xref: /linux/arch/x86/boot/startup/sev-shared.c (revision a5f03880f06a6da6ea5f1d966fffffcb3fc65462)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * AMD Encrypted Register State Support
4  *
5  * Author: Joerg Roedel <jroedel@suse.de>
6  *
7  * This file is not compiled stand-alone. It contains code shared
8  * between the pre-decompression boot code and the running Linux kernel
9  * and is included directly into both code-bases.
10  */
11 
12 #include <asm/setup_data.h>
13 
14 #ifndef __BOOT_COMPRESSED
15 #define error(v)			pr_err(v)
16 #define has_cpuflag(f)			boot_cpu_has(f)
17 #else
18 #undef WARN
19 #define WARN(condition, format...) (!!(condition))
20 #endif
21 
22 /*
23  * SVSM related information:
24  *   During boot, the page tables are set up as identity mapped and later
25  *   changed to use kernel virtual addresses. Maintain separate virtual and
26  *   physical addresses for the CAA to allow SVSM functions to be used during
27  *   early boot, both with identity mapped virtual addresses and proper kernel
28  *   virtual addresses.
29  */
30 u64 boot_svsm_caa_pa __ro_after_init;
31 
32 /*
33  * Since feature negotiation related variables are set early in the boot
34  * process they must reside in the .data section so as not to be zeroed
35  * out when the .bss section is later cleared.
36  *
37  * GHCB protocol version negotiated with the hypervisor.
38  */
39 u16 ghcb_version __ro_after_init;
40 
41 /* Copy of the SNP firmware's CPUID page. */
42 static struct snp_cpuid_table cpuid_table_copy __ro_after_init;
43 
44 /*
45  * These will be initialized based on CPUID table so that non-present
46  * all-zero leaves (for sparse tables) can be differentiated from
47  * invalid/out-of-range leaves. This is needed since all-zero leaves
48  * still need to be post-processed.
49  */
50 static u32 cpuid_std_range_max __ro_after_init;
51 static u32 cpuid_hyp_range_max __ro_after_init;
52 static u32 cpuid_ext_range_max __ro_after_init;
53 
54 bool sev_snp_needs_sfw;
55 
56 void __head __noreturn
57 sev_es_terminate(unsigned int set, unsigned int reason)
58 {
59 	u64 val = GHCB_MSR_TERM_REQ;
60 
61 	/* Tell the hypervisor what went wrong. */
62 	val |= GHCB_SEV_TERM_REASON(set, reason);
63 
64 	/* Request Guest Termination from Hypervisor */
65 	sev_es_wr_ghcb_msr(val);
66 	VMGEXIT();
67 
68 	while (true)
69 		asm volatile("hlt\n" : : : "memory");
70 }
71 
72 /*
73  * The hypervisor features are available from GHCB version 2 onward.
74  */
75 u64 get_hv_features(void)
76 {
77 	u64 val;
78 
79 	if (ghcb_version < 2)
80 		return 0;
81 
82 	sev_es_wr_ghcb_msr(GHCB_MSR_HV_FT_REQ);
83 	VMGEXIT();
84 
85 	val = sev_es_rd_ghcb_msr();
86 	if (GHCB_RESP_CODE(val) != GHCB_MSR_HV_FT_RESP)
87 		return 0;
88 
89 	return GHCB_MSR_HV_FT_RESP_VAL(val);
90 }
91 
92 int svsm_process_result_codes(struct svsm_call *call)
93 {
94 	switch (call->rax_out) {
95 	case SVSM_SUCCESS:
96 		return 0;
97 	case SVSM_ERR_INCOMPLETE:
98 	case SVSM_ERR_BUSY:
99 		return -EAGAIN;
100 	default:
101 		return -EINVAL;
102 	}
103 }
104 
105 /*
106  * Issue a VMGEXIT to call the SVSM:
107  *   - Load the SVSM register state (RAX, RCX, RDX, R8 and R9)
108  *   - Set the CA call pending field to 1
109  *   - Issue VMGEXIT
110  *   - Save the SVSM return register state (RAX, RCX, RDX, R8 and R9)
111  *   - Perform atomic exchange of the CA call pending field
112  *
113  *   - See the "Secure VM Service Module for SEV-SNP Guests" specification for
114  *     details on the calling convention.
