xref: /linux/drivers/hv/mshv_vtl_main.c (revision fab183d632628381b466a41479489541ac0e29a0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2023, Microsoft Corporation.
4  *
5  * Author:
6  *   Roman Kisel <romank@linux.microsoft.com>
7  *   Saurabh Sengar <ssengar@linux.microsoft.com>
8  *   Naman Jain <namjain@linux.microsoft.com>
9  */
10 
11 #include <linux/kernel.h>
12 #include <linux/module.h>
13 #include <linux/miscdevice.h>
14 #include <linux/anon_inodes.h>
15 #include <linux/cpuhotplug.h>
16 #include <linux/count_zeros.h>
17 #include <linux/entry-virt.h>
18 #include <linux/eventfd.h>
19 #include <linux/poll.h>
20 #include <linux/file.h>
21 #include <linux/vmalloc.h>
22 #include <asm/debugreg.h>
23 #include <asm/mshyperv.h>
24 #include <trace/events/ipi.h>
25 #include <uapi/asm/mtrr.h>
26 #include <uapi/linux/mshv.h>
27 #include <hyperv/hvhdk.h>
28 
29 #include "../../kernel/fpu/legacy.h"
30 #include "mshv.h"
31 #include "mshv_vtl.h"
32 #include "hyperv_vmbus.h"
33 
34 MODULE_AUTHOR("Microsoft");
35 MODULE_LICENSE("GPL");
36 MODULE_DESCRIPTION("Microsoft Hyper-V VTL Driver");
37 
38 #define MSHV_ENTRY_REASON_LOWER_VTL_CALL     0x1
39 #define MSHV_ENTRY_REASON_INTERRUPT          0x2
40 #define MSHV_ENTRY_REASON_INTERCEPT          0x3
41 
42 #define MSHV_REAL_OFF_SHIFT	16
43 #define MSHV_PG_OFF_CPU_MASK	(BIT_ULL(MSHV_REAL_OFF_SHIFT) - 1)
44 #define MSHV_RUN_PAGE_OFFSET	0
45 #define MSHV_REG_PAGE_OFFSET	1
46 #define VTL2_VMBUS_SINT_INDEX	7
47 
48 static struct device *mem_dev;
49 
50 static struct tasklet_struct msg_dpc;
51 static wait_queue_head_t fd_wait_queue;
52 static bool has_message;
53 static struct eventfd_ctx *flag_eventfds[HV_EVENT_FLAGS_COUNT];
54 static DEFINE_MUTEX(flag_lock);
55 static bool __read_mostly mshv_has_reg_page;
56 
57 /* hvcall code is of type u16, allocate a bitmap of size (1 << 16) to accommodate it */
58 #define MAX_BITMAP_SIZE ((U16_MAX + 1) / 8)
59 
60 struct mshv_vtl_hvcall_fd {
61 	u8 allow_bitmap[MAX_BITMAP_SIZE];
62 	bool allow_map_initialized;
63 	/*
64 	 * Used to protect hvcall setup in IOCTLs
65 	 */
66 	struct mutex init_mutex;
67 	struct miscdevice *dev;
68 };
69 
70 struct mshv_vtl_poll_file {
71 	struct file *file;
72 	wait_queue_entry_t wait;
73 	wait_queue_head_t *wqh;
74 	poll_table pt;
75 	int cpu;
76 };
77 
78 struct mshv_vtl {
79 	struct device *module_dev;
80 	u64 id;
81 };
82 
83 struct mshv_vtl_per_cpu {
84 	struct mshv_vtl_run *run;
85 	struct page *reg_page;
86 };
87 
88 /* SYNIC_OVERLAY_PAGE_MSR - internal, identical to hv_synic_simp */
89 union hv_synic_overlay_page_msr {
90 	u64 as_uint64;
91 	struct {
92 		u64 enabled: 1;
93 		u64 reserved: 11;
94 		u64 pfn: 52;
95 	} __packed;
96 };
97 
98 static struct mutex mshv_vtl_poll_file_lock;
99 static union hv_register_vsm_page_offsets mshv_vsm_page_offsets;
100 static union hv_register_vsm_capabilities mshv_vsm_capabilities;
101 
102 static DEFINE_PER_CPU(struct mshv_vtl_poll_file, mshv_vtl_poll_file);
103 static DEFINE_PER_CPU(unsigned long long, num_vtl0_transitions);
104 static DEFINE_PER_CPU(struct mshv_vtl_per_cpu, mshv_vtl_per_cpu);
105 
106 static const union hv_input_vtl input_vtl_zero;
107 static const union hv_input_vtl input_vtl_normal = {
108 	.use_target_vtl = 1,
109 };
110 
111 static const struct file_operations mshv_vtl_fops;
112 
113 static long
mshv_ioctl_create_vtl(void __user * user_arg,struct device * module_dev)114 mshv_ioctl_create_vtl(void __user *user_arg, struct device *module_dev)
115 {
116 	struct mshv_vtl *vtl;
117 	struct file *file;
118 	int fd;
119 
120 	vtl = kzalloc_obj(*vtl);
121 	if (!vtl)
122 		return -ENOMEM;
123 
124 	fd = get_unused_fd_flags(O_CLOEXEC);
125 	if (fd < 0) {
126 		kfree(vtl);
127 		return fd;
128 	}
129 	file = anon_inode_getfile("mshv_vtl", &mshv_vtl_fops,
130 				  vtl, O_RDWR);
131 	if (IS_ERR(file)) {
132 		put_unused_fd(fd);
133 		kfree(vtl);
134 		return PTR_ERR(file);
135 	}
136 	vtl->module_dev = module_dev;
137 	fd_install(fd, file);
138 
139 	return fd;
140 }
141 
142 static long
mshv_ioctl_check_extension(void __user * user_arg)143 mshv_ioctl_check_extension(void __user *user_arg)
144 {
145 	u32 arg;
146 
147 	if (copy_from_user(&arg, user_arg, sizeof(arg)))
148 		return -EFAULT;
149 
150 	switch (arg) {
151 	case MSHV_CAP_CORE_API_STABLE:
152 		return 0;
153 	case MSHV_CAP_REGISTER_PAGE:
154 		return mshv_has_reg_page;
155 	case MSHV_CAP_VTL_RETURN_ACTION:
156 		return mshv_vsm_capabilities.return_action_available;
157 	case MSHV_CAP_DR6_SHARED:
158 		return mshv_vsm_capabilities.dr6_shared;
159 	}
160 
161 	return -EOPNOTSUPP;
162 }
163 
164 static long
mshv_dev_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)165 mshv_dev_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
166 {
167 	struct miscdevice *misc = filp->private_data;
168 
169 	switch (ioctl) {
170 	case MSHV_CHECK_EXTENSION:
171 		return mshv_ioctl_check_extension((void __user *)arg);
172 	case MSHV_CREATE_VTL:
173 		return mshv_ioctl_create_vtl((void __user *)arg, misc->this_device);
174 	}
175 
176 	return -ENOTTY;
177 }
178 
179 static const struct file_operations mshv_dev_fops = {
180 	.owner		= THIS_MODULE,
181 	.unlocked_ioctl	= mshv_dev_ioctl,
182 	.llseek		= noop_llseek,
183 };
184 
185 static struct miscdevice mshv_dev = {
186 	.minor = MISC_DYNAMIC_MINOR,
187 	.name = "mshv",
188 	.fops = &mshv_dev_fops,
189 	.mode = 0600,
190 };
191 
mshv_vtl_this_run(void)192 static struct mshv_vtl_run *mshv_vtl_this_run(void)
193 {
194 	return *this_cpu_ptr(&mshv_vtl_per_cpu.run);
195 }
196 
mshv_vtl_cpu_run(int cpu)197 static struct mshv_vtl_run *mshv_vtl_cpu_run(int cpu)
198 {
199 	return *per_cpu_ptr(&mshv_vtl_per_cpu.run, cpu);
200 }
201 
mshv_vtl_cpu_reg_page(int cpu)202 static struct page *mshv_vtl_cpu_reg_page(int cpu)
203 {
204 	return *per_cpu_ptr(&mshv_vtl_per_cpu.reg_page, cpu);
205 }
206 
mshv_vtl_configure_reg_page(struct mshv_vtl_per_cpu * per_cpu)207 static void mshv_vtl_configure_reg_page(struct mshv_vtl_per_cpu *per_cpu)
208 {
209 	struct hv_register_assoc reg_assoc = {};
210 	union hv_synic_overlay_page_msr overlay = {};
211 	struct page *reg_page;
212 
