xref: /openbmc/linux/arch/arm64/kvm/vgic/vgic-mmio-v3.c (revision d9fd5a71)
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_uaccess_read_v3r_typer(struct kvm_vcpu *vcpu,
277 						 gpa_t addr, unsigned int len)
278 {
279 	unsigned long mpidr = kvm_vcpu_get_mpidr_aff(vcpu);
280 	int target_vcpu_id = vcpu->vcpu_id;
281 	u64 value;
282 
283 	value = (u64)(mpidr & GENMASK(23, 0)) << 32;
284 	value |= ((target_vcpu_id & 0xffff) << 8);
285 
286 	if (vgic_has_its(vcpu->kvm))
287 		value |= GICR_TYPER_PLPIS;
288 
289 	/* reporting of the Last bit is not supported for userspace */
290 	return extract_bytes(value, addr & 7, len);
291 }
292 
293 static unsigned long vgic_mmio_read_v3r_iidr(struct kvm_vcpu *vcpu,
294 					     gpa_t addr, unsigned int len)
295 {
296 	return (PRODUCT_ID_KVM << 24) | (IMPLEMENTER_ARM << 0);
297 }
298 
299 static unsigned long vgic_mmio_read_v3_idregs(struct kvm_vcpu *vcpu,
300 					      gpa_t addr, unsigned int len)
301 {
302 	switch (addr & 0xffff) {
303 	case GICD_PIDR2:
304 		/* report a GICv3 compliant implementation */
305 		return 0x3b;
306 	}
307 
308 	return 0;
309 }
310 
311 static unsigned long vgic_v3_uaccess_read_pending(struct kvm_vcpu *vcpu,
312 						  gpa_t addr, unsigned int len)
313 {
314 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
315 	u32 value = 0;
316 	int i;
317 
318 	/*
319 	 * pending state of interrupt is latched in pending_latch variable.
320 	 * Userspace will save and restore pending state and line_level
321 	 * separately.
322 	 * Refer to Documentation/virt/kvm/devices/arm-vgic-v3.rst
323 	 * for handling of ISPENDR and ICPENDR.
324 	 */
325 	for (i = 0; i < len * 8; i++) {
326 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
327 		bool state = irq->pending_latch;
328 
329 		if (irq->hw && vgic_irq_is_sgi(irq->intid)) {
330 			int err;
331 
332 			err = irq_get_irqchip_state(irq->host_irq,
333 						    IRQCHIP_STATE_PENDING,
334 						    &state);
335 			WARN_ON(err);
336 		}
337 
338 		if (state)
339 			value |= (1U << i);
340 
341 		vgic_put_irq(vcpu->kvm, irq);
342 	}
343 
344 	return value;
345 }
346 
347 static int vgic_v3_uaccess_write_pending(struct kvm_vcpu *vcpu,
348 					 gpa_t addr, unsigned int len,
349 					 unsigned long val)
350 {
351 	u32 intid = VGIC_ADDR_TO_INTID(addr, 1);
352 	int i;
353 	unsigned long flags;
354 
355 	for (i = 0; i < len * 8; i++) {
356 		struct vgic_irq *irq = vgic_get_irq(vcpu->kvm, vcpu, intid + i);
357 
358 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
359 		if (test_bit(i, &val)) {
360 			/*
361 			 * pending_latch is set irrespective of irq type
362 			 * (level or edge) to avoid dependency that VM should
363 			 * restore irq config before pending info.
