xref: /openbmc/linux/arch/riscv/kvm/vcpu.c (revision 34fa67e7)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2019 Western Digital Corporation or its affiliates.
4  *
5  * Authors:
6  *     Anup Patel <anup.patel@wdc.com>
7  */
8 
9 #include <linux/bitops.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kdebug.h>
13 #include <linux/module.h>
14 #include <linux/percpu.h>
15 #include <linux/uaccess.h>
16 #include <linux/vmalloc.h>
17 #include <linux/sched/signal.h>
18 #include <linux/fs.h>
19 #include <linux/kvm_host.h>
20 #include <asm/csr.h>
21 #include <asm/hwcap.h>
22 
23 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
24 	KVM_GENERIC_VCPU_STATS(),
25 	STATS_DESC_COUNTER(VCPU, ecall_exit_stat),
26 	STATS_DESC_COUNTER(VCPU, wfi_exit_stat),
27 	STATS_DESC_COUNTER(VCPU, mmio_exit_user),
28 	STATS_DESC_COUNTER(VCPU, mmio_exit_kernel),
29 	STATS_DESC_COUNTER(VCPU, exits)
30 };
31 
32 const struct kvm_stats_header kvm_vcpu_stats_header = {
33 	.name_size = KVM_STATS_NAME_SIZE,
34 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
35 	.id_offset = sizeof(struct kvm_stats_header),
36 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
37 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
38 		       sizeof(kvm_vcpu_stats_desc),
39 };
40 
41 #define KVM_RISCV_ISA_ALLOWED	(riscv_isa_extension_mask(a) | \
42 				 riscv_isa_extension_mask(c) | \
43 				 riscv_isa_extension_mask(d) | \
44 				 riscv_isa_extension_mask(f) | \
45 				 riscv_isa_extension_mask(i) | \
46 				 riscv_isa_extension_mask(m) | \
47 				 riscv_isa_extension_mask(s) | \
48 				 riscv_isa_extension_mask(u))
49 
50 static void kvm_riscv_reset_vcpu(struct kvm_vcpu *vcpu)
51 {
52 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
53 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
54 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
55 	struct kvm_cpu_context *reset_cntx = &vcpu->arch.guest_reset_context;
56 	bool loaded;
57 
58 	/**
59 	 * The preemption should be disabled here because it races with
60 	 * kvm_sched_out/kvm_sched_in(called from preempt notifiers) which
61 	 * also calls vcpu_load/put.
62 	 */
63 	get_cpu();
64 	loaded = (vcpu->cpu != -1);
65 	if (loaded)
66 		kvm_arch_vcpu_put(vcpu);
67 
68 	memcpy(csr, reset_csr, sizeof(*csr));
69 
70 	memcpy(cntx, reset_cntx, sizeof(*cntx));
71 
72 	kvm_riscv_vcpu_fp_reset(vcpu);
73 
74 	kvm_riscv_vcpu_timer_reset(vcpu);
75 
76 	WRITE_ONCE(vcpu->arch.irqs_pending, 0);
77 	WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
78 
79 	/* Reset the guest CSRs for hotplug usecase */
80 	if (loaded)
81 		kvm_arch_vcpu_load(vcpu, smp_processor_id());
82 	put_cpu();
83 }
84 
85 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
86 {
87 	return 0;
88 }
89 
90 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
91 {
92 	struct kvm_cpu_context *cntx;
93 
94 	/* Mark this VCPU never ran */
95 	vcpu->arch.ran_atleast_once = false;
96 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
97 
98 	/* Setup ISA features available to VCPU */
99 	vcpu->arch.isa = riscv_isa_extension_base(NULL) & KVM_RISCV_ISA_ALLOWED;
100 
101 	/* Setup reset state of shadow SSTATUS and HSTATUS CSRs */
102 	cntx = &vcpu->arch.guest_reset_context;
103 	cntx->sstatus = SR_SPP | SR_SPIE;
104 	cntx->hstatus = 0;
105 	cntx->hstatus |= HSTATUS_VTW;
106 	cntx->hstatus |= HSTATUS_SPVP;
107 	cntx->hstatus |= HSTATUS_SPV;
108 
109 	/* Setup VCPU timer */
110 	kvm_riscv_vcpu_timer_init(vcpu);
111 
112 	/* Reset VCPU */
113 	kvm_riscv_reset_vcpu(vcpu);
114 
115 	return 0;
116 }
117 
118 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
119 {
120 	/**
121 	 * vcpu with id 0 is the designated boot cpu.
