xref: /openbmc/linux/arch/arm64/kvm/vgic/vgic-mmio-v3.c (revision 35267cea)
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 && gic_cpuif_has_vsgi())
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 || !gic_cpuif_has_vsgi())
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 bool vgic_mmio_vcpu_rdist_is_last(struct kvm_vcpu *vcpu)
255 {
256 	struct vgic_dist *vgic = &vcpu->kvm->arch.vgic;
257 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
258 	struct vgic_redist_region *iter, *rdreg = vgic_cpu->rdreg;
259 
260 	if (!rdreg)
261 		return false;
262 
263 	if (vgic_cpu->rdreg_index < rdreg->free_index - 1) {
264 		return false;
265 	} else if (rdreg->count && vgic_cpu->rdreg_index == (rdreg->count - 1)) {
266 		struct list_head *rd_regions = &vgic->rd_regions;
267 		gpa_t end = rdreg->base + rdreg->count * KVM_VGIC_V3_REDIST_SIZE;
268 
269 		/*
270 		 * the rdist is the last one of the redist region,
271 		 * check whether there is no other contiguous rdist region
272 		 */
273 		list_for_each_entry(iter, rd_regions, list) {
274 			if (iter->base == end && iter->free_index > 0)
275 				return false;
276 		}
277 	}
278 	return true;
279 }
280 
281 static unsigned long vgic_mmio_read_v3r_typer(struct kvm_vcpu *vcpu,
282 					      gpa_t addr, unsigned int len)
283 {
284 	unsigned long mpidr = kvm_vcpu_get_mpidr_aff(vcpu);
285 	int target_vcpu_id = vcpu->vcpu_id;
286 	u64 value;
287 
288 	value = (u64)(mpidr & GENMASK(23, 0)) << 32;
289 	value |= ((target_vcpu_id & 0xffff) << 8);
290 
291 	if (vgic_has_its(vcpu->kvm))
292 		value |= GICR_TYPER_PLPIS;
293 
294 	if (vgic_mmio_vcpu_rdist_is_last(vcpu))
295 		value |= GICR_TYPER_LAST;
296 
297 	return extract_bytes(value, addr & 7, len);
298 }
299 
300 static unsigned long vgic_mmio_read_v3r_iidr(struct kvm_vcpu *vcpu,
301 					     gpa_t addr, unsigned int len)
302 {
303 	return (PRODUCT_ID_KVM << 24) | (IMPLEMENTER_ARM << 0);
304 }
305 
306 static unsigned long vgic_mmio_read_v3_idregs(struct kvm_vcpu *vcpu,
307 					      gpa_t addr, unsigned int len)
308 {
309 	switch (addr & 0xffff) {
310 	case GICD_PIDR2:
311 		/* report a GICv3 compliant implementation */
312 		return 0x3b;
313 	}
314 
315 	return 0;
316 }
317 
318 static unsigned long vgic_v3_uaccess_read_pending(struct kvm_vcpu *vcpu,
319 						  gpa_t addr, unsigned int len)
320 {
321 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
322 	u32 value = 0;
323 	int i;
324 
325 	/*
326 	 * pending state of interrupt is latched in pending_latch variable.
327 	 * Userspace will save and restore pending state and line_level
328 	 * separately.
329 	 * Refer to Documentation/virt/kvm/devices/arm-vgic-v3.rst
330 	 * for handling of ISPENDR and ICPENDR.
331 	 */
332 	for (i = 0; i < len * 8; i++) {
333 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
334 		bool state = irq->pending_latch;
335 
336 		if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
337 			int err;
338 
339 			err = irq_get_irqchip_state(irq->host_irq,
340 						    IRQCHIP_STATE_PENDING,
341 						    &state);
342 			WARN_ON(err);
343 		}
344 
345 		if (state)
346 			value |= (1U << i);
347 
348 		vgic_put_irq(vcpu->kvm, irq);
349 	}
350 
351 	return value;
352 }
353 
354 static int vgic_v3_uaccess_write_pending(struct kvm_vcpu *vcpu,
355 					 gpa_t addr, unsigned int len,
356 					 unsigned long val)
357 {
358 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
359 	int i;
360 	unsigned long flags;
361 
362 	for (i = 0; i < len * 8; i++) {
363 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
364 
365 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
366 		if (test_bit(i, &val)) {
367 			/*
368 			 * pending_latch is set irrespective of irq type
369 			 * (level or edge) to avoid dependency that VM should
370 			 * restore irq config before pending info.
