xref: /linux/arch/loongarch/kvm/vcpu.c (revision fb89e0fe2dc4246c86ad0eb0fa2b7cb2f8ba9728)
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_update_stolen_time(struct kvm_vcpu *vcpu)
153 {
154 	u32 version;
155 	u64 steal;
156 	gpa_t gpa;
157 	struct kvm_memslots *slots;
158 	struct kvm_steal_time __user *st;
159 	struct gfn_to_hva_cache *ghc;
160 
161 	ghc = &vcpu->arch.st.cache;
162 	gpa = vcpu->arch.st.guest_addr;
163 	if (!(gpa & KVM_STEAL_PHYS_VALID))
164 		return;
165 
166 	gpa &= KVM_STEAL_PHYS_MASK;
167 	slots = kvm_memslots(vcpu->kvm);
168 	if (slots->generation != ghc->generation || gpa != ghc->gpa) {
169 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st))) {
170 			ghc->gpa = INVALID_GPA;
171 			return;
172 		}
173 	}
174 
175 	st = (struct kvm_steal_time __user *)ghc->hva;
176 	if (kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_PREEMPT)) {
177 		unsafe_put_user(0, &st->preempted, out);
178 		vcpu->arch.st.preempted = 0;
179 	}
180 
181 	unsafe_get_user(version, &st->version, out);
182 	if (version & 1)
183 		version += 1; /* first time write, random junk */
184 
185 	version += 1;
186 	unsafe_put_user(version, &st->version, out);
187 	smp_wmb();
188 
189 	unsafe_get_user(steal, &st->steal, out);
190 	steal += current->sched_info.run_delay - vcpu->arch.st.last_steal;
191 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
192 	unsafe_put_user(steal, &st->steal, out);
193 
194 	smp_wmb();
195 	version += 1;
196 	unsafe_put_user(version, &st->version, out);
197 out:
198 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
199 }
200 
201 /*
202  * kvm_check_requests - check and handle pending vCPU requests
203  *
204  * Return: RESUME_GUEST if we should enter the guest
205  *         RESUME_HOST  if we should exit to userspace
206  */
207 static int kvm_check_requests(struct kvm_vcpu *vcpu)
208 {
209 	if (!kvm_request_pending(vcpu))
210 		return RESUME_GUEST;
211 
212 	if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
213 		vcpu->arch.vpid = 0;  /* Drop vpid for this vCPU */
214 
215 	if (kvm_dirty_ring_check_request(vcpu))
216 		return RESUME_HOST;
217 
218 	if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
219 		kvm_update_stolen_time(vcpu);
220 
221 	return RESUME_GUEST;
222 }
223 
224 static void kvm_late_check_requests(struct kvm_vcpu *vcpu)
225 {
226 	lockdep_assert_irqs_disabled();
227 
228 	if (!kvm_request_pending(vcpu))
229 		return;
230 
231 	if (kvm_check_request(KVM_REQ_PMU, vcpu)) {
232 		kvm_own_pmu(vcpu);
233 		vcpu->arch.aux_inuse |= KVM_LARCH_PMU;
234 	}
235 
236 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_GPA, vcpu))
237 		if (vcpu->arch.flush_gpa != INVALID_GPA) {
238 			kvm_flush_tlb_gpa(vcpu, vcpu->arch.flush_gpa);
239 			vcpu->arch.flush_gpa = INVALID_GPA;
240 		}
241 
242 	if (kvm_check_request(KVM_REQ_FPU_LOAD, vcpu)) {
243 		if (kvm_guest_has_lasx(&vcpu->arch))
244 			kvm_own_lasx(vcpu);
245 		else if (kvm_guest_has_lsx(&vcpu->arch))
246 			kvm_own_lsx(vcpu);
247 		else if (kvm_guest_has_fpu(&vcpu->arch))
248 			kvm_own_fpu(vcpu);
249 	}
250 
251 	if (kvm_check_request(KVM_REQ_LBT_LOAD, vcpu))
252 		kvm_own_lbt(vcpu);
253 }
254 
255 /*
256  * Check and handle pending signal and vCPU requests etc
257  * Run with irq enabled and preempt enabled
258  *
259  * Return: RESUME_GUEST if we should enter the guest
260  *         RESUME_HOST  if we should exit to userspace
261  *         < 0 if we should exit to userspace, where the return value
262  *         indicates an error
263  */
264 static int kvm_enter_guest_check(struct kvm_vcpu *vcpu)
265 {
266 	int idx, ret;
267 
268 	/*
269 	 * Check conditions before entering the guest
270 	 */
271 	ret = kvm_xfer_to_guest_mode_handle_work(vcpu);
272 	if (ret < 0)
273 		return ret;
274 
275 	idx = srcu_read_lock(&vcpu->kvm->srcu);
276 	ret = kvm_check_requests(vcpu);
277 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
278 
279 	return ret;
280 }
281 
282 /*
283  * Called with irq enabled
284  *
285  * Return: RESUME_GUEST if we should enter the guest, and irq disabled
286  *         Others if we should exit to userspace
287  */
288 static int kvm_pre_enter_guest(struct kvm_vcpu *vcpu)
289 {
290 	int ret;
291 
292 	do {
293 		ret = kvm_enter_guest_check(vcpu);
294 		if (ret != RESUME_GUEST)
295 			break;
296 
297 		/*
298 		 * Handle vcpu timer, interrupts, check requests and
299 		 * check vmid before vcpu enter guest
300 		 */
301 		local_irq_disable();
302 		kvm_deliver_exception(vcpu);
303 		/* Make sure the vcpu mode has been written */
304 		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
305 		kvm_deliver_intr(vcpu);
306 		kvm_check_vpid(vcpu);
307 
308 		/*
309 		 * Called after function kvm_check_vpid()
310 		 * Since it updates CSR.GSTAT used by kvm_flush_tlb_gpa(),
311 		 * and it may also clear KVM_REQ_TLB_FLUSH_GPA pending bit
312 		 */
313 		kvm_late_check_requests(vcpu);
314 		/* Clear KVM_LARCH_SWCSR_LATEST as CSR will change when enter guest */
315 		vcpu->arch.aux_inuse &= ~KVM_LARCH_SWCSR_LATEST;
316 
317 		if (kvm_request_pending(vcpu) || xfer_to_guest_mode_work_pending()) {
318 			if (vcpu->arch.aux_inuse & KVM_LARCH_PMU) {
319 				kvm_lose_pmu(vcpu);
320 				kvm_make_request(KVM_REQ_PMU, vcpu);
321 			}
322 			/* make sure the vcpu mode has been written */
323 			smp_store_mb(vcpu->mode, OUTSIDE_GUEST_MODE);
324 			local_irq_enable();
325 			ret = -EAGAIN;
326 		}
327 	} while (ret != RESUME_GUEST);
328 
329 	return ret;
330 }
331 
332 /*
333  * Return 1 for resume guest and "<= 0" for resume host.
