xref: /openbmc/linux/arch/riscv/kvm/vcpu.c (revision be2b81b5)
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 
26 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
27 	KVM_GENERIC_VCPU_STATS(),
28 	STATS_DESC_COUNTER(VCPU, ecall_exit_stat),
29 	STATS_DESC_COUNTER(VCPU, wfi_exit_stat),
30 	STATS_DESC_COUNTER(VCPU, mmio_exit_user),
31 	STATS_DESC_COUNTER(VCPU, mmio_exit_kernel),
32 	STATS_DESC_COUNTER(VCPU, csr_exit_user),
33 	STATS_DESC_COUNTER(VCPU, csr_exit_kernel),
34 	STATS_DESC_COUNTER(VCPU, signal_exits),
35 	STATS_DESC_COUNTER(VCPU, exits)
36 };
37 
38 const struct kvm_stats_header kvm_vcpu_stats_header = {
39 	.name_size = KVM_STATS_NAME_SIZE,
40 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
41 	.id_offset = sizeof(struct kvm_stats_header),
42 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
43 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
44 		       sizeof(kvm_vcpu_stats_desc),
45 };
46 
47 #define KVM_RISCV_BASE_ISA_MASK		GENMASK(25, 0)
48 
49 #define KVM_ISA_EXT_ARR(ext)		[KVM_RISCV_ISA_EXT_##ext] = RISCV_ISA_EXT_##ext
50 
51 /* Mapping between KVM ISA Extension ID & Host ISA extension ID */
52 static const unsigned long kvm_isa_ext_arr[] = {
53 	[KVM_RISCV_ISA_EXT_A] = RISCV_ISA_EXT_a,
54 	[KVM_RISCV_ISA_EXT_C] = RISCV_ISA_EXT_c,
55 	[KVM_RISCV_ISA_EXT_D] = RISCV_ISA_EXT_d,
56 	[KVM_RISCV_ISA_EXT_F] = RISCV_ISA_EXT_f,
57 	[KVM_RISCV_ISA_EXT_H] = RISCV_ISA_EXT_h,
58 	[KVM_RISCV_ISA_EXT_I] = RISCV_ISA_EXT_i,
59 	[KVM_RISCV_ISA_EXT_M] = RISCV_ISA_EXT_m,
60 
61 	KVM_ISA_EXT_ARR(SSTC),
62 	KVM_ISA_EXT_ARR(SVINVAL),
63 	KVM_ISA_EXT_ARR(SVPBMT),
64 	KVM_ISA_EXT_ARR(ZIHINTPAUSE),
65 	KVM_ISA_EXT_ARR(ZICBOM),
66 };
67 
68 static unsigned long kvm_riscv_vcpu_base2isa_ext(unsigned long base_ext)
69 {
70 	unsigned long i;
71 
72 	for (i = 0; i < KVM_RISCV_ISA_EXT_MAX; i++) {
73 		if (kvm_isa_ext_arr[i] == base_ext)
74 			return i;
75 	}
76 
77 	return KVM_RISCV_ISA_EXT_MAX;
78 }
79 
80 static bool kvm_riscv_vcpu_isa_enable_allowed(unsigned long ext)
81 {
82 	switch (ext) {
83 	case KVM_RISCV_ISA_EXT_H:
84 		return false;
85 	default:
86 		break;
87 	}
88 
89 	return true;
90 }
91 
92 static bool kvm_riscv_vcpu_isa_disable_allowed(unsigned long ext)
93 {
94 	switch (ext) {
95 	case KVM_RISCV_ISA_EXT_A:
96 	case KVM_RISCV_ISA_EXT_C:
97 	case KVM_RISCV_ISA_EXT_I:
98 	case KVM_RISCV_ISA_EXT_M:
99 	case KVM_RISCV_ISA_EXT_SSTC:
100 	case KVM_RISCV_ISA_EXT_SVINVAL:
101 	case KVM_RISCV_ISA_EXT_ZIHINTPAUSE:
102 		return false;
103 	default:
104 		break;
105 	}
106 
107 	return true;
108 }
109 
110 static void kvm_riscv_reset_vcpu(struct kvm_vcpu *vcpu)
111 {
112 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
113 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
114 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
115 	struct kvm_cpu_context *reset_cntx = &vcpu->arch.guest_reset_context;
116 	bool loaded;
117 
118 	/**
119 	 * The preemption should be disabled here because it races with
120 	 * kvm_sched_out/kvm_sched_in(called from preempt notifiers) which
121 	 * also calls vcpu_load/put.