115  *     - The calling convention loosely follows the Microsoft X64 calling
116  *       convention by putting arguments in RCX, RDX, R8 and R9.
117  *     - RAX specifies the SVSM protocol/callid as input and the return code
118  *       as output.
119  */
120 void svsm_issue_call(struct svsm_call *call, u8 *pending)
121 {
122 	register unsigned long rax asm("rax") = call->rax;
123 	register unsigned long rcx asm("rcx") = call->rcx;
124 	register unsigned long rdx asm("rdx") = call->rdx;
125 	register unsigned long r8  asm("r8")  = call->r8;
126 	register unsigned long r9  asm("r9")  = call->r9;
127 
128 	call->caa->call_pending = 1;
129 
130 	asm volatile("rep; vmmcall\n\t"
131 		     : "+r" (rax), "+r" (rcx), "+r" (rdx), "+r" (r8), "+r" (r9)
132 		     : : "memory");
133 
134 	*pending = xchg(&call->caa->call_pending, *pending);
135 
136 	call->rax_out = rax;
137 	call->rcx_out = rcx;
138 	call->rdx_out = rdx;
139 	call->r8_out  = r8;
140 	call->r9_out  = r9;
141 }
142 
143 int svsm_perform_msr_protocol(struct svsm_call *call)
144 {
145 	u8 pending = 0;
146 	u64 val, resp;
147 
148 	/*
149 	 * When using the MSR protocol, be sure to save and restore
150 	 * the current MSR value.
151 	 */
152 	val = sev_es_rd_ghcb_msr();
153 
154 	sev_es_wr_ghcb_msr(GHCB_MSR_VMPL_REQ_LEVEL(0));
155 
156 	svsm_issue_call(call, &pending);
157 
158 	resp = sev_es_rd_ghcb_msr();
159 
160 	sev_es_wr_ghcb_msr(val);
161 
162 	if (pending)
163 		return -EINVAL;
164 
165 	if (GHCB_RESP_CODE(resp) != GHCB_MSR_VMPL_RESP)
166 		return -EINVAL;
167 
168 	if (GHCB_MSR_VMPL_RESP_VAL(resp))
169 		return -EINVAL;
170 
171 	return svsm_process_result_codes(call);
172 }
173 
174 static int __sev_cpuid_hv(u32 fn, int reg_idx, u32 *reg)
175 {
176 	u64 val;
177 
178 	sev_es_wr_ghcb_msr(GHCB_CPUID_REQ(fn, reg_idx));
179 	VMGEXIT();
180 	val = sev_es_rd_ghcb_msr();
181 	if (GHCB_RESP_CODE(val) != GHCB_MSR_CPUID_RESP)
182 		return -EIO;
183 
184 	*reg = (val >> 32);
185 
186 	return 0;
187 }
188 
189 static int __sev_cpuid_hv_msr(struct cpuid_leaf *leaf)
190 {
191 	int ret;
192 
193 	/*
194 	 * MSR protocol does not support fetching non-zero subfunctions, but is
195 	 * sufficient to handle current early-boot cases. Should that change,
196 	 * make sure to report an error rather than ignoring the index and
197 	 * grabbing random values. If this issue arises in the future, handling
198 	 * can be added here to use GHCB-page protocol for cases that occur late
199 	 * enough in boot that GHCB page is available.
200 	 */
201 	if (cpuid_function_is_indexed(leaf->fn) && leaf->subfn)
202 		return -EINVAL;
203 
204 	ret =         __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EAX, &leaf->eax);
205 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EBX, &leaf->ebx);
206 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_ECX, &leaf->ecx);
207 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EDX, &leaf->edx);
208 
209 	return ret;
210 }
211 
212 
213 
214 /*
215  * This may be called early while still running on the initial identity
216  * mapping. Use RIP-relative addressing to obtain the correct address
217  * while running with the initial identity mapping as well as the
218  * switch-over to kernel virtual addresses later.
