEV_TYPE_ARM_VGIC_V3)
46 		return false;
47 
48 	return dist->has_its;
49 }
50 
51 bool vgic_supports_direct_msis(struct kvm *kvm)
52 {
53 	return (kvm_vgic_global_state.has_gicv4_1 ||
54 		(kvm_vgic_global_state.has_gicv4 && vgic_has_its(kvm)));
55 }
56 
57 /*
58  * The Revision field in the IIDR have the following meanings:
59  *
60  * Revision 2: Interrupt groups are guest-configurable and signaled using
61  * 	       their configured groups.
62  */
63 
64 static unsigned long vgic_mmio_read_v3_misc(struct kvm_vcpu *vcpu,
65 					    gpa_t addr, unsigned int len)
66 {
67 	struct vgic_dist *vgic = &vcpu->kvm->arch.vgic;
68 	u32 value = 0;
69 
70 	switch (addr & 0x0c) {
71 	case GICD_CTLR:
72 		if (vgic->enabled)
73 			value |= GICD_CTLR_ENABLE_SS_G1;
74 		value |= GICD_CTLR_ARE_NS | GICD_CTLR_DS;
75 		if (vgic->nassgireq)
76 			value |= GICD_CTLR_nASSGIreq;
77 		break;
78 	case GICD_TYPER:
79 		value = vgic->nr_spis + VGIC_NR_PRIVATE_IRQS;
80 		value = (value >> 5) - 1;
81 		if (vgic_has_its(vcpu->kvm)) {
82 			value |= (INTERRUPT_ID_BITS_ITS - 1) << 19;
83 			value |= GICD_TYPER_LPIS;
84 		} else {
85 			value |= (INTERRUPT_ID_BITS_SPIS - 1) << 19;
86 		}
87 		break;
88 	case GICD_TYPER2:
89 		if (kvm_vgic_global_state.has_gicv4_1)
90 			value = GICD_TYPER2_nASSGIcap;
91 		break;
92 	case GICD_IIDR:
93 		value = (PRODUCT_ID_KVM << GICD_IIDR_PRODUCT_ID_SHIFT) |
94 			(vgic->implementation_rev << GICD_IIDR_REVISION_SHIFT) |
95 			(IMPLEMENTER_ARM << GICD_IIDR_IMPLEMENTER_SHIFT);
96 		break;
97 	default:
98 		return 0;
99 	}
100 
101 	return value;
102 }
103 
104 static void vgic_mmio_write_v3_misc(struct kvm_vcpu *vcpu,
105 				    gpa_t addr, unsigned int len,
106 				    unsigned long val)
107 {
108 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
109 
110 	switch (addr & 0x0c) {
111 	case GICD_CTLR: {
112 		bool was_enabled, is_hwsgi;
113 
114 		mutex_lock(&vcpu->kvm->lock);
115 
116 		was_enabled = dist->enabled;
117 		is_hwsgi = dist->nassgireq;
118 
119 		dist->enabled = val & GICD_CTLR_ENABLE_SS_G1;
120 
121 		/* Not a GICv4.1? No HW SGIs */
122 		if (!kvm_vgic_global_state.has_gicv4_1)
123 			val &= ~GICD_CTLR_nASSGIreq;
124 
125 		/* Dist stays enabled? nASSGIreq is RO */
126 		if (was_enabled && dist->enabled) {
127 			val &= ~GICD_CTLR_nASSGIreq;
128 			val |= FIELD_PREP(GICD_CTLR_nASSGIreq, is_hwsgi);
129 		}
130 
131 		/* Switching HW SGIs? */
132 		dist->nassgireq = val & GICD_CTLR_nASSGIreq;
133 		if (is_hwsgi != dist->nassgireq)
134 			vgic_v4_configure_vsgis(vcpu->kvm);
135 
136 		if (kvm_vgic_global_state.has_gicv4_1 &&
137 		    was_enabled != dist->enabled)
138 			kvm_make_all_cpus_request(vcpu->kvm, KVM_REQ_RELOAD_GICv4);
139 		else if (!was_enabled && dist->enabled)
140 			vgic_kick_vcpus(vcpu->kvm);
141 
142 		mutex_unlock(&vcpu->kvm->lock);
143 		break;
144 	}
145 	case GICD_TYPER:
146 	case GICD_TYPER2:
147 	case GICD_IIDR:
148 		/* This is at best for documentation purposes... */
149 		return;
150 	}
151 }
152 
153 static int vgic_mmio_uaccess_write_v3_misc(struct kvm_vcpu *vcpu,
154 					   gpa_t addr, unsigned int len,
155 					   unsigned long val)
156 {
157 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
158 
159 	switch (addr & 0x0c) {
160 	case GICD_TYPER2:
161 	case GICD_IIDR:
162 		if (val != vgic_mmio_read_v3_misc(vcpu, addr, len))
163 			return -EINVAL;
164 		return 0;
165 	case GICD_CTLR:
166 		/* Not a GICv4.1? No HW SGIs */
167 		if (!kvm_vgic_global_state.has_gicv4_1)
168 			val &= ~GICD_CTLR_nASSGIreq;
169 
170 		dist->enabled = val & GICD_CTLR_ENABLE_SS_G1;
171 		dist->nassgireq = val & GICD_CTLR_nASSGIreq;
172 		return 0;
173 	}
174 
175 	vgic_mmio_write_v3_misc(vcpu, addr, len, val);
176 	return 0;
177 }
