xref: /linux/arch/x86/boot/startup/sev-shared.c (revision 22bdd6e68bbe270a916233ec5f34a13ae5e80ed9)
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 has_cpuflag(f)			boot_cpu_has(f)
16 #else
17 #undef WARN
18 #define WARN(condition, format...) (!!(condition))
19 #endif
20 
21 /* Copy of the SNP firmware's CPUID page. */
22 static struct snp_cpuid_table cpuid_table_copy __ro_after_init;
23 
24 /*
25  * These will be initialized based on CPUID table so that non-present
26  * all-zero leaves (for sparse tables) can be differentiated from
27  * invalid/out-of-range leaves. This is needed since all-zero leaves
28  * still need to be post-processed.
29  */
30 static u32 cpuid_std_range_max __ro_after_init;
31 static u32 cpuid_hyp_range_max __ro_after_init;
32 static u32 cpuid_ext_range_max __ro_after_init;
33 
34 bool sev_snp_needs_sfw;
35 
36 void __noreturn
sev_es_terminate(unsigned int set,unsigned int reason)37 sev_es_terminate(unsigned int set, unsigned int reason)
38 {
39 	u64 val = GHCB_MSR_TERM_REQ;
40 
41 	/* Tell the hypervisor what went wrong. */
42 	val |= GHCB_SEV_TERM_REASON(set, reason);
43 
44 	/* Request Guest Termination from Hypervisor */
45 	sev_es_wr_ghcb_msr(val);
46 	VMGEXIT();
47 
48 	while (true)
49 		asm volatile("hlt\n" : : : "memory");
50 }
51 
52 /*
53  * The hypervisor features are available from GHCB version 2 onward.
54  */
get_hv_features(void)55 u64 __init get_hv_features(void)
56 {
57 	u64 val;
58 
59 	if (ghcb_version < 2)
60 		return 0;
61 
62 	sev_es_wr_ghcb_msr(GHCB_MSR_HV_FT_REQ);
63 	VMGEXIT();
64 
65 	val = sev_es_rd_ghcb_msr();
66 	if (GHCB_RESP_CODE(val) != GHCB_MSR_HV_FT_RESP)
67 		return 0;
68 
69 	return GHCB_MSR_HV_FT_RESP_VAL(val);
70 }
71 
svsm_process_result_codes(struct svsm_call * call)72 int svsm_process_result_codes(struct svsm_call *call)
73 {
74 	switch (call->rax_out) {
75 	case SVSM_SUCCESS:
76 		return 0;
77 	case SVSM_ERR_INCOMPLETE:
78 	case SVSM_ERR_BUSY:
79 		return -EAGAIN;
80 	default:
81 		return -EINVAL;
82 	}
83 }
84 
85 /*
86  * Issue a VMGEXIT to call the SVSM:
87  *   - Load the SVSM register state (RAX, RCX, RDX, R8 and R9)
88  *   - Set the CA call pending field to 1
89  *   - Issue VMGEXIT
90  *   - Save the SVSM return register state (RAX, RCX, RDX, R8 and R9)
91  *   - Perform atomic exchange of the CA call pending field
92  *
93  *   - See the "Secure VM Service Module for SEV-SNP Guests" specification for
94  *     details on the calling convention.
95  *     - The calling convention loosely follows the Microsoft X64 calling
96  *       convention by putting arguments in RCX, RDX, R8 and R9.
97  *     - RAX specifies the SVSM protocol/callid as input and the return code
98  *       as output.
99  */
svsm_issue_call(struct svsm_call * call,u8 * pending)100 void svsm_issue_call(struct svsm_call *call, u8 *pending)
101 {
102 	register unsigned long rax asm("rax") = call->rax;
103 	register unsigned long rcx asm("rcx") = call->rcx;
104 	register unsigned long rdx asm("rdx") = call->rdx;
105 	register unsigned long r8  asm("r8")  = call->r8;
106 	register unsigned long r9  asm("r9")  = call->r9;
107 
108 	call->caa->call_pending = 1;
109 
110 	asm volatile("rep; vmmcall\n\t"
111 		     : "+r" (rax), "+r" (rcx), "+r" (rdx), "+r" (r8), "+r" (r9)
112 		     : : "memory");
113 
114 	*pending = xchg(&call->caa->call_pending, *pending);
115 
116 	call->rax_out = rax;
117 	call->rcx_out = rcx;
118 	call->rdx_out = rdx;
119 	call->r8_out  = r8;
120 	call->r9_out  = r9;
121 }
122 
svsm_perform_msr_protocol(struct svsm_call * call)123 int svsm_perform_msr_protocol(struct svsm_call *call)
124 {
125 	u8 pending = 0;
126 	u64 val, resp;
127 
128 	/*
129 	 * When using the MSR protocol, be sure to save and restore
130 	 * the current MSR value.