334  */
335 static int kvm_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
336 {
337 	int ret = RESUME_GUEST;
338 	unsigned long estat = vcpu->arch.host_estat;
339 	u32 intr = estat & CSR_ESTAT_IS;
340 	u32 ecode = (estat & CSR_ESTAT_EXC) >> CSR_ESTAT_EXC_SHIFT;
341 
342 	smp_store_mb(vcpu->mode, OUTSIDE_GUEST_MODE);
343 
344 	/* Set a default exit reason */
345 	run->exit_reason = KVM_EXIT_UNKNOWN;
346 
347 	kvm_lose_pmu(vcpu);
348 
349 	guest_timing_exit_irqoff();
350 	guest_state_exit_irqoff();
351 	local_irq_enable();
352 
353 	trace_kvm_exit(vcpu, ecode);
354 	if (ecode) {
355 		ret = kvm_handle_fault(vcpu, ecode);
356 	} else {
357 		WARN(!intr, "vm exiting with suspicious irq\n");
358 		++vcpu->stat.int_exits;
359 	}
360 
361 	if (ret == RESUME_GUEST)
362 		ret = kvm_pre_enter_guest(vcpu);
363 
364 	if (ret != RESUME_GUEST) {
365 		local_irq_disable();
366 		return ret;
367 	}
368 
369 	guest_timing_enter_irqoff();
370 	guest_state_enter_irqoff();
371 	trace_kvm_reenter(vcpu);
372 
373 	return RESUME_GUEST;
374 }
375 
376 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
377 {
378 	return !!(vcpu->arch.irq_pending) &&
379 		vcpu->arch.mp_state.mp_state == KVM_MP_STATE_RUNNABLE;
380 }
381 
382 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
383 {
384 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
385 }
386 
387 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
388 {
389 	unsigned long val;
390 
391 	preempt_disable();
392 	val = gcsr_read(LOONGARCH_CSR_CRMD);
393 	preempt_enable();
394 
395 	return (val & CSR_CRMD_PLV) == PLV_KERN;
396 }
397 
398 #ifdef CONFIG_GUEST_PERF_EVENTS
399 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
400 {
401 	return vcpu->arch.pc;
402 }
403 
404 /*
405  * Returns true if a Performance Monitoring Interrupt (PMI), a.k.a. perf event,
406  * arrived in guest context.  For LoongArch64, if PMU is not passthrough to VM,
407  * any event that arrives while a vCPU is loaded is considered to be "in guest".
408  */
409 bool kvm_arch_pmi_in_guest(struct kvm_vcpu *vcpu)
410 {
411 	return (vcpu && !(vcpu->arch.aux_inuse & KVM_LARCH_PMU));
412 }
413 #endif
414 
415 bool kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu)
416 {
417 	return false;
418 }
419 
420 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
421 {
422 	return VM_FAULT_SIGBUS;
423 }
424 
425 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
426 				  struct kvm_translation *tr)
427 {
428 	return -EINVAL;
429 }
430 
431 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
432 {
433 	int ret;
434 
435 	/* Protect from TOD sync and vcpu_load/put() */
436 	preempt_disable();
437 	ret = kvm_pending_timer(vcpu) ||
438 		kvm_read_hw_gcsr(LOONGARCH_CSR_ESTAT) & (1 << INT_TI);
439 	preempt_enable();
440 
441 	return ret;
442 }
443 
444 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
445 {
446 	int i;
447 
448 	kvm_debug("vCPU Register Dump:\n");
449 	kvm_debug("\tPC = 0x%08lx\n", vcpu->arch.pc);
450 	kvm_debug("\tExceptions: %08lx\n", vcpu->arch.irq_pending);
451 
452 	for (i = 0; i < 32; i += 4) {
453 		kvm_debug("\tGPR%02d: %08lx %08lx %08lx %08lx\n", i,
454 		       vcpu->arch.gprs[i], vcpu->arch.gprs[i + 1],
455 		       vcpu->arch.gprs[i + 2], vcpu->arch.gprs[i + 3]);
456 	}
457 
458 	kvm_debug("\tCRMD: 0x%08lx, ESTAT: 0x%08lx\n",
459 		  kvm_read_hw_gcsr(LOONGARCH_CSR_CRMD),
460 		  kvm_read_hw_gcsr(LOONGARCH_CSR_ESTAT));
461 
462 	kvm_debug("\tERA: 0x%08lx\n", kvm_read_hw_gcsr(LOONGARCH_CSR_ERA));
463 
464 	return 0;
465 }
466 
467 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
468 				struct kvm_mp_state *mp_state)
469 {
470 	*mp_state = vcpu->arch.mp_state;
471 
472 	return 0;
473 }
474 
475 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
476 				struct kvm_mp_state *mp_state)
477 {
478 	int ret = 0;
479 
480 	switch (mp_state->mp_state) {
481 	case KVM_MP_STATE_RUNNABLE:
482 		vcpu->arch.mp_state = *mp_state;
483 		break;
484 	default:
485 		ret = -EINVAL;
486 	}
487 
488 	return ret;
489 }
490 
491 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
492 					struct kvm_guest_debug *dbg)
493 {
494 	if (dbg->control & ~KVM_GUESTDBG_VALID_MASK)
495 		return -EINVAL;
496 
497 	if (dbg->control & KVM_GUESTDBG_ENABLE)
498 		vcpu->guest_debug = dbg->control;
499 	else
500 		vcpu->guest_debug = 0;
501 
502 	return 0;
503 }
504 
