xref: /openbmc/linux/arch/riscv/kvm/vcpu.c (revision 3ca3af7d)
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 
57 	memcpy(csr, reset_csr, sizeof(*csr));
58 
59 	memcpy(cntx, reset_cntx, sizeof(*cntx));
60 
61 	kvm_riscv_vcpu_fp_reset(vcpu);
62 
63 	kvm_riscv_vcpu_timer_reset(vcpu);
64 
65 	WRITE_ONCE(vcpu->arch.irqs_pending, 0);
66 	WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
67 }
68 
69 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
70 {
71 	return 0;
72 }
73 
74 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
75 {
76 	struct kvm_cpu_context *cntx;
77 
78 	/* Mark this VCPU never ran */
79 	vcpu->arch.ran_atleast_once = false;
80 
81 	/* Setup ISA features available to VCPU */
82 	vcpu->arch.isa = riscv_isa_extension_base(NULL) & KVM_RISCV_ISA_ALLOWED;
83 
84 	/* Setup reset state of shadow SSTATUS and HSTATUS CSRs */
85 	cntx = &vcpu->arch.guest_reset_context;
86 	cntx->sstatus = SR_SPP | SR_SPIE;
87 	cntx->hstatus = 0;
88 	cntx->hstatus |= HSTATUS_VTW;
89 	cntx->hstatus |= HSTATUS_SPVP;
90 	cntx->hstatus |= HSTATUS_SPV;
91 
92 	/* Setup VCPU timer */
93 	kvm_riscv_vcpu_timer_init(vcpu);
94 
95 	/* Reset VCPU */
96 	kvm_riscv_reset_vcpu(vcpu);
97 
98 	return 0;
99 }
100 
101 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
102 {
103 }
104 
105 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
106 {
107 	/* Cleanup VCPU timer */
108 	kvm_riscv_vcpu_timer_deinit(vcpu);
109 
110 	/* Flush the pages pre-allocated for Stage2 page table mappings */
111 	kvm_riscv_stage2_flush_cache(vcpu);
112 }
113 
114 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
115 {
116 	return kvm_riscv_vcpu_has_interrupts(vcpu, 1UL << IRQ_VS_TIMER);
117 }
118 
119 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
120 {
121 }
122 
123 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
124 {
125 }
126 
127 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
128 {
129 	return (kvm_riscv_vcpu_has_interrupts(vcpu, -1UL) &&
130 		!vcpu->arch.power_off && !vcpu->arch.pause);
131 }
132 
133 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
134 {
135 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
136 }
137 
138 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
139 {
140 	return (vcpu->arch.guest_context.sstatus & SR_SPP) ? true : false;
141 }
142 
143 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
144 {
145 	return VM_FAULT_SIGBUS;
146 }
147 
148 static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
149 					 const struct kvm_one_reg *reg)
150 {
151 	unsigned long __user *uaddr =
152 			(unsigned long __user *)(unsigned long)reg->addr;
153 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
154 					    KVM_REG_SIZE_MASK |
155 					    KVM_REG_RISCV_CONFIG);
156 	unsigned long reg_val;
157 
158 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
159 		return -EINVAL;
160 
161 	switch (reg_num) {
162 	case KVM_REG_RISCV_CONFIG_REG(isa):
163 		reg_val = vcpu->arch.isa;
164 		break;
165 	default:
166 		return -EINVAL;
167 	}
168 
169 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
170 		return -EFAULT;
171 
172 	return 0;
173 }
174 
175 static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
176 					 const struct kvm_one_reg *reg)
177 {
178 	unsigned long __user *uaddr =
179 			(unsigned long __user *)(unsigned long)reg->addr;
180 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
181 					    KVM_REG_SIZE_MASK |
182 					    KVM_REG_RISCV_CONFIG);
183 	unsigned long reg_val;
184 
185 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
186 		return -EINVAL;
187 
188 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
189 		return -EFAULT;
190 
191 	switch (reg_num) {
192 	case KVM_REG_RISCV_CONFIG_REG(isa):
193 		if (!vcpu->arch.ran_atleast_once) {
194 			vcpu->arch.isa = reg_val;
195 			vcpu->arch.isa &= riscv_isa_extension_base(NULL);
196 			vcpu->arch.isa &= KVM_RISCV_ISA_ALLOWED;
197 			kvm_riscv_vcpu_fp_reset(vcpu);
