xref: /linux/arch/x86/coco/sev/core.c (revision 14bed9bc81bae64db98349319f367bfc7dab0afd)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * AMD Memory Encryption Support
4  *
5  * Copyright (C) 2019 SUSE
6  *
7  * Author: Joerg Roedel <jroedel@suse.de>
8  */
9 
10 #define pr_fmt(fmt)	"SEV: " fmt
11 
12 #include <linux/sched/debug.h>	/* For show_regs() */
13 #include <linux/percpu-defs.h>
14 #include <linux/cc_platform.h>
15 #include <linux/printk.h>
16 #include <linux/mm_types.h>
17 #include <linux/set_memory.h>
18 #include <linux/memblock.h>
19 #include <linux/kernel.h>
20 #include <linux/mm.h>
21 #include <linux/cpumask.h>
22 #include <linux/efi.h>
23 #include <linux/platform_device.h>
24 #include <linux/io.h>
25 #include <linux/psp-sev.h>
26 #include <linux/dmi.h>
27 #include <uapi/linux/sev-guest.h>
28 #include <crypto/gcm.h>
29 
30 #include <asm/init.h>
31 #include <asm/cpu_entry_area.h>
32 #include <asm/stacktrace.h>
33 #include <asm/sev.h>
34 #include <asm/sev-internal.h>
35 #include <asm/insn-eval.h>
36 #include <asm/fpu/xcr.h>
37 #include <asm/processor.h>
38 #include <asm/realmode.h>
39 #include <asm/setup.h>
40 #include <asm/traps.h>
41 #include <asm/svm.h>
42 #include <asm/smp.h>
43 #include <asm/cpu.h>
44 #include <asm/apic.h>
45 #include <asm/cpuid/api.h>
46 #include <asm/cmdline.h>
47 #include <asm/msr.h>
48 
49 /* AP INIT values as documented in the APM2  section "Processor Initialization State" */
50 #define AP_INIT_CS_LIMIT		0xffff
51 #define AP_INIT_DS_LIMIT		0xffff
52 #define AP_INIT_LDTR_LIMIT		0xffff
53 #define AP_INIT_GDTR_LIMIT		0xffff
54 #define AP_INIT_IDTR_LIMIT		0xffff
55 #define AP_INIT_TR_LIMIT		0xffff
56 #define AP_INIT_RFLAGS_DEFAULT		0x2
57 #define AP_INIT_DR6_DEFAULT		0xffff0ff0
58 #define AP_INIT_GPAT_DEFAULT		0x0007040600070406ULL
59 #define AP_INIT_XCR0_DEFAULT		0x1
60 #define AP_INIT_X87_FTW_DEFAULT		0x5555
61 #define AP_INIT_X87_FCW_DEFAULT		0x0040
62 #define AP_INIT_CR0_DEFAULT		0x60000010
63 #define AP_INIT_MXCSR_DEFAULT		0x1f80
64 
65 static const char * const sev_status_feat_names[] = {
66 	[MSR_AMD64_SEV_ENABLED_BIT]		= "SEV",
67 	[MSR_AMD64_SEV_ES_ENABLED_BIT]		= "SEV-ES",
68 	[MSR_AMD64_SEV_SNP_ENABLED_BIT]		= "SEV-SNP",
69 	[MSR_AMD64_SNP_VTOM_BIT]		= "vTom",
70 	[MSR_AMD64_SNP_REFLECT_VC_BIT]		= "ReflectVC",
71 	[MSR_AMD64_SNP_RESTRICTED_INJ_BIT]	= "RI",
72 	[MSR_AMD64_SNP_ALT_INJ_BIT]		= "AI",
73 	[MSR_AMD64_SNP_DEBUG_SWAP_BIT]		= "DebugSwap",
74 	[MSR_AMD64_SNP_PREVENT_HOST_IBS_BIT]	= "NoHostIBS",
75 	[MSR_AMD64_SNP_BTB_ISOLATION_BIT]	= "BTBIsol",
76 	[MSR_AMD64_SNP_VMPL_SSS_BIT]		= "VmplSSS",
77 	[MSR_AMD64_SNP_SECURE_TSC_BIT]		= "SecureTSC",
78 	[MSR_AMD64_SNP_VMGEXIT_PARAM_BIT]	= "VMGExitParam",
79 	[MSR_AMD64_SNP_IBS_VIRT_BIT]		= "IBSVirt",
80 	[MSR_AMD64_SNP_VMSA_REG_PROT_BIT]	= "VMSARegProt",
81 	[MSR_AMD64_SNP_SMT_PROT_BIT]		= "SMTProt",
82 };
83 
84 /*
85  * For Secure TSC guests, the BSP fetches TSC_INFO using SNP guest messaging and
86  * initializes snp_tsc_scale and snp_tsc_offset. These values are replicated
87  * across the APs VMSA fields (TSC_SCALE and TSC_OFFSET).
88  */
89 static u64 snp_tsc_scale __ro_after_init;
90 static u64 snp_tsc_offset __ro_after_init;
91 static unsigned long snp_tsc_freq_khz __ro_after_init;
92 
93 DEFINE_PER_CPU(struct sev_es_runtime_data*, runtime_data);
94 DEFINE_PER_CPU(struct sev_es_save_area *, sev_vmsa);
95 
96 /*
97  * SVSM related information:
98  *   When running under an SVSM, the VMPL that Linux is executing at must be
99  *   non-zero. The VMPL is therefore used to indicate the presence of an SVSM.
100  */
101 u8 snp_vmpl __ro_after_init;
102 EXPORT_SYMBOL_GPL(snp_vmpl);
103 
get_snp_jump_table_addr(void)104 static u64 __init get_snp_jump_table_addr(void)
105 {
106 	struct snp_secrets_page *secrets;
107 	void __iomem *mem;
108 	u64 addr;
109 
110 	mem = ioremap_encrypted(sev_secrets_pa, PAGE_SIZE);
111 	if (!mem) {
112 		pr_err("Unable to locate AP jump table address: failed to map the SNP secrets page.\n");
113 		return 0;
114 	}
115 
116 	secrets = (__force struct snp_secrets_page *)mem;
117 
118 	addr = secrets->os_area.ap_jump_table_pa;
119 	iounmap(mem);
120 
121 	return addr;
122 }
123 
get_jump_table_addr(void)124 static u64 __init get_jump_table_addr(void)
125 {
126 	struct ghcb_state state;
127 	unsigned long flags;
128 	struct ghcb *ghcb;
129 	u64 ret = 0;
130 
131 	if (cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
132 		return get_snp_jump_table_addr();
133 
134 	local_irq_save(flags);
135 
136 	ghcb = __sev_get_ghcb(&state);
137 
138 	vc_ghcb_invalidate(ghcb);
139 	ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_JUMP_TABLE);
140 	ghcb_set_sw_exit_info_1(ghcb, SVM_VMGEXIT_GET_AP_JUMP_TABLE);
141 	ghcb_set_sw_exit_info_2(ghcb, 0);
142 
143 	sev_es_wr_ghcb_msr(__pa(ghcb));
144 	VMGEXIT();
145 
146 	if (ghcb_sw_exit_info_1_is_valid(ghcb) &&
147 	    ghcb_sw_exit_info_2_is_valid(ghcb))
148 		ret = ghcb->save.sw_exit_info_2;
149 
150 	__sev_put_ghcb(&state);
151 
152 	local_irq_restore(flags);
153 
154 	return ret;
155 }
156 
__pval_terminate(u64 pfn,bool action,unsigned int page_size,int ret,u64 svsm_ret)157 static inline void __pval_terminate(u64 pfn, bool action, unsigned int page_size,
158 				    int ret, u64 svsm_ret)
159 {
160 	WARN(1, "PVALIDATE failure: pfn: 0x%llx, action: %u, size: %u, ret: %d, svsm_ret: 0x%llx\n",
161 	     pfn, action, page_size, ret, svsm_ret);
162 
163 	sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PVALIDATE);
164 }
165 
svsm_pval_terminate(struct svsm_pvalidate_call * pc,int ret,u64 svsm_ret)166 static void svsm_pval_terminate(struct svsm_pvalidate_call *pc, int ret, u64 svsm_ret)
167 {
168 	unsigned int page_size;
169 	bool action;
170 	u64 pfn;
171 
172 	pfn = pc->entry[pc->cur_index].pfn;
173 	action = pc->entry[pc->cur_index].action;
174 	page_size = pc->entry[pc->cur_index].page_size;
175 
176 	__pval_terminate(pfn, action, page_size, ret, svsm_ret);
177 }
178 
pval_pages(struct snp_psc_desc * desc)179 static void pval_pages(struct snp_psc_desc *desc)
180 {
181 	struct psc_entry *e;
182 	unsigned long vaddr;
183 	unsigned int size;
184 	unsigned int i;
185 	bool validate;
186 	u64 pfn;
187 	int rc;
188 
189 	for (i = 0; i <= desc->hdr.end_entry; i++) {
190 		e = &desc->entries[i];
191 
192 		pfn = e->gfn;
193 		vaddr = (unsigned long)pfn_to_kaddr(pfn);
194 		size = e->pagesize ? RMP_PG_SIZE_2M : RMP_PG_SIZE_4K;
195 		validate = e->operation == SNP_PAGE_STATE_PRIVATE;
196 
197 		rc = pvalidate(vaddr, size, validate);
198 		if (!rc)
199 			continue;
200 
201 		if (rc == PVALIDATE_FAIL_SIZEMISMATCH && size == RMP_PG_SIZE_2M) {
202 			unsigned long vaddr_end = vaddr + PMD_SIZE;
203 
204 			for (; vaddr < vaddr_end; vaddr += PAGE_SIZE, pfn++) {
205 				rc = pvalidate(vaddr, RMP_PG_SIZE_4K, validate);
206 				if (rc)
207 					__pval_terminate(pfn, validate, RMP_PG_SIZE_4K, rc, 0);
208 			}
209 		} else {
210 			__pval_terminate(pfn, validate, size, rc, 0);
211 		}
212 	}
213 }
214 
svsm_build_ca_from_pfn_range(u64 pfn,u64 pfn_end,bool action,struct svsm_pvalidate_call * pc)215 static u64 svsm_build_ca_from_pfn_range(u64 pfn, u64 pfn_end, bool action,
216 					struct svsm_pvalidate_call *pc)
217 {
218 	struct svsm_pvalidate_entry *pe;
219 
220 	/* Nothing in the CA yet */
221 	pc->num_entries = 0;
222 	pc->cur_index   = 0;
223 
224 	pe = &pc->entry[0];
225 
226 	while (pfn < pfn_end) {
227 		pe->page_size = RMP_PG_SIZE_4K;
228 		pe->action    = action;
229 		pe->ignore_cf = 0;
230 		pe->pfn       = pfn;
231 
232 		pe++;
233 		pfn++;
234 
235 		pc->num_entries++;
236 		if (pc->num_entries == SVSM_PVALIDATE_MAX_COUNT)
237 			break;
238 	}
239 
240 	return pfn;
241 }
242 
svsm_build_ca_from_psc_desc(struct snp_psc_desc * desc,unsigned int desc_entry,struct svsm_pvalidate_call * pc)243 static int svsm_build_ca_from_psc_desc(struct snp_psc_desc *desc, unsigned int desc_entry,
244 				       struct svsm_pvalidate_call *pc)
245 {
246 	struct svsm_pvalidate_entry *pe;
247 	struct psc_entry *e;
248 
249 	/* Nothing in the CA yet */
250 	pc->num_entries = 0;
251 	pc->cur_index   = 0;
252 
253 	pe = &pc->entry[0];
254 	e  = &desc->entries[desc_entry];
255 
256 	while (desc_entry <= desc->hdr.end_entry) {
257 		pe->page_size = e->pagesize ? RMP_PG_SIZE_2M : RMP_PG_SIZE_4K;
258 		pe->action    = e->operation == SNP_PAGE_STATE_PRIVATE;
259 		pe->ignore_cf = 0;
260 		pe->pfn       = e->gfn;
261 
262 		pe++;
263 		e++;
264 
265 		desc_entry++;
266 		pc->num_entries++;
267 		if (pc->num_entries == SVSM_PVALIDATE_MAX_COUNT)
268 			break;
269 	}
270 
271 	return desc_entry;
272 }
273 
svsm_pval_pages(struct snp_psc_desc * desc)274 static void svsm_pval_pages(struct snp_psc_desc *desc)
275 {
276 	struct svsm_pvalidate_entry pv_4k[VMGEXIT_PSC_MAX_ENTRY];
277 	unsigned int i, pv_4k_count = 0;
278 	struct svsm_pvalidate_call *pc;
279 	struct svsm_call call = {};
280 	unsigned long flags;
281 	bool action;
282 	u64 pc_pa;
283 	int ret;
284 
285 	/*
286 	 * This can be called very early in the boot, use native functions in
287 	 * order to avoid paravirt issues.