213 	reg_page = alloc_page(GFP_KERNEL | __GFP_ZERO | __GFP_RETRY_MAYFAIL);
214 	if (!reg_page) {
215 		WARN(1, "failed to allocate register page\n");
216 		return;
217 	}
218 
219 	overlay.enabled = 1;
220 	overlay.pfn = page_to_hvpfn(reg_page);
221 	reg_assoc.name = HV_X64_REGISTER_REG_PAGE;
222 	reg_assoc.value.reg64 = overlay.as_uint64;
223 
224 	if (hv_call_set_vp_registers(HV_VP_INDEX_SELF, HV_PARTITION_ID_SELF,
225 				     1, input_vtl_zero, &reg_assoc)) {
226 		WARN(1, "failed to setup register page\n");
227 		__free_page(reg_page);
228 		return;
229 	}
230 
231 	per_cpu->reg_page = reg_page;
232 	mshv_has_reg_page = true;
233 }
234 
mshv_vtl_synic_enable_regs(unsigned int cpu)235 static void mshv_vtl_synic_enable_regs(unsigned int cpu)
236 {
237 	union hv_synic_sint sint;
238 
239 	sint.as_uint64 = 0;
240 	sint.vector = HYPERVISOR_CALLBACK_VECTOR;
241 	sint.masked = false;
242 	sint.auto_eoi = hv_recommend_using_aeoi();
243 
244 	/* Enable intercepts */
245 	if (!mshv_vsm_capabilities.intercept_page_available)
246 		hv_set_msr(HV_MSR_SINT0 + HV_SYNIC_INTERCEPTION_SINT_INDEX,
247 			   sint.as_uint64);
248 
249 	/* VTL2 Host VSP SINT is (un)masked when the user mode requests that */
250 }
251 
mshv_vtl_get_vsm_regs(void)252 static int mshv_vtl_get_vsm_regs(void)
253 {
254 	struct hv_register_assoc registers[2];
255 	int ret, count = 2;
256 
257 	registers[0].name = HV_REGISTER_VSM_CODE_PAGE_OFFSETS;
258 	registers[1].name = HV_REGISTER_VSM_CAPABILITIES;
259 
260 	ret = hv_call_get_vp_registers(HV_VP_INDEX_SELF, HV_PARTITION_ID_SELF,
261 				       count, input_vtl_zero, registers);
262 	if (ret)
263 		return ret;
264 
265 	mshv_vsm_page_offsets.as_uint64 = registers[0].value.reg64;
266 	mshv_vsm_capabilities.as_uint64 = registers[1].value.reg64;
267 
268 	return ret;
269 }
270 
mshv_vtl_configure_vsm_partition(struct device * dev)271 static int mshv_vtl_configure_vsm_partition(struct device *dev)
272 {
273 	union hv_register_vsm_partition_config config;
274 	struct hv_register_assoc reg_assoc;
275 
276 	config.as_uint64 = 0;
277 	config.default_vtl_protection_mask = HV_MAP_GPA_PERMISSIONS_MASK;
278 	config.enable_vtl_protection = 1;
279 	config.zero_memory_on_reset = 1;
280 	config.intercept_vp_startup = 1;
281 	config.intercept_cpuid_unimplemented = 1;
282 
283 	if (mshv_vsm_capabilities.intercept_page_available) {
284 		dev_dbg(dev, "using intercept page\n");
285 		config.intercept_page = 1;
286 	}
287 
288 	reg_assoc.name = HV_REGISTER_VSM_PARTITION_CONFIG;
289 	reg_assoc.value.reg64 = config.as_uint64;
290 
291 	return hv_call_set_vp_registers(HV_VP_INDEX_SELF, HV_PARTITION_ID_SELF,
292 				       1, input_vtl_zero, &reg_assoc);
293 }
294 
mshv_vtl_vmbus_isr(void)295 static void mshv_vtl_vmbus_isr(void)
296 {
297 	struct hv_per_cpu_context *per_cpu;
298 	struct hv_message *msg;
299 	u32 message_type;
300 	union hv_synic_event_flags *event_flags;
301 	struct eventfd_ctx *eventfd;
302 	u16 i;
303 
304 	per_cpu = this_cpu_ptr(hv_context.cpu_context);
305 	if (smp_processor_id() == 0) {
306 		msg = (struct hv_message *)per_cpu->hyp_synic_message_page + VTL2_VMBUS_SINT_INDEX;
307 		message_type = READ_ONCE(msg->header.message_type);
308 		if (message_type != HVMSG_NONE)
309 			tasklet_schedule(&msg_dpc);
310 	}
311 
312 	event_flags = (union hv_synic_event_flags *)per_cpu->hyp_synic_event_page +
313 			VTL2_VMBUS_SINT_INDEX;
314 	for_each_set_bit(i, event_flags->flags, HV_EVENT_FLAGS_COUNT) {
315 		if (!sync_test_and_clear_bit(i, event_flags->flags))
316 			continue;
317 		rcu_read_lock();
318 		eventfd = READ_ONCE(flag_eventfds[i]);
319 		if (eventfd)
320 			eventfd_signal(eventfd);
321 		rcu_read_unlock();
322 	}
323 
324 	vmbus_isr();
325 }
326 
mshv_vtl_alloc_context(unsigned int cpu)327 static int mshv_vtl_alloc_context(unsigned int cpu)
328 {
329 	struct mshv_vtl_per_cpu *per_cpu = this_cpu_ptr(&mshv_vtl_per_cpu);
330 
331 	per_cpu->run = (struct mshv_vtl_run *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
332 	if (!per_cpu->run)
333 		return -ENOMEM;
334 
335 	if (mshv_vsm_capabilities.intercept_page_available)
336 		mshv_vtl_configure_reg_page(per_cpu);
337 
338 	mshv_vtl_synic_enable_regs(cpu);
339 
340 	return 0;
341 }
342 
343 static int mshv_vtl_cpuhp_online;
344 
hv_vtl_setup_synic(void)345 static int hv_vtl_setup_synic(void)
346 {
347 	int ret;
348 
349 	/* Use our isr to first filter out packets destined for userspace */
350 	hv_setup_vmbus_handler(mshv_vtl_vmbus_isr);
351 
352 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vtl:online",
353 				mshv_vtl_alloc_context, NULL);
354 	if (ret < 0) {
355 		hv_setup_vmbus_handler(vmbus_isr);
356 		return ret;
357 	}
358 
359 	mshv_vtl_cpuhp_online = ret;
360 
361 	return 0;
362 }
363 
hv_vtl_remove_synic(void)364 static void hv_vtl_remove_synic(void)
365 {
366 	cpuhp_remove_state(mshv_vtl_cpuhp_online);
367 	hv_setup_vmbus_handler(vmbus_isr);
368 }
369 
vtl_get_vp_register(struct hv_register_assoc * reg)370 static int vtl_get_vp_register(struct hv_register_assoc *reg)
371 {
372 	return hv_call_get_vp_registers(HV_VP_INDEX_SELF, HV_PARTITION_ID_SELF,
373 					1, input_vtl_normal, reg);
374 }
375 
vtl_set_vp_register(struct hv_register_assoc * reg)376 static int vtl_set_vp_register(struct hv_register_assoc *reg)
377 {
378 	return hv_call_set_vp_registers(HV_VP_INDEX_SELF, HV_PARTITION_ID_SELF,
379 					1, input_vtl_normal, reg);
380 }
381 
mshv_vtl_ioctl_add_vtl0_mem(struct mshv_vtl * vtl,void __user * arg)382 static int mshv_vtl_ioctl_add_vtl0_mem(struct mshv_vtl *vtl, void __user *arg)
383 {
384 	struct mshv_vtl_ram_disposition vtl0_mem;
385 	struct dev_pagemap *pgmap;
386 	void *addr;
387 
388 	if (copy_from_user(&vtl0_mem, arg, sizeof(vtl0_mem)))
389 		return -EFAULT;
390 	if (vtl0_mem.last_pfn <= vtl0_mem.start_pfn) {
391 		dev_err(vtl->module_dev, "range start pfn (%llx) > end pfn (%llx)\n",
392 			vtl0_mem.start_pfn, vtl0_mem.last_pfn);
393 		return -EFAULT;
394 	}
395 
396 	pgmap = kzalloc_obj(*pgmap);
397 	if (!pgmap)
398 		return -ENOMEM;
399 
400 	/*
401 	 * vtl0_mem.last_pfn is excluded in the pagemap range for VTL0 as per design.
