xref: /openbmc/linux/arch/arm64/kvm/vgic/vgic-v2.c (revision 87875c10)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2015, 2016 ARM Ltd.
4  */
5 
6 #include <linux/irqchip/arm-gic.h>
7 #include <linux/kvm.h>
8 #include <linux/kvm_host.h>
9 #include <kvm/arm_vgic.h>
10 #include <asm/kvm_mmu.h>
11 
12 #include "vgic.h"
13 
14 static inline void vgic_v2_write_lr(int lr, u32 val)
15 {
16 	void __iomem *base = kvm_vgic_global_state.vctrl_base;
17 
18 	writel_relaxed(val, base + GICH_LR0 + (lr * 4));
19 }
20 
21 void vgic_v2_init_lrs(void)
22 {
23 	int i;
24 
25 	for (i = 0; i < kvm_vgic_global_state.nr_lr; i++)
26 		vgic_v2_write_lr(i, 0);
27 }
28 
29 void vgic_v2_set_underflow(struct kvm_vcpu *vcpu)
30 {
31 	struct vgic_v2_cpu_if *cpuif = &vcpu->arch.vgic_cpu.vgic_v2;
32 
33 	cpuif->vgic_hcr |= GICH_HCR_UIE;
34 }
35 
36 static bool lr_signals_eoi_mi(u32 lr_val)
37 {
38 	return !(lr_val & GICH_LR_STATE) && (lr_val & GICH_LR_EOI) &&
39 	       !(lr_val & GICH_LR_HW);
40 }
41 
42 /*
43  * transfer the content of the LRs back into the corresponding ap_list:
44  * - active bit is transferred as is
45  * - pending bit is
46  *   - transferred as is in case of edge sensitive IRQs
47  *   - set to the line-level (resample time) for level sensitive IRQs
48  */
49 void vgic_v2_fold_lr_state(struct kvm_vcpu *vcpu)
50 {
51 	struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
52 	struct vgic_v2_cpu_if *cpuif = &vgic_cpu->vgic_v2;
53 	int lr;
54 
55 	DEBUG_SPINLOCK_BUG_ON(!irqs_disabled());
56 
57 	cpuif->vgic_hcr &= ~GICH_HCR_UIE;
58 
59 	for (lr = 0; lr < vgic_cpu->vgic_v2.used_lrs; lr++) {
60 		u32 val = cpuif->vgic_lr[lr];
61 		u32 cpuid, intid = val & GICH_LR_VIRTUALID;
62 		struct vgic_irq *irq;
63 
64 		/* Extract the source vCPU id from the LR */
65 		cpuid = val & GICH_LR_PHYSID_CPUID;
66 		cpuid >>= GICH_LR_PHYSID_CPUID_SHIFT;
67 		cpuid &= 7;
68 
69 		/* Notify fds when the guest EOI'ed a level-triggered SPI */
70 		if (lr_signals_eoi_mi(val) && vgic_valid_spi(vcpu->kvm, intid))
71 			kvm_notify_acked_irq(vcpu->kvm, 0,
72 					     intid - VGIC_NR_PRIVATE_IRQS);
73 
74 		irq = vgic_get_irq(vcpu->kvm, vcpu, intid);
75 
76 		raw_spin_lock(&irq->irq_lock);
77 
78 		/* Always preserve the active bit */
79 		irq->active = !!(val & GICH_LR_ACTIVE_BIT);
80 
81 		if (irq->active && vgic_irq_is_sgi(intid))
82 			irq->active_source = cpuid;
83 
84 		/* Edge is the only case where we preserve the pending bit */
85 		if (irq->config == VGIC_CONFIG_EDGE &&
86 		    (val & GICH_LR_PENDING_BIT)) {
87 			irq->pending_latch = true;
88 
89 			if (vgic_irq_is_sgi(intid))
90 				irq->source |= (1 << cpuid);
91 		}
92 
93 		/*
94 		 * Clear soft pending state when level irqs have been acked.
95 		 */
96 		if (irq->config == VGIC_CONFIG_LEVEL && !(val & GICH_LR_STATE))
97 			irq->pending_latch = false;
98 
99 		/*
100 		 * Level-triggered mapped IRQs are special because we only
101 		 * observe rising edges as input to the VGIC.
102 		 *
103 		 * If the guest never acked the interrupt we have to sample
104 		 * the physical line and set the line level, because the
105 		 * device state could have changed or we simply need to
106 		 * process the still pending interrupt later.
