xref: /openbmc/linux/arch/riscv/kvm/vcpu.c (revision 99a15348)
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 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
94 
95 	/* Mark this VCPU never ran */
96 	vcpu->arch.ran_atleast_once = false;
97 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
98 
99 	/* Setup ISA features available to VCPU */
100 	vcpu->arch.isa = riscv_isa_extension_base(NULL) & KVM_RISCV_ISA_ALLOWED;
101 
102 	/* Setup reset state of shadow SSTATUS and HSTATUS CSRs */
103 	cntx = &vcpu->arch.guest_reset_context;
104 	cntx->sstatus = SR_SPP | SR_SPIE;
105 	cntx->hstatus = 0;
106 	cntx->hstatus |= HSTATUS_VTW;
107 	cntx->hstatus |= HSTATUS_SPVP;
108 	cntx->hstatus |= HSTATUS_SPV;
109 
110 	/* By default, make CY, TM, and IR counters accessible in VU mode */
111 	reset_csr->scounteren = 0x7;
112 
113 	/* Setup VCPU timer */
114 	kvm_riscv_vcpu_timer_init(vcpu);
115 
116 	/* Reset VCPU */
117 	kvm_riscv_reset_vcpu(vcpu);
118 
119 	return 0;
120 }
121 
122 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
123 {
124 	/**
125 	 * vcpu with id 0 is the designated boot cpu.
126 	 * Keep all vcpus with non-zero id in power-off state so that
127 	 * they can be brought up using SBI HSM extension.
128 	 */
129 	if (vcpu->vcpu_idx != 0)
130 		kvm_riscv_vcpu_power_off(vcpu);
131 }
132 
133 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
134 {
135 	/* Cleanup VCPU timer */
136 	kvm_riscv_vcpu_timer_deinit(vcpu);
137 
138 	/* Free unused pages pre-allocated for Stage2 page table mappings */
139 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
140 }
141 
142 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
143 {
144 	return kvm_riscv_vcpu_has_interrupts(vcpu, 1UL << IRQ_VS_TIMER);
145 }
146 
147 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
148 {
149 }
150 
151 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
152 {
153 }
154 
155 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
156 {
157 	return (kvm_riscv_vcpu_has_interrupts(vcpu, -1UL) &&
158 		!vcpu->arch.power_off && !vcpu->arch.pause);
159 }
160 
161 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
162 {
163 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
164 }
165 
166 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
167 {
168 	return (vcpu->arch.guest_context.sstatus & SR_SPP) ? true : false;
169 }
170 
171 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
172 {
173 	return VM_FAULT_SIGBUS;
174 }
175 
176 static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
177 					 const struct kvm_one_reg *reg)
178 {
179 	unsigned long __user *uaddr =
180 			(unsigned long __user *)(unsigned long)reg->addr;
181 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
182 					    KVM_REG_SIZE_MASK |
183 					    KVM_REG_RISCV_CONFIG);
184 	unsigned long reg_val;
185 
186 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
187 		return -EINVAL;
188 
189 	switch (reg_num) {
190 	case KVM_REG_RISCV_CONFIG_REG(isa):
191 		reg_val = vcpu->arch.isa;
192 		break;
193 	default:
194 		return -EINVAL;
195 	}
196 
197 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
198 		return -EFAULT;
199 
200 	return 0;
201 }
202 
203 static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
204 					 const struct kvm_one_reg *reg)
205 {
206 	unsigned long __user *uaddr =
207 			(unsigned long __user *)(unsigned long)reg->addr;
208 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
209 					    KVM_REG_SIZE_MASK |
210 					    KVM_REG_RISCV_CONFIG);
211 	unsigned long reg_val;
212 
213 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
214 		return -EINVAL;
215 
216 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
217 		return -EFAULT;
218 
219 	switch (reg_num) {
220 	case KVM_REG_RISCV_CONFIG_REG(isa):
221 		if (!vcpu->arch.ran_atleast_once) {
222 			vcpu->arch.isa = reg_val;
223 			vcpu->arch.isa &= riscv_isa_extension_base(NULL);
224 			vcpu->arch.isa &= KVM_RISCV_ISA_ALLOWED;
225 			kvm_riscv_vcpu_fp_reset(vcpu);
226 		} else {
227 			return -EOPNOTSUPP;
228 		}
229 		break;
230 	default:
231 		return -EINVAL;
232 	}
233 
234 	return 0;
235 }
236 
237 static int kvm_riscv_vcpu_get_reg_core(struct kvm_vcpu *vcpu,
238 				       const struct kvm_one_reg *reg)
239 {
240 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
241 	unsigned long __user *uaddr =
242 			(unsigned long __user *)(unsigned long)reg->addr;
243 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
244 					    KVM_REG_SIZE_MASK |
245 					    KVM_REG_RISCV_CORE);
246 	unsigned long reg_val;
247 
248 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
249 		return -EINVAL;
250 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
251 		return -EINVAL;
252 
253 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
254 		reg_val = cntx->sepc;
255 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
256 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
257 		reg_val = ((unsigned long *)cntx)[reg_num];
258 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode))
259 		reg_val = (cntx->sstatus & SR_SPP) ?
