xref: /linux/drivers/hv/mshv_root_main.c (revision 12ffd561d2de28825f39e15e8d22346d26b09688)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2024, Microsoft Corporation.
4  *
5  * The main part of the mshv_root module, providing APIs to create
6  * and manage guest partitions.
7  *
8  * Authors: Microsoft Linux virtualization team
9  */
10 
11 #include <linux/entry-virt.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/fs.h>
15 #include <linux/miscdevice.h>
16 #include <linux/slab.h>
17 #include <linux/file.h>
18 #include <linux/anon_inodes.h>
19 #include <linux/mm.h>
20 #include <linux/io.h>
21 #include <linux/cpuhotplug.h>
22 #include <linux/random.h>
23 #include <asm/mshyperv.h>
24 #include <linux/hyperv.h>
25 #include <linux/notifier.h>
26 #include <linux/reboot.h>
27 #include <linux/kexec.h>
28 #include <linux/page-flags.h>
29 #include <linux/crash_dump.h>
30 #include <linux/panic_notifier.h>
31 #include <linux/vmalloc.h>
32 #include <linux/rseq.h>
33 
34 #include "mshv_eventfd.h"
35 #include "mshv.h"
36 #include "mshv_root.h"
37 
38 MODULE_AUTHOR("Microsoft");
39 MODULE_LICENSE("GPL");
40 MODULE_DESCRIPTION("Microsoft Hyper-V root partition VMM interface /dev/mshv");
41 
42 /* TODO move this to another file when debugfs code is added */
43 enum hv_stats_vp_counters {			/* HV_THREAD_COUNTER */
44 #if defined(CONFIG_X86)
45 	VpRootDispatchThreadBlocked			= 202,
46 #elif defined(CONFIG_ARM64)
47 	VpRootDispatchThreadBlocked			= 94,
48 #endif
49 	VpStatsMaxCounter
50 };
51 
52 struct hv_stats_page {
53 	union {
54 		u64 vp_cntrs[VpStatsMaxCounter];		/* VP counters */
55 		u8 data[HV_HYP_PAGE_SIZE];
56 	};
57 } __packed;
58 
59 struct mshv_root mshv_root;
60 
61 enum hv_scheduler_type hv_scheduler_type;
62 
63 /* Once we implement the fast extended hypercall ABI they can go away. */
64 static void * __percpu *root_scheduler_input;
65 static void * __percpu *root_scheduler_output;
66 
67 static long mshv_dev_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg);
68 static int mshv_dev_open(struct inode *inode, struct file *filp);
69 static int mshv_dev_release(struct inode *inode, struct file *filp);
70 static int mshv_vp_release(struct inode *inode, struct file *filp);
71 static long mshv_vp_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg);
72 static int mshv_partition_release(struct inode *inode, struct file *filp);
73 static long mshv_partition_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg);
74 static int mshv_vp_mmap(struct file *file, struct vm_area_struct *vma);
75 static vm_fault_t mshv_vp_fault(struct vm_fault *vmf);
76 static int mshv_init_async_handler(struct mshv_partition *partition);
77 static void mshv_async_hvcall_handler(void *data, u64 *status);
78 
79 static const union hv_input_vtl input_vtl_zero;
80 static const union hv_input_vtl input_vtl_normal = {
81 	.target_vtl = HV_NORMAL_VTL,
82 	.use_target_vtl = 1,
83 };
84 
85 static const struct vm_operations_struct mshv_vp_vm_ops = {
86 	.fault = mshv_vp_fault,
87 };
88 
89 static const struct file_operations mshv_vp_fops = {
90 	.owner = THIS_MODULE,
91 	.release = mshv_vp_release,
92 	.unlocked_ioctl = mshv_vp_ioctl,
93 	.llseek = noop_llseek,
94 	.mmap = mshv_vp_mmap,
95 };
96 
97 static const struct file_operations mshv_partition_fops = {
98 	.owner = THIS_MODULE,
99 	.release = mshv_partition_release,
100 	.unlocked_ioctl = mshv_partition_ioctl,
101 	.llseek = noop_llseek,
102 };
103 
104 static const struct file_operations mshv_dev_fops = {
105 	.owner = THIS_MODULE,
106 	.open = mshv_dev_open,
107 	.release = mshv_dev_release,
108 	.unlocked_ioctl = mshv_dev_ioctl,
109 	.llseek = noop_llseek,
110 };
111 
112 static struct miscdevice mshv_dev = {
113 	.minor = MISC_DYNAMIC_MINOR,
114 	.name = "mshv",
115 	.fops = &mshv_dev_fops,
116 	.mode = 0600,
117 };
118 
119 /*
120  * Only allow hypercalls that have a u64 partition id as the first member of
121  * the input structure.
122  * These are sorted by value.
123  */
124 static u16 mshv_passthru_hvcalls[] = {
125 	HVCALL_GET_PARTITION_PROPERTY,
126 	HVCALL_GET_PARTITION_PROPERTY_EX,
127 	HVCALL_SET_PARTITION_PROPERTY,
128 	HVCALL_INSTALL_INTERCEPT,
129 	HVCALL_GET_VP_REGISTERS,
130 	HVCALL_SET_VP_REGISTERS,
131 	HVCALL_TRANSLATE_VIRTUAL_ADDRESS,
132 	HVCALL_CLEAR_VIRTUAL_INTERRUPT,
133 	HVCALL_REGISTER_INTERCEPT_RESULT,
134 	HVCALL_ASSERT_VIRTUAL_INTERRUPT,
135 	HVCALL_GET_GPA_PAGES_ACCESS_STATES,
136 	HVCALL_SIGNAL_EVENT_DIRECT,
137 	HVCALL_POST_MESSAGE_DIRECT,
138 	HVCALL_GET_VP_CPUID_VALUES,
139 };
140 
141 /*
142  * Only allow hypercalls that are safe to be called by the VMM with the host
143  * partition as target (i.e. HV_PARTITION_ID_SELF). Carefully audit that a
144  * hypercall cannot be misused by the VMM before adding it to this list.
145  */
146 static u16 mshv_self_passthru_hvcalls[] = {
147 	HVCALL_GET_PARTITION_PROPERTY,
148 	HVCALL_GET_PARTITION_PROPERTY_EX,
149 };
150 
151 static bool mshv_hvcall_is_async(u16 code)
152 {
153 	switch (code) {
154 	case HVCALL_SET_PARTITION_PROPERTY:
155 		return true;
156 	default:
157 		break;
158 	}
159 	return false;
160 }
161 
162 static bool mshv_passthru_hvcall_allowed(u16 code, u64 pt_id)
163 {
164 	int i;
165 	int n = ARRAY_SIZE(mshv_passthru_hvcalls);
166 	u16 *allowed_hvcalls = mshv_passthru_hvcalls;
167 
168 	if (pt_id == HV_PARTITION_ID_SELF) {
169 		n = ARRAY_SIZE(mshv_self_passthru_hvcalls);
170 		allowed_hvcalls = mshv_self_passthru_hvcalls;
171 	}
172 
173 	for (i = 0; i < n; ++i)
174 		if (allowed_hvcalls[i] == code)
175 			return true;
176 
177 	return false;
178 }
179 
180 static int mshv_ioctl_passthru_hvcall(struct mshv_partition *partition,
181 				      bool partition_locked,
182 				      void __user *user_args)
183 {
184 	u64 status;
185 	int ret = 0;
186 	bool is_async;
187 	struct mshv_root_hvcall args;
188 	struct page *page;
189 	unsigned int pages_order;
190 	void *input_pg = NULL;
191 	void *output_pg = NULL;
192 	u16 reps_completed;
193 	u64 pt_id = partition ? partition->pt_id : HV_PARTITION_ID_SELF;
194 
195 	if (copy_from_user(&args, user_args, sizeof(args)))
196 		return -EFAULT;
197 
198 	if (args.status || !args.in_ptr || args.in_sz < sizeof(u64) ||
199 	    mshv_field_nonzero(args, rsvd) || args.in_sz > HV_HYP_PAGE_SIZE)
200 		return -EINVAL;
201 
202 	if (args.out_ptr && (!args.out_sz || args.out_sz > HV_HYP_PAGE_SIZE))
203 		return -EINVAL;
204 
205 	if (!mshv_passthru_hvcall_allowed(args.code, pt_id))
206 		return -EINVAL;
207 
208 	is_async = mshv_hvcall_is_async(args.code);
209 	if (is_async) {
210 		/* async hypercalls can only be called from partition fd */
211 		if (!partition || !partition_locked)
212 			return -EINVAL;
213 		ret = mshv_init_async_handler(partition);
214 		if (ret)
215 			return ret;
216 	}
217 
218 	pages_order = args.out_ptr ? 1 : 0;
219 	page = alloc_pages(GFP_KERNEL, pages_order);
220 	if (!page)
221 		return -ENOMEM;
222 	input_pg = page_address(page);
223 
224 	if (args.out_ptr)
225 		output_pg = (char *)input_pg + PAGE_SIZE;
226 	else
227 		output_pg = NULL;
228 
229 	if (copy_from_user(input_pg, (void __user *)args.in_ptr,
230 			   args.in_sz)) {
231 		ret = -EFAULT;
232 		goto free_pages_out;
233 	}
234 
235 	/*
236 	 * NOTE: This only works because all the allowed hypercalls' input
237 	 * structs begin with a u64 partition_id field.
238 	 */
239 	*(u64 *)input_pg = pt_id;
240 
241 	reps_completed = 0;
242 	do {
243 		if (args.reps) {
244 			status = hv_do_rep_hypercall_ex(args.code, args.reps,
245 							0, reps_completed,
246 							input_pg, output_pg);
247 			reps_completed = hv_repcomp(status);
248 		} else {
249 			status = hv_do_hypercall(args.code, input_pg, output_pg);
250 		}
251 
252 		if (hv_result(status) == HV_STATUS_CALL_PENDING) {
253 			if (is_async) {
254 				mshv_async_hvcall_handler(partition, &status);
255 			} else { /* Paranoia check. This shouldn't happen! */
256 				ret = -EBADFD;
257 				goto free_pages_out;
258 			}
259 		}
260 
261 		if (hv_result_success(status))
262 			break;
263 
264 		if (hv_result(status) != HV_STATUS_INSUFFICIENT_MEMORY)
265 			ret = hv_result_to_errno(status);
266 		else
267 			ret = hv_call_deposit_pages(NUMA_NO_NODE,
268 						    pt_id, 1);
269 	} while (!ret);
270 
271 	args.status = hv_result(status);
272 	args.reps = reps_completed;
273 	if (copy_to_user(user_args, &args, sizeof(args)))
274 		ret = -EFAULT;
275 
276 	if (!ret && output_pg &&
277 	    copy_to_user((void __user *)args.out_ptr, output_pg, args.out_sz))
278 		ret = -EFAULT;
279 
280 free_pages_out:
281 	free_pages((unsigned long)input_pg, pages_order);
282 
283 	return ret;
284 }
285 
286 static inline bool is_ghcb_mapping_available(void)
287 {
288 #if IS_ENABLED(CONFIG_X86_64)
289 	return ms_hyperv.ext_features & HV_VP_GHCB_ROOT_MAPPING_AVAILABLE;
290 #else
291 	return 0;
292 #endif
293 }
294 
295 static int mshv_get_vp_registers(u32 vp_index, u64 partition_id, u16 count,
296 				 struct hv_register_assoc *registers)
297 {
298 	return hv_call_get_vp_registers(vp_index, partition_id,
299 					count, input_vtl_zero, registers);
300 }
301 
302 static int mshv_set_vp_registers(u32 vp_index, u64 partition_id, u16 count,
303 				 struct hv_register_assoc *registers)
304 {
305 	return hv_call_set_vp_registers(vp_index, partition_id,
306 					count, input_vtl_zero, registers);
307 }
308 
309 /*
310  * Explicit guest vCPU suspend is asynchronous by nature (as it is requested by
311  * dom0 vCPU for guest vCPU) and thus it can race with "intercept" suspend,
312  * done by the hypervisor.