371 			 */
372 			irq->pending_latch = true;
373 			vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
374 		} else {
375 			irq->pending_latch = false;
376 			raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
377 		}
378 
379 		vgic_put_irq(vcpu->kvm, irq);
380 	}
381 
382 	return 0;
383 }
384 
385 /* We want to avoid outer shareable. */
386 u64 vgic_sanitise_shareability(u64 field)
387 {
388 	switch (field) {
389 	case GIC_BASER_OuterShareable:
390 		return GIC_BASER_InnerShareable;
391 	default:
392 		return field;
393 	}
394 }
395 
396 /* Avoid any inner non-cacheable mapping. */
397 u64 vgic_sanitise_inner_cacheability(u64 field)
398 {
399 	switch (field) {
400 	case GIC_BASER_CACHE_nCnB:
401 	case GIC_BASER_CACHE_nC:
402 		return GIC_BASER_CACHE_RaWb;
403 	default:
404 		return field;
405 	}
406 }
407 
408 /* Non-cacheable or same-as-inner are OK. */
409 u64 vgic_sanitise_outer_cacheability(u64 field)
410 {
411 	switch (field) {
412 	case GIC_BASER_CACHE_SameAsInner:
413 	case GIC_BASER_CACHE_nC:
414 		return field;
415 	default:
416 		return GIC_BASER_CACHE_SameAsInner;
417 	}
418 }
419 
420 u64 vgic_sanitise_field(u64 reg, u64 field_mask, int field_shift,
421 			u64 (*sanitise_fn)(u64))
422 {
423 	u64 field = (reg & field_mask) >> field_shift;
424 
425 	field = sanitise_fn(field) << field_shift;
426 	return (reg & ~field_mask) | field;
427 }
428 
429 #define PROPBASER_RES0_MASK						\
430 	(GENMASK_ULL(63, 59) | GENMASK_ULL(55, 52) | GENMASK_ULL(6, 5))
431 #define PENDBASER_RES0_MASK						\
432 	(BIT_ULL(63) | GENMASK_ULL(61, 59) | GENMASK_ULL(55, 52) |	\
433 	 GENMASK_ULL(15, 12) | GENMASK_ULL(6, 0))
434 
435 static u64 vgic_sanitise_pendbaser(u64 reg)
436 {
437 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_SHAREABILITY_MASK,
438 				  GICR_PENDBASER_SHAREABILITY_SHIFT,
439 				  vgic_sanitise_shareability);
440 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_INNER_CACHEABILITY_MASK,
441 				  GICR_PENDBASER_INNER_CACHEABILITY_SHIFT,
442 				  vgic_sanitise_inner_cacheability);
443 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_OUTER_CACHEABILITY_MASK,
444 				  GICR_PENDBASER_OUTER_CACHEABILITY_SHIFT,
445 				  vgic_sanitise_outer_cacheability);
446 
447 	reg &= ~PENDBASER_RES0_MASK;
448 
449 	return reg;
450 }
451 
452 static u64 vgic_sanitise_propbaser(u64 reg)
453 {
454 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_SHAREABILITY_MASK,
455 				  GICR_PROPBASER_SHAREABILITY_SHIFT,
456 				  vgic_sanitise_shareability);
457 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_INNER_CACHEABILITY_MASK,
458 				  GICR_PROPBASER_INNER_CACHEABILITY_SHIFT,
459 				  vgic_sanitise_inner_cacheability);
460 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_OUTER_CACHEABILITY_MASK,
461 				  GICR_PROPBASER_OUTER_CACHEABILITY_SHIFT,
462 				  vgic_sanitise_outer_cacheability);
463 
464 	reg &= ~PROPBASER_RES0_MASK;
465 	return reg;
466 }
467 
468 static unsigned long vgic_mmio_read_propbase(struct kvm_vcpu *vcpu,
469 					     gpa_t addr, unsigned int len)
470 {
471 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
472 
473 	return extract_bytes(dist->propbaser, addr & 7, len);
474 }
475 
476 static void vgic_mmio_write_propbase(struct kvm_vcpu *vcpu,
477 				     gpa_t addr, unsigned int len,
478 				     unsigned long val)
479 {
480 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
481 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
482 	u64 old_propbaser, propbaser;
483 
484 	/* Storing a value with LPIs already enabled is undefined */
485 	if (vgic_cpu->lpis_enabled)
486 		return;
487 
488 	do {
489 		old_propbaser = READ_ONCE(dist->propbaser);
490 		propbaser = old_propbaser;
491 		propbaser = update_64bit_reg(propbaser, addr & 4, len, val);
492 		propbaser = vgic_sanitise_propbaser(propbaser);
