xref: /linux/arch/loongarch/kvm/vcpu.c (revision 9b486cdd032a90032c6b567ea723595205ca2626)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2020-2023 Loongson Technology Corporation Limited
4  */
5 
6 #include <linux/kvm_host.h>
7 #include <asm/fpu.h>
8 #include <asm/lbt.h>
9 #include <asm/loongarch.h>
10 #include <asm/setup.h>
11 #include <asm/time.h>
12 #include <asm/timex.h>
13 
14 #define CREATE_TRACE_POINTS
15 #include "trace.h"
16 
17 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
18 	KVM_GENERIC_VCPU_STATS(),
19 	STATS_DESC_COUNTER(VCPU, int_exits),
20 	STATS_DESC_COUNTER(VCPU, idle_exits),
21 	STATS_DESC_COUNTER(VCPU, cpucfg_exits),
22 	STATS_DESC_COUNTER(VCPU, signal_exits),
23 	STATS_DESC_COUNTER(VCPU, hypercall_exits),
24 	STATS_DESC_COUNTER(VCPU, ipi_read_exits),
25 	STATS_DESC_COUNTER(VCPU, ipi_write_exits),
26 	STATS_DESC_COUNTER(VCPU, eiointc_read_exits),
27 	STATS_DESC_COUNTER(VCPU, eiointc_write_exits),
28 	STATS_DESC_COUNTER(VCPU, pch_pic_read_exits),
29 	STATS_DESC_COUNTER(VCPU, pch_pic_write_exits)
30 };
31 
32 const struct kvm_stats_header kvm_vcpu_stats_header = {
33 	.name_size = KVM_STATS_NAME_SIZE,
34 	.num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
35 	.id_offset = sizeof(struct kvm_stats_header),
36 	.desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
37 	.data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
38 		       sizeof(kvm_vcpu_stats_desc),
39 };
40 
41 static inline void kvm_save_host_pmu(struct kvm_vcpu *vcpu)
42 {
43 	struct kvm_context *context;
44 
45 	context = this_cpu_ptr(vcpu->kvm->arch.vmcs);
46 	context->perf_cntr[0] = read_csr_perfcntr0();
47 	context->perf_cntr[1] = read_csr_perfcntr1();
48 	context->perf_cntr[2] = read_csr_perfcntr2();
49 	context->perf_cntr[3] = read_csr_perfcntr3();
50 	context->perf_ctrl[0] = write_csr_perfctrl0(0);
51 	context->perf_ctrl[1] = write_csr_perfctrl1(0);
52 	context->perf_ctrl[2] = write_csr_perfctrl2(0);
53 	context->perf_ctrl[3] = write_csr_perfctrl3(0);
54 }
55 
56 static inline void kvm_restore_host_pmu(struct kvm_vcpu *vcpu)
57 {
58 	struct kvm_context *context;
59 
60 	context = this_cpu_ptr(vcpu->kvm->arch.vmcs);
61 	write_csr_perfcntr0(context->perf_cntr[0]);
62 	write_csr_perfcntr1(context->perf_cntr[1]);
63 	write_csr_perfcntr2(context->perf_cntr[2]);
64 	write_csr_perfcntr3(context->perf_cntr[3]);
65 	write_csr_perfctrl0(context->perf_ctrl[0]);
66 	write_csr_perfctrl1(context->perf_ctrl[1]);
67 	write_csr_perfctrl2(context->perf_ctrl[2]);
68 	write_csr_perfctrl3(context->perf_ctrl[3]);
69 }
70 
71 
72 static inline void kvm_save_guest_pmu(struct kvm_vcpu *vcpu)
73 {
74 	struct loongarch_csrs *csr = vcpu->arch.csr;
75 
76 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR0);
77 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR1);
78 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR2);
79 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR3);
80 	kvm_read_clear_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL0);
81 	kvm_read_clear_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL1);
82 	kvm_read_clear_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL2);
83 	kvm_read_clear_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL3);
84 }
85 
86 static inline void kvm_restore_guest_pmu(struct kvm_vcpu *vcpu)
87 {
88 	struct loongarch_csrs *csr = vcpu->arch.csr;
89 
90 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR0);
91 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR1);
92 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR2);
93 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCNTR3);
94 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL0);
95 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL1);
96 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL2);
97 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PERFCTRL3);
98 }
99 
100 static int kvm_own_pmu(struct kvm_vcpu *vcpu)
101 {
102 	unsigned long val;
103 
104 	if (!kvm_guest_has_pmu(&vcpu->arch))
105 		return -EINVAL;
106 
107 	kvm_save_host_pmu(vcpu);
108 
109 	/* Set PM0-PM(num) to guest */
110 	val = read_csr_gcfg() & ~CSR_GCFG_GPERF;
111 	val |= (kvm_get_pmu_num(&vcpu->arch) + 1) << CSR_GCFG_GPERF_SHIFT;
112 	write_csr_gcfg(val);
113 
114 	kvm_restore_guest_pmu(vcpu);
115 
116 	return 0;
117 }
118 
119 static void kvm_lose_pmu(struct kvm_vcpu *vcpu)
120 {
121 	unsigned long val;
122 	struct loongarch_csrs *csr = vcpu->arch.csr;
123 
124 	if (!(vcpu->arch.aux_inuse & KVM_LARCH_PMU))
125 		return;
126 
127 	kvm_save_guest_pmu(vcpu);
128 
129 	/* Disable pmu access from guest */
130 	write_csr_gcfg(read_csr_gcfg() & ~CSR_GCFG_GPERF);
131 
132 	/*
133 	 * Clear KVM_LARCH_PMU if the guest is not using PMU CSRs when
134 	 * exiting the guest, so that the next time trap into the guest.
135 	 * We don't need to deal with PMU CSRs contexts.
136 	 *
137 	 * Otherwise set the request bit KVM_REQ_PMU to restore guest PMU
138 	 * before entering guest VM
139 	 */
140 	val = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL0);
141 	val |= kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL1);
142 	val |= kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL2);
143 	val |= kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL3);
144 	if (!(val & KVM_PMU_EVENT_ENABLED))
145 		vcpu->arch.aux_inuse &= ~KVM_LARCH_PMU;
146 	else
147 		kvm_make_request(KVM_REQ_PMU, vcpu);
148 
149 	kvm_restore_host_pmu(vcpu);
150 }
151 
152 static void kvm_check_pmu(struct kvm_vcpu *vcpu)
153 {
154 	if (kvm_check_request(KVM_REQ_PMU, vcpu)) {
155 		kvm_own_pmu(vcpu);
156 		vcpu->arch.aux_inuse |= KVM_LARCH_PMU;
157 	}
158 }
159 
160 static void kvm_update_stolen_time(struct kvm_vcpu *vcpu)
161 {
162 	u32 version;
163 	u64 steal;
164 	gpa_t gpa;
165 	struct kvm_memslots *slots;
166 	struct kvm_steal_time __user *st;
167 	struct gfn_to_hva_cache *ghc;
168 
169 	ghc = &vcpu->arch.st.cache;
170 	gpa = vcpu->arch.st.guest_addr;
171 	if (!(gpa & KVM_STEAL_PHYS_VALID))
172 		return;
173 
174 	gpa &= KVM_STEAL_PHYS_MASK;
175 	slots = kvm_memslots(vcpu->kvm);
176 	if (slots->generation != ghc->generation || gpa != ghc->gpa) {
177 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st))) {
178 			ghc->gpa = INVALID_GPA;
179 			return;
180 		}
181 	}
182 
183 	st = (struct kvm_steal_time __user *)ghc->hva;
184 	if (kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_PREEMPT)) {
185 		unsafe_put_user(0, &st->preempted, out);
186 		vcpu->arch.st.preempted = 0;
187 	}
188 