402 	 * last_pfn is not reserved or wasted, and reflects 'start_pfn + size' of pagemap range.
403 	 */
404 	pgmap->ranges[0].start = PFN_PHYS(vtl0_mem.start_pfn);
405 	pgmap->ranges[0].end = PFN_PHYS(vtl0_mem.last_pfn) - 1;
406 	pgmap->nr_range = 1;
407 	pgmap->type = MEMORY_DEVICE_GENERIC;
408 
409 	/*
410 	 * Determine the highest page order that can be used for the given memory range.
411 	 * This works best when the range is aligned; i.e. both the start and the length.
412 	 * Clamp to MAX_FOLIO_ORDER to avoid a WARN in memremap_pages() when the range
413 	 * alignment exceeds the maximum supported folio order for this kernel config.
414 	 */
415 	pgmap->vmemmap_shift = min(count_trailing_zeros(vtl0_mem.start_pfn | vtl0_mem.last_pfn),
416 				   MAX_FOLIO_ORDER);
417 	dev_dbg(vtl->module_dev,
418 		"Add VTL0 memory: start: 0x%llx, end_pfn: 0x%llx, page order: %lu\n",
419 		vtl0_mem.start_pfn, vtl0_mem.last_pfn, pgmap->vmemmap_shift);
420 
421 	addr = devm_memremap_pages(mem_dev, pgmap);
422 	if (IS_ERR(addr)) {
423 		dev_err(vtl->module_dev, "devm_memremap_pages error: %ld\n", PTR_ERR(addr));
424 		kfree(pgmap);
425 		return PTR_ERR(addr);
426 	}
427 
428 	/* Don't free pgmap, since it has to stick around until the memory
429 	 * is unmapped, which will never happen as there is no scenario
430 	 * where VTL0 can be released/shutdown without bringing down VTL2.
431 	 */
432 	return 0;
433 }
434 
mshv_vtl_cancel(int cpu)435 static void mshv_vtl_cancel(int cpu)
436 {
437 	int here = get_cpu();
438 
439 	if (here != cpu) {
440 		if (!xchg_relaxed(&mshv_vtl_cpu_run(cpu)->cancel, 1))
441 			smp_send_reschedule(cpu);
442 	} else {
443 		WRITE_ONCE(mshv_vtl_this_run()->cancel, 1);
444 	}
445 	put_cpu();
446 }
447 
mshv_vtl_poll_file_wake(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)448 static int mshv_vtl_poll_file_wake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
449 {
450 	struct mshv_vtl_poll_file *poll_file = container_of(wait, struct mshv_vtl_poll_file, wait);
451 
452 	mshv_vtl_cancel(poll_file->cpu);
453 
454 	return 0;
455 }
456 
mshv_vtl_ptable_queue_proc(struct file * file,wait_queue_head_t * wqh,poll_table * pt)457 static void mshv_vtl_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh, poll_table *pt)
458 {
459 	struct mshv_vtl_poll_file *poll_file = container_of(pt, struct mshv_vtl_poll_file, pt);
460 
461 	WARN_ON(poll_file->wqh);
462 	poll_file->wqh = wqh;
463 	add_wait_queue(wqh, &poll_file->wait);
464 }
465 
mshv_vtl_ioctl_set_poll_file(struct mshv_vtl_set_poll_file __user * user_input)466 static int mshv_vtl_ioctl_set_poll_file(struct mshv_vtl_set_poll_file __user *user_input)
467 {
468 	struct file *file, *old_file;
469 	struct mshv_vtl_poll_file *poll_file;
470 	struct mshv_vtl_set_poll_file input;
471 
472 	if (copy_from_user(&input, user_input, sizeof(input)))
473 		return -EFAULT;
474 
475 	if (input.cpu >= num_possible_cpus() || !cpu_online(input.cpu))
476 		return -EINVAL;
477 	/*
478 	 * CPU Hotplug is not supported in VTL2 in OpenHCL, where this kernel driver exists.
479 	 * CPU is expected to remain online after above cpu_online() check.
480 	 */
481 
482 	file = NULL;
483 	file = fget(input.fd);
484 	if (!file)
485 		return -EBADFD;
486 
487 	poll_file = per_cpu_ptr(&mshv_vtl_poll_file, READ_ONCE(input.cpu));
488 	if (!poll_file)
489 		return -EINVAL;
490 
491 	mutex_lock(&mshv_vtl_poll_file_lock);
492 
493 	if (poll_file->wqh)
494 		remove_wait_queue(poll_file->wqh, &poll_file->wait);
495 	poll_file->wqh = NULL;
496 
497 	old_file = poll_file->file;
498 	poll_file->file = file;
499 	poll_file->cpu = input.cpu;
500 
501 	if (file) {
502 		init_waitqueue_func_entry(&poll_file->wait, mshv_vtl_poll_file_wake);
503 		init_poll_funcptr(&poll_file->pt, mshv_vtl_ptable_queue_proc);
504 		vfs_poll(file, &poll_file->pt);
505 	}
506 
507 	mutex_unlock(&mshv_vtl_poll_file_lock);
508 
509 	if (old_file)
510 		fput(old_file);
511 
512 	return 0;
513 }
514 
515 /* Static table mapping register names to their corresponding actions */
516 static const struct {
517 	enum hv_register_name reg_name;
518 	int debug_reg_num;  /* -1 if not a debug register */
519 	u32 msr_addr;       /* 0 if not an MSR */
520 } reg_table[] = {
521 	/* Debug registers */
522 	{HV_X64_REGISTER_DR0, 0, 0},
523 	{HV_X64_REGISTER_DR1, 1, 0},
524 	{HV_X64_REGISTER_DR2, 2, 0},
525 	{HV_X64_REGISTER_DR3, 3, 0},
526 	{HV_X64_REGISTER_DR6, 6, 0},
527 	/* MTRR MSRs */
528 	{HV_X64_REGISTER_MSR_MTRR_CAP, -1, MSR_MTRRcap},
529 	{HV_X64_REGISTER_MSR_MTRR_DEF_TYPE, -1, MSR_MTRRdefType},
530 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE0, -1, MTRRphysBase_MSR(0)},
531 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE1, -1, MTRRphysBase_MSR(1)},
532 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE2, -1, MTRRphysBase_MSR(2)},
533 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE3, -1, MTRRphysBase_MSR(3)},
534 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE4, -1, MTRRphysBase_MSR(4)},
535 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE5, -1, MTRRphysBase_MSR(5)},
536 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE6, -1, MTRRphysBase_MSR(6)},
537 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE7, -1, MTRRphysBase_MSR(7)},
538 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE8, -1, MTRRphysBase_MSR(8)},
539 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASE9, -1, MTRRphysBase_MSR(9)},
540 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASEA, -1, MTRRphysBase_MSR(0xa)},
541 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASEB, -1, MTRRphysBase_MSR(0xb)},
542 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASEC, -1, MTRRphysBase_MSR(0xc)},
