xref: /linux/arch/x86/hyperv/ivm.c (revision 9591fdb0611dccdeeeeacb99d89f0098737d209b)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Hyper-V Isolation VM interface with paravisor and hypervisor
4  *
5  * Author:
6  *  Tianyu Lan <Tianyu.Lan@microsoft.com>
7  */
8 
9 #include <linux/bitfield.h>
10 #include <linux/types.h>
11 #include <linux/slab.h>
12 #include <linux/cpu.h>
13 #include <linux/export.h>
14 #include <asm/svm.h>
15 #include <asm/sev.h>
16 #include <asm/io.h>
17 #include <asm/coco.h>
18 #include <asm/mem_encrypt.h>
19 #include <asm/set_memory.h>
20 #include <asm/mshyperv.h>
21 #include <asm/hypervisor.h>
22 #include <asm/mtrr.h>
23 #include <asm/io_apic.h>
24 #include <asm/realmode.h>
25 #include <asm/e820/api.h>
26 #include <asm/desc.h>
27 #include <asm/msr.h>
28 #include <uapi/asm/vmx.h>
29 
30 #ifdef CONFIG_AMD_MEM_ENCRYPT
31 
32 #define GHCB_USAGE_HYPERV_CALL	1
33 
34 union hv_ghcb {
35 	struct ghcb ghcb;
36 	struct {
37 		u64 hypercalldata[509];
38 		u64 outputgpa;
39 		union {
40 			union {
41 				struct {
42 					u32 callcode        : 16;
43 					u32 isfast          : 1;
44 					u32 reserved1       : 14;
45 					u32 isnested        : 1;
46 					u32 countofelements : 12;
47 					u32 reserved2       : 4;
48 					u32 repstartindex   : 12;
49 					u32 reserved3       : 4;
50 				};
51 				u64 asuint64;
52 			} hypercallinput;
53 			union {
54 				struct {
55 					u16 callstatus;
56 					u16 reserved1;
57 					u32 elementsprocessed : 12;
58 					u32 reserved2         : 20;
59 				};
60 				u64 asunit64;
61 			} hypercalloutput;
62 		};
63 		u64 reserved2;
64 	} hypercall;
65 } __packed __aligned(HV_HYP_PAGE_SIZE);
66 
67 /* Only used in an SNP VM with the paravisor */
68 static u16 hv_ghcb_version __ro_after_init;
69 
70 /* Functions only used in an SNP VM with the paravisor go here. */
hv_ghcb_hypercall(u64 control,void * input,void * output,u32 input_size)71 u64 hv_ghcb_hypercall(u64 control, void *input, void *output, u32 input_size)
72 {
73 	union hv_ghcb *hv_ghcb;
74 	void **ghcb_base;
75 	unsigned long flags;
76 	u64 status;
77 
78 	if (!hv_ghcb_pg)
79 		return -EFAULT;
80 
81 	WARN_ON(in_nmi());
82 
83 	local_irq_save(flags);
84 	ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg);
85 	hv_ghcb = (union hv_ghcb *)*ghcb_base;
86 	if (!hv_ghcb) {
87 		local_irq_restore(flags);
88 		return -EFAULT;
89 	}
90 
91 	hv_ghcb->ghcb.protocol_version = GHCB_PROTOCOL_MAX;
92 	hv_ghcb->ghcb.ghcb_usage = GHCB_USAGE_HYPERV_CALL;
93 
94 	hv_ghcb->hypercall.outputgpa = (u64)output;
95 	hv_ghcb->hypercall.hypercallinput.asuint64 = 0;
96 	hv_ghcb->hypercall.hypercallinput.callcode = control;
97 
98 	if (input_size)
99 		memcpy(hv_ghcb->hypercall.hypercalldata, input, input_size);
100 
101 	VMGEXIT();
102 
103 	hv_ghcb->ghcb.ghcb_usage = 0xffffffff;
104 	memset(hv_ghcb->ghcb.save.valid_bitmap, 0,
105 	       sizeof(hv_ghcb->ghcb.save.valid_bitmap));
106 
107 	status = hv_ghcb->hypercall.hypercalloutput.callstatus;
108 
109 	local_irq_restore(flags);
110 
111 	return status;
112 }
113 
rd_ghcb_msr(void)114 static inline u64 rd_ghcb_msr(void)
115 {
116 	return native_rdmsrq(MSR_AMD64_SEV_ES_GHCB);
117 }
118 
wr_ghcb_msr(u64 val)119 static inline void wr_ghcb_msr(u64 val)
120 {
121 	native_wrmsrq(MSR_AMD64_SEV_ES_GHCB, val);
122 }
123 
hv_ghcb_hv_call(struct ghcb * ghcb,u64 exit_code,u64 exit_info_1,u64 exit_info_2)124 static enum es_result hv_ghcb_hv_call(struct ghcb *ghcb, u64 exit_code,
125 				   u64 exit_info_1, u64 exit_info_2)
126 {
127 	/* Fill in protocol and format specifiers */
128 	ghcb->protocol_version = hv_ghcb_version;
129 	ghcb->ghcb_usage       = GHCB_DEFAULT_USAGE;
130 
131 	ghcb_set_sw_exit_code(ghcb, exit_code);
132 	ghcb_set_sw_exit_info_1(ghcb, exit_info_1);
133 	ghcb_set_sw_exit_info_2(ghcb, exit_info_2);
134 
135 	VMGEXIT();
136 
137 	if (ghcb->save.sw_exit_info_1 & GENMASK_ULL(31, 0))
138 		return ES_VMM_ERROR;
139 	else
140 		return ES_OK;
141 }
142 
hv_ghcb_terminate(unsigned int set,unsigned int reason)143 void __noreturn hv_ghcb_terminate(unsigned int set, unsigned int reason)