189 	unsafe_get_user(version, &st->version, out);
190 	if (version & 1)
191 		version += 1; /* first time write, random junk */
192 
193 	version += 1;
194 	unsafe_put_user(version, &st->version, out);
195 	smp_wmb();
196 
197 	unsafe_get_user(steal, &st->steal, out);
198 	steal += current->sched_info.run_delay - vcpu->arch.st.last_steal;
199 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
200 	unsafe_put_user(steal, &st->steal, out);
201 
202 	smp_wmb();
203 	version += 1;
204 	unsafe_put_user(version, &st->version, out);
205 out:
206 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
207 }
208 
209 /*
210  * kvm_check_requests - check and handle pending vCPU requests
211  *
212  * Return: RESUME_GUEST if we should enter the guest
213  *         RESUME_HOST  if we should exit to userspace
214  */
215 static int kvm_check_requests(struct kvm_vcpu *vcpu)
216 {
217 	if (!kvm_request_pending(vcpu))
218 		return RESUME_GUEST;
219 
220 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
221 		vcpu->arch.vpid = 0;  /* Drop vpid for this vCPU */
222 
223 	if (kvm_dirty_ring_check_request(vcpu))
224 		return RESUME_HOST;
225 
226 	if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
227 		kvm_update_stolen_time(vcpu);
228 
229 	return RESUME_GUEST;
230 }
231 
232 static void kvm_late_check_requests(struct kvm_vcpu *vcpu)
233 {
234 	lockdep_assert_irqs_disabled();
235 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_GPA, vcpu))
236 		if (vcpu->arch.flush_gpa != INVALID_GPA) {
237 			kvm_flush_tlb_gpa(vcpu, vcpu->arch.flush_gpa);
238 			vcpu->arch.flush_gpa = INVALID_GPA;
239 		}
240 
241 	if (kvm_check_request(KVM_REQ_AUX_LOAD, vcpu)) {
242 		switch (vcpu->arch.aux_ldtype) {
243 		case KVM_LARCH_FPU:
244 			kvm_own_fpu(vcpu);
245 			break;
246 		case KVM_LARCH_LSX:
247 			kvm_own_lsx(vcpu);
248 			break;
249 		case KVM_LARCH_LASX:
250 			kvm_own_lasx(vcpu);
251 			break;
252 		case KVM_LARCH_LBT:
253 			kvm_own_lbt(vcpu);
254 			break;
255 		default:
256 			break;
257 		}
258 
259 		vcpu->arch.aux_ldtype = 0;
260 	}
261 }
262 
263 /*
264  * Check and handle pending signal and vCPU requests etc
265  * Run with irq enabled and preempt enabled
266  *
267  * Return: RESUME_GUEST if we should enter the guest
268  *         RESUME_HOST  if we should exit to userspace
269  *         < 0 if we should exit to userspace, where the return value
270  *         indicates an error
271  */
272 static int kvm_enter_guest_check(struct kvm_vcpu *vcpu)
273 {
274 	int idx, ret;
275 
276 	/*
277 	 * Check conditions before entering the guest
278 	 */
279 	ret = kvm_xfer_to_guest_mode_handle_work(vcpu);
280 	if (ret < 0)
281 		return ret;
282 
283 	idx = srcu_read_lock(&vcpu->kvm->srcu);
284 	ret = kvm_check_requests(vcpu);
285 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
286 
287 	return ret;
288 }
289 
290 /*
291  * Called with irq enabled
292  *
293  * Return: RESUME_GUEST if we should enter the guest, and irq disabled
294  *         Others if we should exit to userspace
295  */
296 static int kvm_pre_enter_guest(struct kvm_vcpu *vcpu)
297 {
298 	int ret;
299 
300 	do {
301 		ret = kvm_enter_guest_check(vcpu);
302 		if (ret != RESUME_GUEST)
303 			break;
304 
305 		/*
306 		 * Handle vcpu timer, interrupts, check requests and
307 		 * check vmid before vcpu enter guest
308 		 */
309 		local_irq_disable();
310 		kvm_deliver_intr(vcpu);
311 		kvm_deliver_exception(vcpu);
312 		/* Make sure the vcpu mode has been written */
313 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
314 		kvm_check_vpid(vcpu);
315 		kvm_check_pmu(vcpu);
316 
317 		/*
318 		 * Called after function kvm_check_vpid()
319 		 * Since it updates CSR.GSTAT used by kvm_flush_tlb_gpa(),
320 		 * and it may also clear KVM_REQ_TLB_FLUSH_GPA pending bit
321 		 */
322 		kvm_late_check_requests(vcpu);
323 		vcpu->arch.host_eentry = csr_read64(LOONGARCH_CSR_EENTRY);
324 		/* Clear KVM_LARCH_SWCSR_LATEST as CSR will change when enter guest */
325 		vcpu->arch.aux_inuse &= ~KVM_LARCH_SWCSR_LATEST;
326 
327 		if (kvm_request_pending(vcpu) || xfer_to_guest_mode_work_pending()) {
328 			if (vcpu->arch.aux_inuse & KVM_LARCH_PMU) {
329 				kvm_lose_pmu(vcpu);
330 				kvm_make_request(KVM_REQ_PMU, vcpu);
331 			}
332 			/* make sure the vcpu mode has been written */
333 			smp_store_mb(vcpu->mode, OUTSIDE_GUEST_MODE);
334 			local_irq_enable();
335 			ret = -EAGAIN;
336 		}
337 	} while (ret != RESUME_GUEST);
338 
339 	return ret;
340 }
341 
342 /*
343  * Return 1 for resume guest and "<= 0" for resume host.
344  */
345 static int kvm_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
346 {
347 	int ret = RESUME_GUEST;
348 	unsigned long estat = vcpu->arch.host_estat;
349 	u32 intr = estat & CSR_ESTAT_IS;
350 	u32 ecode = (estat & CSR_ESTAT_EXC) >> CSR_ESTAT_EXC_SHIFT;
351 
352 	vcpu->mode = OUTSIDE_GUEST_MODE;
353 
354 	/* Set a default exit reason */
355 	run->exit_reason = KVM_EXIT_UNKNOWN;
356 
357 	kvm_lose_pmu(vcpu);
358 
359 	guest_timing_exit_irqoff();
360 	guest_state_exit_irqoff();
361 	local_irq_enable();
362 
363 	trace_kvm_exit(vcpu, ecode);
364 	if (ecode) {
365 		ret = kvm_handle_fault(vcpu, ecode);
366 	} else {
367 		WARN(!intr, "vm exiting with suspicious irq\n");
368 		++vcpu->stat.int_exits;
369 	}
370 
371 	if (ret == RESUME_GUEST)
372 		ret = kvm_pre_enter_guest(vcpu);
373 
374 	if (ret != RESUME_GUEST) {
375 		local_irq_disable();
376 		return ret;
377 	}
378 
379 	guest_timing_enter_irqoff();
380 	guest_state_enter_irqoff();
381 	trace_kvm_reenter(vcpu);
382 
383 	return RESUME_GUEST;
384 }
385 
386 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
387 {
388 	return !!(vcpu->arch.irq_pending) &&
389 		vcpu->arch.mp_state.mp_state == KVM_MP_STATE_RUNNABLE;
390 }
391 
392 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
393 {
394 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
395 }
396 
397 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
398 {
399 	unsigned long val;
400 
401 	preempt_disable();
402 	val = gcsr_read(LOONGARCH_CSR_CRMD);
403 	preempt_enable();
404 
405 	return (val & CSR_PRMD_PPLV) == PLV_KERN;
406 }
407 
408 #ifdef CONFIG_GUEST_PERF_EVENTS
409 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
410 {
411 	return vcpu->arch.pc;
412 }
413 
414 /*
415  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
416  * arrived in guest context.  For LoongArch64, if PMU is not passthrough to VM,
417  * any event that arrives while a vCPU is loaded is considered to be "in guest".