131 	 */
132 	val = sev_es_rd_ghcb_msr();
133 
134 	sev_es_wr_ghcb_msr(GHCB_MSR_VMPL_REQ_LEVEL(0));
135 
136 	svsm_issue_call(call, &pending);
137 
138 	resp = sev_es_rd_ghcb_msr();
139 
140 	sev_es_wr_ghcb_msr(val);
141 
142 	if (pending)
143 		return -EINVAL;
144 
145 	if (GHCB_RESP_CODE(resp) != GHCB_MSR_VMPL_RESP)
146 		return -EINVAL;
147 
148 	if (GHCB_MSR_VMPL_RESP_VAL(resp))
149 		return -EINVAL;
150 
151 	return svsm_process_result_codes(call);
152 }
153 
__sev_cpuid_hv(u32 fn,int reg_idx,u32 * reg)154 static int __sev_cpuid_hv(u32 fn, int reg_idx, u32 *reg)
155 {
156 	u64 val;
157 
158 	sev_es_wr_ghcb_msr(GHCB_CPUID_REQ(fn, reg_idx));
159 	VMGEXIT();
160 	val = sev_es_rd_ghcb_msr();
161 	if (GHCB_RESP_CODE(val) != GHCB_MSR_CPUID_RESP)
162 		return -EIO;
163 
164 	*reg = (val >> 32);
165 
166 	return 0;
167 }
168 
__sev_cpuid_hv_msr(struct cpuid_leaf * leaf)169 static int __sev_cpuid_hv_msr(struct cpuid_leaf *leaf)
170 {
171 	int ret;
172 
173 	/*
174 	 * MSR protocol does not support fetching non-zero subfunctions, but is
175 	 * sufficient to handle current early-boot cases. Should that change,
176 	 * make sure to report an error rather than ignoring the index and
177 	 * grabbing random values. If this issue arises in the future, handling
178 	 * can be added here to use GHCB-page protocol for cases that occur late
179 	 * enough in boot that GHCB page is available.
180 	 */
181 	if (cpuid_function_is_indexed(leaf->fn) && leaf->subfn)
182 		return -EINVAL;
183 
184 	ret =         __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EAX, &leaf->eax);
185 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EBX, &leaf->ebx);
186 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_ECX, &leaf->ecx);
187 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EDX, &leaf->edx);
188 
189 	return ret;
190 }
191 
192 
193 
194 /*
195  * This may be called early while still running on the initial identity
196  * mapping. Use RIP-relative addressing to obtain the correct address
197  * while running with the initial identity mapping as well as the
198  * switch-over to kernel virtual addresses later.
199  */
snp_cpuid_get_table(void)200 const struct snp_cpuid_table *snp_cpuid_get_table(void)
201 {
202 	return rip_rel_ptr(&cpuid_table_copy);
203 }
204 
205 /*
206  * The SNP Firmware ABI, Revision 0.9, Section 7.1, details the use of
207  * XCR0_IN and XSS_IN to encode multiple versions of 0xD subfunctions 0
208  * and 1 based on the corresponding features enabled by a particular
209  * combination of XCR0 and XSS registers so that a guest can look up the
210  * version corresponding to the features currently enabled in its XCR0/XSS
211  * registers. The only values that differ between these versions/table
212  * entries is the enabled XSAVE area size advertised via EBX.
213  *
214  * While hypervisors may choose to make use of this support, it is more
215  * robust/secure for a guest to simply find the entry corresponding to the
216  * base/legacy XSAVE area size (XCR0=1 or XCR0=3), and then calculate the
217  * XSAVE area size using subfunctions 2 through 64, as documented in APM
218  * Volume 3, Rev 3.31, Appendix E.3.8, which is what is done here.
219  *
220  * Since base/legacy XSAVE area size is documented as 0x240, use that value
221  * directly rather than relying on the base size in the CPUID table.
222  *
223  * Return: XSAVE area size on success, 0 otherwise.