144 {
145 	u64 val = GHCB_MSR_TERM_REQ;
146 
147 	/* Tell the hypervisor what went wrong. */
148 	val |= GHCB_SEV_TERM_REASON(set, reason);
149 
150 	/* Request Guest Termination from Hypervisor */
151 	wr_ghcb_msr(val);
152 	VMGEXIT();
153 
154 	while (true)
155 		asm volatile("hlt\n" : : : "memory");
156 }
157 
hv_ghcb_negotiate_protocol(void)158 bool hv_ghcb_negotiate_protocol(void)
159 {
160 	u64 ghcb_gpa;
161 	u64 val;
162 
163 	/* Save ghcb page gpa. */
164 	ghcb_gpa = rd_ghcb_msr();
165 
166 	/* Do the GHCB protocol version negotiation */
167 	wr_ghcb_msr(GHCB_MSR_SEV_INFO_REQ);
168 	VMGEXIT();
169 	val = rd_ghcb_msr();
170 
171 	if (GHCB_MSR_INFO(val) != GHCB_MSR_SEV_INFO_RESP)
172 		return false;
173 
174 	if (GHCB_MSR_PROTO_MAX(val) < GHCB_PROTOCOL_MIN ||
175 	    GHCB_MSR_PROTO_MIN(val) > GHCB_PROTOCOL_MAX)
176 		return false;
177 
178 	hv_ghcb_version = min_t(size_t, GHCB_MSR_PROTO_MAX(val),
179 			     GHCB_PROTOCOL_MAX);
180 
181 	/* Write ghcb page back after negotiating protocol. */
182 	wr_ghcb_msr(ghcb_gpa);
183 	VMGEXIT();
184 
185 	return true;
186 }
187 
hv_ghcb_msr_write(u64 msr,u64 value)188 static void hv_ghcb_msr_write(u64 msr, u64 value)
189 {
190 	union hv_ghcb *hv_ghcb;
191 	void **ghcb_base;
192 	unsigned long flags;
193 
194 	if (!hv_ghcb_pg)
195 		return;
196 
197 	WARN_ON(in_nmi());
198 
199 	local_irq_save(flags);
200 	ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg);
201 	hv_ghcb = (union hv_ghcb *)*ghcb_base;
202 	if (!hv_ghcb) {
203 		local_irq_restore(flags);
204 		return;
205 	}
206 
207 	ghcb_set_rcx(&hv_ghcb->ghcb, msr);
208 	ghcb_set_rax(&hv_ghcb->ghcb, lower_32_bits(value));
209 	ghcb_set_rdx(&hv_ghcb->ghcb, upper_32_bits(value));
210 
211 	if (hv_ghcb_hv_call(&hv_ghcb->ghcb, SVM_EXIT_MSR, 1, 0))
212 		pr_warn("Fail to write msr via ghcb %llx.\n", msr);
213 
214 	local_irq_restore(flags);
215 }
216 
hv_ghcb_msr_read(u64 msr,u64 * value)217 static void hv_ghcb_msr_read(u64 msr, u64 *value)
218 {
219 	union hv_ghcb *hv_ghcb;
220 	void **ghcb_base;
221 	unsigned long flags;
222 
223 	/* Check size of union hv_ghcb here. */
224 	BUILD_BUG_ON(sizeof(union hv_ghcb) != HV_HYP_PAGE_SIZE);
225 
226 	if (!hv_ghcb_pg)
227 		return;
228 
229 	WARN_ON(in_nmi());
230 
231 	local_irq_save(flags);
232 	ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg);
233 	hv_ghcb = (union hv_ghcb *)*ghcb_base;
234 	if (!hv_ghcb) {
235 		local_irq_restore(flags);
236 		return;
237 	}
238 
239 	ghcb_set_rcx(&hv_ghcb->ghcb, msr);
240 	if (hv_ghcb_hv_call(&hv_ghcb->ghcb, SVM_EXIT_MSR, 0, 0))
241 		pr_warn("Fail to read msr via ghcb %llx.\n", msr);
242 	else
243 		*value = (u64)lower_32_bits(hv_ghcb->ghcb.save.rax)
244 			| ((u64)lower_32_bits(hv_ghcb->ghcb.save.rdx) << 32);
245 	local_irq_restore(flags);
246 }
247 
248 /* Only used in a fully enlightened SNP VM, i.e. without the paravisor */
249 static u8 ap_start_input_arg[PAGE_SIZE] __bss_decrypted __aligned(PAGE_SIZE);
250 static u8 ap_start_stack[PAGE_SIZE] __aligned(PAGE_SIZE);
251 static DEFINE_PER_CPU(struct sev_es_save_area *, hv_sev_vmsa);
252 
253 /* Functions only used in an SNP VM without the paravisor go here. */
254 
255 #define hv_populate_vmcb_seg(seg, gdtr_base)			\
256 do {								\
257 	if (seg.selector) {					\
258 		seg.base = 0;					\
259 		seg.limit = HV_AP_SEGMENT_LIMIT;		\
260 		seg.attrib = *(u16 *)(gdtr_base + seg.selector + 5);	\
261 		seg.attrib = (seg.attrib & 0xFF) | ((seg.attrib >> 4) & 0xF00); \
262 	}							\
263 } while (0)							\
264 
snp_set_vmsa(void * va,bool vmsa)265 static int snp_set_vmsa(void *va, bool vmsa)
266 {
267 	u64 attrs;
268 
269 	/*
270 	 * Running at VMPL0 allows the kernel to change the VMSA bit for a page
271 	 * using the RMPADJUST instruction. However, for the instruction to
272 	 * succeed it must target the permissions of a lesser privileged
273 	 * (higher numbered) VMPL level, so use VMPL1 (refer to the RMPADJUST
274 	 * instruction in the AMD64 APM Volume 3).