122 	 * Keep all vcpus with non-zero id in power-off state so that
123 	 * they can be brought up using SBI HSM extension.
124 	 */
125 	if (vcpu->vcpu_idx != 0)
126 		kvm_riscv_vcpu_power_off(vcpu);
127 }
128 
129 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
130 {
131 	/* Cleanup VCPU timer */
132 	kvm_riscv_vcpu_timer_deinit(vcpu);
133 
134 	/* Free unused pages pre-allocated for Stage2 page table mappings */
135 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
136 }
137 
138 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
139 {
140 	return kvm_riscv_vcpu_has_interrupts(vcpu, 1UL << IRQ_VS_TIMER);
141 }
142 
143 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
144 {
145 }
146 
147 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
148 {
149 }
150 
151 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
152 {
153 	return (kvm_riscv_vcpu_has_interrupts(vcpu, -1UL) &&
154 		!vcpu->arch.power_off && !vcpu->arch.pause);
155 }
156 
157 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
158 {
159 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
160 }
161 
162 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
163 {
164 	return (vcpu->arch.guest_context.sstatus & SR_SPP) ? true : false;
165 }
166 
167 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
168 {
169 	return VM_FAULT_SIGBUS;
170 }
171 
172 static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
173 					 const struct kvm_one_reg *reg)
174 {
175 	unsigned long __user *uaddr =
176 			(unsigned long __user *)(unsigned long)reg->addr;
177 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
178 					    KVM_REG_SIZE_MASK |
179 					    KVM_REG_RISCV_CONFIG);
180 	unsigned long reg_val;
181 
182 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
183 		return -EINVAL;
184 
185 	switch (reg_num) {
186 	case KVM_REG_RISCV_CONFIG_REG(isa):
187 		reg_val = vcpu->arch.isa;
188 		break;
189 	default:
190 		return -EINVAL;
191 	}
192 
193 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
194 		return -EFAULT;
195 
196 	return 0;
197 }
198 
199 static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
200 					 const struct kvm_one_reg *reg)
201 {
202 	unsigned long __user *uaddr =
203 			(unsigned long __user *)(unsigned long)reg->addr;
204 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
205 					    KVM_REG_SIZE_MASK |
206 					    KVM_REG_RISCV_CONFIG);
207 	unsigned long reg_val;
208 
209 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
210 		return -EINVAL;
211 
212 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
213 		return -EFAULT;
214 
215 	switch (reg_num) {
216 	case KVM_REG_RISCV_CONFIG_REG(isa):
217 		if (!vcpu->arch.ran_atleast_once) {
218 			vcpu->arch.isa = reg_val;
219 			vcpu->arch.isa &= riscv_isa_extension_base(NULL);
220 			vcpu->arch.isa &= KVM_RISCV_ISA_ALLOWED;
221 			kvm_riscv_vcpu_fp_reset(vcpu);
222 		} else {
223 			return -EOPNOTSUPP;
224 		}
225 		break;
226 	default:
227 		return -EINVAL;
228 	}
229 
230 	return 0;
231 }
232 
233 static int kvm_riscv_vcpu_get_reg_core(struct kvm_vcpu *vcpu,
234 				       const struct kvm_one_reg *reg)
235 {
236 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
237 	unsigned long __user *uaddr =
238 			(unsigned long __user *)(unsigned long)reg->addr;
239 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
240 					    KVM_REG_SIZE_MASK |
241 					    KVM_REG_RISCV_CORE);
242 	unsigned long reg_val;
243 
244 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
245 		return -EINVAL;
246 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
247 		return -EINVAL;
248 
249 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
250 		reg_val = cntx->sepc;
251 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
252 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
253 		reg_val = ((unsigned long *)cntx)[reg_num];
254 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode))
255 		reg_val = (cntx->sstatus & SR_SPP) ?