178 
179 static unsigned long vgic_mmio_read_irouter(struct kvm_vcpu *vcpu,
180 					    gpa_t addr, unsigned int len)
181 {
182 	int intid = VGIC_ADDR_TO_INTID(addr, 64);
183 	struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, NULL, intid);
184 	unsigned long ret = 0;
185 
186 	if (!irq)
187 		return 0;
188 
189 	/* The upper word is RAZ for us. */
190 	if (!(addr & 4))
191 		ret = extract_bytes(READ_ONCE(irq->mpidr), addr & 7, len);
192 
193 	vgic_put_irq(vcpu->kvm, irq);
194 	return ret;
195 }
196 
197 static void vgic_mmio_write_irouter(struct kvm_vcpu *vcpu,
198 				    gpa_t addr, unsigned int len,
199 				    unsigned long val)
200 {
201 	int intid = VGIC_ADDR_TO_INTID(addr, 64);
202 	struct vgic_irq *irq;
203 	unsigned long flags;
204 
205 	/* The upper word is WI for us since we don't implement Aff3. */
206 	if (addr & 4)
207 		return;
208 
209 	irq = vgic_get_irq(vcpu->kvm, NULL, intid);
210 
211 	if (!irq)
212 		return;
213 
214 	raw_spin_lock_irqsave(&irq->irq_lock, flags);
215 
216 	/* We only care about and preserve Aff0, Aff1 and Aff2. */
217 	irq->mpidr = val & GENMASK(23, 0);
218 	irq->target_vcpu = kvm_mpidr_to_vcpu(vcpu->kvm, irq->mpidr);
219 
220 	raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
221 	vgic_put_irq(vcpu->kvm, irq);
222 }
223 
224 static unsigned long vgic_mmio_read_v3r_ctlr(struct kvm_vcpu *vcpu,
225 					     gpa_t addr, unsigned int len)
226 {
227 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
228 
229 	return vgic_cpu->lpis_enabled ? GICR_CTLR_ENABLE_LPIS : 0;
230 }
231 
232 
233 static void vgic_mmio_write_v3r_ctlr(struct kvm_vcpu *vcpu,
234 				     gpa_t addr, unsigned int len,
235 				     unsigned long val)
236 {
237 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
238 	bool was_enabled = vgic_cpu->lpis_enabled;
239 
240 	if (!vgic_has_its(vcpu->kvm))
241 		return;
242 
243 	vgic_cpu->lpis_enabled = val & GICR_CTLR_ENABLE_LPIS;
244 
245 	if (was_enabled && !vgic_cpu->lpis_enabled) {
246 		vgic_flush_pending_lpis(vcpu);
247 		vgic_its_invalidate_cache(vcpu->kvm);
248 	}
249 
250 	if (!was_enabled && vgic_cpu->lpis_enabled)
251 		vgic_enable_lpis(vcpu);
252 }
253 
254 static unsigned long vgic_mmio_read_v3r_typer(struct kvm_vcpu *vcpu,
255 					      gpa_t addr, unsigned int len)
256 {
257 	unsigned long mpidr = kvm_vcpu_get_mpidr_aff(vcpu);
258 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
259 	struct vgic_redist_region *rdreg = vgic_cpu->rdreg;
260 	int target_vcpu_id = vcpu->vcpu_id;
261 	gpa_t last_rdist_typer = rdreg->base + GICR_TYPER +
262 			(rdreg->free_index - 1) * KVM_VGIC_V3_REDIST_SIZE;
263 	u64 value;
264 
265 	value = (u64)(mpidr & GENMASK(23, 0)) << 32;
266 	value |= ((target_vcpu_id & 0xffff) << 8);
267 
268 	if (addr == last_rdist_typer)
269 		value |= GICR_TYPER_LAST;
270 	if (vgic_has_its(vcpu->kvm))
271 		value |= GICR_TYPER_PLPIS;
272 
273 	return extract_bytes(value, addr & 7, len);
274 }
275 
276 static unsigned long vgic_mmio_read_v3r_iidr(struct kvm_vcpu *vcpu,
277 					     gpa_t addr, unsigned int len)
278 {
279 	return (PRODUCT_ID_KVM << 24) | (IMPLEMENTER_ARM << 0);
280 }
281 
282 static unsigned long vgic_mmio_read_v3_idregs(struct kvm_vcpu *vcpu,
283 					      gpa_t addr, unsigned int len)
284 {
285 	switch (addr & 0xffff) {
286 	case GICD_PIDR2:
287 		/* report a GICv3 compliant implementation */
288 		return 0x3b;
289 	}
290 
291 	return 0;
292 }
293 
294 static unsigned long vgic_v3_uaccess_read_pending(struct kvm_vcpu *vcpu,
295 						  gpa_t addr, unsigned int len)
296 {
297 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
298 	u32 value = 0;
299 	int i;
300 
301 	/*
302 	 * pending state of interrupt is latched in pending_latch variable.