418  */
419 bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
420 {
421 	return (vcpu && !(vcpu->arch.aux_inuse & KVM_LARCH_PMU));
422 }
423 #endif
424 
425 bool kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu)
426 {
427 	return false;
428 }
429 
430 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
431 {
432 	return VM_FAULT_SIGBUS;
433 }
434 
435 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
436 				  struct kvm_translation *tr)
437 {
438 	return -EINVAL;
439 }
440 
441 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
442 {
443 	int ret;
444 
445 	/* Protect from TOD sync and vcpu_load/put() */
446 	preempt_disable();
447 	ret = kvm_pending_timer(vcpu) ||
448 		kvm_read_hw_gcsr(LOONGARCH_CSR_ESTAT) & (1 << INT_TI);
449 	preempt_enable();
450 
451 	return ret;
452 }
453 
454 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
455 {
456 	int i;
457 
458 	kvm_debug("vCPU Register Dump:\n");
459 	kvm_debug("\tPC = 0x%08lx\n", vcpu->arch.pc);
460 	kvm_debug("\tExceptions: %08lx\n", vcpu->arch.irq_pending);
461 
462 	for (i = 0; i < 32; i += 4) {
463 		kvm_debug("\tGPR%02d: %08lx %08lx %08lx %08lx\n", i,
464 		       vcpu->arch.gprs[i], vcpu->arch.gprs[i + 1],
465 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
466 	}
467 
468 	kvm_debug("\tCRMD: 0x%08lx, ESTAT: 0x%08lx\n",
469 		  kvm_read_hw_gcsr(LOONGARCH_CSR_CRMD),
470 		  kvm_read_hw_gcsr(LOONGARCH_CSR_ESTAT));
471 
472 	kvm_debug("\tERA: 0x%08lx\n", kvm_read_hw_gcsr(LOONGARCH_CSR_ERA));
473 
474 	return 0;
475 }
476 
477 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
478 				struct kvm_mp_state *mp_state)
479 {
480 	*mp_state = vcpu->arch.mp_state;
481 
482 	return 0;
483 }
484 
485 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
486 				struct kvm_mp_state *mp_state)
487 {
488 	int ret = 0;
489 
490 	switch (mp_state->mp_state) {
491 	case KVM_MP_STATE_RUNNABLE:
492 		vcpu->arch.mp_state = *mp_state;
493 		break;
494 	default:
495 		ret = -EINVAL;
496 	}
497 
498 	return ret;
499 }
500 
501 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
502 					struct kvm_guest_debug *dbg)
503 {
504 	if (dbg->control & ~KVM_GUESTDBG_VALID_MASK)
505 		return -EINVAL;
506 
507 	if (dbg->control & KVM_GUESTDBG_ENABLE)
508 		vcpu->guest_debug = dbg->control;
509 	else
510 		vcpu->guest_debug = 0;
511 
512 	return 0;
513 }
514 
515 static inline int kvm_set_cpuid(struct kvm_vcpu *vcpu, u64 val)
516 {
517 	int cpuid;
518 	struct kvm_phyid_map *map;
519 	struct loongarch_csrs *csr = vcpu->arch.csr;
520 
521 	if (val >= KVM_MAX_PHYID)
522 		return -EINVAL;
523 
524 	map = vcpu->kvm->arch.phyid_map;
525 	cpuid = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_CPUID);
526 
527 	spin_lock(&vcpu->kvm->arch.phyid_map_lock);
528 	if ((cpuid < KVM_MAX_PHYID) && map->phys_map[cpuid].enabled) {
529 		/* Discard duplicated CPUID set operation */
530 		if (cpuid == val) {
531 			spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
532 			return 0;
533 		}
534 
535 		/*
536 		 * CPUID is already set before
537 		 * Forbid changing to a different CPUID at runtime
538 		 */
539 		spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
540 		return -EINVAL;
541 	}
542 
543 	if (map->phys_map[val].enabled) {
544 		/* Discard duplicated CPUID set operation */
545 		if (vcpu == map->phys_map[val].vcpu) {
546 			spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
547 			return 0;
548 		}
549 
550 		/*
551 		 * New CPUID is already set with other vcpu
552 		 * Forbid sharing the same CPUID between different vcpus
553 		 */
554 		spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
555 		return -EINVAL;
556 	}
557 
558 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CPUID, val);
559 	map->phys_map[val].enabled	= true;
560 	map->phys_map[val].vcpu		= vcpu;
561 	spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
562 
563 	return 0;
564 }
565 
566 static inline void kvm_drop_cpuid(struct kvm_vcpu *vcpu)
567 {
568 	int cpuid;
569 	struct kvm_phyid_map *map;
570 	struct loongarch_csrs *csr = vcpu->arch.csr;
571 
572 	map = vcpu->kvm->arch.phyid_map;
573 	cpuid = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_CPUID);
574 
575 	if (cpuid >= KVM_MAX_PHYID)
576 		return;
577 
578 	spin_lock(&vcpu->kvm->arch.phyid_map_lock);
579 	if (map->phys_map[cpuid].enabled) {
580 		map->phys_map[cpuid].vcpu = NULL;
581 		map->phys_map[cpuid].enabled = false;
582 		kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CPUID, KVM_MAX_PHYID);
583 	}
584 	spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
585 }
586 
587 struct kvm_vcpu *kvm_get_vcpu_by_cpuid(struct kvm *kvm, int cpuid)
588 {
589 	struct kvm_phyid_map *map;
590 
591 	if (cpuid >= KVM_MAX_PHYID)
592 		return NULL;
593 
594 	map = kvm->arch.phyid_map;
595 	if (!map->phys_map[cpuid].enabled)
596 		return NULL;
597 
598 	return map->phys_map[cpuid].vcpu;
599 }
600 
601 static int _kvm_getcsr(struct kvm_vcpu *vcpu, unsigned int id, u64 *val)
602 {
603 	unsigned long gintc;
604 	struct loongarch_csrs *csr = vcpu->arch.csr;
605 
606 	if (get_gcsr_flag(id) & INVALID_GCSR)
607 		return -EINVAL;
608 
609 	if (id == LOONGARCH_CSR_ESTAT) {
610 		preempt_disable();
611 		vcpu_load(vcpu);
612 		/*
613 		 * Sync pending interrupts into ESTAT so that interrupt
614 		 * remains during VM migration stage
615 		 */
616 		kvm_deliver_intr(vcpu);
617 		vcpu->arch.aux_inuse &= ~KVM_LARCH_SWCSR_LATEST;
618 		vcpu_put(vcpu);
619 		preempt_enable();
620 
621 		/* ESTAT IP0~IP7 get from GINTC */
622 		gintc = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_GINTC) & 0xff;
623 		*val = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_ESTAT) | (gintc << 2);
624 		return 0;
625 	}
626 
627 	/*
628 	 * Get software CSR state since software state is consistent
629 	 * with hardware for synchronous ioctl
630 	 */
631 	*val = kvm_read_sw_gcsr(csr, id);
632 
633 	return 0;
634 }
635 
636 static int _kvm_setcsr(struct kvm_vcpu *vcpu, unsigned int id, u64 val)
637 {
638 	int ret = 0, gintc;
639 	struct loongarch_csrs *csr = vcpu->arch.csr;
640 
641 	if (get_gcsr_flag(id) & INVALID_GCSR)
642 		return -EINVAL;
643 
644 	if (id == LOONGARCH_CSR_CPUID)
645 		return kvm_set_cpuid(vcpu, val);
646 
647 	if (id == LOONGARCH_CSR_ESTAT) {
648 		/* ESTAT IP0~IP7 inject through GINTC */
649 		gintc = (val >> 2) & 0xff;
650 		kvm_set_sw_gcsr(csr, LOONGARCH_CSR_GINTC, gintc);
651 
652 		gintc = val & ~(0xffUL << 2);
653 		kvm_set_sw_gcsr(csr, LOONGARCH_CSR_ESTAT, gintc);
654 
655 		return ret;
656 	}
657 
658 	kvm_write_sw_gcsr(csr, id, val);
659 
660 	/*
661 	 * After modifying the PMU CSR register value of the vcpu.
662 	 * If the PMU CSRs are used, we need to set KVM_REQ_PMU.
663 	 */
664 	if (id >= LOONGARCH_CSR_PERFCTRL0 && id <= LOONGARCH_CSR_PERFCNTR3) {
665 		unsigned long val;
666 
667 		val = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL0) |
668 		      kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL1) |
669 		      kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL2) |
670 		      kvm_read_sw_gcsr(csr, LOONGARCH_CSR_PERFCTRL3);
671 
672 		if (val & KVM_PMU_EVENT_ENABLED)
673 			kvm_make_request(KVM_REQ_PMU, vcpu);
674 	}
675 
676 	return ret;
677 }
678 
679 static int _kvm_get_cpucfg_mask(int id, u64 *v)
680 {
681 	unsigned int config;
682 
683 	if (id < 0 || id >= KVM_MAX_CPUCFG_REGS)
684 		return -EINVAL;
685 
686 	switch (id) {
687 	case LOONGARCH_CPUCFG0:
688 		*v = GENMASK(31, 0);
689 		return 0;
690 	case LOONGARCH_CPUCFG1:
691 		*v = GENMASK(26, 0);
692 		return 0;
693 	case LOONGARCH_CPUCFG2:
694 		/* CPUCFG2 features unconditionally supported by KVM */
695 		*v = CPUCFG2_FP     | CPUCFG2_FPSP  | CPUCFG2_FPDP     |
696 		     CPUCFG2_FPVERS | CPUCFG2_LLFTP | CPUCFG2_LLFTPREV |
697 		     CPUCFG2_LSPW | CPUCFG2_LAM;
698 		/*
699 		 * For the ISA extensions listed below, if one is supported
700 		 * by the host, then it is also supported by KVM.
701 		 */
702 		if (cpu_has_lsx)
703 			*v |= CPUCFG2_LSX;
704 		if (cpu_has_lasx)
705 			*v |= CPUCFG2_LASX;
706 		if (cpu_has_lbt_x86)
707 			*v |= CPUCFG2_X86BT;
708 		if (cpu_has_lbt_arm)
709 			*v |= CPUCFG2_ARMBT;
710 		if (cpu_has_lbt_mips)
711 			*v |= CPUCFG2_MIPSBT;
712 		if (cpu_has_ptw)
713 			*v |= CPUCFG2_PTW;
714 
715 		config = read_cpucfg(LOONGARCH_CPUCFG2);
716 		*v |= config & (CPUCFG2_FRECIPE | CPUCFG2_DIV32 | CPUCFG2_LAM_BH);
717 		*v |= config & (CPUCFG2_LAMCAS | CPUCFG2_LLACQ_SCREL | CPUCFG2_SCQ);
718 		return 0;
719 	case LOONGARCH_CPUCFG3:
720 		*v = GENMASK(23, 0);
721 
722 		/* VM does not support memory order and SFB setting */
723 		config = read_cpucfg(LOONGARCH_CPUCFG3);
724 		*v &= config & ~(CPUCFG3_SFB);
725 		*v &= config & ~(CPUCFG3_ALDORDER_CAP | CPUCFG3_ASTORDER_CAP | CPUCFG3_SLDORDER_CAP);
726 		return 0;
727 	case LOONGARCH_CPUCFG4:
728 	case LOONGARCH_CPUCFG5:
729 		*v = GENMASK(31, 0);
730 		return 0;
731 	case LOONGARCH_CPUCFG6:
732 		if (cpu_has_pmp)
733 			*v = GENMASK(14, 0);
734 		else
735 			*v = 0;
736 		return 0;
737 	case LOONGARCH_CPUCFG16:
738 		*v = GENMASK(16, 0);
739 		return 0;
740 	case LOONGARCH_CPUCFG17 ... LOONGARCH_CPUCFG20:
741 		*v = GENMASK(30, 0);
742 		return 0;
743 	default:
744 		/*
745 		 * CPUCFG bits should be zero if reserved by HW or not
746 		 * supported by KVM.
