xref: /linux/arch/riscv/kvm/vcpu.c (revision ef815d2cba782e96b9aad9483523d474ed41c62a)
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/entry-kvm.h>
11 #include <linux/errno.h>
12 #include <linux/err.h>
13 #include <linux/kdebug.h>
14 #include <linux/module.h>
15 #include <linux/percpu.h>
16 #include <linux/uaccess.h>
17 #include <linux/vmalloc.h>
18 #include <linux/sched/signal.h>
19 #include <linux/fs.h>
20 #include <linux/kvm_host.h>
21 #include <asm/csr.h>
22 #include <asm/cacheflush.h>
23 #include <asm/hwcap.h>
24 #include <asm/sbi.h>
25 #include <asm/vector.h>
26 #include <asm/kvm_vcpu_vector.h>
27 
28 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
29 	KVM_GENERIC_VCPU_STATS(),
30 	STATS_DESC_COUNTER(VCPU, ecall_exit_stat),
31 	STATS_DESC_COUNTER(VCPU, wfi_exit_stat),
32 	STATS_DESC_COUNTER(VCPU, mmio_exit_user),
33 	STATS_DESC_COUNTER(VCPU, mmio_exit_kernel),
34 	STATS_DESC_COUNTER(VCPU, csr_exit_user),
35 	STATS_DESC_COUNTER(VCPU, csr_exit_kernel),
36 	STATS_DESC_COUNTER(VCPU, signal_exits),
37 	STATS_DESC_COUNTER(VCPU, exits)
38 };
39 
40 const struct kvm_stats_header kvm_vcpu_stats_header = {
41 	.name_size = KVM_STATS_NAME_SIZE,
42 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
43 	.id_offset = sizeof(struct kvm_stats_header),
44 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
45 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
46 		       sizeof(kvm_vcpu_stats_desc),
47 };
48 
49 #define KVM_RISCV_BASE_ISA_MASK		GENMASK(25, 0)
50 
51 #define KVM_ISA_EXT_ARR(ext)		[KVM_RISCV_ISA_EXT_##ext] = RISCV_ISA_EXT_##ext
52 
53 /* Mapping between KVM ISA Extension ID & Host ISA extension ID */
54 static const unsigned long kvm_isa_ext_arr[] = {
55 	[KVM_RISCV_ISA_EXT_A] = RISCV_ISA_EXT_a,
56 	[KVM_RISCV_ISA_EXT_C] = RISCV_ISA_EXT_c,
57 	[KVM_RISCV_ISA_EXT_D] = RISCV_ISA_EXT_d,
58 	[KVM_RISCV_ISA_EXT_F] = RISCV_ISA_EXT_f,
59 	[KVM_RISCV_ISA_EXT_H] = RISCV_ISA_EXT_h,
60 	[KVM_RISCV_ISA_EXT_I] = RISCV_ISA_EXT_i,
61 	[KVM_RISCV_ISA_EXT_M] = RISCV_ISA_EXT_m,
62 	[KVM_RISCV_ISA_EXT_V] = RISCV_ISA_EXT_v,
63 
64 	KVM_ISA_EXT_ARR(SSAIA),
65 	KVM_ISA_EXT_ARR(SSTC),
66 	KVM_ISA_EXT_ARR(SVINVAL),
67 	KVM_ISA_EXT_ARR(SVNAPOT),
68 	KVM_ISA_EXT_ARR(SVPBMT),
69 	KVM_ISA_EXT_ARR(ZBB),
70 	KVM_ISA_EXT_ARR(ZIHINTPAUSE),
71 	KVM_ISA_EXT_ARR(ZICBOM),
72 	KVM_ISA_EXT_ARR(ZICBOZ),
73 };
74 
75 static unsigned long kvm_riscv_vcpu_base2isa_ext(unsigned long base_ext)
76 {
77 	unsigned long i;
78 
79 	for (i = 0; i < KVM_RISCV_ISA_EXT_MAX; i++) {
80 		if (kvm_isa_ext_arr[i] == base_ext)
81 			return i;
82 	}
83 
84 	return KVM_RISCV_ISA_EXT_MAX;
85 }
86 
87 static bool kvm_riscv_vcpu_isa_enable_allowed(unsigned long ext)
88 {
89 	switch (ext) {
90 	case KVM_RISCV_ISA_EXT_H:
91 		return false;
92 	case KVM_RISCV_ISA_EXT_V:
93 		return riscv_v_vstate_ctrl_user_allowed();
94 	default:
95 		break;
96 	}
97 
98 	return true;
99 }
100 
101 static bool kvm_riscv_vcpu_isa_disable_allowed(unsigned long ext)
102 {
103 	switch (ext) {
104 	case KVM_RISCV_ISA_EXT_A:
105 	case KVM_RISCV_ISA_EXT_C:
106 	case KVM_RISCV_ISA_EXT_I:
107 	case KVM_RISCV_ISA_EXT_M:
108 	case KVM_RISCV_ISA_EXT_SSAIA:
109 	case KVM_RISCV_ISA_EXT_SSTC:
110 	case KVM_RISCV_ISA_EXT_SVINVAL:
111 	case KVM_RISCV_ISA_EXT_SVNAPOT:
112 	case KVM_RISCV_ISA_EXT_ZIHINTPAUSE:
113 	case KVM_RISCV_ISA_EXT_ZBB:
114 		return false;
115 	default:
116 		break;
117 	}
118 
119 	return true;
120 }
121 
122 static void kvm_riscv_reset_vcpu(struct kvm_vcpu *vcpu)
123 {
124 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
125 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
126 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
127 	struct kvm_cpu_context *reset_cntx = &vcpu->arch.guest_reset_context;
128 	bool loaded;
129 
130 	/**
131 	 * The preemption should be disabled here because it races with
132 	 * kvm_sched_out/kvm_sched_in(called from preempt notifiers) which
133 	 * also calls vcpu_load/put.