303 	 * Userspace will save and restore pending state and line_level
304 	 * separately.
305 	 * Refer to Documentation/virt/kvm/devices/arm-vgic-v3.rst
306 	 * for handling of ISPENDR and ICPENDR.
307 	 */
308 	for (i = 0; i < len * 8; i++) {
309 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
310 		bool state = irq->pending_latch;
311 
312 		if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
313 			int err;
314 
315 			err = irq_get_irqchip_state(irq->host_irq,
316 						    IRQCHIP_STATE_PENDING,
317 						    &state);
318 			WARN_ON(err);
319 		}
320 
321 		if (state)
322 			value |= (1U << i);
323 
324 		vgic_put_irq(vcpu->kvm, irq);
325 	}
326 
327 	return value;
328 }
329 
330 static int vgic_v3_uaccess_write_pending(struct kvm_vcpu *vcpu,
331 					 gpa_t addr, unsigned int len,
332 					 unsigned long val)
333 {
334 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
335 	int i;
336 	unsigned long flags;
337 
338 	for (i = 0; i < len * 8; i++) {
339 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
340 
341 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
342 		if (test_bit(i, &val)) {
343 			/*
344 			 * pending_latch is set irrespective of irq type
345 			 * (level or edge) to avoid dependency that VM should
346 			 * restore irq config before pending info.
347 			 */
348 			irq->pending_latch = true;
349 			vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
350 		} else {
351 			irq->pending_latch = false;
352 			raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
353 		}
354 
355 		vgic_put_irq(vcpu->kvm, irq);
356 	}
357 
358 	return 0;
359 }
360 
361 /* We want to avoid outer shareable. */
362 u64 vgic_sanitise_shareability(u64 field)
363 {
364 	switch (field) {
365 	case GIC_BASER_OuterShareable:
366 		return GIC_BASER_InnerShareable;
367 	default:
368 		return field;
369 	}
370 }
371 
372 /* Avoid any inner non-cacheable mapping. */
373 u64 vgic_sanitise_inner_cacheability(u64 field)
374 {
375 	switch (field) {
376 	case GIC_BASER_CACHE_nCnB:
377 	case GIC_BASER_CACHE_nC:
378 		return GIC_BASER_CACHE_RaWb;
379 	default:
380 		return field;
381 	}
382 }
383 
384 /* Non-cacheable or same-as-inner are OK. */
385 u64 vgic_sanitise_outer_cacheability(u64 field)
386 {
387 	switch (field) {
388 	case GIC_BASER_CACHE_SameAsInner:
389 	case GIC_BASER_CACHE_nC:
390 		return field;
391 	default:
392 		return GIC_BASER_CACHE_SameAsInner;
393 	}
394 }
395 
396 u64 vgic_sanitise_field(u64 reg, u64 field_mask, int field_shift,
397 			u64 (*sanitise_fn)(u64))
398 {
399 	u64 field = (reg & field_mask) >> field_shift;
400 
401 	field = sanitise_fn(field) << field_shift;
402 	return (reg & ~field_mask) | field;
403 }
404 
405 #define PROPBASER_RES0_MASK						\
406 	(GENMASK_ULL(63, 59) | GENMASK_ULL(55, 52) | GENMASK_ULL(6, 5))
407 #define PENDBASER_RES0_MASK						\
408 	(BIT_ULL(63) | GENMASK_ULL(61, 59) | GENMASK_ULL(55, 52) |	\
409 	 GENMASK_ULL(15, 12) | GENMASK_ULL(6, 0))
410 
411 static u64 vgic_sanitise_pendbaser(u64 reg)
412 {
413 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_SHAREABILITY_MASK,
414 				  GICR_PENDBASER_SHAREABILITY_SHIFT,
415 				  vgic_sanitise_shareability);
416 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_INNER_CACHEABILITY_MASK,
417 				  GICR_PENDBASER_INNER_CACHEABILITY_SHIFT,
418 				  vgic_sanitise_inner_cacheability);
419 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_OUTER_CACHEABILITY_MASK,
420 				  GICR_PENDBASER_OUTER_CACHEABILITY_SHIFT,
421 				  vgic_sanitise_outer_cacheability);
422 
423 	reg &= ~PENDBASER_RES0_MASK;
424 
425 	return reg;
426 }
427 
428 static u64 vgic_sanitise_propbaser(u64 reg)
429 {
430 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_SHAREABILITY_MASK,
431 				  GICR_PROPBASER_SHAREABILITY_SHIFT,
432 				  vgic_sanitise_shareability);
433 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_INNER_CACHEABILITY_MASK,
434 				  GICR_PROPBASER_INNER_CACHEABILITY_SHIFT,
435 				  vgic_sanitise_inner_cacheability);
436 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_OUTER_CACHEABILITY_MASK,
437 				  GICR_PROPBASER_OUTER_CACHEABILITY_SHIFT,
438 				  vgic_sanitise_outer_cacheability);
439 
440 	reg &= ~PROPBASER_RES0_MASK;
441 	return reg;
442 }
443 
444 static unsigned long vgic_mmio_read_propbase(struct kvm_vcpu *vcpu,
445 					     gpa_t addr, unsigned int len)
446 {
447 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
448 
449 	return extract_bytes(dist->propbaser, addr & 7, len);
450 }
451 
452 static void vgic_mmio_write_propbase(struct kvm_vcpu *vcpu,
453 				     gpa_t addr, unsigned int len,
454 				     unsigned long val)
