xref: /linux/arch/riscv/kvm/vcpu.c (revision f85f5ae45ad945270a8884261de8249431e8b5a6)
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/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/cacheflush.h>
22 #include <asm/kvm_vcpu_vector.h>
23 
24 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
25 	KVM_GENERIC_VCPU_STATS(),
26 	STATS_DESC_COUNTER(VCPU, ecall_exit_stat),
27 	STATS_DESC_COUNTER(VCPU, wfi_exit_stat),
28 	STATS_DESC_COUNTER(VCPU, mmio_exit_user),
29 	STATS_DESC_COUNTER(VCPU, mmio_exit_kernel),
30 	STATS_DESC_COUNTER(VCPU, csr_exit_user),
31 	STATS_DESC_COUNTER(VCPU, csr_exit_kernel),
32 	STATS_DESC_COUNTER(VCPU, signal_exits),
33 	STATS_DESC_COUNTER(VCPU, exits)
34 };
35 
36 const struct kvm_stats_header kvm_vcpu_stats_header = {
37 	.name_size = KVM_STATS_NAME_SIZE,
38 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
39 	.id_offset = sizeof(struct kvm_stats_header),
40 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
41 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
42 		       sizeof(kvm_vcpu_stats_desc),
43 };
44 
45 static void kvm_riscv_reset_vcpu(struct kvm_vcpu *vcpu)
46 {
47 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
48 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
49 	struct kvm_cpu_context *cntx = &vcpu->arch.guest_context;
50 	struct kvm_cpu_context *reset_cntx = &vcpu->arch.guest_reset_context;
51 	bool loaded;
52 
53 	/**
54 	 * The preemption should be disabled here because it races with
55 	 * kvm_sched_out/kvm_sched_in(called from preempt notifiers) which
56 	 * also calls vcpu_load/put.
57 	 */
58 	get_cpu();
59 	loaded = (vcpu->cpu != -1);
60 	if (loaded)
61 		kvm_arch_vcpu_put(vcpu);
62 
63 	vcpu->arch.last_exit_cpu = -1;
64 
65 	memcpy(csr, reset_csr, sizeof(*csr));
66 
67 	memcpy(cntx, reset_cntx, sizeof(*cntx));
68 
69 	kvm_riscv_vcpu_fp_reset(vcpu);
70 
71 	kvm_riscv_vcpu_vector_reset(vcpu);
72 
73 	kvm_riscv_vcpu_timer_reset(vcpu);
74 
75 	kvm_riscv_vcpu_aia_reset(vcpu);
76 
77 	bitmap_zero(vcpu->arch.irqs_pending, KVM_RISCV_VCPU_NR_IRQS);
78 	bitmap_zero(vcpu->arch.irqs_pending_mask, KVM_RISCV_VCPU_NR_IRQS);
79 
80 	kvm_riscv_vcpu_pmu_reset(vcpu);
81 
82 	vcpu->arch.hfence_head = 0;
83 	vcpu->arch.hfence_tail = 0;
84 	memset(vcpu->arch.hfence_queue, 0, sizeof(vcpu->arch.hfence_queue));
85 
86 	/* Reset the guest CSRs for hotplug usecase */
87 	if (loaded)
88 		kvm_arch_vcpu_load(vcpu, smp_processor_id());
89 	put_cpu();
90 }
91 
92 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
93 {
94 	return 0;
95 }
96 
97 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
98 {
99 	int rc;
100 	struct kvm_cpu_context *cntx;
101 	struct kvm_vcpu_csr *reset_csr = &vcpu->arch.guest_reset_csr;
102 
103 	/* Mark this VCPU never ran */
104 	vcpu->arch.ran_atleast_once = false;
105 	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
106 	bitmap_zero(vcpu->arch.isa, RISCV_ISA_EXT_MAX);
107 
108 	/* Setup ISA features available to VCPU */
109 	kvm_riscv_vcpu_setup_isa(vcpu);
110 
111 	/* Setup vendor, arch, and implementation details */
112 	vcpu->arch.mvendorid = sbi_get_mvendorid();
113 	vcpu->arch.marchid = sbi_get_marchid();
114 	vcpu->arch.mimpid = sbi_get_mimpid();
115 
116 	/* Setup VCPU hfence queue */
117 	spin_lock_init(&vcpu->arch.hfence_lock);
118 
119 	/* Setup reset state of shadow SSTATUS and HSTATUS CSRs */
120 	cntx = &vcpu->arch.guest_reset_context;