313  * "Intercept" suspend leads to asynchronous message delivery to dom0 which
314  * should be awaited to keep the VP loop consistent (i.e. no message pending
315  * upon VP resume).
316  * VP intercept suspend can't be done when the VP is explicitly suspended
317  * already, and thus can be only two possible race scenarios:
318  *   1. implicit suspend bit set -> explicit suspend bit set -> message sent
319  *   2. implicit suspend bit set -> message sent -> explicit suspend bit set
320  * Checking for implicit suspend bit set after explicit suspend request has
321  * succeeded in either case allows us to reliably identify, if there is a
322  * message to receive and deliver to VMM.
323  */
324 static int
325 mshv_suspend_vp(const struct mshv_vp *vp, bool *message_in_flight)
326 {
327 	struct hv_register_assoc explicit_suspend = {
328 		.name = HV_REGISTER_EXPLICIT_SUSPEND
329 	};
330 	struct hv_register_assoc intercept_suspend = {
331 		.name = HV_REGISTER_INTERCEPT_SUSPEND
332 	};
333 	union hv_explicit_suspend_register *es =
334 		&explicit_suspend.value.explicit_suspend;
335 	union hv_intercept_suspend_register *is =
336 		&intercept_suspend.value.intercept_suspend;
337 	int ret;
338 
339 	es->suspended = 1;
340 
341 	ret = mshv_set_vp_registers(vp->vp_index, vp->vp_partition->pt_id,
342 				    1, &explicit_suspend);
343 	if (ret) {
344 		vp_err(vp, "Failed to explicitly suspend vCPU\n");
345 		return ret;
346 	}
347 
348 	ret = mshv_get_vp_registers(vp->vp_index, vp->vp_partition->pt_id,
349 				    1, &intercept_suspend);
350 	if (ret) {
351 		vp_err(vp, "Failed to get intercept suspend state\n");
352 		return ret;
353 	}
354 
355 	*message_in_flight = is->suspended;
356 
357 	return 0;
358 }
359 
360 /*
361  * This function is used when VPs are scheduled by the hypervisor's
362  * scheduler.
363  *
364  * Caller has to make sure the registers contain cleared
365  * HV_REGISTER_INTERCEPT_SUSPEND and HV_REGISTER_EXPLICIT_SUSPEND registers
366  * exactly in this order (the hypervisor clears them sequentially) to avoid
367  * potential invalid clearing a newly arrived HV_REGISTER_INTERCEPT_SUSPEND
368  * after VP is released from HV_REGISTER_EXPLICIT_SUSPEND in case of the
369  * opposite order.
370  */
371 static long mshv_run_vp_with_hyp_scheduler(struct mshv_vp *vp)
372 {
373 	long ret;
374 	struct hv_register_assoc suspend_regs[2] = {
375 			{ .name = HV_REGISTER_INTERCEPT_SUSPEND },
376 			{ .name = HV_REGISTER_EXPLICIT_SUSPEND }
377 	};
378 	size_t count = ARRAY_SIZE(suspend_regs);
379 
380 	/* Resume VP execution */
381 	ret = mshv_set_vp_registers(vp->vp_index, vp->vp_partition->pt_id,
382 				    count, suspend_regs);
383 	if (ret) {
384 		vp_err(vp, "Failed to resume vp execution. %lx\n", ret);
385 		return ret;
386 	}
387 
388 	ret = wait_event_interruptible(vp->run.vp_suspend_queue,
389 				       vp->run.kicked_by_hv == 1);
390 	if (ret) {
391 		bool message_in_flight;
392 
393 		/*
394 		 * Otherwise the waiting was interrupted by a signal: suspend
395 		 * the vCPU explicitly and copy message in flight (if any).
396 		 */
397 		ret = mshv_suspend_vp(vp, &message_in_flight);
398 		if (ret)
399 			return ret;
400 
401 		/* Return if no message in flight */
402 		if (!message_in_flight)
403 			return -EINTR;
404 
405 		/* Wait for the message in flight. */
406 		wait_event(vp->run.vp_suspend_queue, vp->run.kicked_by_hv == 1);
407 	}
408 
409 	/*
410 	 * Reset the flag to make the wait_event call above work
411 	 * next time.
412 	 */
413 	vp->run.kicked_by_hv = 0;
414 
415 	return 0;
416 }
417 
418 static int
419 mshv_vp_dispatch(struct mshv_vp *vp, u32 flags,
420 		 struct hv_output_dispatch_vp *res)
421 {
422 	struct hv_input_dispatch_vp *input;
423 	struct hv_output_dispatch_vp *output;
424 	u64 status;
425 
426 	preempt_disable();
427 	input = *this_cpu_ptr(root_scheduler_input);
428 	output = *this_cpu_ptr(root_scheduler_output);
429 
430 	memset(input, 0, sizeof(*input));
431 	memset(output, 0, sizeof(*output));
432 
433 	input->partition_id = vp->vp_partition->pt_id;
434 	input->vp_index = vp->vp_index;
435 	input->time_slice = 0; /* Run forever until something happens */
436 	input->spec_ctrl = 0; /* TODO: set sensible flags */
437 	input->flags = flags;
438 
439 	vp->run.flags.root_sched_dispatched = 1;
440 	status = hv_do_hypercall(HVCALL_DISPATCH_VP, input, output);
441 	vp->run.flags.root_sched_dispatched = 0;
442 
443 	*res = *output;
444 	preempt_enable();
445 
446 	if (!hv_result_success(status))
447 		vp_err(vp, "%s: status %s\n", __func__,
448 		       hv_result_to_string(status));
449 
450 	return hv_result_to_errno(status);
451 }
452 
453 static int
454 mshv_vp_clear_explicit_suspend(struct mshv_vp *vp)
455 {
456 	struct hv_register_assoc explicit_suspend = {
457 		.name = HV_REGISTER_EXPLICIT_SUSPEND,
458 		.value.explicit_suspend.suspended = 0,
459 	};
460 	int ret;
461 
462 	ret = mshv_set_vp_registers(vp->vp_index, vp->vp_partition->pt_id,
463 				    1, &explicit_suspend);
464 
465 	if (ret)
466 		vp_err(vp, "Failed to unsuspend\n");
467 
468 	return ret;
469 }
470 
471 #if IS_ENABLED(CONFIG_X86_64)
472 static u64 mshv_vp_interrupt_pending(struct mshv_vp *vp)
473 {
474 	if (!vp->vp_register_page)
475 		return 0;
476 	return vp->vp_register_page->interrupt_vectors.as_uint64;
477 }
478 #else
479 static u64 mshv_vp_interrupt_pending(struct mshv_vp *vp)
480 {
481 	return 0;
482 }
483 #endif
484 
485 static bool mshv_vp_dispatch_thread_blocked(struct mshv_vp *vp)
486 {
487 	struct hv_stats_page **stats = vp->vp_stats_pages;
488 	u64 *self_vp_cntrs = stats[HV_STATS_AREA_SELF]->vp_cntrs;
489 	u64 *parent_vp_cntrs = stats[HV_STATS_AREA_PARENT]->vp_cntrs;
490 
491 	if (self_vp_cntrs[VpRootDispatchThreadBlocked])
492 		return self_vp_cntrs[VpRootDispatchThreadBlocked];
493 	return parent_vp_cntrs[VpRootDispatchThreadBlocked];
494 }
495 
496 static int
497 mshv_vp_wait_for_hv_kick(struct mshv_vp *vp)
498 {
499 	int ret;
500 
501 	ret = wait_event_interruptible(vp->run.vp_suspend_queue,
502 				       (vp->run.kicked_by_hv == 1 &&
503 					!mshv_vp_dispatch_thread_blocked(vp)) ||
504 				       mshv_vp_interrupt_pending(vp));
505 	if (ret)
506 		return -EINTR;
507 
508 	vp->run.flags.root_sched_blocked = 0;
509 	vp->run.kicked_by_hv = 0;
510 
511 	return 0;
512 }
513 
514 /* Must be called with interrupts enabled */
515 static long mshv_run_vp_with_root_scheduler(struct mshv_vp *vp)
516 {
517 	long ret;
518 
519 	if (vp->run.flags.root_sched_blocked) {
520 		/*
521 		 * Dispatch state of this VP is blocked. Need to wait
522 		 * for the hypervisor to clear the blocked state before
523 		 * dispatching it.
524 		 */
525 		ret = mshv_vp_wait_for_hv_kick(vp);
526 		if (ret)
527 			return ret;
528 	}
529 
530 	do {
531 		u32 flags = 0;
532 		struct hv_output_dispatch_vp output;
533 
534 		if (__xfer_to_guest_mode_work_pending()) {
535 			ret = xfer_to_guest_mode_handle_work();
536 			if (ret)
537 				break;
538 		}
539 
540 		if (vp->run.flags.intercept_suspend)
541 			flags |= HV_DISPATCH_VP_FLAG_CLEAR_INTERCEPT_SUSPEND;
542 
543 		if (mshv_vp_interrupt_pending(vp))
544 			flags |= HV_DISPATCH_VP_FLAG_SCAN_INTERRUPT_INJECTION;
545 
546 		ret = mshv_vp_dispatch(vp, flags, &output);
547 		if (ret)
548 			break;
549 
550 		vp->run.flags.intercept_suspend = 0;
551 
552 		if (output.dispatch_state == HV_VP_DISPATCH_STATE_BLOCKED) {
553 			if (output.dispatch_event ==
554 						HV_VP_DISPATCH_EVENT_SUSPEND) {
555 				/*
556 				 * TODO: remove the warning once VP canceling
557 				 *	 is supported
558 				 */
559 				WARN_ONCE(atomic64_read(&vp->run.vp_signaled_count),
560 					  "%s: vp#%d: unexpected explicit suspend\n",
561 					  __func__, vp->vp_index);
562 				/*
563 				 * Need to clear explicit suspend before
564 				 * dispatching.
565 				 * Explicit suspend is either:
566 				 * - set right after the first VP dispatch or
567 				 * - set explicitly via hypercall
568 				 * Since the latter case is not yet supported,
569 				 * simply clear it here.