260 				KVM_RISCV_MODE_S : KVM_RISCV_MODE_U;
261 	else
262 		return -EINVAL;
263 
264 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
265 		return -EFAULT;
266 
267 	return 0;
268 }
269 
270 static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
271 				       const struct kvm_one_reg *reg)
272 {
273 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
274 	unsigned long __user *uaddr =
275 			(unsigned long __user *)(unsigned long)reg->addr;
276 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
277 					    KVM_REG_SIZE_MASK |
278 					    KVM_REG_RISCV_CORE);
279 	unsigned long reg_val;
280 
281 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
282 		return -EINVAL;
283 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
284 		return -EINVAL;
285 
286 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
287 		return -EFAULT;
288 
289 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
290 		cntx->sepc = reg_val;
291 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
292 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
293 		((unsigned long *)cntx)[reg_num] = reg_val;
294 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode)) {
295 		if (reg_val == KVM_RISCV_MODE_S)
296 			cntx->sstatus |= SR_SPP;
297 		else
298 			cntx->sstatus &= ~SR_SPP;
299 	} else
300 		return -EINVAL;
301 
302 	return 0;
303 }
304 
305 static int kvm_riscv_vcpu_get_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 (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
322 		kvm_riscv_vcpu_flush_interrupts(vcpu);
323 		reg_val = (csr->hvip >> VSIP_TO_HVIP_SHIFT) & VSIP_VALID_MASK;
324 	} else
325 		reg_val = ((unsigned long *)csr)[reg_num];
326 
327 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
328 		return -EFAULT;
329 
330 	return 0;
331 }
332 
333 static int kvm_riscv_vcpu_set_reg_csr(struct kvm_vcpu *vcpu,
334 				      const struct kvm_one_reg *reg)
335 {
336 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
337 	unsigned long __user *uaddr =
338 			(unsigned long __user *)(unsigned long)reg->addr;
339 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
340 					    KVM_REG_SIZE_MASK |
341 					    KVM_REG_RISCV_CSR);
342 	unsigned long reg_val;
343 
344 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
345 		return -EINVAL;
346 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
347 		return -EINVAL;
348 
349 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
350 		return -EFAULT;
351 
352 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
353 		reg_val &= VSIP_VALID_MASK;
354 		reg_val <<= VSIP_TO_HVIP_SHIFT;
355 	}
356 
357 	((unsigned long *)csr)[reg_num] = reg_val;
358 
359 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip))
360 		WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
361 
362 	return 0;
363 }
364 
365 static int kvm_riscv_vcpu_set_reg(struct kvm_vcpu *vcpu,
366 				  const struct kvm_one_reg *reg)
367 {
368 	if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CONFIG)
369 		return kvm_riscv_vcpu_set_reg_config(vcpu, reg);
370 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CORE)
371 		return kvm_riscv_vcpu_set_reg_core(vcpu, reg);
372 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CSR)
373 		return kvm_riscv_vcpu_set_reg_csr(vcpu, reg);
374 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_TIMER)
375 		return kvm_riscv_vcpu_set_reg_timer(vcpu, reg);
376 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_F)
377 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
378 						 KVM_REG_RISCV_FP_F);
379 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_D)
380 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