493 	} while (cmpxchg64(&dist->propbaser, old_propbaser,
494 			   propbaser) != old_propbaser);
495 }
496 
497 static unsigned long vgic_mmio_read_pendbase(struct kvm_vcpu *vcpu,
498 					     gpa_t addr, unsigned int len)
499 {
500 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
501 	u64 value = vgic_cpu->pendbaser;
502 
503 	value &= ~GICR_PENDBASER_PTZ;
504 
505 	return extract_bytes(value, addr & 7, len);
506 }
507 
508 static void vgic_mmio_write_pendbase(struct kvm_vcpu *vcpu,
509 				     gpa_t addr, unsigned int len,
510 				     unsigned long val)
511 {
512 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
513 	u64 old_pendbaser, pendbaser;
514 
515 	/* Storing a value with LPIs already enabled is undefined */
516 	if (vgic_cpu->lpis_enabled)
517 		return;
518 
519 	do {
520 		old_pendbaser = READ_ONCE(vgic_cpu->pendbaser);
521 		pendbaser = old_pendbaser;
522 		pendbaser = update_64bit_reg(pendbaser, addr & 4, len, val);
523 		pendbaser = vgic_sanitise_pendbaser(pendbaser);
524 	} while (cmpxchg64(&vgic_cpu->pendbaser, old_pendbaser,
525 			   pendbaser) != old_pendbaser);
526 }
527 
528 /*
529  * The GICv3 per-IRQ registers are split to control PPIs and SGIs in the
530  * redistributors, while SPIs are covered by registers in the distributor
531  * block. Trying to set private IRQs in this block gets ignored.
532  * We take some special care here to fix the calculation of the register
533  * offset.
534  */
535 #define REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(off, rd, wr, ur, uw, bpi, acc) \
536 	{								\
537 		.reg_offset = off,					\
538 		.bits_per_irq = bpi,					\
539 		.len = (bpi * VGIC_NR_PRIVATE_IRQS) / 8,		\
540 		.access_flags = acc,					\
541 		.read = vgic_mmio_read_raz,				\
542 		.write = vgic_mmio_write_wi,				\
543 	}, {								\
544 		.reg_offset = off + (bpi * VGIC_NR_PRIVATE_IRQS) / 8,	\
545 		.bits_per_irq = bpi,					\
546 		.len = (bpi * (1024 - VGIC_NR_PRIVATE_IRQS)) / 8,	\
547 		.access_flags = acc,					\
548 		.read = rd,						\
549 		.write = wr,						\
550 		.uaccess_read = ur,					\
551 		.uaccess_write = uw,					\
552 	}
553 
554 static const struct vgic_register_region vgic_v3_dist_registers[] = {
555 	REGISTER_DESC_WITH_LENGTH_UACCESS(GICD_CTLR,
556 		vgic_mmio_read_v3_misc, vgic_mmio_write_v3_misc,
557 		NULL, vgic_mmio_uaccess_write_v3_misc,
558 		16, VGIC_ACCESS_32bit),
559 	REGISTER_DESC_WITH_LENGTH(GICD_STATUSR,
560 		vgic_mmio_read_rao, vgic_mmio_write_wi, 4,
561 		VGIC_ACCESS_32bit),
562 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGROUPR,
563 		vgic_mmio_read_group, vgic_mmio_write_group, NULL, NULL, 1,
564 		VGIC_ACCESS_32bit),
565 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISENABLER,
566 		vgic_mmio_read_enable, vgic_mmio_write_senable,
567 		NULL, vgic_uaccess_write_senable, 1,
568 		VGIC_ACCESS_32bit),
569 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICENABLER,
570 		vgic_mmio_read_enable, vgic_mmio_write_cenable,
571 	       NULL, vgic_uaccess_write_cenable, 1,
572 		VGIC_ACCESS_32bit),
573 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISPENDR,
574 		vgic_mmio_read_pending, vgic_mmio_write_spending,
575 		vgic_v3_uaccess_read_pending, vgic_v3_uaccess_write_pending, 1,
576 		VGIC_ACCESS_32bit),
577 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICPENDR,
578 		vgic_mmio_read_pending, vgic_mmio_write_cpending,
579 		vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 1,
580 		VGIC_ACCESS_32bit),
581 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISACTIVER,
582 		vgic_mmio_read_active, vgic_mmio_write_sactive,
583 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 1,
584 		VGIC_ACCESS_32bit),
585 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICACTIVER,
586 		vgic_mmio_read_active, vgic_mmio_write_cactive,