224  */
snp_cpuid_calc_xsave_size(u64 xfeatures_en,bool compacted)225 static u32 snp_cpuid_calc_xsave_size(u64 xfeatures_en, bool compacted)
226 {
227 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
228 	u64 xfeatures_found = 0;
229 	u32 xsave_size = 0x240;
230 	int i;
231 
232 	for (i = 0; i < cpuid_table->count; i++) {
233 		const struct snp_cpuid_fn *e = &cpuid_table->fn[i];
234 
235 		if (!(e->eax_in == 0xD && e->ecx_in > 1 && e->ecx_in < 64))
236 			continue;
237 		if (!(xfeatures_en & (BIT_ULL(e->ecx_in))))
238 			continue;
239 		if (xfeatures_found & (BIT_ULL(e->ecx_in)))
240 			continue;
241 
242 		xfeatures_found |= (BIT_ULL(e->ecx_in));
243 
244 		if (compacted)
245 			xsave_size += e->eax;
246 		else
247 			xsave_size = max(xsave_size, e->eax + e->ebx);
248 	}
249 
250 	/*
251 	 * Either the guest set unsupported XCR0/XSS bits, or the corresponding
252 	 * entries in the CPUID table were not present. This is not a valid
253 	 * state to be in.
254 	 */
255 	if (xfeatures_found != (xfeatures_en & GENMASK_ULL(63, 2)))
256 		return 0;
257 
258 	return xsave_size;
259 }
260 
261 static bool
snp_cpuid_get_validated_func(struct cpuid_leaf * leaf)262 snp_cpuid_get_validated_func(struct cpuid_leaf *leaf)
263 {
264 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
265 	int i;
266 
267 	for (i = 0; i < cpuid_table->count; i++) {
268 		const struct snp_cpuid_fn *e = &cpuid_table->fn[i];
269 
270 		if (e->eax_in != leaf->fn)
271 			continue;
272 
273 		if (cpuid_function_is_indexed(leaf->fn) && e->ecx_in != leaf->subfn)
274 			continue;
275 
276 		/*
277 		 * For 0xD subfunctions 0 and 1, only use the entry corresponding
278 		 * to the base/legacy XSAVE area size (XCR0=1 or XCR0=3, XSS=0).
279 		 * See the comments above snp_cpuid_calc_xsave_size() for more
280 		 * details.
281 		 */
282 		if (e->eax_in == 0xD && (e->ecx_in == 0 || e->ecx_in == 1))
283 			if (!(e->xcr0_in == 1 || e->xcr0_in == 3) || e->xss_in)
284 				continue;
285 
286 		leaf->eax = e->eax;
287 		leaf->ebx = e->ebx;
288 		leaf->ecx = e->ecx;
289 		leaf->edx = e->edx;
290 
291 		return true;
292 	}
293 
294 	return false;
295 }
296 
snp_cpuid_hv_msr(void * ctx,struct cpuid_leaf * leaf)297 static void snp_cpuid_hv_msr(void *ctx, struct cpuid_leaf *leaf)
298 {
299 	if (__sev_cpuid_hv_msr(leaf))
300 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID_HV);
301 }
302 
303 static int
snp_cpuid_postprocess(void (* cpuid_fn)(void * ctx,struct cpuid_leaf * leaf),void * ctx,struct cpuid_leaf * leaf)304 snp_cpuid_postprocess(void (*cpuid_fn)(void *ctx, struct cpuid_leaf *leaf),
305 		      void *ctx, struct cpuid_leaf *leaf)
306 {
307 	struct cpuid_leaf leaf_hv = *leaf;
308 
309 	switch (leaf->fn) {
310 	case 0x1:
311 		cpuid_fn(ctx, &leaf_hv);
312 
313 		/* initial APIC ID */
314 		leaf->ebx = (leaf_hv.ebx & GENMASK(31, 24)) | (leaf->ebx & GENMASK(23, 0));
315 		/* APIC enabled bit */
316 		leaf->edx = (leaf_hv.edx & BIT(9)) | (leaf->edx & ~BIT(9));
317 
318 		/* OSXSAVE enabled bit */
319 		if (native_read_cr4() & X86_CR4_OSXSAVE)
320 			leaf->ecx |= BIT(27);
321 		break;
322 	case 0x7:
323 		/* OSPKE enabled bit */
324 		leaf->ecx &= ~BIT(4);
325 		if (native_read_cr4() & X86_CR4_PKE)
326 			leaf->ecx |= BIT(4);
327 		break;
328 	case 0xB:
329 		leaf_hv.subfn = 0;
330 		cpuid_fn(ctx, &leaf_hv);
331 
332 		/* extended APIC ID */
333 		leaf->edx = leaf_hv.edx;
334 		break;
335 	case 0xD: {
336 		bool compacted = false;
337 		u64 xcr0 = 1, xss = 0;
338 		u32 xsave_size;
339 
340 		if (leaf->subfn != 0 && leaf->subfn != 1)
341 			return 0;
342 
343 		if (native_read_cr4() & X86_CR4_OSXSAVE)
344 			xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
345 		if (leaf->subfn == 1) {
346 			/* Get XSS value if XSAVES is enabled. */
347 			if (leaf->eax & BIT(3)) {
348 				unsigned long lo, hi;
349 
350 				asm volatile("rdmsr" : "=a" (lo), "=d" (hi)
351 						     : "c" (MSR_IA32_XSS));
352 				xss = (hi << 32) | lo;
353 			}
354 
355 			/*
356 			 * The PPR and APM aren't clear on what size should be
357 			 * encoded in 0xD:0x1:EBX when compaction is not enabled
358 			 * by either XSAVEC (feature bit 1) or XSAVES (feature
359 			 * bit 3) since SNP-capable hardware has these feature
360 			 * bits fixed as 1. KVM sets it to 0 in this case, but
361 			 * to avoid this becoming an issue it's safer to simply
362 			 * treat this as unsupported for SNP guests.