288 	 */
289 	flags = native_local_irq_save();
290 
291 	/*
292 	 * The SVSM calling area (CA) can support processing 510 entries at a
293 	 * time. Loop through the Page State Change descriptor until the CA is
294 	 * full or the last entry in the descriptor is reached, at which time
295 	 * the SVSM is invoked. This repeats until all entries in the descriptor
296 	 * are processed.
297 	 */
298 	call.caa = svsm_get_caa();
299 
300 	pc = (struct svsm_pvalidate_call *)call.caa->svsm_buffer;
301 	pc_pa = svsm_get_caa_pa() + offsetof(struct svsm_ca, svsm_buffer);
302 
303 	/* Protocol 0, Call ID 1 */
304 	call.rax = SVSM_CORE_CALL(SVSM_CORE_PVALIDATE);
305 	call.rcx = pc_pa;
306 
307 	for (i = 0; i <= desc->hdr.end_entry;) {
308 		i = svsm_build_ca_from_psc_desc(desc, i, pc);
309 
310 		do {
311 			ret = svsm_perform_call_protocol(&call);
312 			if (!ret)
313 				continue;
314 
315 			/*
316 			 * Check if the entry failed because of an RMP mismatch (a
317 			 * PVALIDATE at 2M was requested, but the page is mapped in
318 			 * the RMP as 4K).
319 			 */
320 
321 			if (call.rax_out == SVSM_PVALIDATE_FAIL_SIZEMISMATCH &&
322 			    pc->entry[pc->cur_index].page_size == RMP_PG_SIZE_2M) {
323 				/* Save this entry for post-processing at 4K */
324 				pv_4k[pv_4k_count++] = pc->entry[pc->cur_index];
325 
326 				/* Skip to the next one unless at the end of the list */
327 				pc->cur_index++;
328 				if (pc->cur_index < pc->num_entries)
329 					ret = -EAGAIN;
330 				else
331 					ret = 0;
332 			}
333 		} while (ret == -EAGAIN);
334 
335 		if (ret)
336 			svsm_pval_terminate(pc, ret, call.rax_out);
337 	}
338 
339 	/* Process any entries that failed to be validated at 2M and validate them at 4K */
340 	for (i = 0; i < pv_4k_count; i++) {
341 		u64 pfn, pfn_end;
342 
343 		action  = pv_4k[i].action;
344 		pfn     = pv_4k[i].pfn;
345 		pfn_end = pfn + 512;
346 
347 		while (pfn < pfn_end) {
348 			pfn = svsm_build_ca_from_pfn_range(pfn, pfn_end, action, pc);
349 
350 			ret = svsm_perform_call_protocol(&call);
351 			if (ret)
352 				svsm_pval_terminate(pc, ret, call.rax_out);
353 		}
354 	}
355 
356 	native_local_irq_restore(flags);
357 }
358 
pvalidate_pages(struct snp_psc_desc * desc)359 static void pvalidate_pages(struct snp_psc_desc *desc)
360 {
361 	if (snp_vmpl)
362 		svsm_pval_pages(desc);
363 	else
364 		pval_pages(desc);
365 }
366 
vmgexit_psc(struct ghcb * ghcb,struct snp_psc_desc * desc)367 static int vmgexit_psc(struct ghcb *ghcb, struct snp_psc_desc *desc)
368 {
369 	int cur_entry, end_entry, ret = 0;
370 	struct snp_psc_desc *data;
371 	struct es_em_ctxt ctxt;
372 
373 	vc_ghcb_invalidate(ghcb);
374 
375 	/* Copy the input desc into GHCB shared buffer */
376 	data = (struct snp_psc_desc *)ghcb->shared_buffer;
377 	memcpy(ghcb->shared_buffer, desc, min_t(int, GHCB_SHARED_BUF_SIZE, sizeof(*desc)));
378 
379 	/*
380 	 * As per the GHCB specification, the hypervisor can resume the guest
381 	 * before processing all the entries. Check whether all the entries
382 	 * are processed. If not, then keep retrying. Note, the hypervisor
383 	 * will update the data memory directly to indicate the status, so
384 	 * reference the data->hdr everywhere.
385 	 *
386 	 * The strategy here is to wait for the hypervisor to change the page
387 	 * state in the RMP table before guest accesses the memory pages. If the
388 	 * page state change was not successful, then later memory access will
389 	 * result in a crash.
390 	 */
391 	cur_entry = data->hdr.cur_entry;
392 	end_entry = data->hdr.end_entry;
393 
394 	while (data->hdr.cur_entry <= data->hdr.end_entry) {
395 		ghcb_set_sw_scratch(ghcb, (u64)__pa(data));
396 
397 		/* This will advance the shared buffer data points to. */
398 		ret = sev_es_ghcb_hv_call(ghcb, &ctxt, SVM_VMGEXIT_PSC, 0, 0);
399 
400 		/*
401 		 * Page State Change VMGEXIT can pass error code through
402 		 * exit_info_2.
403 		 */
404 		if (WARN(ret || ghcb->save.sw_exit_info_2,
405 			 "SNP: PSC failed ret=%d exit_info_2=%llx\n",
406 			 ret, ghcb->save.sw_exit_info_2)) {
407 			ret = 1;
408 			goto out;
409 		}
410 
411 		/* Verify that reserved bit is not set */
412 		if (WARN(data->hdr.reserved, "Reserved bit is set in the PSC header\n")) {
413 			ret = 1;
414 			goto out;
415 		}
416 
417 		/*
418 		 * Sanity check that entry processing is not going backwards.
419 		 * This will happen only if hypervisor is tricking us.
420 		 */
421 		if (WARN(data->hdr.end_entry > end_entry || cur_entry > data->hdr.cur_entry,
422 "SNP: PSC processing going backward, end_entry %d (got %d) cur_entry %d (got %d)\n",
423 			 end_entry, data->hdr.end_entry, cur_entry, data->hdr.cur_entry)) {
424 			ret = 1;
425 			goto out;
426 		}
427 	}
428 
429 out:
430 	return ret;
431 }
432 
__set_pages_state(struct snp_psc_desc * data,unsigned long vaddr,unsigned long vaddr_end,int op)433 static unsigned long __set_pages_state(struct snp_psc_desc *data, unsigned long vaddr,
434 				       unsigned long vaddr_end, int op)
435 {
436 	struct ghcb_state state;
437 	bool use_large_entry;
438 	struct psc_hdr *hdr;
439 	struct psc_entry *e;
440 	unsigned long flags;
441 	unsigned long pfn;
442 	struct ghcb *ghcb;
443 	int i;
444 
445 	hdr = &data->hdr;
446 	e = data->entries;
447 
448 	memset(data, 0, sizeof(*data));
449 	i = 0;
450 
451 	while (vaddr < vaddr_end && i < ARRAY_SIZE(data->entries)) {
452 		hdr->end_entry = i;
453 
454 		if (is_vmalloc_addr((void *)vaddr)) {
455 			pfn = vmalloc_to_pfn((void *)vaddr);
456 			use_large_entry = false;
457 		} else {
458 			pfn = __pa(vaddr) >> PAGE_SHIFT;
459 			use_large_entry = true;
460 		}
461 
462 		e->gfn = pfn;
463 		e->operation = op;
464 
465 		if (use_large_entry && IS_ALIGNED(vaddr, PMD_SIZE) &&
466 		    (vaddr_end - vaddr) >= PMD_SIZE) {
467 			e->pagesize = RMP_PG_SIZE_2M;
468 			vaddr += PMD_SIZE;
469 		} else {
470 			e->pagesize = RMP_PG_SIZE_4K;
471 			vaddr += PAGE_SIZE;
472 		}
473 
474 		e++;
475 		i++;
476 	}
477 
478 	/* Page validation must be rescinded before changing to shared */
479 	if (op == SNP_PAGE_STATE_SHARED)
480 		pvalidate_pages(data);
481 
482 	local_irq_save(flags);
483 
484 	if (sev_cfg.ghcbs_initialized)
485 		ghcb = __sev_get_ghcb(&state);
486 	else
487 		ghcb = boot_ghcb;
488 
489 	/* Invoke the hypervisor to perform the page state changes */
490 	if (!ghcb || vmgexit_psc(ghcb, data))
491 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PSC);
492 
493 	if (sev_cfg.ghcbs_initialized)
494 		__sev_put_ghcb(&state);
495 
496 	local_irq_restore(flags);
497 
498 	/* Page validation must be performed after changing to private */
499 	if (op == SNP_PAGE_STATE_PRIVATE)
500 		pvalidate_pages(data);
501 
502 	return vaddr;
503 }
504 
set_pages_state(unsigned long vaddr,unsigned long npages,int op)505 static void set_pages_state(unsigned long vaddr, unsigned long npages, int op)
506 {
507 	struct snp_psc_desc desc;
508 	unsigned long vaddr_end;
509 
510 	/* Use the MSR protocol when a GHCB is not available. */
511 	if (!boot_ghcb)
512 		return early_set_pages_state(vaddr, __pa(vaddr), npages, op);
513 
514 	vaddr = vaddr & PAGE_MASK;
515 	vaddr_end = vaddr + (npages << PAGE_SHIFT);
516 
517 	while (vaddr < vaddr_end)
518 		vaddr = __set_pages_state(&desc, vaddr, vaddr_end, op);
519 }
520 
snp_set_memory_shared(unsigned long vaddr,unsigned long npages)521 void snp_set_memory_shared(unsigned long vaddr, unsigned long npages)
522 {
523 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
524 		return;
525 
526 	set_pages_state(vaddr, npages, SNP_PAGE_STATE_SHARED);
527 }
528 
snp_set_memory_private(unsigned long vaddr,unsigned long npages)529 void snp_set_memory_private(unsigned long vaddr, unsigned long npages)
530 {
531 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
532 		return;
533 
534 	set_pages_state(vaddr, npages, SNP_PAGE_STATE_PRIVATE);
535 }
536 
snp_accept_memory(phys_addr_t start,phys_addr_t end)537 void snp_accept_memory(phys_addr_t start, phys_addr_t end)
538 {
539 	unsigned long vaddr, npages;
540 
541 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
542 		return;
543 
544 	vaddr = (unsigned long)__va(start);
545 	npages = (end - start) >> PAGE_SHIFT;
546 
547 	set_pages_state(vaddr, npages, SNP_PAGE_STATE_PRIVATE);
548 }
549 
vmgexit_ap_control(u64 event,struct sev_es_save_area * vmsa,u32 apic_id)550 static int vmgexit_ap_control(u64 event, struct sev_es_save_area *vmsa, u32 apic_id)
551 {
552 	bool create = event != SVM_VMGEXIT_AP_DESTROY;
553 	struct ghcb_state state;
554 	unsigned long flags;
555 	struct ghcb *ghcb;
556 	int ret = 0;
557 
558 	local_irq_save(flags);
559 
560 	ghcb = __sev_get_ghcb(&state);
561 
562 	vc_ghcb_invalidate(ghcb);
563 
564 	if (create)
565 		ghcb_set_rax(ghcb, vmsa->sev_features);
566 
567 	ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_CREATION);
568 	ghcb_set_sw_exit_info_1(ghcb,
569 				((u64)apic_id << 32)	|
570 				((u64)snp_vmpl << 16)	|
571 				event);
572 	ghcb_set_sw_exit_info_2(ghcb, __pa(vmsa));
573 
574 	sev_es_wr_ghcb_msr(__pa(ghcb));
575 	VMGEXIT();
576 
577 	if (!ghcb_sw_exit_info_1_is_valid(ghcb) ||
578 	    lower_32_bits(ghcb->save.sw_exit_info_1)) {
579 		pr_err("SNP AP %s error\n", (create ? "CREATE" : "DESTROY"));
580 		ret = -EINVAL;
581 	}
582 
583 	__sev_put_ghcb(&state);
584 
585 	local_irq_restore(flags);
586 
587 	return ret;
588 }
589 
snp_set_vmsa(void * va,void * caa,int apic_id,bool make_vmsa)590 static int snp_set_vmsa(void *va, void *caa, int apic_id, bool make_vmsa)
591 {
592 	int ret;
593 
594 	if (snp_vmpl) {
595 		struct svsm_call call = {};
596 		unsigned long flags;
597 
598 		local_irq_save(flags);
599 
600 		call.caa = this_cpu_read(svsm_caa);
601 		call.rcx = __pa(va);
602 
603 		if (make_vmsa) {
604 			/* Protocol 0, Call ID 2 */
605 			call.rax = SVSM_CORE_CALL(SVSM_CORE_CREATE_VCPU);
606 			call.rdx = __pa(caa);
607 			call.r8  = apic_id;
608 		} else {
609 			/* Protocol 0, Call ID 3 */
610 			call.rax = SVSM_CORE_CALL(SVSM_CORE_DELETE_VCPU);
611 		}
612 
613 		ret = svsm_perform_call_protocol(&call);
614 
615 		local_irq_restore(flags);
616 	} else {
617 		/*
618 		 * If the kernel runs at VMPL0, it can change the VMSA
619 		 * bit for a page using the RMPADJUST instruction.