107 		 *
108 		 * If this causes us to lower the level, we have to also clear
109 		 * the physical active state, since we will otherwise never be
110 		 * told when the interrupt becomes asserted again.
111 		 *
112 		 * Another case is when the interrupt requires a helping hand
113 		 * on deactivation (no HW deactivation, for example).
114 		 */
115 		if (vgic_irq_is_mapped_level(irq)) {
116 			bool resample = false;
117 
118 			if (val & GICH_LR_PENDING_BIT) {
119 				irq->line_level = vgic_get_phys_line_level(irq);
120 				resample = !irq->line_level;
121 			} else if (vgic_irq_needs_resampling(irq) &&
122 				   !(irq->active || irq->pending_latch)) {
123 				resample = true;
124 			}
125 
126 			if (resample)
127 				vgic_irq_set_phys_active(irq, false);
128 		}
129 
130 		raw_spin_unlock(&irq->irq_lock);
131 		vgic_put_irq(vcpu->kvm, irq);
132 	}
133 
134 	cpuif->used_lrs = 0;
135 }
136 
137 /*
138  * Populates the particular LR with the state of a given IRQ:
139  * - for an edge sensitive IRQ the pending state is cleared in struct vgic_irq
140  * - for a level sensitive IRQ the pending state value is unchanged;
141  *   it is dictated directly by the input level
142  *
143  * If @irq describes an SGI with multiple sources, we choose the
144  * lowest-numbered source VCPU and clear that bit in the source bitmap.
145  *
146  * The irq_lock must be held by the caller.
147  */
148 void vgic_v2_populate_lr(struct kvm_vcpu *vcpu, struct vgic_irq *irq, int lr)
149 {
150 	u32 val = irq->intid;
151 	bool allow_pending = true;
152 
153 	if (irq->active) {
154 		val |= GICH_LR_ACTIVE_BIT;
155 		if (vgic_irq_is_sgi(irq->intid))
156 			val |= irq->active_source << GICH_LR_PHYSID_CPUID_SHIFT;
157 		if (vgic_irq_is_multi_sgi(irq)) {
158 			allow_pending = false;
159 			val |= GICH_LR_EOI;
160 		}
161 	}
162 
163 	if (irq->group)
164 		val |= GICH_LR_GROUP1;
165 
166 	if (irq->hw && !vgic_irq_needs_resampling(irq)) {
167 		val |= GICH_LR_HW;
168 		val |= irq->hwintid << GICH_LR_PHYSID_CPUID_SHIFT;
169 		/*
170 		 * Never set pending+active on a HW interrupt, as the
171 		 * pending state is kept at the physical distributor
172 		 * level.
173 		 */
174 		if (irq->active)
175 			allow_pending = false;
176 	} else {
177 		if (irq->config == VGIC_CONFIG_LEVEL) {
178 			val |= GICH_LR_EOI;
179 
180 			/*
181 			 * Software resampling doesn't work very well
182 			 * if we allow P+A, so let's not do that.
183 			 */
184 			if (irq->active)
185 				allow_pending = false;
186 		}
187 	}
188 
189 	if (allow_pending && irq_is_pending(irq)) {
190 		val |= GICH_LR_PENDING_BIT;
191 
192 		if (irq->config == VGIC_CONFIG_EDGE)
193 			irq->pending_latch = false;
194 
195 		if (vgic_irq_is_sgi(irq->intid)) {
196 			u32 src = ffs(irq->source);
197 
198 			if (WARN_RATELIMIT(!src, "No SGI source for INTID %d\n",
199 					   irq->intid))
200 				return;
201 
202 			val |= (src - 1) << GICH_LR_PHYSID_CPUID_SHIFT;
203 			irq->source &= ~(1 << (src - 1));
204 			if (irq->source) {
205 				irq->pending_latch = true;
206 				val |= GICH_LR_EOI;
207 			}
208 		}
209 	}
210 
211 	/*
212 	 * Level-triggered mapped IRQs are special because we only observe
213 	 * rising edges as input to the VGIC.  We therefore lower the line
214 	 * level here, so that we can take new virtual IRQs.  See
215 	 * vgic_v2_fold_lr_state for more info.