364 			 */
365 			irq->pending_latch = true;
366 			vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
367 		} else {
368 			irq->pending_latch = false;
369 			raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
370 		}
371 
372 		vgic_put_irq(vcpu->kvm, irq);
373 	}
374 
375 	return 0;
376 }
377 
378 /* We want to avoid outer shareable. */
379 u64 vgic_sanitise_shareability(u64 field)
380 {
381 	switch (field) {
382 	case GIC_BASER_OuterShareable:
383 		return GIC_BASER_InnerShareable;
384 	default:
385 		return field;
386 	}
387 }
388 
389 /* Avoid any inner non-cacheable mapping. */
390 u64 vgic_sanitise_inner_cacheability(u64 field)
391 {
392 	switch (field) {
393 	case GIC_BASER_CACHE_nCnB:
394 	case GIC_BASER_CACHE_nC:
395 		return GIC_BASER_CACHE_RaWb;
396 	default:
397 		return field;
398 	}
399 }
400 
401 /* Non-cacheable or same-as-inner are OK. */
402 u64 vgic_sanitise_outer_cacheability(u64 field)
403 {
404 	switch (field) {
405 	case GIC_BASER_CACHE_SameAsInner:
406 	case GIC_BASER_CACHE_nC:
407 		return field;
408 	default:
409 		return GIC_BASER_CACHE_SameAsInner;
410 	}
411 }
412 
413 u64 vgic_sanitise_field(u64 reg, u64 field_mask, int field_shift,
414 			u64 (*sanitise_fn)(u64))
415 {
416 	u64 field = (reg & field_mask) >> field_shift;
417 
418 	field = sanitise_fn(field) << field_shift;
419 	return (reg & ~field_mask) | field;
420 }
421 
422 #define PROPBASER_RES0_MASK						\
423 	(GENMASK_ULL(63, 59) | GENMASK_ULL(55, 52) | GENMASK_ULL(6, 5))
424 #define PENDBASER_RES0_MASK						\
425 	(BIT_ULL(63) | GENMASK_ULL(61, 59) | GENMASK_ULL(55, 52) |	\
426 	 GENMASK_ULL(15, 12) | GENMASK_ULL(6, 0))
427 
428 static u64 vgic_sanitise_pendbaser(u64 reg)
429 {
430 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_SHAREABILITY_MASK,
431 				  GICR_PENDBASER_SHAREABILITY_SHIFT,
432 				  vgic_sanitise_shareability);
433 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_INNER_CACHEABILITY_MASK,
434 				  GICR_PENDBASER_INNER_CACHEABILITY_SHIFT,
435 				  vgic_sanitise_inner_cacheability);
436 	reg = vgic_sanitise_field(reg, GICR_PENDBASER_OUTER_CACHEABILITY_MASK,
437 				  GICR_PENDBASER_OUTER_CACHEABILITY_SHIFT,
438 				  vgic_sanitise_outer_cacheability);
439 
440 	reg &= ~PENDBASER_RES0_MASK;
441 
442 	return reg;
443 }
444 
445 static u64 vgic_sanitise_propbaser(u64 reg)
446 {
447 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_SHAREABILITY_MASK,
448 				  GICR_PROPBASER_SHAREABILITY_SHIFT,
449 				  vgic_sanitise_shareability);
450 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_INNER_CACHEABILITY_MASK,
451 				  GICR_PROPBASER_INNER_CACHEABILITY_SHIFT,
452 				  vgic_sanitise_inner_cacheability);
453 	reg = vgic_sanitise_field(reg, GICR_PROPBASER_OUTER_CACHEABILITY_MASK,
454 				  GICR_PROPBASER_OUTER_CACHEABILITY_SHIFT,
455 				  vgic_sanitise_outer_cacheability);
456 
457 	reg &= ~PROPBASER_RES0_MASK;
458 	return reg;
459 }
460 
461 static unsigned long vgic_mmio_read_propbase(struct kvm_vcpu *vcpu,
462 					     gpa_t addr, unsigned int len)
463 {
464 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
465 
466 	return extract_bytes(dist->propbaser, addr & 7, len);
467 }
468 
469 static void vgic_mmio_write_propbase(struct kvm_vcpu *vcpu,
470 				     gpa_t addr, unsigned int len,
471 				     unsigned long val)
472 {
473 	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
474 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
475 	u64 old_propbaser, propbaser;
476 
477 	/* Storing a value with LPIs already enabled is undefined */
478 	if (vgic_cpu->lpis_enabled)
479 		return;
480 
481 	do {
482 		old_propbaser = READ_ONCE(dist->propbaser);
483 		propbaser = old_propbaser;
484 		propbaser = update_64bit_reg(propbaser, addr & 4, len, val);
485 		propbaser = vgic_sanitise_propbaser(propbaser);
486 	} while (cmpxchg64(&dist->propbaser, old_propbaser,
487 			   propbaser) != old_propbaser);
488 }
489 
490 static unsigned long vgic_mmio_read_pendbase(struct kvm_vcpu *vcpu,
491 					     gpa_t addr, unsigned int len)
492 {
493 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
494 	u64 value = vgic_cpu->pendbaser;
495 
496 	value &= ~GICR_PENDBASER_PTZ;
497 
498 	return extract_bytes(value, addr & 7, len);
499 }
500 
501 static void vgic_mmio_write_pendbase(struct kvm_vcpu *vcpu,
502 				     gpa_t addr, unsigned int len,
503 				     unsigned long val)
504 {
505 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
506 	u64 old_pendbaser, pendbaser;
507 
508 	/* Storing a value with LPIs already enabled is undefined */
509 	if (vgic_cpu->lpis_enabled)
510 		return;
511 
512 	do {
513 		old_pendbaser = READ_ONCE(vgic_cpu->pendbaser);
514 		pendbaser = old_pendbaser;
515 		pendbaser = update_64bit_reg(pendbaser, addr & 4, len, val);
516 		pendbaser = vgic_sanitise_pendbaser(pendbaser);
517 	} while (cmpxchg64(&vgic_cpu->pendbaser, old_pendbaser,
518 			   pendbaser) != old_pendbaser);
519 }
520 
521 /*
522  * The GICv3 per-IRQ registers are split to control PPIs and SGIs in the
523  * redistributors, while SPIs are covered by registers in the distributor
524  * block. Trying to set private IRQs in this block gets ignored.