121 	cntx->sstatus = SR_SPP | SR_SPIE;
122 	cntx->hstatus = 0;
123 	cntx->hstatus |= HSTATUS_VTW;
124 	cntx->hstatus |= HSTATUS_SPVP;
125 	cntx->hstatus |= HSTATUS_SPV;
126 
127 	if (kvm_riscv_vcpu_alloc_vector_context(vcpu, cntx))
128 		return -ENOMEM;
129 
130 	/* By default, make CY, TM, and IR counters accessible in VU mode */
131 	reset_csr->scounteren = 0x7;
132 
133 	/* Setup VCPU timer */
134 	kvm_riscv_vcpu_timer_init(vcpu);
135 
136 	/* setup performance monitoring */
137 	kvm_riscv_vcpu_pmu_init(vcpu);
138 
139 	/* Setup VCPU AIA */
140 	rc = kvm_riscv_vcpu_aia_init(vcpu);
141 	if (rc)
142 		return rc;
143 
144 	/* Reset VCPU */
145 	kvm_riscv_reset_vcpu(vcpu);
146 
147 	return 0;
148 }
149 
150 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
151 {
152 	/**
153 	 * vcpu with id 0 is the designated boot cpu.
154 	 * Keep all vcpus with non-zero id in power-off state so that
155 	 * they can be brought up using SBI HSM extension.
156 	 */
157 	if (vcpu->vcpu_idx != 0)
158 		kvm_riscv_vcpu_power_off(vcpu);
159 }
160 
161 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
162 {
163 	/* Cleanup VCPU AIA context */
164 	kvm_riscv_vcpu_aia_deinit(vcpu);
165 
166 	/* Cleanup VCPU timer */
167 	kvm_riscv_vcpu_timer_deinit(vcpu);
168 
169 	kvm_riscv_vcpu_pmu_deinit(vcpu);
170 
171 	/* Free unused pages pre-allocated for G-stage page table mappings */
172 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
173 
174 	/* Free vector context space for host and guest kernel */
175 	kvm_riscv_vcpu_free_vector_context(vcpu);
176 }
177 
178 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
179 {
180 	return kvm_riscv_vcpu_timer_pending(vcpu);
181 }
182 
183 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
184 {
185 	kvm_riscv_aia_wakeon_hgei(vcpu, true);
186 }
187 
188 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
189 {
190 	kvm_riscv_aia_wakeon_hgei(vcpu, false);
191 }
192 
193 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
194 {
195 	return (kvm_riscv_vcpu_has_interrupts(vcpu, -1UL) &&
196 		!vcpu->arch.power_off && !vcpu->arch.pause);
197 }
198 
199 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
200 {
201 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
202 }
203 
204 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
205 {
206 	return (vcpu->arch.guest_context.sstatus & SR_SPP) ? true : false;
207 }
208 
209 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
210 {
211 	return VM_FAULT_SIGBUS;
212 }
213 
214 long kvm_arch_vcpu_async_ioctl(struct file *filp,
215 			       unsigned int ioctl, unsigned long arg)
216 {
217 	struct kvm_vcpu *vcpu = filp->private_data;
218 	void __user *argp = (void __user *)arg;
219 
220 	if (ioctl == KVM_INTERRUPT) {
221 		struct kvm_interrupt irq;
222 
223 		if (copy_from_user(&irq, argp, sizeof(irq)))
224 			return -EFAULT;
225 
226 		if (irq.irq == KVM_INTERRUPT_SET)
227 			return kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT);
228 		else
229 			return kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
230 	}
231 
232 	return -ENOIOCTLCMD;
233 }
234 
235 long kvm_arch_vcpu_ioctl(struct file *filp,
236 			 unsigned int ioctl, unsigned long arg)
237 {
238 	struct kvm_vcpu *vcpu = filp->private_data;
239 	void __user *argp = (void __user *)arg;
240 	long r = -EINVAL;
241 
242 	switch (ioctl) {
243 	case KVM_SET_ONE_REG:
244 	case KVM_GET_ONE_REG: {
245 		struct kvm_one_reg reg;
246 
247 		r = -EFAULT;
248 		if (copy_from_user(&reg, argp, sizeof(reg)))
249 			break;
250 
251 		if (ioctl == KVM_SET_ONE_REG)
252 			r = kvm_riscv_vcpu_set_reg(vcpu, &reg);
253 		else
254 			r = kvm_riscv_vcpu_get_reg(vcpu, &reg);
255 		break;
256 	}
257 	case KVM_GET_REG_LIST: {
258 		struct kvm_reg_list __user *user_list = argp;
259 		struct kvm_reg_list reg_list;
260 		unsigned int n;
261 
262 		r = -EFAULT;
263 		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
264 			break;
265 		n = reg_list.n;
266 		reg_list.n = kvm_riscv_vcpu_num_regs(vcpu);
267 		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
268 			break;
269 		r = -E2BIG;
270 		if (n < reg_list.n)
271 			break;
272 		r = kvm_riscv_vcpu_copy_reg_indices(vcpu, user_list->reg);
273 		break;
274 	}
275 	default:
276 		break;
277 	}
278 
279 	return r;
280 }
281 
282 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
283 				  struct kvm_sregs *sregs)
284 {
285 	return -EINVAL;
286 }
287 
288 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
289 				  struct kvm_sregs *sregs)
290 {
291 	return -EINVAL;
292 }
293 
294 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
295 {
296 	return -EINVAL;
297 }
298 
299 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
300 {
301 	return -EINVAL;
302 }
303 
304 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
305 				  struct kvm_translation *tr)
306 {
307 	return -EINVAL;
308 }
309 
310 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
311 {
312 	return -EINVAL;
313 }
314 
315 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
316 {
317 	return -EINVAL;
318 }
319 
320 void kvm_riscv_vcpu_flush_interrupts(struct kvm_vcpu *vcpu)
321 {
322 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
323 	unsigned long mask, val;
324 
325 	if (READ_ONCE(vcpu->arch.irqs_pending_mask[0])) {
326 		mask = xchg_acquire(&vcpu->arch.irqs_pending_mask[0], 0);
327 		val = READ_ONCE(vcpu->arch.irqs_pending[0]) & mask;
328 
329 		csr->hvip &= ~mask;
330 		csr->hvip |= val;
331 	}
332 
333 	/* Flush AIA high interrupts */
334 	kvm_riscv_vcpu_aia_flush_interrupts(vcpu);
335 }
336 
337 void kvm_riscv_vcpu_sync_interrupts(struct kvm_vcpu *vcpu)
338 {
339 	unsigned long hvip;
340 	struct kvm_vcpu_arch *v = &vcpu->arch;
341 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
342 
343 	/* Read current HVIP and VSIE CSRs */
344 	csr->vsie = csr_read(CSR_VSIE);
345 
346 	/* Sync-up HVIP.VSSIP bit changes does by Guest */
347 	hvip = csr_read(CSR_HVIP);
348 	if ((csr->hvip ^ hvip) & (1UL << IRQ_VS_SOFT)) {
349 		if (hvip & (1UL << IRQ_VS_SOFT)) {
350 			if (!test_and_set_bit(IRQ_VS_SOFT,
351 					      v->irqs_pending_mask))
352 				set_bit(IRQ_VS_SOFT, v->irqs_pending);
353 		} else {
354 			if (!test_and_set_bit(IRQ_VS_SOFT,
355 					      v->irqs_pending_mask))
356 				clear_bit(IRQ_VS_SOFT, v->irqs_pending);
357 		}
358 	}
359 
360 	/* Sync-up AIA high interrupts */
361 	kvm_riscv_vcpu_aia_sync_interrupts(vcpu);
362 
363 	/* Sync-up timer CSRs */
364 	kvm_riscv_vcpu_timer_sync(vcpu);
365 }
366 
367 int kvm_riscv_vcpu_set_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
368 {
369 	/*
370 	 * We only allow VS-mode software, timer, and external
371 	 * interrupts when irq is one of the local interrupts
372 	 * defined by RISC-V privilege specification.