381 						 KVM_REG_RISCV_FP_D);
382 
383 	return -EINVAL;
384 }
385 
386 static int kvm_riscv_vcpu_get_reg(struct kvm_vcpu *vcpu,
387 				  const struct kvm_one_reg *reg)
388 {
389 	if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CONFIG)
390 		return kvm_riscv_vcpu_get_reg_config(vcpu, reg);
391 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CORE)
392 		return kvm_riscv_vcpu_get_reg_core(vcpu, reg);
393 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_CSR)
394 		return kvm_riscv_vcpu_get_reg_csr(vcpu, reg);
395 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_TIMER)
396 		return kvm_riscv_vcpu_get_reg_timer(vcpu, reg);
397 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_F)
398 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
399 						 KVM_REG_RISCV_FP_F);
400 	else if ((reg->id & KVM_REG_RISCV_TYPE_MASK) == KVM_REG_RISCV_FP_D)
401 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
402 						 KVM_REG_RISCV_FP_D);
403 
404 	return -EINVAL;
405 }
406 
407 long kvm_arch_vcpu_async_ioctl(struct file *filp,
408 			       unsigned int ioctl, unsigned long arg)
409 {
410 	struct kvm_vcpu *vcpu = filp->private_data;
411 	void __user *argp = (void __user *)arg;
412 
413 	if (ioctl == KVM_INTERRUPT) {
414 		struct kvm_interrupt irq;
415 
416 		if (copy_from_user(&irq, argp, sizeof(irq)))
417 			return -EFAULT;
418 
419 		if (irq.irq == KVM_INTERRUPT_SET)
420 			return kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT);
421 		else
422 			return kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
423 	}
424 
425 	return -ENOIOCTLCMD;
426 }
427 
428 long kvm_arch_vcpu_ioctl(struct file *filp,
429 			 unsigned int ioctl, unsigned long arg)
430 {
431 	struct kvm_vcpu *vcpu = filp->private_data;
432 	void __user *argp = (void __user *)arg;
433 	long r = -EINVAL;
434 
435 	switch (ioctl) {
436 	case KVM_SET_ONE_REG:
437 	case KVM_GET_ONE_REG: {
438 		struct kvm_one_reg reg;
439 
440 		r = -EFAULT;
441 		if (copy_from_user(&reg, argp, sizeof(reg)))
442 			break;
443 
444 		if (ioctl == KVM_SET_ONE_REG)
445 			r = kvm_riscv_vcpu_set_reg(vcpu, &reg);
446 		else
447 			r = kvm_riscv_vcpu_get_reg(vcpu, &reg);
448 		break;
449 	}
450 	default:
451 		break;
452 	}
453 
454 	return r;
455 }
456 
457 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
458 				  struct kvm_sregs *sregs)
459 {
460 	return -EINVAL;
461 }
462 
463 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
464 				  struct kvm_sregs *sregs)
465 {
466 	return -EINVAL;
467 }
468 
469 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
470 {
471 	return -EINVAL;
472 }
473 
474 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
475 {
476 	return -EINVAL;
477 }
478 
479 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
480 				  struct kvm_translation *tr)
481 {
482 	return -EINVAL;
483 }
484 
485 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
486 {
487 	return -EINVAL;
488 }
489 
490 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
491 {
492 	return -EINVAL;
493 }
494 
495 void kvm_riscv_vcpu_flush_interrupts(struct kvm_vcpu *vcpu)
496 {
497 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
498 	unsigned long mask, val;
499 
500 	if (READ_ONCE(vcpu->arch.irqs_pending_mask)) {
501 		mask = xchg_acquire(&vcpu->arch.irqs_pending_mask, 0);
502 		val = READ_ONCE(vcpu->arch.irqs_pending) & mask;
503 
504 		csr->hvip &= ~mask;
505 		csr->hvip |= val;
506 	}
507 }
508 
509 void kvm_riscv_vcpu_sync_interrupts(struct kvm_vcpu *vcpu)
510 {
511 	unsigned long hvip;
512 	struct kvm_vcpu_arch *v = &vcpu->arch;
513 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
514 
515 	/* Read current HVIP and VSIE CSRs */