134 	 */
135 	get_cpu();
136 	loaded = (vcpu->cpu != -1);
137 	if (loaded)
138 		kvm_arch_vcpu_put(vcpu);
139 
140 	vcpu->arch.last_exit_cpu = -1;
141 
142 	memcpy(csr, reset_csr, sizeof(*csr));
143 
144 	memcpy(cntx, reset_cntx, sizeof(*cntx));
145 
146 	kvm_riscv_vcpu_fp_reset(vcpu);
147 
148 	kvm_riscv_vcpu_vector_reset(vcpu);
149 
150 	kvm_riscv_vcpu_timer_reset(vcpu);
151 
152 	kvm_riscv_vcpu_aia_reset(vcpu);
153 
154 	bitmap_zero(vcpu->arch.irqs_pending, KVM_RISCV_VCPU_NR_IRQS);
155 	bitmap_zero(vcpu->arch.irqs_pending_mask, KVM_RISCV_VCPU_NR_IRQS);
156 
157 	kvm_riscv_vcpu_pmu_reset(vcpu);
158 
159 	vcpu->arch.hfence_head = 0;
160 	vcpu->arch.hfence_tail = 0;
161 	memset(vcpu->arch.hfence_queue, 0, sizeof(vcpu->arch.hfence_queue));
162 
163 	/* Reset the guest CSRs for hotplug usecase */
164 	if (loaded)
165 		kvm_arch_vcpu_load(vcpu, smp_processor_id());
166 	put_cpu();
167 }
168 
169 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
170 {
171 	return 0;
172 }
173 
174 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
175 {
176 	int rc;
177 	struct kvm_cpu_context *cntx;
178 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
179 	unsigned long host_isa, i;
180 
181 	/* Mark this VCPU never ran */
182 	vcpu->arch.ran_atleast_once = false;
183 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
184 	bitmap_zero(vcpu->arch.isa, RISCV_ISA_EXT_MAX);
185 
186 	/* Setup ISA features available to VCPU */
187 	for (i = 0; i < ARRAY_SIZE(kvm_isa_ext_arr); i++) {
188 		host_isa = kvm_isa_ext_arr[i];
189 		if (__riscv_isa_extension_available(NULL, host_isa) &&
190 		    kvm_riscv_vcpu_isa_enable_allowed(i))
191 			set_bit(host_isa, vcpu->arch.isa);
192 	}
193 
194 	/* Setup vendor, arch, and implementation details */
195 	vcpu->arch.mvendorid = sbi_get_mvendorid();
196 	vcpu->arch.marchid = sbi_get_marchid();
197 	vcpu->arch.mimpid = sbi_get_mimpid();
198 
199 	/* Setup VCPU hfence queue */
200 	spin_lock_init(&vcpu->arch.hfence_lock);
201 
202 	/* Setup reset state of shadow SSTATUS and HSTATUS CSRs */
203 	cntx = &vcpu->arch.guest_reset_context;
204 	cntx->sstatus = SR_SPP | SR_SPIE;
205 	cntx->hstatus = 0;
206 	cntx->hstatus |= HSTATUS_VTW;
207 	cntx->hstatus |= HSTATUS_SPVP;
208 	cntx->hstatus |= HSTATUS_SPV;
209 
210 	if (kvm_riscv_vcpu_alloc_vector_context(vcpu, cntx))
211 		return -ENOMEM;
212 
213 	/* By default, make CY, TM, and IR counters accessible in VU mode */
214 	reset_csr->scounteren = 0x7;
215 
216 	/* Setup VCPU timer */
217 	kvm_riscv_vcpu_timer_init(vcpu);
218 
219 	/* setup performance monitoring */
220 	kvm_riscv_vcpu_pmu_init(vcpu);
221 
222 	/* Setup VCPU AIA */
223 	rc = kvm_riscv_vcpu_aia_init(vcpu);
224 	if (rc)
225 		return rc;
226 
227 	/* Reset VCPU */
228 	kvm_riscv_reset_vcpu(vcpu);
229 
230 	return 0;
231 }
232 
233 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
234 {
235 	/**
236 	 * vcpu with id 0 is the designated boot cpu.
237 	 * Keep all vcpus with non-zero id in power-off state so that
238 	 * they can be brought up using SBI HSM extension.
239 	 */
240 	if (vcpu->vcpu_idx != 0)
241 		kvm_riscv_vcpu_power_off(vcpu);
242 }
243 
244 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
245 {
246 	/* Cleanup VCPU AIA context */
247 	kvm_riscv_vcpu_aia_deinit(vcpu);
248 
249 	/* Cleanup VCPU timer */
250 	kvm_riscv_vcpu_timer_deinit(vcpu);
251 
252 	kvm_riscv_vcpu_pmu_deinit(vcpu);
253 
254 	/* Free unused pages pre-allocated for G-stage page table mappings */
255 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
256 
257 	/* Free vector context space for host and guest kernel */
258 	kvm_riscv_vcpu_free_vector_context(vcpu);
259 }
260 
261 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
262 {
263 	return kvm_riscv_vcpu_timer_pending(vcpu);
264 }
265 
266 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
267 {
268 	kvm_riscv_aia_wakeon_hgei(vcpu, true);
269 }
270 
271 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
272 {
273 	kvm_riscv_aia_wakeon_hgei(vcpu, false);
274 }
275 
276 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
277 {
278 	return (kvm_riscv_vcpu_has_interrupts(vcpu, -1UL) &&
279 		!vcpu->arch.power_off && !vcpu->arch.pause);
280 }
281 
282 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
283 {
284 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
285 }
286 
287 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
288 {
289 	return (vcpu->arch.guest_context.sstatus & SR_SPP) ? true : false;
290 }
291 
292 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
293 {
294 	return VM_FAULT_SIGBUS;
295 }
296 
297 static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
298 					 const struct kvm_one_reg *reg)
299 {
300 	unsigned long __user *uaddr =
301 			(unsigned long __user *)(unsigned long)reg->addr;
302 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
303 					    KVM_REG_SIZE_MASK |
304 					    KVM_REG_RISCV_CONFIG);
305 	unsigned long reg_val;
306 
307 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
308 		return -EINVAL;
309 
310 	switch (reg_num) {
311 	case KVM_REG_RISCV_CONFIG_REG(isa):
312 		reg_val = vcpu->arch.isa[0] & KVM_RISCV_BASE_ISA_MASK;
313 		break;
314 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
315 		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOM))
316 			return -EINVAL;
317 		reg_val = riscv_cbom_block_size;
318 		break;
319 	case KVM_REG_RISCV_CONFIG_REG(zicboz_block_size):
320 		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOZ))
321 			return -EINVAL;
322 		reg_val = riscv_cboz_block_size;
323 		break;
324 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
325 		reg_val = vcpu->arch.mvendorid;
326 		break;
327 	case KVM_REG_RISCV_CONFIG_REG(marchid):
328 		reg_val = vcpu->arch.marchid;
329 		break;
330 	case KVM_REG_RISCV_CONFIG_REG(mimpid):
331 		reg_val = vcpu->arch.mimpid;
332 		break;
333 	default:
334 		return -EINVAL;
335 	}
336 
337 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
338 		return -EFAULT;
339 
340 	return 0;
341 }
342 
343 static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
344 					 const struct kvm_one_reg *reg)
345 {
346 	unsigned long __user *uaddr =
347 			(unsigned long __user *)(unsigned long)reg->addr;
348 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
349 					    KVM_REG_SIZE_MASK |
350 					    KVM_REG_RISCV_CONFIG);
351 	unsigned long i, isa_ext, reg_val;
352 
353 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
354 		return -EINVAL;
355 
356 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
357 		return -EFAULT;
358 
359 	switch (reg_num) {
360 	case KVM_REG_RISCV_CONFIG_REG(isa):
361 		/*
362 		 * This ONE REG interface is only defined for
363 		 * single letter extensions.