219  */
220 const struct snp_cpuid_table *snp_cpuid_get_table(void)
221 {
222 	return rip_rel_ptr(&cpuid_table_copy);
223 }
224 
225 /*
226  * The SNP Firmware ABI, Revision 0.9, Section 7.1, details the use of
227  * XCR0_IN and XSS_IN to encode multiple versions of 0xD subfunctions 0
228  * and 1 based on the corresponding features enabled by a particular
229  * combination of XCR0 and XSS registers so that a guest can look up the
230  * version corresponding to the features currently enabled in its XCR0/XSS
231  * registers. The only values that differ between these versions/table
232  * entries is the enabled XSAVE area size advertised via EBX.
233  *
234  * While hypervisors may choose to make use of this support, it is more
235  * robust/secure for a guest to simply find the entry corresponding to the
236  * base/legacy XSAVE area size (XCR0=1 or XCR0=3), and then calculate the
237  * XSAVE area size using subfunctions 2 through 64, as documented in APM
238  * Volume 3, Rev 3.31, Appendix E.3.8, which is what is done here.
239  *
240  * Since base/legacy XSAVE area size is documented as 0x240, use that value
241  * directly rather than relying on the base size in the CPUID table.
242  *
243  * Return: XSAVE area size on success, 0 otherwise.
244  */
245 static u32 __head snp_cpuid_calc_xsave_size(u64 xfeatures_en, bool compacted)
246 {
247 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
248 	u64 xfeatures_found = 0;
249 	u32 xsave_size = 0x240;
250 	int i;
251 
252 	for (i = 0; i < cpuid_table->count; i++) {
253 		const struct snp_cpuid_fn *e = &cpuid_table->fn[i];
254 
255 		if (!(e->eax_in == 0xD && e->ecx_in > 1 && e->ecx_in < 64))
256 			continue;
257 		if (!(xfeatures_en & (BIT_ULL(e->ecx_in))))
258 			continue;
259 		if (xfeatures_found & (BIT_ULL(e->ecx_in)))
260 			continue;
261 
262 		xfeatures_found |= (BIT_ULL(e->ecx_in));
263 
264 		if (compacted)
265 			xsave_size += e->eax;
266 		else
267 			xsave_size = max(xsave_size, e->eax + e->ebx);
268 	}
269 
270 	/*
271 	 * Either the guest set unsupported XCR0/XSS bits, or the corresponding
272 	 * entries in the CPUID table were not present. This is not a valid
273 	 * state to be in.
274 	 */
275 	if (xfeatures_found != (xfeatures_en & GENMASK_ULL(63, 2)))
276 		return 0;
277 
278 	return xsave_size;
279 }
280 
281 static bool __head
282 snp_cpuid_get_validated_func(struct cpuid_leaf *leaf)
283 {
284 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
285 	int i;
286 
287 	for (i = 0; i < cpuid_table->count; i++) {
288 		const struct snp_cpuid_fn *e = &cpuid_table->fn[i];
289 
290 		if (e->eax_in != leaf->fn)
291 			continue;
292 
293 		if (cpuid_function_is_indexed(leaf->fn) && e->ecx_in != leaf->subfn)
294 			continue;
295 
296 		/*
297 		 * For 0xD subfunctions 0 and 1, only use the entry corresponding
298 		 * to the base/legacy XSAVE area size (XCR0=1 or XCR0=3, XSS=0).
299 		 * See the comments above snp_cpuid_calc_xsave_size() for more
300 		 * details.