747 		 */
748 		*v = 0;
749 		return 0;
750 	}
751 }
752 
753 static int kvm_check_cpucfg(int id, u64 val)
754 {
755 	int ret;
756 	u32 host;
757 	u64 mask = 0;
758 
759 	ret = _kvm_get_cpucfg_mask(id, &mask);
760 	if (ret)
761 		return ret;
762 
763 	if (val & ~mask)
764 		/* Unsupported features and/or the higher 32 bits should not be set */
765 		return -EINVAL;
766 
767 	switch (id) {
768 	case LOONGARCH_CPUCFG1:
769 		if ((val & CPUCFG1_MSGINT) && !cpu_has_msgint)
770 			return -EINVAL;
771 		return 0;
772 	case LOONGARCH_CPUCFG2:
773 		if (!(val & CPUCFG2_LLFTP))
774 			/* Guests must have a constant timer */
775 			return -EINVAL;
776 		if ((val & CPUCFG2_FP) && (!(val & CPUCFG2_FPSP) || !(val & CPUCFG2_FPDP)))
777 			/* Single and double float point must both be set when FP is enabled */
778 			return -EINVAL;
779 		if ((val & CPUCFG2_LSX) && !(val & CPUCFG2_FP))
780 			/* LSX architecturally implies FP but val does not satisfy that */
781 			return -EINVAL;
782 		if ((val & CPUCFG2_LASX) && !(val & CPUCFG2_LSX))
783 			/* LASX architecturally implies LSX and FP but val does not satisfy that */
784 			return -EINVAL;
785 		return 0;
786 	case LOONGARCH_CPUCFG3:
787 		host = read_cpucfg(LOONGARCH_CPUCFG3);
788 		if ((val & CPUCFG3_RVAMAX) > (host & CPUCFG3_RVAMAX))
789 			return -EINVAL;
790 		if ((val & CPUCFG3_SPW_LVL) > (host & CPUCFG3_SPW_LVL))
791 			return -EINVAL;
792 		return 0;
793 	case LOONGARCH_CPUCFG6:
794 		if (val & CPUCFG6_PMP) {
795 			host = read_cpucfg(LOONGARCH_CPUCFG6);
796 			if ((val & CPUCFG6_PMBITS) != (host & CPUCFG6_PMBITS))
797 				return -EINVAL;
798 			if ((val & CPUCFG6_PMNUM) > (host & CPUCFG6_PMNUM))
799 				return -EINVAL;
800 			if ((val & CPUCFG6_UPM) && !(host & CPUCFG6_UPM))
801 				return -EINVAL;
802 		}
803 		return 0;
804 	default:
805 		/*
806 		 * Values for the other CPUCFG IDs are not being further validated
807 		 * besides the mask check above.
808 		 */
809 		return 0;
810 	}
811 }
812 
813 static int kvm_get_one_reg(struct kvm_vcpu *vcpu,
814 		const struct kvm_one_reg *reg, u64 *v)
815 {
816 	int id, ret = 0;
817 	u64 type = reg->id & KVM_REG_LOONGARCH_MASK;
818 
819 	switch (type) {
820 	case KVM_REG_LOONGARCH_CSR:
821 		id = KVM_GET_IOC_CSR_IDX(reg->id);
822 		ret = _kvm_getcsr(vcpu, id, v);
823 		break;
824 	case KVM_REG_LOONGARCH_CPUCFG:
825 		id = KVM_GET_IOC_CPUCFG_IDX(reg->id);
826 		if (id >= 0 && id < KVM_MAX_CPUCFG_REGS)
827 			*v = vcpu->arch.cpucfg[id];
828 		else
829 			ret = -EINVAL;
830 		break;
831 	case KVM_REG_LOONGARCH_LBT:
832 		if (!kvm_guest_has_lbt(&vcpu->arch))
833 			return -ENXIO;
834 
835 		switch (reg->id) {
836 		case KVM_REG_LOONGARCH_LBT_SCR0:
837 			*v = vcpu->arch.lbt.scr0;
838 			break;
839 		case KVM_REG_LOONGARCH_LBT_SCR1:
840 			*v = vcpu->arch.lbt.scr1;
841 			break;
842 		case KVM_REG_LOONGARCH_LBT_SCR2:
843 			*v = vcpu->arch.lbt.scr2;
844 			break;
845 		case KVM_REG_LOONGARCH_LBT_SCR3:
846 			*v = vcpu->arch.lbt.scr3;
847 			break;
848 		case KVM_REG_LOONGARCH_LBT_EFLAGS:
849 			*v = vcpu->arch.lbt.eflags;
850 			break;
851 		case KVM_REG_LOONGARCH_LBT_FTOP:
852 			*v = vcpu->arch.fpu.ftop;
853 			break;
854 		default:
855 			ret = -EINVAL;
856 			break;
857 		}
858 		break;
859 	case KVM_REG_LOONGARCH_KVM:
860 		switch (reg->id) {
861 		case KVM_REG_LOONGARCH_COUNTER:
862 			*v = get_cycles() + vcpu->kvm->arch.time_offset;
863 			break;
864 		case KVM_REG_LOONGARCH_DEBUG_INST:
865 			*v = INSN_HVCL | KVM_HCALL_SWDBG;
866 			break;
867 		default:
868 			ret = -EINVAL;
869 			break;
870 		}
871 		break;
872 	default:
873 		ret = -EINVAL;
874 		break;
875 	}
876 
877 	return ret;
878 }
879 
880 static int kvm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
881 {
882 	int ret = 0;
883 	u64 v, size = reg->id & KVM_REG_SIZE_MASK;
884 
885 	switch (size) {
886 	case KVM_REG_SIZE_U64:
887 		ret = kvm_get_one_reg(vcpu, reg, &v);
888 		if (ret)
889 			return ret;
890 		ret = put_user(v, (u64 __user *)(long)reg->addr);
891 		break;
892 	default:
893 		ret = -EINVAL;
894 		break;
895 	}
896 
897 	return ret;
898 }
899 
900 static int kvm_set_one_reg(struct kvm_vcpu *vcpu,
901 			const struct kvm_one_reg *reg, u64 v)
902 {
903 	int id, ret = 0;
904 	u64 type = reg->id & KVM_REG_LOONGARCH_MASK;
905 
906 	switch (type) {
907 	case KVM_REG_LOONGARCH_CSR:
908 		id = KVM_GET_IOC_CSR_IDX(reg->id);
909 		ret = _kvm_setcsr(vcpu, id, v);
910 		break;
911 	case KVM_REG_LOONGARCH_CPUCFG:
912 		id = KVM_GET_IOC_CPUCFG_IDX(reg->id);
913 		ret = kvm_check_cpucfg(id, v);
914 		if (ret)
915 			break;
916 		vcpu->arch.cpucfg[id] = (u32)v;
917 		if (id == LOONGARCH_CPUCFG6)
918 			vcpu->arch.max_pmu_csrid =
919 				LOONGARCH_CSR_PERFCTRL0 + 2 * kvm_get_pmu_num(&vcpu->arch) + 1;
920 		break;
921 	case KVM_REG_LOONGARCH_LBT:
922 		if (!kvm_guest_has_lbt(&vcpu->arch))
923 			return -ENXIO;
924 
925 		switch (reg->id) {
926 		case KVM_REG_LOONGARCH_LBT_SCR0:
927 			vcpu->arch.lbt.scr0 = v;
928 			break;
929 		case KVM_REG_LOONGARCH_LBT_SCR1:
930 			vcpu->arch.lbt.scr1 = v;
931 			break;
932 		case KVM_REG_LOONGARCH_LBT_SCR2:
933 			vcpu->arch.lbt.scr2 = v;
934 			break;
935 		case KVM_REG_LOONGARCH_LBT_SCR3:
936 			vcpu->arch.lbt.scr3 = v;
937 			break;
938 		case KVM_REG_LOONGARCH_LBT_EFLAGS:
939 			vcpu->arch.lbt.eflags = v;
940 			break;
941 		case KVM_REG_LOONGARCH_LBT_FTOP:
942 			vcpu->arch.fpu.ftop = v;
943 			break;
944 		default:
945 			ret = -EINVAL;
946 			break;
947 		}
948 		break;
949 	case KVM_REG_LOONGARCH_KVM:
950 		switch (reg->id) {
951 		case KVM_REG_LOONGARCH_COUNTER:
952 			/*
953 			 * gftoffset is relative with board, not vcpu
954 			 * only set for the first time for smp system
955 			 */
956 			if (vcpu->vcpu_id == 0)
957 				vcpu->kvm->arch.time_offset = (signed long)(v - get_cycles());
958 			break;
959 		case KVM_REG_LOONGARCH_VCPU_RESET:
960 			vcpu->arch.st.guest_addr = 0;
961 			memset(&vcpu->arch.irq_pending, 0, sizeof(vcpu->arch.irq_pending));
962 			memset(&vcpu->arch.irq_clear, 0, sizeof(vcpu->arch.irq_clear));
963 
964 			/*
965 			 * When vCPU reset, clear the ESTAT and GINTC registers
966 			 * Other CSR registers are cleared with function _kvm_setcsr().