505 static inline int kvm_set_cpuid(struct kvm_vcpu *vcpu, u64 val)
506 {
507 	int cpuid;
508 	struct kvm_phyid_map *map;
509 	struct loongarch_csrs *csr = vcpu->arch.csr;
510 
511 	if (val >= KVM_MAX_PHYID)
512 		return -EINVAL;
513 
514 	map = vcpu->kvm->arch.phyid_map;
515 	cpuid = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_CPUID);
516 
517 	spin_lock(&vcpu->kvm->arch.phyid_map_lock);
518 	if ((cpuid < KVM_MAX_PHYID) && map->phys_map[cpuid].enabled) {
519 		/* Discard duplicated CPUID set operation */
520 		if (cpuid == val) {
521 			spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
522 			return 0;
523 		}
524 
525 		/*
526 		 * CPUID is already set before
527 		 * Forbid changing to a different CPUID at runtime
528 		 */
529 		spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
530 		return -EINVAL;
531 	}
532 
533 	if (map->phys_map[val].enabled) {
534 		/* Discard duplicated CPUID set operation */
535 		if (vcpu == map->phys_map[val].vcpu) {
536 			spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
537 			return 0;
538 		}
539 
540 		/*
541 		 * New CPUID is already set with other vcpu
542 		 * Forbid sharing the same CPUID between different vcpus
543 		 */
544 		spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
545 		return -EINVAL;
546 	}
547 
548 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CPUID, val);
549 	map->phys_map[val].enabled	= true;
550 	map->phys_map[val].vcpu		= vcpu;
551 	spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
552 
553 	return 0;
554 }
555 
556 static inline void kvm_drop_cpuid(struct kvm_vcpu *vcpu)
557 {
558 	int cpuid;
559 	struct kvm_phyid_map *map;
560 	struct loongarch_csrs *csr = vcpu->arch.csr;
561 
562 	map = vcpu->kvm->arch.phyid_map;
563 	cpuid = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_CPUID);
564 
565 	if (cpuid >= KVM_MAX_PHYID)
566 		return;
567 
568 	spin_lock(&vcpu->kvm->arch.phyid_map_lock);
569 	if (map->phys_map[cpuid].enabled) {
570 		map->phys_map[cpuid].vcpu = NULL;
571 		map->phys_map[cpuid].enabled = false;
572 		kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CPUID, KVM_MAX_PHYID);
573 	}
574 	spin_unlock(&vcpu->kvm->arch.phyid_map_lock);
575 }
576 
577 struct kvm_vcpu *kvm_get_vcpu_by_cpuid(struct kvm *kvm, int cpuid)
578 {
579 	struct kvm_phyid_map *map;
580 
581 	if (cpuid < 0)
582 		return NULL;
583 
584 	if (cpuid >= KVM_MAX_PHYID)
585 		return NULL;
586 
587 	map = kvm->arch.phyid_map;
588 	if (!map->phys_map[cpuid].enabled)
589 		return NULL;
590 
591 	return map->phys_map[cpuid].vcpu;
592 }
593 
594 static int _kvm_getcsr(struct kvm_vcpu *vcpu, unsigned int id, u64 *val)
595 {
596 	unsigned long estat, gintc;
597 	struct loongarch_csrs *csr = vcpu->arch.csr;
598 
599 	if (get_gcsr_flag(id) & INVALID_GCSR)
600 		return -EINVAL;
601 
602 	if (id == LOONGARCH_CSR_ESTAT) {
603 		preempt_disable();
604 		vcpu_load(vcpu);
605 		/*
606 		 * Sync pending interrupts into ESTAT so that interrupt
607 		 * remains during VM migration stage
608 		 */
609 		kvm_deliver_intr(vcpu);
610 		vcpu->arch.aux_inuse &= ~KVM_LARCH_SWCSR_LATEST;
611 		vcpu_put(vcpu);
612 		preempt_enable();
613 
614 		/* ESTAT IP0~IP7 get from GINTC */
615 		gintc = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_GINTC) & KVM_GINTC_IRQ_MASK;
616 		estat = kvm_read_sw_gcsr(csr, LOONGARCH_CSR_ESTAT) & ~KVM_ESTAT_EXTI_MASK;
617 		*val = estat | (gintc << VIP_DELTA);
618 		return 0;
619 	}
620 
621 	/*
622 	 * Get software CSR state since software state is consistent
623 	 * with hardware for synchronous ioctl
624 	 */
625 	*val = kvm_read_sw_gcsr(csr, id);
626 
627 	return 0;
628 }
629 
630 static int _kvm_setcsr(struct kvm_vcpu *vcpu, unsigned int id, u64 val)
631 {
632 	int ret = 0;
633 	unsigned long estat, gintc;
634 	struct loongarch_csrs *csr = vcpu->arch.csr;
635 
636 	if (get_gcsr_flag(id) & INVALID_GCSR)
637 		return -EINVAL;
638 
639 	if (id == LOONGARCH_CSR_CPUID)
640 		return kvm_set_cpuid(vcpu, val);
641 
642 	if (id == LOONGARCH_CSR_ESTAT) {
643 		/* ESTAT IP0~IP7 inject through GINTC */
644 		gintc = (val >> VIP_DELTA) & KVM_GINTC_IRQ_MASK;
645 		gintc |= kvm_read_sw_gcsr(csr, LOONGARCH_CSR_GINTC) & ~KVM_GINTC_IRQ_MASK;
646 		kvm_write_sw_gcsr(csr, LOONGARCH_CSR_GINTC, gintc);
647 
648 		/* Only set valid ESTAT bits */
649 		estat = val & ~KVM_ESTAT_EXTI_MASK;