525  * We take some special care here to fix the calculation of the register
526  * offset.
527  */
528 #define REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(off, rd, wr, ur, uw, bpi, acc) \
529 	{								\
530 		.reg_offset = off,					\
531 		.bits_per_irq = bpi,					\
532 		.len = (bpi * VGIC_NR_PRIVATE_IRQS) / 8,		\
533 		.access_flags = acc,					\
534 		.read = vgic_mmio_read_raz,				\
535 		.write = vgic_mmio_write_wi,				\
536 	}, {								\
537 		.reg_offset = off + (bpi * VGIC_NR_PRIVATE_IRQS) / 8,	\
538 		.bits_per_irq = bpi,					\
539 		.len = (bpi * (1024 - VGIC_NR_PRIVATE_IRQS)) / 8,	\
540 		.access_flags = acc,					\
541 		.read = rd,						\
542 		.write = wr,						\
543 		.uaccess_read = ur,					\
544 		.uaccess_write = uw,					\
545 	}
546 
547 static const struct vgic_register_region vgic_v3_dist_registers[] = {
548 	REGISTER_DESC_WITH_LENGTH_UACCESS(GICD_CTLR,
549 		vgic_mmio_read_v3_misc, vgic_mmio_write_v3_misc,
550 		NULL, vgic_mmio_uaccess_write_v3_misc,
551 		16, VGIC_ACCESS_32bit),
552 	REGISTER_DESC_WITH_LENGTH(GICD_STATUSR,
553 		vgic_mmio_read_rao, vgic_mmio_write_wi, 4,
554 		VGIC_ACCESS_32bit),
555 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGROUPR,
556 		vgic_mmio_read_group, vgic_mmio_write_group, NULL, NULL, 1,
557 		VGIC_ACCESS_32bit),
558 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISENABLER,
559 		vgic_mmio_read_enable, vgic_mmio_write_senable,
560 		NULL, vgic_uaccess_write_senable, 1,
561 		VGIC_ACCESS_32bit),
562 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICENABLER,
563 		vgic_mmio_read_enable, vgic_mmio_write_cenable,
564 	       NULL, vgic_uaccess_write_cenable, 1,
565 		VGIC_ACCESS_32bit),
566 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISPENDR,
567 		vgic_mmio_read_pending, vgic_mmio_write_spending,
568 		vgic_v3_uaccess_read_pending, vgic_v3_uaccess_write_pending, 1,
569 		VGIC_ACCESS_32bit),
570 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICPENDR,
571 		vgic_mmio_read_pending, vgic_mmio_write_cpending,
572 		vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 1,
573 		VGIC_ACCESS_32bit),
574 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ISACTIVER,
575 		vgic_mmio_read_active, vgic_mmio_write_sactive,
576 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 1,
577 		VGIC_ACCESS_32bit),
578 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICACTIVER,
579 		vgic_mmio_read_active, vgic_mmio_write_cactive,
580 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive,
581 		1, VGIC_ACCESS_32bit),
582 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IPRIORITYR,
583 		vgic_mmio_read_priority, vgic_mmio_write_priority, NULL, NULL,
584 		8, VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
585 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ITARGETSR,
586 		vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 8,
587 		VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
588 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_ICFGR,
589 		vgic_mmio_read_config, vgic_mmio_write_config, NULL, NULL, 2,
590 		VGIC_ACCESS_32bit),
591 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IGRPMODR,
592 		vgic_mmio_read_raz, vgic_mmio_write_wi, NULL, NULL, 1,
593 		VGIC_ACCESS_32bit),
594 	REGISTER_DESC_WITH_BITS_PER_IRQ_SHARED(GICD_IROUTER,