373 	 */
374 	if (irq < IRQ_LOCAL_MAX &&
375 	    irq != IRQ_VS_SOFT &&
376 	    irq != IRQ_VS_TIMER &&
377 	    irq != IRQ_VS_EXT)
378 		return -EINVAL;
379 
380 	set_bit(irq, vcpu->arch.irqs_pending);
381 	smp_mb__before_atomic();
382 	set_bit(irq, vcpu->arch.irqs_pending_mask);
383 
384 	kvm_vcpu_kick(vcpu);
385 
386 	return 0;
387 }
388 
389 int kvm_riscv_vcpu_unset_interrupt(struct kvm_vcpu *vcpu, unsigned int irq)
390 {
391 	/*
392 	 * We only allow VS-mode software, timer, and external
393 	 * interrupts when irq is one of the local interrupts
394 	 * defined by RISC-V privilege specification.
395 	 */
396 	if (irq < IRQ_LOCAL_MAX &&
397 	    irq != IRQ_VS_SOFT &&
398 	    irq != IRQ_VS_TIMER &&
399 	    irq != IRQ_VS_EXT)
400 		return -EINVAL;
401 
402 	clear_bit(irq, vcpu->arch.irqs_pending);
403 	smp_mb__before_atomic();
404 	set_bit(irq, vcpu->arch.irqs_pending_mask);
405 
406 	return 0;
407 }
408 
409 bool kvm_riscv_vcpu_has_interrupts(struct kvm_vcpu *vcpu, u64 mask)
410 {
411 	unsigned long ie;
412 
413 	ie = ((vcpu->arch.guest_csr.vsie & VSIP_VALID_MASK)
414 		<< VSIP_TO_HVIP_SHIFT) & (unsigned long)mask;
415 	ie |= vcpu->arch.guest_csr.vsie & ~IRQ_LOCAL_MASK &
416 		(unsigned long)mask;
417 	if (READ_ONCE(vcpu->arch.irqs_pending[0]) & ie)
418 		return true;
419 
420 	/* Check AIA high interrupts */
421 	return kvm_riscv_vcpu_aia_has_interrupts(vcpu, mask);
422 }
423 
424 void kvm_riscv_vcpu_power_off(struct kvm_vcpu *vcpu)
425 {
426 	vcpu->arch.power_off = true;
427 	kvm_make_request(KVM_REQ_SLEEP, vcpu);
428 	kvm_vcpu_kick(vcpu);
429 }
430 
431 void kvm_riscv_vcpu_power_on(struct kvm_vcpu *vcpu)
432 {
433 	vcpu->arch.power_off = false;
434 	kvm_vcpu_wake_up(vcpu);
435 }
436 
437 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
438 				    struct kvm_mp_state *mp_state)
439 {
440 	if (vcpu->arch.power_off)
441 		mp_state->mp_state = KVM_MP_STATE_STOPPED;
442 	else
443 		mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
444 
445 	return 0;
446 }
447 
448 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
449 				    struct kvm_mp_state *mp_state)
450 {
451 	int ret = 0;
452 
453 	switch (mp_state->mp_state) {
454 	case KVM_MP_STATE_RUNNABLE:
455 		vcpu->arch.power_off = false;
456 		break;
457 	case KVM_MP_STATE_STOPPED:
458 		kvm_riscv_vcpu_power_off(vcpu);
459 		break;
460 	default:
461 		ret = -EINVAL;
462 	}
463 
464 	return ret;
465 }
466 
467 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
468 					struct kvm_guest_debug *dbg)
469 {
470 	/* TODO; To be implemented later. */
471 	return -EINVAL;
472 }
473 
474 static void kvm_riscv_vcpu_update_config(const unsigned long *isa)
475 {
476 	u64 henvcfg = 0;
477 
478 	if (riscv_isa_extension_available(isa, SVPBMT))
479 		henvcfg |= ENVCFG_PBMTE;
480 
481 	if (riscv_isa_extension_available(isa, SSTC))
482 		henvcfg |= ENVCFG_STCE;
483 