216 	 */
217 	if (vgic_irq_is_mapped_level(irq) && (val & GICH_LR_PENDING_BIT))
218 		irq->line_level = false;
219 
220 	/* The GICv2 LR only holds five bits of priority. */
221 	val |= (irq->priority >> 3) << GICH_LR_PRIORITY_SHIFT;
222 
223 	vcpu->arch.vgic_cpu.vgic_v2.vgic_lr[lr] = val;
224 }
225 
226 void vgic_v2_clear_lr(struct kvm_vcpu *vcpu, int lr)
227 {
228 	vcpu->arch.vgic_cpu.vgic_v2.vgic_lr[lr] = 0;
229 }
230 
231 void vgic_v2_set_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcrp)
232 {
233 	struct vgic_v2_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v2;
234 	u32 vmcr;
235 
236 	vmcr = (vmcrp->grpen0 << GICH_VMCR_ENABLE_GRP0_SHIFT) &
237 		GICH_VMCR_ENABLE_GRP0_MASK;
238 	vmcr |= (vmcrp->grpen1 << GICH_VMCR_ENABLE_GRP1_SHIFT) &
239 		GICH_VMCR_ENABLE_GRP1_MASK;
240 	vmcr |= (vmcrp->ackctl << GICH_VMCR_ACK_CTL_SHIFT) &
241 		GICH_VMCR_ACK_CTL_MASK;
242 	vmcr |= (vmcrp->fiqen << GICH_VMCR_FIQ_EN_SHIFT) &
243 		GICH_VMCR_FIQ_EN_MASK;
244 	vmcr |= (vmcrp->cbpr << GICH_VMCR_CBPR_SHIFT) &
245 		GICH_VMCR_CBPR_MASK;
246 	vmcr |= (vmcrp->eoim << GICH_VMCR_EOI_MODE_SHIFT) &
247 		GICH_VMCR_EOI_MODE_MASK;
248 	vmcr |= (vmcrp->abpr << GICH_VMCR_ALIAS_BINPOINT_SHIFT) &
249 		GICH_VMCR_ALIAS_BINPOINT_MASK;
250 	vmcr |= (vmcrp->bpr << GICH_VMCR_BINPOINT_SHIFT) &
251 		GICH_VMCR_BINPOINT_MASK;
252 	vmcr |= ((vmcrp->pmr >> GICV_PMR_PRIORITY_SHIFT) <<
253 		 GICH_VMCR_PRIMASK_SHIFT) & GICH_VMCR_PRIMASK_MASK;
254 
255 	cpu_if->vgic_vmcr = vmcr;
256 }
257 
258 void vgic_v2_get_vmcr(struct kvm_vcpu *vcpu, struct vgic_vmcr *vmcrp)
259 {
260 	struct vgic_v2_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v2;
261 	u32 vmcr;
262 
263 	vmcr = cpu_if->vgic_vmcr;
264 
265 	vmcrp->grpen0 = (vmcr & GICH_VMCR_ENABLE_GRP0_MASK) >>
266 		GICH_VMCR_ENABLE_GRP0_SHIFT;
267 	vmcrp->grpen1 = (vmcr & GICH_VMCR_ENABLE_GRP1_MASK) >>
268 		GICH_VMCR_ENABLE_GRP1_SHIFT;
269 	vmcrp->ackctl = (vmcr & GICH_VMCR_ACK_CTL_MASK) >>
270 		GICH_VMCR_ACK_CTL_SHIFT;
271 	vmcrp->fiqen = (vmcr & GICH_VMCR_FIQ_EN_MASK) >>
272 		GICH_VMCR_FIQ_EN_SHIFT;
273 	vmcrp->cbpr = (vmcr & GICH_VMCR_CBPR_MASK) >>
274 		GICH_VMCR_CBPR_SHIFT;
275 	vmcrp->eoim = (vmcr & GICH_VMCR_EOI_MODE_MASK) >>
276 		GICH_VMCR_EOI_MODE_SHIFT;
277 
278 	vmcrp->abpr = (vmcr & GICH_VMCR_ALIAS_BINPOINT_MASK) >>
279 			GICH_VMCR_ALIAS_BINPOINT_SHIFT;
280 	vmcrp->bpr  = (vmcr & GICH_VMCR_BINPOINT_MASK) >>
281 			GICH_VMCR_BINPOINT_SHIFT;
282 	vmcrp->pmr  = ((vmcr & GICH_VMCR_PRIMASK_MASK) >>
283 			GICH_VMCR_PRIMASK_SHIFT) << GICV_PMR_PRIORITY_SHIFT;
284 }
285 
286 void vgic_v2_enable(struct kvm_vcpu *vcpu)
287 {
288 	/*
289 	 * By forcing VMCR to zero, the GIC will restore the binary
290 	 * points to their reset values. Anything else resets to zero
291 	 * anyway.