543 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASED, -1, MTRRphysBase_MSR(0xd)},
544 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASEE, -1, MTRRphysBase_MSR(0xe)},
545 	{HV_X64_REGISTER_MSR_MTRR_PHYS_BASEF, -1, MTRRphysBase_MSR(0xf)},
546 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK0, -1, MTRRphysMask_MSR(0)},
547 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK1, -1, MTRRphysMask_MSR(1)},
548 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK2, -1, MTRRphysMask_MSR(2)},
549 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK3, -1, MTRRphysMask_MSR(3)},
550 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK4, -1, MTRRphysMask_MSR(4)},
551 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK5, -1, MTRRphysMask_MSR(5)},
552 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK6, -1, MTRRphysMask_MSR(6)},
553 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK7, -1, MTRRphysMask_MSR(7)},
554 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK8, -1, MTRRphysMask_MSR(8)},
555 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASK9, -1, MTRRphysMask_MSR(9)},
556 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASKA, -1, MTRRphysMask_MSR(0xa)},
557 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASKB, -1, MTRRphysMask_MSR(0xb)},
558 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASKC, -1, MTRRphysMask_MSR(0xc)},
559 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASKD, -1, MTRRphysMask_MSR(0xd)},
560 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASKE, -1, MTRRphysMask_MSR(0xe)},
561 	{HV_X64_REGISTER_MSR_MTRR_PHYS_MASKF, -1, MTRRphysMask_MSR(0xf)},
562 	{HV_X64_REGISTER_MSR_MTRR_FIX64K00000, -1, MSR_MTRRfix64K_00000},
563 	{HV_X64_REGISTER_MSR_MTRR_FIX16K80000, -1, MSR_MTRRfix16K_80000},
564 	{HV_X64_REGISTER_MSR_MTRR_FIX16KA0000, -1, MSR_MTRRfix16K_A0000},
565 	{HV_X64_REGISTER_MSR_MTRR_FIX4KC0000, -1, MSR_MTRRfix4K_C0000},
566 	{HV_X64_REGISTER_MSR_MTRR_FIX4KC8000, -1, MSR_MTRRfix4K_C8000},
567 	{HV_X64_REGISTER_MSR_MTRR_FIX4KD0000, -1, MSR_MTRRfix4K_D0000},
568 	{HV_X64_REGISTER_MSR_MTRR_FIX4KD8000, -1, MSR_MTRRfix4K_D8000},
569 	{HV_X64_REGISTER_MSR_MTRR_FIX4KE0000, -1, MSR_MTRRfix4K_E0000},
570 	{HV_X64_REGISTER_MSR_MTRR_FIX4KE8000, -1, MSR_MTRRfix4K_E8000},
571 	{HV_X64_REGISTER_MSR_MTRR_FIX4KF0000, -1, MSR_MTRRfix4K_F0000},
572 	{HV_X64_REGISTER_MSR_MTRR_FIX4KF8000, -1, MSR_MTRRfix4K_F8000},
573 };
574 
mshv_vtl_get_set_reg(struct hv_register_assoc * regs,bool set)575 static int mshv_vtl_get_set_reg(struct hv_register_assoc *regs, bool set)
576 {
577 	u64 *reg64;
578 	enum hv_register_name gpr_name;
579 	int i;
580 
581 	gpr_name = regs->name;
582 	reg64 = &regs->value.reg64;
583 
584 	/* Search for the register in the table */
585 	for (i = 0; i < ARRAY_SIZE(reg_table); i++) {
586 		if (reg_table[i].reg_name != gpr_name)
587 			continue;
588 		if (reg_table[i].debug_reg_num != -1) {
589 			/* Handle debug registers */
590 			if (gpr_name == HV_X64_REGISTER_DR6 &&
591 			    !mshv_vsm_capabilities.dr6_shared)
592 				goto hypercall;
593 			if (set)
594 				native_set_debugreg(reg_table[i].debug_reg_num, *reg64);
595 			else
596 				*reg64 = native_get_debugreg(reg_table[i].debug_reg_num);
597 		} else {
598 			/* Handle MSRs */
599 			if (set)
600 				wrmsrq(reg_table[i].msr_addr, *reg64);
601 			else
602 				rdmsrq(reg_table[i].msr_addr, *reg64);
603 		}
604 		return 0;
605 	}
606 
607 hypercall:
608 	return 1;
609 }
610 
mshv_vtl_return(struct mshv_vtl_cpu_context * vtl0)611 static void mshv_vtl_return(struct mshv_vtl_cpu_context *vtl0)
612 {
613 	struct hv_vp_assist_page *hvp;
614 
615 	hvp = hv_vp_assist_page[smp_processor_id()];
616 
617 	/*
618 	 * Process signal event direct set in the run page, if any.
619 	 */
620 	if (mshv_vsm_capabilities.return_action_available) {
621 		u32 offset = READ_ONCE(mshv_vtl_this_run()->vtl_ret_action_size);
622 
623 		WRITE_ONCE(mshv_vtl_this_run()->vtl_ret_action_size, 0);
624 
625 		/*
626 		 * Hypervisor will take care of clearing out the actions
627 		 * set in the assist page.
628 		 */
629 		memcpy(hvp->vtl_ret_actions,
630 		       mshv_vtl_this_run()->vtl_ret_actions,
631 		       min_t(u32, offset, sizeof(hvp->vtl_ret_actions)));
632 	}
633 
634 	mshv_vtl_return_call(vtl0);
635 }
636 
mshv_vtl_process_intercept(void)637 static bool mshv_vtl_process_intercept(void)
638 {
639 	struct hv_per_cpu_context *mshv_cpu;
640 	void *synic_message_page;
641 	struct hv_message *msg;
642 	u32 message_type;
643 
644 	mshv_cpu = this_cpu_ptr(hv_context.cpu_context);
645 	synic_message_page = mshv_cpu->hyp_synic_message_page;
646 	if (unlikely(!synic_message_page))
647 		return true;
648 
649 	msg = (struct hv_message *)synic_message_page + HV_SYNIC_INTERCEPTION_SINT_INDEX;
650 	message_type = READ_ONCE(msg->header.message_type);
651 	if (message_type == HVMSG_NONE)
652 		return true;
653 
654 	memcpy(mshv_vtl_this_run()->exit_message, msg, sizeof(*msg));
655 	vmbus_signal_eom(msg, message_type);
656 
657 	return false;
658 }
659 
mshv_vtl_ioctl_return_to_lower_vtl(void)660 static int mshv_vtl_ioctl_return_to_lower_vtl(void)
661 {
662 	preempt_disable();
663 	for (;;) {
664 		unsigned long irq_flags;
665 		struct hv_vp_assist_page *hvp;
666 		int ret;
667 
668 		if (__xfer_to_guest_mode_work_pending()) {
669 			preempt_enable();
670 			ret = xfer_to_guest_mode_handle_work();
671 			if (ret)
672 				return ret;
673 			preempt_disable();
674 		}
675 
676 		local_irq_save(irq_flags);
677 		if (READ_ONCE(mshv_vtl_this_run()->cancel)) {
678 			local_irq_restore(irq_flags);
679 			preempt_enable();
680 			return -EINTR;
681 		}
682 
683 		mshv_vtl_return(&mshv_vtl_this_run()->cpu_context);
684 		local_irq_restore(irq_flags);
685 