516 	csr->vsie = csr_read(CSR_VSIE);
517 
518 	/* Sync-up HVIP.VSSIP bit changes does by Guest */
519 	hvip = csr_read(CSR_HVIP);
520 	if ((csr->hvip ^ hvip) & (1UL << IRQ_VS_SOFT)) {
521 		if (hvip & (1UL << IRQ_VS_SOFT)) {
522 			if (!test_and_set_bit(IRQ_VS_SOFT,
523 					      &v->irqs_pending_mask))
524 				set_bit(IRQ_VS_SOFT, &v->irqs_pending);
525 		} else {
526 			if (!test_and_set_bit(IRQ_VS_SOFT,
527 					      &v->irqs_pending_mask))
528 				clear_bit(IRQ_VS_SOFT, &v->irqs_pending);
529 		}
530 	}
531 }
532 
533 int kvm_riscv_vcpu_set_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
534 {
535 	if (irq != IRQ_VS_SOFT &&
536 	    irq != IRQ_VS_TIMER &&
537 	    irq != IRQ_VS_EXT)
538 		return -EINVAL;
539 
540 	set_bit(irq, &vcpu->arch.irqs_pending);
541 	smp_mb__before_atomic();
542 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
543 
544 	kvm_vcpu_kick(vcpu);
545 
546 	return 0;
547 }
548 
549 int kvm_riscv_vcpu_unset_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
550 {
551 	if (irq != IRQ_VS_SOFT &&
552 	    irq != IRQ_VS_TIMER &&
553 	    irq != IRQ_VS_EXT)
554 		return -EINVAL;
555 
556 	clear_bit(irq, &vcpu->arch.irqs_pending);
557 	smp_mb__before_atomic();
558 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
559 
560 	return 0;
561 }
562 
563 bool kvm_riscv_vcpu_has_interrupts(struct kvm_vcpu *vcpu, unsigned long mask)
564 {
565 	unsigned long ie = ((vcpu->arch.guest_csr.vsie & VSIP_VALID_MASK)
566 			    << VSIP_TO_HVIP_SHIFT) & mask;
567 
568 	return (READ_ONCE(vcpu->arch.irqs_pending) & ie) ? true : false;
569 }
570 
571 void kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu)
572 {
573 	vcpu->arch.power_off = true;
574 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
575 	kvm_vcpu_kick(vcpu);
576 }
577 
578 void kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu)
579 {
580 	vcpu->arch.power_off = false;
581 	kvm_vcpu_wake_up(vcpu);
582 }
583 
584 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
585 				    struct kvm_mp_state *mp_state)
586 {
587 	if (vcpu->arch.power_off)
588 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
589 	else
590 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
591 
592 	return 0;
593 }
594 
595 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
596 				    struct kvm_mp_state *mp_state)
597 {
598 	int ret = 0;
599 
600 	switch (mp_state->mp_state) {
601 	case KVM_MP_STATE_RUNNABLE:
602 		vcpu->arch.power_off = false;
603 		break;
604 	case KVM_MP_STATE_STOPPED:
605 		kvm_riscv_vcpu_power_off(vcpu);
606 		break;
607 	default:
608 		ret = -EINVAL;
609 	}
610 
611 	return ret;
612 }
613 
614 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
615 					struct kvm_guest_debug *dbg)
616 {
617 	/* TODO; To be implemented later. */
618 	return -EINVAL;
619 }
620 
621 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
622 {
623 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
624 
625 	csr_write(CSR_VSSTATUS, csr->vsstatus);
626 	csr_write(CSR_VSIE, csr->vsie);
627 	csr_write(CSR_VSTVEC, csr->vstvec);
628 	csr_write(CSR_VSSCRATCH, csr->vsscratch);
629 	csr_write(CSR_VSEPC, csr->vsepc);
630 	csr_write(CSR_VSCAUSE, csr->vscause);
631 	csr_write(CSR_VSTVAL, csr->vstval);
632 	csr_write(CSR_HVIP, csr->hvip);
633 	csr_write(CSR_VSATP, csr->vsatp);
634 
635 	kvm_riscv_stage2_update_hgatp(vcpu);
636 
637 	kvm_riscv_vcpu_timer_restore(vcpu);
638 
639 	kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context);
640 	kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context,
641 					vcpu->arch.isa);
642 
643 	vcpu->cpu = cpu;
644 }
645 
646 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