570 				 */
571 				ret = mshv_vp_clear_explicit_suspend(vp);
572 				if (ret)
573 					break;
574 
575 				ret = mshv_vp_wait_for_hv_kick(vp);
576 				if (ret)
577 					break;
578 			} else {
579 				vp->run.flags.root_sched_blocked = 1;
580 				ret = mshv_vp_wait_for_hv_kick(vp);
581 				if (ret)
582 					break;
583 			}
584 		} else {
585 			/* HV_VP_DISPATCH_STATE_READY */
586 			if (output.dispatch_event ==
587 						HV_VP_DISPATCH_EVENT_INTERCEPT)
588 				vp->run.flags.intercept_suspend = 1;
589 		}
590 	} while (!vp->run.flags.intercept_suspend);
591 
592 	rseq_virt_userspace_exit();
593 
594 	return ret;
595 }
596 
597 static_assert(sizeof(struct hv_message) <= MSHV_RUN_VP_BUF_SZ,
598 	      "sizeof(struct hv_message) must not exceed MSHV_RUN_VP_BUF_SZ");
599 
600 static struct mshv_mem_region *
601 mshv_partition_region_by_gfn(struct mshv_partition *partition, u64 gfn)
602 {
603 	struct mshv_mem_region *region;
604 
605 	hlist_for_each_entry(region, &partition->pt_mem_regions, hnode) {
606 		if (gfn >= region->start_gfn &&
607 		    gfn < region->start_gfn + region->nr_pages)
608 			return region;
609 	}
610 
611 	return NULL;
612 }
613 
614 static struct mshv_mem_region *
615 mshv_partition_region_by_gfn_get(struct mshv_partition *p, u64 gfn)
616 {
617 	struct mshv_mem_region *region;
618 
619 	spin_lock(&p->pt_mem_regions_lock);
620 	region = mshv_partition_region_by_gfn(p, gfn);
621 	if (!region || !mshv_region_get(region)) {
622 		spin_unlock(&p->pt_mem_regions_lock);
623 		return NULL;
624 	}
625 	spin_unlock(&p->pt_mem_regions_lock);
626 
627 	return region;
628 }
629 
630 /**
631  * mshv_handle_gpa_intercept - Handle GPA (Guest Physical Address) intercepts.
632  * @vp: Pointer to the virtual processor structure.
633  *
634  * This function processes GPA intercepts by identifying the memory region
635  * corresponding to the intercepted GPA, aligning the page offset, and
636  * mapping the required pages. It ensures that the region is valid and
637  * handles faults efficiently by mapping multiple pages at once.
638  *
639  * Return: true if the intercept was handled successfully, false otherwise.
640  */
641 static bool mshv_handle_gpa_intercept(struct mshv_vp *vp)
642 {
643 	struct mshv_partition *p = vp->vp_partition;
644 	struct mshv_mem_region *region;
645 	bool ret;
646 	u64 gfn;
647 #if defined(CONFIG_X86_64)
648 	struct hv_x64_memory_intercept_message *msg =
649 		(struct hv_x64_memory_intercept_message *)
650 		vp->vp_intercept_msg_page->u.payload;
651 #elif defined(CONFIG_ARM64)
652 	struct hv_arm64_memory_intercept_message *msg =
653 		(struct hv_arm64_memory_intercept_message *)
654 		vp->vp_intercept_msg_page->u.payload;
655 #endif
656 
657 	gfn = HVPFN_DOWN(msg->guest_physical_address);
658 
659 	region = mshv_partition_region_by_gfn_get(p, gfn);
660 	if (!region)
661 		return false;
662 
663 	/* Only movable memory ranges are supported for GPA intercepts */
664 	if (region->type == MSHV_REGION_TYPE_MEM_MOVABLE)
665 		ret = mshv_region_handle_gfn_fault(region, gfn);
666 	else
667 		ret = false;
668 
669 	mshv_region_put(region);
670 
671 	return ret;
672 }
673 
674 static bool mshv_vp_handle_intercept(struct mshv_vp *vp)
675 {
676 	switch (vp->vp_intercept_msg_page->header.message_type) {
677 	case HVMSG_GPA_INTERCEPT:
678 		return mshv_handle_gpa_intercept(vp);
679 	}
680 	return false;
681 }
682 
683 static long mshv_vp_ioctl_run_vp(struct mshv_vp *vp, void __user *ret_msg)
684 {
685 	long rc;
686 
687 	do {
688 		if (hv_scheduler_type == HV_SCHEDULER_TYPE_ROOT)
689 			rc = mshv_run_vp_with_root_scheduler(vp);
690 		else
691 			rc = mshv_run_vp_with_hyp_scheduler(vp);
692 	} while (rc == 0 && mshv_vp_handle_intercept(vp));
693 
694 	if (rc)
695 		return rc;
696 
697 	if (copy_to_user(ret_msg, vp->vp_intercept_msg_page,
698 			 sizeof(struct hv_message)))
699 		rc = -EFAULT;
700 
701 	return rc;
702 }
703 
704 static int
705 mshv_vp_ioctl_get_set_state_pfn(struct mshv_vp *vp,
706 				struct hv_vp_state_data state_data,
707 				unsigned long user_pfn, size_t page_count,
708 				bool is_set)
709 {
710 	int completed, ret = 0;
711 	unsigned long check;
712 	struct page **pages;
713 
714 	if (page_count > INT_MAX)
715 		return -EINVAL;
716 	/*
717 	 * Check the arithmetic for wraparound/overflow.
718 	 * The last page address in the buffer is:
719 	 * (user_pfn + (page_count - 1)) * PAGE_SIZE
720 	 */
721 	if (check_add_overflow(user_pfn, (page_count - 1), &check))
722 		return -EOVERFLOW;
723 	if (check_mul_overflow(check, PAGE_SIZE, &check))
724 		return -EOVERFLOW;
725 
726 	/* Pin user pages so hypervisor can copy directly to them */
727 	pages = kcalloc(page_count, sizeof(struct page *), GFP_KERNEL);
728 	if (!pages)
729 		return -ENOMEM;
730 
731 	for (completed = 0; completed < page_count; completed += ret) {
732 		unsigned long user_addr = (user_pfn + completed) * PAGE_SIZE;
733 		int remaining = page_count - completed;
734 
735 		ret = pin_user_pages_fast(user_addr, remaining, FOLL_WRITE,
736 					  &pages[completed]);
737 		if (ret < 0) {
738 			vp_err(vp, "%s: Failed to pin user pages error %i\n",
739 			       __func__, ret);
740 			goto unpin_pages;
741 		}
742 	}
743 
744 	if (is_set)
745 		ret = hv_call_set_vp_state(vp->vp_index,
746 					   vp->vp_partition->pt_id,
747 					   state_data, page_count, pages,
748 					   0, NULL);
749 	else
750 		ret = hv_call_get_vp_state(vp->vp_index,
751 					   vp->vp_partition->pt_id,
752 					   state_data, page_count, pages,
753 					   NULL);
754 
755 unpin_pages:
756 	unpin_user_pages(pages, completed);
757 	kfree(pages);
758 	return ret;
759 }
760 
761 static long
762 mshv_vp_ioctl_get_set_state(struct mshv_vp *vp,
763 			    struct mshv_get_set_vp_state __user *user_args,
764 			    bool is_set)
765 {
766 	struct mshv_get_set_vp_state args;
767 	long ret = 0;
768 	union hv_output_get_vp_state vp_state;
769 	u32 data_sz;
770 	struct hv_vp_state_data state_data = {};
771 
772 	if (copy_from_user(&args, user_args, sizeof(args)))
773 		return -EFAULT;
774 
775 	if (args.type >= MSHV_VP_STATE_COUNT || mshv_field_nonzero(args, rsvd) ||
776 	    !args.buf_sz || !PAGE_ALIGNED(args.buf_sz) ||
777 	    !PAGE_ALIGNED(args.buf_ptr))
778 		return -EINVAL;
779 
780 	if (!access_ok((void __user *)args.buf_ptr, args.buf_sz))
781 		return -EFAULT;
782 
783 	switch (args.type) {
784 	case MSHV_VP_STATE_LAPIC:
785 		state_data.type = HV_GET_SET_VP_STATE_LAPIC_STATE;
786 		data_sz = HV_HYP_PAGE_SIZE;
787 		break;
788 	case MSHV_VP_STATE_XSAVE:
789 	{
790 		u64 data_sz_64;
791 
792 		ret = hv_call_get_partition_property(vp->vp_partition->pt_id,
793 						     HV_PARTITION_PROPERTY_XSAVE_STATES,
794 						     &state_data.xsave.states.as_uint64);
795 		if (ret)
796 			return ret;
797 
798 		ret = hv_call_get_partition_property(vp->vp_partition->pt_id,
799 						     HV_PARTITION_PROPERTY_MAX_XSAVE_DATA_SIZE,
800 						     &data_sz_64);
801 		if (ret)
802 			return ret;
803 
804 		data_sz = (u32)data_sz_64;
805 		state_data.xsave.flags = 0;
806 		/* Always request legacy states */
807 		state_data.xsave.states.legacy_x87 = 1;
808 		state_data.xsave.states.legacy_sse = 1;
809 		state_data.type = HV_GET_SET_VP_STATE_XSAVE;
810 		break;
811 	}
812 	case MSHV_VP_STATE_SIMP:
813 		state_data.type = HV_GET_SET_VP_STATE_SIM_PAGE;
814 		data_sz = HV_HYP_PAGE_SIZE;
815 		break;
816 	case MSHV_VP_STATE_SIEFP:
817 		state_data.type = HV_GET_SET_VP_STATE_SIEF_PAGE;
818 		data_sz = HV_HYP_PAGE_SIZE;
819 		break;
820 	case MSHV_VP_STATE_SYNTHETIC_TIMERS:
821 		state_data.type = HV_GET_SET_VP_STATE_SYNTHETIC_TIMERS;
822 		data_sz = sizeof(vp_state.synthetic_timers_state);
823 		break;
824 	default:
825 		return -EINVAL;
826 	}
827 
828 	if (copy_to_user(&user_args->buf_sz, &data_sz, sizeof(user_args->buf_sz)))
829 		return -EFAULT;
830 
831 	if (data_sz > args.buf_sz)
832 		return -EINVAL;
833 
834 	/* If the data is transmitted via pfns, delegate to helper */
835 	if (state_data.type & HV_GET_SET_VP_STATE_TYPE_PFN) {
836 		unsigned long user_pfn = PFN_DOWN(args.buf_ptr);
837 		size_t page_count = PFN_DOWN(args.buf_sz);
838 
839 		return mshv_vp_ioctl_get_set_state_pfn(vp, state_data, user_pfn,
840 						       page_count, is_set);
841 	}
842 
843 	/* Paranoia check - this shouldn't happen! */
844 	if (data_sz > sizeof(vp_state)) {
845 		vp_err(vp, "Invalid vp state data size!\n");
846 		return -EINVAL;
847 	}
848 
849 	if (is_set) {
850 		if (copy_from_user(&vp_state, (__user void *)args.buf_ptr, data_sz))
851 			return -EFAULT;
852 
853 		return hv_call_set_vp_state(vp->vp_index,
854 					    vp->vp_partition->pt_id,
855 					    state_data, 0, NULL,
856 					    sizeof(vp_state), (u8 *)&vp_state);
857 	}
858 
859 	ret = hv_call_get_vp_state(vp->vp_index, vp->vp_partition->pt_id,
860 				   state_data, 0, NULL, &vp_state);
861 	if (ret)
862 		return ret;
863 
864 	if (copy_to_user((void __user *)args.buf_ptr, &vp_state, data_sz))
865 		return -EFAULT;
866 
867 	return 0;
868 }
869 
870 static long
871 mshv_vp_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
872 {
873 	struct mshv_vp *vp = filp->private_data;
874 	long r = -ENOTTY;
875 
876 	if (mutex_lock_killable(&vp->vp_mutex))
877 		return -EINTR;
878 
879 	switch (ioctl) {
880 	case MSHV_RUN_VP:
881 		r = mshv_vp_ioctl_run_vp(vp, (void __user *)arg);
882 		break;
883 	case MSHV_GET_VP_STATE:
884 		r = mshv_vp_ioctl_get_set_state(vp, (void __user *)arg, false);
885 		break;
886 	case MSHV_SET_VP_STATE:
887 		r = mshv_vp_ioctl_get_set_state(vp, (void __user *)arg, true);
888 		break;