595 		vgic_mmio_read_irouter, vgic_mmio_write_irouter, NULL, NULL, 64,
596 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
597 	REGISTER_DESC_WITH_LENGTH(GICD_IDREGS,
598 		vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48,
599 		VGIC_ACCESS_32bit),
600 };
601 
602 static const struct vgic_register_region vgic_v3_rd_registers[] = {
603 	/* RD_base registers */
604 	REGISTER_DESC_WITH_LENGTH(GICR_CTLR,
605 		vgic_mmio_read_v3r_ctlr, vgic_mmio_write_v3r_ctlr, 4,
606 		VGIC_ACCESS_32bit),
607 	REGISTER_DESC_WITH_LENGTH(GICR_STATUSR,
608 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
609 		VGIC_ACCESS_32bit),
610 	REGISTER_DESC_WITH_LENGTH(GICR_IIDR,
611 		vgic_mmio_read_v3r_iidr, vgic_mmio_write_wi, 4,
612 		VGIC_ACCESS_32bit),
613 	REGISTER_DESC_WITH_LENGTH_UACCESS(GICR_TYPER,
614 		vgic_mmio_read_v3r_typer, vgic_mmio_write_wi,
615 		vgic_uaccess_read_v3r_typer, vgic_mmio_uaccess_write_wi, 8,
616 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
617 	REGISTER_DESC_WITH_LENGTH(GICR_WAKER,
618 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
619 		VGIC_ACCESS_32bit),
620 	REGISTER_DESC_WITH_LENGTH(GICR_PROPBASER,
621 		vgic_mmio_read_propbase, vgic_mmio_write_propbase, 8,
622 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
623 	REGISTER_DESC_WITH_LENGTH(GICR_PENDBASER,
624 		vgic_mmio_read_pendbase, vgic_mmio_write_pendbase, 8,
625 		VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
626 	REGISTER_DESC_WITH_LENGTH(GICR_IDREGS,
627 		vgic_mmio_read_v3_idregs, vgic_mmio_write_wi, 48,
628 		VGIC_ACCESS_32bit),
629 	/* SGI_base registers */
630 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGROUPR0,
631 		vgic_mmio_read_group, vgic_mmio_write_group, 4,
632 		VGIC_ACCESS_32bit),
633 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISENABLER0,
634 		vgic_mmio_read_enable, vgic_mmio_write_senable,
635 		NULL, vgic_uaccess_write_senable, 4,
636 		VGIC_ACCESS_32bit),
637 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICENABLER0,
638 		vgic_mmio_read_enable, vgic_mmio_write_cenable,
639 		NULL, vgic_uaccess_write_cenable, 4,
640 		VGIC_ACCESS_32bit),
641 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISPENDR0,
642 		vgic_mmio_read_pending, vgic_mmio_write_spending,
643 		vgic_v3_uaccess_read_pending, vgic_v3_uaccess_write_pending, 4,
644 		VGIC_ACCESS_32bit),
645 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICPENDR0,
646 		vgic_mmio_read_pending, vgic_mmio_write_cpending,
647 		vgic_mmio_read_raz, vgic_mmio_uaccess_write_wi, 4,
648 		VGIC_ACCESS_32bit),
649 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ISACTIVER0,
650 		vgic_mmio_read_active, vgic_mmio_write_sactive,
651 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_sactive, 4,
652 		VGIC_ACCESS_32bit),
653 	REGISTER_DESC_WITH_LENGTH_UACCESS(SZ_64K + GICR_ICACTIVER0,
654 		vgic_mmio_read_active, vgic_mmio_write_cactive,
655 		vgic_uaccess_read_active, vgic_mmio_uaccess_write_cactive, 4,
656 		VGIC_ACCESS_32bit),
657 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IPRIORITYR0,
658 		vgic_mmio_read_priority, vgic_mmio_write_priority, 32,
659 		VGIC_ACCESS_32bit | VGIC_ACCESS_8bit),
660 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_ICFGR0,
661 		vgic_mmio_read_config, vgic_mmio_write_config, 8,
662 		VGIC_ACCESS_32bit),