122 	 */
123 	get_cpu();
124 	loaded = (vcpu->cpu != -1);
125 	if (loaded)
126 		kvm_arch_vcpu_put(vcpu);
127 
128 	vcpu->arch.last_exit_cpu = -1;
129 
130 	memcpy(csr, reset_csr, sizeof(*csr));
131 
132 	memcpy(cntx, reset_cntx, sizeof(*cntx));
133 
134 	kvm_riscv_vcpu_fp_reset(vcpu);
135 
136 	kvm_riscv_vcpu_timer_reset(vcpu);
137 
138 	WRITE_ONCE(vcpu->arch.irqs_pending, 0);
139 	WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
140 
141 	vcpu->arch.hfence_head = 0;
142 	vcpu->arch.hfence_tail = 0;
143 	memset(vcpu->arch.hfence_queue, 0, sizeof(vcpu->arch.hfence_queue));
144 
145 	/* Reset the guest CSRs for hotplug usecase */
146 	if (loaded)
147 		kvm_arch_vcpu_load(vcpu, smp_processor_id());
148 	put_cpu();
149 }
150 
151 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
152 {
153 	return 0;
154 }
155 
156 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
157 {
158 	struct kvm_cpu_context *cntx;
159 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
160 	unsigned long host_isa, i;
161 
162 	/* Mark this VCPU never ran */
163 	vcpu->arch.ran_atleast_once = false;
164 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
165 	bitmap_zero(vcpu->arch.isa, RISCV_ISA_EXT_MAX);
166 
167 	/* Setup ISA features available to VCPU */
168 	for (i = 0; i < ARRAY_SIZE(kvm_isa_ext_arr); i++) {
169 		host_isa = kvm_isa_ext_arr[i];
170 		if (__riscv_isa_extension_available(NULL, host_isa) &&
171 		    kvm_riscv_vcpu_isa_enable_allowed(i))
172 			set_bit(host_isa, vcpu->arch.isa);
173 	}
174 
175 	/* Setup vendor, arch, and implementation details */
176 	vcpu->arch.mvendorid = sbi_get_mvendorid();
177 	vcpu->arch.marchid = sbi_get_marchid();
178 	vcpu->arch.mimpid = sbi_get_mimpid();
179 
180 	/* Setup VCPU hfence queue */
181 	spin_lock_init(&vcpu->arch.hfence_lock);
182 
183 	/* Setup reset state of shadow SSTATUS and HSTATUS CSRs */
184 	cntx = &vcpu->arch.guest_reset_context;
185 	cntx->sstatus = SR_SPP | SR_SPIE;
186 	cntx->hstatus = 0;
187 	cntx->hstatus |= HSTATUS_VTW;
188 	cntx->hstatus |= HSTATUS_SPVP;
189 	cntx->hstatus |= HSTATUS_SPV;
190 
191 	/* By default, make CY, TM, and IR counters accessible in VU mode */
192 	reset_csr->scounteren = 0x7;
193 
194 	/* Setup VCPU timer */
195 	kvm_riscv_vcpu_timer_init(vcpu);
196 
197 	/* Reset VCPU */
198 	kvm_riscv_reset_vcpu(vcpu);
199 
200 	return 0;
201 }
202 
203 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
204 {
205 	/**
206 	 * vcpu with id 0 is the designated boot cpu.
207 	 * Keep all vcpus with non-zero id in power-off state so that
208 	 * they can be brought up using SBI HSM extension.
209 	 */
210 	if (vcpu->vcpu_idx != 0)
211 		kvm_riscv_vcpu_power_off(vcpu);
212 }
213 
214 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
215 {
216 	/* Cleanup VCPU timer */
217 	kvm_riscv_vcpu_timer_deinit(vcpu);
218 
219 	/* Free unused pages pre-allocated for G-stage page table mappings */
220 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
221 }
222 
223 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
224 {
225 	return kvm_riscv_vcpu_timer_pending(vcpu);
226 }
227 
228 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
229 {
230 }
231 
232 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
233 {
234 }
235 
236 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
237 {
238 	return (kvm_riscv_vcpu_has_interrupts(vcpu, -1UL) &&
239 		!vcpu->arch.power_off && !vcpu->arch.pause);
240 }
241 
242 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
243 {
244 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
245 }
246 
247 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
248 {
249 	return (vcpu->arch.guest_context.sstatus & SR_SPP) ? true : false;
250 }
251 
252 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
253 {
254 	return VM_FAULT_SIGBUS;
255 }
256 
257 static int kvm_riscv_vcpu_get_reg_config(struct kvm_vcpu *vcpu,
258 					 const struct kvm_one_reg *reg)
259 {
260 	unsigned long __user *uaddr =
261 			(unsigned long __user *)(unsigned long)reg->addr;
262 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
263 					    KVM_REG_SIZE_MASK |
264 					    KVM_REG_RISCV_CONFIG);
265 	unsigned long reg_val;
266 
267 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
268 		return -EINVAL;
269 
270 	switch (reg_num) {
271 	case KVM_REG_RISCV_CONFIG_REG(isa):
272 		reg_val = vcpu->arch.isa[0] & KVM_RISCV_BASE_ISA_MASK;
273 		break;
274 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
275 		if (!riscv_isa_extension_available(vcpu->arch.isa, ZICBOM))
276 			return -EINVAL;
277 		reg_val = riscv_cbom_block_size;
278 		break;
279 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
280 		reg_val = vcpu->arch.mvendorid;
281 		break;
282 	case KVM_REG_RISCV_CONFIG_REG(marchid):
283 		reg_val = vcpu->arch.marchid;
284 		break;
285 	case KVM_REG_RISCV_CONFIG_REG(mimpid):
286 		reg_val = vcpu->arch.mimpid;
287 		break;
288 	default:
289 		return -EINVAL;
290 	}
291 
292 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
293 		return -EFAULT;
294 
295 	return 0;
296 }
297 
298 static int kvm_riscv_vcpu_set_reg_config(struct kvm_vcpu *vcpu,
299 					 const struct kvm_one_reg *reg)
300 {
301 	unsigned long __user *uaddr =
302 			(unsigned long __user *)(unsigned long)reg->addr;
303 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
304 					    KVM_REG_SIZE_MASK |
305 					    KVM_REG_RISCV_CONFIG);
306 	unsigned long i, isa_ext, reg_val;
307 
308 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
309 		return -EINVAL;
310 
311 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
312 		return -EFAULT;
313 
314 	switch (reg_num) {
315 	case KVM_REG_RISCV_CONFIG_REG(isa):
316 		/*
317 		 * This ONE REG interface is only defined for
318 		 * single letter extensions.