256 				KVM_RISCV_MODE_S : KVM_RISCV_MODE_U;
257 	else
258 		return -EINVAL;
259 
260 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
261 		return -EFAULT;
262 
263 	return 0;
264 }
265 
266 static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
267 				       const struct kvm_one_reg *reg)
268 {
269 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
270 	unsigned long __user *uaddr =
271 			(unsigned long __user *)(unsigned long)reg->addr;
272 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
273 					    KVM_REG_SIZE_MASK |
274 					    KVM_REG_RISCV_CORE);
275 	unsigned long reg_val;
276 
277 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
278 		return -EINVAL;
279 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
280 		return -EINVAL;
281 
282 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
283 		return -EFAULT;
284 
285 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
286 		cntx->sepc = reg_val;
287 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
288 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
289 		((unsigned long *)cntx)[reg_num] = reg_val;
290 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode)) {
291 		if (reg_val == KVM_RISCV_MODE_S)
292 			cntx->sstatus |= SR_SPP;
293 		else
294 			cntx->sstatus &= ~SR_SPP;
295 	} else
296 		return -EINVAL;
297 
298 	return 0;
299 }
300 
301 static int kvm_riscv_vcpu_get_reg_csr(struct kvm_vcpu *vcpu,
302 				      const struct kvm_one_reg *reg)
303 {
304 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
305 	unsigned long __user *uaddr =
306 			(unsigned long __user *)(unsigned long)reg->addr;
307 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
308 					    KVM_REG_SIZE_MASK |
309 					    KVM_REG_RISCV_CSR);
310 	unsigned long reg_val;
311 
312 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
313 		return -EINVAL;
314 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
315 		return -EINVAL;
316 
317 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
318 		kvm_riscv_vcpu_flush_interrupts(vcpu);
319 		reg_val = (csr->hvip >> VSIP_TO_HVIP_SHIFT) & VSIP_VALID_MASK;
320 	} else
321 		reg_val = ((unsigned long *)csr)[reg_num];
322 
323 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
324 		return -EFAULT;
325 
326 	return 0;
327 }
328 
329 static int kvm_riscv_vcpu_set_reg_csr(struct kvm_vcpu *vcpu,
330 				      const struct kvm_one_reg *reg)
331 {
332 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
333 	unsigned long __user *uaddr =
334 			(unsigned long __user *)(unsigned long)reg->addr;
335 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
336 					    KVM_REG_SIZE_MASK |
337 					    KVM_REG_RISCV_CSR);
338 	unsigned long reg_val;
339 
340 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
341 		return -EINVAL;
342 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
343 		return -EINVAL;
344 
345 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
346 		return -EFAULT;
347 
348 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
349 		reg_val &= VSIP_VALID_MASK;
350 		reg_val <<= VSIP_TO_HVIP_SHIFT;
351 	}
352 
353 	((unsigned long *)csr)[reg_num] = reg_val;
354 
355 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip))
356 		WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
357 
358 	return 0;
359 }
360 
361 static int kvm_riscv_vcpu_set_reg(struct kvm_vcpu *vcpu,
362 				  const struct kvm_one_reg *reg)
363 {
364 	if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CONFIG)
365 		return kvm_riscv_vcpu_set_reg_config(vcpu, reg);
366 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CORE)
367 		return kvm_riscv_vcpu_set_reg_core(vcpu, reg);