647 {
648 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
649 
650 	vcpu->cpu = -1;
651 
652 	kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context,
653 				     vcpu->arch.isa);
654 	kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context);
655 
656 	csr_write(CSR_HGATP, 0);
657 
658 	csr->vsstatus = csr_read(CSR_VSSTATUS);
659 	csr->vsie = csr_read(CSR_VSIE);
660 	csr->vstvec = csr_read(CSR_VSTVEC);
661 	csr->vsscratch = csr_read(CSR_VSSCRATCH);
662 	csr->vsepc = csr_read(CSR_VSEPC);
663 	csr->vscause = csr_read(CSR_VSCAUSE);
664 	csr->vstval = csr_read(CSR_VSTVAL);
665 	csr->hvip = csr_read(CSR_HVIP);
666 	csr->vsatp = csr_read(CSR_VSATP);
667 }
668 
669 static void kvm_riscv_check_vcpu_requests(struct kvm_vcpu *vcpu)
670 {
671 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
672 
673 	if (kvm_request_pending(vcpu)) {
674 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) {
675 			rcuwait_wait_event(wait,
676 				(!vcpu->arch.power_off) && (!vcpu->arch.pause),
677 				TASK_INTERRUPTIBLE);
678 
679 			if (vcpu->arch.power_off || vcpu->arch.pause) {
680 				/*
681 				 * Awaken to handle a signal, request to
682 				 * sleep again later.
683 				 */
684 				kvm_make_request(KVM_REQ_SLEEP, vcpu);
685 			}
686 		}
687 
688 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
689 			kvm_riscv_reset_vcpu(vcpu);
690 
691 		if (kvm_check_request(KVM_REQ_UPDATE_HGATP, vcpu))
692 			kvm_riscv_stage2_update_hgatp(vcpu);
693 
694 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
695 			__kvm_riscv_hfence_gvma_all();
696 	}
697 }
698 
699 static void kvm_riscv_update_hvip(struct kvm_vcpu *vcpu)
700 {
701 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
702 
703 	csr_write(CSR_HVIP, csr->hvip);
704 }
705 
706 /*
707  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
708  * the vCPU is running.
709  *
710  * This must be noinstr as instrumentation may make use of RCU, and this is not
711  * safe during the EQS.
712  */
713 static void noinstr kvm_riscv_vcpu_enter_exit(struct kvm_vcpu *vcpu)
714 {
715 	guest_state_enter_irqoff();
716 	__kvm_riscv_switch_to(&vcpu->arch);
717 	guest_state_exit_irqoff();
718 }
719 
720 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
721 {
722 	int ret;
723 	struct kvm_cpu_trap trap;
724 	struct kvm_run *run = vcpu->run;
725 
726 	/* Mark this VCPU ran at least once */
727 	vcpu->arch.ran_atleast_once = true;
728 
729 	vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
730 
731 	/* Process MMIO value returned from user-space */
732 	if (run->exit_reason == KVM_EXIT_MMIO) {
733 		ret = kvm_riscv_vcpu_mmio_return(vcpu, vcpu->run);
734 		if (ret) {
735 			srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
736 			return ret;
737 		}
738 	}
739 
740 	/* Process SBI value returned from user-space */
741 	if (run->exit_reason == KVM_EXIT_RISCV_SBI) {
742 		ret = kvm_riscv_vcpu_sbi_return(vcpu, vcpu->run);
743 		if (ret) {
744 			srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
745 			return ret;
746 		}
747 	}
748 
749 	if (run->immediate_exit) {
750 		srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
751 		return -EINTR;
752 	}
753 
754 	vcpu_load(vcpu);
755 
756 	kvm_sigset_activate(vcpu);
757 
758 	ret = 1;
759 	run->exit_reason = KVM_EXIT_UNKNOWN;
760 	while (ret > 0) {
761 		/* Check conditions before entering the guest */
762 		cond_resched();
763 
764 		kvm_riscv_stage2_vmid_update(vcpu);
765 
766 		kvm_riscv_check_vcpu_requests(vcpu);
767 
768 		preempt_disable();
769 
770 		local_irq_disable();
771 
772 		/*
773 		 * Exit if we have a signal pending so that we can deliver
774 		 * the signal to user space.