319 		 */
320 		if (fls(reg_val) >= RISCV_ISA_EXT_BASE)
321 			return -EINVAL;
322 
323 		if (!vcpu->arch.ran_atleast_once) {
324 			/* Ignore the enable/disable request for certain extensions */
325 			for (i = 0; i < RISCV_ISA_EXT_BASE; i++) {
326 				isa_ext = kvm_riscv_vcpu_base2isa_ext(i);
327 				if (isa_ext >= KVM_RISCV_ISA_EXT_MAX) {
328 					reg_val &= ~BIT(i);
329 					continue;
330 				}
331 				if (!kvm_riscv_vcpu_isa_enable_allowed(isa_ext))
332 					if (reg_val & BIT(i))
333 						reg_val &= ~BIT(i);
334 				if (!kvm_riscv_vcpu_isa_disable_allowed(isa_ext))
335 					if (!(reg_val & BIT(i)))
336 						reg_val |= BIT(i);
337 			}
338 			reg_val &= riscv_isa_extension_base(NULL);
339 			/* Do not modify anything beyond single letter extensions */
340 			reg_val = (vcpu->arch.isa[0] & ~KVM_RISCV_BASE_ISA_MASK) |
341 				  (reg_val & KVM_RISCV_BASE_ISA_MASK);
342 			vcpu->arch.isa[0] = reg_val;
343 			kvm_riscv_vcpu_fp_reset(vcpu);
344 		} else {
345 			return -EOPNOTSUPP;
346 		}
347 		break;
348 	case KVM_REG_RISCV_CONFIG_REG(zicbom_block_size):
349 		return -EOPNOTSUPP;
350 	case KVM_REG_RISCV_CONFIG_REG(mvendorid):
351 		if (!vcpu->arch.ran_atleast_once)
352 			vcpu->arch.mvendorid = reg_val;
353 		else
354 			return -EBUSY;
355 		break;
356 	case KVM_REG_RISCV_CONFIG_REG(marchid):
357 		if (!vcpu->arch.ran_atleast_once)
358 			vcpu->arch.marchid = reg_val;
359 		else
360 			return -EBUSY;
361 		break;
362 	case KVM_REG_RISCV_CONFIG_REG(mimpid):
363 		if (!vcpu->arch.ran_atleast_once)
364 			vcpu->arch.mimpid = reg_val;
365 		else
366 			return -EBUSY;
367 		break;
368 	default:
369 		return -EINVAL;
370 	}
371 
372 	return 0;
373 }
374 
375 static int kvm_riscv_vcpu_get_reg_core(struct kvm_vcpu *vcpu,
376 				       const struct kvm_one_reg *reg)
377 {
378 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
379 	unsigned long __user *uaddr =
380 			(unsigned long __user *)(unsigned long)reg->addr;
381 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
382 					    KVM_REG_SIZE_MASK |
383 					    KVM_REG_RISCV_CORE);
384 	unsigned long reg_val;
385 
386 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
387 		return -EINVAL;
388 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
389 		return -EINVAL;
390 
391 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
392 		reg_val = cntx->sepc;
393 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
394 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
395 		reg_val = ((unsigned long *)cntx)[reg_num];
396 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode))
397 		reg_val = (cntx->sstatus & SR_SPP) ?