967 			 */
968 			kvm_write_sw_gcsr(vcpu->arch.csr, LOONGARCH_CSR_GINTC, 0);
969 			kvm_write_sw_gcsr(vcpu->arch.csr, LOONGARCH_CSR_ESTAT, 0);
970 			break;
971 		default:
972 			ret = -EINVAL;
973 			break;
974 		}
975 		break;
976 	default:
977 		ret = -EINVAL;
978 		break;
979 	}
980 
981 	return ret;
982 }
983 
984 static int kvm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
985 {
986 	int ret = 0;
987 	u64 v, size = reg->id & KVM_REG_SIZE_MASK;
988 
989 	switch (size) {
990 	case KVM_REG_SIZE_U64:
991 		ret = get_user(v, (u64 __user *)(long)reg->addr);
992 		if (ret)
993 			return ret;
994 		break;
995 	default:
996 		return -EINVAL;
997 	}
998 
999 	return kvm_set_one_reg(vcpu, reg, v);
1000 }
1001 
1002 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1003 {
1004 	return -ENOIOCTLCMD;
1005 }
1006 
1007 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
1008 {
1009 	return -ENOIOCTLCMD;
1010 }
1011 
1012 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1013 {
1014 	int i;
1015 
1016 	for (i = 0; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1017 		regs->gpr[i] = vcpu->arch.gprs[i];
1018 
1019 	regs->pc = vcpu->arch.pc;
1020 
1021 	return 0;
1022 }
1023 
1024 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1025 {
1026 	int i;
1027 
1028 	for (i = 1; i < ARRAY_SIZE(vcpu->arch.gprs); i++)
1029 		vcpu->arch.gprs[i] = regs->gpr[i];
1030 
1031 	vcpu->arch.gprs[0] = 0; /* zero is special, and cannot be set. */
1032 	vcpu->arch.pc = regs->pc;
1033 
1034 	return 0;
1035 }
1036 
1037 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1038 				     struct kvm_enable_cap *cap)
1039 {
1040 	/* FPU is enabled by default, will support LSX/LASX later. */
1041 	return -EINVAL;
1042 }
1043 
1044 static int kvm_loongarch_cpucfg_has_attr(struct kvm_vcpu *vcpu,
1045 					 struct kvm_device_attr *attr)
1046 {
1047 	switch (attr->attr) {
1048 	case LOONGARCH_CPUCFG2:
1049 	case LOONGARCH_CPUCFG6:
1050 		return 0;
1051 	case CPUCFG_KVM_FEATURE:
1052 		return 0;
1053 	default:
1054 		return -ENXIO;
1055 	}
1056 
1057 	return -ENXIO;
1058 }
1059 
1060 static int kvm_loongarch_pvtime_has_attr(struct kvm_vcpu *vcpu,
1061 					 struct kvm_device_attr *attr)
1062 {
1063 	if (!kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_STEAL_TIME)
1064 			|| attr->attr != KVM_LOONGARCH_VCPU_PVTIME_GPA)
1065 		return -ENXIO;
1066 
1067 	return 0;
1068 }
1069 
1070 static int kvm_loongarch_vcpu_has_attr(struct kvm_vcpu *vcpu,
1071 				       struct kvm_device_attr *attr)
1072 {
1073 	int ret = -ENXIO;
1074 
1075 	switch (attr->group) {
1076 	case KVM_LOONGARCH_VCPU_CPUCFG:
1077 		ret = kvm_loongarch_cpucfg_has_attr(vcpu, attr);
1078 		break;
1079 	case KVM_LOONGARCH_VCPU_PVTIME_CTRL:
1080 		ret = kvm_loongarch_pvtime_has_attr(vcpu, attr);
1081 		break;
1082 	default:
1083 		break;
1084 	}
1085 
1086 	return ret;
1087 }
1088 
1089 static int kvm_loongarch_cpucfg_get_attr(struct kvm_vcpu *vcpu,
1090 					 struct kvm_device_attr *attr)
1091 {
1092 	int ret = 0;
1093 	uint64_t val;
1094 	uint64_t __user *uaddr = (uint64_t __user *)attr->addr;
1095 
1096 	switch (attr->attr) {
1097 	case 0 ... (KVM_MAX_CPUCFG_REGS - 1):
1098 		ret = _kvm_get_cpucfg_mask(attr->attr, &val);
1099 		if (ret)
1100 			return ret;
1101 		break;
1102 	case CPUCFG_KVM_FEATURE:
1103 		val = vcpu->kvm->arch.pv_features & LOONGARCH_PV_FEAT_MASK;
1104 		break;
1105 	default:
1106 		return -ENXIO;
1107 	}
1108 
1109 	put_user(val, uaddr);
1110 
1111 	return ret;
1112 }
1113 
1114 static int kvm_loongarch_pvtime_get_attr(struct kvm_vcpu *vcpu,
1115 					 struct kvm_device_attr *attr)
1116 {
1117 	u64 gpa;
1118 	u64 __user *user = (u64 __user *)attr->addr;
1119 
1120 	if (!kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_STEAL_TIME)
1121 			|| attr->attr != KVM_LOONGARCH_VCPU_PVTIME_GPA)
1122 		return -ENXIO;
1123 
1124 	gpa = vcpu->arch.st.guest_addr;
1125 	if (put_user(gpa, user))
1126 		return -EFAULT;
1127 
1128 	return 0;
1129 }
1130 
1131 static int kvm_loongarch_vcpu_get_attr(struct kvm_vcpu *vcpu,
1132 				       struct kvm_device_attr *attr)
1133 {
1134 	int ret = -ENXIO;
1135 
1136 	switch (attr->group) {
1137 	case KVM_LOONGARCH_VCPU_CPUCFG:
1138 		ret = kvm_loongarch_cpucfg_get_attr(vcpu, attr);
1139 		break;
1140 	case KVM_LOONGARCH_VCPU_PVTIME_CTRL:
1141 		ret = kvm_loongarch_pvtime_get_attr(vcpu, attr);
1142 		break;
1143 	default:
1144 		break;
1145 	}
1146 
1147 	return ret;
1148 }
1149 
1150 static int kvm_loongarch_cpucfg_set_attr(struct kvm_vcpu *vcpu,
1151 					 struct kvm_device_attr *attr)
1152 {
1153 	u64 val, valid;
1154 	u64 __user *user = (u64 __user *)attr->addr;
1155 	struct kvm *kvm = vcpu->kvm;
1156 
1157 	switch (attr->attr) {
1158 	case CPUCFG_KVM_FEATURE:
1159 		if (get_user(val, user))
1160 			return -EFAULT;
1161 
1162 		valid = LOONGARCH_PV_FEAT_MASK;
1163 		if (val & ~valid)
1164 			return -EINVAL;
1165 
1166 		/* All vCPUs need set the same PV features */
1167 		if ((kvm->arch.pv_features & LOONGARCH_PV_FEAT_UPDATED)
1168 				&& ((kvm->arch.pv_features & valid) != val))
1169 			return -EINVAL;
1170 		kvm->arch.pv_features = val | LOONGARCH_PV_FEAT_UPDATED;
1171 		return 0;
1172 	default:
1173 		return -ENXIO;
1174 	}
1175 }
1176 
1177 static int kvm_loongarch_pvtime_set_attr(struct kvm_vcpu *vcpu,
1178 					 struct kvm_device_attr *attr)
1179 {
1180 	int idx, ret = 0;
1181 	u64 gpa, __user *user = (u64 __user *)attr->addr;
1182 	struct kvm *kvm = vcpu->kvm;
1183 
1184 	if (!kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_STEAL_TIME)
1185 			|| attr->attr != KVM_LOONGARCH_VCPU_PVTIME_GPA)
1186 		return -ENXIO;
1187 
1188 	if (get_user(gpa, user))
1189 		return -EFAULT;
1190 
1191 	if (gpa & ~(KVM_STEAL_PHYS_MASK | KVM_STEAL_PHYS_VALID))
1192 		return -EINVAL;
1193 
1194 	if (!(gpa & KVM_STEAL_PHYS_VALID)) {
1195 		vcpu->arch.st.guest_addr = gpa;
1196 		return 0;
1197 	}
1198 
1199 	/* Check the address is in a valid memslot */
1200 	idx = srcu_read_lock(&kvm->srcu);
1201 	if (kvm_is_error_hva(gfn_to_hva(kvm, gpa >> PAGE_SHIFT)))
1202 		ret = -EINVAL;
1203 	srcu_read_unlock(&kvm->srcu, idx);
1204 
1205 	if (!ret) {
1206 		vcpu->arch.st.guest_addr = gpa;
1207 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
1208 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1209 	}
1210 
1211 	return ret;
1212 }
1213 
1214 static int kvm_loongarch_vcpu_set_attr(struct kvm_vcpu *vcpu,
1215 				       struct kvm_device_attr *attr)
1216 {
1217 	int ret = -ENXIO;
1218 
1219 	switch (attr->group) {
1220 	case KVM_LOONGARCH_VCPU_CPUCFG:
1221 		ret = kvm_loongarch_cpucfg_set_attr(vcpu, attr);
1222 		break;
1223 	case KVM_LOONGARCH_VCPU_PVTIME_CTRL:
1224 		ret = kvm_loongarch_pvtime_set_attr(vcpu, attr);
1225 		break;
1226 	default:
1227 		break;
1228 	}
1229 
1230 	return ret;
1231 }
1232 
1233 long kvm_arch_vcpu_ioctl(struct file *filp,
1234 			 unsigned int ioctl, unsigned long arg)
1235 {
1236 	long r;
1237 	struct kvm_device_attr attr;
1238 	void __user *argp = (void __user *)arg;
1239 	struct kvm_vcpu *vcpu = filp->private_data;
1240 
1241 	/*
1242 	 * Only software CSR should be modified
1243 	 *
1244 	 * If any hardware CSR register is modified, vcpu_load/vcpu_put pair
1245 	 * should be used. Since CSR registers owns by this vcpu, if switch
1246 	 * to other vcpus, other vcpus need reload CSR registers.