484 	if (riscv_isa_extension_available(isa, ZICBOM))
485 		henvcfg |= (ENVCFG_CBIE | ENVCFG_CBCFE);
486 
487 	if (riscv_isa_extension_available(isa, ZICBOZ))
488 		henvcfg |= ENVCFG_CBZE;
489 
490 	csr_write(CSR_HENVCFG, henvcfg);
491 #ifdef CONFIG_32BIT
492 	csr_write(CSR_HENVCFGH, henvcfg >> 32);
493 #endif
494 }
495 
496 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
497 {
498 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
499 
500 	csr_write(CSR_VSSTATUS, csr->vsstatus);
501 	csr_write(CSR_VSIE, csr->vsie);
502 	csr_write(CSR_VSTVEC, csr->vstvec);
503 	csr_write(CSR_VSSCRATCH, csr->vsscratch);
504 	csr_write(CSR_VSEPC, csr->vsepc);
505 	csr_write(CSR_VSCAUSE, csr->vscause);
506 	csr_write(CSR_VSTVAL, csr->vstval);
507 	csr_write(CSR_HVIP, csr->hvip);
508 	csr_write(CSR_VSATP, csr->vsatp);
509 
510 	kvm_riscv_vcpu_update_config(vcpu->arch.isa);
511 
512 	kvm_riscv_gstage_update_hgatp(vcpu);
513 
514 	kvm_riscv_vcpu_timer_restore(vcpu);
515 
516 	kvm_riscv_vcpu_host_fp_save(&vcpu->arch.host_context);
517 	kvm_riscv_vcpu_guest_fp_restore(&vcpu->arch.guest_context,
518 					vcpu->arch.isa);
519 	kvm_riscv_vcpu_host_vector_save(&vcpu->arch.host_context);
520 	kvm_riscv_vcpu_guest_vector_restore(&vcpu->arch.guest_context,
521 					    vcpu->arch.isa);
522 
523 	kvm_riscv_vcpu_aia_load(vcpu, cpu);
524 
525 	vcpu->cpu = cpu;
526 }
527 
528 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
529 {
530 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
531 
532 	vcpu->cpu = -1;
533 
534 	kvm_riscv_vcpu_aia_put(vcpu);
535 
536 	kvm_riscv_vcpu_guest_fp_save(&vcpu->arch.guest_context,
537 				     vcpu->arch.isa);
538 	kvm_riscv_vcpu_host_fp_restore(&vcpu->arch.host_context);
539 
540 	kvm_riscv_vcpu_timer_save(vcpu);
541 	kvm_riscv_vcpu_guest_vector_save(&vcpu->arch.guest_context,
542 					 vcpu->arch.isa);
543 	kvm_riscv_vcpu_host_vector_restore(&vcpu->arch.host_context);
544 
545 	csr->vsstatus = csr_read(CSR_VSSTATUS);
546 	csr->vsie = csr_read(CSR_VSIE);
547 	csr->vstvec = csr_read(CSR_VSTVEC);
548 	csr->vsscratch = csr_read(CSR_VSSCRATCH);
549 	csr->vsepc = csr_read(CSR_VSEPC);
550 	csr->vscause = csr_read(CSR_VSCAUSE);
551 	csr->vstval = csr_read(CSR_VSTVAL);
552 	csr->hvip = csr_read(CSR_HVIP);
553 	csr->vsatp = csr_read(CSR_VSATP);
554 }
555 
556 static void kvm_riscv_check_vcpu_requests(struct kvm_vcpu *vcpu)
557 {
558 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
559 
560 	if (kvm_request_pending(vcpu)) {
561 		if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) {
562 			kvm_vcpu_srcu_read_unlock(vcpu);
563 			rcuwait_wait_event(wait,
564 				(!vcpu->arch.power_off) && (!vcpu->arch.pause),
565 				TASK_INTERRUPTIBLE);
566 			kvm_vcpu_srcu_read_lock(vcpu);
567 
568 			if (vcpu->arch.power_off || vcpu->arch.pause) {
569 				/*
570 				 * Awaken to handle a signal, request to
571 				 * sleep again later.