198 		} else {
199 			return -EOPNOTSUPP;
200 		}
201 		break;
202 	default:
203 		return -EINVAL;
204 	}
205 
206 	return 0;
207 }
208 
209 static int kvm_riscv_vcpu_get_reg_core(struct kvm_vcpu *vcpu,
210 				       const struct kvm_one_reg *reg)
211 {
212 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
213 	unsigned long __user *uaddr =
214 			(unsigned long __user *)(unsigned long)reg->addr;
215 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
216 					    KVM_REG_SIZE_MASK |
217 					    KVM_REG_RISCV_CORE);
218 	unsigned long reg_val;
219 
220 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
221 		return -EINVAL;
222 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
223 		return -EINVAL;
224 
225 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
226 		reg_val = cntx->sepc;
227 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
228 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
229 		reg_val = ((unsigned long *)cntx)[reg_num];
230 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode))
231 		reg_val = (cntx->sstatus & SR_SPP) ?
232 				KVM_RISCV_MODE_S : KVM_RISCV_MODE_U;
233 	else
234 		return -EINVAL;
235 
236 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
237 		return -EFAULT;
238 
239 	return 0;
240 }
241 
242 static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
243 				       const struct kvm_one_reg *reg)
244 {
245 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
246 	unsigned long __user *uaddr =
247 			(unsigned long __user *)(unsigned long)reg->addr;
248 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
249 					    KVM_REG_SIZE_MASK |
250 					    KVM_REG_RISCV_CORE);
251 	unsigned long reg_val;
252 
253 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
254 		return -EINVAL;
255 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
256 		return -EINVAL;
257 
258 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
259 		return -EFAULT;
260 
261 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
262 		cntx->sepc = reg_val;
263 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
264 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
265 		((unsigned long *)cntx)[reg_num] = reg_val;
266 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode)) {
267 		if (reg_val == KVM_RISCV_MODE_S)
268 			cntx->sstatus |= SR_SPP;
269 		else
270 			cntx->sstatus &= ~SR_SPP;
271 	} else
272 		return -EINVAL;
273 
274 	return 0;
275 }
276 
277 static int kvm_riscv_vcpu_get_reg_csr(struct kvm_vcpu *vcpu,
278 				      const struct kvm_one_reg *reg)
279 {
280 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
281 	unsigned long __user *uaddr =
282 			(unsigned long __user *)(unsigned long)reg->addr;
283 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
284 					    KVM_REG_SIZE_MASK |
285 					    KVM_REG_RISCV_CSR);
286 	unsigned long reg_val;
287 
288 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
289 		return -EINVAL;
290 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
291 		return -EINVAL;
292 
293 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
294 		kvm_riscv_vcpu_flush_interrupts(vcpu);
295 		reg_val = (csr->hvip >> VSIP_TO_HVIP_SHIFT) & VSIP_VALID_MASK;
296 	} else
297 		reg_val = ((unsigned long *)csr)[reg_num];
298 
299 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
300 		return -EFAULT;
301 
302 	return 0;
303 }
304 
305 static int kvm_riscv_vcpu_set_reg_csr(struct kvm_vcpu *vcpu,
306 				      const struct kvm_one_reg *reg)
307 {
308 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
309 	unsigned long __user *uaddr =
310 			(unsigned long __user *)(unsigned long)reg->addr;
311 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
312 					    KVM_REG_SIZE_MASK |
313 					    KVM_REG_RISCV_CSR);
314 	unsigned long reg_val;
315 
316 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
317 		return -EINVAL;
318 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
319 		return -EINVAL;