368 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CSR)
369 		return kvm_riscv_vcpu_set_reg_csr(vcpu, reg);
370 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_TIMER)
371 		return kvm_riscv_vcpu_set_reg_timer(vcpu, reg);
372 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_F)
373 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
374 						 KVM_REG_RISCV_FP_F);
375 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_D)
376 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
377 						 KVM_REG_RISCV_FP_D);
378 
379 	return -EINVAL;
380 }
381 
382 static int kvm_riscv_vcpu_get_reg(struct kvm_vcpu *vcpu,
383 				  const struct kvm_one_reg *reg)
384 {
385 	if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CONFIG)
386 		return kvm_riscv_vcpu_get_reg_config(vcpu, reg);
387 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CORE)
388 		return kvm_riscv_vcpu_get_reg_core(vcpu, reg);
389 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CSR)
390 		return kvm_riscv_vcpu_get_reg_csr(vcpu, reg);
391 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_TIMER)
392 		return kvm_riscv_vcpu_get_reg_timer(vcpu, reg);
393 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_F)
394 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
395 						 KVM_REG_RISCV_FP_F);
396 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_D)
397 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
398 						 KVM_REG_RISCV_FP_D);
399 
400 	return -EINVAL;
401 }
402 
403 long kvm_arch_vcpu_async_ioctl(struct file *filp,
404 			       unsigned int ioctl, unsigned long arg)
405 {
406 	struct kvm_vcpu *vcpu = filp->private_data;
407 	void __user *argp = (void __user *)arg;
408 
409 	if (ioctl == KVM_INTERRUPT) {
410 		struct kvm_interrupt irq;
411 
412 		if (copy_from_user(&irq, argp, sizeof(irq)))
413 			return -EFAULT;
414 
415 		if (irq.irq == KVM_INTERRUPT_SET)
416 			return kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT);
417 		else
418 			return kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
419 	}
420 
421 	return -ENOIOCTLCMD;
422 }
423 
424 long kvm_arch_vcpu_ioctl(struct file *filp,
425 			 unsigned int ioctl, unsigned long arg)
426 {
427 	struct kvm_vcpu *vcpu = filp->private_data;
428 	void __user *argp = (void __user *)arg;
429 	long r = -EINVAL;
430 
431 	switch (ioctl) {
432 	case KVM_SET_ONE_REG:
433 	case KVM_GET_ONE_REG: {
434 		struct kvm_one_reg reg;
435 
436 		r = -EFAULT;
437 		if (copy_from_user(&reg, argp, sizeof(reg)))
438 			break;
439 
440 		if (ioctl == KVM_SET_ONE_REG)
441 			r = kvm_riscv_vcpu_set_reg(vcpu, &reg);
442 		else
443 			r = kvm_riscv_vcpu_get_reg(vcpu, &reg);
444 		break;
445 	}
446 	default:
447 		break;
448 	}
449 
450 	return r;
451 }
452 
453 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
454 				  struct kvm_sregs *sregs)
455 {
456 	return -EINVAL;
457 }
458 
459 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
460 				  struct kvm_sregs *sregs)
461 {
462 	return -EINVAL;
463 }
464 
465 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
466 {
467 	return -EINVAL;
468 }
469 
470 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
471 {
472 	return -EINVAL;
473 }
474 
475 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
476 				  struct kvm_translation *tr)
477 {
478 	return -EINVAL;
479 }
480 
481 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
482 {
483 	return -EINVAL;
484 }
485 
486 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
487 {
488 	return -EINVAL;
489 }
490 
491 void kvm_riscv_vcpu_flush_interrupts(struct kvm_vcpu *vcpu)
492 {
493 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
494 	unsigned long mask, val;
495 