455 {
456 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
457 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
458 	u64 old_propbaser, propbaser;
459 
460 	/* Storing a value with LPIs already enabled is undefined */
461 	if (vgic_cpu->lpis_enabled)
462 		return;
463 
464 	do {
465 		old_propbaser = READ_ONCE(dist->propbaser);
466 		propbaser = old_propbaser;
467 		propbaser = update_64bit_reg(propbaser, addr & 4, len, val);
468 		propbaser = vgic_sanitise_propbaser(propbaser);
469 	} while (cmpxchg64(&dist->propbaser, old_propbaser,
470 			   propbaser) != old_propbaser);
471 }
472 
473 static unsigned long vgic_mmio_read_pendbase(struct kvm_vcpu *vcpu,
474 					     gpa_t addr, unsigned int len)
475 {
476 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
477 	u64 value = vgic_cpu->pendbaser;
478 
479 	value &= ~GICR_PENDBASER_PTZ;
480 
481 	return extract_bytes(value, addr & 7, len);
482 }
483 
484 static void vgic_mmio_write_pendbase(struct kvm_vcpu *vcpu,
485 				     gpa_t addr, unsigned int len,
486 				     unsigned long val)
487 {
488 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
489 	u64 old_pendbaser, pendbaser;
490 
491 	/* Storing a value with LPIs already enabled is undefined */
492 	if (vgic_cpu->lpis_enabled)
493 		return;
494 
495 	do {
496 		old_pendbaser = READ_ONCE(vgic_cpu->pendbaser);
497 		pendbaser = old_pendbaser;
498 		pendbaser = update_64bit_reg(pendbaser, addr & 4, len, val);
499 		pendbaser = vgic_sanitise_pendbaser(pendbaser);
500 	} while (cmpxchg64(&vgic_cpu->pendbaser, old_pendbaser,
501 			   pendbaser) != old_pendbaser);
502 }
503 
504 /*
505  * The GICv3 per-IRQ registers are split to control PPIs and SGIs in the
506  * redistributors, while SPIs are covered by registers in the distributor
507  * block. Trying to set private IRQs in this block gets ignored.
508  * We take some special care here to fix the calculation of the register
509  * offset.
510  */
511 #define REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(off, rd, wr, ur, uw, bpi, acc) \
512 	{								\
513 		.reg_offset = off,					\
514 		.bits_per_irq = bpi,					\
515 		.len = (bpi * VGIC_NR_PRIVATE_IRQS) / 8,		\
516 		.access_flags = acc,					\
517 		.read = vgic_mmio_read_raz,				\
518 		.write = vgic_mmio_write_wi,				\
519 	}, {								\
520 		.reg_offset = off + (bpi * VGIC_NR_PRIVATE_IRQS) / 8,	\
521 		.bits_per_irq = bpi,					\
522 		.len = (bpi * (1024 - VGIC_NR_PRIVATE_IRQS)) / 8,	\
523 		.access_flags = acc,					\
524 		.read = rd,						\
525 		.write = wr,						\
526 		.uaccess_read = ur,					\
527 		.uaccess_write = uw,					\
528 	}
529 
530 static const struct vgic_register_region vgic_v3_dist_registers[] = {
531 	REGISTER_DESC_WITH_LENGTH_UACCESS(GICD_CTLR,
532 		vgic_mmio_read_v3_misc, vgic_mmio_write_v3_misc,
533 		NULL, vgic_mmio_uaccess_write_v3_misc,
534 		16, VGIC_ACCESS_32bit),
535 	REGISTER_DESC_WITH_LENGTH(GICD_STATUSR,
536 		vgic_mmio_read_rao, vgic_mmio_write_wi, 4,
537 		VGIC_ACCESS_32bit),
538 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGROUPR,
539 		vgic_mmio_read_group, vgic_mmio_write_group, NULL, NULL, 1,
540 		VGIC_ACCESS_32bit),
541 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISENABLER,
542 		vgic_mmio_read_enable, vgic_mmio_write_senable,
543 		NULL, vgic_uaccess_write_senable, 1,
544 		VGIC_ACCESS_32bit),
545 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICENABLER,
546 		vgic_mmio_read_enable, vgic_mmio_write_cenable,
547 	       NULL, vgic_uaccess_write_cenable, 1,
548 		VGIC_ACCESS_32bit),
549 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISPENDR,
550 		vgic_mmio_read_pending, vgic_mmio_write_spending,
551 		vgic_v3_uaccess_read_pending, vgic_v3_uaccess_write_pending, 1,
552 		VGIC_ACCESS_32bit),
553 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICPENDR,
554 		vgic_mmio_read_pending, vgic_mmio_write_cpending,
555 		vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 1,
556 		VGIC_ACCESS_32bit),
557 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISACTIVER,
558 		vgic_mmio_read_active, vgic_mmio_write_sactive,
559 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 1,
560 		VGIC_ACCESS_32bit),
561 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICACTIVER,
562 		vgic_mmio_read_active, vgic_mmio_write_cactive,
563 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive,
564 		1, VGIC_ACCESS_32bit),
565 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IPRIORITYR,
566 		vgic_mmio_read_priority, vgic_mmio_write_priority, NULL, NULL,
567 		8, VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