650 		estat &= CSR_ESTAT_IS | CSR_ESTAT_EXC | CSR_ESTAT_ESUBCODE;
651 		if (!kvm_guest_has_msgint(&vcpu->arch))
652 			estat &= ~CPU_AVEC;
653 		kvm_write_sw_gcsr(csr, LOONGARCH_CSR_ESTAT, estat);
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 	if (put_user(val, uaddr))
1110 		return -EFAULT;
1111 
1112 	return ret;
1113 }
1114 
1115 static int kvm_loongarch_pvtime_get_attr(struct kvm_vcpu *vcpu,
1116 					 struct kvm_device_attr *attr)
1117 {
1118 	u64 gpa;
1119 	u64 __user *user = (u64 __user *)attr->addr;
1120 
1121 	if (!kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_STEAL_TIME)
1122 			|| attr->attr != KVM_LOONGARCH_VCPU_PVTIME_GPA)
1123 		return -ENXIO;
1124 
1125 	gpa = vcpu->arch.st.guest_addr;
1126 	if (put_user(gpa, user))
1127 		return -EFAULT;
1128 
1129 	return 0;
1130 }
1131 
1132 static int kvm_loongarch_vcpu_get_attr(struct kvm_vcpu *vcpu,
1133 				       struct kvm_device_attr *attr)
1134 {
1135 	int ret = -ENXIO;
1136 
1137 	switch (attr->group) {
1138 	case KVM_LOONGARCH_VCPU_CPUCFG:
1139 		ret = kvm_loongarch_cpucfg_get_attr(vcpu, attr);
1140 		break;
1141 	case KVM_LOONGARCH_VCPU_PVTIME_CTRL:
1142 		ret = kvm_loongarch_pvtime_get_attr(vcpu, attr);
1143 		break;
1144 	default:
1145 		break;
1146 	}
1147 
1148 	return ret;
1149 }
1150 
1151 static int kvm_loongarch_cpucfg_set_attr(struct kvm_vcpu *vcpu,
1152 					 struct kvm_device_attr *attr)
1153 {
1154 	u64 val, valid;
1155 	u64 __user *user = (u64 __user *)attr->addr;
1156 	struct kvm *kvm = vcpu->kvm;
1157 
1158 	switch (attr->attr) {
1159 	case CPUCFG_KVM_FEATURE:
1160 		if (get_user(val, user))
1161 			return -EFAULT;
1162 
1163 		valid = LOONGARCH_PV_FEAT_MASK;
1164 		if (val & ~valid)
1165 			return -EINVAL;
1166 
1167 		/* All vCPUs need set the same PV features */
1168 		if ((kvm->arch.pv_features & LOONGARCH_PV_FEAT_UPDATED)
1169 				&& ((kvm->arch.pv_features & valid) != val))
1170 			return -EINVAL;
1171 		kvm->arch.pv_features = val | LOONGARCH_PV_FEAT_UPDATED;
1172 		return 0;
1173 	default:
1174 		return -ENXIO;
1175 	}
1176 }
1177 
1178 static int kvm_loongarch_pvtime_set_attr(struct kvm_vcpu *vcpu,
1179 					 struct kvm_device_attr *attr)
1180 {
1181 	int idx, ret = 0;
1182 	u64 gpa, __user *user = (u64 __user *)attr->addr;
1183 	struct kvm *kvm = vcpu->kvm;
1184 
1185 	if (!kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_STEAL_TIME)
1186 			|| attr->attr != KVM_LOONGARCH_VCPU_PVTIME_GPA)
1187 		return -ENXIO;
1188 
1189 	if (get_user(gpa, user))
1190 		return -EFAULT;
1191 
1192 	if (gpa & ~(KVM_STEAL_PHYS_MASK | KVM_STEAL_PHYS_VALID))
1193 		return -EINVAL;
1194 
1195 	if (!(gpa & KVM_STEAL_PHYS_VALID)) {
1196 		vcpu->arch.st.guest_addr = gpa;
1197 		return 0;
1198 	}
1199 
1200 	/* Check the address is in a valid memslot */
1201 	idx = srcu_read_lock(&kvm->srcu);
1202 	if (kvm_is_error_hva(gfn_to_hva(kvm, gpa >> PAGE_SHIFT)))
1203 		ret = -EINVAL;
1204 	srcu_read_unlock(&kvm->srcu, idx);
1205 
1206 	if (!ret) {
1207 		vcpu->arch.st.guest_addr = gpa;
1208 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
1209 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1210 	}
1211 
1212 	return ret;
1213 }
1214 
1215 static int kvm_loongarch_vcpu_set_attr(struct kvm_vcpu *vcpu,
1216 				       struct kvm_device_attr *attr)
1217 {
1218 	int ret = -ENXIO;
1219 
1220 	switch (attr->group) {
1221 	case KVM_LOONGARCH_VCPU_CPUCFG:
1222 		ret = kvm_loongarch_cpucfg_set_attr(vcpu, attr);
1223 		break;
1224 	case KVM_LOONGARCH_VCPU_PVTIME_CTRL:
1225 		ret = kvm_loongarch_pvtime_set_attr(vcpu, attr);
1226 		break;
1227 	default:
1228 		break;
1229 	}
1230 
1231 	return ret;
1232 }
1233 
1234 long kvm_arch_vcpu_ioctl(struct file *filp,
1235 			 unsigned int ioctl, unsigned long arg)
1236 {
1237 	long r;
1238 	struct kvm_device_attr attr;
1239 	void __user *argp = (void __user *)arg;
1240 	struct kvm_vcpu *vcpu = filp->private_data;
1241 
1242 	/*
1243 	 * Only software CSR should be modified
1244 	 *
1245 	 * If any hardware CSR register is modified, vcpu_load/vcpu_put pair
1246 	 * should be used. Since CSR registers owns by this vcpu, if switch
1247 	 * to other vcpus, other vcpus need reload CSR registers.
1248 	 *
1249 	 * If software CSR is modified, bit KVM_LARCH_HWCSR_USABLE should
1250 	 * be clear in vcpu->arch.aux_inuse, and vcpu_load will check
1251 	 * aux_inuse flag and reload CSR registers form software.