889 	case MSHV_ROOT_HVCALL:
890 		r = mshv_ioctl_passthru_hvcall(vp->vp_partition, false,
891 					       (void __user *)arg);
892 		break;
893 	default:
894 		vp_warn(vp, "Invalid ioctl: %#x\n", ioctl);
895 		break;
896 	}
897 	mutex_unlock(&vp->vp_mutex);
898 
899 	return r;
900 }
901 
902 static vm_fault_t mshv_vp_fault(struct vm_fault *vmf)
903 {
904 	struct mshv_vp *vp = vmf->vma->vm_file->private_data;
905 
906 	switch (vmf->vma->vm_pgoff) {
907 	case MSHV_VP_MMAP_OFFSET_REGISTERS:
908 		vmf->page = virt_to_page(vp->vp_register_page);
909 		break;
910 	case MSHV_VP_MMAP_OFFSET_INTERCEPT_MESSAGE:
911 		vmf->page = virt_to_page(vp->vp_intercept_msg_page);
912 		break;
913 	case MSHV_VP_MMAP_OFFSET_GHCB:
914 		vmf->page = virt_to_page(vp->vp_ghcb_page);
915 		break;
916 	default:
917 		return VM_FAULT_SIGBUS;
918 	}
919 
920 	get_page(vmf->page);
921 
922 	return 0;
923 }
924 
925 static int mshv_vp_mmap(struct file *file, struct vm_area_struct *vma)
926 {
927 	struct mshv_vp *vp = file->private_data;
928 
929 	switch (vma->vm_pgoff) {
930 	case MSHV_VP_MMAP_OFFSET_REGISTERS:
931 		if (!vp->vp_register_page)
932 			return -ENODEV;
933 		break;
934 	case MSHV_VP_MMAP_OFFSET_INTERCEPT_MESSAGE:
935 		if (!vp->vp_intercept_msg_page)
936 			return -ENODEV;
937 		break;
938 	case MSHV_VP_MMAP_OFFSET_GHCB:
939 		if (!vp->vp_ghcb_page)
940 			return -ENODEV;
941 		break;
942 	default:
943 		return -EINVAL;
944 	}
945 
946 	vma->vm_ops = &mshv_vp_vm_ops;
947 	return 0;
948 }
949 
950 static int
951 mshv_vp_release(struct inode *inode, struct file *filp)
952 {
953 	struct mshv_vp *vp = filp->private_data;
954 
955 	/* Rest of VP cleanup happens in destroy_partition() */
956 	mshv_partition_put(vp->vp_partition);
957 	return 0;
958 }
959 
960 static void mshv_vp_stats_unmap(u64 partition_id, u32 vp_index,
961 				void *stats_pages[])
962 {
963 	union hv_stats_object_identity identity = {
964 		.vp.partition_id = partition_id,
965 		.vp.vp_index = vp_index,
966 	};
967 
968 	identity.vp.stats_area_type = HV_STATS_AREA_SELF;
969 	hv_unmap_stats_page(HV_STATS_OBJECT_VP, NULL, &identity);
970 
971 	identity.vp.stats_area_type = HV_STATS_AREA_PARENT;
972 	hv_unmap_stats_page(HV_STATS_OBJECT_VP, NULL, &identity);
973 }
974 
975 static int mshv_vp_stats_map(u64 partition_id, u32 vp_index,
976 			     void *stats_pages[])
977 {
978 	union hv_stats_object_identity identity = {
979 		.vp.partition_id = partition_id,
980 		.vp.vp_index = vp_index,
981 	};
982 	int err;
983 
984 	identity.vp.stats_area_type = HV_STATS_AREA_SELF;
985 	err = hv_map_stats_page(HV_STATS_OBJECT_VP, &identity,
986 				&stats_pages[HV_STATS_AREA_SELF]);
987 	if (err)
988 		return err;
989 
990 	identity.vp.stats_area_type = HV_STATS_AREA_PARENT;
991 	err = hv_map_stats_page(HV_STATS_OBJECT_VP, &identity,
992 				&stats_pages[HV_STATS_AREA_PARENT]);
993 	if (err)
994 		goto unmap_self;
995 
996 	return 0;
997 
998 unmap_self:
999 	identity.vp.stats_area_type = HV_STATS_AREA_SELF;
1000 	hv_unmap_stats_page(HV_STATS_OBJECT_VP, NULL, &identity);
1001 	return err;
1002 }
1003 
1004 static long
1005 mshv_partition_ioctl_create_vp(struct mshv_partition *partition,
1006 			       void __user *arg)
1007 {
1008 	struct mshv_create_vp args;
1009 	struct mshv_vp *vp;
1010 	struct page *intercept_msg_page, *register_page, *ghcb_page;
1011 	void *stats_pages[2];
1012 	long ret;
1013 
1014 	if (copy_from_user(&args, arg, sizeof(args)))
1015 		return -EFAULT;
1016 
1017 	if (args.vp_index >= MSHV_MAX_VPS)
1018 		return -EINVAL;
1019 
1020 	if (partition->pt_vp_array[args.vp_index])
1021 		return -EEXIST;
1022 
1023 	ret = hv_call_create_vp(NUMA_NO_NODE, partition->pt_id, args.vp_index,
1024 				0 /* Only valid for root partition VPs */);
1025 	if (ret)
1026 		return ret;
1027 
1028 	ret = hv_map_vp_state_page(partition->pt_id, args.vp_index,
1029 				   HV_VP_STATE_PAGE_INTERCEPT_MESSAGE,
1030 				   input_vtl_zero, &intercept_msg_page);
1031 	if (ret)
1032 		goto destroy_vp;
1033 
1034 	if (!mshv_partition_encrypted(partition)) {
1035 		ret = hv_map_vp_state_page(partition->pt_id, args.vp_index,
1036 					   HV_VP_STATE_PAGE_REGISTERS,
1037 					   input_vtl_zero, &register_page);
1038 		if (ret)
1039 			goto unmap_intercept_message_page;
1040 	}
1041 
1042 	if (mshv_partition_encrypted(partition) &&
1043 	    is_ghcb_mapping_available()) {
1044 		ret = hv_map_vp_state_page(partition->pt_id, args.vp_index,
1045 					   HV_VP_STATE_PAGE_GHCB,
1046 					   input_vtl_normal, &ghcb_page);
1047 		if (ret)
1048 			goto unmap_register_page;
1049 	}
1050 
1051 	/*
1052 	 * This mapping of the stats page is for detecting if dispatch thread
1053 	 * is blocked - only relevant for root scheduler
1054 	 */
1055 	if (hv_scheduler_type == HV_SCHEDULER_TYPE_ROOT) {
1056 		ret = mshv_vp_stats_map(partition->pt_id, args.vp_index,
1057 					stats_pages);
1058 		if (ret)
1059 			goto unmap_ghcb_page;
1060 	}
1061 
1062 	vp = kzalloc(sizeof(*vp), GFP_KERNEL);
1063 	if (!vp)
1064 		goto unmap_stats_pages;
1065 
1066 	vp->vp_partition = mshv_partition_get(partition);
1067 	if (!vp->vp_partition) {
1068 		ret = -EBADF;
1069 		goto free_vp;
1070 	}
1071 
1072 	mutex_init(&vp->vp_mutex);
1073 	init_waitqueue_head(&vp->run.vp_suspend_queue);
1074 	atomic64_set(&vp->run.vp_signaled_count, 0);
1075 
1076 	vp->vp_index = args.vp_index;
1077 	vp->vp_intercept_msg_page = page_to_virt(intercept_msg_page);
1078 	if (!mshv_partition_encrypted(partition))
1079 		vp->vp_register_page = page_to_virt(register_page);
1080 
1081 	if (mshv_partition_encrypted(partition) && is_ghcb_mapping_available())
1082 		vp->vp_ghcb_page = page_to_virt(ghcb_page);
1083 
1084 	if (hv_scheduler_type == HV_SCHEDULER_TYPE_ROOT)
1085 		memcpy(vp->vp_stats_pages, stats_pages, sizeof(stats_pages));
1086 
1087 	/*
1088 	 * Keep anon_inode_getfd last: it installs fd in the file struct and
1089 	 * thus makes the state accessible in user space.
1090 	 */
1091 	ret = anon_inode_getfd("mshv_vp", &mshv_vp_fops, vp,
1092 			       O_RDWR | O_CLOEXEC);
1093 	if (ret < 0)
1094 		goto put_partition;
1095 
1096 	/* already exclusive with the partition mutex for all ioctls */
1097 	partition->pt_vp_count++;
1098 	partition->pt_vp_array[args.vp_index] = vp;
1099 
1100 	return ret;
1101 
1102 put_partition:
1103 	mshv_partition_put(partition);
1104 free_vp:
1105 	kfree(vp);
1106 unmap_stats_pages:
1107 	if (hv_scheduler_type == HV_SCHEDULER_TYPE_ROOT)
1108 		mshv_vp_stats_unmap(partition->pt_id, args.vp_index, stats_pages);
1109 unmap_ghcb_page:
1110 	if (mshv_partition_encrypted(partition) && is_ghcb_mapping_available())
1111 		hv_unmap_vp_state_page(partition->pt_id, args.vp_index,
1112 				       HV_VP_STATE_PAGE_GHCB, ghcb_page,
1113 				       input_vtl_normal);
1114 unmap_register_page:
1115 	if (!mshv_partition_encrypted(partition))
1116 		hv_unmap_vp_state_page(partition->pt_id, args.vp_index,
1117 				       HV_VP_STATE_PAGE_REGISTERS,
1118 				       register_page, input_vtl_zero);
1119 unmap_intercept_message_page:
1120 	hv_unmap_vp_state_page(partition->pt_id, args.vp_index,
1121 			       HV_VP_STATE_PAGE_INTERCEPT_MESSAGE,
1122 			       intercept_msg_page, input_vtl_zero);
1123 destroy_vp:
1124 	hv_call_delete_vp(partition->pt_id, args.vp_index);
1125 	return ret;
1126 }
1127 
1128 static int mshv_init_async_handler(struct mshv_partition *partition)
1129 {
1130 	if (completion_done(&partition->async_hypercall)) {
1131 		pt_err(partition,
1132 		       "Cannot issue async hypercall while another one in progress!\n");
1133 		return -EPERM;
1134 	}
1135 
1136 	reinit_completion(&partition->async_hypercall);
1137 	return 0;
1138 }
1139 
1140 static void mshv_async_hvcall_handler(void *data, u64 *status)
1141 {
1142 	struct mshv_partition *partition = data;
1143 
1144 	wait_for_completion(&partition->async_hypercall);
1145 	pt_dbg(partition, "Async hypercall completed!\n");
1146 
1147 	*status = partition->async_hypercall_status;
1148 }
1149 
1150 /*
1151  * NB: caller checks and makes sure mem->size is page aligned
1152  * Returns: 0 with regionpp updated on success, or -errno
1153  */
1154 static int mshv_partition_create_region(struct mshv_partition *partition,
1155 					struct mshv_user_mem_region *mem,
1156 					struct mshv_mem_region **regionpp,
1157 					bool is_mmio)
1158 {
1159 	struct mshv_mem_region *rg;
1160 	u64 nr_pages = HVPFN_DOWN(mem->size);
1161 
1162 	/* Reject overlapping regions */
1163 	spin_lock(&partition->pt_mem_regions_lock);
1164 	hlist_for_each_entry(rg, &partition->pt_mem_regions, hnode) {
1165 		if (mem->guest_pfn + nr_pages <= rg->start_gfn ||
1166 		    rg->start_gfn + rg->nr_pages <= mem->guest_pfn)
1167 			continue;
1168 		spin_unlock(&partition->pt_mem_regions_lock);
1169 		return -EEXIST;
1170 	}
1171 	spin_unlock(&partition->pt_mem_regions_lock);
1172 
1173 	rg = mshv_region_create(mem->guest_pfn, nr_pages,
1174 				mem->userspace_addr, mem->flags);
1175 	if (IS_ERR(rg))
1176 		return PTR_ERR(rg);
1177 
1178 	if (is_mmio)
1179 		rg->type = MSHV_REGION_TYPE_MMIO;
1180 	else if (mshv_partition_encrypted(partition) ||
1181 		 !mshv_region_movable_init(rg))
1182 		rg->type = MSHV_REGION_TYPE_MEM_PINNED;
1183 	else
1184 		rg->type = MSHV_REGION_TYPE_MEM_MOVABLE;
1185 
1186 	rg->partition = partition;
1187 
1188 	*regionpp = rg;
1189 
1190 	return 0;
1191 }
1192 
1193 /**
1194  * mshv_prepare_pinned_region - Pin and map memory regions
1195  * @region: Pointer to the memory region structure
1196  *
1197  * This function processes memory regions that are explicitly marked as pinned.