686 		hvp = hv_vp_assist_page[smp_processor_id()];
687 		this_cpu_inc(num_vtl0_transitions);
688 		switch (hvp->vtl_entry_reason) {
689 		case MSHV_ENTRY_REASON_INTERRUPT:
690 			if (!mshv_vsm_capabilities.intercept_page_available &&
691 			    likely(!mshv_vtl_process_intercept()))
692 				goto done;
693 			break;
694 
695 		case MSHV_ENTRY_REASON_INTERCEPT:
696 			WARN_ON(!mshv_vsm_capabilities.intercept_page_available);
697 			memcpy(mshv_vtl_this_run()->exit_message, hvp->intercept_message,
698 			       sizeof(hvp->intercept_message));
699 			goto done;
700 
701 		default:
702 			panic("unknown entry reason: %d", hvp->vtl_entry_reason);
703 		}
704 	}
705 
706 done:
707 	preempt_enable();
708 
709 	return 0;
710 }
711 
712 static long
mshv_vtl_ioctl_get_regs(void __user * user_args)713 mshv_vtl_ioctl_get_regs(void __user *user_args)
714 {
715 	struct mshv_vp_registers args;
716 	struct hv_register_assoc reg;
717 	long ret;
718 
719 	if (copy_from_user(&args, user_args, sizeof(args)))
720 		return -EFAULT;
721 
722 	/*  This IOCTL supports processing only one register at a time. */
723 	if (args.count != 1)
724 		return -EINVAL;
725 
726 	if (copy_from_user(&reg, (void __user *)args.regs_ptr,
727 			   sizeof(reg)))
728 		return -EFAULT;
729 
730 	ret = mshv_vtl_get_set_reg(&reg, false);
731 	if (!ret)
732 		goto copy_args; /* No need of hypercall */
733 	ret = vtl_get_vp_register(&reg);
734 	if (ret)
735 		return ret;
736 
737 copy_args:
738 	if (copy_to_user((void __user *)args.regs_ptr, &reg, sizeof(reg)))
739 		ret = -EFAULT;
740 
741 	return ret;
742 }
743 
744 static long
mshv_vtl_ioctl_set_regs(void __user * user_args)745 mshv_vtl_ioctl_set_regs(void __user *user_args)
746 {
747 	struct mshv_vp_registers args;
748 	struct hv_register_assoc reg;
749 	long ret;
750 
751 	if (copy_from_user(&args, user_args, sizeof(args)))
752 		return -EFAULT;
753 
754 	/*  This IOCTL supports processing only one register at a time. */
755 	if (args.count != 1)
756 		return -EINVAL;
757 
758 	if (copy_from_user(&reg, (void __user *)args.regs_ptr, sizeof(reg)))
759 		return -EFAULT;
760 
761 	ret = mshv_vtl_get_set_reg(&reg, true);
762 	if (!ret)
763 		return ret; /* No need of hypercall */
764 	ret = vtl_set_vp_register(&reg);
765 
766 	return ret;
767 }
768 
769 static long
mshv_vtl_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)770 mshv_vtl_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
771 {
772 	long ret;
773 	struct mshv_vtl *vtl = filp->private_data;
774 
775 	switch (ioctl) {
776 	case MSHV_SET_POLL_FILE:
777 		ret = mshv_vtl_ioctl_set_poll_file((struct mshv_vtl_set_poll_file __user *)arg);
778 		break;
779 	case MSHV_GET_VP_REGISTERS:
780 		ret = mshv_vtl_ioctl_get_regs((void __user *)arg);
781 		break;
782 	case MSHV_SET_VP_REGISTERS:
783 		ret = mshv_vtl_ioctl_set_regs((void __user *)arg);
784 		break;
785 	case MSHV_RETURN_TO_LOWER_VTL:
786 		ret = mshv_vtl_ioctl_return_to_lower_vtl();
787 		break;
788 	case MSHV_ADD_VTL0_MEMORY:
789 		ret = mshv_vtl_ioctl_add_vtl0_mem(vtl, (void __user *)arg);
790 		break;
791 	default:
792 		dev_err(vtl->module_dev, "invalid vtl ioctl: %#x\n", ioctl);
793 		ret = -ENOTTY;
794 	}
795 
796 	return ret;
797 }
798 
mshv_vtl_fault(struct vm_fault * vmf)799 static vm_fault_t mshv_vtl_fault(struct vm_fault *vmf)
800 {
801 	struct page *page;
802 	int cpu = vmf->pgoff & MSHV_PG_OFF_CPU_MASK;
803 	int real_off = vmf->pgoff >> MSHV_REAL_OFF_SHIFT;
804 
805 	if (cpu >= nr_cpu_ids || !cpu_online(cpu))
806 		return VM_FAULT_SIGBUS;
807 	/*
808 	 * CPU Hotplug is not supported in VTL2 in OpenHCL, where this kernel driver exists.
809 	 * CPU is expected to remain online after above cpu_online() check.
810 	 */
811 
812 	if (real_off == MSHV_RUN_PAGE_OFFSET) {
813 		page = virt_to_page(mshv_vtl_cpu_run(cpu));
814 	} else if (real_off == MSHV_REG_PAGE_OFFSET) {
815 		if (!mshv_has_reg_page)
816 			return VM_FAULT_SIGBUS;
817 		page = mshv_vtl_cpu_reg_page(cpu);
818 	} else {
819 		return VM_FAULT_NOPAGE;
820 	}
821 
822 	get_page(page);
823 	vmf->page = page;
824 
825 	return 0;
826 }
827 
828 static const struct vm_operations_struct mshv_vtl_vm_ops = {
829 	.fault = mshv_vtl_fault,
830 };
831 
mshv_vtl_mmap(struct file * filp,struct vm_area_struct * vma)832 static int mshv_vtl_mmap(struct file *filp, struct vm_area_struct *vma)
833 {
834 	vma->vm_ops = &mshv_vtl_vm_ops;
835 
836 	return 0;
837 }
838 
mshv_vtl_release(struct inode * inode,struct file * filp)839 static int mshv_vtl_release(struct inode *inode, struct file *filp)
840 {
841 	struct mshv_vtl *vtl = filp->private_data;
842 
843 	kfree(vtl);
844 
845 	return 0;
846 }
847 
848 static const struct file_operations mshv_vtl_fops = {
849 	.owner = THIS_MODULE,
850 	.unlocked_ioctl = mshv_vtl_ioctl,
851 	.release = mshv_vtl_release,
852 	.mmap = mshv_vtl_mmap,
853 };
854 
mshv_vtl_synic_mask_vmbus_sint(void * info)855 static void mshv_vtl_synic_mask_vmbus_sint(void *info)
856 {
857 	union hv_synic_sint sint;
858 	const u8 *mask = info;
859 
860 	sint.as_uint64 = 0;
861 	sint.vector = HYPERVISOR_CALLBACK_VECTOR;
862 	sint.masked = (*mask != 0);
863 	sint.auto_eoi = hv_recommend_using_aeoi();
864 
865 	hv_set_msr(HV_MSR_SINT0 + VTL2_VMBUS_SINT_INDEX,
866 		   sint.as_uint64);
867 
868 	if (!sint.masked)
869 		pr_debug("%s: Unmasking VTL2 VMBUS SINT on VP %d\n", __func__, smp_processor_id());
870 	else
871 		pr_debug("%s: Masking VTL2 VMBUS SINT on VP %d\n", __func__, smp_processor_id());
872 }
873 
mshv_vtl_read_remote(void * buffer)874 static void mshv_vtl_read_remote(void *buffer)
875 {
876 	struct hv_per_cpu_context *mshv_cpu = this_cpu_ptr(hv_context.cpu_context);
877 	struct hv_message *msg = (struct hv_message *)mshv_cpu->hyp_synic_message_page +