1252 	 */
1253 
1254 	switch (ioctl) {
1255 	case KVM_SET_ONE_REG:
1256 	case KVM_GET_ONE_REG: {
1257 		struct kvm_one_reg reg;
1258 
1259 		r = -EFAULT;
1260 		if (copy_from_user(&reg, argp, sizeof(reg)))
1261 			break;
1262 		if (ioctl == KVM_SET_ONE_REG) {
1263 			r = kvm_set_reg(vcpu, &reg);
1264 			vcpu->arch.aux_inuse &= ~KVM_LARCH_HWCSR_USABLE;
1265 		} else
1266 			r = kvm_get_reg(vcpu, &reg);
1267 		break;
1268 	}
1269 	case KVM_ENABLE_CAP: {
1270 		struct kvm_enable_cap cap;
1271 
1272 		r = -EFAULT;
1273 		if (copy_from_user(&cap, argp, sizeof(cap)))
1274 			break;
1275 		r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1276 		break;
1277 	}
1278 	case KVM_HAS_DEVICE_ATTR: {
1279 		r = -EFAULT;
1280 		if (copy_from_user(&attr, argp, sizeof(attr)))
1281 			break;
1282 		r = kvm_loongarch_vcpu_has_attr(vcpu, &attr);
1283 		break;
1284 	}
1285 	case KVM_GET_DEVICE_ATTR: {
1286 		r = -EFAULT;
1287 		if (copy_from_user(&attr, argp, sizeof(attr)))
1288 			break;
1289 		r = kvm_loongarch_vcpu_get_attr(vcpu, &attr);
1290 		break;
1291 	}
1292 	case KVM_SET_DEVICE_ATTR: {
1293 		r = -EFAULT;
1294 		if (copy_from_user(&attr, argp, sizeof(attr)))
1295 			break;
1296 		r = kvm_loongarch_vcpu_set_attr(vcpu, &attr);
1297 		break;
1298 	}
1299 	default:
1300 		r = -ENOIOCTLCMD;
1301 		break;
1302 	}
1303 
1304 	return r;
1305 }
1306 
1307 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1308 {
1309 	int i = 0;
1310 
1311 	fpu->fcc = vcpu->arch.fpu.fcc;
1312 	fpu->fcsr = vcpu->arch.fpu.fcsr;
1313 	for (i = 0; i < NUM_FPU_REGS; i++)
1314 		memcpy(&fpu->fpr[i], &vcpu->arch.fpu.fpr[i], sizeof(union fpureg));
1315 
1316 	return 0;
1317 }
1318 
1319 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1320 {
1321 	int i = 0;
1322 
1323 	vcpu->arch.fpu.fcc = fpu->fcc;
1324 	vcpu->arch.fpu.fcsr = fpu->fcsr;
1325 	for (i = 0; i < NUM_FPU_REGS; i++)
1326 		memcpy(&vcpu->arch.fpu.fpr[i], &fpu->fpr[i], sizeof(union fpureg));
1327 
1328 	return 0;
1329 }
1330 
1331 #ifdef CONFIG_CPU_HAS_LBT
1332 int kvm_own_lbt(struct kvm_vcpu *vcpu)
1333 {
1334 	if (!(vcpu->arch.aux_inuse & KVM_LARCH_LBT)) {
1335 		set_csr_euen(CSR_EUEN_LBTEN);
1336 		_restore_lbt(&vcpu->arch.lbt);
1337 		vcpu->arch.aux_inuse |= KVM_LARCH_LBT;
1338 	}
1339 
1340 	return 0;
1341 }
1342 
1343 static void kvm_lose_lbt(struct kvm_vcpu *vcpu)
1344 {
1345 	preempt_disable();
1346 	if (vcpu->arch.aux_inuse & KVM_LARCH_LBT) {
1347 		_save_lbt(&vcpu->arch.lbt);
1348 		clear_csr_euen(CSR_EUEN_LBTEN);
1349 		vcpu->arch.aux_inuse &= ~KVM_LARCH_LBT;
1350 	}
1351 	preempt_enable();
1352 }
1353 
1354 static void kvm_check_fcsr(struct kvm_vcpu *vcpu, unsigned long fcsr)
1355 {
1356 	/*
1357 	 * If TM is enabled, top register save/restore will
1358 	 * cause lbt exception, here enable lbt in advance
1359 	 */
1360 	if (fcsr & FPU_CSR_TM)
1361 		kvm_own_lbt(vcpu);
1362 }
1363 
1364 static void kvm_check_fcsr_alive(struct kvm_vcpu *vcpu)
1365 {
1366 	if (vcpu->arch.aux_inuse & KVM_LARCH_FPU) {
1367 		if (vcpu->arch.aux_inuse & KVM_LARCH_LBT)
1368 			return;
1369 		kvm_check_fcsr(vcpu, read_fcsr(LOONGARCH_FCSR0));
1370 	}
1371 }
1372 #else
1373 static inline void kvm_lose_lbt(struct kvm_vcpu *vcpu) { }
1374 static inline void kvm_check_fcsr(struct kvm_vcpu *vcpu, unsigned long fcsr) { }
1375 static inline void kvm_check_fcsr_alive(struct kvm_vcpu *vcpu) { }
1376 #endif
1377 
1378 /* Enable FPU and restore context */
1379 void kvm_own_fpu(struct kvm_vcpu *vcpu)
1380 {
1381 	/*
1382 	 * Enable FPU for guest
1383 	 * Set FR and FRE according to guest context
1384 	 */
1385 	kvm_check_fcsr(vcpu, vcpu->arch.fpu.fcsr);
1386 	set_csr_euen(CSR_EUEN_FPEN);
1387 
1388 	kvm_restore_fpu(&vcpu->arch.fpu);
1389 	vcpu->arch.aux_inuse |= KVM_LARCH_FPU;
1390 	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
1391 }
1392 
1393 #ifdef CONFIG_CPU_HAS_LSX
1394 /* Enable LSX and restore context */
1395 int kvm_own_lsx(struct kvm_vcpu *vcpu)
1396 {
1397 	/* Enable LSX for guest */
1398 	kvm_check_fcsr(vcpu, vcpu->arch.fpu.fcsr);
1399 	set_csr_euen(CSR_EUEN_LSXEN | CSR_EUEN_FPEN);
1400 
1401 	kvm_restore_lsx(&vcpu->arch.fpu);
1402 	vcpu->arch.aux_inuse |= KVM_LARCH_FPU;
1403 	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_LSX);
1404 
1405 	return 0;
1406 }