1198  * Pinned regions are preallocated, mapped upfront, and do not rely on fault-based
1199  * population. The function ensures the region is properly populated, handles
1200  * encryption requirements for SNP partitions if applicable, maps the region,
1201  * and performs necessary sharing or eviction operations based on the mapping
1202  * result.
1203  *
1204  * Return: 0 on success, negative error code on failure.
1205  */
1206 static int mshv_prepare_pinned_region(struct mshv_mem_region *region)
1207 {
1208 	struct mshv_partition *partition = region->partition;
1209 	int ret;
1210 
1211 	ret = mshv_region_pin(region);
1212 	if (ret) {
1213 		pt_err(partition, "Failed to pin memory region: %d\n",
1214 		       ret);
1215 		goto err_out;
1216 	}
1217 
1218 	/*
1219 	 * For an SNP partition it is a requirement that for every memory region
1220 	 * that we are going to map for this partition we should make sure that
1221 	 * host access to that region is released. This is ensured by doing an
1222 	 * additional hypercall which will update the SLAT to release host
1223 	 * access to guest memory regions.
1224 	 */
1225 	if (mshv_partition_encrypted(partition)) {
1226 		ret = mshv_region_unshare(region);
1227 		if (ret) {
1228 			pt_err(partition,
1229 			       "Failed to unshare memory region (guest_pfn: %llu): %d\n",
1230 			       region->start_gfn, ret);
1231 			goto invalidate_region;
1232 		}
1233 	}
1234 
1235 	ret = mshv_region_map(region);
1236 	if (ret && mshv_partition_encrypted(partition)) {
1237 		int shrc;
1238 
1239 		shrc = mshv_region_share(region);
1240 		if (!shrc)
1241 			goto invalidate_region;
1242 
1243 		pt_err(partition,
1244 		       "Failed to share memory region (guest_pfn: %llu): %d\n",
1245 		       region->start_gfn, shrc);
1246 		/*
1247 		 * Don't unpin if marking shared failed because pages are no
1248 		 * longer mapped in the host, ie root, anymore.
1249 		 */
1250 		goto err_out;
1251 	}
1252 
1253 	return 0;
1254 
1255 invalidate_region:
1256 	mshv_region_invalidate(region);
1257 err_out:
1258 	return ret;
1259 }
1260 
1261 /*
1262  * This maps two things: guest RAM and for pci passthru mmio space.
1263  *
1264  * mmio:
1265  *  - vfio overloads vm_pgoff to store the mmio start pfn/spa.
1266  *  - Two things need to happen for mapping mmio range:
1267  *	1. mapped in the uaddr so VMM can access it.
1268  *	2. mapped in the hwpt (gfn <-> mmio phys addr) so guest can access it.
1269  *
1270  *   This function takes care of the second. The first one is managed by vfio,
1271  *   and hence is taken care of via vfio_pci_mmap_fault().
1272  */
1273 static long
1274 mshv_map_user_memory(struct mshv_partition *partition,
1275 		     struct mshv_user_mem_region mem)
1276 {
1277 	struct mshv_mem_region *region;
1278 	struct vm_area_struct *vma;
1279 	bool is_mmio;
1280 	ulong mmio_pfn;
1281 	long ret;
1282 
1283 	if (mem.flags & BIT(MSHV_SET_MEM_BIT_UNMAP) ||
1284 	    !access_ok((const void __user *)mem.userspace_addr, mem.size))
1285 		return -EINVAL;
1286 
1287 	mmap_read_lock(current->mm);
1288 	vma = vma_lookup(current->mm, mem.userspace_addr);
1289 	is_mmio = vma ? !!(vma->vm_flags & (VM_IO | VM_PFNMAP)) : 0;
1290 	mmio_pfn = is_mmio ? vma->vm_pgoff : 0;
1291 	mmap_read_unlock(current->mm);
1292 
1293 	if (!vma)
1294 		return -EINVAL;
1295 
1296 	ret = mshv_partition_create_region(partition, &mem, &region,
1297 					   is_mmio);
1298 	if (ret)
1299 		return ret;
1300 
1301 	switch (region->type) {
1302 	case MSHV_REGION_TYPE_MEM_PINNED:
1303 		ret = mshv_prepare_pinned_region(region);
1304 		break;
1305 	case MSHV_REGION_TYPE_MEM_MOVABLE:
1306 		/*
1307 		 * For movable memory regions, remap with no access to let
1308 		 * the hypervisor track dirty pages, enabling pre-copy live
1309 		 * migration.
1310 		 */
1311 		ret = hv_call_map_gpa_pages(partition->pt_id,
1312 					    region->start_gfn,
1313 					    region->nr_pages,
1314 					    HV_MAP_GPA_NO_ACCESS, NULL);
1315 		break;
1316 	case MSHV_REGION_TYPE_MMIO:
1317 		ret = hv_call_map_mmio_pages(partition->pt_id,
1318 					     region->start_gfn,
1319 					     mmio_pfn,
1320 					     region->nr_pages);
1321 		break;
1322 	}
1323 
1324 	if (ret)
1325 		goto errout;
1326 
1327 	spin_lock(&partition->pt_mem_regions_lock);
1328 	hlist_add_head(&region->hnode, &partition->pt_mem_regions);
1329 	spin_unlock(&partition->pt_mem_regions_lock);
1330 
1331 	return 0;
1332 
1333 errout:
1334 	vfree(region);
1335 	return ret;
1336 }
1337 
1338 /* Called for unmapping both the guest ram and the mmio space */
1339 static long
1340 mshv_unmap_user_memory(struct mshv_partition *partition,
1341 		       struct mshv_user_mem_region mem)
1342 {
1343 	struct mshv_mem_region *region;
1344 
1345 	if (!(mem.flags & BIT(MSHV_SET_MEM_BIT_UNMAP)))
1346 		return -EINVAL;
1347 
1348 	spin_lock(&partition->pt_mem_regions_lock);
1349 
1350 	region = mshv_partition_region_by_gfn(partition, mem.guest_pfn);
1351 	if (!region) {
1352 		spin_unlock(&partition->pt_mem_regions_lock);
1353 		return -ENOENT;
1354 	}
1355 
1356 	/* Paranoia check */
1357 	if (region->start_uaddr != mem.userspace_addr ||
1358 	    region->start_gfn != mem.guest_pfn ||
1359 	    region->nr_pages != HVPFN_DOWN(mem.size)) {
1360 		spin_unlock(&partition->pt_mem_regions_lock);
1361 		return -EINVAL;
1362 	}
1363 
1364 	hlist_del(&region->hnode);
1365 
1366 	spin_unlock(&partition->pt_mem_regions_lock);
1367 
1368 	mshv_region_put(region);
1369 
1370 	return 0;
1371 }
1372 
1373 static long
1374 mshv_partition_ioctl_set_memory(struct mshv_partition *partition,
1375 				struct mshv_user_mem_region __user *user_mem)
1376 {
1377 	struct mshv_user_mem_region mem;
1378 
1379 	if (copy_from_user(&mem, user_mem, sizeof(mem)))
1380 		return -EFAULT;
1381 
1382 	if (!mem.size ||
1383 	    !PAGE_ALIGNED(mem.size) ||
1384 	    !PAGE_ALIGNED(mem.userspace_addr) ||
1385 	    (mem.flags & ~MSHV_SET_MEM_FLAGS_MASK) ||
1386 	    mshv_field_nonzero(mem, rsvd))
1387 		return -EINVAL;
1388 
1389 	if (mem.flags & BIT(MSHV_SET_MEM_BIT_UNMAP))
1390 		return mshv_unmap_user_memory(partition, mem);
1391 
1392 	return mshv_map_user_memory(partition, mem);
1393 }
1394 
1395 static long
1396 mshv_partition_ioctl_ioeventfd(struct mshv_partition *partition,
1397 			       void __user *user_args)
1398 {
1399 	struct mshv_user_ioeventfd args;
1400 
1401 	if (copy_from_user(&args, user_args, sizeof(args)))
1402 		return -EFAULT;
1403 
1404 	return mshv_set_unset_ioeventfd(partition, &args);
1405 }
1406 
1407 static long
1408 mshv_partition_ioctl_irqfd(struct mshv_partition *partition,
1409 			   void __user *user_args)
1410 {
1411 	struct mshv_user_irqfd args;
1412 
1413 	if (copy_from_user(&args, user_args, sizeof(args)))
1414 		return -EFAULT;
1415 
1416 	return mshv_set_unset_irqfd(partition, &args);
1417 }
1418 
1419 static long
1420 mshv_partition_ioctl_get_gpap_access_bitmap(struct mshv_partition *partition,
1421 					    void __user *user_args)
1422 {
1423 	struct mshv_gpap_access_bitmap args;
1424 	union hv_gpa_page_access_state *states;
1425 	long ret, i;
1426 	union hv_gpa_page_access_state_flags hv_flags = {};
1427 	u8 hv_type_mask;
1428 	ulong bitmap_buf_sz, states_buf_sz;
1429 	int written = 0;
1430 
1431 	if (copy_from_user(&args, user_args, sizeof(args)))
1432 		return -EFAULT;
1433 
1434 	if (args.access_type >= MSHV_GPAP_ACCESS_TYPE_COUNT ||
1435 	    args.access_op >= MSHV_GPAP_ACCESS_OP_COUNT ||
1436 	    mshv_field_nonzero(args, rsvd) || !args.page_count ||
1437 	    !args.bitmap_ptr)
1438 		return -EINVAL;
1439 
1440 	if (check_mul_overflow(args.page_count, sizeof(*states), &states_buf_sz))
1441 		return -E2BIG;
1442 
1443 	/* Num bytes needed to store bitmap; one bit per page rounded up */
1444 	bitmap_buf_sz = DIV_ROUND_UP(args.page_count, 8);