620 		 * However, for the instruction to succeed it must
621 		 * target the permissions of a lesser privileged (higher
622 		 * numbered) VMPL level, so use VMPL1.
623 		 */
624 		u64 attrs = 1;
625 
626 		if (make_vmsa)
627 			attrs |= RMPADJUST_VMSA_PAGE_BIT;
628 
629 		ret = rmpadjust((unsigned long)va, RMP_PG_SIZE_4K, attrs);
630 	}
631 
632 	return ret;
633 }
634 
snp_cleanup_vmsa(struct sev_es_save_area * vmsa,int apic_id)635 static void snp_cleanup_vmsa(struct sev_es_save_area *vmsa, int apic_id)
636 {
637 	int err;
638 
639 	err = snp_set_vmsa(vmsa, NULL, apic_id, false);
640 	if (err)
641 		pr_err("clear VMSA page failed (%u), leaking page\n", err);
642 	else
643 		free_page((unsigned long)vmsa);
644 }
645 
set_pte_enc(pte_t * kpte,int level,void * va)646 static void set_pte_enc(pte_t *kpte, int level, void *va)
647 {
648 	struct pte_enc_desc d = {
649 		.kpte	   = kpte,
650 		.pte_level = level,
651 		.va	   = va,
652 		.encrypt   = true
653 	};
654 
655 	prepare_pte_enc(&d);
656 	set_pte_enc_mask(kpte, d.pfn, d.new_pgprot);
657 }
658 
unshare_all_memory(void)659 static void unshare_all_memory(void)
660 {
661 	unsigned long addr, end, size, ghcb;
662 	struct sev_es_runtime_data *data;
663 	unsigned int npages, level;
664 	bool skipped_addr;
665 	pte_t *pte;
666 	int cpu;
667 
668 	/* Unshare the direct mapping. */
669 	addr = PAGE_OFFSET;
670 	end  = PAGE_OFFSET + get_max_mapped();
671 
672 	while (addr < end) {
673 		pte = lookup_address(addr, &level);
674 		size = page_level_size(level);
675 		npages = size / PAGE_SIZE;
676 		skipped_addr = false;
677 
678 		if (!pte || !pte_decrypted(*pte) || pte_none(*pte)) {
679 			addr += size;
680 			continue;
681 		}
682 
683 		/*
684 		 * Ensure that all the per-CPU GHCBs are made private at the
685 		 * end of the unsharing loop so that the switch to the slower
686 		 * MSR protocol happens last.
687 		 */
688 		for_each_possible_cpu(cpu) {
689 			data = per_cpu(runtime_data, cpu);
690 			ghcb = (unsigned long)&data->ghcb_page;
691 
692 			/* Handle the case of a huge page containing the GHCB page */
693 			if (addr <= ghcb && ghcb < addr + size) {
694 				skipped_addr = true;
695 				break;
696 			}
697 		}
698 
699 		if (!skipped_addr) {
700 			set_pte_enc(pte, level, (void *)addr);
701 			snp_set_memory_private(addr, npages);
702 		}
703 		addr += size;
704 	}
705 
706 	/* Unshare all bss decrypted memory. */
707 	addr = (unsigned long)__start_bss_decrypted;
708 	end  = (unsigned long)__start_bss_decrypted_unused;
709 	npages = (end - addr) >> PAGE_SHIFT;
710 
711 	for (; addr < end; addr += PAGE_SIZE) {
712 		pte = lookup_address(addr, &level);
713 		if (!pte || !pte_decrypted(*pte) || pte_none(*pte))
714 			continue;
715 
716 		set_pte_enc(pte, level, (void *)addr);
717 	}
718 	addr = (unsigned long)__start_bss_decrypted;
719 	snp_set_memory_private(addr, npages);
720 
721 	__flush_tlb_all();
722 }
723 
724 /* Stop new private<->shared conversions */
snp_kexec_begin(void)725 void snp_kexec_begin(void)
726 {
727 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
728 		return;
729 
730 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
731 		return;
732 
733 	/*
734 	 * Crash kernel ends up here with interrupts disabled: can't wait for
735 	 * conversions to finish.
736 	 *
737 	 * If race happened, just report and proceed.
738 	 */
739 	if (!set_memory_enc_stop_conversion())
740 		pr_warn("Failed to stop shared<->private conversions\n");
741 }
742 
743 /*
744  * Shutdown all APs except the one handling kexec/kdump and clearing
745  * the VMSA tag on AP's VMSA pages as they are not being used as
746  * VMSA page anymore.
747  */
shutdown_all_aps(void)748 static void shutdown_all_aps(void)
749 {
750 	struct sev_es_save_area *vmsa;
751 	int apic_id, this_cpu, cpu;
752 
753 	this_cpu = get_cpu();
754 
755 	/*
756 	 * APs are already in HLT loop when enc_kexec_finish() callback
757 	 * is invoked.
758 	 */
759 	for_each_present_cpu(cpu) {
760 		vmsa = per_cpu(sev_vmsa, cpu);
761 
762 		/*
763 		 * The BSP or offlined APs do not have guest allocated VMSA
764 		 * and there is no need  to clear the VMSA tag for this page.
765 		 */
766 		if (!vmsa)
767 			continue;
768 
769 		/*
770 		 * Cannot clear the VMSA tag for the currently running vCPU.
771 		 */
772 		if (this_cpu == cpu) {
773 			unsigned long pa;
774 			struct page *p;
775 
776 			pa = __pa(vmsa);
777 			/*
778 			 * Mark the VMSA page of the running vCPU as offline
779 			 * so that is excluded and not touched by makedumpfile
780 			 * while generating vmcore during kdump.
781 			 */
782 			p = pfn_to_online_page(pa >> PAGE_SHIFT);
783 			if (p)
784 				__SetPageOffline(p);
785 			continue;
786 		}
787 
788 		apic_id = cpuid_to_apicid[cpu];
789 
790 		/*
791 		 * Issue AP destroy to ensure AP gets kicked out of guest mode
792 		 * to allow using RMPADJUST to remove the VMSA tag on it's
793 		 * VMSA page.
794 		 */
795 		vmgexit_ap_control(SVM_VMGEXIT_AP_DESTROY, vmsa, apic_id);
796 		snp_cleanup_vmsa(vmsa, apic_id);
797 	}
798 
799 	put_cpu();
800 }
801 
snp_kexec_finish(void)802 void snp_kexec_finish(void)
803 {
804 	struct sev_es_runtime_data *data;
805 	unsigned long size, addr;
806 	unsigned int level, cpu;
807 	struct ghcb *ghcb;
808 	pte_t *pte;
809 
810 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
811 		return;
812 
813 	if (!IS_ENABLED(CONFIG_KEXEC_CORE))
814 		return;
815 
816 	shutdown_all_aps();
817 
818 	unshare_all_memory();
819 
820 	/*
821 	 * Switch to using the MSR protocol to change per-CPU GHCBs to
822 	 * private. All the per-CPU GHCBs have been switched back to private,
823 	 * so can't do any more GHCB calls to the hypervisor beyond this point
824 	 * until the kexec'ed kernel starts running.
825 	 */
826 	boot_ghcb = NULL;
827 	sev_cfg.ghcbs_initialized = false;
828 
829 	for_each_possible_cpu(cpu) {
830 		data = per_cpu(runtime_data, cpu);
831 		ghcb = &data->ghcb_page;
832 		pte = lookup_address((unsigned long)ghcb, &level);
833 		size = page_level_size(level);
834 		/* Handle the case of a huge page containing the GHCB page */
835 		addr = (unsigned long)ghcb & page_level_mask(level);
836 		set_pte_enc(pte, level, (void *)addr);
837 		snp_set_memory_private(addr, (size / PAGE_SIZE));
838 	}
839 }
840 
841 #define __ATTR_BASE		(SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK)
842 #define INIT_CS_ATTRIBS		(__ATTR_BASE | SVM_SELECTOR_READ_MASK | SVM_SELECTOR_CODE_MASK)
843 #define INIT_DS_ATTRIBS		(__ATTR_BASE | SVM_SELECTOR_WRITE_MASK)
844 
845 #define INIT_LDTR_ATTRIBS	(SVM_SELECTOR_P_MASK | 2)
846 #define INIT_TR_ATTRIBS		(SVM_SELECTOR_P_MASK | 3)
847 
snp_alloc_vmsa_page(int cpu)848 static void *snp_alloc_vmsa_page(int cpu)
849 {
850 	struct page *p;
851 
852 	/*
853 	 * Allocate VMSA page to work around the SNP erratum where the CPU will
854 	 * incorrectly signal an RMP violation #PF if a large page (2MB or 1GB)
855 	 * collides with the RMP entry of VMSA page. The recommended workaround
856 	 * is to not use a large page.