275 	 */
276 	attrs = 1;
277 	if (vmsa)
278 		attrs |= RMPADJUST_VMSA_PAGE_BIT;
279 
280 	return rmpadjust((unsigned long)va, RMP_PG_SIZE_4K, attrs);
281 }
282 
snp_cleanup_vmsa(struct sev_es_save_area * vmsa)283 static void snp_cleanup_vmsa(struct sev_es_save_area *vmsa)
284 {
285 	int err;
286 
287 	err = snp_set_vmsa(vmsa, false);
288 	if (err)
289 		pr_err("clear VMSA page failed (%u), leaking page\n", err);
290 	else
291 		free_page((unsigned long)vmsa);
292 }
293 
hv_snp_boot_ap(u32 apic_id,unsigned long start_ip,unsigned int cpu)294 int hv_snp_boot_ap(u32 apic_id, unsigned long start_ip, unsigned int cpu)
295 {
296 	struct sev_es_save_area *vmsa = (struct sev_es_save_area *)
297 		__get_free_page(GFP_KERNEL | __GFP_ZERO);
298 	struct sev_es_save_area *cur_vmsa;
299 	struct desc_ptr gdtr;
300 	u64 ret, retry = 5;
301 	struct hv_enable_vp_vtl *start_vp_input;
302 	unsigned long flags;
303 	int vp_index;
304 
305 	if (!vmsa)
306 		return -ENOMEM;
307 
308 	/* Find the Hyper-V VP index which might be not the same as APIC ID */
309 	vp_index = hv_apicid_to_vp_index(apic_id);
310 	if (vp_index < 0 || vp_index > ms_hyperv.max_vp_index)
311 		return -EINVAL;
312 
313 	native_store_gdt(&gdtr);
314 
315 	vmsa->gdtr.base = gdtr.address;
316 	vmsa->gdtr.limit = gdtr.size;
317 
318 	asm volatile("movl %%es, %%eax;" : "=a" (vmsa->es.selector));
319 	hv_populate_vmcb_seg(vmsa->es, vmsa->gdtr.base);
320 
321 	asm volatile("movl %%cs, %%eax;" : "=a" (vmsa->cs.selector));
322 	hv_populate_vmcb_seg(vmsa->cs, vmsa->gdtr.base);
323 
324 	asm volatile("movl %%ss, %%eax;" : "=a" (vmsa->ss.selector));
325 	hv_populate_vmcb_seg(vmsa->ss, vmsa->gdtr.base);
326 
327 	asm volatile("movl %%ds, %%eax;" : "=a" (vmsa->ds.selector));
328 	hv_populate_vmcb_seg(vmsa->ds, vmsa->gdtr.base);
329 
330 	vmsa->efer = native_read_msr(MSR_EFER);
331 
332 	vmsa->cr4 = native_read_cr4();
333 	vmsa->cr3 = __native_read_cr3();
334 	vmsa->cr0 = native_read_cr0();
335 
336 	vmsa->xcr0 = 1;
337 	vmsa->g_pat = HV_AP_INIT_GPAT_DEFAULT;
338 	vmsa->rip = (u64)secondary_startup_64_no_verify;
339 	vmsa->rsp = (u64)&ap_start_stack[PAGE_SIZE];
340 
341 	/*
342 	 * Set the SNP-specific fields for this VMSA:
343 	 *   VMPL level
344 	 *   SEV_FEATURES (matches the SEV STATUS MSR right shifted 2 bits)
345 	 */
346 	vmsa->vmpl = 0;
347 	vmsa->sev_features = sev_status >> 2;
348 
349 	ret = snp_set_vmsa(vmsa, true);
350 	if (ret) {
351 		pr_err("RMPADJUST(%llx) failed: %llx\n", (u64)vmsa, ret);
352 		free_page((u64)vmsa);
353 		return ret;
354 	}
355 
356 	local_irq_save(flags);
357 	start_vp_input = (struct hv_enable_vp_vtl *)ap_start_input_arg;
358 	memset(start_vp_input, 0, sizeof(*start_vp_input));
359 	start_vp_input->partition_id = -1;
360 	start_vp_input->vp_index = vp_index;
361 	start_vp_input->target_vtl.target_vtl = ms_hyperv.vtl;
362 	*(u64 *)&start_vp_input->vp_context = __pa(vmsa) | 1;
363 
364 	do {
365 		ret = hv_do_hypercall(HVCALL_START_VP,
366 				      start_vp_input, NULL);
367 	} while (hv_result(ret) == HV_STATUS_TIME_OUT && retry--);
368 
369 	local_irq_restore(flags);
370 
371 	if (!hv_result_success(ret)) {
372 		pr_err("HvCallStartVirtualProcessor failed: %llx\n", ret);
373 		snp_cleanup_vmsa(vmsa);
374 		vmsa = NULL;
375 	}
376 
377 	cur_vmsa = per_cpu(hv_sev_vmsa, cpu);
378 	/* Free up any previous VMSA page */