572 				 */
573 				kvm_make_request(KVM_REQ_SLEEP, vcpu);
574 			}
575 		}
576 
577 		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
578 			kvm_riscv_reset_vcpu(vcpu);
579 
580 		if (kvm_check_request(KVM_REQ_UPDATE_HGATP, vcpu))
581 			kvm_riscv_gstage_update_hgatp(vcpu);
582 
583 		if (kvm_check_request(KVM_REQ_FENCE_I, vcpu))
584 			kvm_riscv_fence_i_process(vcpu);
585 
586 		/*
587 		 * The generic KVM_REQ_TLB_FLUSH is same as
588 		 * KVM_REQ_HFENCE_GVMA_VMID_ALL
589 		 */
590 		if (kvm_check_request(KVM_REQ_HFENCE_GVMA_VMID_ALL, vcpu))
591 			kvm_riscv_hfence_gvma_vmid_all_process(vcpu);
592 
593 		if (kvm_check_request(KVM_REQ_HFENCE_VVMA_ALL, vcpu))
594 			kvm_riscv_hfence_vvma_all_process(vcpu);
595 
596 		if (kvm_check_request(KVM_REQ_HFENCE, vcpu))
597 			kvm_riscv_hfence_process(vcpu);
598 	}
599 }
600 
601 static void kvm_riscv_update_hvip(struct kvm_vcpu *vcpu)
602 {
603 	struct kvm_vcpu_csr *csr = &vcpu->arch.guest_csr;
604 
605 	csr_write(CSR_HVIP, csr->hvip);
606 	kvm_riscv_vcpu_aia_update_hvip(vcpu);
607 }
608 
609 /*
610  * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
611  * the vCPU is running.
612  *
613  * This must be noinstr as instrumentation may make use of RCU, and this is not
614  * safe during the EQS.
615  */
616 static void noinstr kvm_riscv_vcpu_enter_exit(struct kvm_vcpu *vcpu)
617 {
618 	guest_state_enter_irqoff();
619 	__kvm_riscv_switch_to(&vcpu->arch);
620 	vcpu->arch.last_exit_cpu = vcpu->cpu;
621 	guest_state_exit_irqoff();
622 }
623 
624 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
625 {
626 	int ret;
627 	struct kvm_cpu_trap trap;
628 	struct kvm_run *run = vcpu->run;
629 
630 	/* Mark this VCPU ran at least once */
631 	vcpu->arch.ran_atleast_once = true;
632 
633 	kvm_vcpu_srcu_read_lock(vcpu);
634 
635 	switch (run->exit_reason) {
636 	case KVM_EXIT_MMIO:
637 		/* Process MMIO value returned from user-space */
638 		ret = kvm_riscv_vcpu_mmio_return(vcpu, vcpu->run);
639 		break;
640 	case KVM_EXIT_RISCV_SBI:
641 		/* Process SBI value returned from user-space */
642 		ret = kvm_riscv_vcpu_sbi_return(vcpu, vcpu->run);
643 		break;
644 	case KVM_EXIT_RISCV_CSR:
645 		/* Process CSR value returned from user-space */
646 		ret = kvm_riscv_vcpu_csr_return(vcpu, vcpu->run);
647 		break;
648 	default:
649 		ret = 0;
650 		break;
651 	}
652 	if (ret) {
653 		kvm_vcpu_srcu_read_unlock(vcpu);
654 		return ret;
655 	}
656 
657 	if (run->immediate_exit) {
658 		kvm_vcpu_srcu_read_unlock(vcpu);
659 		return -EINTR;
660 	}
661 
662 	vcpu_load(vcpu);
663 
664 	kvm_sigset_activate(vcpu);
665 
666 	ret = 1;
667 	run->exit_reason = KVM_EXIT_UNKNOWN;
668 	while (ret > 0) {
669 		/* Check conditions before entering the guest */
670 		ret = xfer_to_guest_mode_handle_work(vcpu);
671 		if (ret)
672 			continue;
673 		ret = 1;
674 
675 		kvm_riscv_gstage_vmid_update(vcpu);
676 
677 		kvm_riscv_check_vcpu_requests(vcpu);
678 
679 		preempt_disable();
680 
681 		/* Update AIA HW state before entering guest */
682 		ret = kvm_riscv_vcpu_aia_update(vcpu);
683 		if (ret <= 0) {
684 			preempt_enable();
685 			continue;
686 		}
687 
688 		local_irq_disable();
689 
690 		/*
691 		 * Ensure we set mode to IN_GUEST_MODE after we disable
692 		 * interrupts and before the final VCPU requests check.