320 
321 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
322 		return -EFAULT;
323 
324 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
325 		reg_val &= VSIP_VALID_MASK;
326 		reg_val <<= VSIP_TO_HVIP_SHIFT;
327 	}
328 
329 	((unsigned long *)csr)[reg_num] = reg_val;
330 
331 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip))
332 		WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
333 
334 	return 0;
335 }
336 
337 static int kvm_riscv_vcpu_set_reg(struct kvm_vcpu *vcpu,
338 				  const struct kvm_one_reg *reg)
339 {
340 	if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CONFIG)
341 		return kvm_riscv_vcpu_set_reg_config(vcpu, reg);
342 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CORE)
343 		return kvm_riscv_vcpu_set_reg_core(vcpu, reg);
344 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CSR)
345 		return kvm_riscv_vcpu_set_reg_csr(vcpu, reg);
346 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_TIMER)
347 		return kvm_riscv_vcpu_set_reg_timer(vcpu, reg);
348 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_F)
349 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
350 						 KVM_REG_RISCV_FP_F);
351 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_D)
352 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
353 						 KVM_REG_RISCV_FP_D);
354 
355 	return -EINVAL;
356 }
357 
358 static int kvm_riscv_vcpu_get_reg(struct kvm_vcpu *vcpu,
359 				  const struct kvm_one_reg *reg)
360 {
361 	if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CONFIG)
362 		return kvm_riscv_vcpu_get_reg_config(vcpu, reg);
363 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CORE)
364 		return kvm_riscv_vcpu_get_reg_core(vcpu, reg);
365 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CSR)
366 		return kvm_riscv_vcpu_get_reg_csr(vcpu, reg);
367 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_TIMER)
368 		return kvm_riscv_vcpu_get_reg_timer(vcpu, reg);
369 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_F)
370 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
371 						 KVM_REG_RISCV_FP_F);
372 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_D)
373 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
374 						 KVM_REG_RISCV_FP_D);
375 
376 	return -EINVAL;
377 }
378 
379 long kvm_arch_vcpu_async_ioctl(struct file *filp,
380 			       unsigned int ioctl, unsigned long arg)
381 {
382 	struct kvm_vcpu *vcpu = filp->private_data;
383 	void __user *argp = (void __user *)arg;
384 
385 	if (ioctl == KVM_INTERRUPT) {
386 		struct kvm_interrupt irq;
387 
388 		if (copy_from_user(&irq, argp, sizeof(irq)))
389 			return -EFAULT;
390 
391 		if (irq.irq == KVM_INTERRUPT_SET)
392 			return kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT);
393 		else
394 			return kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
395 	}
396 
397 	return -ENOIOCTLCMD;
398 }
399 
400 long kvm_arch_vcpu_ioctl(struct file *filp,
401 			 unsigned int ioctl, unsigned long arg)
402 {
403 	struct kvm_vcpu *vcpu = filp->private_data;
404 	void __user *argp = (void __user *)arg;
405 	long r = -EINVAL;
406 
407 	switch (ioctl) {
408 	case KVM_SET_ONE_REG:
409 	case KVM_GET_ONE_REG: {
410 		struct kvm_one_reg reg;
411 
412 		r = -EFAULT;
413 		if (copy_from_user(&reg, argp, sizeof(reg)))
414 			break;
415 
416 		if (ioctl == KVM_SET_ONE_REG)
417 			r = kvm_riscv_vcpu_set_reg(vcpu, &reg);
418 		else
419 			r = kvm_riscv_vcpu_get_reg(vcpu, &reg);
420 		break;
421 	}
422 	default:
423 		break;
424 	}
425 
426 	return r;
427 }
428 
429 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
430 				  struct kvm_sregs *sregs)
431 {
432 	return -EINVAL;
433 }
434 
435 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
436 				  struct kvm_sregs *sregs)
437 {
438 	return -EINVAL;
439 }
440 
441 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
442 {
443 	return -EINVAL;
444 }
445 
446 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