379 	if (cur_vmsa)
380 		snp_cleanup_vmsa(cur_vmsa);
381 
382 	/* Record the current VMSA page */
383 	per_cpu(hv_sev_vmsa, cpu) = vmsa;
384 
385 	return ret;
386 }
387 
hv_snp_hypercall(u64 control,u64 param1,u64 param2)388 u64 hv_snp_hypercall(u64 control, u64 param1, u64 param2)
389 {
390 	u64 hv_status;
391 
392 	register u64 __r8 asm("r8") = param2;
393 	asm volatile("vmmcall"
394 		     : "=a" (hv_status), ASM_CALL_CONSTRAINT,
395 		       "+c" (control), "+d" (param1), "+r" (__r8)
396 		     : : "cc", "memory", "r9", "r10", "r11");
397 
398 	return hv_status;
399 }
400 
401 #else
hv_ghcb_msr_write(u64 msr,u64 value)402 static inline void hv_ghcb_msr_write(u64 msr, u64 value) {}
hv_ghcb_msr_read(u64 msr,u64 * value)403 static inline void hv_ghcb_msr_read(u64 msr, u64 *value) {}
hv_snp_hypercall(u64 control,u64 param1,u64 param2)404 u64 hv_snp_hypercall(u64 control, u64 param1, u64 param2) { return U64_MAX; }
405 #endif /* CONFIG_AMD_MEM_ENCRYPT */
406 
407 #ifdef CONFIG_INTEL_TDX_GUEST
hv_tdx_msr_write(u64 msr,u64 val)408 static void hv_tdx_msr_write(u64 msr, u64 val)
409 {
410 	struct tdx_module_args args = {
411 		.r10 = TDX_HYPERCALL_STANDARD,
412 		.r11 = EXIT_REASON_MSR_WRITE,
413 		.r12 = msr,
414 		.r13 = val,
415 	};
416 
417 	u64 ret = __tdx_hypercall(&args);
418 
419 	WARN_ONCE(ret, "Failed to emulate MSR write: %lld\n", ret);
420 }
421 
hv_tdx_msr_read(u64 msr,u64 * val)422 static void hv_tdx_msr_read(u64 msr, u64 *val)
423 {
424 	struct tdx_module_args args = {
425 		.r10 = TDX_HYPERCALL_STANDARD,
426 		.r11 = EXIT_REASON_MSR_READ,
427 		.r12 = msr,
428 	};
429 
430 	u64 ret = __tdx_hypercall(&args);
431 
432 	if (WARN_ONCE(ret, "Failed to emulate MSR read: %lld\n", ret))
433 		*val = 0;
434 	else
435 		*val = args.r11;
436 }
437 
hv_tdx_hypercall(u64 control,u64 param1,u64 param2)438 u64 hv_tdx_hypercall(u64 control, u64 param1, u64 param2)
439 {
440 	struct tdx_module_args args = { };
441 
442 	args.r10 = control;
443 	args.rdx = param1;
444 	args.r8  = param2;
445 
446 	(void)__tdx_hypercall(&args);
447 
448 	return args.r11;
449 }
450 
451 #else
hv_tdx_msr_write(u64 msr,u64 value)452 static inline void hv_tdx_msr_write(u64 msr, u64 value) {}
hv_tdx_msr_read(u64 msr,u64 * value)453 static inline void hv_tdx_msr_read(u64 msr, u64 *value) {}
hv_tdx_hypercall(u64 control,u64 param1,u64 param2)454 u64 hv_tdx_hypercall(u64 control, u64 param1, u64 param2) { return U64_MAX; }
455 #endif /* CONFIG_INTEL_TDX_GUEST */
456 
457 #if defined(CONFIG_AMD_MEM_ENCRYPT) || defined(CONFIG_INTEL_TDX_GUEST)
hv_ivm_msr_write(u64 msr,u64 value)458 void hv_ivm_msr_write(u64 msr, u64 value)
459 {
460 	if (!ms_hyperv.paravisor_present)
461 		return;
462 
463 	if (hv_isolation_type_tdx())
464 		hv_tdx_msr_write(msr, value);
465 	else if (hv_isolation_type_snp())
466 		hv_ghcb_msr_write(msr, value);
467 }
468 
hv_ivm_msr_read(u64 msr,u64 * value)469 void hv_ivm_msr_read(u64 msr, u64 *value)
470 {
471 	if (!ms_hyperv.paravisor_present)
472 		return;
473 
474 	if (hv_isolation_type_tdx())
475 		hv_tdx_msr_read(msr, value);
476 	else if (hv_isolation_type_snp())
477 		hv_ghcb_msr_read(msr, value);
478 }
479 
480 /*
481  * Keep track of the PFN regions which were shared with the host. The access
482  * must be revoked upon kexec/kdump (see hv_ivm_clear_host_access()).