1247 	 *
1248 	 * If software CSR is modified, bit KVM_LARCH_HWCSR_USABLE should
1249 	 * be clear in vcpu->arch.aux_inuse, and vcpu_load will check
1250 	 * aux_inuse flag and reload CSR registers form software.
1251 	 */
1252 
1253 	switch (ioctl) {
1254 	case KVM_SET_ONE_REG:
1255 	case KVM_GET_ONE_REG: {
1256 		struct kvm_one_reg reg;
1257 
1258 		r = -EFAULT;
1259 		if (copy_from_user(&reg, argp, sizeof(reg)))
1260 			break;
1261 		if (ioctl == KVM_SET_ONE_REG) {
1262 			r = kvm_set_reg(vcpu, &reg);
1263 			vcpu->arch.aux_inuse &= ~KVM_LARCH_HWCSR_USABLE;
1264 		} else
1265 			r = kvm_get_reg(vcpu, &reg);
1266 		break;
1267 	}
1268 	case KVM_ENABLE_CAP: {
1269 		struct kvm_enable_cap cap;
1270 
1271 		r = -EFAULT;
1272 		if (copy_from_user(&cap, argp, sizeof(cap)))
1273 			break;
1274 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1275 		break;
1276 	}
1277 	case KVM_HAS_DEVICE_ATTR: {
1278 		r = -EFAULT;
1279 		if (copy_from_user(&attr, argp, sizeof(attr)))
1280 			break;
1281 		r = kvm_loongarch_vcpu_has_attr(vcpu, &attr);
1282 		break;
1283 	}
1284 	case KVM_GET_DEVICE_ATTR: {
1285 		r = -EFAULT;
1286 		if (copy_from_user(&attr, argp, sizeof(attr)))
1287 			break;
1288 		r = kvm_loongarch_vcpu_get_attr(vcpu, &attr);
1289 		break;
1290 	}
1291 	case KVM_SET_DEVICE_ATTR: {
1292 		r = -EFAULT;
1293 		if (copy_from_user(&attr, argp, sizeof(attr)))
1294 			break;
1295 		r = kvm_loongarch_vcpu_set_attr(vcpu, &attr);
1296 		break;
1297 	}
1298 	default:
1299 		r = -ENOIOCTLCMD;
1300 		break;
1301 	}
1302 
1303 	return r;
1304 }
1305 
1306 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1307 {
1308 	int i = 0;
1309 
1310 	fpu->fcc = vcpu->arch.fpu.fcc;
1311 	fpu->fcsr = vcpu->arch.fpu.fcsr;
1312 	for (i = 0; i < NUM_FPU_REGS; i++)
1313 		memcpy(&fpu->fpr[i], &vcpu->arch.fpu.fpr[i], FPU_REG_WIDTH / 64);
1314 
1315 	return 0;
1316 }
1317 
1318 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1319 {
1320 	int i = 0;
1321 
1322 	vcpu->arch.fpu.fcc = fpu->fcc;
1323 	vcpu->arch.fpu.fcsr = fpu->fcsr;
1324 	for (i = 0; i < NUM_FPU_REGS; i++)
1325 		memcpy(&vcpu->arch.fpu.fpr[i], &fpu->fpr[i], FPU_REG_WIDTH / 64);
1326 
1327 	return 0;
1328 }
1329 
1330 #ifdef CONFIG_CPU_HAS_LBT
1331 int kvm_own_lbt(struct kvm_vcpu *vcpu)
1332 {
1333 	if (!(vcpu->arch.aux_inuse & KVM_LARCH_LBT)) {
1334 		set_csr_euen(CSR_EUEN_LBTEN);
1335 		_restore_lbt(&vcpu->arch.lbt);
1336 		vcpu->arch.aux_inuse |= KVM_LARCH_LBT;
1337 	}
1338 
1339 	return 0;
1340 }
1341 
1342 static void kvm_lose_lbt(struct kvm_vcpu *vcpu)
1343 {
1344 	preempt_disable();
1345 	if (vcpu->arch.aux_inuse & KVM_LARCH_LBT) {
1346 		_save_lbt(&vcpu->arch.lbt);
1347 		clear_csr_euen(CSR_EUEN_LBTEN);
1348 		vcpu->arch.aux_inuse &= ~KVM_LARCH_LBT;
1349 	}
1350 	preempt_enable();
1351 }
1352 
1353 static void kvm_check_fcsr(struct kvm_vcpu *vcpu, unsigned long fcsr)
1354 {
1355 	/*
1356 	 * If TM is enabled, top register save/restore will
1357 	 * cause lbt exception, here enable lbt in advance
1358 	 */
1359 	if (fcsr & FPU_CSR_TM)
1360 		kvm_own_lbt(vcpu);
1361 }
1362 
1363 static void kvm_check_fcsr_alive(struct kvm_vcpu *vcpu)
1364 {
1365 	if (vcpu->arch.aux_inuse & KVM_LARCH_FPU) {
1366 		if (vcpu->arch.aux_inuse & KVM_LARCH_LBT)
1367 			return;
1368 		kvm_check_fcsr(vcpu, read_fcsr(LOONGARCH_FCSR0));
1369 	}
1370 }
1371 #else
1372 static inline void kvm_lose_lbt(struct kvm_vcpu *vcpu) { }
1373 static inline void kvm_check_fcsr(struct kvm_vcpu *vcpu, unsigned long fcsr) { }
1374 static inline void kvm_check_fcsr_alive(struct kvm_vcpu *vcpu) { }
1375 #endif
1376 
1377 /* Enable FPU and restore context */
1378 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1379 {
1380 	/*
1381 	 * Enable FPU for guest
1382 	 * Set FR and FRE according to guest context
1383 	 */
1384 	kvm_check_fcsr(vcpu, vcpu->arch.fpu.fcsr);
1385 	set_csr_euen(CSR_EUEN_FPEN);
1386 
1387 	kvm_restore_fpu(&vcpu->arch.fpu);
1388 	vcpu->arch.aux_inuse |= KVM_LARCH_FPU;
1389 	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1390 }
1391 
1392 #ifdef CONFIG_CPU_HAS_LSX
1393 /* Enable LSX and restore context */
1394 int kvm_own_lsx(struct kvm_vcpu *vcpu)
1395 {
1396 	/* Enable LSX for guest */
1397 	kvm_check_fcsr(vcpu, vcpu->arch.fpu.fcsr);
1398 	set_csr_euen(CSR_EUEN_LSXEN | CSR_EUEN_FPEN);
1399 	switch (vcpu->arch.aux_inuse & KVM_LARCH_FPU) {
1400 	case KVM_LARCH_FPU:
1401 		/*
1402 		 * Guest FPU state already loaded,
1403 		 * only restore upper LSX state
1404 		 */
1405 		_restore_lsx_upper(&vcpu->arch.fpu);
1406 		break;
1407 	default:
1408 		/* Neither FP or LSX already active,
1409 		 * restore full LSX state
1410 		 */
1411 		kvm_restore_lsx(&vcpu->arch.fpu);
1412 		break;
1413 	}
1414 
1415 	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_LSX);
1416 	vcpu->arch.aux_inuse |= KVM_LARCH_LSX | KVM_LARCH_FPU;
1417 
1418 	return 0;
1419 }
1420 #endif
1421 
1422 #ifdef CONFIG_CPU_HAS_LASX
1423 /* Enable LASX and restore context */
1424 int kvm_own_lasx(struct kvm_vcpu *vcpu)
1425 {
1426 	kvm_check_fcsr(vcpu, vcpu->arch.fpu.fcsr);
1427 	set_csr_euen(CSR_EUEN_FPEN | CSR_EUEN_LSXEN | CSR_EUEN_LASXEN);
1428 	switch (vcpu->arch.aux_inuse & (KVM_LARCH_FPU | KVM_LARCH_LSX)) {
1429 	case KVM_LARCH_LSX:
1430 	case KVM_LARCH_LSX | KVM_LARCH_FPU:
1431 		/* Guest LSX state already loaded, only restore upper LASX state */
1432 		_restore_lasx_upper(&vcpu->arch.fpu);
1433 		break;
1434 	case KVM_LARCH_FPU:
1435 		/* Guest FP state already loaded, only restore upper LSX & LASX state */
1436 		_restore_lsx_upper(&vcpu->arch.fpu);
1437 		_restore_lasx_upper(&vcpu->arch.fpu);
1438 		break;
1439 	default:
1440 		/* Neither FP or LSX already active, restore full LASX state */