587 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive,
588 		1, VGIC_ACCESS_32bit),
589 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IPRIORITYR,
590 		vgic_mmio_read_priority, vgic_mmio_write_priority, NULL, NULL,
591 		8, VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
592 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ITARGETSR,
593 		vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 8,
594 		VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
595 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICFGR,
596 		vgic_mmio_read_config, vgic_mmio_write_config, NULL, NULL, 2,
597 		VGIC_ACCESS_32bit),
598 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGRPMODR,
599 		vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 1,
600 		VGIC_ACCESS_32bit),
601 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IROUTER,
602 		vgic_mmio_read_irouter, vgic_mmio_write_irouter, NULL, NULL, 64,
603 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
604 	REGISTER_DESC_WITH_LENGTH(GICD_IDREGS,
605 		vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48,
606 		VGIC_ACCESS_32bit),
607 };
608 
609 static const struct vgic_register_region vgic_v3_rd_registers[] = {
610 	/* RD_base registers */
611 	REGISTER_DESC_WITH_LENGTH(GICR_CTLR,
612 		vgic_mmio_read_v3r_ctlr, vgic_mmio_write_v3r_ctlr, 4,
613 		VGIC_ACCESS_32bit),
614 	REGISTER_DESC_WITH_LENGTH(GICR_STATUSR,
615 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
616 		VGIC_ACCESS_32bit),
617 	REGISTER_DESC_WITH_LENGTH(GICR_IIDR,
618 		vgic_mmio_read_v3r_iidr, vgic_mmio_write_wi, 4,
619 		VGIC_ACCESS_32bit),
620 	REGISTER_DESC_WITH_LENGTH_UACCESS(GICR_TYPER,
621 		vgic_mmio_read_v3r_typer, vgic_mmio_write_wi,
622 		NULL, vgic_mmio_uaccess_write_wi, 8,
623 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
624 	REGISTER_DESC_WITH_LENGTH(GICR_WAKER,
625 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
626 		VGIC_ACCESS_32bit),
627 	REGISTER_DESC_WITH_LENGTH(GICR_PROPBASER,
628 		vgic_mmio_read_propbase, vgic_mmio_write_propbase, 8,
629 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
630 	REGISTER_DESC_WITH_LENGTH(GICR_PENDBASER,
631 		vgic_mmio_read_pendbase, vgic_mmio_write_pendbase, 8,
632 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
633 	REGISTER_DESC_WITH_LENGTH(GICR_IDREGS,
634 		vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48,
635 		VGIC_ACCESS_32bit),
636 	/* SGI_base registers */
637 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGROUPR0,
638 		vgic_mmio_read_group, vgic_mmio_write_group, 4,
639 		VGIC_ACCESS_32bit),
640 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISENABLER0,
641 		vgic_mmio_read_enable, vgic_mmio_write_senable,
642 		NULL, vgic_uaccess_write_senable, 4,
643 		VGIC_ACCESS_32bit),
644 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICENABLER0,
645 		vgic_mmio_read_enable, vgic_mmio_write_cenable,
646 		NULL, vgic_uaccess_write_cenable, 4,
647 		VGIC_ACCESS_32bit),
648 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISPENDR0,
649 		vgic_mmio_read_pending, vgic_mmio_write_spending,
650 		vgic_v3_uaccess_read_pending, vgic_v3_uaccess_write_pending, 4,
651 		VGIC_ACCESS_32bit),
652 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICPENDR0,
653 		vgic_mmio_read_pending, vgic_mmio_write_cpending,
654 		vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 4,
655 		VGIC_ACCESS_32bit),
656 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISACTIVER0,
657 		vgic_mmio_read_active, vgic_mmio_write_sactive,
658 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 4,
659 		VGIC_ACCESS_32bit),
660 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICACTIVER0,
661 		vgic_mmio_read_active, vgic_mmio_write_cactive,
662 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive, 4,