398 				KVM_RISCV_MODE_S : KVM_RISCV_MODE_U;
399 	else
400 		return -EINVAL;
401 
402 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
403 		return -EFAULT;
404 
405 	return 0;
406 }
407 
408 static int kvm_riscv_vcpu_set_reg_core(struct kvm_vcpu *vcpu,
409 				       const struct kvm_one_reg *reg)
410 {
411 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
412 	unsigned long __user *uaddr =
413 			(unsigned long __user *)(unsigned long)reg->addr;
414 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
415 					    KVM_REG_SIZE_MASK |
416 					    KVM_REG_RISCV_CORE);
417 	unsigned long reg_val;
418 
419 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
420 		return -EINVAL;
421 	if (reg_num >= sizeof(struct kvm_riscv_core) / sizeof(unsigned long))
422 		return -EINVAL;
423 
424 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
425 		return -EFAULT;
426 
427 	if (reg_num == KVM_REG_RISCV_CORE_REG(regs.pc))
428 		cntx->sepc = reg_val;
429 	else if (KVM_REG_RISCV_CORE_REG(regs.pc) < reg_num &&
430 		 reg_num <= KVM_REG_RISCV_CORE_REG(regs.t6))
431 		((unsigned long *)cntx)[reg_num] = reg_val;
432 	else if (reg_num == KVM_REG_RISCV_CORE_REG(mode)) {
433 		if (reg_val == KVM_RISCV_MODE_S)
434 			cntx->sstatus |= SR_SPP;
435 		else
436 			cntx->sstatus &= ~SR_SPP;
437 	} else
438 		return -EINVAL;
439 
440 	return 0;
441 }
442 
443 static int kvm_riscv_vcpu_get_reg_csr(struct kvm_vcpu *vcpu,
444 				      const struct kvm_one_reg *reg)
445 {
446 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
447 	unsigned long __user *uaddr =
448 			(unsigned long __user *)(unsigned long)reg->addr;
449 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
450 					    KVM_REG_SIZE_MASK |
451 					    KVM_REG_RISCV_CSR);
452 	unsigned long reg_val;
453 
454 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
455 		return -EINVAL;
456 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
457 		return -EINVAL;
458 
459 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
460 		kvm_riscv_vcpu_flush_interrupts(vcpu);
461 		reg_val = (csr->hvip >> VSIP_TO_HVIP_SHIFT) & VSIP_VALID_MASK;
462 	} else
463 		reg_val = ((unsigned long *)csr)[reg_num];
464 
465 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
466 		return -EFAULT;
467 
468 	return 0;
469 }
470 
471 static int kvm_riscv_vcpu_set_reg_csr(struct kvm_vcpu *vcpu,
472 				      const struct kvm_one_reg *reg)
473 {
474 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
475 	unsigned long __user *uaddr =
476 			(unsigned long __user *)(unsigned long)reg->addr;
477 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
478 					    KVM_REG_SIZE_MASK |
479 					    KVM_REG_RISCV_CSR);
480 	unsigned long reg_val;
481 
482 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
483 		return -EINVAL;
484 	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
485 		return -EINVAL;
486 
487 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
488 		return -EFAULT;
489 
490 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip)) {
491 		reg_val &= VSIP_VALID_MASK;
492 		reg_val <<= VSIP_TO_HVIP_SHIFT;
493 	}
494 
495 	((unsigned long *)csr)[reg_num] = reg_val;
496 
497 	if (reg_num == KVM_REG_RISCV_CSR_REG(sip))
498 		WRITE_ONCE(vcpu->arch.irqs_pending_mask, 0);
499 
500 	return 0;
501 }
502 
503 static int kvm_riscv_vcpu_get_reg_isa_ext(struct kvm_vcpu *vcpu,
504 					  const struct kvm_one_reg *reg)
505 {
506 	unsigned long __user *uaddr =
507 			(unsigned long __user *)(unsigned long)reg->addr;
508 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
509 					    KVM_REG_SIZE_MASK |
510 					    KVM_REG_RISCV_ISA_EXT);
511 	unsigned long reg_val = 0;
512 	unsigned long host_isa_ext;
513 
514 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
515 		return -EINVAL;
516 
517 	if (reg_num >= KVM_RISCV_ISA_EXT_MAX ||
518 	    reg_num >= ARRAY_SIZE(kvm_isa_ext_arr))
519 		return -EINVAL;
520 
521 	host_isa_ext = kvm_isa_ext_arr[reg_num];
522 	if (__riscv_isa_extension_available(vcpu->arch.isa, host_isa_ext))
523 		reg_val = 1; /* Mark the given extension as available */
524 
525 	if (copy_to_user(uaddr, &reg_val, KVM_REG_SIZE(reg->id)))
526 		return -EFAULT;
527 
528 	return 0;
529 }