1407 #endif
1408 
1409 #ifdef CONFIG_CPU_HAS_LASX
1410 /* Enable LASX and restore context */
1411 int kvm_own_lasx(struct kvm_vcpu *vcpu)
1412 {
1413 	kvm_check_fcsr(vcpu, vcpu->arch.fpu.fcsr);
1414 	set_csr_euen(CSR_EUEN_FPEN | CSR_EUEN_LSXEN | CSR_EUEN_LASXEN);
1415 
1416 	kvm_restore_lasx(&vcpu->arch.fpu);
1417 	vcpu->arch.aux_inuse |= KVM_LARCH_FPU;
1418 	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_LASX);
1419 
1420 	return 0;
1421 }
1422 #endif
1423 
1424 /* Save context and disable FPU */
1425 void kvm_lose_fpu(struct kvm_vcpu *vcpu)
1426 {
1427 	preempt_disable();
1428 
1429 	if (!(vcpu->arch.aux_inuse & KVM_LARCH_FPU))
1430 		goto done;
1431 
1432 	kvm_check_fcsr_alive(vcpu);
1433 	if (kvm_guest_has_lasx(&vcpu->arch)) {
1434 		kvm_save_lasx(&vcpu->arch.fpu);
1435 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_LASX);
1436 
1437 		/* Disable LASX & LSX & FPU */
1438 		clear_csr_euen(CSR_EUEN_FPEN | CSR_EUEN_LSXEN | CSR_EUEN_LASXEN);
1439 	} else if (kvm_guest_has_lsx(&vcpu->arch)) {
1440 		kvm_save_lsx(&vcpu->arch.fpu);
1441 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_LSX);
1442 
1443 		/* Disable LSX & FPU */
1444 		clear_csr_euen(CSR_EUEN_FPEN | CSR_EUEN_LSXEN);
1445 	} else if (vcpu->arch.aux_inuse & KVM_LARCH_FPU) {
1446 		kvm_save_fpu(&vcpu->arch.fpu);
1447 		trace_kvm_aux(vcpu, KVM_TRACE_AUX_SAVE, KVM_TRACE_AUX_FPU);
1448 
1449 		/* Disable FPU */
1450 		clear_csr_euen(CSR_EUEN_FPEN);
1451 	}
1452 	vcpu->arch.aux_inuse &= ~KVM_LARCH_FPU;
1453 
1454 done:
1455 	kvm_lose_lbt(vcpu);
1456 
1457 	preempt_enable();
1458 }
1459 
1460 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq)
1461 {
1462 	int intr = (int)irq->irq;
1463 	unsigned int vector = abs(intr);
1464 
1465 	if (vector >= EXCCODE_INT_NUM)
1466 		return -EINVAL;
1467 
1468 	if (!kvm_guest_has_msgint(&vcpu->arch) && (vector == INT_AVEC))
1469 		return -EINVAL;
1470 
1471 	if (intr > 0)
1472 		kvm_queue_irq(vcpu, intr);
1473 	else if (intr < 0)
1474 		kvm_dequeue_irq(vcpu, -intr);
1475 	else {
1476 		kvm_err("%s: invalid interrupt ioctl %d\n", __func__, irq->irq);
1477 		return -EINVAL;
1478 	}
1479 
1480 	kvm_vcpu_kick(vcpu);
1481 
1482 	return 0;
1483 }
1484 
1485 long kvm_arch_vcpu_unlocked_ioctl(struct file *filp, unsigned int ioctl,
1486 				  unsigned long arg)
1487 {
1488 	void __user *argp = (void __user *)arg;
1489 	struct kvm_vcpu *vcpu = filp->private_data;
1490 
1491 	if (ioctl == KVM_INTERRUPT) {
1492 		struct kvm_interrupt irq;
1493 
1494 		if (copy_from_user(&irq, argp, sizeof(irq)))
1495 			return -EFAULT;
1496 
1497 		kvm_debug("[%d] %s: irq: %d\n", vcpu->vcpu_id, __func__, irq.irq);
1498 
1499 		return kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1500 	}
1501 
1502 	return -ENOIOCTLCMD;
1503 }
1504 
1505 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
1506 {
1507 	return 0;
1508 }
1509 
1510 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
1511 {
1512 	unsigned long timer_hz;
1513 	struct loongarch_csrs *csr;
1514 
1515 	vcpu->arch.vpid = 0;
1516 	vcpu->arch.flush_gpa = INVALID_GPA;
1517 
1518 	hrtimer_setup(&vcpu->arch.swtimer, kvm_swtimer_wakeup, CLOCK_MONOTONIC,
1519 		      HRTIMER_MODE_ABS_PINNED_HARD);
1520 
1521 	/* Get GPA (=HVA) of PGD for kvm hypervisor */
1522 	vcpu->arch.kvm_pgd = __pa(vcpu->kvm->arch.pgd);
1523 
1524 	/*
1525 	 * Get PGD for primary mmu, virtual address is used since there is
1526 	 * memory access after loading from CSR_PGD in tlb exception fast path.
1527 	 */
1528 	vcpu->arch.host_pgd = (unsigned long)vcpu->kvm->mm->pgd;
1529 
1530 	vcpu->arch.handle_exit = kvm_handle_exit;
1531 	vcpu->arch.guest_eentry = (unsigned long)kvm_loongarch_ops->exc_entry;
1532 	vcpu->arch.csr = kzalloc_obj(struct loongarch_csrs);
1533 	if (!vcpu->arch.csr)
1534 		return -ENOMEM;
1535 
1536 	/*
1537 	 * All kvm exceptions share one exception entry, and host <-> guest
1538 	 * switch also switch ECFG.VS field, keep host ECFG.VS info here.
1539 	 */
1540 	vcpu->arch.host_ecfg = (read_csr_ecfg() & CSR_ECFG_VS);
1541 
1542 	/* Init */
1543 	vcpu->arch.last_sched_cpu = -1;
1544 
1545 	/* Init ipi_state lock */
1546 	spin_lock_init(&vcpu->arch.ipi_state.lock);
1547 
1548 	/*
1549 	 * Initialize guest register state to valid architectural reset state.