364 		 */
365 		if (fls(reg_val) >= RISCV_ISA_EXT_BASE)
366 			return -EINVAL;
367 
368 		if (!vcpu->arch.ran_atleast_once) {
369 			/* Ignore the enable/disable request for certain extensions */
370 			for (i = 0; i < RISCV_ISA_EXT_BASE; i++) {
371 				isa_ext = kvm_riscv_vcpu_base2isa_ext(i);
372 				if (isa_ext >= KVM_RISCV_ISA_EXT_MAX) {
373 					reg_val &= ~BIT(i);
374 					continue;
375 				}
376 				if (!kvm_riscv_vcpu_isa_enable_allowed(isa_ext))
377 					if (reg_val & BIT(i))
378 						reg_val &= ~BIT(i);
379 				if (!kvm_riscv_vcpu_isa_disable_allowed(isa_ext))
380 					if (!(reg_val & BIT(i)))
381 						reg_val |= BIT(i);
382 			}
383 			reg_val &= riscv_isa_extension_base(NULL);
384 			/* Do not modify anything beyond single letter extensions */
385 			reg_val = (vcpu->arch.isa[0] & ~KVM_RISCV_BASE_ISA_MASK) |
386 				  (reg_val & KVM_RISCV_BASE_ISA_MASK);
387 			vcpu->arch.isa[0] = reg_val;
388 			kvm_riscv_vcpu_fp_reset(vcpu);
389 		} else {
390 			return -EOPNOTSUPP;
391 		}
392 		break;
393 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
394 		return -EOPNOTSUPP;
395 	case KVM_REG_RISCV_CONFIG_REG(zicboz_block_size):
396 		return -EOPNOTSUPP;
397 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
398 		if (!vcpu->arch.ran_atleast_once)
399 			vcpu->arch.mvendorid = reg_val;
400 		else
401 			return -EBUSY;
402 		break;
403 	case KVM_REG_RISCV_CONFIG_REG(marchid):
404 		if (!vcpu->arch.ran_atleast_once)
405 			vcpu->arch.marchid = reg_val;
406 		else
407 			return -EBUSY;
408 		break;
409 	case KVM_REG_RISCV_CONFIG_REG(mimpid):
410 		if (!vcpu->arch.ran_atleast_once)
411 			vcpu->arch.mimpid = reg_val;
412 		else
413 			return -EBUSY;
414 		break;
415 	default:
416 		return -EINVAL;
417 	}
418 
419 	return 0;
420 }
421 
422 static int kvm_riscv_vcpu_get_reg_core(struct kvm_vcpu *vcpu,
423 				       const struct kvm_one_reg *reg)
424 {
425 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
426 	unsigned long __user *uaddr =
427 			(unsigned long __user *)(unsigned long)reg->addr;
428 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
429 					    KVM_REG_SIZE_MASK |
430 					    KVM_REG_RISCV_CORE);
431 	unsigned long reg_val;
432 
433 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
434 		return -EINVAL;
435 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
436 		return -EINVAL;
437 
438 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
439 		reg_val = cntx->sepc;
440 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
441 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
442 		reg_val = ((unsigned long *)cntx)[reg_num];
443 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode))
444 		reg_val = (cntx->sstatus & SR_SPP) ?