568 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ITARGETSR,
569 		vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 8,
570 		VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
571 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICFGR,
572 		vgic_mmio_read_config, vgic_mmio_write_config, NULL, NULL, 2,
573 		VGIC_ACCESS_32bit),
574 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGRPMODR,
575 		vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 1,
576 		VGIC_ACCESS_32bit),
577 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IROUTER,
578 		vgic_mmio_read_irouter, vgic_mmio_write_irouter, NULL, NULL, 64,
579 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
580 	REGISTER_DESC_WITH_LENGTH(GICD_IDREGS,
581 		vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48,
582 		VGIC_ACCESS_32bit),
583 };
584 
585 static const struct vgic_register_region vgic_v3_rd_registers[] = {
586 	/* RD_base registers */
587 	REGISTER_DESC_WITH_LENGTH(GICR_CTLR,
588 		vgic_mmio_read_v3r_ctlr, vgic_mmio_write_v3r_ctlr, 4,
589 		VGIC_ACCESS_32bit),
590 	REGISTER_DESC_WITH_LENGTH(GICR_STATUSR,
591 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
592 		VGIC_ACCESS_32bit),
593 	REGISTER_DESC_WITH_LENGTH(GICR_IIDR,
594 		vgic_mmio_read_v3r_iidr, vgic_mmio_write_wi, 4,
595 		VGIC_ACCESS_32bit),
596 	REGISTER_DESC_WITH_LENGTH(GICR_TYPER,
597 		vgic_mmio_read_v3r_typer, vgic_mmio_write_wi, 8,
598 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
599 	REGISTER_DESC_WITH_LENGTH(GICR_WAKER,
600 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
601 		VGIC_ACCESS_32bit),
602 	REGISTER_DESC_WITH_LENGTH(GICR_PROPBASER,
603 		vgic_mmio_read_propbase, vgic_mmio_write_propbase, 8,
604 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
605 	REGISTER_DESC_WITH_LENGTH(GICR_PENDBASER,
606 		vgic_mmio_read_pendbase, vgic_mmio_write_pendbase, 8,
607 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
608 	REGISTER_DESC_WITH_LENGTH(GICR_IDREGS,
609 		vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48,
610 		VGIC_ACCESS_32bit),
611 	/* SGI_base registers */
612 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGROUPR0,
613 		vgic_mmio_read_group, vgic_mmio_write_group, 4,
614 		VGIC_ACCESS_32bit),
615 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISENABLER0,
616 		vgic_mmio_read_enable, vgic_mmio_write_senable,
617 		NULL, vgic_uaccess_write_senable, 4,
618 		VGIC_ACCESS_32bit),
619 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICENABLER0,
620 		vgic_mmio_read_enable, vgic_mmio_write_cenable,
621 		NULL, vgic_uaccess_write_cenable, 4,
622 		VGIC_ACCESS_32bit),
623 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISPENDR0,
624 		vgic_mmio_read_pending, vgic_mmio_write_spending,
625 		vgic_v3_uaccess_read_pending, vgic_v3_uaccess_write_pending, 4,
626 		VGIC_ACCESS_32bit),
627 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICPENDR0,
628 		vgic_mmio_read_pending, vgic_mmio_write_cpending,
629 		vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 4,
630 		VGIC_ACCESS_32bit),
631 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISACTIVER0,
632 		vgic_mmio_read_active, vgic_mmio_write_sactive,
633 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 4,
634 		VGIC_ACCESS_32bit),
635 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICACTIVER0,
636 		vgic_mmio_read_active, vgic_mmio_write_cactive,
637 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive, 4,
638 		VGIC_ACCESS_32bit),
639 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IPRIORITYR0,
640 		vgic_mmio_read_priority, vgic_mmio_write_priority, 32,
641 		VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
642 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_ICFGR0,
643 		vgic_mmio_read_config, vgic_mmio_write_config, 8,
644 		VGIC_ACCESS_32bit),
645 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGRPMODR0,
646 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
647 		VGIC_ACCESS_32bit),
648 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_NSACR,
649 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
650 		VGIC_ACCESS_32bit),
651 };
652 
653 unsigned int vgic_v3_init_dist_iodev(struct vgic_io_device *dev)
654 {
655 	dev->regions = vgic_v3_dist_registers;
656 	dev->nr_regions = ARRAY_SIZE(vgic_v3_dist_registers);
657 
658 	kvm_iodevice_init(&dev->dev, &kvm_io_gic_ops);
659 
660 	return SZ_64K;
661 }
662 
663 /**
664  * vgic_register_redist_iodev - register a single redist iodev
665  * @vcpu:    The VCPU to which the redistributor belongs
666  *
667  * Register a KVM iodev for this VCPU's redistributor using the address
668  * provided.