483  */
484 struct hv_enc_pfn_region {
485 	struct list_head list;
486 	u64 pfn;
487 	int count;
488 };
489 
490 static LIST_HEAD(hv_list_enc);
491 static DEFINE_RAW_SPINLOCK(hv_list_enc_lock);
492 
hv_list_enc_add(const u64 * pfn_list,int count)493 static int hv_list_enc_add(const u64 *pfn_list, int count)
494 {
495 	struct hv_enc_pfn_region *ent;
496 	unsigned long flags;
497 	u64 pfn;
498 	int i;
499 
500 	for (i = 0; i < count; i++) {
501 		pfn = pfn_list[i];
502 
503 		raw_spin_lock_irqsave(&hv_list_enc_lock, flags);
504 		/* Check if the PFN already exists in some region first */
505 		list_for_each_entry(ent, &hv_list_enc, list) {
506 			if ((ent->pfn <= pfn) && (ent->pfn + ent->count - 1 >= pfn))
507 				/* Nothing to do - pfn is already in the list */
508 				goto unlock_done;
509 		}
510 
511 		/*
512 		 * Check if the PFN is adjacent to an existing region. Growing
513 		 * a region can make it adjacent to another one but merging is
514 		 * not (yet) implemented for simplicity. A PFN cannot be added
515 		 * to two regions to keep the logic in hv_list_enc_remove()
516 		 * correct.
517 		 */
518 		list_for_each_entry(ent, &hv_list_enc, list) {
519 			if (ent->pfn + ent->count == pfn) {
520 				/* Grow existing region up */
521 				ent->count++;
522 				goto unlock_done;
523 			} else if (pfn + 1 == ent->pfn) {
524 				/* Grow existing region down */
525 				ent->pfn--;
526 				ent->count++;
527 				goto unlock_done;
528 			}
529 		}
530 		raw_spin_unlock_irqrestore(&hv_list_enc_lock, flags);
531 
532 		/* No adjacent region found -- create a new one */
533 		ent = kzalloc(sizeof(struct hv_enc_pfn_region), GFP_KERNEL);
534 		if (!ent)
535 			return -ENOMEM;
536 
537 		ent->pfn = pfn;
538 		ent->count = 1;
539 
540 		raw_spin_lock_irqsave(&hv_list_enc_lock, flags);
541 		list_add(&ent->list, &hv_list_enc);
542 
543 unlock_done:
544 		raw_spin_unlock_irqrestore(&hv_list_enc_lock, flags);
545 	}
546 
547 	return 0;
548 }
549 
hv_list_enc_remove(const u64 * pfn_list,int count)550 static int hv_list_enc_remove(const u64 *pfn_list, int count)
551 {
552 	struct hv_enc_pfn_region *ent, *t;
553 	struct hv_enc_pfn_region new_region;
554 	unsigned long flags;
555 	u64 pfn;
556 	int i;
557 
558 	for (i = 0; i < count; i++) {
559 		pfn = pfn_list[i];
560 
561 		raw_spin_lock_irqsave(&hv_list_enc_lock, flags);
562 		list_for_each_entry_safe(ent, t, &hv_list_enc, list) {
563 			if (pfn == ent->pfn + ent->count - 1) {
564 				/* Removing tail pfn */
565 				ent->count--;
566 				if (!ent->count) {
567 					list_del(&ent->list);
568 					kfree(ent);
569 				}
570 				goto unlock_done;
571 			} else if (pfn == ent->pfn) {
572 				/* Removing head pfn */
573 				ent->count--;
574 				ent->pfn++;
575 				if (!ent->count) {
576 					list_del(&ent->list);
577 					kfree(ent);
578 				}
579 				goto unlock_done;
580 			} else if (pfn > ent->pfn && pfn < ent->pfn + ent->count - 1) {
581 				/*
582 				 * Removing a pfn in the middle. Cut off the tail
583 				 * of the existing region and create a template for
584 				 * the new one.