445 				KVM_RISCV_MODE_S : KVM_RISCV_MODE_U;
446 	else
447 		return -EINVAL;
448 
449 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
450 		return -EFAULT;
451 
452 	return 0;
453 }
454 
455 static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
456 				       const struct kvm_one_reg *reg)
457 {
458 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
459 	unsigned long __user *uaddr =
460 			(unsigned long __user *)(unsigned long)reg->addr;
461 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
462 					    KVM_REG_SIZE_MASK |
463 					    KVM_REG_RISCV_CORE);
464 	unsigned long reg_val;
465 
466 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
467 		return -EINVAL;
468 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
469 		return -EINVAL;
470 
471 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
472 		return -EFAULT;
473 
474 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
475 		cntx->sepc = reg_val;
476 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
477 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
478 		((unsigned long *)cntx)[reg_num] = reg_val;
479 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode)) {
480 		if (reg_val == KVM_RISCV_MODE_S)
481 			cntx->sstatus |= SR_SPP;
482 		else
483 			cntx->sstatus &= ~SR_SPP;
484 	} else
485 		return -EINVAL;
486 
487 	return 0;
488 }
489 
490 static int kvm_riscv_vcpu_general_get_csr(struct kvm_vcpu *vcpu,
491 					  unsigned long reg_num,
492 					  unsigned long *out_val)
493 {
494 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
495 
496 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
497 		return -EINVAL;
498 
499 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
500 		kvm_riscv_vcpu_flush_interrupts(vcpu);
501 		*out_val = (csr->hvip >> VSIP_TO_HVIP_SHIFT) & VSIP_VALID_MASK;
502 		*out_val |= csr->hvip & ~IRQ_LOCAL_MASK;
503 	} else
504 		*out_val = ((unsigned long *)csr)[reg_num];
505 
506 	return 0;
507 }
508 
509 static inline int kvm_riscv_vcpu_general_set_csr(struct kvm_vcpu *vcpu,
510 						 unsigned long reg_num,
511 						 unsigned long reg_val)
512 {
513 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
514 
515 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
516 		return -EINVAL;
517 
518 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
519 		reg_val &= VSIP_VALID_MASK;
520 		reg_val <<= VSIP_TO_HVIP_SHIFT;
521 	}
522 
523 	((unsigned long *)csr)[reg_num] = reg_val;
524 
525 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip))
526 		WRITE_ONCE(vcpu->arch.irqs_pending_mask[0], 0);
527 
528 	return 0;
529 }
530 
531 static int kvm_riscv_vcpu_get_reg_csr(struct kvm_vcpu *vcpu,
532 				      const struct kvm_one_reg *reg)
533 {
534 	int rc;
535 	unsigned long __user *uaddr =
536 			(unsigned long __user *)(unsigned long)reg->addr;
537 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
538 					    KVM_REG_SIZE_MASK |
539 					    KVM_REG_RISCV_CSR);
540 	unsigned long reg_val, reg_subtype;
541 
542 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
543 		return -EINVAL;
544 
545 	reg_subtype = reg_num & KVM_REG_RISCV_SUBTYPE_MASK;
546 	reg_num &= ~KVM_REG_RISCV_SUBTYPE_MASK;
547 	switch (reg_subtype) {
548 	case KVM_REG_RISCV_CSR_GENERAL:
549 		rc = kvm_riscv_vcpu_general_get_csr(vcpu, reg_num, &reg_val);
550 		break;
551 	case KVM_REG_RISCV_CSR_AIA:
552 		rc = kvm_riscv_vcpu_aia_get_csr(vcpu, reg_num, &reg_val);
553 		break;
554 	default:
555 		rc = -EINVAL;
556 		break;
557 	}
558 	if (rc)
559 		return rc;
560 
561 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
562 		return -EFAULT;
563 
564 	return 0;
565 }
566 
567 static int kvm_riscv_vcpu_set_reg_csr(struct kvm_vcpu *vcpu,
568 				      const struct kvm_one_reg *reg)
569 {
570 	int rc;
571 	unsigned long __user *uaddr =
572 			(unsigned long __user *)(unsigned long)reg->addr;
573 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
574 					    KVM_REG_SIZE_MASK |
575 					    KVM_REG_RISCV_CSR);
576 	unsigned long reg_val, reg_subtype;
577 
578 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
579 		return -EINVAL;
580 
581 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
582 		return -EFAULT;
583 
584 	reg_subtype = reg_num & KVM_REG_RISCV_SUBTYPE_MASK;
585 	reg_num &= ~KVM_REG_RISCV_SUBTYPE_MASK;
586 	switch (reg_subtype) {
587 	case KVM_REG_RISCV_CSR_GENERAL:
588 		rc = kvm_riscv_vcpu_general_set_csr(vcpu, reg_num, reg_val);
589 		break;
590 	case KVM_REG_RISCV_CSR_AIA:
591 		rc = kvm_riscv_vcpu_aia_set_csr(vcpu, reg_num, reg_val);
592 		break;
593 	default:
594 		rc = -EINVAL;
595 		break;
596 	}
597 	if (rc)
598 		return rc;
599 
600 	return 0;
601 }
602 
603 static int kvm_riscv_vcpu_get_reg_isa_ext(struct kvm_vcpu *vcpu,
604 					  const struct kvm_one_reg *reg)
605 {
606 	unsigned long __user *uaddr =
607 			(unsigned long __user *)(unsigned long)reg->addr;
608 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
609 					    KVM_REG_SIZE_MASK |
610 					    KVM_REG_RISCV_ISA_EXT);
611 	unsigned long reg_val = 0;