1550 	 */
1551 	timer_hz = calc_const_freq();
1552 	kvm_init_timer(vcpu, timer_hz);
1553 
1554 	/* Set Initialize mode for guest */
1555 	csr = vcpu->arch.csr;
1556 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CRMD, CSR_CRMD_DA);
1557 
1558 	/* Set cpuid */
1559 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_TMID, vcpu->vcpu_id);
1560 	kvm_write_sw_gcsr(csr, LOONGARCH_CSR_CPUID, KVM_MAX_PHYID);
1561 
1562 	/* Start with no pending virtual guest interrupts */
1563 	csr->csrs[LOONGARCH_CSR_GINTC] = 0;
1564 
1565 	return 0;
1566 }
1567 
1568 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1569 {
1570 }
1571 
1572 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1573 {
1574 	int cpu;
1575 	struct kvm_context *context;
1576 
1577 	hrtimer_cancel(&vcpu->arch.swtimer);
1578 	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
1579 	kvm_drop_cpuid(vcpu);
1580 	kfree(vcpu->arch.csr);
1581 
1582 	/*
1583 	 * If the vCPU is freed and reused as another vCPU, we don't want the
1584 	 * matching pointer wrongly hanging around in last_vcpu.
1585 	 */
1586 	for_each_possible_cpu(cpu) {
1587 		context = per_cpu_ptr(vcpu->kvm->arch.vmcs, cpu);
1588 		if (context->last_vcpu == vcpu)
1589 			context->last_vcpu = NULL;
1590 	}
1591 }
1592 
1593 static int _kvm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1594 {
1595 	bool migrated;
1596 	struct kvm_context *context;
1597 	struct loongarch_csrs *csr = vcpu->arch.csr;
1598 
1599 	/*
1600 	 * Have we migrated to a different CPU?
1601 	 * If so, any old guest TLB state may be stale.
1602 	 */
1603 	migrated = (vcpu->arch.last_sched_cpu != cpu);
1604 
1605 	/*
1606 	 * Was this the last vCPU to run on this CPU?
1607 	 * If not, any old guest state from this vCPU will have been clobbered.
1608 	 */
1609 	context = per_cpu_ptr(vcpu->kvm->arch.vmcs, cpu);
1610 	if (migrated || (context->last_vcpu != vcpu)) {
1611 		context->last_vcpu = vcpu;
1612 		vcpu->arch.aux_inuse &= ~KVM_LARCH_HWCSR_USABLE;
1613 		vcpu->arch.host_eentry = csr_read64(LOONGARCH_CSR_EENTRY);
1614 	}
1615 
1616 	/* Restore timer state regardless */
1617 	kvm_restore_timer(vcpu);
1618 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1619 
1620 	/* Don't bother restoring registers multiple times unless necessary */
1621 	if (vcpu->arch.aux_inuse & KVM_LARCH_HWCSR_USABLE)
1622 		return 0;
1623 
1624 	write_csr_gcntc((ulong)vcpu->kvm->arch.time_offset);
1625 
1626 	/* Restore guest CSR registers */
1627 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_CRMD);
1628 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PRMD);
1629 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_EUEN);
1630 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_MISC);
1631 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ECFG);
1632 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ERA);
1633 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_BADV);
1634 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_BADI);
1635 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_EENTRY);
1636 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBIDX);
1637 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBEHI);
1638 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBELO0);
1639 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBELO1);
1640 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ASID);
1641 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PGDL);
1642 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PGDH);
1643 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PWCTL0);
1644 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_PWCTL1);
1645 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_STLBPGSIZE);
1646 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_RVACFG);
1647 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_CPUID);
1648 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS0);
1649 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS1);
1650 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS2);
1651 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS3);
1652 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS4);
1653 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS5);
1654 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS6);
1655 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_KS7);
1656 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TMID);
1657 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_CNTC);
1658 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRENTRY);
1659 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRBADV);
1660 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRERA);
1661 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRSAVE);
1662 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO0);
1663 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO1);
1664 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBREHI);
1665 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_TLBRPRMD);
1666 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN0);
1667 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN1);
1668 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN2);
1669 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_DMWIN3);
1670 	kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_LLBCTL);
1671 
1672 	if (kvm_guest_has_msgint(&vcpu->arch)) {
1673 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_IPR);
1674 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR0);
1675 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR1);
1676 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR2);
1677 		kvm_restore_hw_gcsr(csr, LOONGARCH_CSR_ISR3);
1678 	}
1679 
1680 	/* Restore Root.GINTC from unused Guest.GINTC register */
1681 	write_csr_gintc(csr->csrs[LOONGARCH_CSR_GINTC]);
1682 	write_csr_gstat(csr->csrs[LOONGARCH_CSR_GSTAT]);
1683 
1684 	/*
1685 	 * We should clear linked load bit to break interrupted atomics. This
1686 	 * prevents a SC on the next vCPU from succeeding by matching a LL on
1687 	 * the previous vCPU.
1688 	 */
1689 	if (vcpu->kvm->created_vcpus > 1)
1690 		set_gcsr_llbctl(CSR_LLBCTL_WCLLB);
1691 
1692 	vcpu->arch.aux_inuse |= KVM_LARCH_HWCSR_USABLE;
1693 
1694 	return 0;
1695 }
1696 
1697 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1698 {
1699 	unsigned long flags;
1700 
1701 	local_irq_save(flags);
1702 	/* Restore guest state to registers */
1703 	_kvm_vcpu_load(vcpu, cpu);
1704 	local_irq_restore(flags);
1705 }
1706 
1707 static int _kvm_vcpu_put(struct kvm_vcpu *vcpu, int cpu)
1708 {
1709 	struct loongarch_csrs *csr = vcpu->arch.csr;
1710 
1711 	kvm_lose_fpu(vcpu);
1712 
1713 	/*
1714 	 * Update CSR state from hardware if software CSR state is stale,
1715 	 * most CSR registers are kept unchanged during process context
1716 	 * switch except CSR registers like remaining timer tick value and
1717 	 * injected interrupt state.