663 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_IGRPMODR0,
664 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
665 		VGIC_ACCESS_32bit),
666 	REGISTER_DESC_WITH_LENGTH(SZ_64K + GICR_NSACR,
667 		vgic_mmio_read_raz, vgic_mmio_write_wi, 4,
668 		VGIC_ACCESS_32bit),
669 };
670 
671 unsigned int vgic_v3_init_dist_iodev(struct vgic_io_device *dev)
672 {
673 	dev->regions = vgic_v3_dist_registers;
674 	dev->nr_regions = ARRAY_SIZE(vgic_v3_dist_registers);
675 
676 	kvm_iodevice_init(&dev->dev, &kvm_io_gic_ops);
677 
678 	return SZ_64K;
679 }
680 
681 /**
682  * vgic_register_redist_iodev - register a single redist iodev
683  * @vcpu:    The VCPU to which the redistributor belongs
684  *
685  * Register a KVM iodev for this VCPU's redistributor using the address
686  * provided.
687  *
688  * Return 0 on success, -ERRNO otherwise.
689  */
690 int vgic_register_redist_iodev(struct kvm_vcpu *vcpu)
691 {
692 	struct kvm *kvm = vcpu->kvm;
693 	struct vgic_dist *vgic = &kvm->arch.vgic;
694 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
695 	struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev;
696 	struct vgic_redist_region *rdreg;
697 	gpa_t rd_base;
698 	int ret;
699 
700 	if (!IS_VGIC_ADDR_UNDEF(vgic_cpu->rd_iodev.base_addr))
701 		return 0;
702 
703 	/*
704 	 * We may be creating VCPUs before having set the base address for the
705 	 * redistributor region, in which case we will come back to this
706 	 * function for all VCPUs when the base address is set.  Just return
707 	 * without doing any work for now.
708 	 */
709 	rdreg = vgic_v3_rdist_free_slot(&vgic->rd_regions);
710 	if (!rdreg)
711 		return 0;
712 
713 	if (!vgic_v3_check_base(kvm))
714 		return -EINVAL;
715 
716 	vgic_cpu->rdreg = rdreg;
717 
718 	rd_base = rdreg->base + rdreg->free_index * KVM_VGIC_V3_REDIST_SIZE;
719 
720 	kvm_iodevice_init(&rd_dev->dev, &kvm_io_gic_ops);
721 	rd_dev->base_addr = rd_base;
722 	rd_dev->iodev_type = IODEV_REDIST;
723 	rd_dev->regions = vgic_v3_rd_registers;
724 	rd_dev->nr_regions = ARRAY_SIZE(vgic_v3_rd_registers);
725 	rd_dev->redist_vcpu = vcpu;
726 
727 	mutex_lock(&kvm->slots_lock);
728 	ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS, rd_base,
729 				      2 * SZ_64K, &rd_dev->dev);
730 	mutex_unlock(&kvm->slots_lock);
731 
732 	if (ret)
733 		return ret;
734 
735 	rdreg->free_index++;
736 	return 0;
737 }
738 
739 static void vgic_unregister_redist_iodev(struct kvm_vcpu *vcpu)
740 {
741 	struct vgic_io_device *rd_dev = &vcpu->arch.vgic_cpu.rd_iodev;
742 
743 	kvm_io_bus_unregister_dev(vcpu->kvm, KVM_MMIO_BUS, &rd_dev->dev);
744 }
745 
746 static int vgic_register_all_redist_iodevs(struct kvm *kvm)
747 {
748 	struct kvm_vcpu *vcpu;
749 	int c, ret = 0;
750 
751 	kvm_for_each_vcpu(c, vcpu, kvm) {
752 		ret = vgic_register_redist_iodev(vcpu);
753 		if (ret)
754 			break;
755 	}
756 
757 	if (ret) {
758 		/* The current c failed, so we start with the previous one. */
759 		mutex_lock(&kvm->slots_lock);
760 		for (c--; c >= 0; c--) {
761 			vcpu = kvm_get_vcpu(kvm, c);
762 			vgic_unregister_redist_iodev(vcpu);
763 		}
764 		mutex_unlock(&kvm->slots_lock);
765 	}
766 
767 	return ret;
768 }
769 
770 /**
771  * vgic_v3_insert_redist_region - Insert a new redistributor region
772  *
773  * Performs various checks before inserting the rdist region in the list.