663 		VGIC_ACCESS_32bit),
664 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IPRIORITYR0,
665 		vgic_mmio_read_priority, vgic_mmio_write_priority, 32,
666 		VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
667 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_ICFGR0,
668 		vgic_mmio_read_config, vgic_mmio_write_config, 8,
669 		VGIC_ACCESS_32bit),
670 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGRPMODR0,
671 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
672 		VGIC_ACCESS_32bit),
673 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_NSACR,
674 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
675 		VGIC_ACCESS_32bit),
676 };
677 
678 unsigned int vgic_v3_init_dist_iodev(struct vgic_io_device *dev)
679 {
680 	dev->regions = vgic_v3_dist_registers;
681 	dev->nr_regions = ARRAY_SIZE(vgic_v3_dist_registers);
682 
683 	kvm_iodevice_init(&dev->dev, &kvm_io_gic_ops);
684 
685 	return SZ_64K;
686 }
687 
688 /**
689  * vgic_register_redist_iodev - register a single redist iodev
690  * @vcpu:    The VCPU to which the redistributor belongs
691  *
692  * Register a KVM iodev for this VCPU's redistributor using the address
693  * provided.
694  *
695  * Return 0 on success, -ERRNO otherwise.
696  */
697 int vgic_register_redist_iodev(struct kvm_vcpu *vcpu)
698 {
699 	struct kvm *kvm = vcpu->kvm;
700 	struct vgic_dist *vgic = &kvm->arch.vgic;
701 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
702 	struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev;
703 	struct vgic_redist_region *rdreg;
704 	gpa_t rd_base;
705 	int ret;
706 
707 	if (!IS_VGIC_ADDR_UNDEF(vgic_cpu->rd_iodev.base_addr))
708 		return 0;
709 
710 	/*
711 	 * We may be creating VCPUs before having set the base address for the
712 	 * redistributor region, in which case we will come back to this
713 	 * function for all VCPUs when the base address is set.  Just return
714 	 * without doing any work for now.
715 	 */
716 	rdreg = vgic_v3_rdist_free_slot(&vgic->rd_regions);
717 	if (!rdreg)
718 		return 0;
719 
720 	if (!vgic_v3_check_base(kvm))
721 		return -EINVAL;
722 
723 	vgic_cpu->rdreg = rdreg;
724 	vgic_cpu->rdreg_index = rdreg->free_index;
725 
726 	rd_base = rdreg->base + rdreg->free_index * KVM_VGIC_V3_REDIST_SIZE;
727 
728 	kvm_iodevice_init(&rd_dev->dev, &kvm_io_gic_ops);
729 	rd_dev->base_addr = rd_base;
730 	rd_dev->iodev_type = IODEV_REDIST;
731 	rd_dev->regions = vgic_v3_rd_registers;
732 	rd_dev->nr_regions = ARRAY_SIZE(vgic_v3_rd_registers);
733 	rd_dev->redist_vcpu = vcpu;
734 
735 	mutex_lock(&kvm->slots_lock);
736 	ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, rd_base,
737 				      2 * SZ_64K, &rd_dev->dev);
738 	mutex_unlock(&kvm->slots_lock);
739 
740 	if (ret)
741 		return ret;
742 
743 	rdreg->free_index++;
744 	return 0;
745 }
746 
747 static void vgic_unregister_redist_iodev(struct kvm_vcpu *vcpu)
748 {
749 	struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev;
750 
751 	kvm_io_bus_unregister_dev(vcpu->kvm, KVM_MMIO_BUS, &rd_dev->dev);
752 }
753 
754 static int vgic_register_all_redist_iodevs(struct kvm *kvm)
755 {
756 	struct kvm_vcpu *vcpu;
757 	int c, ret = 0;
758 
759 	kvm_for_each_vcpu(c, vcpu, kvm) {
760 		ret = vgic_register_redist_iodev(vcpu);
761 		if (ret)
762 			break;
763 	}
764 
765 	if (ret) {
766 		/* The current c failed, so we start with the previous one. */
767 		mutex_lock(&kvm->slots_lock);
768 		for (c--; c >= 0; c--) {
769 			vcpu = kvm_get_vcpu(kvm, c);
770 			vgic_unregister_redist_iodev(vcpu);
771 		}
772 		mutex_unlock(&kvm->slots_lock);
773 	}
774 
775 	return ret;
776 }
777 
778 /**
779  * vgic_v3_alloc_redist_region - Allocate a new redistributor region
780  *
781  * Performs various checks before inserting the rdist region in the list.