496 	if (READ_ONCE(vcpu->arch.irqs_pending_mask)) {
497 		mask = xchg_acquire(&vcpu->arch.irqs_pending_mask, 0);
498 		val = READ_ONCE(vcpu->arch.irqs_pending) & mask;
499 
500 		csr->hvip &= ~mask;
501 		csr->hvip |= val;
502 	}
503 }
504 
505 void kvm_riscv_vcpu_sync_interrupts(struct kvm_vcpu *vcpu)
506 {
507 	unsigned long hvip;
508 	struct kvm_vcpu_arch *v = &vcpu->arch;
509 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
510 
511 	/* Read current HVIP and VSIE CSRs */
512 	csr->vsie = csr_read(CSR_VSIE);
513 
514 	/* Sync-up HVIP.VSSIP bit changes does by Guest */
515 	hvip = csr_read(CSR_HVIP);
516 	if ((csr->hvip ^ hvip) & (1UL << IRQ_VS_SOFT)) {
517 		if (hvip & (1UL << IRQ_VS_SOFT)) {
518 			if (!test_and_set_bit(IRQ_VS_SOFT,
519 					      &v->irqs_pending_mask))
520 				set_bit(IRQ_VS_SOFT, &v->irqs_pending);
521 		} else {
522 			if (!test_and_set_bit(IRQ_VS_SOFT,
523 					      &v->irqs_pending_mask))
524 				clear_bit(IRQ_VS_SOFT, &v->irqs_pending);
525 		}
526 	}
527 }
528 
529 int kvm_riscv_vcpu_set_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
530 {
531 	if (irq != IRQ_VS_SOFT &&
532 	    irq != IRQ_VS_TIMER &&
533 	    irq != IRQ_VS_EXT)
534 		return -EINVAL;
535 
536 	set_bit(irq, &vcpu->arch.irqs_pending);
537 	smp_mb__before_atomic();
538 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
539 
540 	kvm_vcpu_kick(vcpu);
541 
542 	return 0;
543 }
544 
545 int kvm_riscv_vcpu_unset_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
546 {
547 	if (irq != IRQ_VS_SOFT &&
548 	    irq != IRQ_VS_TIMER &&
549 	    irq != IRQ_VS_EXT)
550 		return -EINVAL;
551 
552 	clear_bit(irq, &vcpu->arch.irqs_pending);
553 	smp_mb__before_atomic();
554 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
555 
556 	return 0;
557 }
558 
559 bool kvm_riscv_vcpu_has_interrupts(struct kvm_vcpu *vcpu, unsigned long mask)
560 {
561 	unsigned long ie = ((vcpu->arch.guest_csr.vsie & VSIP_VALID_MASK)
562 			    << VSIP_TO_HVIP_SHIFT) & mask;
563 
564 	return (READ_ONCE(vcpu->arch.irqs_pending) & ie) ? true : false;
565 }
566 
567 void kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu)
568 {
569 	vcpu->arch.power_off = true;
570 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
571 	kvm_vcpu_kick(vcpu);
572 }
573 
574 void kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu)
575 {
576 	vcpu->arch.power_off = false;
577 	kvm_vcpu_wake_up(vcpu);
578 }
579 
580 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
581 				    struct kvm_mp_state *mp_state)
582 {
583 	if (vcpu->arch.power_off)
584 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
585 	else
586 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
587 
588 	return 0;
589 }
590 
591 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
592 				    struct kvm_mp_state *mp_state)
593 {
594 	int ret = 0;
595 
596 	switch (mp_state->mp_state) {
597 	case KVM_MP_STATE_RUNNABLE:
598 		vcpu->arch.power_off = false;
599 		break;
600 	case KVM_MP_STATE_STOPPED:
601 		kvm_riscv_vcpu_power_off(vcpu);
602 		break;
603 	default:
604 		ret = -EINVAL;
605 	}
606 
607 	return ret;
608 }
609 
610 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
611 					struct kvm_guest_debug *dbg)
612 {
613 	/* TODO; To be implemented later. */
614 	return -EINVAL;
615 }
616 
617 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
618 {
619 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
620 
621 	csr_write(CSR_VSSTATUS, csr->vsstatus);
622 	csr_write(CSR_VSIE, csr->vsie);
623 	csr_write(CSR_VSTVEC, csr->vstvec);
624 	csr_write(CSR_VSSCRATCH, csr->vsscratch);
625 	csr_write(CSR_VSEPC, csr->vsepc);
626 	csr_write(CSR_VSCAUSE, csr->vscause);
627 	csr_write(CSR_VSTVAL, csr->vstval);