878 					VTL2_VMBUS_SINT_INDEX;
879 	u32 message_type = READ_ONCE(msg->header.message_type);
880 
881 	WRITE_ONCE(has_message, false);
882 	if (message_type == HVMSG_NONE)
883 		return;
884 
885 	memcpy(buffer, msg, sizeof(*msg));
886 	vmbus_signal_eom(msg, message_type);
887 }
888 
889 static bool vtl_synic_mask_vmbus_sint_masked = true;
890 
mshv_vtl_sint_read(struct file * filp,char __user * arg,size_t size,loff_t * offset)891 static ssize_t mshv_vtl_sint_read(struct file *filp, char __user *arg, size_t size, loff_t *offset)
892 {
893 	struct hv_message msg = {};
894 	int ret;
895 
896 	if (size < sizeof(msg))
897 		return -EINVAL;
898 
899 	for (;;) {
900 		smp_call_function_single(VMBUS_CONNECT_CPU, mshv_vtl_read_remote, &msg, true);
901 		if (msg.header.message_type != HVMSG_NONE)
902 			break;
903 
904 		if (READ_ONCE(vtl_synic_mask_vmbus_sint_masked))
905 			return 0; /* EOF */
906 
907 		if (filp->f_flags & O_NONBLOCK)
908 			return -EAGAIN;
909 
910 		ret = wait_event_interruptible(fd_wait_queue,
911 					       READ_ONCE(has_message) ||
912 						READ_ONCE(vtl_synic_mask_vmbus_sint_masked));
913 		if (ret)
914 			return ret;
915 	}
916 
917 	if (copy_to_user(arg, &msg, sizeof(msg)))
918 		return -EFAULT;
919 
920 	return sizeof(msg);
921 }
922 
mshv_vtl_sint_poll(struct file * filp,poll_table * wait)923 static __poll_t mshv_vtl_sint_poll(struct file *filp, poll_table *wait)
924 {
925 	__poll_t mask = 0;
926 
927 	poll_wait(filp, &fd_wait_queue, wait);
928 	if (READ_ONCE(has_message) || READ_ONCE(vtl_synic_mask_vmbus_sint_masked))
929 		mask |= EPOLLIN | EPOLLRDNORM;
930 
931 	return mask;
932 }
933 
mshv_vtl_sint_on_msg_dpc(unsigned long data)934 static void mshv_vtl_sint_on_msg_dpc(unsigned long data)
935 {
936 	WRITE_ONCE(has_message, true);
937 	wake_up_interruptible_poll(&fd_wait_queue, EPOLLIN);
938 }
939 
mshv_vtl_sint_ioctl_post_msg(struct mshv_vtl_sint_post_msg __user * arg)940 static int mshv_vtl_sint_ioctl_post_msg(struct mshv_vtl_sint_post_msg __user *arg)
941 {
942 	struct mshv_vtl_sint_post_msg message;
943 	u8 payload[HV_MESSAGE_PAYLOAD_BYTE_COUNT];
944 
945 	if (copy_from_user(&message, arg, sizeof(message)))
946 		return -EFAULT;
947 	if (message.payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
948 		return -EINVAL;
949 	if (copy_from_user(payload, (void __user *)message.payload_ptr,
950 			   message.payload_size))
951 		return -EFAULT;
952 
953 	return hv_post_message((union hv_connection_id)message.connection_id,
954 			       message.message_type, (void *)payload,
955 			       message.payload_size);
956 }
957 
mshv_vtl_sint_ioctl_signal_event(struct mshv_vtl_signal_event __user * arg)958 static int mshv_vtl_sint_ioctl_signal_event(struct mshv_vtl_signal_event __user *arg)
959 {
960 	u64 input, status;
961 	struct mshv_vtl_signal_event signal_event;
962 
963 	if (copy_from_user(&signal_event, arg, sizeof(signal_event)))
964 		return -EFAULT;
965 
966 	input = signal_event.connection_id | ((u64)signal_event.flag << 32);
967 
968 	status = hv_do_fast_hypercall8(HVCALL_SIGNAL_EVENT, input);
969 
970 	return hv_result_to_errno(status);
971 }
972 
mshv_vtl_sint_ioctl_set_eventfd(struct mshv_vtl_set_eventfd __user * arg)973 static int mshv_vtl_sint_ioctl_set_eventfd(struct mshv_vtl_set_eventfd __user *arg)
974 {
975 	struct mshv_vtl_set_eventfd set_eventfd;
976 	struct eventfd_ctx *eventfd, *old_eventfd;
977 
978 	if (copy_from_user(&set_eventfd, arg, sizeof(set_eventfd)))
979 		return -EFAULT;
980 	if (set_eventfd.flag >= HV_EVENT_FLAGS_COUNT)
981 		return -EINVAL;
982 
983 	eventfd = NULL;
984 	if (set_eventfd.fd >= 0) {
985 		eventfd = eventfd_ctx_fdget(set_eventfd.fd);
986 		if (IS_ERR(eventfd))
987 			return PTR_ERR(eventfd);
988 	}
989 
990 	guard(mutex)(&flag_lock);
991 	old_eventfd = READ_ONCE(flag_eventfds[set_eventfd.flag]);
992 	WRITE_ONCE(flag_eventfds[set_eventfd.flag], eventfd);
993 
994 	if (old_eventfd) {
995 		synchronize_rcu();
996 		eventfd_ctx_put(old_eventfd);
997 	}
998 
999 	return 0;
1000 }
1001 
mshv_vtl_sint_ioctl_pause_msg_stream(struct mshv_sint_mask __user * arg)1002 static int mshv_vtl_sint_ioctl_pause_msg_stream(struct mshv_sint_mask __user *arg)
1003 {
1004 	static DEFINE_MUTEX(vtl2_vmbus_sint_mask_mutex);
1005 	struct mshv_sint_mask mask;
1006 
1007 	if (copy_from_user(&mask, arg, sizeof(mask)))
1008 		return -EFAULT;
1009 	guard(mutex)(&vtl2_vmbus_sint_mask_mutex);
1010 	on_each_cpu(mshv_vtl_synic_mask_vmbus_sint, &mask.mask, 1);
1011 	WRITE_ONCE(vtl_synic_mask_vmbus_sint_masked, mask.mask != 0);
1012 	if (mask.mask)
1013 		wake_up_interruptible_poll(&fd_wait_queue, EPOLLIN);
1014 
1015 	return 0;
1016 }
1017 
mshv_vtl_sint_ioctl(struct file * f,unsigned int cmd,unsigned long arg)1018 static long mshv_vtl_sint_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1019 {
1020 	switch (cmd) {
1021 	case MSHV_SINT_POST_MESSAGE:
1022 		return mshv_vtl_sint_ioctl_post_msg((struct mshv_vtl_sint_post_msg __user *)arg);
1023 	case MSHV_SINT_SIGNAL_EVENT:
1024 		return mshv_vtl_sint_ioctl_signal_event((struct mshv_vtl_signal_event __user *)arg);
1025 	case MSHV_SINT_SET_EVENTFD:
1026 		return mshv_vtl_sint_ioctl_set_eventfd((struct mshv_vtl_set_eventfd __user *)arg);
1027 	case MSHV_SINT_PAUSE_MESSAGE_STREAM:
1028 		return mshv_vtl_sint_ioctl_pause_msg_stream((struct mshv_sint_mask __user *)arg);
1029 	default:
1030 		return -ENOIOCTLCMD;
1031 	}
1032 }
1033 
1034 static const struct file_operations mshv_vtl_sint_ops = {
1035 	.owner = THIS_MODULE,
1036 	.read = mshv_vtl_sint_read,
1037 	.poll = mshv_vtl_sint_poll,
1038 	.unlocked_ioctl = mshv_vtl_sint_ioctl,
1039 };
1040 
1041 static struct miscdevice mshv_vtl_sint_dev = {
1042 	.name = "mshv_sint",
1043 	.fops = &mshv_vtl_sint_ops,