301 		 */
302 		if (e->eax_in == 0xD && (e->ecx_in == 0 || e->ecx_in == 1))
303 			if (!(e->xcr0_in == 1 || e->xcr0_in == 3) || e->xss_in)
304 				continue;
305 
306 		leaf->eax = e->eax;
307 		leaf->ebx = e->ebx;
308 		leaf->ecx = e->ecx;
309 		leaf->edx = e->edx;
310 
311 		return true;
312 	}
313 
314 	return false;
315 }
316 
317 static void snp_cpuid_hv_msr(void *ctx, struct cpuid_leaf *leaf)
318 {
319 	if (__sev_cpuid_hv_msr(leaf))
320 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID_HV);
321 }
322 
323 static int __head
324 snp_cpuid_postprocess(void (*cpuid_fn)(void *ctx, struct cpuid_leaf *leaf),
325 		      void *ctx, struct cpuid_leaf *leaf)
326 {
327 	struct cpuid_leaf leaf_hv = *leaf;
328 
329 	switch (leaf->fn) {
330 	case 0x1:
331 		cpuid_fn(ctx, &leaf_hv);
332 
333 		/* initial APIC ID */
334 		leaf->ebx = (leaf_hv.ebx & GENMASK(31, 24)) | (leaf->ebx & GENMASK(23, 0));
335 		/* APIC enabled bit */
336 		leaf->edx = (leaf_hv.edx & BIT(9)) | (leaf->edx & ~BIT(9));
337 
338 		/* OSXSAVE enabled bit */
339 		if (native_read_cr4() & X86_CR4_OSXSAVE)
340 			leaf->ecx |= BIT(27);
341 		break;
342 	case 0x7:
343 		/* OSPKE enabled bit */
344 		leaf->ecx &= ~BIT(4);
345 		if (native_read_cr4() & X86_CR4_PKE)
346 			leaf->ecx |= BIT(4);
347 		break;
348 	case 0xB:
349 		leaf_hv.subfn = 0;
350 		cpuid_fn(ctx, &leaf_hv);
351 
352 		/* extended APIC ID */
353 		leaf->edx = leaf_hv.edx;
354 		break;
355 	case 0xD: {
356 		bool compacted = false;
357 		u64 xcr0 = 1, xss = 0;
358 		u32 xsave_size;
359 
360 		if (leaf->subfn != 0 && leaf->subfn != 1)
361 			return 0;
362 
363 		if (native_read_cr4() & X86_CR4_OSXSAVE)
364 			xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
365 		if (leaf->subfn == 1) {
366 			/* Get XSS value if XSAVES is enabled. */
367 			if (leaf->eax & BIT(3)) {
368 				unsigned long lo, hi;
369 
370 				asm volatile("rdmsr" : "=a" (lo), "=d" (hi)
371 						     : "c" (MSR_IA32_XSS));
372 				xss = (hi << 32) | lo;
373 			}
374 
375 			/*
376 			 * The PPR and APM aren't clear on what size should be
377 			 * encoded in 0xD:0x1:EBX when compaction is not enabled
378 			 * by either XSAVEC (feature bit 1) or XSAVES (feature
379 			 * bit 3) since SNP-capable hardware has these feature
380 			 * bits fixed as 1. KVM sets it to 0 in this case, but
381 			 * to avoid this becoming an issue it's safer to simply
382 			 * treat this as unsupported for SNP guests.
383 			 */
384 			if (!(leaf->eax & (BIT(1) | BIT(3))))
385 				return -EINVAL;
386 
387 			compacted = true;
388 		}
389 
390 		xsave_size = snp_cpuid_calc_xsave_size(xcr0 | xss, compacted);
391 		if (!xsave_size)
392 			return -EINVAL;
393 
394 		leaf->ebx = xsave_size;
395 		}
396 		break;
397 	case 0x8000001E:
398 		cpuid_fn(ctx, &leaf_hv);
399 
400 		/* extended APIC ID */
401 		leaf->eax = leaf_hv.eax;
402 		/* compute ID */
403 		leaf->ebx = (leaf->ebx & GENMASK(31, 8)) | (leaf_hv.ebx & GENMASK(7, 0));
404 		/* node ID */
405 		leaf->ecx = (leaf->ecx & GENMASK(31, 8)) | (leaf_hv.ecx & GENMASK(7, 0));
406 		break;
407 	default:
408 		/* No fix-ups needed, use values as-is. */
409 		break;
410 	}
411 
412 	return 0;
413 }
414 
415 /*
416  * Returns -EOPNOTSUPP if feature not enabled. Any other non-zero return value
417  * should be treated as fatal by caller.