1441 		kvm_restore_lasx(&vcpu->arch.fpu);
1442 		break;
1443 	}
1444 
1445 	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_LASX);
1446 	vcpu->arch.aux_inuse |= KVM_LARCH_LASX | KVM_LARCH_LSX | KVM_LARCH_FPU;
1447 
1448 	return 0;
1449 }
1450 #endif
1451 
1452 /* Save context and disable FPU */
1453 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1454 {
1455 	preempt_disable();
1456 
1457 	kvm_check_fcsr_alive(vcpu);
1458 	if (vcpu->arch.aux_inuse & KVM_LARCH_LASX) {
1459 		kvm_save_lasx(&vcpu->arch.fpu);
1460 		vcpu->arch.aux_inuse &= ~(KVM_LARCH_LSX | KVM_LARCH_FPU | KVM_LARCH_LASX);
1461 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_LASX);
1462 
1463 		/* Disable LASX & LSX & FPU */
1464 		clear_csr_euen(CSR_EUEN_FPEN | CSR_EUEN_LSXEN | CSR_EUEN_LASXEN);
1465 	} else if (vcpu->arch.aux_inuse & KVM_LARCH_LSX) {
1466 		kvm_save_lsx(&vcpu->arch.fpu);
1467 		vcpu->arch.aux_inuse &= ~(KVM_LARCH_LSX | KVM_LARCH_FPU);
1468 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_LSX);
1469 
1470 		/* Disable LSX & FPU */
1471 		clear_csr_euen(CSR_EUEN_FPEN | CSR_EUEN_LSXEN);
1472 	} else if (vcpu->arch.aux_inuse & KVM_LARCH_FPU) {
1473 		kvm_save_fpu(&vcpu->arch.fpu);
1474 		vcpu->arch.aux_inuse &= ~KVM_LARCH_FPU;
1475 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1476 
1477 		/* Disable FPU */
1478 		clear_csr_euen(CSR_EUEN_FPEN);
1479 	}
1480 	kvm_lose_lbt(vcpu);
1481 
1482 	preempt_enable();
1483 }
1484 
1485 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1486 {
1487 	int intr = (int)irq->irq;
1488 
1489 	if (intr > 0)
1490 		kvm_queue_irq(vcpu, intr);
1491 	else if (intr < 0)
1492 		kvm_dequeue_irq(vcpu, -intr);
1493 	else {
1494 		kvm_err("%s: invalid interrupt ioctl %d\n", __func__, irq->irq);
1495 		return -EINVAL;
1496 	}
1497 
1498 	kvm_vcpu_kick(vcpu);
1499 
1500 	return 0;
1501 }
1502 
1503 long kvm_arch_vcpu_unlocked_ioctl(struct file *filp, unsigned int ioctl,
1504 				  unsigned long arg)
1505 {
1506 	void __user *argp = (void __user *)arg;
1507 	struct kvm_vcpu *vcpu = filp->private_data;
1508 
1509 	if (ioctl == KVM_INTERRUPT) {
1510 		struct kvm_interrupt irq;
1511 
1512 		if (copy_from_user(&irq, argp, sizeof(irq)))
1513 			return -EFAULT;
1514 
1515 		kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__, irq.irq);
1516 
1517 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1518 	}
1519 
1520 	return -ENOIOCTLCMD;
1521 }
1522 
1523 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
1524 {
1525 	return 0;
1526 }
1527 
1528 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
1529 {
1530 	unsigned long timer_hz;
1531 	struct loongarch_csrs *csr;
1532 
1533 	vcpu->arch.vpid = 0;
1534 	vcpu->arch.flush_gpa = INVALID_GPA;
1535 
1536 	hrtimer_setup(&vcpu->arch.swtimer, kvm_swtimer_wakeup, CLOCK_MONOTONIC,
1537 		      HRTIMER_MODE_ABS_PINNED_HARD);
1538 
1539 	/* Get GPA (=HVA) of PGD for kvm hypervisor */
1540 	vcpu->arch.kvm_pgd = __pa(vcpu->kvm->arch.pgd);
1541 
1542 	/*
1543 	 * Get PGD for primary mmu, virtual address is used since there is
1544 	 * memory access after loading from CSR_PGD in tlb exception fast path.
1545 	 */
1546 	vcpu->arch.host_pgd = (unsigned long)vcpu->kvm->mm->pgd;
1547 
1548 	vcpu->arch.handle_exit = kvm_handle_exit;
1549 	vcpu->arch.guest_eentry = (unsigned long)kvm_loongarch_ops->exc_entry;
1550 	vcpu->arch.csr = kzalloc(sizeof(struct loongarch_csrs), GFP_KERNEL);
1551 	if (!vcpu->arch.csr)
1552 		return -ENOMEM;
1553 
1554 	/*
1555 	 * All kvm exceptions share one exception entry, and host <-> guest
1556 	 * switch also switch ECFG.VS field, keep host ECFG.VS info here.
1557 	 */
1558 	vcpu->arch.host_ecfg = (read_csr_ecfg() & CSR_ECFG_VS);
1559 
1560 	/* Init */
1561 	vcpu->arch.last_sched_cpu = -1;
1562 
1563 	/* Init ipi_state lock */
1564 	spin_lock_init(&vcpu->arch.ipi_state.lock);
1565 
1566 	/*
1567 	 * Initialize guest register state to valid architectural reset state.
1568 	 */
1569 	timer_hz = calc_const_freq();
1570 	kvm_init_timer(vcpu, timer_hz);
1571 
1572 	/* Set Initialize mode for guest */
1573 	csr = vcpu->arch.csr;
1574 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CRMD, CSR_CRMD_DA);
1575 
1576 	/* Set cpuid */
1577 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_TMID, vcpu->vcpu_id);
1578 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CPUID, KVM_MAX_PHYID);
1579 
1580 	/* Start with no pending virtual guest interrupts */
1581 	csr->csrs[LOONGARCH_CSR_GINTC] = 0;
1582 
1583 	return 0;
1584 }
1585 
1586 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1587 {
1588 }
1589 
1590 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1591 {
1592 	int cpu;
1593 	struct kvm_context *context;
1594 
1595 	hrtimer_cancel(&vcpu->arch.swtimer);
1596 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
1597 	kvm_drop_cpuid(vcpu);
1598 	kfree(vcpu->arch.csr);
1599 
1600 	/*
1601 	 * If the vCPU is freed and reused as another vCPU, we don't want the
1602 	 * matching pointer wrongly hanging around in last_vcpu.
1603 	 */
1604 	for_each_possible_cpu(cpu) {
1605 		context = per_cpu_ptr(vcpu->kvm->arch.vmcs, cpu);
1606 		if (context->last_vcpu == vcpu)
1607 			context->last_vcpu = NULL;
1608 	}
1609 }
1610 
1611 static int _kvm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1612 {
1613 	bool migrated;
1614 	struct kvm_context *context;
1615 	struct loongarch_csrs *csr = vcpu->arch.csr;
1616 
1617 	/*
1618 	 * Have we migrated to a different CPU?
1619 	 * If so, any old guest TLB state may be stale.
1620 	 */
1621 	migrated = (vcpu->arch.last_sched_cpu != cpu);
1622 
1623 	/*
1624 	 * Was this the last vCPU to run on this CPU?
1625 	 * If not, any old guest state from this vCPU will have been clobbered.