782  * Those tests depend on whether the size of the rdist region is known
783  * (ie. count != 0). The list is sorted by rdist region index.
784  *
785  * @kvm: kvm handle
786  * @index: redist region index
787  * @base: base of the new rdist region
788  * @count: number of redistributors the region is made of (0 in the old style
789  * single region, whose size is induced from the number of vcpus)
790  *
791  * Return 0 on success, < 0 otherwise
792  */
793 static int vgic_v3_alloc_redist_region(struct kvm *kvm, uint32_t index,
794 				       gpa_t base, uint32_t count)
795 {
796 	struct vgic_dist *d = &kvm->arch.vgic;
797 	struct vgic_redist_region *rdreg;
798 	struct list_head *rd_regions = &d->rd_regions;
799 	size_t size = count * KVM_VGIC_V3_REDIST_SIZE;
800 	int ret;
801 
802 	/* cross the end of memory ? */
803 	if (base + size < base)
804 		return -EINVAL;
805 
806 	if (list_empty(rd_regions)) {
807 		if (index != 0)
808 			return -EINVAL;
809 	} else {
810 		rdreg = list_last_entry(rd_regions,
811 					struct vgic_redist_region, list);
812 
813 		/* Don't mix single region and discrete redist regions */
814 		if (!count && rdreg->count)
815 			return -EINVAL;
816 
817 		if (!count)
818 			return -EEXIST;
819 
820 		if (index != rdreg->index + 1)
821 			return -EINVAL;
822 	}
823 
824 	/*
825 	 * For legacy single-region redistributor regions (!count),
826 	 * check that the redistributor region does not overlap with the
827 	 * distributor's address space.
828 	 */
829 	if (!count && !IS_VGIC_ADDR_UNDEF(d->vgic_dist_base) &&
830 		vgic_dist_overlap(kvm, base, size))
831 		return -EINVAL;
832 
833 	/* collision with any other rdist region? */
834 	if (vgic_v3_rdist_overlap(kvm, base, size))
835 		return -EINVAL;
836 
837 	rdreg = kzalloc(sizeof(*rdreg), GFP_KERNEL);
838 	if (!rdreg)
839 		return -ENOMEM;
840 
841 	rdreg->base = VGIC_ADDR_UNDEF;
842 
843 	ret = vgic_check_ioaddr(kvm, &rdreg->base, base, SZ_64K);
844 	if (ret)
845 		goto free;
846 
847 	rdreg->base = base;
848 	rdreg->count = count;
849 	rdreg->free_index = 0;
850 	rdreg->index = index;
851 
852 	list_add_tail(&rdreg->list, rd_regions);
853 	return 0;
854 free:
855 	kfree(rdreg);
856 	return ret;
857 }
858 
859 void vgic_v3_free_redist_region(struct vgic_redist_region *rdreg)
860 {
861 	list_del(&rdreg->list);
862 	kfree(rdreg);
863 }
864 
865 int vgic_v3_set_redist_base(struct kvm *kvm, u32 index, u64 addr, u32 count)
866 {
867 	int ret;
868 
869 	ret = vgic_v3_alloc_redist_region(kvm, index, addr, count);
870 	if (ret)
871 		return ret;
872 
873 	/*
874 	 * Register iodevs for each existing VCPU.  Adding more VCPUs
875 	 * afterwards will register the iodevs when needed.