292 	 */
293 	vcpu->arch.vgic_cpu.vgic_v2.vgic_vmcr = 0;
294 
295 	/* Get the show on the road... */
296 	vcpu->arch.vgic_cpu.vgic_v2.vgic_hcr = GICH_HCR_EN;
297 }
298 
299 /* check for overlapping regions and for regions crossing the end of memory */
300 static bool vgic_v2_check_base(gpa_t dist_base, gpa_t cpu_base)
301 {
302 	if (dist_base + KVM_VGIC_V2_DIST_SIZE < dist_base)
303 		return false;
304 	if (cpu_base + KVM_VGIC_V2_CPU_SIZE < cpu_base)
305 		return false;
306 
307 	if (dist_base + KVM_VGIC_V2_DIST_SIZE <= cpu_base)
308 		return true;
309 	if (cpu_base + KVM_VGIC_V2_CPU_SIZE <= dist_base)
310 		return true;
311 
312 	return false;
313 }
314 
315 int vgic_v2_map_resources(struct kvm *kvm)
316 {
317 	struct vgic_dist *dist = &kvm->arch.vgic;
318 	int ret = 0;
319 
320 	if (IS_VGIC_ADDR_UNDEF(dist->vgic_dist_base) ||
321 	    IS_VGIC_ADDR_UNDEF(dist->vgic_cpu_base)) {
322 		kvm_err("Need to set vgic cpu and dist addresses first\n");
323 		return -ENXIO;
324 	}
325 
326 	if (!vgic_v2_check_base(dist->vgic_dist_base, dist->vgic_cpu_base)) {
327 		kvm_err("VGIC CPU and dist frames overlap\n");
328 		return -EINVAL;
329 	}
330 
331 	/*
332 	 * Initialize the vgic if this hasn't already been done on demand by
333 	 * accessing the vgic state from userspace.
334 	 */
335 	ret = vgic_init(kvm);
336 	if (ret) {
337 		kvm_err("Unable to initialize VGIC dynamic data structures\n");
338 		return ret;
339 	}
340 
341 	ret = vgic_register_dist_iodev(kvm, dist->vgic_dist_base, VGIC_V2);
342 	if (ret) {
343 		kvm_err("Unable to register VGIC MMIO regions\n");
344 		return ret;
345 	}
346 
347 	if (!static_branch_unlikely(&vgic_v2_cpuif_trap)) {
348 		ret = kvm_phys_addr_ioremap(kvm, dist->vgic_cpu_base,
349 					    kvm_vgic_global_state.vcpu_base,
350 					    KVM_VGIC_V2_CPU_SIZE, true);
351 		if (ret) {
352 			kvm_err("Unable to remap VGIC CPU to VCPU\n");
353 			return ret;
354 		}
355 	}
356 
357 	return 0;
358 }
359 
360 DEFINE_STATIC_KEY_FALSE(vgic_v2_cpuif_trap);
361 
362 /**
363  * vgic_v2_probe - probe for a VGICv2 compatible interrupt controller
364  * @info:	pointer to the GIC description
365  *
366  * Returns 0 if the VGICv2 has been probed successfully, returns an error code
367  * otherwise
368  */
369 int vgic_v2_probe(const struct gic_kvm_info *info)
370 {
371 	int ret;
372 	u32 vtr;
373 
374 	if (!info->vctrl.start) {
375 		kvm_err("GICH not present in the firmware table\n");
376 		return -ENXIO;
377 	}
378 
379 	if (!PAGE_ALIGNED(info->vcpu.start) ||
380 	    !PAGE_ALIGNED(resource_size(&info->vcpu))) {
381 		kvm_info("GICV region size/alignment is unsafe, using trapping (reduced performance)\n");
382 
383 		ret = create_hyp_io_mappings(info->vcpu.start,
384 					     resource_size(&info->vcpu),
385 					     &kvm_vgic_global_state.vcpu_base_va,
386 					     &kvm_vgic_global_state.vcpu_hyp_va);
387 		if (ret) {
388 			kvm_err("Cannot map GICV into hyp\n");
389 			goto out;
390 		}
391 
392 		static_branch_enable(&vgic_v2_cpuif_trap);
393 	}
394 
395 	ret = create_hyp_io_mappings(info->vctrl.start,
396 				     resource_size(&info->vctrl),
397 				     &kvm_vgic_global_state.vctrl_base,
398 				     &kvm_vgic_global_state.vctrl_hyp);