1044 	.mode = 0600,
1045 	.minor = MISC_DYNAMIC_MINOR,
1046 };
1047 
mshv_vtl_hvcall_dev_open(struct inode * node,struct file * f)1048 static int mshv_vtl_hvcall_dev_open(struct inode *node, struct file *f)
1049 {
1050 	struct miscdevice *dev = f->private_data;
1051 	struct mshv_vtl_hvcall_fd *fd;
1052 
1053 	if (!capable(CAP_SYS_ADMIN))
1054 		return -EPERM;
1055 
1056 	fd = vzalloc(sizeof(*fd));
1057 	if (!fd)
1058 		return -ENOMEM;
1059 	fd->dev = dev;
1060 	f->private_data = fd;
1061 	mutex_init(&fd->init_mutex);
1062 
1063 	return 0;
1064 }
1065 
mshv_vtl_hvcall_dev_release(struct inode * node,struct file * f)1066 static int mshv_vtl_hvcall_dev_release(struct inode *node, struct file *f)
1067 {
1068 	struct mshv_vtl_hvcall_fd *fd;
1069 
1070 	fd = f->private_data;
1071 	if (fd) {
1072 		vfree(fd);
1073 		f->private_data = NULL;
1074 	}
1075 
1076 	return 0;
1077 }
1078 
mshv_vtl_hvcall_do_setup(struct mshv_vtl_hvcall_fd * fd,struct mshv_vtl_hvcall_setup __user * hvcall_setup_user)1079 static int mshv_vtl_hvcall_do_setup(struct mshv_vtl_hvcall_fd *fd,
1080 				    struct mshv_vtl_hvcall_setup __user *hvcall_setup_user)
1081 {
1082 	struct mshv_vtl_hvcall_setup hvcall_setup;
1083 
1084 	guard(mutex)(&fd->init_mutex);
1085 
1086 	if (fd->allow_map_initialized) {
1087 		dev_err(fd->dev->this_device,
1088 			"Hypercall allow map has already been set, pid %d\n",
1089 			current->pid);
1090 		return -EINVAL;
1091 	}
1092 
1093 	if (copy_from_user(&hvcall_setup, hvcall_setup_user,
1094 			   sizeof(struct mshv_vtl_hvcall_setup))) {
1095 		return -EFAULT;
1096 	}
1097 	if (hvcall_setup.bitmap_array_size > ARRAY_SIZE(fd->allow_bitmap))
1098 		return -EINVAL;
1099 
1100 	if (copy_from_user(&fd->allow_bitmap,
1101 			   (void __user *)hvcall_setup.allow_bitmap_ptr,
1102 			   hvcall_setup.bitmap_array_size)) {
1103 		return -EFAULT;
1104 	}
1105 
1106 	dev_info(fd->dev->this_device, "Hypercall allow map has been set, pid %d\n",
1107 		 current->pid);
1108 	fd->allow_map_initialized = true;
1109 	return 0;
1110 }
1111 
mshv_vtl_hvcall_is_allowed(struct mshv_vtl_hvcall_fd * fd,u16 call_code)1112 static bool mshv_vtl_hvcall_is_allowed(struct mshv_vtl_hvcall_fd *fd, u16 call_code)
1113 {
1114 	return test_bit(call_code, (unsigned long *)fd->allow_bitmap);
1115 }
1116 
mshv_vtl_hvcall_call(struct mshv_vtl_hvcall_fd * fd,struct mshv_vtl_hvcall __user * hvcall_user)1117 static int mshv_vtl_hvcall_call(struct mshv_vtl_hvcall_fd *fd,
1118 				struct mshv_vtl_hvcall __user *hvcall_user)
1119 {
1120 	struct mshv_vtl_hvcall hvcall;
1121 	void *in, *out;
1122 	int ret;
1123 
1124 	if (copy_from_user(&hvcall, hvcall_user, sizeof(struct mshv_vtl_hvcall)))
1125 		return -EFAULT;
1126 	if (hvcall.input_size > HV_HYP_PAGE_SIZE)
1127 		return -EINVAL;
1128 	if (hvcall.output_size > HV_HYP_PAGE_SIZE)
1129 		return -EINVAL;
1130 
1131 	/*
1132 	 * By default, all hypercalls are not allowed.
1133 	 * The user mode code has to set up the allow bitmap once.
1134 	 */
1135 
1136 	if (!mshv_vtl_hvcall_is_allowed(fd, hvcall.control & 0xFFFF)) {
1137 		dev_err(fd->dev->this_device,
1138 			"Hypercall with control data %#llx isn't allowed\n",
1139 			hvcall.control);
1140 		return -EPERM;
1141 	}
1142 
1143 	/*
1144 	 * This may create a problem for Confidential VM (CVM) usecase where we need to use
1145 	 * Hyper-V driver allocated per-cpu input and output pages (hyperv_pcpu_input_arg and
1146 	 * hyperv_pcpu_output_arg) for making a hypervisor call.
1147 	 *
1148 	 * TODO: Take care of this when CVM support is added.
1149 	 */
1150 	in = (void *)__get_free_page(GFP_KERNEL);
1151 	out = (void *)__get_free_page(GFP_KERNEL);
1152 	if (!in || !out) {
1153 		ret = -ENOMEM;
1154 		goto free_pages;
1155 	}
1156 
1157 	if (copy_from_user(in, (void __user *)hvcall.input_ptr, hvcall.input_size)) {
1158 		ret = -EFAULT;
1159 		goto free_pages;
1160 	}
1161 
1162 	/*
1163 	 * The caller supplies output_size, so clear the range copied back to
1164 	 * userspace in case the hypercall writes fewer bytes than requested.
1165 	 */
1166 	memset(out, 0, hvcall.output_size);
1167 
1168 	hvcall.status = hv_do_hypercall(hvcall.control, in, out);
1169 
1170 	if (copy_to_user((void __user *)hvcall.output_ptr, out, hvcall.output_size)) {
1171 		ret = -EFAULT;
1172 		goto free_pages;
1173 	}
1174 	ret = put_user(hvcall.status, &hvcall_user->status);
1175 free_pages:
1176 	free_page((unsigned long)in);
1177 	free_page((unsigned long)out);
1178 
1179 	return ret;
1180 }
1181 
mshv_vtl_hvcall_dev_ioctl(struct file * f,unsigned int cmd,unsigned long arg)1182 static long mshv_vtl_hvcall_dev_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1183 {
1184 	struct mshv_vtl_hvcall_fd *fd = f->private_data;
1185 
1186 	switch (cmd) {
1187 	case MSHV_HVCALL_SETUP:
1188 		return mshv_vtl_hvcall_do_setup(fd, (struct mshv_vtl_hvcall_setup __user *)arg);
1189 	case MSHV_HVCALL:
1190 		return mshv_vtl_hvcall_call(fd, (struct mshv_vtl_hvcall __user *)arg);
1191 	default:
1192 		break;
1193 	}
1194 
1195 	return -ENOIOCTLCMD;
1196 }
1197 
1198 static const struct file_operations mshv_vtl_hvcall_dev_file_ops = {
1199 	.owner = THIS_MODULE,
1200 	.open = mshv_vtl_hvcall_dev_open,
1201 	.release = mshv_vtl_hvcall_dev_release,
1202 	.unlocked_ioctl = mshv_vtl_hvcall_dev_ioctl,
1203 };
1204 
1205 static struct miscdevice mshv_vtl_hvcall_dev = {
1206 	.name = "mshv_hvcall",
1207 	.nodename = "mshv_hvcall",
1208 	.fops = &mshv_vtl_hvcall_dev_file_ops,
1209 	.mode = 0600,
1210 	.minor = MISC_DYNAMIC_MINOR,
1211 };
1212 
mshv_vtl_low_open(struct inode * inodep,struct file * filp)1213 static int mshv_vtl_low_open(struct inode *inodep, struct file *filp)
1214 {
1215 	pid_t pid = task_pid_vnr(current);