447 {
448 	return -EINVAL;
449 }
450 
451 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
452 				  struct kvm_translation *tr)
453 {
454 	return -EINVAL;
455 }
456 
457 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
458 {
459 	return -EINVAL;
460 }
461 
462 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
463 {
464 	return -EINVAL;
465 }
466 
467 void kvm_riscv_vcpu_flush_interrupts(struct kvm_vcpu *vcpu)
468 {
469 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
470 	unsigned long mask, val;
471 
472 	if (READ_ONCE(vcpu->arch.irqs_pending_mask)) {
473 		mask = xchg_acquire(&vcpu->arch.irqs_pending_mask, 0);
474 		val = READ_ONCE(vcpu->arch.irqs_pending) & mask;
475 
476 		csr->hvip &= ~mask;
477 		csr->hvip |= val;
478 	}
479 }
480 
481 void kvm_riscv_vcpu_sync_interrupts(struct kvm_vcpu *vcpu)
482 {
483 	unsigned long hvip;
484 	struct kvm_vcpu_arch *v = &vcpu->arch;
485 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
486 
487 	/* Read current HVIP and VSIE CSRs */
488 	csr->vsie = csr_read(CSR_VSIE);
489 
490 	/* Sync-up HVIP.VSSIP bit changes does by Guest */
491 	hvip = csr_read(CSR_HVIP);
492 	if ((csr->hvip ^ hvip) & (1UL << IRQ_VS_SOFT)) {
493 		if (hvip & (1UL << IRQ_VS_SOFT)) {
494 			if (!test_and_set_bit(IRQ_VS_SOFT,
495 					      &v->irqs_pending_mask))
496 				set_bit(IRQ_VS_SOFT, &v->irqs_pending);
497 		} else {
498 			if (!test_and_set_bit(IRQ_VS_SOFT,
499 					      &v->irqs_pending_mask))
500 				clear_bit(IRQ_VS_SOFT, &v->irqs_pending);
501 		}
502 	}
503 }
504 
505 int kvm_riscv_vcpu_set_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
506 {
507 	if (irq != IRQ_VS_SOFT &&
508 	    irq != IRQ_VS_TIMER &&
509 	    irq != IRQ_VS_EXT)
510 		return -EINVAL;
511 
512 	set_bit(irq, &vcpu->arch.irqs_pending);
513 	smp_mb__before_atomic();
514 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
515 
516 	kvm_vcpu_kick(vcpu);
517 
518 	return 0;
519 }
520 
521 int kvm_riscv_vcpu_unset_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
522 {
523 	if (irq != IRQ_VS_SOFT &&
524 	    irq != IRQ_VS_TIMER &&
525 	    irq != IRQ_VS_EXT)
526 		return -EINVAL;
527 
528 	clear_bit(irq, &vcpu->arch.irqs_pending);
529 	smp_mb__before_atomic();
530 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
531 
532 	return 0;
533 }
534 
535 bool kvm_riscv_vcpu_has_interrupts(struct kvm_vcpu *vcpu, unsigned long mask)
536 {
537 	unsigned long ie = ((vcpu->arch.guest_csr.vsie & VSIP_VALID_MASK)
538 			    << VSIP_TO_HVIP_SHIFT) & mask;
539 
540 	return (READ_ONCE(vcpu->arch.irqs_pending) & ie) ? true : false;
541 }
542 
543 void kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu)
544 {
545 	vcpu->arch.power_off = true;
546 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
547 	kvm_vcpu_kick(vcpu);
548 }
549 
550 void kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu)
551 {
552 	vcpu->arch.power_off = false;
553 	kvm_vcpu_wake_up(vcpu);
554 }
555 
556 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
557 				    struct kvm_mp_state *mp_state)
558 {
559 	if (vcpu->arch.power_off)
560 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
561 	else
562 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
563 
564 	return 0;
565 }
566 
567 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
568 				    struct kvm_mp_state *mp_state)
569 {
570 	int ret = 0;
571 
572 	switch (mp_state->mp_state) {
573 	case KVM_MP_STATE_RUNNABLE:
574 		vcpu->arch.power_off = false;
575 		break;
576 	case KVM_MP_STATE_STOPPED:
577 		kvm_riscv_vcpu_power_off(vcpu);
578 		break;
579 	default:
580 		ret = -EINVAL;
581 	}
582 
583 	return ret;
584 }
585 
586 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
587 					struct kvm_guest_debug *dbg)
588 {
589 	/* TODO; To be implemented later. */
590 	return -EINVAL;
591 }
592 
593 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
594 {
595 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
596 