585 				 */
586 				new_region.pfn = pfn + 1;
587 				new_region.count = ent->count - (pfn - ent->pfn + 1);
588 				ent->count = pfn - ent->pfn;
589 				goto unlock_split;
590 			}
591 
592 		}
593 unlock_done:
594 		raw_spin_unlock_irqrestore(&hv_list_enc_lock, flags);
595 		continue;
596 
597 unlock_split:
598 		raw_spin_unlock_irqrestore(&hv_list_enc_lock, flags);
599 
600 		ent = kzalloc(sizeof(struct hv_enc_pfn_region), GFP_KERNEL);
601 		if (!ent)
602 			return -ENOMEM;
603 
604 		ent->pfn = new_region.pfn;
605 		ent->count = new_region.count;
606 
607 		raw_spin_lock_irqsave(&hv_list_enc_lock, flags);
608 		list_add(&ent->list, &hv_list_enc);
609 		raw_spin_unlock_irqrestore(&hv_list_enc_lock, flags);
610 	}
611 
612 	return 0;
613 }
614 
615 /* Stop new private<->shared conversions */
hv_vtom_kexec_begin(void)616 static void hv_vtom_kexec_begin(void)
617 {
618 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
619 		return;
620 
621 	/*
622 	 * Crash kernel reaches here with interrupts disabled: can't wait for
623 	 * conversions to finish.
624 	 *
625 	 * If race happened, just report and proceed.
626 	 */
627 	if (!set_memory_enc_stop_conversion())
628 		pr_warn("Failed to stop shared<->private conversions\n");
629 }
630 
hv_vtom_kexec_finish(void)631 static void hv_vtom_kexec_finish(void)
632 {
633 	struct hv_gpa_range_for_visibility *input;
634 	struct hv_enc_pfn_region *ent;
635 	unsigned long flags;
636 	u64 hv_status;
637 	int cur, i;
638 
639 	local_irq_save(flags);
640 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
641 
642 	if (unlikely(!input))
643 		goto out;
644 
645 	list_for_each_entry(ent, &hv_list_enc, list) {
646 		for (i = 0, cur = 0; i < ent->count; i++) {
647 			input->gpa_page_list[cur] = ent->pfn + i;
648 			cur++;
649 
650 			if (cur == HV_MAX_MODIFY_GPA_REP_COUNT || i == ent->count - 1) {
651 				input->partition_id = HV_PARTITION_ID_SELF;
652 				input->host_visibility = VMBUS_PAGE_NOT_VISIBLE;
653 				input->reserved0 = 0;
654 				input->reserved1 = 0;
655 				hv_status = hv_do_rep_hypercall(
656 					HVCALL_MODIFY_SPARSE_GPA_PAGE_HOST_VISIBILITY,
657 					cur, 0, input, NULL);
658 				WARN_ON_ONCE(!hv_result_success(hv_status));
659 				cur = 0;
660 			}
661 		}
662 
663 	}
664 
665 out:
666 	local_irq_restore(flags);
667 }
668 
669 /*
670  * hv_mark_gpa_visibility - Set pages visible to host via hvcall.
671  *
672  * In Isolation VM, all guest memory is encrypted from host and guest
673  * needs to set memory visible to host via hvcall before sharing memory
674  * with host.
675  */
hv_mark_gpa_visibility(u16 count,const u64 pfn[],enum hv_mem_host_visibility visibility)676 static int hv_mark_gpa_visibility(u16 count, const u64 pfn[],
677 			   enum hv_mem_host_visibility visibility)
678 {
679 	struct hv_gpa_range_for_visibility *input;
680 	u64 hv_status;
681 	unsigned long flags;
682 	int ret;
683 
684 	/* no-op if partition isolation is not enabled */
685 	if (!hv_is_isolation_supported())
686 		return 0;
687 
688 	if (count > HV_MAX_MODIFY_GPA_REP_COUNT) {
689 		pr_err("Hyper-V: GPA count:%d exceeds supported:%lu\n", count,
690 			HV_MAX_MODIFY_GPA_REP_COUNT);
691 		return -EINVAL;
692 	}
693 
694 	if (visibility == VMBUS_PAGE_NOT_VISIBLE)
695 		ret = hv_list_enc_remove(pfn, count);
696 	else
697 		ret = hv_list_enc_add(pfn, count);
698 	if (ret)
699 		return ret;
700 
701 	local_irq_save(flags);
702 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
703 
704 	if (unlikely(!input)) {
705 		local_irq_restore(flags);
706 		return -EINVAL;
707 	}
708 
709 	input->partition_id = HV_PARTITION_ID_SELF;
710 	input->host_visibility = visibility;
711 	input->reserved0 = 0;
712 	input->reserved1 = 0;
713 	memcpy((void *)input->gpa_page_list, pfn, count * sizeof(*pfn));
714 	hv_status = hv_do_rep_hypercall(
715 			HVCALL_MODIFY_SPARSE_GPA_PAGE_HOST_VISIBILITY, count,
716 			0, input, NULL);
717 	local_irq_restore(flags);
718 
719 	if (hv_result_success(hv_status))
720 		return 0;
721 
722 	if (visibility == VMBUS_PAGE_NOT_VISIBLE)
723 		ret = hv_list_enc_add(pfn, count);
724 	else
725 		ret = hv_list_enc_remove(pfn, count);
726 	/*
727 	 * There's no good way to recover from -ENOMEM here, the accounting is
728 	 * wrong either way.