693 		 * See the comment in kvm_vcpu_exiting_guest_mode() and
694 		 * Documentation/virt/kvm/vcpu-requests.rst
695 		 */
696 		vcpu->mode = IN_GUEST_MODE;
697 
698 		kvm_vcpu_srcu_read_unlock(vcpu);
699 		smp_mb__after_srcu_read_unlock();
700 
701 		/*
702 		 * We might have got VCPU interrupts updated asynchronously
703 		 * so update it in HW.
704 		 */
705 		kvm_riscv_vcpu_flush_interrupts(vcpu);
706 
707 		/* Update HVIP CSR for current CPU */
708 		kvm_riscv_update_hvip(vcpu);
709 
710 		if (ret <= 0 ||
711 		    kvm_riscv_gstage_vmid_ver_changed(&vcpu->kvm->arch.vmid) ||
712 		    kvm_request_pending(vcpu) ||
713 		    xfer_to_guest_mode_work_pending()) {
714 			vcpu->mode = OUTSIDE_GUEST_MODE;
715 			local_irq_enable();
716 			preempt_enable();
717 			kvm_vcpu_srcu_read_lock(vcpu);
718 			continue;
719 		}
720 
721 		/*
722 		 * Cleanup stale TLB enteries
723 		 *
724 		 * Note: This should be done after G-stage VMID has been
725 		 * updated using kvm_riscv_gstage_vmid_ver_changed()
726 		 */
727 		kvm_riscv_local_tlb_sanitize(vcpu);
728 
729 		guest_timing_enter_irqoff();
730 
731 		kvm_riscv_vcpu_enter_exit(vcpu);
732 
733 		vcpu->mode = OUTSIDE_GUEST_MODE;
734 		vcpu->stat.exits++;
735 
736 		/*
737 		 * Save SCAUSE, STVAL, HTVAL, and HTINST because we might
738 		 * get an interrupt between __kvm_riscv_switch_to() and
739 		 * local_irq_enable() which can potentially change CSRs.
740 		 */
741 		trap.sepc = vcpu->arch.guest_context.sepc;
742 		trap.scause = csr_read(CSR_SCAUSE);
743 		trap.stval = csr_read(CSR_STVAL);
744 		trap.htval = csr_read(CSR_HTVAL);
745 		trap.htinst = csr_read(CSR_HTINST);
746 
747 		/* Syncup interrupts state with HW */
748 		kvm_riscv_vcpu_sync_interrupts(vcpu);
749 
750 		/*
751 		 * We must ensure that any pending interrupts are taken before
752 		 * we exit guest timing so that timer ticks are accounted as
753 		 * guest time. Transiently unmask interrupts so that any
754 		 * pending interrupts are taken.
755 		 *
756 		 * There's no barrier which ensures that pending interrupts are
757 		 * recognised, so we just hope that the CPU takes any pending
758 		 * interrupts between the enable and disable.
759 		 */
760 		local_irq_enable();
761 		local_irq_disable();
762 
763 		guest_timing_exit_irqoff();
764 
765 		local_irq_enable();
766 
767 		preempt_enable();
768 
769 		kvm_vcpu_srcu_read_lock(vcpu);
770 
771 		ret = kvm_riscv_vcpu_exit(vcpu, run, &trap);
772 	}
773 
774 	kvm_sigset_deactivate(vcpu);
775 
776 	vcpu_put(vcpu);
777 
778 	kvm_vcpu_srcu_read_unlock(vcpu);
779 
780 	return ret;
781 }
782