530 
531 static int kvm_riscv_vcpu_set_reg_isa_ext(struct kvm_vcpu *vcpu,
532 					  const struct kvm_one_reg *reg)
533 {
534 	unsigned long __user *uaddr =
535 			(unsigned long __user *)(unsigned long)reg->addr;
536 	unsigned long reg_num = reg->id & ~(KVM_REG_ARCH_MASK |
537 					    KVM_REG_SIZE_MASK |
538 					    KVM_REG_RISCV_ISA_EXT);
539 	unsigned long reg_val;
540 	unsigned long host_isa_ext;
541 
542 	if (KVM_REG_SIZE(reg->id) != sizeof(unsigned long))
543 		return -EINVAL;
544 
545 	if (reg_num >= KVM_RISCV_ISA_EXT_MAX ||
546 	    reg_num >= ARRAY_SIZE(kvm_isa_ext_arr))
547 		return -EINVAL;
548 
549 	if (copy_from_user(&reg_val, uaddr, KVM_REG_SIZE(reg->id)))
550 		return -EFAULT;
551 
552 	host_isa_ext = kvm_isa_ext_arr[reg_num];
553 	if (!__riscv_isa_extension_available(NULL, host_isa_ext))
554 		return	-EOPNOTSUPP;
555 
556 	if (!vcpu->arch.ran_atleast_once) {
557 		/*
558 		 * All multi-letter extension and a few single letter
559 		 * extension can be disabled
560 		 */
561 		if (reg_val == 1 &&
562 		    kvm_riscv_vcpu_isa_enable_allowed(reg_num))
563 			set_bit(host_isa_ext, vcpu->arch.isa);
564 		else if (!reg_val &&
565 			 kvm_riscv_vcpu_isa_disable_allowed(reg_num))
566 			clear_bit(host_isa_ext, vcpu->arch.isa);
567 		else
568 			return -EINVAL;
569 		kvm_riscv_vcpu_fp_reset(vcpu);
570 	} else {
571 		return -EOPNOTSUPP;
572 	}
573 
574 	return 0;
575 }
576 
577 static int kvm_riscv_vcpu_set_reg(struct kvm_vcpu *vcpu,
578 				  const struct kvm_one_reg *reg)
579 {
580 	switch (reg->id & KVM_REG_RISCV_TYPE_MASK) {
581 	case KVM_REG_RISCV_CONFIG:
582 		return kvm_riscv_vcpu_set_reg_config(vcpu, reg);
583 	case KVM_REG_RISCV_CORE:
584 		return kvm_riscv_vcpu_set_reg_core(vcpu, reg);
585 	case KVM_REG_RISCV_CSR:
586 		return kvm_riscv_vcpu_set_reg_csr(vcpu, reg);
587 	case KVM_REG_RISCV_TIMER:
588 		return kvm_riscv_vcpu_set_reg_timer(vcpu, reg);
589 	case KVM_REG_RISCV_FP_F:
590 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
591 						 KVM_REG_RISCV_FP_F);
592 	case KVM_REG_RISCV_FP_D:
593 		return kvm_riscv_vcpu_set_reg_fp(vcpu, reg,
594 						 KVM_REG_RISCV_FP_D);
595 	case KVM_REG_RISCV_ISA_EXT:
596 		return kvm_riscv_vcpu_set_reg_isa_ext(vcpu, reg);
597 	default:
598 		break;
599 	}
600 
601 	return -EINVAL;
602 }
603 
604 static int kvm_riscv_vcpu_get_reg(struct kvm_vcpu *vcpu,
605 				  const struct kvm_one_reg *reg)
606 {
607 	switch (reg->id & KVM_REG_RISCV_TYPE_MASK) {
608 	case KVM_REG_RISCV_CONFIG:
609 		return kvm_riscv_vcpu_get_reg_config(vcpu, reg);
610 	case KVM_REG_RISCV_CORE:
611 		return kvm_riscv_vcpu_get_reg_core(vcpu, reg);
612 	case KVM_REG_RISCV_CSR:
613 		return kvm_riscv_vcpu_get_reg_csr(vcpu, reg);
614 	case KVM_REG_RISCV_TIMER:
615 		return kvm_riscv_vcpu_get_reg_timer(vcpu, reg);
616 	case KVM_REG_RISCV_FP_F:
617 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
618 						 KVM_REG_RISCV_FP_F);
619 	case KVM_REG_RISCV_FP_D:
620 		return kvm_riscv_vcpu_get_reg_fp(vcpu, reg,
621 						 KVM_REG_RISCV_FP_D);
622 	case KVM_REG_RISCV_ISA_EXT:
623 		return kvm_riscv_vcpu_get_reg_isa_ext(vcpu, reg);
624 	default:
625 		break;
626 	}
627 
628 	return -EINVAL;
629 }
630 
631 long kvm_arch_vcpu_async_ioctl(struct file *filp,
632 			       unsigned int ioctl, unsigned long arg)
633 {
634 	struct kvm_vcpu *vcpu = filp->private_data;
635 	void __user *argp = (void __user *)arg;
636 
637 	if (ioctl == KVM_INTERRUPT) {
638 		struct kvm_interrupt irq;
639 
640 		if (copy_from_user(&irq, argp, sizeof(irq)))
641 			return -EFAULT;
642 
643 		if (irq.irq == KVM_INTERRUPT_SET)
644 			return kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT);
645 		else
646 			return kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
647 	}
648 
649 	return -ENOIOCTLCMD;
650 }
651 
652 long kvm_arch_vcpu_ioctl(struct file *filp,
653 			 unsigned int ioctl, unsigned long arg)
654 {
655 	struct kvm_vcpu *vcpu = filp->private_data;
656 	void __user *argp = (void __user *)arg;
657 	long r = -EINVAL;
658 
659 	switch (ioctl) {
660 	case KVM_SET_ONE_REG:
661 	case KVM_GET_ONE_REG: {
662 		struct kvm_one_reg reg;
663 
664 		r = -EFAULT;
665 		if (copy_from_user(&reg, argp, sizeof(reg)))
666 			break;
667 