1626 	 */
1627 	context = per_cpu_ptr(vcpu->kvm->arch.vmcs, cpu);
1628 	if (migrated || (context->last_vcpu != vcpu))
1629 		vcpu->arch.aux_inuse &= ~KVM_LARCH_HWCSR_USABLE;
1630 	context->last_vcpu = vcpu;
1631 
1632 	/* Restore timer state regardless */
1633 	kvm_restore_timer(vcpu);
1634 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1635 
1636 	/* Don't bother restoring registers multiple times unless necessary */
1637 	if (vcpu->arch.aux_inuse & KVM_LARCH_HWCSR_USABLE)
1638 		return 0;
1639 
1640 	write_csr_gcntc((ulong)vcpu->kvm->arch.time_offset);
1641 
1642 	/* Restore guest CSR registers */
1643 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_CRMD);
1644 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PRMD);
1645 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_EUEN);
1646 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_MISC);
1647 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ECFG);
1648 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ERA);
1649 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_BADV);
1650 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_BADI);
1651 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_EENTRY);
1652 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBIDX);
1653 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBEHI);
1654 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBELO0);
1655 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBELO1);
1656 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ASID);
1657 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PGDL);
1658 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PGDH);
1659 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PWCTL0);
1660 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PWCTL1);
1661 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_STLBPGSIZE);
1662 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_RVACFG);
1663 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_CPUID);
1664 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS0);
1665 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS1);
1666 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS2);
1667 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS3);
1668 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS4);
1669 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS5);
1670 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS6);
1671 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS7);
1672 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TMID);
1673 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_CNTC);
1674 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRENTRY);
1675 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRBADV);
1676 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRERA);
1677 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRSAVE);
1678 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO0);
1679 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO1);
1680 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBREHI);
1681 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRPRMD);
1682 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN0);
1683 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN1);
1684 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN2);
1685 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN3);
1686 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_LLBCTL);
1687 
1688 	if (kvm_guest_has_msgint(&vcpu->arch)) {
1689 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_IPR);
1690 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR0);
1691 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR1);
1692 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR2);
1693 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR3);
1694 	}
1695 
1696 	/* Restore Root.GINTC from unused Guest.GINTC register */
1697 	write_csr_gintc(csr->csrs[LOONGARCH_CSR_GINTC]);
1698 
1699 	/*
1700 	 * We should clear linked load bit to break interrupted atomics. This
1701 	 * prevents a SC on the next vCPU from succeeding by matching a LL on
1702 	 * the previous vCPU.
1703 	 */
1704 	if (vcpu->kvm->created_vcpus > 1)
1705 		set_gcsr_llbctl(CSR_LLBCTL_WCLLB);
1706 
1707 	vcpu->arch.aux_inuse |= KVM_LARCH_HWCSR_USABLE;
1708 
1709 	return 0;
1710 }
1711 
1712 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1713 {
1714 	unsigned long flags;
1715 
1716 	local_irq_save(flags);
1717 	/* Restore guest state to registers */
1718 	_kvm_vcpu_load(vcpu, cpu);
1719 	local_irq_restore(flags);
1720 }
1721 
1722 static int _kvm_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
1723 {
1724 	struct loongarch_csrs *csr = vcpu->arch.csr;
1725 
1726 	kvm_lose_fpu(vcpu);
1727 
1728 	/*
1729 	 * Update CSR state from hardware if software CSR state is stale,
1730 	 * most CSR registers are kept unchanged during process context
1731 	 * switch except CSR registers like remaining timer tick value and
1732 	 * injected interrupt state.
1733 	 */
1734 	if (vcpu->arch.aux_inuse & KVM_LARCH_SWCSR_LATEST)
1735 		goto out;
1736 
1737 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_CRMD);
1738 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRMD);
1739 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_EUEN);
1740 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_MISC);
1741 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ECFG);
1742 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ERA);
1743 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_BADV);
1744 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_BADI);
1745 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_EENTRY);
1746 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBIDX);
1747 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBEHI);
1748 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBELO0);
1749 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBELO1);
1750 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ASID);
1751 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PGDL);
1752 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PGDH);
1753 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PWCTL0);
1754 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PWCTL1);
1755 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_STLBPGSIZE);
1756 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_RVACFG);
1757 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_CPUID);
1758 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRCFG1);
1759 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRCFG2);
1760 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRCFG3);
1761 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS0);
1762 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS1);
1763 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS2);
1764 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS3);
1765 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS4);
1766 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS5);
1767 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS6);
1768 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS7);
1769 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TMID);
1770 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_CNTC);
1771 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_LLBCTL);
1772 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRENTRY);
1773 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRBADV);
1774 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRERA);
1775 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRSAVE);
1776 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO0);
1777 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO1);
1778 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBREHI);
1779 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRPRMD);
1780 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN0);
1781 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN1);
1782 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN2);
1783 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN3);
1784 
1785 	if (kvm_guest_has_msgint(&vcpu->arch)) {
1786 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_IPR);
1787 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR0);
1788 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR1);
1789 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR2);
1790 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR3);
1791 	}
1792 
1793 	vcpu->arch.aux_inuse |= KVM_LARCH_SWCSR_LATEST;
1794 
1795 out:
1796 	kvm_save_timer(vcpu);
1797 	/* Save Root.GINTC into unused Guest.GINTC register */
1798 	csr->csrs[LOONGARCH_CSR_GINTC] = read_csr_gintc();
1799 
1800 	return 0;
1801 }
1802 
1803 static void kvm_vcpu_set_pv_preempted(struct kvm_vcpu *vcpu)
1804 {
1805 	gpa_t gpa;
1806 	struct gfn_to_hva_cache *ghc;
1807 	struct kvm_memslots *slots;
1808 	struct kvm_steal_time __user *st;
1809 
1810 	gpa = vcpu->arch.st.guest_addr;
1811 	if (!(gpa & KVM_STEAL_PHYS_VALID))
1812 		return;
1813 
1814 	/* vCPU may be preempted for many times */
1815 	if (vcpu->arch.st.preempted)
1816 		return;
1817 
1818 	/* This happens on process exit */
1819 	if (unlikely(current->mm != vcpu->kvm->mm))
1820 		return;
1821 
1822 	gpa &= KVM_STEAL_PHYS_MASK;
1823 	ghc = &vcpu->arch.st.cache;
1824 	slots = kvm_memslots(vcpu->kvm);
1825 	if (slots->generation != ghc->generation || gpa != ghc->gpa) {
1826 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st))) {
1827 			ghc->gpa = INVALID_GPA;
1828 			return;
1829 		}
1830 	}
1831 
1832 	st = (struct kvm_steal_time __user *)ghc->hva;
1833 	unsafe_put_user(KVM_VCPU_PREEMPTED, &st->preempted, out);
1834 	vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
1835 out:
1836 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
1837 }
1838 
1839 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1840 {
1841 	int cpu, idx;
1842 	unsigned long flags;
1843 
1844 	if (vcpu->preempted && kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_PREEMPT)) {
1845 		/*
1846 		 * Take the srcu lock as memslots will be accessed to check
1847 		 * the gfn cache generation against the memslots generation.
1848 		 */
1849 		idx = srcu_read_lock(&vcpu->kvm->srcu);
1850 		kvm_vcpu_set_pv_preempted(vcpu);
1851 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
1852 	}
1853 
1854 	local_irq_save(flags);
1855 	cpu = smp_processor_id();
1856 	vcpu->arch.last_sched_cpu = cpu;
1857 
1858 	/* Save guest state in registers */
1859 	_kvm_vcpu_put(vcpu, cpu);
1860 	local_irq_restore(flags);
1861 }
1862 
1863 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1864 {
1865 	int r = -EINTR;
1866 	struct kvm_run *run = vcpu->run;
1867 
1868 	if (vcpu->mmio_needed) {
1869 		if (!vcpu->mmio_is_write)
1870 			kvm_complete_mmio_read(vcpu, run);
1871 		vcpu->mmio_needed = 0;
1872 	}
1873 
1874 	switch (run->exit_reason) {
1875 	case KVM_EXIT_HYPERCALL:
1876 		kvm_complete_user_service(vcpu, run);
1877 		break;
1878 	case KVM_EXIT_LOONGARCH_IOCSR:
1879 		if (!run->iocsr_io.is_write)
1880 			kvm_complete_iocsr_read(vcpu, run);
1881 		break;
1882 	}
1883 
1884 	if (!vcpu->wants_to_run)
1885 		return r;
1886 
1887 	/* Clear exit_reason */
1888 	run->exit_reason = KVM_EXIT_UNKNOWN;
1889 	lose_fpu(1);
1890 	vcpu_load(vcpu);
1891 	kvm_sigset_activate(vcpu);
1892 	r = kvm_pre_enter_guest(vcpu);
1893 	if (r != RESUME_GUEST)
1894 		goto out;
1895 
1896 	guest_timing_enter_irqoff();
1897 	guest_state_enter_irqoff();
1898 	trace_kvm_enter(vcpu);
1899 	r = kvm_loongarch_ops->enter_guest(run, vcpu);
1900 
1901 	trace_kvm_out(vcpu);
1902 	/*
1903 	 * Guest exit is already recorded at kvm_handle_exit()
1904 	 * return value must not be RESUME_GUEST
1905 	 */
1906 	local_irq_enable();
1907 out:
1908 	kvm_sigset_deactivate(vcpu);
1909 	vcpu_put(vcpu);
1910 
1911 	return r;
1912 }
1913