597 	csr_write(CSR_VSSTATUS, csr->vsstatus);
598 	csr_write(CSR_VSIE, csr->vsie);
599 	csr_write(CSR_VSTVEC, csr->vstvec);
600 	csr_write(CSR_VSSCRATCH, csr->vsscratch);
601 	csr_write(CSR_VSEPC, csr->vsepc);
602 	csr_write(CSR_VSCAUSE, csr->vscause);
603 	csr_write(CSR_VSTVAL, csr->vstval);
604 	csr_write(CSR_HVIP, csr->hvip);
605 	csr_write(CSR_VSATP, csr->vsatp);
606 
607 	kvm_riscv_stage2_update_hgatp(vcpu);
608 
609 	kvm_riscv_vcpu_timer_restore(vcpu);
610 
611 	kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context);
612 	kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context,
613 					vcpu->arch.isa);
614 
615 	vcpu->cpu = cpu;
616 }
617 
618 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
619 {
620 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
621 
622 	vcpu->cpu = -1;
623 
624 	kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context,
625 				     vcpu->arch.isa);
626 	kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context);
627 
628 	csr_write(CSR_HGATP, 0);
629 
630 	csr->vsstatus = csr_read(CSR_VSSTATUS);
631 	csr->vsie = csr_read(CSR_VSIE);
632 	csr->vstvec = csr_read(CSR_VSTVEC);
633 	csr->vsscratch = csr_read(CSR_VSSCRATCH);
634 	csr->vsepc = csr_read(CSR_VSEPC);
635 	csr->vscause = csr_read(CSR_VSCAUSE);
636 	csr->vstval = csr_read(CSR_VSTVAL);
637 	csr->hvip = csr_read(CSR_HVIP);
638 	csr->vsatp = csr_read(CSR_VSATP);
639 }
640 
641 static void kvm_riscv_check_vcpu_requests(struct kvm_vcpu *vcpu)
642 {
643 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
644 
645 	if (kvm_request_pending(vcpu)) {
646 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) {
647 			rcuwait_wait_event(wait,
648 				(!vcpu->arch.power_off) && (!vcpu->arch.pause),
649 				TASK_INTERRUPTIBLE);
650 
651 			if (vcpu->arch.power_off || vcpu->arch.pause) {
652 				/*
653 				 * Awaken to handle a signal, request to
654 				 * sleep again later.
655 				 */
656 				kvm_make_request(KVM_REQ_SLEEP, vcpu);
657 			}
658 		}
659 
660 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
661 			kvm_riscv_reset_vcpu(vcpu);
662 
663 		if (kvm_check_request(KVM_REQ_UPDATE_HGATP, vcpu))
664 			kvm_riscv_stage2_update_hgatp(vcpu);
665 
666 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
667 			__kvm_riscv_hfence_gvma_all();
668 	}
669 }
670 
671 static void kvm_riscv_update_hvip(struct kvm_vcpu *vcpu)
672 {
673 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
674 
675 	csr_write(CSR_HVIP, csr->hvip);
676 }
677 
678 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
679 {
680 	int ret;
681 	struct kvm_cpu_trap trap;
682 	struct kvm_run *run = vcpu->run;
683 
684 	/* Mark this VCPU ran at least once */
685 	vcpu->arch.ran_atleast_once = true;
686 
687 	vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
688 
689 	/* Process MMIO value returned from user-space */
690 	if (run->exit_reason == KVM_EXIT_MMIO) {
691 		ret = kvm_riscv_vcpu_mmio_return(vcpu, vcpu->run);
692 		if (ret) {
693 			srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
694 			return ret;
695 		}
696 	}
697 
698 	/* Process SBI value returned from user-space */
699 	if (run->exit_reason == KVM_EXIT_RISCV_SBI) {
700 		ret = kvm_riscv_vcpu_sbi_return(vcpu, vcpu->run);
701 		if (ret) {
702 			srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
703 			return ret;
704 		}
705 	}
706 
707 	if (run->immediate_exit) {
708 		srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
709 		return -EINTR;
710 	}
711 
712 	vcpu_load(vcpu);
713 
714 	kvm_sigset_activate(vcpu);
715 
716 	ret = 1;
717 	run->exit_reason = KVM_EXIT_UNKNOWN;
718 	while (ret > 0) {
719 		/* Check conditions before entering the guest */
720 		cond_resched();
721 
722 		kvm_riscv_stage2_vmid_update(vcpu);
723 
724 		kvm_riscv_check_vcpu_requests(vcpu);
725 
726 		preempt_disable();
727 
728 		local_irq_disable();
729 
730 		/*
731 		 * Exit if we have a signal pending so that we can deliver
732 		 * the signal to user space.