1445 
1446 	/* Sanity check */
1447 	if (bitmap_buf_sz > states_buf_sz)
1448 		return -EBADFD;
1449 
1450 	switch (args.access_type) {
1451 	case MSHV_GPAP_ACCESS_TYPE_ACCESSED:
1452 		hv_type_mask = 1;
1453 		if (args.access_op == MSHV_GPAP_ACCESS_OP_CLEAR) {
1454 			hv_flags.clear_accessed = 1;
1455 			/* not accessed implies not dirty */
1456 			hv_flags.clear_dirty = 1;
1457 		} else { /* MSHV_GPAP_ACCESS_OP_SET */
1458 			hv_flags.set_accessed = 1;
1459 		}
1460 		break;
1461 	case MSHV_GPAP_ACCESS_TYPE_DIRTY:
1462 		hv_type_mask = 2;
1463 		if (args.access_op == MSHV_GPAP_ACCESS_OP_CLEAR) {
1464 			hv_flags.clear_dirty = 1;
1465 		} else { /* MSHV_GPAP_ACCESS_OP_SET */
1466 			hv_flags.set_dirty = 1;
1467 			/* dirty implies accessed */
1468 			hv_flags.set_accessed = 1;
1469 		}
1470 		break;
1471 	}
1472 
1473 	states = vzalloc(states_buf_sz);
1474 	if (!states)
1475 		return -ENOMEM;
1476 
1477 	ret = hv_call_get_gpa_access_states(partition->pt_id, args.page_count,
1478 					    args.gpap_base, hv_flags, &written,
1479 					    states);
1480 	if (ret)
1481 		goto free_return;
1482 
1483 	/*
1484 	 * Overwrite states buffer with bitmap - the bits in hv_type_mask
1485 	 * correspond to bitfields in hv_gpa_page_access_state
1486 	 */
1487 	for (i = 0; i < written; ++i)
1488 		__assign_bit(i, (ulong *)states,
1489 			     states[i].as_uint8 & hv_type_mask);
1490 
1491 	/* zero the unused bits in the last byte(s) of the returned bitmap */
1492 	for (i = written; i < bitmap_buf_sz * 8; ++i)
1493 		__clear_bit(i, (ulong *)states);
1494 
1495 	if (copy_to_user((void __user *)args.bitmap_ptr, states, bitmap_buf_sz))
1496 		ret = -EFAULT;
1497 
1498 free_return:
1499 	vfree(states);
1500 	return ret;
1501 }
1502 
1503 static long
1504 mshv_partition_ioctl_set_msi_routing(struct mshv_partition *partition,
1505 				     void __user *user_args)
1506 {
1507 	struct mshv_user_irq_entry *entries = NULL;
1508 	struct mshv_user_irq_table args;
1509 	long ret;
1510 
1511 	if (copy_from_user(&args, user_args, sizeof(args)))
1512 		return -EFAULT;
1513 
1514 	if (args.nr > MSHV_MAX_GUEST_IRQS ||
1515 	    mshv_field_nonzero(args, rsvd))
1516 		return -EINVAL;
1517 
1518 	if (args.nr) {
1519 		struct mshv_user_irq_table __user *urouting = user_args;
1520 
1521 		entries = vmemdup_user(urouting->entries,
1522 				       array_size(sizeof(*entries),
1523 						  args.nr));
1524 		if (IS_ERR(entries))
1525 			return PTR_ERR(entries);
1526 	}
1527 	ret = mshv_update_routing_table(partition, entries, args.nr);
1528 	kvfree(entries);
1529 
1530 	return ret;
1531 }
1532 
1533 static long
1534 mshv_partition_ioctl_initialize(struct mshv_partition *partition)
1535 {
1536 	long ret;
1537 
1538 	if (partition->pt_initialized)
1539 		return 0;
1540 
1541 	ret = hv_call_initialize_partition(partition->pt_id);
1542 	if (ret)
1543 		goto withdraw_mem;
1544 
1545 	partition->pt_initialized = true;
1546 
1547 	return 0;
1548 
1549 withdraw_mem:
1550 	hv_call_withdraw_memory(U64_MAX, NUMA_NO_NODE, partition->pt_id);
1551 
1552 	return ret;
1553 }
1554 
1555 static long
1556 mshv_partition_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1557 {
1558 	struct mshv_partition *partition = filp->private_data;
1559 	long ret;
1560 	void __user *uarg = (void __user *)arg;
1561 
1562 	if (mutex_lock_killable(&partition->pt_mutex))
1563 		return -EINTR;
1564 
1565 	switch (ioctl) {
1566 	case MSHV_INITIALIZE_PARTITION:
1567 		ret = mshv_partition_ioctl_initialize(partition);
1568 		break;
1569 	case MSHV_SET_GUEST_MEMORY:
1570 		ret = mshv_partition_ioctl_set_memory(partition, uarg);
1571 		break;
1572 	case MSHV_CREATE_VP:
1573 		ret = mshv_partition_ioctl_create_vp(partition, uarg);
1574 		break;
1575 	case MSHV_IRQFD:
1576 		ret = mshv_partition_ioctl_irqfd(partition, uarg);
1577 		break;
1578 	case MSHV_IOEVENTFD:
1579 		ret = mshv_partition_ioctl_ioeventfd(partition, uarg);
1580 		break;
1581 	case MSHV_SET_MSI_ROUTING:
1582 		ret = mshv_partition_ioctl_set_msi_routing(partition, uarg);
1583 		break;
1584 	case MSHV_GET_GPAP_ACCESS_BITMAP:
1585 		ret = mshv_partition_ioctl_get_gpap_access_bitmap(partition,
1586 								  uarg);
1587 		break;
1588 	case MSHV_ROOT_HVCALL:
1589 		ret = mshv_ioctl_passthru_hvcall(partition, true, uarg);
1590 		break;
1591 	default:
1592 		ret = -ENOTTY;
1593 	}
1594 
1595 	mutex_unlock(&partition->pt_mutex);
1596 	return ret;
1597 }
1598 
1599 static int
1600 disable_vp_dispatch(struct mshv_vp *vp)
1601 {
1602 	int ret;
1603 	struct hv_register_assoc dispatch_suspend = {
1604 		.name = HV_REGISTER_DISPATCH_SUSPEND,
1605 		.value.dispatch_suspend.suspended = 1,
1606 	};
1607 
1608 	ret = mshv_set_vp_registers(vp->vp_index, vp->vp_partition->pt_id,
1609 				    1, &dispatch_suspend);
1610 	if (ret)
1611 		vp_err(vp, "failed to suspend\n");
1612 
1613 	return ret;
1614 }
1615 
1616 static int
1617 get_vp_signaled_count(struct mshv_vp *vp, u64 *count)
1618 {
1619 	int ret;
1620 	struct hv_register_assoc root_signal_count = {
1621 		.name = HV_REGISTER_VP_ROOT_SIGNAL_COUNT,
1622 	};
1623 
1624 	ret = mshv_get_vp_registers(vp->vp_index, vp->vp_partition->pt_id,
1625 				    1, &root_signal_count);
1626 
1627 	if (ret) {
1628 		vp_err(vp, "Failed to get root signal count");
1629 		*count = 0;
1630 		return ret;
1631 	}
1632 
1633 	*count = root_signal_count.value.reg64;
1634 
1635 	return ret;
1636 }
1637 
1638 static void
1639 drain_vp_signals(struct mshv_vp *vp)
1640 {
1641 	u64 hv_signal_count;
1642 	u64 vp_signal_count;
1643 
1644 	get_vp_signaled_count(vp, &hv_signal_count);
1645 
1646 	vp_signal_count = atomic64_read(&vp->run.vp_signaled_count);
1647 
1648 	/*
1649 	 * There should be at most 1 outstanding notification, but be extra
1650 	 * careful anyway.
1651 	 */
1652 	while (hv_signal_count != vp_signal_count) {
1653 		WARN_ON(hv_signal_count - vp_signal_count != 1);
1654 
1655 		if (wait_event_interruptible(vp->run.vp_suspend_queue,
1656 					     vp->run.kicked_by_hv == 1))
1657 			break;
1658 		vp->run.kicked_by_hv = 0;
1659 		vp_signal_count = atomic64_read(&vp->run.vp_signaled_count);
1660 	}
1661 }
1662 
1663 static void drain_all_vps(const struct mshv_partition *partition)
1664 {
1665 	int i;
1666 	struct mshv_vp *vp;
1667 
1668 	/*
1669 	 * VPs are reachable from ISR. It is safe to not take the partition
1670 	 * lock because nobody else can enter this function and drop the
1671 	 * partition from the list.
1672 	 */
1673 	for (i = 0; i < MSHV_MAX_VPS; i++) {
1674 		vp = partition->pt_vp_array[i];
1675 		if (!vp)
1676 			continue;
1677 		/*
1678 		 * Disable dispatching of the VP in the hypervisor. After this
1679 		 * the hypervisor guarantees it won't generate any signals for
1680 		 * the VP and the hypervisor's VP signal count won't change.
1681 		 */
1682 		disable_vp_dispatch(vp);
1683 		drain_vp_signals(vp);
1684 	}
1685 }
1686 
1687 static void
1688 remove_partition(struct mshv_partition *partition)
1689 {
1690 	spin_lock(&mshv_root.pt_ht_lock);
1691 	hlist_del_rcu(&partition->pt_hnode);
1692 	spin_unlock(&mshv_root.pt_ht_lock);
1693 
1694 	synchronize_rcu();
1695 }
1696 
1697 /*
1698  * Tear down a partition and remove it from the list.
1699  * Partition's refcount must be 0
1700  */
1701 static void destroy_partition(struct mshv_partition *partition)
1702 {
1703 	struct mshv_vp *vp;
1704 	struct mshv_mem_region *region;
1705 	struct hlist_node *n;
1706 	int i;
1707 
1708 	if (refcount_read(&partition->pt_ref_count)) {
1709 		pt_err(partition,
1710 		       "Attempt to destroy partition but refcount > 0\n");
1711 		return;
1712 	}
1713 
1714 	if (partition->pt_initialized) {
1715 		/*
1716 		 * We only need to drain signals for root scheduler. This should be
1717 		 * done before removing the partition from the partition list.