399 	if (ret) {
400 		kvm_err("Cannot map VCTRL into hyp\n");
401 		goto out;
402 	}
403 
404 	vtr = readl_relaxed(kvm_vgic_global_state.vctrl_base + GICH_VTR);
405 	kvm_vgic_global_state.nr_lr = (vtr & 0x3f) + 1;
406 
407 	ret = kvm_register_vgic_device(KVM_DEV_TYPE_ARM_VGIC_V2);
408 	if (ret) {
409 		kvm_err("Cannot register GICv2 KVM device\n");
410 		goto out;
411 	}
412 
413 	kvm_vgic_global_state.can_emulate_gicv2 = true;
414 	kvm_vgic_global_state.vcpu_base = info->vcpu.start;
415 	kvm_vgic_global_state.type = VGIC_V2;
416 	kvm_vgic_global_state.max_gic_vcpus = VGIC_V2_MAX_CPUS;
417 
418 	kvm_debug("vgic-v2@%llx\n", info->vctrl.start);
419 
420 	return 0;
421 out:
422 	if (kvm_vgic_global_state.vctrl_base)
423 		iounmap(kvm_vgic_global_state.vctrl_base);
424 	if (kvm_vgic_global_state.vcpu_base_va)
425 		iounmap(kvm_vgic_global_state.vcpu_base_va);
426 
427 	return ret;
428 }
429 
430 static void save_lrs(struct kvm_vcpu *vcpu, void __iomem *base)
431 {
432 	struct vgic_v2_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v2;
433 	u64 used_lrs = cpu_if->used_lrs;
434 	u64 elrsr;
435 	int i;
436 
437 	elrsr = readl_relaxed(base + GICH_ELRSR0);
438 	if (unlikely(used_lrs > 32))
439 		elrsr |= ((u64)readl_relaxed(base + GICH_ELRSR1)) << 32;
440 
441 	for (i = 0; i < used_lrs; i++) {
442 		if (elrsr & (1UL << i))
443 			cpu_if->vgic_lr[i] &= ~GICH_LR_STATE;
444 		else
445 			cpu_if->vgic_lr[i] = readl_relaxed(base + GICH_LR0 + (i * 4));
446 
447 		writel_relaxed(0, base + GICH_LR0 + (i * 4));
448 	}
449 }
450 
451 void vgic_v2_save_state(struct kvm_vcpu *vcpu)
452 {
453 	void __iomem *base = kvm_vgic_global_state.vctrl_base;
454 	u64 used_lrs = vcpu->arch.vgic_cpu.vgic_v2.used_lrs;
455 
456 	if (!base)
457 		return;
458 
459 	if (used_lrs) {
460 		save_lrs(vcpu, base);
461 		writel_relaxed(0, base + GICH_HCR);
462 	}
463 }
464 
465 void vgic_v2_restore_state(struct kvm_vcpu *vcpu)
466 {
467 	struct vgic_v2_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v2;
468 	void __iomem *base = kvm_vgic_global_state.vctrl_base;
469 	u64 used_lrs = cpu_if->used_lrs;
470 	int i;
471 
472 	if (!base)
473 		return;
474 
475 	if (used_lrs) {
476 		writel_relaxed(cpu_if->vgic_hcr, base + GICH_HCR);
477 		for (i = 0; i < used_lrs; i++) {
478 			writel_relaxed(cpu_if->vgic_lr[i],
479 				       base + GICH_LR0 + (i * 4));
480 		}
481 	}
482 }
483 
484 void vgic_v2_load(struct kvm_vcpu *vcpu)
485 {
486 	struct vgic_v2_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v2;
487 
488 	writel_relaxed(cpu_if->vgic_vmcr,
489 		       kvm_vgic_global_state.vctrl_base + GICH_VMCR);
490 	writel_relaxed(cpu_if->vgic_apr,
491 		       kvm_vgic_global_state.vctrl_base + GICH_APR);
492 }
493 
494 void vgic_v2_vmcr_sync(struct kvm_vcpu *vcpu)
495 {
496 	struct vgic_v2_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v2;
497 
498 	cpu_if->vgic_vmcr = readl_relaxed(kvm_vgic_global_state.vctrl_base + GICH_VMCR);
499 }
500 
501 void vgic_v2_put(struct kvm_vcpu *vcpu)
502 {
503 	struct vgic_v2_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v2;
504 
505 	vgic_v2_vmcr_sync(vcpu);
506 	cpu_if->vgic_apr = readl_relaxed(kvm_vgic_global_state.vctrl_base + GICH_APR);
507 }
508