669  *
670  * Return 0 on success, -ERRNO otherwise.
671  */
672 int vgic_register_redist_iodev(struct kvm_vcpu *vcpu)
673 {
674 	struct kvm *kvm = vcpu->kvm;
675 	struct vgic_dist *vgic = &kvm->arch.vgic;
676 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
677 	struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev;
678 	struct vgic_redist_region *rdreg;
679 	gpa_t rd_base;
680 	int ret;
681 
682 	if (!IS_VGIC_ADDR_UNDEF(vgic_cpu->rd_iodev.base_addr))
683 		return 0;
684 
685 	/*
686 	 * We may be creating VCPUs before having set the base address for the
687 	 * redistributor region, in which case we will come back to this
688 	 * function for all VCPUs when the base address is set.  Just return
689 	 * without doing any work for now.
690 	 */
691 	rdreg = vgic_v3_rdist_free_slot(&vgic->rd_regions);
692 	if (!rdreg)
693 		return 0;
694 
695 	if (!vgic_v3_check_base(kvm))
696 		return -EINVAL;
697 
698 	vgic_cpu->rdreg = rdreg;
699 
700 	rd_base = rdreg->base + rdreg->free_index * KVM_VGIC_V3_REDIST_SIZE;
701 
702 	kvm_iodevice_init(&rd_dev->dev, &kvm_io_gic_ops);
703 	rd_dev->base_addr = rd_base;
704 	rd_dev->iodev_type = IODEV_REDIST;
705 	rd_dev->regions = vgic_v3_rd_registers;
706 	rd_dev->nr_regions = ARRAY_SIZE(vgic_v3_rd_registers);
707 	rd_dev->redist_vcpu = vcpu;
708 
709 	mutex_lock(&kvm->slots_lock);
710 	ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, rd_base,
711 				      2 * SZ_64K, &rd_dev->dev);
712 	mutex_unlock(&kvm->slots_lock);
713 
714 	if (ret)
715 		return ret;
716 
717 	rdreg->free_index++;
718 	return 0;
719 }
720 
721 static void vgic_unregister_redist_iodev(struct kvm_vcpu *vcpu)
722 {
723 	struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev;
724 
725 	kvm_io_bus_unregister_dev(vcpu->kvm, KVM_MMIO_BUS, &rd_dev->dev);
726 }
727 
728 static int vgic_register_all_redist_iodevs(struct kvm *kvm)
729 {
730 	struct kvm_vcpu *vcpu;
731 	int c, ret = 0;
732 
733 	kvm_for_each_vcpu(c, vcpu, kvm) {
734 		ret = vgic_register_redist_iodev(vcpu);
735 		if (ret)
736 			break;
737 	}
738 
739 	if (ret) {
740 		/* The current c failed, so we start with the previous one. */
741 		mutex_lock(&kvm->slots_lock);
742 		for (c--; c >= 0; c--) {
743 			vcpu = kvm_get_vcpu(kvm, c);
744 			vgic_unregister_redist_iodev(vcpu);
745 		}
746 		mutex_unlock(&kvm->slots_lock);
747 	}
748 
749 	return ret;
750 }
751 
752 /**
753  * vgic_v3_insert_redist_region - Insert a new redistributor region
754  *
755  * Performs various checks before inserting the rdist region in the list.
756  * Those tests depend on whether the size of the rdist region is known
757  * (ie. count != 0). The list is sorted by rdist region index.
758  *
759  * @kvm: kvm handle
760  * @index: redist region index
761  * @base: base of the new rdist region
762  * @count: number of redistributors the region is made of (0 in the old style
763  * single region, whose size is induced from the number of vcpus)
764  *
765  * Return 0 on success, < 0 otherwise
766  */
767 static int vgic_v3_insert_redist_region(struct kvm *kvm, uint32_t index,
768 					gpa_t base, uint32_t count)
769 {
770 	struct vgic_dist *d = &kvm->arch.vgic;
771 	struct vgic_redist_region *rdreg;
772 	struct list_head *rd_regions = &d->rd_regions;
773 	size_t size = count * KVM_VGIC_V3_REDIST_SIZE;
774 	int ret;
775 
776 	/* single rdist region already set ?*/
777 	if (!count && !list_empty(rd_regions))
778 		return -EINVAL;
779 
780 	/* cross the end of memory ? */
781 	if (base + size < base)
782 		return -EINVAL;
783 
784 	if (list_empty(rd_regions)) {
785 		if (index != 0)
786 			return -EINVAL;
787 	} else {
788 		rdreg = list_last_entry(rd_regions,
789 					struct vgic_redist_region, list);
790 		if (index != rdreg->index + 1)
791 			return -EINVAL;
792 
793 		/* Cannot add an explicitly sized regions after legacy region */
794 		if (!rdreg->count)
795 			return -EINVAL;
796 	}
797 
798 	/*
799 	 * For legacy single-region redistributor regions (!count),
800 	 * check that the redistributor region does not overlap with the
801 	 * distributor's address space.