876 	 */
877 	ret = vgic_register_all_redist_iodevs(kvm);
878 	if (ret) {
879 		struct vgic_redist_region *rdreg;
880 
881 		rdreg = vgic_v3_rdist_region_from_index(kvm, index);
882 		vgic_v3_free_redist_region(rdreg);
883 		return ret;
884 	}
885 
886 	return 0;
887 }
888 
889 int vgic_v3_has_attr_regs(struct kvm_device *dev, struct kvm_device_attr *attr)
890 {
891 	const struct vgic_register_region *region;
892 	struct vgic_io_device iodev;
893 	struct vgic_reg_attr reg_attr;
894 	struct kvm_vcpu *vcpu;
895 	gpa_t addr;
896 	int ret;
897 
898 	ret = vgic_v3_parse_attr(dev, attr, &reg_attr);
899 	if (ret)
900 		return ret;
901 
902 	vcpu = reg_attr.vcpu;
903 	addr = reg_attr.addr;
904 
905 	switch (attr->group) {
906 	case KVM_DEV_ARM_VGIC_GRP_DIST_REGS:
907 		iodev.regions = vgic_v3_dist_registers;
908 		iodev.nr_regions = ARRAY_SIZE(vgic_v3_dist_registers);
909 		iodev.base_addr = 0;
910 		break;
911 	case KVM_DEV_ARM_VGIC_GRP_REDIST_REGS:{
912 		iodev.regions = vgic_v3_rd_registers;
913 		iodev.nr_regions = ARRAY_SIZE(vgic_v3_rd_registers);
914 		iodev.base_addr = 0;
915 		break;
916 	}
917 	case KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS: {
918 		u64 reg, id;
919 
920 		id = (attr->attr & KVM_DEV_ARM_VGIC_SYSREG_INSTR_MASK);
921 		return vgic_v3_has_cpu_sysregs_attr(vcpu, 0, id, &reg);
922 	}
923 	default:
924 		return -ENXIO;
925 	}
926 
927 	/* We only support aligned 32-bit accesses. */
928 	if (addr & 3)
929 		return -ENXIO;
930 
931 	region = vgic_get_mmio_region(vcpu, &iodev, addr, sizeof(u32));
932 	if (!region)
933 		return -ENXIO;
934 
935 	return 0;
936 }
937 /*
938  * Compare a given affinity (level 1-3 and a level 0 mask, from the SGI
939  * generation register ICC_SGI1R_EL1) with a given VCPU.
940  * If the VCPU's MPIDR matches, return the level0 affinity, otherwise
941  * return -1.
942  */
943 static int match_mpidr(u64 sgi_aff, u16 sgi_cpu_mask, struct kvm_vcpu *vcpu)
944 {
945 	unsigned long affinity;
946 	int level0;
947 
948 	/*
949 	 * Split the current VCPU's MPIDR into affinity level 0 and the
950 	 * rest as this is what we have to compare against.
951 	 */
952 	affinity = kvm_vcpu_get_mpidr_aff(vcpu);
953 	level0 = MPIDR_AFFINITY_LEVEL(affinity, 0);
954 	affinity &= ~MPIDR_LEVEL_MASK;
955 
956 	/* bail out if the upper three levels don't match */
957 	if (sgi_aff != affinity)
958 		return -1;
959 
960 	/* Is this VCPU's bit set in the mask ? */
961 	if (!(sgi_cpu_mask & BIT(level0)))
962 		return -1;
963 
964 	return level0;
965 }
966 
967 /*
968  * The ICC_SGI* registers encode the affinity differently from the MPIDR,
969  * so provide a wrapper to use the existing defines to isolate a certain
970  * affinity level.
971  */
972 #define SGI_AFFINITY_LEVEL(reg, level) \
973 	((((reg) & ICC_SGI1R_AFFINITY_## level ##_MASK) \
974 	>> ICC_SGI1R_AFFINITY_## level ##_SHIFT) << MPIDR_LEVEL_SHIFT(level))
975 
976 /**
977  * vgic_v3_dispatch_sgi - handle SGI requests from VCPUs
978  * @vcpu: The VCPU requesting a SGI
979  * @reg: The value written into ICC_{ASGI1,SGI0,SGI1}R by that VCPU
980  * @allow_group1: Does the sysreg access allow generation of G1 SGIs
981  *
982  * With GICv3 (and ARE=1) CPUs trigger SGIs by writing to a system register.
983  * This will trap in sys_regs.c and call this function.
984  * This ICC_SGI1R_EL1 register contains the upper three affinity levels of the
985  * target processors as well as a bitmask of 16 Aff0 CPUs.
986  * If the interrupt routing mode bit is not set, we iterate over all VCPUs to
987  * check for matching ones. If this bit is set, we signal all, but not the
988  * calling VCPU.