857 	 *
858 	 * Allocate an 8k page which is also 8k-aligned.
859 	 */
860 	p = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL_ACCOUNT | __GFP_ZERO, 1);
861 	if (!p)
862 		return NULL;
863 
864 	split_page(p, 1);
865 
866 	/* Free the first 4k. This page may be 2M/1G aligned and cannot be used. */
867 	__free_page(p);
868 
869 	return page_address(p + 1);
870 }
871 
wakeup_cpu_via_vmgexit(u32 apic_id,unsigned long start_ip,unsigned int cpu)872 static int wakeup_cpu_via_vmgexit(u32 apic_id, unsigned long start_ip, unsigned int cpu)
873 {
874 	struct sev_es_save_area *cur_vmsa, *vmsa;
875 	struct svsm_ca *caa;
876 	u8 sipi_vector;
877 	int ret;
878 	u64 cr4;
879 
880 	/*
881 	 * The hypervisor SNP feature support check has happened earlier, just check
882 	 * the AP_CREATION one here.
883 	 */
884 	if (!(sev_hv_features & GHCB_HV_FT_SNP_AP_CREATION))
885 		return -EOPNOTSUPP;
886 
887 	/*
888 	 * Verify the desired start IP against the known trampoline start IP
889 	 * to catch any future new trampolines that may be introduced that
890 	 * would require a new protected guest entry point.
891 	 */
892 	if (WARN_ONCE(start_ip != real_mode_header->trampoline_start,
893 		      "Unsupported SNP start_ip: %lx\n", start_ip))
894 		return -EINVAL;
895 
896 	/* Override start_ip with known protected guest start IP */
897 	start_ip = real_mode_header->sev_es_trampoline_start;
898 	cur_vmsa = per_cpu(sev_vmsa, cpu);
899 
900 	/*
901 	 * A new VMSA is created each time because there is no guarantee that
902 	 * the current VMSA is the kernels or that the vCPU is not running. If
903 	 * an attempt was done to use the current VMSA with a running vCPU, a
904 	 * #VMEXIT of that vCPU would wipe out all of the settings being done
905 	 * here.
906 	 */
907 	vmsa = (struct sev_es_save_area *)snp_alloc_vmsa_page(cpu);
908 	if (!vmsa)
909 		return -ENOMEM;
910 
911 	/* If an SVSM is present, the SVSM per-CPU CAA will be !NULL */
912 	caa = per_cpu(svsm_caa, cpu);
913 
914 	/* CR4 should maintain the MCE value */
915 	cr4 = native_read_cr4() & X86_CR4_MCE;
916 
917 	/* Set the CS value based on the start_ip converted to a SIPI vector */
918 	sipi_vector		= (start_ip >> 12);
919 	vmsa->cs.base		= sipi_vector << 12;
920 	vmsa->cs.limit		= AP_INIT_CS_LIMIT;
921 	vmsa->cs.attrib		= INIT_CS_ATTRIBS;
922 	vmsa->cs.selector	= sipi_vector << 8;
923 
924 	/* Set the RIP value based on start_ip */
925 	vmsa->rip		= start_ip & 0xfff;
926 
927 	/* Set AP INIT defaults as documented in the APM */
928 	vmsa->ds.limit		= AP_INIT_DS_LIMIT;
929 	vmsa->ds.attrib		= INIT_DS_ATTRIBS;
930 	vmsa->es		= vmsa->ds;
931 	vmsa->fs		= vmsa->ds;
932 	vmsa->gs		= vmsa->ds;
933 	vmsa->ss		= vmsa->ds;
934 
935 	vmsa->gdtr.limit	= AP_INIT_GDTR_LIMIT;
936 	vmsa->ldtr.limit	= AP_INIT_LDTR_LIMIT;
937 	vmsa->ldtr.attrib	= INIT_LDTR_ATTRIBS;
938 	vmsa->idtr.limit	= AP_INIT_IDTR_LIMIT;
939 	vmsa->tr.limit		= AP_INIT_TR_LIMIT;
940 	vmsa->tr.attrib		= INIT_TR_ATTRIBS;
941 
942 	vmsa->cr4		= cr4;
943 	vmsa->cr0		= AP_INIT_CR0_DEFAULT;
944 	vmsa->dr7		= DR7_RESET_VALUE;
945 	vmsa->dr6		= AP_INIT_DR6_DEFAULT;
946 	vmsa->rflags		= AP_INIT_RFLAGS_DEFAULT;
947 	vmsa->g_pat		= AP_INIT_GPAT_DEFAULT;
948 	vmsa->xcr0		= AP_INIT_XCR0_DEFAULT;
949 	vmsa->mxcsr		= AP_INIT_MXCSR_DEFAULT;
950 	vmsa->x87_ftw		= AP_INIT_X87_FTW_DEFAULT;
951 	vmsa->x87_fcw		= AP_INIT_X87_FCW_DEFAULT;
952 
953 	/* SVME must be set. */
954 	vmsa->efer		= EFER_SVME;
955 
956 	/*
957 	 * Set the SNP-specific fields for this VMSA:
958 	 *   VMPL level
959 	 *   SEV_FEATURES (matches the SEV STATUS MSR right shifted 2 bits)
960 	 */
961 	vmsa->vmpl		= snp_vmpl;
962 	vmsa->sev_features	= sev_status >> 2;
963 
964 	/* Populate AP's TSC scale/offset to get accurate TSC values. */
965 	if (cc_platform_has(CC_ATTR_GUEST_SNP_SECURE_TSC)) {
966 		vmsa->tsc_scale = snp_tsc_scale;
967 		vmsa->tsc_offset = snp_tsc_offset;
968 	}
969 
970 	/* Switch the page over to a VMSA page now that it is initialized */
971 	ret = snp_set_vmsa(vmsa, caa, apic_id, true);
972 	if (ret) {
973 		pr_err("set VMSA page failed (%u)\n", ret);
974 		free_page((unsigned long)vmsa);
975 
976 		return -EINVAL;
977 	}
978 
979 	/* Issue VMGEXIT AP Creation NAE event */
980 	ret = vmgexit_ap_control(SVM_VMGEXIT_AP_CREATE, vmsa, apic_id);
981 	if (ret) {
982 		snp_cleanup_vmsa(vmsa, apic_id);
983 		vmsa = NULL;
984 	}
985 
986 	/* Free up any previous VMSA page */
987 	if (cur_vmsa)
988 		snp_cleanup_vmsa(cur_vmsa, apic_id);
989 
990 	/* Record the current VMSA page */
991 	per_cpu(sev_vmsa, cpu) = vmsa;
992 
993 	return ret;
994 }
995 
snp_set_wakeup_secondary_cpu(void)996 void __init snp_set_wakeup_secondary_cpu(void)
997 {
998 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
999 		return;
1000 
1001 	/*
1002 	 * Always set this override if SNP is enabled. This makes it the
1003 	 * required method to start APs under SNP. If the hypervisor does
1004 	 * not support AP creation, then no APs will be started.
1005 	 */
1006 	apic_update_callback(wakeup_secondary_cpu, wakeup_cpu_via_vmgexit);
1007 }
1008 
sev_es_setup_ap_jump_table(struct real_mode_header * rmh)1009 int __init sev_es_setup_ap_jump_table(struct real_mode_header *rmh)
1010 {
1011 	u16 startup_cs, startup_ip;
1012 	phys_addr_t jump_table_pa;
1013 	u64 jump_table_addr;
1014 	u16 __iomem *jump_table;
1015 
1016 	jump_table_addr = get_jump_table_addr();
1017 
1018 	/* On UP guests there is no jump table so this is not a failure */
1019 	if (!jump_table_addr)
1020 		return 0;
1021 
1022 	/* Check if AP Jump Table is page-aligned */
1023 	if (jump_table_addr & ~PAGE_MASK)
1024 		return -EINVAL;
1025 
1026 	jump_table_pa = jump_table_addr & PAGE_MASK;
1027 
1028 	startup_cs = (u16)(rmh->trampoline_start >> 4);
1029 	startup_ip = (u16)(rmh->sev_es_trampoline_start -
1030 			   rmh->trampoline_start);
1031 
1032 	jump_table = ioremap_encrypted(jump_table_pa, PAGE_SIZE);
1033 	if (!jump_table)
1034 		return -EIO;
1035 
1036 	writew(startup_ip, &jump_table[0]);
1037 	writew(startup_cs, &jump_table[1]);
1038 
1039 	iounmap(jump_table);
1040 
1041 	return 0;
1042 }
1043 
1044 /*
1045  * This is needed by the OVMF UEFI firmware which will use whatever it finds in
1046  * the GHCB MSR as its GHCB to talk to the hypervisor. So make sure the per-cpu
1047  * runtime GHCBs used by the kernel are also mapped in the EFI page-table.
1048  *
1049  * When running under SVSM the CA page is needed too, so map it as well.
1050  */
sev_es_efi_map_ghcbs_cas(pgd_t * pgd)1051 int __init sev_es_efi_map_ghcbs_cas(pgd_t *pgd)
1052 {
1053 	unsigned long address, pflags, pflags_enc;
1054 	struct sev_es_runtime_data *data;
1055 	int cpu;
1056 	u64 pfn;
1057 
1058 	if (!cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT))
1059 		return 0;
1060 
1061 	pflags = _PAGE_NX | _PAGE_RW;
1062 	pflags_enc = cc_mkenc(pflags);
1063 
1064 	for_each_possible_cpu(cpu) {
1065 		data = per_cpu(runtime_data, cpu);
1066 
1067 		address = __pa(&data->ghcb_page);
1068 		pfn = address >> PAGE_SHIFT;
1069 
1070 		if (kernel_map_pages_in_pgd(pgd, pfn, address, 1, pflags))
1071 			return 1;
1072 
1073 		if (snp_vmpl) {
1074 			address = per_cpu(svsm_caa_pa, cpu);
1075 			if (!address)
1076 				return 1;
1077 
1078 			pfn = address >> PAGE_SHIFT;
1079 			if (kernel_map_pages_in_pgd(pgd, pfn, address, 1, pflags_enc))
1080 				return 1;
1081 		}
1082 	}
1083 
1084 	return 0;
1085 }
1086 
snp_register_per_cpu_ghcb(void)1087 static void snp_register_per_cpu_ghcb(void)
1088 {
1089 	struct sev_es_runtime_data *data;
1090 	struct ghcb *ghcb;
1091 
1092 	data = this_cpu_read(runtime_data);
1093 	ghcb = &data->ghcb_page;
1094 
1095 	snp_register_ghcb_early(__pa(ghcb));
1096 }
1097 
setup_ghcb(void)1098 void setup_ghcb(void)
1099 {
1100 	if (!cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT))
1101 		return;
1102 
1103 	/*
1104 	 * Check whether the runtime #VC exception handler is active. It uses
1105 	 * the per-CPU GHCB page which is set up by sev_es_init_vc_handling().
1106 	 *
1107 	 * If SNP is active, register the per-CPU GHCB page so that the runtime
1108 	 * exception handler can use it.
1109 	 */
1110 	if (initial_vc_handler == (unsigned long)kernel_exc_vmm_communication) {
1111 		if (cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
1112 			snp_register_per_cpu_ghcb();
1113 
1114 		sev_cfg.ghcbs_initialized = true;
1115 
1116 		return;
1117 	}
1118 
1119 	/*
1120 	 * Make sure the hypervisor talks a supported protocol.