729 	 */
730 	WARN_ON_ONCE(ret);
731 
732 	return -EFAULT;
733 }
734 
735 /*
736  * When transitioning memory between encrypted and decrypted, the caller
737  * of set_memory_encrypted() or set_memory_decrypted() is responsible for
738  * ensuring that the memory isn't in use and isn't referenced while the
739  * transition is in progress.  The transition has multiple steps, and the
740  * memory is in an inconsistent state until all steps are complete. A
741  * reference while the state is inconsistent could result in an exception
742  * that can't be cleanly fixed up.
743  *
744  * But the Linux kernel load_unaligned_zeropad() mechanism could cause a
745  * stray reference that can't be prevented by the caller, so Linux has
746  * specific code to handle this case. But when the #VC and #VE exceptions
747  * routed to a paravisor, the specific code doesn't work. To avoid this
748  * problem, mark the pages as "not present" while the transition is in
749  * progress. If load_unaligned_zeropad() causes a stray reference, a normal
750  * page fault is generated instead of #VC or #VE, and the page-fault-based
751  * handlers for load_unaligned_zeropad() resolve the reference.  When the
752  * transition is complete, hv_vtom_set_host_visibility() marks the pages
753  * as "present" again.
754  */
hv_vtom_clear_present(unsigned long kbuffer,int pagecount,bool enc)755 static int hv_vtom_clear_present(unsigned long kbuffer, int pagecount, bool enc)
756 {
757 	return set_memory_np(kbuffer, pagecount);
758 }
759 
760 /*
761  * hv_vtom_set_host_visibility - Set specified memory visible to host.
762  *
763  * In Isolation VM, all guest memory is encrypted from host and guest
764  * needs to set memory visible to host via hvcall before sharing memory
765  * with host. This function works as wrap of hv_mark_gpa_visibility()
766  * with memory base and size.
767  */
hv_vtom_set_host_visibility(unsigned long kbuffer,int pagecount,bool enc)768 static int hv_vtom_set_host_visibility(unsigned long kbuffer, int pagecount, bool enc)
769 {
770 	enum hv_mem_host_visibility visibility = enc ?
771 			VMBUS_PAGE_NOT_VISIBLE : VMBUS_PAGE_VISIBLE_READ_WRITE;
772 	u64 *pfn_array;
773 	phys_addr_t paddr;
774 	int i, pfn, err;
775 	void *vaddr;
776 	int ret = 0;
777 
778 	pfn_array = kmalloc(HV_HYP_PAGE_SIZE, GFP_KERNEL);
779 	if (!pfn_array) {
780 		ret = -ENOMEM;
781 		goto err_set_memory_p;
782 	}
783 
784 	for (i = 0, pfn = 0; i < pagecount; i++) {
785 		/*
786 		 * Use slow_virt_to_phys() because the PRESENT bit has been
787 		 * temporarily cleared in the PTEs.  slow_virt_to_phys() works
788 		 * without the PRESENT bit while virt_to_hvpfn() or similar
789 		 * does not.
790 		 */
791 		vaddr = (void *)kbuffer + (i * HV_HYP_PAGE_SIZE);
792 		paddr = slow_virt_to_phys(vaddr);
793 		pfn_array[pfn] = paddr >> HV_HYP_PAGE_SHIFT;
794 		pfn++;
795 
796 		if (pfn == HV_MAX_MODIFY_GPA_REP_COUNT || i == pagecount - 1) {
797 			ret = hv_mark_gpa_visibility(pfn, pfn_array,
798 						     visibility);
799 			if (ret)
800 				goto err_free_pfn_array;
801 			pfn = 0;
802 		}
803 	}
804 
805 err_free_pfn_array:
806 	kfree(pfn_array);
807 
808 err_set_memory_p:
809 	/*
810 	 * Set the PTE PRESENT bits again to revert what hv_vtom_clear_present()
811 	 * did. Do this even if there is an error earlier in this function in
812 	 * order to avoid leaving the memory range in a "broken" state. Setting
813 	 * the PRESENT bits shouldn't fail, but return an error if it does.
814 	 */
815 	err = set_memory_p(kbuffer, pagecount);
816 	if (err && !ret)
817 		ret = err;
818 
819 	return ret;
820 }
821 
hv_vtom_tlb_flush_required(bool private)822 static bool hv_vtom_tlb_flush_required(bool private)
823 {
824 	/*
825 	 * Since hv_vtom_clear_present() marks the PTEs as "not present"
826 	 * and flushes the TLB, they can't be in the TLB. That makes the
827 	 * flush controlled by this function redundant, so return "false".