1718 		 */
1719 		if (hv_scheduler_type == HV_SCHEDULER_TYPE_ROOT)
1720 			drain_all_vps(partition);
1721 
1722 		/* Remove vps */
1723 		for (i = 0; i < MSHV_MAX_VPS; ++i) {
1724 			vp = partition->pt_vp_array[i];
1725 			if (!vp)
1726 				continue;
1727 
1728 			if (hv_scheduler_type == HV_SCHEDULER_TYPE_ROOT)
1729 				mshv_vp_stats_unmap(partition->pt_id, vp->vp_index,
1730 						    (void **)vp->vp_stats_pages);
1731 
1732 			if (vp->vp_register_page) {
1733 				(void)hv_unmap_vp_state_page(partition->pt_id,
1734 							     vp->vp_index,
1735 							     HV_VP_STATE_PAGE_REGISTERS,
1736 							     virt_to_page(vp->vp_register_page),
1737 							     input_vtl_zero);
1738 				vp->vp_register_page = NULL;
1739 			}
1740 
1741 			(void)hv_unmap_vp_state_page(partition->pt_id,
1742 						     vp->vp_index,
1743 						     HV_VP_STATE_PAGE_INTERCEPT_MESSAGE,
1744 						     virt_to_page(vp->vp_intercept_msg_page),
1745 						     input_vtl_zero);
1746 			vp->vp_intercept_msg_page = NULL;
1747 
1748 			if (vp->vp_ghcb_page) {
1749 				(void)hv_unmap_vp_state_page(partition->pt_id,
1750 							     vp->vp_index,
1751 							     HV_VP_STATE_PAGE_GHCB,
1752 							     virt_to_page(vp->vp_ghcb_page),
1753 							     input_vtl_normal);
1754 				vp->vp_ghcb_page = NULL;
1755 			}
1756 
1757 			kfree(vp);
1758 
1759 			partition->pt_vp_array[i] = NULL;
1760 		}
1761 
1762 		/* Deallocates and unmaps everything including vcpus, GPA mappings etc */
1763 		hv_call_finalize_partition(partition->pt_id);
1764 
1765 		partition->pt_initialized = false;
1766 	}
1767 
1768 	remove_partition(partition);
1769 
1770 	hlist_for_each_entry_safe(region, n, &partition->pt_mem_regions,
1771 				  hnode) {
1772 		hlist_del(&region->hnode);
1773 		mshv_region_put(region);
1774 	}
1775 
1776 	/* Withdraw and free all pages we deposited */
1777 	hv_call_withdraw_memory(U64_MAX, NUMA_NO_NODE, partition->pt_id);
1778 	hv_call_delete_partition(partition->pt_id);
1779 
1780 	mshv_free_routing_table(partition);
1781 	kfree(partition);
1782 }
1783 
1784 struct
1785 mshv_partition *mshv_partition_get(struct mshv_partition *partition)
1786 {
1787 	if (refcount_inc_not_zero(&partition->pt_ref_count))
1788 		return partition;
1789 	return NULL;
1790 }
1791 
1792 struct
1793 mshv_partition *mshv_partition_find(u64 partition_id)
1794 	__must_hold(RCU)
1795 {
1796 	struct mshv_partition *p;
1797 
1798 	hash_for_each_possible_rcu(mshv_root.pt_htable, p, pt_hnode,
1799 				   partition_id)
1800 		if (p->pt_id == partition_id)
1801 			return p;
1802 
1803 	return NULL;
1804 }
1805 
1806 void
1807 mshv_partition_put(struct mshv_partition *partition)
1808 {
1809 	if (refcount_dec_and_test(&partition->pt_ref_count))
1810 		destroy_partition(partition);
1811 }
1812 
1813 static int
1814 mshv_partition_release(struct inode *inode, struct file *filp)
1815 {
1816 	struct mshv_partition *partition = filp->private_data;
1817 
1818 	mshv_eventfd_release(partition);
1819 
1820 	cleanup_srcu_struct(&partition->pt_irq_srcu);
1821 
1822 	mshv_partition_put(partition);
1823 
1824 	return 0;
1825 }
1826 
1827 static int
1828 add_partition(struct mshv_partition *partition)
1829 {
1830 	spin_lock(&mshv_root.pt_ht_lock);
1831 
1832 	hash_add_rcu(mshv_root.pt_htable, &partition->pt_hnode,
1833 		     partition->pt_id);
1834 
1835 	spin_unlock(&mshv_root.pt_ht_lock);
1836 
1837 	return 0;
1838 }
1839 
1840 static_assert(MSHV_NUM_CPU_FEATURES_BANKS ==
1841 	      HV_PARTITION_PROCESSOR_FEATURES_BANKS);
1842 
1843 static long mshv_ioctl_process_pt_flags(void __user *user_arg, u64 *pt_flags,
1844 					struct hv_partition_creation_properties *cr_props,
1845 					union hv_partition_isolation_properties *isol_props)
1846 {
1847 	int i;
1848 	struct mshv_create_partition_v2 args;
1849 	union hv_partition_processor_features *disabled_procs;
1850 	union hv_partition_processor_xsave_features *disabled_xsave;
1851 
1852 	/* First, copy v1 struct in case user is on previous versions */
1853 	if (copy_from_user(&args, user_arg,
1854 			   sizeof(struct mshv_create_partition)))
1855 		return -EFAULT;
1856 
1857 	if ((args.pt_flags & ~MSHV_PT_FLAGS_MASK) ||
1858 	    args.pt_isolation >= MSHV_PT_ISOLATION_COUNT)
1859 		return -EINVAL;
1860 
1861 	disabled_procs = &cr_props->disabled_processor_features;
1862 	disabled_xsave = &cr_props->disabled_processor_xsave_features;
1863 
1864 	/* Check if user provided newer struct with feature fields */
1865 	if (args.pt_flags & BIT_ULL(MSHV_PT_BIT_CPU_AND_XSAVE_FEATURES)) {
1866 		if (copy_from_user(&args, user_arg, sizeof(args)))
1867 			return -EFAULT;
1868 
1869 		/* Re-validate v1 fields after second copy_from_user() */
1870 		if ((args.pt_flags & ~MSHV_PT_FLAGS_MASK) ||
1871 		    args.pt_isolation >= MSHV_PT_ISOLATION_COUNT)
1872 			return -EINVAL;
1873 
1874 		if (args.pt_num_cpu_fbanks != MSHV_NUM_CPU_FEATURES_BANKS ||
1875 		    mshv_field_nonzero(args, pt_rsvd) ||
1876 		    mshv_field_nonzero(args, pt_rsvd1))
1877 			return -EINVAL;
1878 
1879 		/*
1880 		 * Note this assumes MSHV_NUM_CPU_FEATURES_BANKS will never
1881 		 * change and equals HV_PARTITION_PROCESSOR_FEATURES_BANKS
1882 		 * (i.e. 2).
1883 		 *
1884 		 * Further banks (index >= 2) will be modifiable as 'early'
1885 		 * properties via the set partition property hypercall.
1886 		 */
1887 		for (i = 0; i < HV_PARTITION_PROCESSOR_FEATURES_BANKS; i++)
1888 			disabled_procs->as_uint64[i] = args.pt_cpu_fbanks[i];
1889 
1890 #if IS_ENABLED(CONFIG_X86_64)
1891 		disabled_xsave->as_uint64 = args.pt_disabled_xsave;
1892 #else
1893 		/*
1894 		 * In practice this field is ignored on arm64, but safer to
1895 		 * zero it in case it is ever used.
1896 		 */
1897 		disabled_xsave->as_uint64 = 0;
1898 
1899 		if (mshv_field_nonzero(args, pt_rsvd2))
1900 			return -EINVAL;
1901 #endif
1902 	} else {
1903 		/*
1904 		 * v1 behavior: try to enable everything. The hypervisor will
1905 		 * disable features that are not supported. The banks can be
1906 		 * queried via the get partition property hypercall.
1907 		 */
1908 		for (i = 0; i < HV_PARTITION_PROCESSOR_FEATURES_BANKS; i++)
1909 			disabled_procs->as_uint64[i] = 0;
1910 
1911 		disabled_xsave->as_uint64 = 0;
1912 	}
1913 
1914 	/* Only support EXO partitions */
1915 	*pt_flags = HV_PARTITION_CREATION_FLAG_EXO_PARTITION |
1916 		    HV_PARTITION_CREATION_FLAG_INTERCEPT_MESSAGE_PAGE_ENABLED;
1917 
1918 	if (args.pt_flags & BIT_ULL(MSHV_PT_BIT_LAPIC))
1919 		*pt_flags |= HV_PARTITION_CREATION_FLAG_LAPIC_ENABLED;
1920 	if (args.pt_flags & BIT_ULL(MSHV_PT_BIT_X2APIC))
1921 		*pt_flags |= HV_PARTITION_CREATION_FLAG_X2APIC_CAPABLE;
1922 	if (args.pt_flags & BIT_ULL(MSHV_PT_BIT_GPA_SUPER_PAGES))
1923 		*pt_flags |= HV_PARTITION_CREATION_FLAG_GPA_SUPER_PAGES_ENABLED;
1924 
1925 	isol_props->as_uint64 = 0;
1926 
1927 	switch (args.pt_isolation) {
1928 	case MSHV_PT_ISOLATION_NONE:
1929 		isol_props->isolation_type = HV_PARTITION_ISOLATION_TYPE_NONE;
1930 		break;
1931 	}
1932 
1933 	return 0;
1934 }
1935 
1936 static long
1937 mshv_ioctl_create_partition(void __user *user_arg, struct device *module_dev)
1938 {
1939 	u64 creation_flags;
1940 	struct hv_partition_creation_properties creation_properties;
1941 	union hv_partition_isolation_properties isolation_properties;
1942 	struct mshv_partition *partition;
1943 	long ret;
1944 
1945 	ret = mshv_ioctl_process_pt_flags(user_arg, &creation_flags,
1946 					  &creation_properties,
1947 					  &isolation_properties);
1948 	if (ret)
1949 		return ret;
1950 
1951 	partition = kzalloc(sizeof(*partition), GFP_KERNEL);
1952 	if (!partition)
1953 		return -ENOMEM;
1954 
1955 	partition->pt_module_dev = module_dev;
1956 	partition->isolation_type = isolation_properties.isolation_type;
1957 
1958 	refcount_set(&partition->pt_ref_count, 1);
1959 
1960 	mutex_init(&partition->pt_mutex);
1961 
1962 	mutex_init(&partition->pt_irq_lock);
1963 
1964 	init_completion(&partition->async_hypercall);
1965 
1966 	INIT_HLIST_HEAD(&partition->irq_ack_notifier_list);
1967 
1968 	INIT_HLIST_HEAD(&partition->pt_devices);
1969 
1970 	spin_lock_init(&partition->pt_mem_regions_lock);
1971 	INIT_HLIST_HEAD(&partition->pt_mem_regions);
1972 
1973 	mshv_eventfd_init(partition);
1974 
1975 	ret = init_srcu_struct(&partition->pt_irq_srcu);
1976 	if (ret)
1977 		goto free_partition;
1978 
1979 	ret = hv_call_create_partition(creation_flags,
1980 				       creation_properties,
1981 				       isolation_properties,
1982 				       &partition->pt_id);
1983 	if (ret)
1984 		goto cleanup_irq_srcu;
1985 
1986 	ret = add_partition(partition);
1987 	if (ret)
1988 		goto delete_partition;
1989 
1990 	ret = mshv_init_async_handler(partition);
1991 	if (!ret) {
1992 		ret = FD_ADD(O_CLOEXEC, anon_inode_getfile("mshv_partition",
1993 							   &mshv_partition_fops,
1994 							   partition, O_RDWR));
1995 		if (ret >= 0)
1996 			return ret;
1997 	}
1998 	remove_partition(partition);
1999 delete_partition:
2000 	hv_call_delete_partition(partition->pt_id);
2001 cleanup_irq_srcu:
2002 	cleanup_srcu_struct(&partition->pt_irq_srcu);
2003 free_partition:
2004 	kfree(partition);
2005 
2006 	return ret;
2007 }
2008 
2009 static long mshv_dev_ioctl(struct file *filp, unsigned int ioctl,
2010 			   unsigned long arg)
2011 {