628 	csr_write(CSR_HVIP, csr->hvip);
629 	csr_write(CSR_VSATP, csr->vsatp);
630 
631 	kvm_riscv_stage2_update_hgatp(vcpu);
632 
633 	kvm_riscv_vcpu_timer_restore(vcpu);
634 
635 	kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context);
636 	kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context,
637 					vcpu->arch.isa);
638 
639 	vcpu->cpu = cpu;
640 }
641 
642 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
643 {
644 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
645 
646 	vcpu->cpu = -1;
647 
648 	kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context,
649 				     vcpu->arch.isa);
650 	kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context);
651 
652 	csr_write(CSR_HGATP, 0);
653 
654 	csr->vsstatus = csr_read(CSR_VSSTATUS);
655 	csr->vsie = csr_read(CSR_VSIE);
656 	csr->vstvec = csr_read(CSR_VSTVEC);
657 	csr->vsscratch = csr_read(CSR_VSSCRATCH);
658 	csr->vsepc = csr_read(CSR_VSEPC);
659 	csr->vscause = csr_read(CSR_VSCAUSE);
660 	csr->vstval = csr_read(CSR_VSTVAL);
661 	csr->hvip = csr_read(CSR_HVIP);
662 	csr->vsatp = csr_read(CSR_VSATP);
663 }
664 
665 static void kvm_riscv_check_vcpu_requests(struct kvm_vcpu *vcpu)
666 {
667 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
668 
669 	if (kvm_request_pending(vcpu)) {
670 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) {
671 			rcuwait_wait_event(wait,
672 				(!vcpu->arch.power_off) && (!vcpu->arch.pause),
673 				TASK_INTERRUPTIBLE);
674 
675 			if (vcpu->arch.power_off || vcpu->arch.pause) {
676 				/*
677 				 * Awaken to handle a signal, request to
678 				 * sleep again later.
679 				 */
680 				kvm_make_request(KVM_REQ_SLEEP, vcpu);
681 			}
682 		}
683 
684 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
685 			kvm_riscv_reset_vcpu(vcpu);
686 
687 		if (kvm_check_request(KVM_REQ_UPDATE_HGATP, vcpu))
688 			kvm_riscv_stage2_update_hgatp(vcpu);
689 
690 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
691 			__kvm_riscv_hfence_gvma_all();
692 	}
693 }
694 
695 static void kvm_riscv_update_hvip(struct kvm_vcpu *vcpu)
696 {
697 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
698 
699 	csr_write(CSR_HVIP, csr->hvip);
700 }
701 
702 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
703 {
704 	int ret;
705 	struct kvm_cpu_trap trap;
706 	struct kvm_run *run = vcpu->run;
707 
708 	/* Mark this VCPU ran at least once */
709 	vcpu->arch.ran_atleast_once = true;
710 
711 	vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
712 
713 	/* Process MMIO value returned from user-space */
714 	if (run->exit_reason == KVM_EXIT_MMIO) {
715 		ret = kvm_riscv_vcpu_mmio_return(vcpu, vcpu->run);
716 		if (ret) {
717 			srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
718 			return ret;
719 		}
720 	}
721 
722 	/* Process SBI value returned from user-space */
723 	if (run->exit_reason == KVM_EXIT_RISCV_SBI) {
724 		ret = kvm_riscv_vcpu_sbi_return(vcpu, vcpu->run);
725 		if (ret) {
726 			srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
727 			return ret;
728 		}
729 	}
730 
731 	if (run->immediate_exit) {
732 		srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
733 		return -EINTR;
734 	}
735 
736 	vcpu_load(vcpu);
737 
738 	kvm_sigset_activate(vcpu);
739 
740 	ret = 1;
741 	run->exit_reason = KVM_EXIT_UNKNOWN;
742 	while (ret > 0) {
743 		/* Check conditions before entering the guest */
744 		cond_resched();
745 
746 		kvm_riscv_stage2_vmid_update(vcpu);
747 
748 		kvm_riscv_check_vcpu_requests(vcpu);
749 
750 		preempt_disable();
751 
752 		local_irq_disable();
753 
754 		/*
755 		 * Exit if we have a signal pending so that we can deliver
756 		 * the signal to user space.