733 		 */
734 		if (signal_pending(current)) {
735 			ret = -EINTR;
736 			run->exit_reason = KVM_EXIT_INTR;
737 		}
738 
739 		/*
740 		 * Ensure we set mode to IN_GUEST_MODE after we disable
741 		 * interrupts and before the final VCPU requests check.
742 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
743 		 * Documentation/virtual/kvm/vcpu-requests.rst
744 		 */
745 		vcpu->mode = IN_GUEST_MODE;
746 
747 		srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
748 		smp_mb__after_srcu_read_unlock();
749 
750 		/*
751 		 * We might have got VCPU interrupts updated asynchronously
752 		 * so update it in HW.
753 		 */
754 		kvm_riscv_vcpu_flush_interrupts(vcpu);
755 
756 		/* Update HVIP CSR for current CPU */
757 		kvm_riscv_update_hvip(vcpu);
758 
759 		if (ret <= 0 ||
760 		    kvm_riscv_stage2_vmid_ver_changed(&vcpu->kvm->arch.vmid) ||
761 		    kvm_request_pending(vcpu)) {
762 			vcpu->mode = OUTSIDE_GUEST_MODE;
763 			local_irq_enable();
764 			preempt_enable();
765 			vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
766 			continue;
767 		}
768 
769 		guest_enter_irqoff();
770 
771 		__kvm_riscv_switch_to(&vcpu->arch);
772 
773 		vcpu->mode = OUTSIDE_GUEST_MODE;
774 		vcpu->stat.exits++;
775 
776 		/*
777 		 * Save SCAUSE, STVAL, HTVAL, and HTINST because we might
778 		 * get an interrupt between __kvm_riscv_switch_to() and
779 		 * local_irq_enable() which can potentially change CSRs.
780 		 */
781 		trap.sepc = vcpu->arch.guest_context.sepc;
782 		trap.scause = csr_read(CSR_SCAUSE);
783 		trap.stval = csr_read(CSR_STVAL);
784 		trap.htval = csr_read(CSR_HTVAL);
785 		trap.htinst = csr_read(CSR_HTINST);
786 
787 		/* Syncup interrupts state with HW */
788 		kvm_riscv_vcpu_sync_interrupts(vcpu);
789 
790 		/*
791 		 * We may have taken a host interrupt in VS/VU-mode (i.e.
792 		 * while executing the guest). This interrupt is still
793 		 * pending, as we haven't serviced it yet!
794 		 *
795 		 * We're now back in HS-mode with interrupts disabled
796 		 * so enabling the interrupts now will have the effect
797 		 * of taking the interrupt again, in HS-mode this time.
798 		 */
799 		local_irq_enable();
800 
801 		/*
802 		 * We do local_irq_enable() before calling guest_exit() so
803 		 * that if a timer interrupt hits while running the guest
804 		 * we account that tick as being spent in the guest. We
805 		 * enable preemption after calling guest_exit() so that if
806 		 * we get preempted we make sure ticks after that is not
807 		 * counted as guest time.
808 		 */
809 		guest_exit();
810 
811 		preempt_enable();
812 
813 		vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
814 
815 		ret = kvm_riscv_vcpu_exit(vcpu, run, &trap);
816 	}
817 
818 	kvm_sigset_deactivate(vcpu);
819 
820 	vcpu_put(vcpu);
821 
822 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
823 
824 	return ret;
825 }
826