774  * Those tests depend on whether the size of the rdist region is known
775  * (ie. count != 0). The list is sorted by rdist region index.
776  *
777  * @kvm: kvm handle
778  * @index: redist region index
779  * @base: base of the new rdist region
780  * @count: number of redistributors the region is made of (0 in the old style
781  * single region, whose size is induced from the number of vcpus)
782  *
783  * Return 0 on success, < 0 otherwise
784  */
785 static int vgic_v3_insert_redist_region(struct kvm *kvm, uint32_t index,
786 					gpa_t base, uint32_t count)
787 {
788 	struct vgic_dist *d = &kvm->arch.vgic;
789 	struct vgic_redist_region *rdreg;
790 	struct list_head *rd_regions = &d->rd_regions;
791 	size_t size = count * KVM_VGIC_V3_REDIST_SIZE;
792 	int ret;
793 
794 	/* single rdist region already set ?*/
795 	if (!count && !list_empty(rd_regions))
796 		return -EINVAL;
797 
798 	/* cross the end of memory ? */
799 	if (base + size < base)
800 		return -EINVAL;
801 
802 	if (list_empty(rd_regions)) {
803 		if (index != 0)
804 			return -EINVAL;
805 	} else {
806 		rdreg = list_last_entry(rd_regions,
807 					struct vgic_redist_region, list);
808 		if (index != rdreg->index + 1)
809 			return -EINVAL;
810 
811 		/* Cannot add an explicitly sized regions after legacy region */
812 		if (!rdreg->count)
813 			return -EINVAL;
814 	}
815 
816 	/*
817 	 * For legacy single-region redistributor regions (!count),
818 	 * check that the redistributor region does not overlap with the
819 	 * distributor's address space.
820 	 */
821 	if (!count && !IS_VGIC_ADDR_UNDEF(d->vgic_dist_base) &&
822 		vgic_dist_overlap(kvm, base, size))
823 		return -EINVAL;
824 
825 	/* collision with any other rdist region? */
826 	if (vgic_v3_rdist_overlap(kvm, base, size))
827 		return -EINVAL;
828 
829 	rdreg = kzalloc(sizeof(*rdreg), GFP_KERNEL);
830 	if (!rdreg)
831 		return -ENOMEM;
832 
833 	rdreg->base = VGIC_ADDR_UNDEF;
834 
835 	ret = vgic_check_ioaddr(kvm, &rdreg->base, base, SZ_64K);
836 	if (ret)
837 		goto free;
838 
839 	rdreg->base = base;
840 	rdreg->count = count;
841 	rdreg->free_index = 0;
842 	rdreg->index = index;
843 
844 	list_add_tail(&rdreg->list, rd_regions);
845 	return 0;
846 free:
847 	kfree(rdreg);
848 	return ret;
849 }
850 
851 int vgic_v3_set_redist_base(struct kvm *kvm, u32 index, u64 addr, u32 count)
852 {
853 	int ret;
854 
855 	ret = vgic_v3_insert_redist_region(kvm, index, addr, count);
856 	if (ret)
857 		return ret;
858 
859 	/*
860 	 * Register iodevs for each existing VCPU.  Adding more VCPUs
861 	 * afterwards will register the iodevs when needed.