418  */
419 int __head snp_cpuid(void (*cpuid_fn)(void *ctx, struct cpuid_leaf *leaf),
420 		     void *ctx, struct cpuid_leaf *leaf)
421 {
422 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
423 
424 	if (!cpuid_table->count)
425 		return -EOPNOTSUPP;
426 
427 	if (!snp_cpuid_get_validated_func(leaf)) {
428 		/*
429 		 * Some hypervisors will avoid keeping track of CPUID entries
430 		 * where all values are zero, since they can be handled the
431 		 * same as out-of-range values (all-zero). This is useful here
432 		 * as well as it allows virtually all guest configurations to
433 		 * work using a single SNP CPUID table.
434 		 *
435 		 * To allow for this, there is a need to distinguish between
436 		 * out-of-range entries and in-range zero entries, since the
437 		 * CPUID table entries are only a template that may need to be
438 		 * augmented with additional values for things like
439 		 * CPU-specific information during post-processing. So if it's
440 		 * not in the table, set the values to zero. Then, if they are
441 		 * within a valid CPUID range, proceed with post-processing
442 		 * using zeros as the initial values. Otherwise, skip
443 		 * post-processing and just return zeros immediately.
444 		 */
445 		leaf->eax = leaf->ebx = leaf->ecx = leaf->edx = 0;
446 
447 		/* Skip post-processing for out-of-range zero leafs. */
448 		if (!(leaf->fn <= cpuid_std_range_max ||
449 		      (leaf->fn >= 0x40000000 && leaf->fn <= cpuid_hyp_range_max) ||
450 		      (leaf->fn >= 0x80000000 && leaf->fn <= cpuid_ext_range_max)))
451 			return 0;
452 	}
453 
454 	return snp_cpuid_postprocess(cpuid_fn, ctx, leaf);
455 }
456 
457 /*
458  * Boot VC Handler - This is the first VC handler during boot, there is no GHCB
459  * page yet, so it only supports the MSR based communication with the
460  * hypervisor and only the CPUID exit-code.
461  */
462 void __head do_vc_no_ghcb(struct pt_regs *regs, unsigned long exit_code)
463 {
464 	unsigned int subfn = lower_bits(regs->cx, 32);
465 	unsigned int fn = lower_bits(regs->ax, 32);
466 	u16 opcode = *(unsigned short *)regs->ip;
467 	struct cpuid_leaf leaf;
468 	int ret;
469 
470 	/* Only CPUID is supported via MSR protocol */
471 	if (exit_code != SVM_EXIT_CPUID)
472 		goto fail;
473 
474 	/* Is it really a CPUID insn? */
475 	if (opcode != 0xa20f)
476 		goto fail;
477 
478 	leaf.fn = fn;
479 	leaf.subfn = subfn;
480 
481 	ret = snp_cpuid(snp_cpuid_hv_msr, NULL, &leaf);
482 	if (!ret)
483 		goto cpuid_done;
484 
485 	if (ret != -EOPNOTSUPP)
486 		goto fail;
487 
488 	if (__sev_cpuid_hv_msr(&leaf))
489 		goto fail;
490 
491 cpuid_done:
492 	regs->ax = leaf.eax;
493 	regs->bx = leaf.ebx;
494 	regs->cx = leaf.ecx;
495 	regs->dx = leaf.edx;
496 
497 	/*
498 	 * This is a VC handler and the #VC is only raised when SEV-ES is
499 	 * active, which means SEV must be active too. Do sanity checks on the
500 	 * CPUID results to make sure the hypervisor does not trick the kernel
501 	 * into the no-sev path. This could map sensitive data unencrypted and
502 	 * make it accessible to the hypervisor.
503 	 *
504 	 * In particular, check for:
505 	 *	- Availability of CPUID leaf 0x8000001f
506 	 *	- SEV CPUID bit.
507 	 *
508 	 * The hypervisor might still report the wrong C-bit position, but this
509 	 * can't be checked here.
510 	 */
511 
512 	if (fn == 0x80000000 && (regs->ax < 0x8000001f))
513 		/* SEV leaf check */
514 		goto fail;
515 	else if ((fn == 0x8000001f && !(regs->ax & BIT(1))))
516 		/* SEV bit */
517 		goto fail;
518 
519 	/* Skip over the CPUID two-byte opcode */
520 	regs->ip += 2;
521 
522 	return;
523 
524 fail:
525 	/* Terminate the guest */
526 	sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
527 }
528 
529 struct cc_setup_data {
530 	struct setup_data header;
531 	u32 cc_blob_address;
532 };
533 
534 /*
535  * Search for a Confidential Computing blob passed in as a setup_data entry
536  * via the Linux Boot Protocol.