1121 	 * This gets called only in the BSP boot phase.
1122 	 */
1123 	if (!sev_es_negotiate_protocol())
1124 		sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
1125 
1126 	/*
1127 	 * Clear the boot_ghcb. The first exception comes in before the bss
1128 	 * section is cleared.
1129 	 */
1130 	memset(&boot_ghcb_page, 0, PAGE_SIZE);
1131 
1132 	/* Alright - Make the boot-ghcb public */
1133 	boot_ghcb = &boot_ghcb_page;
1134 
1135 	/* SNP guest requires that GHCB GPA must be registered. */
1136 	if (cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
1137 		snp_register_ghcb_early(__pa(&boot_ghcb_page));
1138 }
1139 
1140 #ifdef CONFIG_HOTPLUG_CPU
sev_es_ap_hlt_loop(void)1141 static void sev_es_ap_hlt_loop(void)
1142 {
1143 	struct ghcb_state state;
1144 	struct ghcb *ghcb;
1145 
1146 	ghcb = __sev_get_ghcb(&state);
1147 
1148 	while (true) {
1149 		vc_ghcb_invalidate(ghcb);
1150 		ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_HLT_LOOP);
1151 		ghcb_set_sw_exit_info_1(ghcb, 0);
1152 		ghcb_set_sw_exit_info_2(ghcb, 0);
1153 
1154 		sev_es_wr_ghcb_msr(__pa(ghcb));
1155 		VMGEXIT();
1156 
1157 		/* Wakeup signal? */
1158 		if (ghcb_sw_exit_info_2_is_valid(ghcb) &&
1159 		    ghcb->save.sw_exit_info_2)
1160 			break;
1161 	}
1162 
1163 	__sev_put_ghcb(&state);
1164 }
1165 
1166 /*
1167  * Play_dead handler when running under SEV-ES. This is needed because
1168  * the hypervisor can't deliver an SIPI request to restart the AP.
1169  * Instead the kernel has to issue a VMGEXIT to halt the VCPU until the
1170  * hypervisor wakes it up again.
1171  */
sev_es_play_dead(void)1172 static void sev_es_play_dead(void)
1173 {
1174 	play_dead_common();
1175 
1176 	/* IRQs now disabled */
1177 
1178 	sev_es_ap_hlt_loop();
1179 
1180 	/*
1181 	 * If we get here, the VCPU was woken up again. Jump to CPU
1182 	 * startup code to get it back online.
1183 	 */
1184 	soft_restart_cpu();
1185 }
1186 #else  /* CONFIG_HOTPLUG_CPU */
1187 #define sev_es_play_dead	native_play_dead
1188 #endif /* CONFIG_HOTPLUG_CPU */
1189 
1190 #ifdef CONFIG_SMP
sev_es_setup_play_dead(void)1191 static void __init sev_es_setup_play_dead(void)
1192 {
1193 	smp_ops.play_dead = sev_es_play_dead;
1194 }
1195 #else
sev_es_setup_play_dead(void)1196 static inline void sev_es_setup_play_dead(void) { }
1197 #endif
1198 
alloc_runtime_data(int cpu)1199 static void __init alloc_runtime_data(int cpu)
1200 {
1201 	struct sev_es_runtime_data *data;
1202 
1203 	data = memblock_alloc_node(sizeof(*data), PAGE_SIZE, cpu_to_node(cpu));
1204 	if (!data)
1205 		panic("Can't allocate SEV-ES runtime data");
1206 
1207 	per_cpu(runtime_data, cpu) = data;
1208 
1209 	if (snp_vmpl) {
1210 		struct svsm_ca *caa;
1211 
1212 		/* Allocate the SVSM CA page if an SVSM is present */
1213 		caa = memblock_alloc_or_panic(sizeof(*caa), PAGE_SIZE);
1214 
1215 		per_cpu(svsm_caa, cpu) = caa;
1216 		per_cpu(svsm_caa_pa, cpu) = __pa(caa);
1217 	}
1218 }
1219 
init_ghcb(int cpu)1220 static void __init init_ghcb(int cpu)
1221 {
1222 	struct sev_es_runtime_data *data;
1223 	int err;
1224 
1225 	data = per_cpu(runtime_data, cpu);
1226 
1227 	err = early_set_memory_decrypted((unsigned long)&data->ghcb_page,
1228 					 sizeof(data->ghcb_page));
1229 	if (err)
1230 		panic("Can't map GHCBs unencrypted");
1231 
1232 	memset(&data->ghcb_page, 0, sizeof(data->ghcb_page));
1233 
1234 	data->ghcb_active = false;
1235 	data->backup_ghcb_active = false;
1236 }
1237 
sev_es_init_vc_handling(void)1238 void __init sev_es_init_vc_handling(void)
1239 {
1240 	int cpu;
1241 
1242 	BUILD_BUG_ON(offsetof(struct sev_es_runtime_data, ghcb_page) % PAGE_SIZE);
1243 
1244 	if (!cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT))
1245 		return;
1246 
1247 	if (!sev_es_check_cpu_features())
1248 		panic("SEV-ES CPU Features missing");
1249 
1250 	/*
1251 	 * SNP is supported in v2 of the GHCB spec which mandates support for HV
1252 	 * features.
1253 	 */
1254 	if (cc_platform_has(CC_ATTR_GUEST_SEV_SNP)) {
1255 		sev_hv_features = get_hv_features();
1256 
1257 		if (!(sev_hv_features & GHCB_HV_FT_SNP))
1258 			sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
1259 	}
1260 
1261 	/* Initialize per-cpu GHCB pages */
1262 	for_each_possible_cpu(cpu) {
1263 		alloc_runtime_data(cpu);
1264 		init_ghcb(cpu);
1265 	}
1266 
1267 	/* If running under an SVSM, switch to the per-cpu CA */
1268 	if (snp_vmpl) {
1269 		struct svsm_call call = {};
1270 		unsigned long flags;
1271 		int ret;
1272 
1273 		local_irq_save(flags);
1274 
1275 		/*
1276 		 * SVSM_CORE_REMAP_CA call:
1277 		 *   RAX = 0 (Protocol=0, CallID=0)
1278 		 *   RCX = New CA GPA
1279 		 */
1280 		call.caa = svsm_get_caa();
1281 		call.rax = SVSM_CORE_CALL(SVSM_CORE_REMAP_CA);
1282 		call.rcx = this_cpu_read(svsm_caa_pa);
1283 		ret = svsm_perform_call_protocol(&call);
1284 		if (ret)
1285 			panic("Can't remap the SVSM CA, ret=%d, rax_out=0x%llx\n",
1286 			      ret, call.rax_out);
1287 
1288 		sev_cfg.use_cas = true;
1289 
1290 		local_irq_restore(flags);
1291 	}
1292 
1293 	sev_es_setup_play_dead();
1294 
1295 	/* Secondary CPUs use the runtime #VC handler */
1296 	initial_vc_handler = (unsigned long)kernel_exc_vmm_communication;
1297 }
1298 
1299 /*
1300  * SEV-SNP guests should only execute dmi_setup() if EFI_CONFIG_TABLES are
1301  * enabled, as the alternative (fallback) logic for DMI probing in the legacy
1302  * ROM region can cause a crash since this region is not pre-validated.
1303  */
snp_dmi_setup(void)1304 void __init snp_dmi_setup(void)
1305 {
1306 	if (efi_enabled(EFI_CONFIG_TABLES))
1307 		dmi_setup();
1308 }
1309 
dump_cpuid_table(void)1310 static void dump_cpuid_table(void)
1311 {
1312 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
1313 	int i = 0;
1314 
1315 	pr_info("count=%d reserved=0x%x reserved2=0x%llx\n",
1316 		cpuid_table->count, cpuid_table->__reserved1, cpuid_table->__reserved2);
1317 
1318 	for (i = 0; i < SNP_CPUID_COUNT_MAX; i++) {
1319 		const struct snp_cpuid_fn *fn = &cpuid_table->fn[i];
1320 
1321 		pr_info("index=%3d fn=0x%08x subfn=0x%08x: eax=0x%08x ebx=0x%08x ecx=0x%08x edx=0x%08x xcr0_in=0x%016llx xss_in=0x%016llx reserved=0x%016llx\n",
1322 			i, fn->eax_in, fn->ecx_in, fn->eax, fn->ebx, fn->ecx,
1323 			fn->edx, fn->xcr0_in, fn->xss_in, fn->__reserved);
1324 	}
1325 }
1326 
1327 /*
1328  * It is useful from an auditing/testing perspective to provide an easy way
1329  * for the guest owner to know that the CPUID table has been initialized as
1330  * expected, but that initialization happens too early in boot to print any
1331  * sort of indicator, and there's not really any other good place to do it,
1332  * so do it here.
1333  *
1334  * If running as an SNP guest, report the current VM privilege level (VMPL).
1335  */
report_snp_info(void)1336 static int __init report_snp_info(void)
1337 {
1338 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
1339 
1340 	if (cpuid_table->count) {
1341 		pr_info("Using SNP CPUID table, %d entries present.\n",
1342 			cpuid_table->count);
1343 
1344 		if (sev_cfg.debug)
1345 			dump_cpuid_table();
1346 	}
1347 
1348 	if (cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
1349 		pr_info("SNP running at VMPL%u.\n", snp_vmpl);
1350 
1351 	return 0;
1352 }
1353 arch_initcall(report_snp_info);
1354 
update_attest_input(struct svsm_call * call,struct svsm_attest_call * input)1355 static void update_attest_input(struct svsm_call *call, struct svsm_attest_call *input)
1356 {
1357 	/* If (new) lengths have been returned, propagate them up */
1358 	if (call->rcx_out != call->rcx)
1359 		input->manifest_buf.len = call->rcx_out;
1360 
1361 	if (call->rdx_out != call->rdx)
1362 		input->certificates_buf.len = call->rdx_out;
1363 
1364 	if (call->r8_out != call->r8)
1365 		input->report_buf.len = call->r8_out;
1366 }
1367 
snp_issue_svsm_attest_req(u64 call_id,struct svsm_call * call,struct svsm_attest_call * input)1368 int snp_issue_svsm_attest_req(u64 call_id, struct svsm_call *call,
1369 			      struct svsm_attest_call *input)
1370 {
1371 	struct svsm_attest_call *ac;
1372 	unsigned long flags;
1373 	u64 attest_call_pa;
1374 	int ret;
1375 
1376 	if (!snp_vmpl)
1377 		return -EINVAL;
1378 
1379 	local_irq_save(flags);
1380 
1381 	call->caa = svsm_get_caa();
1382 
1383 	ac = (struct svsm_attest_call *)call->caa->svsm_buffer;
1384 	attest_call_pa = svsm_get_caa_pa() + offsetof(struct svsm_ca, svsm_buffer);
1385 
1386 	*ac = *input;
1387 
1388 	/*
1389 	 * Set input registers for the request and set RDX and R8 to known
1390 	 * values in order to detect length values being returned in them.