1718 	 */
1719 	if (vcpu->arch.aux_inuse & KVM_LARCH_SWCSR_LATEST)
1720 		goto out;
1721 
1722 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_CRMD);
1723 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRMD);
1724 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_EUEN);
1725 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_MISC);
1726 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ECFG);
1727 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ERA);
1728 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_BADV);
1729 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_BADI);
1730 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_EENTRY);
1731 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBIDX);
1732 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBEHI);
1733 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBELO0);
1734 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBELO1);
1735 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ASID);
1736 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PGDL);
1737 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PGDH);
1738 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PWCTL0);
1739 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PWCTL1);
1740 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_STLBPGSIZE);
1741 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_RVACFG);
1742 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_CPUID);
1743 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRCFG1);
1744 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRCFG2);
1745 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_PRCFG3);
1746 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS0);
1747 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS1);
1748 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS2);
1749 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS3);
1750 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS4);
1751 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS5);
1752 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS6);
1753 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_KS7);
1754 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TMID);
1755 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_CNTC);
1756 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_LLBCTL);
1757 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRENTRY);
1758 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRBADV);
1759 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRERA);
1760 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRSAVE);
1761 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO0);
1762 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRELO1);
1763 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBREHI);
1764 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_TLBRPRMD);
1765 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN0);
1766 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN1);
1767 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN2);
1768 	kvm_save_hw_gcsr(csr, LOONGARCH_CSR_DMWIN3);
1769 
1770 	if (kvm_guest_has_msgint(&vcpu->arch)) {
1771 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_IPR);
1772 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR0);
1773 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR1);
1774 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR2);
1775 		kvm_save_hw_gcsr(csr, LOONGARCH_CSR_ISR3);
1776 	}
1777 
1778 	csr->csrs[LOONGARCH_CSR_GSTAT] = read_csr_gstat();
1779 	vcpu->arch.aux_inuse |= KVM_LARCH_SWCSR_LATEST;
1780 
1781 out:
1782 	kvm_save_timer(vcpu);
1783 	/* Save Root.GINTC into unused Guest.GINTC register */
1784 	csr->csrs[LOONGARCH_CSR_GINTC] = read_csr_gintc();
1785 
1786 	return 0;
1787 }
1788 
1789 static void kvm_vcpu_set_pv_preempted(struct kvm_vcpu *vcpu)
1790 {
1791 	gpa_t gpa;
1792 	struct gfn_to_hva_cache *ghc;
1793 	struct kvm_memslots *slots;
1794 	struct kvm_steal_time __user *st;
1795 
1796 	gpa = vcpu->arch.st.guest_addr;
1797 	if (!(gpa & KVM_STEAL_PHYS_VALID))
1798 		return;
1799 
1800 	/* vCPU may be preempted for many times */
1801 	if (vcpu->arch.st.preempted)
1802 		return;
1803 
1804 	/* This happens on process exit */
1805 	if (unlikely(current->mm != vcpu->kvm->mm))
1806 		return;
1807 
1808 	gpa &= KVM_STEAL_PHYS_MASK;
1809 	ghc = &vcpu->arch.st.cache;
1810 	slots = kvm_memslots(vcpu->kvm);
1811 	if (slots->generation != ghc->generation || gpa != ghc->gpa) {
1812 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st))) {
1813 			ghc->gpa = INVALID_GPA;
1814 			return;
1815 		}
1816 	}
1817 
1818 	st = (struct kvm_steal_time __user *)ghc->hva;
1819 	unsafe_put_user(KVM_VCPU_PREEMPTED, &st->preempted, out);
1820 	vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
1821 out:
1822 	mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
1823 }
1824 
1825 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1826 {
1827 	int cpu, idx;
1828 	unsigned long flags;
1829 
1830 	if (vcpu->preempted && kvm_guest_has_pv_feature(vcpu, KVM_FEATURE_PREEMPT)) {
1831 		/*
1832 		 * Take the srcu lock as memslots will be accessed to check
1833 		 * the gfn cache generation against the memslots generation.
1834 		 */
1835 		idx = srcu_read_lock(&vcpu->kvm->srcu);
1836 		kvm_vcpu_set_pv_preempted(vcpu);
1837 		srcu_read_unlock(&vcpu->kvm->srcu, idx);
1838 	}
1839 
1840 	local_irq_save(flags);
1841 	cpu = smp_processor_id();
1842 	vcpu->arch.last_sched_cpu = cpu;
1843 
1844 	/* Save guest state in registers */
1845 	_kvm_vcpu_put(vcpu, cpu);
1846 	local_irq_restore(flags);
1847 }
1848 
1849 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1850 {
1851 	int r = -EINTR;
1852 	struct kvm_run *run = vcpu->run;
1853 
1854 	if (vcpu->mmio_needed) {
1855 		if (!vcpu->mmio_is_write)
1856 			kvm_complete_mmio_read(vcpu, run);
1857 		vcpu->mmio_needed = 0;
1858 	}
1859 
1860 	switch (run->exit_reason) {
1861 	case KVM_EXIT_HYPERCALL:
1862 		kvm_complete_user_service(vcpu, run);
1863 		break;
1864 	case KVM_EXIT_LOONGARCH_IOCSR:
1865 		if (!run->iocsr_io.is_write)
1866 			kvm_complete_iocsr_read(vcpu, run);
1867 		break;
1868 	}
1869 
1870 	if (!vcpu->wants_to_run)
1871 		return r;
1872 
1873 	/* Clear exit_reason */
1874 	run->exit_reason = KVM_EXIT_UNKNOWN;
1875 	lose_fpu(1);
1876 	vcpu_load(vcpu);
1877 	kvm_sigset_activate(vcpu);
1878 	r = kvm_pre_enter_guest(vcpu);
1879 	if (r != RESUME_GUEST)
1880 		goto out;
1881 
1882 	guest_timing_enter_irqoff();
1883 	guest_state_enter_irqoff();
1884 	trace_kvm_enter(vcpu);
1885 	r = kvm_loongarch_ops->enter_guest(run, vcpu);
1886 
1887 	trace_kvm_out(vcpu);
1888 	/*
1889 	 * Guest exit is already recorded at kvm_handle_exit()
1890 	 * return value must not be RESUME_GUEST
1891 	 */
1892 	local_irq_enable();
1893 out:
1894 	kvm_sigset_deactivate(vcpu);
1895 	vcpu_put(vcpu);
1896 
1897 	return r;
1898 }
1899