537  */
538 static __head
539 struct cc_blob_sev_info *find_cc_blob_setup_data(struct boot_params *bp)
540 {
541 	struct cc_setup_data *sd = NULL;
542 	struct setup_data *hdr;
543 
544 	hdr = (struct setup_data *)bp->hdr.setup_data;
545 
546 	while (hdr) {
547 		if (hdr->type == SETUP_CC_BLOB) {
548 			sd = (struct cc_setup_data *)hdr;
549 			return (struct cc_blob_sev_info *)(unsigned long)sd->cc_blob_address;
550 		}
551 		hdr = (struct setup_data *)hdr->next;
552 	}
553 
554 	return NULL;
555 }
556 
557 /*
558  * Initialize the kernel's copy of the SNP CPUID table, and set up the
559  * pointer that will be used to access it.
560  *
561  * Maintaining a direct mapping of the SNP CPUID table used by firmware would
562  * be possible as an alternative, but the approach is brittle since the
563  * mapping needs to be updated in sync with all the changes to virtual memory
564  * layout and related mapping facilities throughout the boot process.
565  */
566 static void __head setup_cpuid_table(const struct cc_blob_sev_info *cc_info)
567 {
568 	const struct snp_cpuid_table *cpuid_table_fw, *cpuid_table;
569 	int i;
570 
571 	if (!cc_info || !cc_info->cpuid_phys || cc_info->cpuid_len < PAGE_SIZE)
572 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID);
573 
574 	cpuid_table_fw = (const struct snp_cpuid_table *)cc_info->cpuid_phys;
575 	if (!cpuid_table_fw->count || cpuid_table_fw->count > SNP_CPUID_COUNT_MAX)
576 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID);
577 
578 	cpuid_table = snp_cpuid_get_table();
579 	memcpy((void *)cpuid_table, cpuid_table_fw, sizeof(*cpuid_table));
580 
581 	/* Initialize CPUID ranges for range-checking. */
582 	for (i = 0; i < cpuid_table->count; i++) {
583 		const struct snp_cpuid_fn *fn = &cpuid_table->fn[i];
584 
585 		if (fn->eax_in == 0x0)
586 			cpuid_std_range_max = fn->eax;
587 		else if (fn->eax_in == 0x40000000)
588 			cpuid_hyp_range_max = fn->eax;
589 		else if (fn->eax_in == 0x80000000)
590 			cpuid_ext_range_max = fn->eax;
591 	}
592 }
593 
594 static int __head svsm_call_msr_protocol(struct svsm_call *call)
595 {
596 	int ret;
597 
598 	do {
599 		ret = svsm_perform_msr_protocol(call);
600 	} while (ret == -EAGAIN);
601 
602 	return ret;
603 }
604 
605 static void __head svsm_pval_4k_page(unsigned long paddr, bool validate)
606 {
607 	struct svsm_pvalidate_call *pc;
608 	struct svsm_call call = {};
609 	unsigned long flags;
610 	u64 pc_pa;
611 
612 	/*
613 	 * This can be called very early in the boot, use native functions in
614 	 * order to avoid paravirt issues.
615 	 */
616 	flags = native_local_irq_save();
617 
618 	call.caa = svsm_get_caa();
619 
620 	pc = (struct svsm_pvalidate_call *)call.caa->svsm_buffer;
621 	pc_pa = svsm_get_caa_pa() + offsetof(struct svsm_ca, svsm_buffer);
622 
623 	pc->num_entries = 1;
624 	pc->cur_index   = 0;
625 	pc->entry[0].page_size = RMP_PG_SIZE_4K;
626 	pc->entry[0].action    = validate;
627 	pc->entry[0].ignore_cf = 0;
628 	pc->entry[0].rsvd      = 0;
629 	pc->entry[0].pfn       = paddr >> PAGE_SHIFT;
630 
631 	/* Protocol 0, Call ID 1 */
632 	call.rax = SVSM_CORE_CALL(SVSM_CORE_PVALIDATE);
633 	call.rcx = pc_pa;
634 
635 	/*
636 	 * Use the MSR protocol exclusively, so that this code is usable in
637 	 * startup code where VA/PA translations of the GHCB page's address may
638 	 * be problematic.