2012 	struct miscdevice *misc = filp->private_data;
2013 
2014 	switch (ioctl) {
2015 	case MSHV_CREATE_PARTITION:
2016 		return mshv_ioctl_create_partition((void __user *)arg,
2017 						misc->this_device);
2018 	case MSHV_ROOT_HVCALL:
2019 		return mshv_ioctl_passthru_hvcall(NULL, false,
2020 					(void __user *)arg);
2021 	}
2022 
2023 	return -ENOTTY;
2024 }
2025 
2026 static int
2027 mshv_dev_open(struct inode *inode, struct file *filp)
2028 {
2029 	return 0;
2030 }
2031 
2032 static int
2033 mshv_dev_release(struct inode *inode, struct file *filp)
2034 {
2035 	return 0;
2036 }
2037 
2038 static int mshv_cpuhp_online;
2039 static int mshv_root_sched_online;
2040 
2041 static const char *scheduler_type_to_string(enum hv_scheduler_type type)
2042 {
2043 	switch (type) {
2044 	case HV_SCHEDULER_TYPE_LP:
2045 		return "classic scheduler without SMT";
2046 	case HV_SCHEDULER_TYPE_LP_SMT:
2047 		return "classic scheduler with SMT";
2048 	case HV_SCHEDULER_TYPE_CORE_SMT:
2049 		return "core scheduler";
2050 	case HV_SCHEDULER_TYPE_ROOT:
2051 		return "root scheduler";
2052 	default:
2053 		return "unknown scheduler";
2054 	};
2055 }
2056 
2057 /* TODO move this to hv_common.c when needed outside */
2058 static int __init hv_retrieve_scheduler_type(enum hv_scheduler_type *out)
2059 {
2060 	struct hv_input_get_system_property *input;
2061 	struct hv_output_get_system_property *output;
2062 	unsigned long flags;
2063 	u64 status;
2064 
2065 	local_irq_save(flags);
2066 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
2067 	output = *this_cpu_ptr(hyperv_pcpu_output_arg);
2068 
2069 	memset(input, 0, sizeof(*input));
2070 	memset(output, 0, sizeof(*output));
2071 	input->property_id = HV_SYSTEM_PROPERTY_SCHEDULER_TYPE;
2072 
2073 	status = hv_do_hypercall(HVCALL_GET_SYSTEM_PROPERTY, input, output);
2074 	if (!hv_result_success(status)) {
2075 		local_irq_restore(flags);
2076 		pr_err("%s: %s\n", __func__, hv_result_to_string(status));
2077 		return hv_result_to_errno(status);
2078 	}
2079 
2080 	*out = output->scheduler_type;
2081 	local_irq_restore(flags);
2082 
2083 	return 0;
2084 }
2085 
2086 /* Retrieve and stash the supported scheduler type */
2087 static int __init mshv_retrieve_scheduler_type(struct device *dev)
2088 {
2089 	int ret = 0;
2090 
2091 	if (hv_l1vh_partition())
2092 		hv_scheduler_type = HV_SCHEDULER_TYPE_CORE_SMT;
2093 	else
2094 		ret = hv_retrieve_scheduler_type(&hv_scheduler_type);
2095 
2096 	if (ret)
2097 		return ret;
2098 
2099 	dev_info(dev, "Hypervisor using %s\n",
2100 		 scheduler_type_to_string(hv_scheduler_type));
2101 
2102 	switch (hv_scheduler_type) {
2103 	case HV_SCHEDULER_TYPE_CORE_SMT:
2104 	case HV_SCHEDULER_TYPE_LP_SMT:
2105 	case HV_SCHEDULER_TYPE_ROOT:
2106 	case HV_SCHEDULER_TYPE_LP:
2107 		/* Supported scheduler, nothing to do */
2108 		break;
2109 	default:
2110 		dev_err(dev, "unsupported scheduler 0x%x, bailing.\n",
2111 			hv_scheduler_type);
2112 		return -EOPNOTSUPP;
2113 	}
2114 
2115 	return 0;
2116 }
2117 
2118 static int mshv_root_scheduler_init(unsigned int cpu)
2119 {
2120 	void **inputarg, **outputarg, *p;
2121 
2122 	inputarg = (void **)this_cpu_ptr(root_scheduler_input);
2123 	outputarg = (void **)this_cpu_ptr(root_scheduler_output);
2124 
2125 	/* Allocate two consecutive pages. One for input, one for output. */
2126 	p = kmalloc(2 * HV_HYP_PAGE_SIZE, GFP_KERNEL);
2127 	if (!p)
2128 		return -ENOMEM;
2129 
2130 	*inputarg = p;
2131 	*outputarg = (char *)p + HV_HYP_PAGE_SIZE;
2132 
2133 	return 0;
2134 }
2135 
2136 static int mshv_root_scheduler_cleanup(unsigned int cpu)
2137 {
2138 	void *p, **inputarg, **outputarg;
2139 
2140 	inputarg = (void **)this_cpu_ptr(root_scheduler_input);
2141 	outputarg = (void **)this_cpu_ptr(root_scheduler_output);
2142 
2143 	p = *inputarg;
2144 
2145 	*inputarg = NULL;
2146 	*outputarg = NULL;
2147 
2148 	kfree(p);
2149 
2150 	return 0;
2151 }
2152 
2153 /* Must be called after retrieving the scheduler type */
2154 static int
2155 root_scheduler_init(struct device *dev)
2156 {
2157 	int ret;
2158 
2159 	if (hv_scheduler_type != HV_SCHEDULER_TYPE_ROOT)
2160 		return 0;
2161 
2162 	root_scheduler_input = alloc_percpu(void *);
2163 	root_scheduler_output = alloc_percpu(void *);
2164 
2165 	if (!root_scheduler_input || !root_scheduler_output) {
2166 		dev_err(dev, "Failed to allocate root scheduler buffers\n");
2167 		ret = -ENOMEM;
2168 		goto out;
2169 	}
2170 
2171 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "mshv_root_sched",
2172 				mshv_root_scheduler_init,
2173 				mshv_root_scheduler_cleanup);
2174 
2175 	if (ret < 0) {
2176 		dev_err(dev, "Failed to setup root scheduler state: %i\n", ret);
2177 		goto out;
2178 	}
2179 
2180 	mshv_root_sched_online = ret;
2181 
2182 	return 0;
2183 
2184 out:
2185 	free_percpu(root_scheduler_input);
2186 	free_percpu(root_scheduler_output);
2187 	return ret;
2188 }
2189 
2190 static void
2191 root_scheduler_deinit(void)
2192 {
2193 	if (hv_scheduler_type != HV_SCHEDULER_TYPE_ROOT)
2194 		return;
2195 
2196 	cpuhp_remove_state(mshv_root_sched_online);
2197 	free_percpu(root_scheduler_input);
2198 	free_percpu(root_scheduler_output);
2199 }
2200 
2201 static int mshv_reboot_notify(struct notifier_block *nb,
2202 			      unsigned long code, void *unused)
2203 {
2204 	cpuhp_remove_state(mshv_cpuhp_online);
2205 	return 0;
2206 }
2207 
2208 struct notifier_block mshv_reboot_nb = {
2209 	.notifier_call = mshv_reboot_notify,
2210 };
2211 
2212 static void mshv_root_partition_exit(void)
2213 {
2214 	unregister_reboot_notifier(&mshv_reboot_nb);
2215 	root_scheduler_deinit();
2216 }
2217 
2218 static int __init mshv_root_partition_init(struct device *dev)
2219 {
2220 	int err;
2221 
2222 	err = root_scheduler_init(dev);
2223 	if (err)
2224 		return err;
2225 
2226 	err = register_reboot_notifier(&mshv_reboot_nb);
2227 	if (err)
2228 		goto root_sched_deinit;
2229 
2230 	return 0;
2231 
2232 root_sched_deinit:
2233 	root_scheduler_deinit();
2234 	return err;
2235 }
2236 
2237 static void mshv_init_vmm_caps(struct device *dev)
2238 {
2239 	/*
2240 	 * This can only fail here if HVCALL_GET_PARTITION_PROPERTY_EX or
2241 	 * HV_PARTITION_PROPERTY_VMM_CAPABILITIES are not supported. In that
2242 	 * case it's valid to proceed as if all vmm_caps are disabled (zero).
2243 	 */
2244 	if (hv_call_get_partition_property_ex(HV_PARTITION_ID_SELF,
2245 					      HV_PARTITION_PROPERTY_VMM_CAPABILITIES,
2246 					      0, &mshv_root.vmm_caps,
2247 					      sizeof(mshv_root.vmm_caps)))
2248 		dev_warn(dev, "Unable to get VMM capabilities\n");
2249 
2250 	dev_dbg(dev, "vmm_caps = %#llx\n", mshv_root.vmm_caps.as_uint64[0]);
2251 }
2252 
2253 static int __init mshv_parent_partition_init(void)
2254 {
2255 	int ret;
2256 	struct device *dev;
2257 	union hv_hypervisor_version_info version_info;
2258 
2259 	if (!hv_parent_partition() || is_kdump_kernel())
2260 		return -ENODEV;
2261 
2262 	if (hv_get_hypervisor_version(&version_info))
2263 		return -ENODEV;
2264 
2265 	ret = misc_register(&mshv_dev);
2266 	if (ret)
2267 		return ret;
2268 
2269 	dev = mshv_dev.this_device;
2270 
2271 	if (version_info.build_number < MSHV_HV_MIN_VERSION ||
2272 	    version_info.build_number > MSHV_HV_MAX_VERSION) {
2273 		dev_err(dev, "Running on unvalidated Hyper-V version\n");
2274 		dev_err(dev, "Versions: current: %u  min: %u  max: %u\n",
2275 			version_info.build_number, MSHV_HV_MIN_VERSION,
2276 			MSHV_HV_MAX_VERSION);
2277 	}
2278 
2279 	mshv_root.synic_pages = alloc_percpu(struct hv_synic_pages);
2280 	if (!mshv_root.synic_pages) {
2281 		dev_err(dev, "Failed to allocate percpu synic page\n");
2282 		ret = -ENOMEM;
2283 		goto device_deregister;
2284 	}
2285 
2286 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "mshv_synic",
2287 				mshv_synic_init,
2288 				mshv_synic_cleanup);
2289 	if (ret < 0) {
2290 		dev_err(dev, "Failed to setup cpu hotplug state: %i\n", ret);
2291 		goto free_synic_pages;
2292 	}
2293 
2294 	mshv_cpuhp_online = ret;
2295 
2296 	ret = mshv_retrieve_scheduler_type(dev);
2297 	if (ret)
2298 		goto remove_cpu_state;
2299 
2300 	if (hv_root_partition())
2301 		ret = mshv_root_partition_init(dev);
2302 	if (ret)
2303 		goto remove_cpu_state;
2304 
2305 	mshv_init_vmm_caps(dev);
2306 
2307 	ret = mshv_irqfd_wq_init();
2308 	if (ret)
2309 		goto exit_partition;
2310 
2311 	spin_lock_init(&mshv_root.pt_ht_lock);
2312 	hash_init(mshv_root.pt_htable);
2313 
2314 	hv_setup_mshv_handler(mshv_isr);
2315 
2316 	return 0;
2317 
2318 exit_partition:
2319 	if (hv_root_partition())
2320 		mshv_root_partition_exit();
2321 remove_cpu_state:
2322 	cpuhp_remove_state(mshv_cpuhp_online);
2323 free_synic_pages:
2324 	free_percpu(mshv_root.synic_pages);
2325 device_deregister:
2326 	misc_deregister(&mshv_dev);
2327 	return ret;
2328 }
2329 
2330 static void __exit mshv_parent_partition_exit(void)
2331 {
2332 	hv_setup_mshv_handler(NULL);
2333 	mshv_port_table_fini();
2334 	misc_deregister(&mshv_dev);
2335 	mshv_irqfd_wq_cleanup();
2336 	if (hv_root_partition())
2337 		mshv_root_partition_exit();
2338 	cpuhp_remove_state(mshv_cpuhp_online);
2339 	free_percpu(mshv_root.synic_pages);
2340 }
2341 
2342 module_init(mshv_parent_partition_init);
2343 module_exit(mshv_parent_partition_exit);
2344