1391 	 */
1392 	call->rax = call_id;
1393 	call->rcx = attest_call_pa;
1394 	call->rdx = -1;
1395 	call->r8 = -1;
1396 	ret = svsm_perform_call_protocol(call);
1397 	update_attest_input(call, input);
1398 
1399 	local_irq_restore(flags);
1400 
1401 	return ret;
1402 }
1403 EXPORT_SYMBOL_GPL(snp_issue_svsm_attest_req);
1404 
snp_issue_guest_request(struct snp_guest_req * req)1405 static int snp_issue_guest_request(struct snp_guest_req *req)
1406 {
1407 	struct snp_req_data *input = &req->input;
1408 	struct ghcb_state state;
1409 	struct es_em_ctxt ctxt;
1410 	unsigned long flags;
1411 	struct ghcb *ghcb;
1412 	int ret;
1413 
1414 	req->exitinfo2 = SEV_RET_NO_FW_CALL;
1415 
1416 	/*
1417 	 * __sev_get_ghcb() needs to run with IRQs disabled because it is using
1418 	 * a per-CPU GHCB.
1419 	 */
1420 	local_irq_save(flags);
1421 
1422 	ghcb = __sev_get_ghcb(&state);
1423 	if (!ghcb) {
1424 		ret = -EIO;
1425 		goto e_restore_irq;
1426 	}
1427 
1428 	vc_ghcb_invalidate(ghcb);
1429 
1430 	if (req->exit_code == SVM_VMGEXIT_EXT_GUEST_REQUEST) {
1431 		ghcb_set_rax(ghcb, input->data_gpa);
1432 		ghcb_set_rbx(ghcb, input->data_npages);
1433 	}
1434 
1435 	ret = sev_es_ghcb_hv_call(ghcb, &ctxt, req->exit_code, input->req_gpa, input->resp_gpa);
1436 	if (ret)
1437 		goto e_put;
1438 
1439 	req->exitinfo2 = ghcb->save.sw_exit_info_2;
1440 	switch (req->exitinfo2) {
1441 	case 0:
1442 		break;
1443 
1444 	case SNP_GUEST_VMM_ERR(SNP_GUEST_VMM_ERR_BUSY):
1445 		ret = -EAGAIN;
1446 		break;
1447 
1448 	case SNP_GUEST_VMM_ERR(SNP_GUEST_VMM_ERR_INVALID_LEN):
1449 		/* Number of expected pages are returned in RBX */
1450 		if (req->exit_code == SVM_VMGEXIT_EXT_GUEST_REQUEST) {
1451 			input->data_npages = ghcb_get_rbx(ghcb);
1452 			ret = -ENOSPC;
1453 			break;
1454 		}
1455 		fallthrough;
1456 	default:
1457 		ret = -EIO;
1458 		break;
1459 	}
1460 
1461 e_put:
1462 	__sev_put_ghcb(&state);
1463 e_restore_irq:
1464 	local_irq_restore(flags);
1465 
1466 	return ret;
1467 }
1468 
1469 /**
1470  * snp_svsm_vtpm_probe() - Probe if SVSM provides a vTPM device
1471  *
1472  * Check that there is SVSM and that it supports at least TPM_SEND_COMMAND
1473  * which is the only request used so far.
1474  *
1475  * Return: true if the platform provides a vTPM SVSM device, false otherwise.
1476  */
snp_svsm_vtpm_probe(void)1477 static bool snp_svsm_vtpm_probe(void)
1478 {
1479 	struct svsm_call call = {};
1480 
1481 	/* The vTPM device is available only if a SVSM is present */
1482 	if (!snp_vmpl)
1483 		return false;
1484 
1485 	call.caa = svsm_get_caa();
1486 	call.rax = SVSM_VTPM_CALL(SVSM_VTPM_QUERY);
1487 
1488 	if (svsm_perform_call_protocol(&call))
1489 		return false;
1490 
1491 	/* Check platform commands contains TPM_SEND_COMMAND - platform command 8 */
1492 	return call.rcx_out & BIT_ULL(8);
1493 }
1494 
1495 /**
1496  * snp_svsm_vtpm_send_command() - Execute a vTPM operation on SVSM
1497  * @buffer: A buffer used to both send the command and receive the response.
1498  *
1499  * Execute a SVSM_VTPM_CMD call as defined by
1500  * "Secure VM Service Module for SEV-SNP Guests" Publication # 58019 Revision: 1.00
1501  *
1502  * All command request/response buffers have a common structure as specified by
1503  * the following table:
1504  *     Byte      Size       In/Out    Description
1505  *     Offset    (Bytes)
1506  *     0x000     4          In        Platform command
1507  *                          Out       Platform command response size
1508  *
1509  * Each command can build upon this common request/response structure to create
1510  * a structure specific to the command. See include/linux/tpm_svsm.h for more
1511  * details.
1512  *
1513  * Return: 0 on success, -errno on failure
1514  */
snp_svsm_vtpm_send_command(u8 * buffer)1515 int snp_svsm_vtpm_send_command(u8 *buffer)
1516 {
1517 	struct svsm_call call = {};
1518 
1519 	call.caa = svsm_get_caa();
1520 	call.rax = SVSM_VTPM_CALL(SVSM_VTPM_CMD);
1521 	call.rcx = __pa(buffer);
1522 
1523 	return svsm_perform_call_protocol(&call);
1524 }
1525 EXPORT_SYMBOL_GPL(snp_svsm_vtpm_send_command);
1526 
1527 static struct platform_device sev_guest_device = {
1528 	.name		= "sev-guest",
1529 	.id		= -1,
1530 };
1531 
1532 static struct platform_device tpm_svsm_device = {
1533 	.name		= "tpm-svsm",
1534 	.id		= -1,
1535 };
1536 
snp_init_platform_device(void)1537 static int __init snp_init_platform_device(void)
1538 {
1539 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
1540 		return -ENODEV;
1541 
1542 	if (platform_device_register(&sev_guest_device))
1543 		return -ENODEV;
1544 
1545 	if (snp_svsm_vtpm_probe() &&
1546 	    platform_device_register(&tpm_svsm_device))
1547 		return -ENODEV;
1548 
1549 	pr_info("SNP guest platform devices initialized.\n");
1550 	return 0;
1551 }
1552 device_initcall(snp_init_platform_device);
1553 
sev_show_status(void)1554 void sev_show_status(void)
1555 {
1556 	int i;
1557 
1558 	pr_info("Status: ");
1559 	for (i = 0; i < MSR_AMD64_SNP_RESV_BIT; i++) {
1560 		if (sev_status & BIT_ULL(i)) {
1561 			if (!sev_status_feat_names[i])
1562 				continue;
1563 
1564 			pr_cont("%s ", sev_status_feat_names[i]);
1565 		}
1566 	}
1567 	pr_cont("\n");
1568 }
1569 
snp_update_svsm_ca(void)1570 void __init snp_update_svsm_ca(void)
1571 {
1572 	if (!snp_vmpl)
1573 		return;
1574 
1575 	/* Update the CAA to a proper kernel address */
1576 	boot_svsm_caa = &boot_svsm_ca_page;
1577 }
1578 
1579 #ifdef CONFIG_SYSFS
vmpl_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)1580 static ssize_t vmpl_show(struct kobject *kobj,
1581 			 struct kobj_attribute *attr, char *buf)
1582 {
1583 	return sysfs_emit(buf, "%d\n", snp_vmpl);
1584 }
1585 
1586 static struct kobj_attribute vmpl_attr = __ATTR_RO(vmpl);
1587 
1588 static struct attribute *vmpl_attrs[] = {
1589 	&vmpl_attr.attr,
1590 	NULL
1591 };
1592 
1593 static struct attribute_group sev_attr_group = {
1594 	.attrs = vmpl_attrs,
1595 };
1596 
sev_sysfs_init(void)1597 static int __init sev_sysfs_init(void)
1598 {
1599 	struct kobject *sev_kobj;
1600 	struct device *dev_root;
1601 	int ret;
1602 
1603 	if (!cc_platform_has(CC_ATTR_GUEST_SEV_SNP))
1604 		return -ENODEV;
1605 
1606 	dev_root = bus_get_dev_root(&cpu_subsys);
1607 	if (!dev_root)
1608 		return -ENODEV;
1609 
1610 	sev_kobj = kobject_create_and_add("sev", &dev_root->kobj);
1611 	put_device(dev_root);
1612 
1613 	if (!sev_kobj)
1614 		return -ENOMEM;
1615 
1616 	ret = sysfs_create_group(sev_kobj, &sev_attr_group);
1617 	if (ret)
1618 		kobject_put(sev_kobj);
1619 
1620 	return ret;
1621 }
1622 arch_initcall(sev_sysfs_init);
1623 #endif // CONFIG_SYSFS
1624 
free_shared_pages(void * buf,size_t sz)1625 static void free_shared_pages(void *buf, size_t sz)
1626 {
1627 	unsigned int npages = PAGE_ALIGN(sz) >> PAGE_SHIFT;
1628 	int ret;
1629 
1630 	if (!buf)
1631 		return;
1632 
1633 	ret = set_memory_encrypted((unsigned long)buf, npages);
1634 	if (ret) {
1635 		WARN_ONCE(ret, "failed to restore encryption mask (leak it)\n");
1636 		return;
1637 	}
1638 
1639 	__free_pages(virt_to_page(buf), get_order(sz));
1640 }
1641 
alloc_shared_pages(size_t sz)1642 static void *alloc_shared_pages(size_t sz)
1643 {
1644 	unsigned int npages = PAGE_ALIGN(sz) >> PAGE_SHIFT;
1645 	struct page *page;
1646 	int ret;
1647 
1648 	page = alloc_pages(GFP_KERNEL_ACCOUNT, get_order(sz));
1649 	if (!page)
1650 		return NULL;
1651 
1652 	ret = set_memory_decrypted((unsigned long)page_address(page), npages);
1653 	if (ret) {
1654 		pr_err("failed to mark page shared, ret=%d\n", ret);
1655 		__free_pages(page, get_order(sz));
1656 		return NULL;
1657 	}
1658 
1659 	return page_address(page);
1660 }
1661 
get_vmpck(int id,struct snp_secrets_page * secrets,u32 ** seqno)1662 static u8 *get_vmpck(int id, struct snp_secrets_page *secrets, u32 **seqno)
1663 {
1664 	u8 *key = NULL;
1665 
1666 	switch (id) {
1667 	case 0:
1668 		*seqno = &secrets->os_area.msg_seqno_0;
1669 		key = secrets->vmpck0;
1670 		break;
1671 	case 1:
1672 		*seqno = &secrets->os_area.msg_seqno_1;
1673 		key = secrets->vmpck1;
1674 		break;
1675 	case 2:
1676 		*seqno = &secrets->os_area.msg_seqno_2;
1677 		key = secrets->vmpck2;
1678 		break;
1679 	case 3:
1680 		*seqno = &secrets->os_area.msg_seqno_3;
1681 		key = secrets->vmpck3;
1682 		break;
1683 	default:
1684 		break;
1685 	}
1686 
1687 	return key;
1688 }
1689 
snp_init_crypto(u8 * key,size_t keylen)1690 static struct aesgcm_ctx *snp_init_crypto(u8 *key, size_t keylen)
1691 {
1692 	struct aesgcm_ctx *ctx;
1693 
1694 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1695 	if (!ctx)
1696 		return NULL;
1697 
1698 	if (aesgcm_expandkey(ctx, key, keylen, AUTHTAG_LEN)) {
1699 		pr_err("Crypto context initialization failed\n");
1700 		kfree(ctx);
1701 		return NULL;
1702 	}
1703 
1704 	return ctx;
1705 }
1706 