802 	 */
803 	if (!count && !IS_VGIC_ADDR_UNDEF(d->vgic_dist_base) &&
804 		vgic_dist_overlap(kvm, base, size))
805 		return -EINVAL;
806 
807 	/* collision with any other rdist region? */
808 	if (vgic_v3_rdist_overlap(kvm, base, size))
809 		return -EINVAL;
810 
811 	rdreg = kzalloc(sizeof(*rdreg), GFP_KERNEL);
812 	if (!rdreg)
813 		return -ENOMEM;
814 
815 	rdreg->base = VGIC_ADDR_UNDEF;
816 
817 	ret = vgic_check_ioaddr(kvm, &rdreg->base, base, SZ_64K);
818 	if (ret)
819 		goto free;
820 
821 	rdreg->base = base;
822 	rdreg->count = count;
823 	rdreg->free_index = 0;
824 	rdreg->index = index;
825 
826 	list_add_tail(&rdreg->list, rd_regions);
827 	return 0;
828 free:
829 	kfree(rdreg);
830 	return ret;
831 }
832 
833 int vgic_v3_set_redist_base(struct kvm *kvm, u32 index, u64 addr, u32 count)
834 {
835 	int ret;
836 
837 	ret = vgic_v3_insert_redist_region(kvm, index, addr, count);
838 	if (ret)
839 		return ret;
840 
841 	/*
842 	 * Register iodevs for each existing VCPU.  Adding more VCPUs
843 	 * afterwards will register the iodevs when needed.
844 	 */
845 	ret = vgic_register_all_redist_iodevs(kvm);
846 	if (ret)
847 		return ret;
848 
849 	return 0;
850 }
851 
852 int vgic_v3_has_attr_regs(struct kvm_device *dev, struct kvm_device_attr *attr)
853 {
854 	const struct vgic_register_region *region;
855 	struct vgic_io_device iodev;
856 	struct vgic_reg_attr reg_attr;
857 	struct kvm_vcpu *vcpu;
858 	gpa_t addr;
859 	int ret;
860 
861 	ret = vgic_v3_parse_attr(dev, attr, &reg_attr);
862 	if (ret)
863 		return ret;
864 
865 	vcpu = reg_attr.vcpu;
866 	addr = reg_attr.addr;
867 
868 	switch (attr->group) {
869 	case KVM_DEV_ARM_VGIC_GRP_DIST_REGS:
870 		iodev.regions = vgic_v3_dist_registers;
871 		iodev.nr_regions = ARRAY_SIZE(vgic_v3_dist_registers);
872 		iodev.base_addr = 0;
873 		break;
874 	case KVM_DEV_ARM_VGIC_GRP_REDIST_REGS:{
875 		iodev.regions = vgic_v3_rd_registers;
876 		iodev.nr_regions = ARRAY_SIZE(vgic_v3_rd_registers);
877 		iodev.base_addr = 0;
878 		break;
879 	}
880 	case KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS: {
881 		u64 reg, id;
882 
883 		id = (attr->attr & KVM_DEV_ARM_VGIC_SYSREG_INSTR_MASK);
884 		return vgic_v3_has_cpu_sysregs_attr(vcpu, 0, id, &reg);
885 	}
886 	default:
887 		return -ENXIO;
888 	}
889 
890 	/* We only support aligned 32-bit accesses. */
891 	if (addr & 3)
892 		return -ENXIO;
893 
894 	region = vgic_get_mmio_region(vcpu, &iodev, addr, sizeof(u32));
895 	if (!region)
896 		return -ENXIO;
897 
898 	return 0;
899 }
900 /*
901  * Compare a given affinity (level 1-3 and a level 0 mask, from the SGI
902  * generation register ICC_SGI1R_EL1) with a given VCPU.
903  * If the VCPU's MPIDR matches, return the level0 affinity, otherwise
904  * return -1.
905  */
906 static int match_mpidr(u64 sgi_aff, u16 sgi_cpu_mask, struct kvm_vcpu *vcpu)
907 {
908 	unsigned long affinity;
909 	int level0;
910 
911 	/*
912 	 * Split the current VCPU's MPIDR into affinity level 0 and the
913 	 * rest as this is what we have to compare against.
914 	 */
915 	affinity = kvm_vcpu_get_mpidr_aff(vcpu);
916 	level0 = MPIDR_AFFINITY_LEVEL(affinity, 0);
917 	affinity &= ~MPIDR_LEVEL_MASK;
918 
919 	/* bail out if the upper three levels don't match */
920 	if (sgi_aff != affinity)
921 		return -1;
922 
923 	/* Is this VCPU's bit set in the mask ? */
924 	if (!(sgi_cpu_mask & BIT(level0)))
925 		return -1;
926 
927 	return level0;
928 }
929 
930 /*
931  * The ICC_SGI* registers encode the affinity differently from the MPIDR,
932  * so provide a wrapper to use the existing defines to isolate a certain
933  * affinity level.
934  */
935 #define SGI_AFFINITY_LEVEL(reg, level) \
936 	((((reg) & ICC_SGI1R_AFFINITY_## level ##_MASK) \
937 	>> ICC_SGI1R_AFFINITY_## level ##_SHIFT) << MPIDR_LEVEL_SHIFT(level))
938 
939 /**
940  * vgic_v3_dispatch_sgi - handle SGI requests from VCPUs
941  * @vcpu: The VCPU requesting a SGI
942  * @reg: The value written into ICC_{ASGI1,SGI0,SGI1}R by that VCPU
943  * @allow_group1: Does the sysreg access allow generation of G1 SGIs
944  *
945  * With GICv3 (and ARE=1) CPUs trigger SGIs by writing to a system register.