639 	 */
640 	if (svsm_call_msr_protocol(&call))
641 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PVALIDATE);
642 
643 	native_local_irq_restore(flags);
644 }
645 
646 static void __head pvalidate_4k_page(unsigned long vaddr, unsigned long paddr,
647 				     bool validate)
648 {
649 	int ret;
650 
651 	if (snp_vmpl) {
652 		svsm_pval_4k_page(paddr, validate);
653 	} else {
654 		ret = pvalidate(vaddr, RMP_PG_SIZE_4K, validate);
655 		if (ret)
656 			sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PVALIDATE);
657 	}
658 
659 	/*
660 	 * If validating memory (making it private) and affected by the
661 	 * cache-coherency vulnerability, perform the cache eviction mitigation.
662 	 */
663 	if (validate && sev_snp_needs_sfw)
664 		sev_evict_cache((void *)vaddr, 1);
665 }
666 
667 /*
668  * Maintain the GPA of the SVSM Calling Area (CA) in order to utilize the SVSM
669  * services needed when not running in VMPL0.
670  */
671 static bool __head svsm_setup_ca(const struct cc_blob_sev_info *cc_info,
672 				 void *page)
673 {
674 	struct snp_secrets_page *secrets_page;
675 	struct snp_cpuid_table *cpuid_table;
676 	unsigned int i;
677 	u64 caa;
678 
679 	BUILD_BUG_ON(sizeof(*secrets_page) != PAGE_SIZE);
680 
681 	/*
682 	 * Check if running at VMPL0.
683 	 *
684 	 * Use RMPADJUST (see the rmpadjust() function for a description of what
685 	 * the instruction does) to update the VMPL1 permissions of a page. If
686 	 * the guest is running at VMPL0, this will succeed and implies there is
687 	 * no SVSM. If the guest is running at any other VMPL, this will fail.
688 	 * Linux SNP guests only ever run at a single VMPL level so permission mask
689 	 * changes of a lesser-privileged VMPL are a don't-care.
690 	 *
691 	 * Use a rip-relative reference to obtain the proper address, since this
692 	 * routine is running identity mapped when called, both by the decompressor
693 	 * code and the early kernel code.
694 	 */
695 	if (!rmpadjust((unsigned long)page, RMP_PG_SIZE_4K, 1))
696 		return false;
697 
698 	/*
699 	 * Not running at VMPL0, ensure everything has been properly supplied
700 	 * for running under an SVSM.
701 	 */
702 	if (!cc_info || !cc_info->secrets_phys || cc_info->secrets_len != PAGE_SIZE)
703 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SECRETS_PAGE);
704 
705 	secrets_page = (struct snp_secrets_page *)cc_info->secrets_phys;
706 	if (!secrets_page->svsm_size)
707 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_NO_SVSM);
708 
709 	if (!secrets_page->svsm_guest_vmpl)
710 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SVSM_VMPL0);
711 
712 	snp_vmpl = secrets_page->svsm_guest_vmpl;
713 
714 	caa = secrets_page->svsm_caa;
715 
716 	/*
717 	 * An open-coded PAGE_ALIGNED() in order to avoid including
718 	 * kernel-proper headers into the decompressor.
719 	 */
720 	if (caa & (PAGE_SIZE - 1))
721 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SVSM_CAA);
722 
723 	boot_svsm_caa_pa = caa;
724 
725 	/* Advertise the SVSM presence via CPUID. */
726 	cpuid_table = (struct snp_cpuid_table *)snp_cpuid_get_table();
727 	for (i = 0; i < cpuid_table->count; i++) {
728 		struct snp_cpuid_fn *fn = &cpuid_table->fn[i];
729 
730 		if (fn->eax_in == 0x8000001f)
731 			fn->eax |= BIT(28);
732 	}
733 
734 	return true;
735 }
736