snp_msg_init(struct snp_msg_desc * mdesc,int vmpck_id)1707 int snp_msg_init(struct snp_msg_desc *mdesc, int vmpck_id)
1708 {
1709 	/* Adjust the default VMPCK key based on the executing VMPL level */
1710 	if (vmpck_id == -1)
1711 		vmpck_id = snp_vmpl;
1712 
1713 	mdesc->vmpck = get_vmpck(vmpck_id, mdesc->secrets, &mdesc->os_area_msg_seqno);
1714 	if (!mdesc->vmpck) {
1715 		pr_err("Invalid VMPCK%d communication key\n", vmpck_id);
1716 		return -EINVAL;
1717 	}
1718 
1719 	/* Verify that VMPCK is not zero. */
1720 	if (!memchr_inv(mdesc->vmpck, 0, VMPCK_KEY_LEN)) {
1721 		pr_err("Empty VMPCK%d communication key\n", vmpck_id);
1722 		return -EINVAL;
1723 	}
1724 
1725 	mdesc->vmpck_id = vmpck_id;
1726 
1727 	mdesc->ctx = snp_init_crypto(mdesc->vmpck, VMPCK_KEY_LEN);
1728 	if (!mdesc->ctx)
1729 		return -ENOMEM;
1730 
1731 	return 0;
1732 }
1733 EXPORT_SYMBOL_GPL(snp_msg_init);
1734 
snp_msg_alloc(void)1735 struct snp_msg_desc *snp_msg_alloc(void)
1736 {
1737 	struct snp_msg_desc *mdesc;
1738 	void __iomem *mem;
1739 
1740 	BUILD_BUG_ON(sizeof(struct snp_guest_msg) > PAGE_SIZE);
1741 
1742 	mdesc = kzalloc(sizeof(struct snp_msg_desc), GFP_KERNEL);
1743 	if (!mdesc)
1744 		return ERR_PTR(-ENOMEM);
1745 
1746 	mem = ioremap_encrypted(sev_secrets_pa, PAGE_SIZE);
1747 	if (!mem)
1748 		goto e_free_mdesc;
1749 
1750 	mdesc->secrets = (__force struct snp_secrets_page *)mem;
1751 
1752 	/* Allocate the shared page used for the request and response message. */
1753 	mdesc->request = alloc_shared_pages(sizeof(struct snp_guest_msg));
1754 	if (!mdesc->request)
1755 		goto e_unmap;
1756 
1757 	mdesc->response = alloc_shared_pages(sizeof(struct snp_guest_msg));
1758 	if (!mdesc->response)
1759 		goto e_free_request;
1760 
1761 	return mdesc;
1762 
1763 e_free_request:
1764 	free_shared_pages(mdesc->request, sizeof(struct snp_guest_msg));
1765 e_unmap:
1766 	iounmap(mem);
1767 e_free_mdesc:
1768 	kfree(mdesc);
1769 
1770 	return ERR_PTR(-ENOMEM);
1771 }
1772 EXPORT_SYMBOL_GPL(snp_msg_alloc);
1773 
snp_msg_free(struct snp_msg_desc * mdesc)1774 void snp_msg_free(struct snp_msg_desc *mdesc)
1775 {
1776 	if (!mdesc)
1777 		return;
1778 
1779 	kfree(mdesc->ctx);
1780 	free_shared_pages(mdesc->response, sizeof(struct snp_guest_msg));
1781 	free_shared_pages(mdesc->request, sizeof(struct snp_guest_msg));
1782 	iounmap((__force void __iomem *)mdesc->secrets);
1783 
1784 	memset(mdesc, 0, sizeof(*mdesc));
1785 	kfree(mdesc);
1786 }
1787 EXPORT_SYMBOL_GPL(snp_msg_free);
1788 
1789 /* Mutex to serialize the shared buffer access and command handling. */
1790 static DEFINE_MUTEX(snp_cmd_mutex);
1791 
1792 /*
1793  * If an error is received from the host or AMD Secure Processor (ASP) there
1794  * are two options. Either retry the exact same encrypted request or discontinue
1795  * using the VMPCK.
1796  *
1797  * This is because in the current encryption scheme GHCB v2 uses AES-GCM to
1798  * encrypt the requests. The IV for this scheme is the sequence number. GCM
1799  * cannot tolerate IV reuse.
1800  *
1801  * The ASP FW v1.51 only increments the sequence numbers on a successful
1802  * guest<->ASP back and forth and only accepts messages at its exact sequence
1803  * number.
1804  *
1805  * So if the sequence number were to be reused the encryption scheme is
1806  * vulnerable. If the sequence number were incremented for a fresh IV the ASP
1807  * will reject the request.
1808  */
snp_disable_vmpck(struct snp_msg_desc * mdesc)1809 static void snp_disable_vmpck(struct snp_msg_desc *mdesc)
1810 {
1811 	pr_alert("Disabling VMPCK%d communication key to prevent IV reuse.\n",
1812 		  mdesc->vmpck_id);
1813 	memzero_explicit(mdesc->vmpck, VMPCK_KEY_LEN);
1814 	mdesc->vmpck = NULL;
1815 }
1816 
__snp_get_msg_seqno(struct snp_msg_desc * mdesc)1817 static inline u64 __snp_get_msg_seqno(struct snp_msg_desc *mdesc)
1818 {
1819 	u64 count;
1820 
1821 	lockdep_assert_held(&snp_cmd_mutex);
1822 
1823 	/* Read the current message sequence counter from secrets pages */
1824 	count = *mdesc->os_area_msg_seqno;
1825 
1826 	return count + 1;
1827 }
1828 
1829 /* Return a non-zero on success */
snp_get_msg_seqno(struct snp_msg_desc * mdesc)1830 static u64 snp_get_msg_seqno(struct snp_msg_desc *mdesc)
1831 {
1832 	u64 count = __snp_get_msg_seqno(mdesc);
1833 
1834 	/*
1835 	 * The message sequence counter for the SNP guest request is a  64-bit
1836 	 * value but the version 2 of GHCB specification defines a 32-bit storage
1837 	 * for it. If the counter exceeds the 32-bit value then return zero.
1838 	 * The caller should check the return value, but if the caller happens to
1839 	 * not check the value and use it, then the firmware treats zero as an
1840 	 * invalid number and will fail the  message request.
1841 	 */
1842 	if (count >= UINT_MAX) {
1843 		pr_err("request message sequence counter overflow\n");
1844 		return 0;
1845 	}
1846 
1847 	return count;
1848 }
1849 
snp_inc_msg_seqno(struct snp_msg_desc * mdesc)1850 static void snp_inc_msg_seqno(struct snp_msg_desc *mdesc)
1851 {
1852 	/*
1853 	 * The counter is also incremented by the PSP, so increment it by 2
1854 	 * and save in secrets page.
1855 	 */
1856 	*mdesc->os_area_msg_seqno += 2;
1857 }
1858 
verify_and_dec_payload(struct snp_msg_desc * mdesc,struct snp_guest_req * req)1859 static int verify_and_dec_payload(struct snp_msg_desc *mdesc, struct snp_guest_req *req)
1860 {
1861 	struct snp_guest_msg *resp_msg = &mdesc->secret_response;
1862 	struct snp_guest_msg *req_msg = &mdesc->secret_request;
1863 	struct snp_guest_msg_hdr *req_msg_hdr = &req_msg->hdr;
1864 	struct snp_guest_msg_hdr *resp_msg_hdr = &resp_msg->hdr;
1865 	struct aesgcm_ctx *ctx = mdesc->ctx;
1866 	u8 iv[GCM_AES_IV_SIZE] = {};
1867 
1868 	pr_debug("response [seqno %lld type %d version %d sz %d]\n",
1869 		 resp_msg_hdr->msg_seqno, resp_msg_hdr->msg_type, resp_msg_hdr->msg_version,
1870 		 resp_msg_hdr->msg_sz);
1871 
1872 	/* Copy response from shared memory to encrypted memory. */
1873 	memcpy(resp_msg, mdesc->response, sizeof(*resp_msg));
1874 
1875 	/* Verify that the sequence counter is incremented by 1 */
1876 	if (unlikely(resp_msg_hdr->msg_seqno != (req_msg_hdr->msg_seqno + 1)))
1877 		return -EBADMSG;
1878 
1879 	/* Verify response message type and version number. */
1880 	if (resp_msg_hdr->msg_type != (req_msg_hdr->msg_type + 1) ||
1881 	    resp_msg_hdr->msg_version != req_msg_hdr->msg_version)
1882 		return -EBADMSG;
1883 
1884 	/*
1885 	 * If the message size is greater than our buffer length then return
1886 	 * an error.
1887 	 */
1888 	if (unlikely((resp_msg_hdr->msg_sz + ctx->authsize) > req->resp_sz))
1889 		return -EBADMSG;
1890 
1891 	/* Decrypt the payload */
1892 	memcpy(iv, &resp_msg_hdr->msg_seqno, min(sizeof(iv), sizeof(resp_msg_hdr->msg_seqno)));
1893 	if (!aesgcm_decrypt(ctx, req->resp_buf, resp_msg->payload, resp_msg_hdr->msg_sz,
1894 			    &resp_msg_hdr->algo, AAD_LEN, iv, resp_msg_hdr->authtag))
1895 		return -EBADMSG;
1896 
1897 	return 0;
1898 }
1899 
enc_payload(struct snp_msg_desc * mdesc,u64 seqno,struct snp_guest_req * req)1900 static int enc_payload(struct snp_msg_desc *mdesc, u64 seqno, struct snp_guest_req *req)
1901 {
1902 	struct snp_guest_msg *msg = &mdesc->secret_request;
1903 	struct snp_guest_msg_hdr *hdr = &msg->hdr;
1904 	struct aesgcm_ctx *ctx = mdesc->ctx;
1905 	u8 iv[GCM_AES_IV_SIZE] = {};
1906 
1907 	memset(msg, 0, sizeof(*msg));
1908 
1909 	hdr->algo = SNP_AEAD_AES_256_GCM;
1910 	hdr->hdr_version = MSG_HDR_VER;
1911 	hdr->hdr_sz = sizeof(*hdr);
1912 	hdr->msg_type = req->msg_type;
1913 	hdr->msg_version = req->msg_version;
1914 	hdr->msg_seqno = seqno;
1915 	hdr->msg_vmpck = req->vmpck_id;
1916 	hdr->msg_sz = req->req_sz;
1917 
1918 	/* Verify the sequence number is non-zero */
1919 	if (!hdr->msg_seqno)
1920 		return -ENOSR;
1921 
1922 	pr_debug("request [seqno %lld type %d version %d sz %d]\n",
1923 		 hdr->msg_seqno, hdr->msg_type, hdr->msg_version, hdr->msg_sz);
1924 
1925 	if (WARN_ON((req->req_sz + ctx->authsize) > sizeof(msg->payload)))
1926 		return -EBADMSG;
1927 
1928 	memcpy(iv, &hdr->msg_seqno, min(sizeof(iv), sizeof(hdr->msg_seqno)));
1929 	aesgcm_encrypt(ctx, msg->payload, req->req_buf, req->req_sz, &hdr->algo,
1930 		       AAD_LEN, iv, hdr->authtag);
1931 
1932 	return 0;
1933 }
1934 
__handle_guest_request(struct snp_msg_desc * mdesc,struct snp_guest_req * req)1935 static int __handle_guest_request(struct snp_msg_desc *mdesc, struct snp_guest_req *req)
1936 {
1937 	unsigned long req_start = jiffies;
1938 	unsigned int override_npages = 0;
1939 	u64 override_err = 0;
1940 	int rc;
1941 
1942 retry_request:
1943 	/*
1944 	 * Call firmware to process the request. In this function the encrypted
1945 	 * message enters shared memory with the host. So after this call the
1946 	 * sequence number must be incremented or the VMPCK must be deleted to
1947 	 * prevent reuse of the IV.