612 	unsigned long host_isa_ext;
613 
614 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
615 		return -EINVAL;
616 
617 	if (reg_num >= KVM_RISCV_ISA_EXT_MAX ||
618 	    reg_num >= ARRAY_SIZE(kvm_isa_ext_arr))
619 		return -EINVAL;
620 
621 	host_isa_ext = kvm_isa_ext_arr[reg_num];
622 	if (__riscv_isa_extension_available(vcpu->arch.isa, host_isa_ext))
623 		reg_val = 1; /* Mark the given extension as available */
624 
625 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
626 		return -EFAULT;
627 
628 	return 0;
629 }
630 
631 static int kvm_riscv_vcpu_set_reg_isa_ext(struct kvm_vcpu *vcpu,
632 					  const struct kvm_one_reg *reg)
633 {
634 	unsigned long __user *uaddr =
635 			(unsigned long __user *)(unsigned long)reg->addr;
636 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
637 					    KVM_REG_SIZE_MASK |
638 					    KVM_REG_RISCV_ISA_EXT);
639 	unsigned long reg_val;
640 	unsigned long host_isa_ext;
641 
642 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
643 		return -EINVAL;
644 
645 	if (reg_num >= KVM_RISCV_ISA_EXT_MAX ||
646 	    reg_num >= ARRAY_SIZE(kvm_isa_ext_arr))
647 		return -EINVAL;
648 
649 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
650 		return -EFAULT;
651 
652 	host_isa_ext = kvm_isa_ext_arr[reg_num];
653 	if (!__riscv_isa_extension_available(NULL, host_isa_ext))
654 		return	-EOPNOTSUPP;
655 
656 	if (!vcpu->arch.ran_atleast_once) {
657 		/*
658 		 * All multi-letter extension and a few single letter
659 		 * extension can be disabled
660 		 */
661 		if (reg_val == 1 &&
662 		    kvm_riscv_vcpu_isa_enable_allowed(reg_num))
663 			set_bit(host_isa_ext, vcpu->arch.isa);
664 		else if (!reg_val &&
665 			 kvm_riscv_vcpu_isa_disable_allowed(reg_num))
666 			clear_bit(host_isa_ext, vcpu->arch.isa);
667 		else
668 			return -EINVAL;
669 		kvm_riscv_vcpu_fp_reset(vcpu);
670 	} else {
671 		return -EOPNOTSUPP;
672 	}
673 
674 	return 0;
675 }
676 
677 static int kvm_riscv_vcpu_set_reg(struct kvm_vcpu *vcpu,
678 				  const struct kvm_one_reg *reg)
679 {
680 	switch (reg->id & KVM_REG_RISCV_TYPE_MASK) {
681 	case KVM_REG_RISCV_CONFIG:
682 		return kvm_riscv_vcpu_set_reg_config(vcpu, reg);
683 	case KVM_REG_RISCV_CORE:
684 		return kvm_riscv_vcpu_set_reg_core(vcpu, reg);
685 	case KVM_REG_RISCV_CSR:
686 		return kvm_riscv_vcpu_set_reg_csr(vcpu, reg);
687 	case KVM_REG_RISCV_TIMER:
688 		return kvm_riscv_vcpu_set_reg_timer(vcpu, reg);
689 	case KVM_REG_RISCV_FP_F:
690 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
691 						 KVM_REG_RISCV_FP_F);
692 	case KVM_REG_RISCV_FP_D:
693 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
694 						 KVM_REG_RISCV_FP_D);
695 	case KVM_REG_RISCV_ISA_EXT:
696 		return kvm_riscv_vcpu_set_reg_isa_ext(vcpu, reg);
697 	case KVM_REG_RISCV_SBI_EXT:
698 		return kvm_riscv_vcpu_set_reg_sbi_ext(vcpu, reg);
699 	case KVM_REG_RISCV_VECTOR:
700 		return kvm_riscv_vcpu_set_reg_vector(vcpu, reg,
701 						 KVM_REG_RISCV_VECTOR);
702 	default:
703 		break;
704 	}
705 
706 	return -EINVAL;
707 }
708 
709 static int kvm_riscv_vcpu_get_reg(struct kvm_vcpu *vcpu,
710 				  const struct kvm_one_reg *reg)
711 {
712 	switch (reg->id & KVM_REG_RISCV_TYPE_MASK) {
713 	case KVM_REG_RISCV_CONFIG:
714 		return kvm_riscv_vcpu_get_reg_config(vcpu, reg);
715 	case KVM_REG_RISCV_CORE:
716 		return kvm_riscv_vcpu_get_reg_core(vcpu, reg);
717 	case KVM_REG_RISCV_CSR:
718 		return kvm_riscv_vcpu_get_reg_csr(vcpu, reg);
719 	case KVM_REG_RISCV_TIMER:
720 		return kvm_riscv_vcpu_get_reg_timer(vcpu, reg);
721 	case KVM_REG_RISCV_FP_F:
722 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
723 						 KVM_REG_RISCV_FP_F);
724 	case KVM_REG_RISCV_FP_D:
725 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
726 						 KVM_REG_RISCV_FP_D);
727 	case KVM_REG_RISCV_ISA_EXT:
728 		return kvm_riscv_vcpu_get_reg_isa_ext(vcpu, reg);
729 	case KVM_REG_RISCV_SBI_EXT:
730 		return kvm_riscv_vcpu_get_reg_sbi_ext(vcpu, reg);
731 	case KVM_REG_RISCV_VECTOR:
732 		return kvm_riscv_vcpu_get_reg_vector(vcpu, reg,
733 						 KVM_REG_RISCV_VECTOR);
734 	default:
735 		break;
736 	}
737 
738 	return -EINVAL;
739 }
740 
741 long kvm_arch_vcpu_async_ioctl(struct file *filp,
742 			       unsigned int ioctl, unsigned long arg)
743 {
744 	struct kvm_vcpu *vcpu = filp->private_data;
745 	void __user *argp = (void __user *)arg;
746 
747 	if (ioctl == KVM_INTERRUPT) {
748 		struct kvm_interrupt irq;
749 
750 		if (copy_from_user(&irq, argp, sizeof(irq)))
751 			return -EFAULT;
752 
753 		if (irq.irq == KVM_INTERRUPT_SET)
754 			return kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT);
755 		else
756 			return kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
757 	}
758 
759 	return -ENOIOCTLCMD;
760 }
761 
762 long kvm_arch_vcpu_ioctl(struct file *filp,
763 			 unsigned int ioctl, unsigned long arg)
764 {
765 	struct kvm_vcpu *vcpu = filp->private_data;
766 	void __user *argp = (void __user *)arg;
767 	long r = -EINVAL;
768 
769 	switch (ioctl) {
770 	case KVM_SET_ONE_REG:
771 	case KVM_GET_ONE_REG: {
772 		struct kvm_one_reg reg;
773 
774 		r = -EFAULT;
775 		if (copy_from_user(&reg, argp, sizeof(reg)))
776 			break;
777 
778 		if (ioctl == KVM_SET_ONE_REG)
779 			r = kvm_riscv_vcpu_set_reg(vcpu, &reg);
780 		else
781 			r = kvm_riscv_vcpu_get_reg(vcpu, &reg);
782 		break;
783 	}
784 	default:
785 		break;
786 	}
787 
788 	return r;
789 }
790 
791 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
792 				  struct kvm_sregs *sregs)
793 {
794 	return -EINVAL;
795 }
796 
797 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
798 				  struct kvm_sregs *sregs)
799 {
800 	return -EINVAL;
801 }
802 
803 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
804 {
805 	return -EINVAL;
806 }
807 
808 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
809 {
810 	return -EINVAL;
811 }
812 
813 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
814 				  struct kvm_translation *tr)
815 {
816 	return -EINVAL;
817 }
818 
819 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
820 {
821 	return -EINVAL;
822 }
823 
824 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
825 {
826 	return -EINVAL;
827 }
828 
829 void kvm_riscv_vcpu_flush_interrupts(struct kvm_vcpu *vcpu)
830 {
831 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
832 	unsigned long mask, val;
833 
834 	if (READ_ONCE(vcpu->arch.irqs_pending_mask[0])) {
835 		mask = xchg_acquire(&vcpu->arch.irqs_pending_mask[0], 0);
836 		val = READ_ONCE(vcpu->arch.irqs_pending[0]) & mask;
837 
838 		csr->hvip &= ~mask;
839 		csr->hvip |= val;
840 	}
841 
842 	/* Flush AIA high interrupts */
843 	kvm_riscv_vcpu_aia_flush_interrupts(vcpu);
844 }
845 
846 void kvm_riscv_vcpu_sync_interrupts(struct kvm_vcpu *vcpu)
847 {
848 	unsigned long hvip;
849 	struct kvm_vcpu_arch *v = &vcpu->arch;
850 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
851 
852 	/* Read current HVIP and VSIE CSRs */
853 	csr->vsie = csr_read(CSR_VSIE);
854 
855 	/* Sync-up HVIP.VSSIP bit changes does by Guest */
856 	hvip = csr_read(CSR_HVIP);
857 	if ((csr->hvip ^ hvip) & (1UL << IRQ_VS_SOFT)) {
858 		if (hvip & (1UL << IRQ_VS_SOFT)) {
859 			if (!test_and_set_bit(IRQ_VS_SOFT,
860 					      v->irqs_pending_mask))
861 				set_bit(IRQ_VS_SOFT, v->irqs_pending);
862 		} else {
863 			if (!test_and_set_bit(IRQ_VS_SOFT,
864 					      v->irqs_pending_mask))
865 				clear_bit(IRQ_VS_SOFT, v->irqs_pending);
866 		}
867 	}
868 
869 	/* Sync-up AIA high interrupts */
870 	kvm_riscv_vcpu_aia_sync_interrupts(vcpu);
871 
872 	/* Sync-up timer CSRs */
873 	kvm_riscv_vcpu_timer_sync(vcpu);
874 }
875 
876 int kvm_riscv_vcpu_set_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
877 {
878 	/*
879 	 * We only allow VS-mode software, timer, and external
880 	 * interrupts when irq is one of the local interrupts
881 	 * defined by RISC-V privilege specification.
882 	 */
883 	if (irq < IRQ_LOCAL_MAX &&
884 	    irq != IRQ_VS_SOFT &&
885 	    irq != IRQ_VS_TIMER &&
886 	    irq != IRQ_VS_EXT)
887 		return -EINVAL;
888 
889 	set_bit(irq, vcpu->arch.irqs_pending);
890 	smp_mb__before_atomic();
891 	set_bit(irq, vcpu->arch.irqs_pending_mask);
892 
893 	kvm_vcpu_kick(vcpu);
894 
895 	return 0;
896 }
897 
898 int kvm_riscv_vcpu_unset_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
899 {
900 	/*
901 	 * We only allow VS-mode software, timer, and external
902 	 * interrupts when irq is one of the local interrupts
903 	 * defined by RISC-V privilege specification.
904 	 */
905 	if (irq < IRQ_LOCAL_MAX &&
906 	    irq != IRQ_VS_SOFT &&
907 	    irq != IRQ_VS_TIMER &&
908 	    irq != IRQ_VS_EXT)
909 		return -EINVAL;
910 
911 	clear_bit(irq, vcpu->arch.irqs_pending);
912 	smp_mb__before_atomic();
913 	set_bit(irq, vcpu->arch.irqs_pending_mask);
914 
915 	return 0;
916 }
917 
918 bool kvm_riscv_vcpu_has_interrupts(struct kvm_vcpu *vcpu, u64 mask)
919 {
920 	unsigned long ie;
921 
922 	ie = ((vcpu->arch.guest_csr.vsie & VSIP_VALID_MASK)
923 		<< VSIP_TO_HVIP_SHIFT) & (unsigned long)mask;
924 	ie |= vcpu->arch.guest_csr.vsie & ~IRQ_LOCAL_MASK &
925 		(unsigned long)mask;
926 	if (READ_ONCE(vcpu->arch.irqs_pending[0]) & ie)
927 		return true;
928 
929 	/* Check AIA high interrupts */
930 	return kvm_riscv_vcpu_aia_has_interrupts(vcpu, mask);
931 }
932 
933 void kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu)
934 {
935 	vcpu->arch.power_off = true;
936 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
937 	kvm_vcpu_kick(vcpu);
938 }
939 
940 void kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu)
941 {
942 	vcpu->arch.power_off = false;
943 	kvm_vcpu_wake_up(vcpu);
944 }
945 
946 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
947 				    struct kvm_mp_state *mp_state)
948 {
949 	if (vcpu->arch.power_off)
950 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
951 	else
952 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
953 
954 	return 0;
955 }
956 
957 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
958 				    struct kvm_mp_state *mp_state)
959 {
960 	int ret = 0;