668 		if (ioctl == KVM_SET_ONE_REG)
669 			r = kvm_riscv_vcpu_set_reg(vcpu, &reg);
670 		else
671 			r = kvm_riscv_vcpu_get_reg(vcpu, &reg);
672 		break;
673 	}
674 	default:
675 		break;
676 	}
677 
678 	return r;
679 }
680 
681 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
682 				  struct kvm_sregs *sregs)
683 {
684 	return -EINVAL;
685 }
686 
687 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
688 				  struct kvm_sregs *sregs)
689 {
690 	return -EINVAL;
691 }
692 
693 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
694 {
695 	return -EINVAL;
696 }
697 
698 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
699 {
700 	return -EINVAL;
701 }
702 
703 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
704 				  struct kvm_translation *tr)
705 {
706 	return -EINVAL;
707 }
708 
709 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
710 {
711 	return -EINVAL;
712 }
713 
714 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
715 {
716 	return -EINVAL;
717 }
718 
719 void kvm_riscv_vcpu_flush_interrupts(struct kvm_vcpu *vcpu)
720 {
721 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
722 	unsigned long mask, val;
723 
724 	if (READ_ONCE(vcpu->arch.irqs_pending_mask)) {
725 		mask = xchg_acquire(&vcpu->arch.irqs_pending_mask, 0);
726 		val = READ_ONCE(vcpu->arch.irqs_pending) & mask;
727 
728 		csr->hvip &= ~mask;
729 		csr->hvip |= val;
730 	}
731 }
732 
733 void kvm_riscv_vcpu_sync_interrupts(struct kvm_vcpu *vcpu)
734 {
735 	unsigned long hvip;
736 	struct kvm_vcpu_arch *v = &vcpu->arch;
737 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
738 
739 	/* Read current HVIP and VSIE CSRs */
740 	csr->vsie = csr_read(CSR_VSIE);
741 
742 	/* Sync-up HVIP.VSSIP bit changes does by Guest */
743 	hvip = csr_read(CSR_HVIP);
744 	if ((csr->hvip ^ hvip) & (1UL << IRQ_VS_SOFT)) {
745 		if (hvip & (1UL << IRQ_VS_SOFT)) {
746 			if (!test_and_set_bit(IRQ_VS_SOFT,
747 					      &v->irqs_pending_mask))
748 				set_bit(IRQ_VS_SOFT, &v->irqs_pending);
749 		} else {
750 			if (!test_and_set_bit(IRQ_VS_SOFT,
751 					      &v->irqs_pending_mask))
752 				clear_bit(IRQ_VS_SOFT, &v->irqs_pending);
753 		}
754 	}
755 
756 	/* Sync-up timer CSRs */
757 	kvm_riscv_vcpu_timer_sync(vcpu);
758 }
759 
760 int kvm_riscv_vcpu_set_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
761 {
762 	if (irq != IRQ_VS_SOFT &&
763 	    irq != IRQ_VS_TIMER &&
764 	    irq != IRQ_VS_EXT)
765 		return -EINVAL;
766 
767 	set_bit(irq, &vcpu->arch.irqs_pending);
768 	smp_mb__before_atomic();
769 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
770 
771 	kvm_vcpu_kick(vcpu);
772 
773 	return 0;
774 }
775 
776 int kvm_riscv_vcpu_unset_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
777 {
778 	if (irq != IRQ_VS_SOFT &&
779 	    irq != IRQ_VS_TIMER &&
780 	    irq != IRQ_VS_EXT)
781 		return -EINVAL;
782 
783 	clear_bit(irq, &vcpu->arch.irqs_pending);
784 	smp_mb__before_atomic();
785 	set_bit(irq, &vcpu->arch.irqs_pending_mask);
786 
787 	return 0;
788 }
789 
790 bool kvm_riscv_vcpu_has_interrupts(struct kvm_vcpu *vcpu, unsigned long mask)
791 {
792 	unsigned long ie = ((vcpu->arch.guest_csr.vsie & VSIP_VALID_MASK)
793 			    << VSIP_TO_HVIP_SHIFT) & mask;
794 
795 	return (READ_ONCE(vcpu->arch.irqs_pending) & ie) ? true : false;
796 }
797 
798 void kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu)
799 {
800 	vcpu->arch.power_off = true;
801 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
802 	kvm_vcpu_kick(vcpu);
803 }
804 
805 void kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu)
806 {
807 	vcpu->arch.power_off = false;
808 	kvm_vcpu_wake_up(vcpu);
809 }
810 
811 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
812 				    struct kvm_mp_state *mp_state)
813 {
814 	if (vcpu->arch.power_off)
815 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
816 	else
817 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
818 
819 	return 0;
820 }
821 
822 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
823 				    struct kvm_mp_state *mp_state)
824 {
825 	int ret = 0;
826 
827 	switch (mp_state->mp_state) {
828 	case KVM_MP_STATE_RUNNABLE:
829 		vcpu->arch.power_off = false;
830 		break;
831 	case KVM_MP_STATE_STOPPED:
832 		kvm_riscv_vcpu_power_off(vcpu);
833 		break;
834 	default:
835 		ret = -EINVAL;
836 	}
837 
838 	return ret;
839 }
840 
841 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
842 					struct kvm_guest_debug *dbg)
843 {
844 	/* TODO; To be implemented later. */
845 	return -EINVAL;
846 }
847 
848 static void kvm_riscv_vcpu_update_config(const unsigned long *isa)
849 {
850 	u64 henvcfg = 0;
851 
852 	if (riscv_isa_extension_available(isa, SVPBMT))
853 		henvcfg |= ENVCFG_PBMTE;
854 
855 	if (riscv_isa_extension_available(isa, SSTC))
856 		henvcfg |= ENVCFG_STCE;
857 
858 	if (riscv_isa_extension_available(isa, ZICBOM))
859 		henvcfg |= (ENVCFG_CBIE | ENVCFG_CBCFE);
860 
861 	csr_write(CSR_HENVCFG, henvcfg);
862 #ifdef CONFIG_32BIT
863 	csr_write(CSR_HENVCFGH, henvcfg >> 32);
864 #endif
865 }
866 
867 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
868 {
869 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
870 
871 	csr_write(CSR_VSSTATUS, csr->vsstatus);
872 	csr_write(CSR_VSIE, csr->vsie);
873 	csr_write(CSR_VSTVEC, csr->vstvec);
874 	csr_write(CSR_VSSCRATCH, csr->vsscratch);
875 	csr_write(CSR_VSEPC, csr->vsepc);
876 	csr_write(CSR_VSCAUSE, csr->vscause);
877 	csr_write(CSR_VSTVAL, csr->vstval);
878 	csr_write(CSR_HVIP, csr->hvip);
879 	csr_write(CSR_VSATP, csr->vsatp);
880 
881 	kvm_riscv_vcpu_update_config(vcpu->arch.isa);
882 
883 	kvm_riscv_gstage_update_hgatp(vcpu);
884 
885 	kvm_riscv_vcpu_timer_restore(vcpu);
886 
887 	kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context);
888 	kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context,
889 					vcpu->arch.isa);
890 
891 	vcpu->cpu = cpu;
892 }
893 
894 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
895 {
896 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
897 
898 	vcpu->cpu = -1;
899 
900 	kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context,
901 				     vcpu->arch.isa);
902 	kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context);
903 
904 	kvm_riscv_vcpu_timer_save(vcpu);
905 
906 	csr->vsstatus = csr_read(CSR_VSSTATUS);
907 	csr->vsie = csr_read(CSR_VSIE);
908 	csr->vstvec = csr_read(CSR_VSTVEC);
909 	csr->vsscratch = csr_read(CSR_VSSCRATCH);
910 	csr->vsepc = csr_read(CSR_VSEPC);
911 	csr->vscause = csr_read(CSR_VSCAUSE);
912 	csr->vstval = csr_read(CSR_VSTVAL);
913 	csr->hvip = csr_read(CSR_HVIP);
914 	csr->vsatp = csr_read(CSR_VSATP);
915 }
916 
917 static void kvm_riscv_check_vcpu_requests(struct kvm_vcpu *vcpu)
918 {
919 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
920 
921 	if (kvm_request_pending(vcpu)) {
922 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) {
923 			kvm_vcpu_srcu_read_unlock(vcpu);
924 			rcuwait_wait_event(wait,
925 				(!vcpu->arch.power_off) && (!vcpu->arch.pause),
926 				TASK_INTERRUPTIBLE);
927 			kvm_vcpu_srcu_read_lock(vcpu);
928 
929 			if (vcpu->arch.power_off || vcpu->arch.pause) {
930 				/*
931 				 * Awaken to handle a signal, request to
932 				 * sleep again later.
933 				 */
934 				kvm_make_request(KVM_REQ_SLEEP, vcpu);
935 			}
936 		}
937 
938 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
939 			kvm_riscv_reset_vcpu(vcpu);
940 
941 		if (kvm_check_request(KVM_REQ_UPDATE_HGATP, vcpu))
942 			kvm_riscv_gstage_update_hgatp(vcpu);
943 
944 		if (kvm_check_request(KVM_REQ_FENCE_I, vcpu))
945 			kvm_riscv_fence_i_process(vcpu);
946 
947 		/*
948 		 * The generic KVM_REQ_TLB_FLUSH is same as
949 		 * KVM_REQ_HFENCE_GVMA_VMID_ALL
950 		 */
951 		if (kvm_check_request(KVM_REQ_HFENCE_GVMA_VMID_ALL, vcpu))
952 			kvm_riscv_hfence_gvma_vmid_all_process(vcpu);
953 
954 		if (kvm_check_request(KVM_REQ_HFENCE_VVMA_ALL, vcpu))
955 			kvm_riscv_hfence_vvma_all_process(vcpu);
956 
957 		if (kvm_check_request(KVM_REQ_HFENCE, vcpu))
958 			kvm_riscv_hfence_process(vcpu);
959 	}
960 }
961 
962 static void kvm_riscv_update_hvip(struct kvm_vcpu *vcpu)
963 {
964 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
965 
966 	csr_write(CSR_HVIP, csr->hvip);
967 }
968 
969 /*
970  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
971  * the vCPU is running.
972  *
973  * This must be noinstr as instrumentation may make use of RCU, and this is not
974  * safe during the EQS.