1216 	uid_t uid = current_uid().val;
1217 	int ret = 0;
1218 
1219 	pr_debug("%s: Opening VTL low, task group %d, uid %d\n", __func__, pid, uid);
1220 
1221 	if (capable(CAP_SYS_ADMIN)) {
1222 		filp->private_data = inodep;
1223 	} else {
1224 		pr_err("%s: VTL low open failed: CAP_SYS_ADMIN required. task group %d, uid %d",
1225 		       __func__, pid, uid);
1226 		ret = -EPERM;
1227 	}
1228 
1229 	return ret;
1230 }
1231 
can_fault(struct vm_fault * vmf,unsigned long size,unsigned long * pfn)1232 static bool can_fault(struct vm_fault *vmf, unsigned long size, unsigned long *pfn)
1233 {
1234 	unsigned long mask = size - 1;
1235 	unsigned long start = vmf->address & ~mask;
1236 	unsigned long end = start + size;
1237 	bool is_valid;
1238 
1239 	is_valid = (vmf->address & mask) == ((vmf->pgoff << PAGE_SHIFT) & mask) &&
1240 		start >= vmf->vma->vm_start &&
1241 		end <= vmf->vma->vm_end;
1242 
1243 	if (is_valid)
1244 		*pfn = vmf->pgoff & ~(mask >> PAGE_SHIFT);
1245 
1246 	return is_valid;
1247 }
1248 
mshv_vtl_low_huge_fault(struct vm_fault * vmf,unsigned int order)1249 static vm_fault_t mshv_vtl_low_huge_fault(struct vm_fault *vmf, unsigned int order)
1250 {
1251 	unsigned long pfn = vmf->pgoff;
1252 	vm_fault_t ret = VM_FAULT_FALLBACK;
1253 
1254 	switch (order) {
1255 	case 0:
1256 		return vmf_insert_mixed(vmf->vma, vmf->address, pfn);
1257 
1258 	case PMD_ORDER:
1259 		if (can_fault(vmf, PMD_SIZE, &pfn))
1260 			ret = vmf_insert_pfn_pmd(vmf, pfn, vmf->flags & FAULT_FLAG_WRITE);
1261 		return ret;
1262 
1263 	case PUD_ORDER:
1264 		if (can_fault(vmf, PUD_SIZE, &pfn))
1265 			ret = vmf_insert_pfn_pud(vmf, pfn, vmf->flags & FAULT_FLAG_WRITE);
1266 		return ret;
1267 
1268 	default:
1269 		return VM_FAULT_SIGBUS;
1270 	}
1271 }
1272 
mshv_vtl_low_fault(struct vm_fault * vmf)1273 static vm_fault_t mshv_vtl_low_fault(struct vm_fault *vmf)
1274 {
1275 	return mshv_vtl_low_huge_fault(vmf, 0);
1276 }
1277 
1278 static const struct vm_operations_struct mshv_vtl_low_vm_ops = {
1279 	.fault = mshv_vtl_low_fault,
1280 	.huge_fault = mshv_vtl_low_huge_fault,
1281 };
1282 
mshv_vtl_low_mmap(struct file * filp,struct vm_area_struct * vma)1283 static int mshv_vtl_low_mmap(struct file *filp, struct vm_area_struct *vma)
1284 {
1285 	vma->vm_ops = &mshv_vtl_low_vm_ops;
1286 	vm_flags_set(vma, VM_HUGEPAGE | VM_MIXEDMAP);
1287 
1288 	return 0;
1289 }
1290 
1291 static const struct file_operations mshv_vtl_low_file_ops = {
1292 	.owner		= THIS_MODULE,
1293 	.open		= mshv_vtl_low_open,
1294 	.mmap		= mshv_vtl_low_mmap,
1295 };
1296 
1297 static struct miscdevice mshv_vtl_low = {
1298 	.name = "mshv_vtl_low",
1299 	.nodename = "mshv_vtl_low",
1300 	.fops = &mshv_vtl_low_file_ops,
1301 	.mode = 0600,
1302 	.minor = MISC_DYNAMIC_MINOR,
1303 };
1304 
mshv_vtl_init(void)1305 static int __init mshv_vtl_init(void)
1306 {
1307 	int ret;
1308 	struct device *dev = mshv_dev.this_device;
1309 
1310 	/*
1311 	 * This creates /dev/mshv which provides functionality to create VTLs and partitions.
1312 	 */
1313 	ret = misc_register(&mshv_dev);
1314 	if (ret) {
1315 		dev_err(dev, "mshv device register failed: %d\n", ret);
1316 		goto free_dev;
1317 	}
1318 
1319 	tasklet_init(&msg_dpc, mshv_vtl_sint_on_msg_dpc, 0);
1320 	init_waitqueue_head(&fd_wait_queue);
1321 
1322 	if (mshv_vtl_get_vsm_regs()) {
1323 		dev_emerg(dev, "Unable to get VSM capabilities !!\n");
1324 		ret = -ENODEV;
1325 		goto free_dev;
1326 	}
1327 	if (mshv_vtl_configure_vsm_partition(dev)) {
1328 		dev_emerg(dev, "VSM configuration failed !!\n");
1329 		ret = -ENODEV;
1330 		goto free_dev;
1331 	}
1332 
1333 	mshv_vtl_return_call_init(mshv_vsm_page_offsets.vtl_return_offset);
1334 	ret = hv_vtl_setup_synic();
1335 	if (ret)
1336 		goto free_dev;
1337 
1338 	/*
1339 	 * mshv_sint device adds VMBus relay ioctl support.
1340 	 * This provides a channel for VTL0 to communicate with VTL2.
1341 	 */
1342 	ret = misc_register(&mshv_vtl_sint_dev);
1343 	if (ret)
1344 		goto free_synic;
1345 
1346 	/*
1347 	 * mshv_hvcall device adds interface to enable userspace for direct hypercalls support.
1348 	 */
1349 	ret = misc_register(&mshv_vtl_hvcall_dev);
1350 	if (ret)
1351 		goto free_sint;
1352 
1353 	/*
1354 	 * mshv_vtl_low device is used to map VTL0 address space to a user-mode process in VTL2.
1355 	 * It implements mmap() to allow a user-mode process in VTL2 to map to the address of VTL0.
1356 	 */
1357 	ret = misc_register(&mshv_vtl_low);
1358 	if (ret)
1359 		goto free_hvcall;
1360 
1361 	/*
1362 	 * "mshv vtl mem dev" device is later used to setup VTL0 memory.
1363 	 */
1364 	mem_dev = kzalloc_obj(*mem_dev);
1365 	if (!mem_dev) {
1366 		ret = -ENOMEM;
1367 		goto free_low;
1368 	}
1369 
1370 	mutex_init(&mshv_vtl_poll_file_lock);
1371 
1372 	device_initialize(mem_dev);
1373 	dev_set_name(mem_dev, "mshv vtl mem dev");
1374 	ret = device_add(mem_dev);
1375 	if (ret) {
1376 		dev_err(dev, "mshv vtl mem dev add: %d\n", ret);
1377 		goto free_mem;
1378 	}
1379 
1380 	return 0;
1381 
1382 free_mem:
1383 	kfree(mem_dev);
1384 free_low:
1385 	misc_deregister(&mshv_vtl_low);
1386 free_hvcall:
1387 	misc_deregister(&mshv_vtl_hvcall_dev);
1388 free_sint:
1389 	misc_deregister(&mshv_vtl_sint_dev);
1390 free_synic:
1391 	hv_vtl_remove_synic();
1392 free_dev:
1393 	misc_deregister(&mshv_dev);
1394 
1395 	return ret;
1396 }
1397 
mshv_vtl_exit(void)1398 static void __exit mshv_vtl_exit(void)
1399 {
1400 	device_del(mem_dev);
1401 	kfree(mem_dev);
1402 	misc_deregister(&mshv_vtl_low);
1403 	misc_deregister(&mshv_vtl_hvcall_dev);
1404 	misc_deregister(&mshv_vtl_sint_dev);
1405 	hv_vtl_remove_synic();
1406 	misc_deregister(&mshv_dev);
1407 }
1408 
1409 module_init(mshv_vtl_init);
1410 module_exit(mshv_vtl_exit);
1411