757 		 */
758 		if (signal_pending(current)) {
759 			ret = -EINTR;
760 			run->exit_reason = KVM_EXIT_INTR;
761 		}
762 
763 		/*
764 		 * Ensure we set mode to IN_GUEST_MODE after we disable
765 		 * interrupts and before the final VCPU requests check.
766 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
767 		 * Documentation/virt/kvm/vcpu-requests.rst
768 		 */
769 		vcpu->mode = IN_GUEST_MODE;
770 
771 		srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
772 		smp_mb__after_srcu_read_unlock();
773 
774 		/*
775 		 * We might have got VCPU interrupts updated asynchronously
776 		 * so update it in HW.
777 		 */
778 		kvm_riscv_vcpu_flush_interrupts(vcpu);
779 
780 		/* Update HVIP CSR for current CPU */
781 		kvm_riscv_update_hvip(vcpu);
782 
783 		if (ret <= 0 ||
784 		    kvm_riscv_stage2_vmid_ver_changed(&vcpu->kvm->arch.vmid) ||
785 		    kvm_request_pending(vcpu)) {
786 			vcpu->mode = OUTSIDE_GUEST_MODE;
787 			local_irq_enable();
788 			preempt_enable();
789 			vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
790 			continue;
791 		}
792 
793 		guest_enter_irqoff();
794 
795 		__kvm_riscv_switch_to(&vcpu->arch);
796 
797 		vcpu->mode = OUTSIDE_GUEST_MODE;
798 		vcpu->stat.exits++;
799 
800 		/*
801 		 * Save SCAUSE, STVAL, HTVAL, and HTINST because we might
802 		 * get an interrupt between __kvm_riscv_switch_to() and
803 		 * local_irq_enable() which can potentially change CSRs.
804 		 */
805 		trap.sepc = vcpu->arch.guest_context.sepc;
806 		trap.scause = csr_read(CSR_SCAUSE);
807 		trap.stval = csr_read(CSR_STVAL);
808 		trap.htval = csr_read(CSR_HTVAL);
809 		trap.htinst = csr_read(CSR_HTINST);
810 
811 		/* Syncup interrupts state with HW */
812 		kvm_riscv_vcpu_sync_interrupts(vcpu);
813 
814 		/*
815 		 * We may have taken a host interrupt in VS/VU-mode (i.e.
816 		 * while executing the guest). This interrupt is still
817 		 * pending, as we haven't serviced it yet!
818 		 *
819 		 * We're now back in HS-mode with interrupts disabled
820 		 * so enabling the interrupts now will have the effect
821 		 * of taking the interrupt again, in HS-mode this time.
822 		 */
823 		local_irq_enable();
824 
825 		/*
826 		 * We do local_irq_enable() before calling guest_exit() so
827 		 * that if a timer interrupt hits while running the guest
828 		 * we account that tick as being spent in the guest. We
829 		 * enable preemption after calling guest_exit() so that if
830 		 * we get preempted we make sure ticks after that is not
831 		 * counted as guest time.
832 		 */
833 		guest_exit();
834 
835 		preempt_enable();
836 
837 		vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
838 
839 		ret = kvm_riscv_vcpu_exit(vcpu, run, &trap);
840 	}
841 
842 	kvm_sigset_deactivate(vcpu);
843 
844 	vcpu_put(vcpu);
845 
846 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
847 
848 	return ret;
849 }
850