862 	 */
863 	ret = vgic_register_all_redist_iodevs(kvm);
864 	if (ret)
865 		return ret;
866 
867 	return 0;
868 }
869 
870 int vgic_v3_has_attr_regs(struct kvm_device *dev, struct kvm_device_attr *attr)
871 {
872 	const struct vgic_register_region *region;
873 	struct vgic_io_device iodev;
874 	struct vgic_reg_attr reg_attr;
875 	struct kvm_vcpu *vcpu;
876 	gpa_t addr;
877 	int ret;
878 
879 	ret = vgic_v3_parse_attr(dev, attr, &reg_attr);
880 	if (ret)
881 		return ret;
882 
883 	vcpu = reg_attr.vcpu;
884 	addr = reg_attr.addr;
885 
886 	switch (attr->group) {
887 	case KVM_DEV_ARM_VGIC_GRP_DIST_REGS:
888 		iodev.regions = vgic_v3_dist_registers;
889 		iodev.nr_regions = ARRAY_SIZE(vgic_v3_dist_registers);
890 		iodev.base_addr = 0;
891 		break;
892 	case KVM_DEV_ARM_VGIC_GRP_REDIST_REGS:{
893 		iodev.regions = vgic_v3_rd_registers;
894 		iodev.nr_regions = ARRAY_SIZE(vgic_v3_rd_registers);
895 		iodev.base_addr = 0;
896 		break;
897 	}
898 	case KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS: {
899 		u64 reg, id;
900 
901 		id = (attr->attr & KVM_DEV_ARM_VGIC_SYSREG_INSTR_MASK);
902 		return vgic_v3_has_cpu_sysregs_attr(vcpu, 0, id, &reg);
903 	}
904 	default:
905 		return -ENXIO;
906 	}
907 
908 	/* We only support aligned 32-bit accesses. */
909 	if (addr & 3)
910 		return -ENXIO;
911 
912 	region = vgic_get_mmio_region(vcpu, &iodev, addr, sizeof(u32));
913 	if (!region)
914 		return -ENXIO;
915 
916 	return 0;
917 }
918 /*
919  * Compare a given affinity (level 1-3 and a level 0 mask, from the SGI
920  * generation register ICC_SGI1R_EL1) with a given VCPU.
921  * If the VCPU's MPIDR matches, return the level0 affinity, otherwise
922  * return -1.
923  */
924 static int match_mpidr(u64 sgi_aff, u16 sgi_cpu_mask, struct kvm_vcpu *vcpu)
925 {
926 	unsigned long affinity;
927 	int level0;
928 
929 	/*
930 	 * Split the current VCPU's MPIDR into affinity level 0 and the
931 	 * rest as this is what we have to compare against.
932 	 */
933 	affinity = kvm_vcpu_get_mpidr_aff(vcpu);
934 	level0 = MPIDR_AFFINITY_LEVEL(affinity, 0);
935 	affinity &= ~MPIDR_LEVEL_MASK;
936 
937 	/* bail out if the upper three levels don't match */
938 	if (sgi_aff != affinity)
939 		return -1;
940 
941 	/* Is this VCPU's bit set in the mask ? */
942 	if (!(sgi_cpu_mask & BIT(level0)))
943 		return -1;
944 
945 	return level0;
946 }
947 
948 /*
949  * The ICC_SGI* registers encode the affinity differently from the MPIDR,
950  * so provide a wrapper to use the existing defines to isolate a certain
951  * affinity level.
952  */
953 #define SGI_AFFINITY_LEVEL(reg, level) \
954 	((((reg) & ICC_SGI1R_AFFINITY_## level ##_MASK) \
955 	>> ICC_SGI1R_AFFINITY_## level ##_SHIFT) << MPIDR_LEVEL_SHIFT(level))
956 
957 /**
958  * vgic_v3_dispatch_sgi - handle SGI requests from VCPUs
959  * @vcpu: The VCPU requesting a SGI
960  * @reg: The value written into ICC_{ASGI1,SGI0,SGI1}R by that VCPU
961  * @allow_group1: Does the sysreg access allow generation of G1 SGIs
962  *
963  * With GICv3 (and ARE=1) CPUs trigger SGIs by writing to a system register.
964  * This will trap in sys_regs.c and call this function.
965  * This ICC_SGI1R_EL1 register contains the upper three affinity levels of the
966  * target processors as well as a bitmask of 16 Aff0 CPUs.
967  * If the interrupt routing mode bit is not set, we iterate over all VCPUs to
968  * check for matching ones. If this bit is set, we signal all, but not the
969  * calling VCPU.