961 
962 	switch (mp_state->mp_state) {
963 	case KVM_MP_STATE_RUNNABLE:
964 		vcpu->arch.power_off = false;
965 		break;
966 	case KVM_MP_STATE_STOPPED:
967 		kvm_riscv_vcpu_power_off(vcpu);
968 		break;
969 	default:
970 		ret = -EINVAL;
971 	}
972 
973 	return ret;
974 }
975 
976 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
977 					struct kvm_guest_debug *dbg)
978 {
979 	/* TODO; To be implemented later. */
980 	return -EINVAL;
981 }
982 
983 static void kvm_riscv_vcpu_update_config(const unsigned long *isa)
984 {
985 	u64 henvcfg = 0;
986 
987 	if (riscv_isa_extension_available(isa, SVPBMT))
988 		henvcfg |= ENVCFG_PBMTE;
989 
990 	if (riscv_isa_extension_available(isa, SSTC))
991 		henvcfg |= ENVCFG_STCE;
992 
993 	if (riscv_isa_extension_available(isa, ZICBOM))
994 		henvcfg |= (ENVCFG_CBIE | ENVCFG_CBCFE);
995 
996 	if (riscv_isa_extension_available(isa, ZICBOZ))
997 		henvcfg |= ENVCFG_CBZE;
998 
999 	csr_write(CSR_HENVCFG, henvcfg);
1000 #ifdef CONFIG_32BIT
1001 	csr_write(CSR_HENVCFGH, henvcfg >> 32);
1002 #endif
1003 }
1004 
1005 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1006 {
1007 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
1008 
1009 	csr_write(CSR_VSSTATUS, csr->vsstatus);
1010 	csr_write(CSR_VSIE, csr->vsie);
1011 	csr_write(CSR_VSTVEC, csr->vstvec);
1012 	csr_write(CSR_VSSCRATCH, csr->vsscratch);
1013 	csr_write(CSR_VSEPC, csr->vsepc);
1014 	csr_write(CSR_VSCAUSE, csr->vscause);
1015 	csr_write(CSR_VSTVAL, csr->vstval);
1016 	csr_write(CSR_HVIP, csr->hvip);
1017 	csr_write(CSR_VSATP, csr->vsatp);
1018 
1019 	kvm_riscv_vcpu_update_config(vcpu->arch.isa);
1020 
1021 	kvm_riscv_gstage_update_hgatp(vcpu);
1022 
1023 	kvm_riscv_vcpu_timer_restore(vcpu);
1024 
1025 	kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context);
1026 	kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context,
1027 					vcpu->arch.isa);
1028 	kvm_riscv_vcpu_host_vector_save(&vcpu->arch.host_context);
1029 	kvm_riscv_vcpu_guest_vector_restore(&vcpu->arch.guest_context,
1030 					    vcpu->arch.isa);
1031 
1032 	kvm_riscv_vcpu_aia_load(vcpu, cpu);
1033 
1034 	vcpu->cpu = cpu;
1035 }
1036 
1037 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1038 {
1039 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
1040 
1041 	vcpu->cpu = -1;
1042 
1043 	kvm_riscv_vcpu_aia_put(vcpu);
1044 
1045 	kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context,
1046 				     vcpu->arch.isa);
1047 	kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context);
1048 
1049 	kvm_riscv_vcpu_timer_save(vcpu);
1050 	kvm_riscv_vcpu_guest_vector_save(&vcpu->arch.guest_context,
1051 					 vcpu->arch.isa);
1052 	kvm_riscv_vcpu_host_vector_restore(&vcpu->arch.host_context);
1053 
1054 	csr->vsstatus = csr_read(CSR_VSSTATUS);
1055 	csr->vsie = csr_read(CSR_VSIE);
1056 	csr->vstvec = csr_read(CSR_VSTVEC);
1057 	csr->vsscratch = csr_read(CSR_VSSCRATCH);
1058 	csr->vsepc = csr_read(CSR_VSEPC);
1059 	csr->vscause = csr_read(CSR_VSCAUSE);
1060 	csr->vstval = csr_read(CSR_VSTVAL);
1061 	csr->hvip = csr_read(CSR_HVIP);
1062 	csr->vsatp = csr_read(CSR_VSATP);
1063 }
1064 
1065 static void kvm_riscv_check_vcpu_requests(struct kvm_vcpu *vcpu)
1066 {
1067 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
1068 
1069 	if (kvm_request_pending(vcpu)) {
1070 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) {
1071 			kvm_vcpu_srcu_read_unlock(vcpu);
1072 			rcuwait_wait_event(wait,
1073 				(!vcpu->arch.power_off) && (!vcpu->arch.pause),
1074 				TASK_INTERRUPTIBLE);
1075 			kvm_vcpu_srcu_read_lock(vcpu);
1076 
1077 			if (vcpu->arch.power_off || vcpu->arch.pause) {
1078 				/*
1079 				 * Awaken to handle a signal, request to
1080 				 * sleep again later.
1081 				 */
1082 				kvm_make_request(KVM_REQ_SLEEP, vcpu);
1083 			}
1084 		}
1085 
1086 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1087 			kvm_riscv_reset_vcpu(vcpu);
1088 
1089 		if (kvm_check_request(KVM_REQ_UPDATE_HGATP, vcpu))
1090 			kvm_riscv_gstage_update_hgatp(vcpu);
1091 
1092 		if (kvm_check_request(KVM_REQ_FENCE_I, vcpu))
1093 			kvm_riscv_fence_i_process(vcpu);
1094 
1095 		/*
1096 		 * The generic KVM_REQ_TLB_FLUSH is same as
1097 		 * KVM_REQ_HFENCE_GVMA_VMID_ALL
1098 		 */
1099 		if (kvm_check_request(KVM_REQ_HFENCE_GVMA_VMID_ALL, vcpu))
1100 			kvm_riscv_hfence_gvma_vmid_all_process(vcpu);
1101 
1102 		if (kvm_check_request(KVM_REQ_HFENCE_VVMA_ALL, vcpu))
1103 			kvm_riscv_hfence_vvma_all_process(vcpu);
1104 
1105 		if (kvm_check_request(KVM_REQ_HFENCE, vcpu))
1106 			kvm_riscv_hfence_process(vcpu);
1107 	}
1108 }
1109 
1110 static void kvm_riscv_update_hvip(struct kvm_vcpu *vcpu)
1111 {
1112 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
1113 
1114 	csr_write(CSR_HVIP, csr->hvip);
1115 	kvm_riscv_vcpu_aia_update_hvip(vcpu);
1116 }
1117 
1118 /*
1119  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
1120  * the vCPU is running.
1121  *
1122  * This must be noinstr as instrumentation may make use of RCU, and this is not
1123  * safe during the EQS.