946  * This will trap in sys_regs.c and call this function.
947  * This ICC_SGI1R_EL1 register contains the upper three affinity levels of the
948  * target processors as well as a bitmask of 16 Aff0 CPUs.
949  * If the interrupt routing mode bit is not set, we iterate over all VCPUs to
950  * check for matching ones. If this bit is set, we signal all, but not the
951  * calling VCPU.
952  */
953 void vgic_v3_dispatch_sgi(struct kvm_vcpu *vcpu, u64 reg, bool allow_group1)
954 {
955 	struct kvm *kvm = vcpu->kvm;
956 	struct kvm_vcpu *c_vcpu;
957 	u16 target_cpus;
958 	u64 mpidr;
959 	int sgi, c;
960 	int vcpu_id = vcpu->vcpu_id;
961 	bool broadcast;
962 	unsigned long flags;
963 
964 	sgi = (reg & ICC_SGI1R_SGI_ID_MASK) >> ICC_SGI1R_SGI_ID_SHIFT;
965 	broadcast = reg & BIT_ULL(ICC_SGI1R_IRQ_ROUTING_MODE_BIT);
966 	target_cpus = (reg & ICC_SGI1R_TARGET_LIST_MASK) >> ICC_SGI1R_TARGET_LIST_SHIFT;
967 	mpidr = SGI_AFFINITY_LEVEL(reg, 3);
968 	mpidr |= SGI_AFFINITY_LEVEL(reg, 2);
969 	mpidr |= SGI_AFFINITY_LEVEL(reg, 1);
970 
971 	/*
972 	 * We iterate over all VCPUs to find the MPIDRs matching the request.
973 	 * If we have handled one CPU, we clear its bit to detect early
974 	 * if we are already finished. This avoids iterating through all
975 	 * VCPUs when most of the times we just signal a single VCPU.
976 	 */
977 	kvm_for_each_vcpu(c, c_vcpu, kvm) {
978 		struct vgic_irq *irq;
979 
980 		/* Exit early if we have dealt with all requested CPUs */
981 		if (!broadcast && target_cpus == 0)
982 			break;
983 
984 		/* Don't signal the calling VCPU */
985 		if (broadcast && c == vcpu_id)
986 			continue;
987 
988 		if (!broadcast) {
989 			int level0;
990 
991 			level0 = match_mpidr(mpidr, target_cpus, c_vcpu);
992 			if (level0 == -1)
993 				continue;
994 
995 			/* remove this matching VCPU from the mask */
996 			target_cpus &= ~BIT(level0);
997 		}
998 
999 		irq = vgic_get_irq(vcpu->kvm, c_vcpu, sgi);
1000 
1001 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
1002 
1003 		/*
1004 		 * An access targetting Group0 SGIs can only generate
1005 		 * those, while an access targetting Group1 SGIs can
1006 		 * generate interrupts of either group.
1007 		 */
1008 		if (!irq->group || allow_group1) {
1009 			if (!irq->hw) {
1010 				irq->pending_latch = true;
1011 				vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
1012 			} else {
1013 				/* HW SGI? Ask the GIC to inject it */
1014 				int err;
1015 				err = irq_set_irqchip_state(irq->host_irq,
1016 							    IRQCHIP_STATE_PENDING,
1017 							    true);
1018 				WARN_RATELIMIT(err, "IRQ %d", irq->host_irq);
1019 				raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
1020 			}
1021 		} else {
1022 			raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
1023 		}
1024 
1025 		vgic_put_irq(vcpu->kvm, irq);
1026 	}
1027 }
1028 
1029 int vgic_v3_dist_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1030 			 int offset, u32 *val)
1031 {
1032 	struct vgic_io_device dev = {
1033 		.regions = vgic_v3_dist_registers,
1034 		.nr_regions = ARRAY_SIZE(vgic_v3_dist_registers),
1035 	};
1036 
1037 	return vgic_uaccess(vcpu, &dev, is_write, offset, val);
1038 }
1039 
1040 int vgic_v3_redist_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1041 			   int offset, u32 *val)
1042 {
1043 	struct vgic_io_device rd_dev = {
1044 		.regions = vgic_v3_rd_registers,
1045 		.nr_regions = ARRAY_SIZE(vgic_v3_rd_registers),
1046 	};
1047 
1048 	return vgic_uaccess(vcpu, &rd_dev, is_write, offset, val);
1049 }
1050 
1051 int vgic_v3_line_level_info_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1052 				    u32 intid, u64 *val)
1053 {
1054 	if (intid % 32)
1055 		return -EINVAL;
1056 
1057 	if (is_write)
1058 		vgic_write_irq_line_level_info(vcpu, intid, *val);
1059 	else
1060 		*val = vgic_read_irq_line_level_info(vcpu, intid);
1061 
1062 	return 0;
1063 }
1064