970  */
971 void vgic_v3_dispatch_sgi(struct kvm_vcpu *vcpu, u64 reg, bool allow_group1)
972 {
973 	struct kvm *kvm = vcpu->kvm;
974 	struct kvm_vcpu *c_vcpu;
975 	u16 target_cpus;
976 	u64 mpidr;
977 	int sgi, c;
978 	int vcpu_id = vcpu->vcpu_id;
979 	bool broadcast;
980 	unsigned long flags;
981 
982 	sgi = (reg & ICC_SGI1R_SGI_ID_MASK) >> ICC_SGI1R_SGI_ID_SHIFT;
983 	broadcast = reg & BIT_ULL(ICC_SGI1R_IRQ_ROUTING_MODE_BIT);
984 	target_cpus = (reg & ICC_SGI1R_TARGET_LIST_MASK) >> ICC_SGI1R_TARGET_LIST_SHIFT;
985 	mpidr = SGI_AFFINITY_LEVEL(reg, 3);
986 	mpidr |= SGI_AFFINITY_LEVEL(reg, 2);
987 	mpidr |= SGI_AFFINITY_LEVEL(reg, 1);
988 
989 	/*
990 	 * We iterate over all VCPUs to find the MPIDRs matching the request.
991 	 * If we have handled one CPU, we clear its bit to detect early
992 	 * if we are already finished. This avoids iterating through all
993 	 * VCPUs when most of the times we just signal a single VCPU.
994 	 */
995 	kvm_for_each_vcpu(c, c_vcpu, kvm) {
996 		struct vgic_irq *irq;
997 
998 		/* Exit early if we have dealt with all requested CPUs */
999 		if (!broadcast && target_cpus == 0)
1000 			break;
1001 
1002 		/* Don't signal the calling VCPU */
1003 		if (broadcast && c == vcpu_id)
1004 			continue;
1005 
1006 		if (!broadcast) {
1007 			int level0;
1008 
1009 			level0 = match_mpidr(mpidr, target_cpus, c_vcpu);
1010 			if (level0 == -1)
1011 				continue;
1012 
1013 			/* remove this matching VCPU from the mask */
1014 			target_cpus &= ~BIT(level0);
1015 		}
1016 
1017 		irq = vgic_get_irq(vcpu->kvm, c_vcpu, sgi);
1018 
1019 		raw_spin_lock_irqsave(&irq->irq_lock, flags);
1020 
1021 		/*
1022 		 * An access targeting Group0 SGIs can only generate
1023 		 * those, while an access targeting Group1 SGIs can
1024 		 * generate interrupts of either group.
1025 		 */
1026 		if (!irq->group || allow_group1) {
1027 			if (!irq->hw) {
1028 				irq->pending_latch = true;
1029 				vgic_queue_irq_unlock(vcpu->kvm, irq, flags);
1030 			} else {
1031 				/* HW SGI? Ask the GIC to inject it */
1032 				int err;
1033 				err = irq_set_irqchip_state(irq->host_irq,
1034 							    IRQCHIP_STATE_PENDING,
1035 							    true);
1036 				WARN_RATELIMIT(err, "IRQ %d", irq->host_irq);
1037 				raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
1038 			}
1039 		} else {
1040 			raw_spin_unlock_irqrestore(&irq->irq_lock, flags);
1041 		}
1042 
1043 		vgic_put_irq(vcpu->kvm, irq);
1044 	}
1045 }
1046 
1047 int vgic_v3_dist_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1048 			 int offset, u32 *val)
1049 {
1050 	struct vgic_io_device dev = {
1051 		.regions = vgic_v3_dist_registers,
1052 		.nr_regions = ARRAY_SIZE(vgic_v3_dist_registers),
1053 	};
1054 
1055 	return vgic_uaccess(vcpu, &dev, is_write, offset, val);
1056 }
1057 
1058 int vgic_v3_redist_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1059 			   int offset, u32 *val)
1060 {
1061 	struct vgic_io_device rd_dev = {
1062 		.regions = vgic_v3_rd_registers,
1063 		.nr_regions = ARRAY_SIZE(vgic_v3_rd_registers),
1064 	};
1065 
1066 	return vgic_uaccess(vcpu, &rd_dev, is_write, offset, val);
1067 }
1068 
1069 int vgic_v3_line_level_info_uaccess(struct kvm_vcpu *vcpu, bool is_write,
1070 				    u32 intid, u64 *val)
1071 {
1072 	if (intid % 32)
1073 		return -EINVAL;
1074 
1075 	if (is_write)
1076 		vgic_write_irq_line_level_info(vcpu, intid, *val);
1077 	else
1078 		*val = vgic_read_irq_line_level_info(vcpu, intid);
1079 
1080 	return 0;
1081 }
1082