775 		 */
776 		if (signal_pending(current)) {
777 			ret = -EINTR;
778 			run->exit_reason = KVM_EXIT_INTR;
779 		}
780 
781 		/*
782 		 * Ensure we set mode to IN_GUEST_MODE after we disable
783 		 * interrupts and before the final VCPU requests check.
784 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
785 		 * Documentation/virt/kvm/vcpu-requests.rst
786 		 */
787 		vcpu->mode = IN_GUEST_MODE;
788 
789 		srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
790 		smp_mb__after_srcu_read_unlock();
791 
792 		/*
793 		 * We might have got VCPU interrupts updated asynchronously
794 		 * so update it in HW.
795 		 */
796 		kvm_riscv_vcpu_flush_interrupts(vcpu);
797 
798 		/* Update HVIP CSR for current CPU */
799 		kvm_riscv_update_hvip(vcpu);
800 
801 		if (ret <= 0 ||
802 		    kvm_riscv_stage2_vmid_ver_changed(&vcpu->kvm->arch.vmid) ||
803 		    kvm_request_pending(vcpu)) {
804 			vcpu->mode = OUTSIDE_GUEST_MODE;
805 			local_irq_enable();
806 			preempt_enable();
807 			vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
808 			continue;
809 		}
810 
811 		guest_timing_enter_irqoff();
812 
813 		kvm_riscv_vcpu_enter_exit(vcpu);
814 
815 		vcpu->mode = OUTSIDE_GUEST_MODE;
816 		vcpu->stat.exits++;
817 
818 		/*
819 		 * Save SCAUSE, STVAL, HTVAL, and HTINST because we might
820 		 * get an interrupt between __kvm_riscv_switch_to() and
821 		 * local_irq_enable() which can potentially change CSRs.
822 		 */
823 		trap.sepc = vcpu->arch.guest_context.sepc;
824 		trap.scause = csr_read(CSR_SCAUSE);
825 		trap.stval = csr_read(CSR_STVAL);
826 		trap.htval = csr_read(CSR_HTVAL);
827 		trap.htinst = csr_read(CSR_HTINST);
828 
829 		/* Syncup interrupts state with HW */
830 		kvm_riscv_vcpu_sync_interrupts(vcpu);
831 
832 		/*
833 		 * We must ensure that any pending interrupts are taken before
834 		 * we exit guest timing so that timer ticks are accounted as
835 		 * guest time. Transiently unmask interrupts so that any
836 		 * pending interrupts are taken.
837 		 *
838 		 * There's no barrier which ensures that pending interrupts are
839 		 * recognised, so we just hope that the CPU takes any pending
840 		 * interrupts between the enable and disable.
841 		 */
842 		local_irq_enable();
843 		local_irq_disable();
844 
845 		guest_timing_exit_irqoff();
846 
847 		local_irq_enable();
848 
849 		preempt_enable();
850 
851 		vcpu->arch.srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
852 
853 		ret = kvm_riscv_vcpu_exit(vcpu, run, &trap);
854 	}
855 
856 	kvm_sigset_deactivate(vcpu);
857 
858 	vcpu_put(vcpu);
859 
860 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->arch.srcu_idx);
861 
862 	return ret;
863 }
864