1124  */
1125 static void noinstr kvm_riscv_vcpu_enter_exit(struct kvm_vcpu *vcpu)
1126 {
1127 	guest_state_enter_irqoff();
1128 	__kvm_riscv_switch_to(&vcpu->arch);
1129 	vcpu->arch.last_exit_cpu = vcpu->cpu;
1130 	guest_state_exit_irqoff();
1131 }
1132 
1133 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1134 {
1135 	int ret;
1136 	struct kvm_cpu_trap trap;
1137 	struct kvm_run *run = vcpu->run;
1138 
1139 	/* Mark this VCPU ran at least once */
1140 	vcpu->arch.ran_atleast_once = true;
1141 
1142 	kvm_vcpu_srcu_read_lock(vcpu);
1143 
1144 	switch (run->exit_reason) {
1145 	case KVM_EXIT_MMIO:
1146 		/* Process MMIO value returned from user-space */
1147 		ret = kvm_riscv_vcpu_mmio_return(vcpu, vcpu->run);
1148 		break;
1149 	case KVM_EXIT_RISCV_SBI:
1150 		/* Process SBI value returned from user-space */
1151 		ret = kvm_riscv_vcpu_sbi_return(vcpu, vcpu->run);
1152 		break;
1153 	case KVM_EXIT_RISCV_CSR:
1154 		/* Process CSR value returned from user-space */
1155 		ret = kvm_riscv_vcpu_csr_return(vcpu, vcpu->run);
1156 		break;
1157 	default:
1158 		ret = 0;
1159 		break;
1160 	}
1161 	if (ret) {
1162 		kvm_vcpu_srcu_read_unlock(vcpu);
1163 		return ret;
1164 	}
1165 
1166 	if (run->immediate_exit) {
1167 		kvm_vcpu_srcu_read_unlock(vcpu);
1168 		return -EINTR;
1169 	}
1170 
1171 	vcpu_load(vcpu);
1172 
1173 	kvm_sigset_activate(vcpu);
1174 
1175 	ret = 1;
1176 	run->exit_reason = KVM_EXIT_UNKNOWN;
1177 	while (ret > 0) {
1178 		/* Check conditions before entering the guest */
1179 		ret = xfer_to_guest_mode_handle_work(vcpu);
1180 		if (ret)
1181 			continue;
1182 		ret = 1;
1183 
1184 		kvm_riscv_gstage_vmid_update(vcpu);
1185 
1186 		kvm_riscv_check_vcpu_requests(vcpu);
1187 
1188 		preempt_disable();
1189 
1190 		/* Update AIA HW state before entering guest */
1191 		ret = kvm_riscv_vcpu_aia_update(vcpu);
1192 		if (ret <= 0) {
1193 			preempt_enable();
1194 			continue;
1195 		}
1196 
1197 		local_irq_disable();
1198 
1199 		/*
1200 		 * Ensure we set mode to IN_GUEST_MODE after we disable
1201 		 * interrupts and before the final VCPU requests check.
1202 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
1203 		 * Documentation/virt/kvm/vcpu-requests.rst
1204 		 */
1205 		vcpu->mode = IN_GUEST_MODE;
1206 
1207 		kvm_vcpu_srcu_read_unlock(vcpu);
1208 		smp_mb__after_srcu_read_unlock();
1209 
1210 		/*
1211 		 * We might have got VCPU interrupts updated asynchronously
1212 		 * so update it in HW.
1213 		 */
1214 		kvm_riscv_vcpu_flush_interrupts(vcpu);
1215 
1216 		/* Update HVIP CSR for current CPU */
1217 		kvm_riscv_update_hvip(vcpu);
1218 
1219 		if (ret <= 0 ||
1220 		    kvm_riscv_gstage_vmid_ver_changed(&vcpu->kvm->arch.vmid) ||
1221 		    kvm_request_pending(vcpu) ||
1222 		    xfer_to_guest_mode_work_pending()) {
1223 			vcpu->mode = OUTSIDE_GUEST_MODE;
1224 			local_irq_enable();
1225 			preempt_enable();
1226 			kvm_vcpu_srcu_read_lock(vcpu);
1227 			continue;
1228 		}
1229 
1230 		/*
1231 		 * Cleanup stale TLB enteries
1232 		 *
1233 		 * Note: This should be done after G-stage VMID has been
1234 		 * updated using kvm_riscv_gstage_vmid_ver_changed()
1235 		 */
1236 		kvm_riscv_local_tlb_sanitize(vcpu);
1237 
1238 		guest_timing_enter_irqoff();
1239 
1240 		kvm_riscv_vcpu_enter_exit(vcpu);
1241 
1242 		vcpu->mode = OUTSIDE_GUEST_MODE;
1243 		vcpu->stat.exits++;
1244 
1245 		/*
1246 		 * Save SCAUSE, STVAL, HTVAL, and HTINST because we might
1247 		 * get an interrupt between __kvm_riscv_switch_to() and
1248 		 * local_irq_enable() which can potentially change CSRs.
1249 		 */
1250 		trap.sepc = vcpu->arch.guest_context.sepc;
1251 		trap.scause = csr_read(CSR_SCAUSE);
1252 		trap.stval = csr_read(CSR_STVAL);
1253 		trap.htval = csr_read(CSR_HTVAL);
1254 		trap.htinst = csr_read(CSR_HTINST);
1255 
1256 		/* Syncup interrupts state with HW */
1257 		kvm_riscv_vcpu_sync_interrupts(vcpu);
1258 
1259 		/*
1260 		 * We must ensure that any pending interrupts are taken before
1261 		 * we exit guest timing so that timer ticks are accounted as
1262 		 * guest time. Transiently unmask interrupts so that any
1263 		 * pending interrupts are taken.
1264 		 *
1265 		 * There's no barrier which ensures that pending interrupts are
1266 		 * recognised, so we just hope that the CPU takes any pending
1267 		 * interrupts between the enable and disable.
1268 		 */
1269 		local_irq_enable();
1270 		local_irq_disable();
1271 
1272 		guest_timing_exit_irqoff();
1273 
1274 		local_irq_enable();
1275 
1276 		preempt_enable();
1277 
1278 		kvm_vcpu_srcu_read_lock(vcpu);
1279 
1280 		ret = kvm_riscv_vcpu_exit(vcpu, run, &trap);
1281 	}
1282 
1283 	kvm_sigset_deactivate(vcpu);
1284 
1285 	vcpu_put(vcpu);
1286 
1287 	kvm_vcpu_srcu_read_unlock(vcpu);
1288 
1289 	return ret;
1290 }
1291