975  */
976 static void noinstr kvm_riscv_vcpu_enter_exit(struct kvm_vcpu *vcpu)
977 {
978 	guest_state_enter_irqoff();
979 	__kvm_riscv_switch_to(&vcpu->arch);
980 	vcpu->arch.last_exit_cpu = vcpu->cpu;
981 	guest_state_exit_irqoff();
982 }
983 
984 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
985 {
986 	int ret;
987 	struct kvm_cpu_trap trap;
988 	struct kvm_run *run = vcpu->run;
989 
990 	/* Mark this VCPU ran at least once */
991 	vcpu->arch.ran_atleast_once = true;
992 
993 	kvm_vcpu_srcu_read_lock(vcpu);
994 
995 	switch (run->exit_reason) {
996 	case KVM_EXIT_MMIO:
997 		/* Process MMIO value returned from user-space */
998 		ret = kvm_riscv_vcpu_mmio_return(vcpu, vcpu->run);
999 		break;
1000 	case KVM_EXIT_RISCV_SBI:
1001 		/* Process SBI value returned from user-space */
1002 		ret = kvm_riscv_vcpu_sbi_return(vcpu, vcpu->run);
1003 		break;
1004 	case KVM_EXIT_RISCV_CSR:
1005 		/* Process CSR value returned from user-space */
1006 		ret = kvm_riscv_vcpu_csr_return(vcpu, vcpu->run);
1007 		break;
1008 	default:
1009 		ret = 0;
1010 		break;
1011 	}
1012 	if (ret) {
1013 		kvm_vcpu_srcu_read_unlock(vcpu);
1014 		return ret;
1015 	}
1016 
1017 	if (run->immediate_exit) {
1018 		kvm_vcpu_srcu_read_unlock(vcpu);
1019 		return -EINTR;
1020 	}
1021 
1022 	vcpu_load(vcpu);
1023 
1024 	kvm_sigset_activate(vcpu);
1025 
1026 	ret = 1;
1027 	run->exit_reason = KVM_EXIT_UNKNOWN;
1028 	while (ret > 0) {
1029 		/* Check conditions before entering the guest */
1030 		ret = xfer_to_guest_mode_handle_work(vcpu);
1031 		if (ret)
1032 			continue;
1033 		ret = 1;
1034 
1035 		kvm_riscv_gstage_vmid_update(vcpu);
1036 
1037 		kvm_riscv_check_vcpu_requests(vcpu);
1038 
1039 		local_irq_disable();
1040 
1041 		/*
1042 		 * Ensure we set mode to IN_GUEST_MODE after we disable
1043 		 * interrupts and before the final VCPU requests check.
1044 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
1045 		 * Documentation/virt/kvm/vcpu-requests.rst
1046 		 */
1047 		vcpu->mode = IN_GUEST_MODE;
1048 
1049 		kvm_vcpu_srcu_read_unlock(vcpu);
1050 		smp_mb__after_srcu_read_unlock();
1051 
1052 		/*
1053 		 * We might have got VCPU interrupts updated asynchronously
1054 		 * so update it in HW.
1055 		 */
1056 		kvm_riscv_vcpu_flush_interrupts(vcpu);
1057 
1058 		/* Update HVIP CSR for current CPU */
1059 		kvm_riscv_update_hvip(vcpu);
1060 
1061 		if (ret <= 0 ||
1062 		    kvm_riscv_gstage_vmid_ver_changed(&vcpu->kvm->arch.vmid) ||
1063 		    kvm_request_pending(vcpu) ||
1064 		    xfer_to_guest_mode_work_pending()) {
1065 			vcpu->mode = OUTSIDE_GUEST_MODE;
1066 			local_irq_enable();
1067 			kvm_vcpu_srcu_read_lock(vcpu);
1068 			continue;
1069 		}
1070 
1071 		/*
1072 		 * Cleanup stale TLB enteries
1073 		 *
1074 		 * Note: This should be done after G-stage VMID has been
1075 		 * updated using kvm_riscv_gstage_vmid_ver_changed()
1076 		 */
1077 		kvm_riscv_local_tlb_sanitize(vcpu);
1078 
1079 		guest_timing_enter_irqoff();
1080 
1081 		kvm_riscv_vcpu_enter_exit(vcpu);
1082 
1083 		vcpu->mode = OUTSIDE_GUEST_MODE;
1084 		vcpu->stat.exits++;
1085 
1086 		/*
1087 		 * Save SCAUSE, STVAL, HTVAL, and HTINST because we might
1088 		 * get an interrupt between __kvm_riscv_switch_to() and
1089 		 * local_irq_enable() which can potentially change CSRs.
1090 		 */
1091 		trap.sepc = vcpu->arch.guest_context.sepc;
1092 		trap.scause = csr_read(CSR_SCAUSE);
1093 		trap.stval = csr_read(CSR_STVAL);
1094 		trap.htval = csr_read(CSR_HTVAL);
1095 		trap.htinst = csr_read(CSR_HTINST);
1096 
1097 		/* Syncup interrupts state with HW */
1098 		kvm_riscv_vcpu_sync_interrupts(vcpu);
1099 
1100 		preempt_disable();
1101 
1102 		/*
1103 		 * We must ensure that any pending interrupts are taken before
1104 		 * we exit guest timing so that timer ticks are accounted as
1105 		 * guest time. Transiently unmask interrupts so that any
1106 		 * pending interrupts are taken.
1107 		 *
1108 		 * There's no barrier which ensures that pending interrupts are
1109 		 * recognised, so we just hope that the CPU takes any pending
1110 		 * interrupts between the enable and disable.
1111 		 */
1112 		local_irq_enable();
1113 		local_irq_disable();
1114 
1115 		guest_timing_exit_irqoff();
1116 
1117 		local_irq_enable();
1118 
1119 		preempt_enable();
1120 
1121 		kvm_vcpu_srcu_read_lock(vcpu);
1122 
1123 		ret = kvm_riscv_vcpu_exit(vcpu, run, &trap);
1124 	}
1125 
1126 	kvm_sigset_deactivate(vcpu);
1127 
1128 	vcpu_put(vcpu);
1129 
1130 	kvm_vcpu_srcu_read_unlock(vcpu);
1131 
1132 	return ret;
1133 }
1134