828 	 */
829 	return false;
830 }
831 
hv_vtom_cache_flush_required(void)832 static bool hv_vtom_cache_flush_required(void)
833 {
834 	return false;
835 }
836 
hv_is_private_mmio(u64 addr)837 static bool hv_is_private_mmio(u64 addr)
838 {
839 	/*
840 	 * Hyper-V always provides a single IO-APIC in a guest VM.
841 	 * When a paravisor is used, it is emulated by the paravisor
842 	 * in the guest context and must be mapped private.
843 	 */
844 	if (addr >= HV_IOAPIC_BASE_ADDRESS &&
845 	    addr < (HV_IOAPIC_BASE_ADDRESS + PAGE_SIZE))
846 		return true;
847 
848 	/* Same with a vTPM */
849 	if (addr >= VTPM_BASE_ADDRESS &&
850 	    addr < (VTPM_BASE_ADDRESS + PAGE_SIZE))
851 		return true;
852 
853 	return false;
854 }
855 
hv_vtom_init(void)856 void __init hv_vtom_init(void)
857 {
858 	enum hv_isolation_type type = hv_get_isolation_type();
859 
860 	switch (type) {
861 	case HV_ISOLATION_TYPE_VBS:
862 		fallthrough;
863 	/*
864 	 * By design, a VM using vTOM doesn't see the SEV setting,
865 	 * so SEV initialization is bypassed and sev_status isn't set.
866 	 * Set it here to indicate a vTOM VM.
867 	 *
868 	 * Note: if CONFIG_AMD_MEM_ENCRYPT is not set, sev_status is
869 	 * defined as 0ULL, to which we can't assigned a value.
870 	 */
871 #ifdef CONFIG_AMD_MEM_ENCRYPT
872 	case HV_ISOLATION_TYPE_SNP:
873 		sev_status = MSR_AMD64_SNP_VTOM;
874 		cc_vendor = CC_VENDOR_AMD;
875 		break;
876 #endif
877 
878 	case HV_ISOLATION_TYPE_TDX:
879 		cc_vendor = CC_VENDOR_INTEL;
880 		break;
881 
882 	default:
883 		panic("hv_vtom_init: unsupported isolation type %d\n", type);
884 	}
885 
886 	cc_set_mask(ms_hyperv.shared_gpa_boundary);
887 	physical_mask &= ms_hyperv.shared_gpa_boundary - 1;
888 
889 	x86_platform.hyper.is_private_mmio = hv_is_private_mmio;
890 	x86_platform.guest.enc_cache_flush_required = hv_vtom_cache_flush_required;
891 	x86_platform.guest.enc_tlb_flush_required = hv_vtom_tlb_flush_required;
892 	x86_platform.guest.enc_status_change_prepare = hv_vtom_clear_present;
893 	x86_platform.guest.enc_status_change_finish = hv_vtom_set_host_visibility;
894 	x86_platform.guest.enc_kexec_begin = hv_vtom_kexec_begin;
895 	x86_platform.guest.enc_kexec_finish = hv_vtom_kexec_finish;
896 
897 	/* Set WB as the default cache mode. */
898 	guest_force_mtrr_state(NULL, 0, MTRR_TYPE_WRBACK);
899 }
900 
901 #endif /* defined(CONFIG_AMD_MEM_ENCRYPT) || defined(CONFIG_INTEL_TDX_GUEST) */
902 
hv_get_isolation_type(void)903 enum hv_isolation_type hv_get_isolation_type(void)
904 {
905 	if (!(ms_hyperv.priv_high & HV_ISOLATION))
906 		return HV_ISOLATION_TYPE_NONE;
907 	return FIELD_GET(HV_ISOLATION_TYPE, ms_hyperv.isolation_config_b);
908 }
909 EXPORT_SYMBOL_GPL(hv_get_isolation_type);
910 
911 /*
912  * hv_is_isolation_supported - Check system runs in the Hyper-V
913  * isolation VM.
914  */
hv_is_isolation_supported(void)915 bool hv_is_isolation_supported(void)
916 {
917 	if (!cpu_feature_enabled(X86_FEATURE_HYPERVISOR))
918 		return false;
919 
920 	if (!hypervisor_is_type(X86_HYPER_MS_HYPERV))
921 		return false;
922 
923 	return hv_get_isolation_type() != HV_ISOLATION_TYPE_NONE;
924 }
925 
926 DEFINE_STATIC_KEY_FALSE(isolation_type_snp);
927 
928 /*
929  * hv_isolation_type_snp - Check if the system runs in an AMD SEV-SNP based
930  * isolation VM.
931  */
hv_isolation_type_snp(void)932 bool hv_isolation_type_snp(void)
933 {
934 	return static_branch_unlikely(&isolation_type_snp);
935 }
936 
937 DEFINE_STATIC_KEY_FALSE(isolation_type_tdx);
938 /*
939  * hv_isolation_type_tdx - Check if the system runs in an Intel TDX based
940  * isolated VM.
941  */
hv_isolation_type_tdx(void)942 bool hv_isolation_type_tdx(void)
943 {
944 	return static_branch_unlikely(&isolation_type_tdx);
945 }
946