1948 	 */
1949 	rc = snp_issue_guest_request(req);
1950 	switch (rc) {
1951 	case -ENOSPC:
1952 		/*
1953 		 * If the extended guest request fails due to having too
1954 		 * small of a certificate data buffer, retry the same
1955 		 * guest request without the extended data request in
1956 		 * order to increment the sequence number and thus avoid
1957 		 * IV reuse.
1958 		 */
1959 		override_npages = req->input.data_npages;
1960 		req->exit_code	= SVM_VMGEXIT_GUEST_REQUEST;
1961 
1962 		/*
1963 		 * Override the error to inform callers the given extended
1964 		 * request buffer size was too small and give the caller the
1965 		 * required buffer size.
1966 		 */
1967 		override_err = SNP_GUEST_VMM_ERR(SNP_GUEST_VMM_ERR_INVALID_LEN);
1968 
1969 		/*
1970 		 * If this call to the firmware succeeds, the sequence number can
1971 		 * be incremented allowing for continued use of the VMPCK. If
1972 		 * there is an error reflected in the return value, this value
1973 		 * is checked further down and the result will be the deletion
1974 		 * of the VMPCK and the error code being propagated back to the
1975 		 * user as an ioctl() return code.
1976 		 */
1977 		goto retry_request;
1978 
1979 	/*
1980 	 * The host may return SNP_GUEST_VMM_ERR_BUSY if the request has been
1981 	 * throttled. Retry in the driver to avoid returning and reusing the
1982 	 * message sequence number on a different message.
1983 	 */
1984 	case -EAGAIN:
1985 		if (jiffies - req_start > SNP_REQ_MAX_RETRY_DURATION) {
1986 			rc = -ETIMEDOUT;
1987 			break;
1988 		}
1989 		schedule_timeout_killable(SNP_REQ_RETRY_DELAY);
1990 		goto retry_request;
1991 	}
1992 
1993 	/*
1994 	 * Increment the message sequence number. There is no harm in doing
1995 	 * this now because decryption uses the value stored in the response
1996 	 * structure and any failure will wipe the VMPCK, preventing further
1997 	 * use anyway.
1998 	 */
1999 	snp_inc_msg_seqno(mdesc);
2000 
2001 	if (override_err) {
2002 		req->exitinfo2 = override_err;
2003 
2004 		/*
2005 		 * If an extended guest request was issued and the supplied certificate
2006 		 * buffer was not large enough, a standard guest request was issued to
2007 		 * prevent IV reuse. If the standard request was successful, return -EIO
2008 		 * back to the caller as would have originally been returned.
2009 		 */
2010 		if (!rc && override_err == SNP_GUEST_VMM_ERR(SNP_GUEST_VMM_ERR_INVALID_LEN))
2011 			rc = -EIO;
2012 	}
2013 
2014 	if (override_npages)
2015 		req->input.data_npages = override_npages;
2016 
2017 	return rc;
2018 }
2019 
snp_send_guest_request(struct snp_msg_desc * mdesc,struct snp_guest_req * req)2020 int snp_send_guest_request(struct snp_msg_desc *mdesc, struct snp_guest_req *req)
2021 {
2022 	u64 seqno;
2023 	int rc;
2024 
2025 	/*
2026 	 * enc_payload() calls aesgcm_encrypt(), which can potentially offload to HW.
2027 	 * The offload's DMA SG list of data to encrypt has to be in linear mapping.
2028 	 */
2029 	if (!virt_addr_valid(req->req_buf) || !virt_addr_valid(req->resp_buf)) {
2030 		pr_warn("AES-GSM buffers must be in linear mapping");
2031 		return -EINVAL;
2032 	}
2033 
2034 	guard(mutex)(&snp_cmd_mutex);
2035 
2036 	/* Check if the VMPCK is not empty */
2037 	if (!mdesc->vmpck || !memchr_inv(mdesc->vmpck, 0, VMPCK_KEY_LEN)) {
2038 		pr_err_ratelimited("VMPCK is disabled\n");
2039 		return -ENOTTY;
2040 	}
2041 
2042 	/* Get message sequence and verify that its a non-zero */
2043 	seqno = snp_get_msg_seqno(mdesc);
2044 	if (!seqno)
2045 		return -EIO;
2046 
2047 	/* Clear shared memory's response for the host to populate. */
2048 	memset(mdesc->response, 0, sizeof(struct snp_guest_msg));
2049 
2050 	/* Encrypt the userspace provided payload in mdesc->secret_request. */
2051 	rc = enc_payload(mdesc, seqno, req);
2052 	if (rc)
2053 		return rc;
2054 
2055 	/*
2056 	 * Write the fully encrypted request to the shared unencrypted
2057 	 * request page.
2058 	 */
2059 	memcpy(mdesc->request, &mdesc->secret_request, sizeof(mdesc->secret_request));
2060 
2061 	/* Initialize the input address for guest request */
2062 	req->input.req_gpa = __pa(mdesc->request);
2063 	req->input.resp_gpa = __pa(mdesc->response);
2064 	req->input.data_gpa = req->certs_data ? __pa(req->certs_data) : 0;
2065 
2066 	rc = __handle_guest_request(mdesc, req);
2067 	if (rc) {
2068 		if (rc == -EIO &&
2069 		    req->exitinfo2 == SNP_GUEST_VMM_ERR(SNP_GUEST_VMM_ERR_INVALID_LEN))
2070 			return rc;
2071 
2072 		pr_alert("Detected error from ASP request. rc: %d, exitinfo2: 0x%llx\n",
2073 			 rc, req->exitinfo2);
2074 
2075 		snp_disable_vmpck(mdesc);
2076 		return rc;
2077 	}
2078 
2079 	rc = verify_and_dec_payload(mdesc, req);
2080 	if (rc) {
2081 		pr_alert("Detected unexpected decode failure from ASP. rc: %d\n", rc);
2082 		snp_disable_vmpck(mdesc);
2083 		return rc;
2084 	}
2085 
2086 	return 0;
2087 }
2088 EXPORT_SYMBOL_GPL(snp_send_guest_request);
2089 
snp_get_tsc_info(void)2090 static int __init snp_get_tsc_info(void)
2091 {
2092 	struct snp_tsc_info_resp *tsc_resp;
2093 	struct snp_tsc_info_req *tsc_req;
2094 	struct snp_msg_desc *mdesc;
2095 	struct snp_guest_req req = {};
2096 	int rc = -ENOMEM;
2097 
2098 	tsc_req = kzalloc(sizeof(*tsc_req), GFP_KERNEL);
2099 	if (!tsc_req)
2100 		return rc;
2101 
2102 	/*
2103 	 * The intermediate response buffer is used while decrypting the
2104 	 * response payload. Make sure that it has enough space to cover
2105 	 * the authtag.
2106 	 */
2107 	tsc_resp = kzalloc(sizeof(*tsc_resp) + AUTHTAG_LEN, GFP_KERNEL);
2108 	if (!tsc_resp)
2109 		goto e_free_tsc_req;
2110 
2111 	mdesc = snp_msg_alloc();
2112 	if (IS_ERR_OR_NULL(mdesc))
2113 		goto e_free_tsc_resp;
2114 
2115 	rc = snp_msg_init(mdesc, snp_vmpl);
2116 	if (rc)
2117 		goto e_free_mdesc;
2118 
2119 	req.msg_version = MSG_HDR_VER;
2120 	req.msg_type = SNP_MSG_TSC_INFO_REQ;
2121 	req.vmpck_id = snp_vmpl;
2122 	req.req_buf = tsc_req;
2123 	req.req_sz = sizeof(*tsc_req);
2124 	req.resp_buf = (void *)tsc_resp;
2125 	req.resp_sz = sizeof(*tsc_resp) + AUTHTAG_LEN;
2126 	req.exit_code = SVM_VMGEXIT_GUEST_REQUEST;
2127 
2128 	rc = snp_send_guest_request(mdesc, &req);
2129 	if (rc)
2130 		goto e_request;
2131 
2132 	pr_debug("%s: response status 0x%x scale 0x%llx offset 0x%llx factor 0x%x\n",
2133 		 __func__, tsc_resp->status, tsc_resp->tsc_scale, tsc_resp->tsc_offset,
2134 		 tsc_resp->tsc_factor);
2135 
2136 	if (!tsc_resp->status) {
2137 		snp_tsc_scale = tsc_resp->tsc_scale;
2138 		snp_tsc_offset = tsc_resp->tsc_offset;
2139 	} else {
2140 		pr_err("Failed to get TSC info, response status 0x%x\n", tsc_resp->status);
2141 		rc = -EIO;
2142 	}
2143 
2144 e_request:
2145 	/* The response buffer contains sensitive data, explicitly clear it. */
2146 	memzero_explicit(tsc_resp, sizeof(*tsc_resp) + AUTHTAG_LEN);
2147 e_free_mdesc:
2148 	snp_msg_free(mdesc);
2149 e_free_tsc_resp:
2150 	kfree(tsc_resp);
2151 e_free_tsc_req:
2152 	kfree(tsc_req);
2153 
2154 	return rc;
2155 }
2156 
snp_secure_tsc_prepare(void)2157 void __init snp_secure_tsc_prepare(void)
2158 {
2159 	if (!cc_platform_has(CC_ATTR_GUEST_SNP_SECURE_TSC))
2160 		return;
2161 
2162 	if (snp_get_tsc_info()) {
2163 		pr_alert("Unable to retrieve Secure TSC info from ASP\n");
2164 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SECURE_TSC);
2165 	}
2166 
2167 	pr_debug("SecureTSC enabled");
2168 }
2169 
securetsc_get_tsc_khz(void)2170 static unsigned long securetsc_get_tsc_khz(void)
2171 {
2172 	return snp_tsc_freq_khz;
2173 }
2174 
snp_secure_tsc_init(void)2175 void __init snp_secure_tsc_init(void)
2176 {
2177 	struct snp_secrets_page *secrets;
2178 	unsigned long tsc_freq_mhz;
2179 	void *mem;
2180 
2181 	if (!cc_platform_has(CC_ATTR_GUEST_SNP_SECURE_TSC))
2182 		return;
2183 
2184 	mem = early_memremap_encrypted(sev_secrets_pa, PAGE_SIZE);
2185 	if (!mem) {
2186 		pr_err("Unable to get TSC_FACTOR: failed to map the SNP secrets page.\n");
2187 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SECURE_TSC);
2188 	}
2189 
2190 	secrets = (__force struct snp_secrets_page *)mem;
2191 
2192 	setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
2193 	rdmsrq(MSR_AMD64_GUEST_TSC_FREQ, tsc_freq_mhz);
2194 
2195 	/* Extract the GUEST TSC MHZ from BIT[17:0], rest is reserved space */
2196 	tsc_freq_mhz &= GENMASK_ULL(17, 0);
2197 
2198 	snp_tsc_freq_khz = SNP_SCALE_TSC_FREQ(tsc_freq_mhz * 1000, secrets->tsc_factor);
2199 
2200 	x86_platform.calibrate_cpu = securetsc_get_tsc_khz;
2201 	x86_platform.calibrate_tsc = securetsc_get_tsc_khz;
2202 
2203 	early_memunmap(mem, PAGE_SIZE);
2204 }
2205