989  */
990 void vgic_v3_dispatch_sgi(struct kvm_vcpu *vcpu, u64 reg, bool allow_group1)
991 {
992 	struct kvm *kvm = vcpu->kvm;
993 	struct kvm_vcpu *c_vcpu;
994 	u16 target_cpus;
995 	u64 mpidr;
996 	int sgi, c;
997 	int vcpu_id = vcpu->vcpu_id;
998 	bool broadcast;
999 	unsigned long flags;
1000 
1001 	sgi = (reg & ICC_SGI1R_SGI_ID_MASK) >> ICC_SGI1R_SGI_ID_SHIFT;
1002 	broadcast = reg & BIT_ULL(ICC_SGI1R_IRQ_ROUTING_MODE_BIT);
1003 	target_cpus = (reg & ICC_SGI1R_TARGET_LIST_MASK) >> ICC_SGI1R_TARGET_LIST_SHIFT;
1004 	mpidr = SGI_AFFINITY_LEVEL(reg, 3);
1005 	mpidr |= SGI_AFFINITY_LEVEL(reg, 2);
1006 	mpidr |= SGI_AFFINITY_LEVEL(reg, 1);
1007 
1008 	/*
1009 	 * We iterate over all VCPUs to find the MPIDRs matching the request.
1010 	 * If we have handled one CPU, we clear its bit to detect early
1011 	 * if we are already finished. This avoids iterating through all
1012 	 * VCPUs when most of the times we just signal a single VCPU.
1013 	 */
1014 	kvm_for_each_vcpu(c, c_vcpu, kvm) {
1015 		struct vgic_irq *irq;
1016 
1017 		/* Exit early if we have dealt with all requested CPUs */
1018 		if (!broadcast && target_cpus == 0)
1019 			break;
1020 
1021 		/* Don't signal the calling VCPU */
1022 		if (broadcast && c == vcpu_id)
1023 			continue;
1024 
1025 		if (!broadcast) {
1026 			int level0;
1027 
1028 			level0 = match_mpidr(mpidr, target_cpus, c_vcpu);
1029 			if (level0 == -1)
1030 				continue;
1031 
1032 			/* remove this matching VCPU from the mask */
1033 			target_cpus &= ~BIT(level0);
1034 		}
1035 
1036 		irq = vgic_get_irq(vcpu->kvm, c_vcpu, sgi);
1037 
1038 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
1039 
1040 		/*
1041 		 * An access targeting Group0 SGIs can only generate
1042 		 * those, while an access targeting Group1 SGIs can
1043 		 * generate interrupts of either group.
1044 		 */
1045 		if (!irq->group || allow_group1) {
1046 			if (!irq->hw) {
1047 				irq->pending_latch = true;
1048 				vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
1049 			} else {
1050 				/* HW SGI? Ask the GIC to inject it */
1051 				int err;
1052 				err = irq_set_irqchip_state(irq->host_irq,
1053 							    IRQCHIP_STATE_PENDING,
1054 							    true);
1055 				WARN_RATELIMIT(err, "IRQ %d", irq->host_irq);
1056 				raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
1057 			}
1058 		} else {
1059 			raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
1060 		}
1061 
1062 		vgic_put_irq(vcpu->kvm, irq);
1063 	}
1064 }
1065 
1066 int vgic_v3_dist_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1067 			 int offset, u32 *val)
1068 {
1069 	struct vgic_io_device dev = {
1070 		.regions = vgic_v3_dist_registers,
1071 		.nr_regions = ARRAY_SIZE(vgic_v3_dist_registers),
1072 	};
1073 
1074 	return vgic_uaccess(vcpu, &dev, is_write, offset, val);
1075 }
1076 
1077 int vgic_v3_redist_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1078 			   int offset, u32 *val)
1079 {
1080 	struct vgic_io_device rd_dev = {
1081 		.regions = vgic_v3_rd_registers,
1082 		.nr_regions = ARRAY_SIZE(vgic_v3_rd_registers),
1083 	};
1084 
1085 	return vgic_uaccess(vcpu, &rd_dev, is_write, offset, val);
1086 }
1087 
1088 int vgic_v3_line_level_info_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1089 				    u32 intid, u64 *val)
1090 {
1091 	if (intid % 32)
1092 		return -EINVAL;
1093 
1094 	if (is_write)
1095 		vgic_write_irq_line_level_info(vcpu, intid, *val);
1096 	else
1097 		*val = vgic_read_irq_line_level_info(vcpu, intid);
1098 
1099 	return 0;
1100 }
1101