363 			 */
364 			if (!(leaf->eax & (BIT(1) | BIT(3))))
365 				return -EINVAL;
366 
367 			compacted = true;
368 		}
369 
370 		xsave_size = snp_cpuid_calc_xsave_size(xcr0 | xss, compacted);
371 		if (!xsave_size)
372 			return -EINVAL;
373 
374 		leaf->ebx = xsave_size;
375 		}
376 		break;
377 	case 0x8000001E:
378 		cpuid_fn(ctx, &leaf_hv);
379 
380 		/* extended APIC ID */
381 		leaf->eax = leaf_hv.eax;
382 		/* compute ID */
383 		leaf->ebx = (leaf->ebx & GENMASK(31, 8)) | (leaf_hv.ebx & GENMASK(7, 0));
384 		/* node ID */
385 		leaf->ecx = (leaf->ecx & GENMASK(31, 8)) | (leaf_hv.ecx & GENMASK(7, 0));
386 		break;
387 	default:
388 		/* No fix-ups needed, use values as-is. */
389 		break;
390 	}
391 
392 	return 0;
393 }
394 
395 /*
396  * Returns -EOPNOTSUPP if feature not enabled. Any other non-zero return value
397  * should be treated as fatal by caller.
398  */
snp_cpuid(void (* cpuid_fn)(void * ctx,struct cpuid_leaf * leaf),void * ctx,struct cpuid_leaf * leaf)399 int snp_cpuid(void (*cpuid_fn)(void *ctx, struct cpuid_leaf *leaf),
400 	      void *ctx, struct cpuid_leaf *leaf)
401 {
402 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
403 
404 	if (!cpuid_table->count)
405 		return -EOPNOTSUPP;
406 
407 	if (!snp_cpuid_get_validated_func(leaf)) {
408 		/*
409 		 * Some hypervisors will avoid keeping track of CPUID entries
410 		 * where all values are zero, since they can be handled the
411 		 * same as out-of-range values (all-zero). This is useful here
412 		 * as well as it allows virtually all guest configurations to
413 		 * work using a single SNP CPUID table.
414 		 *
415 		 * To allow for this, there is a need to distinguish between
416 		 * out-of-range entries and in-range zero entries, since the
417 		 * CPUID table entries are only a template that may need to be
418 		 * augmented with additional values for things like
419 		 * CPU-specific information during post-processing. So if it's
420 		 * not in the table, set the values to zero. Then, if they are
421 		 * within a valid CPUID range, proceed with post-processing
422 		 * using zeros as the initial values. Otherwise, skip
423 		 * post-processing and just return zeros immediately.
424 		 */
425 		leaf->eax = leaf->ebx = leaf->ecx = leaf->edx = 0;
426 
427 		/* Skip post-processing for out-of-range zero leafs. */
428 		if (!(leaf->fn <= cpuid_std_range_max ||
429 		      (leaf->fn >= 0x40000000 && leaf->fn <= cpuid_hyp_range_max) ||
430 		      (leaf->fn >= 0x80000000 && leaf->fn <= cpuid_ext_range_max)))
431 			return 0;
432 	}
433 
434 	return snp_cpuid_postprocess(cpuid_fn, ctx, leaf);
435 }
436 
437 /*
438  * Boot VC Handler - This is the first VC handler during boot, there is no GHCB
439  * page yet, so it only supports the MSR based communication with the
440  * hypervisor and only the CPUID exit-code.
441  */
do_vc_no_ghcb(struct pt_regs * regs,unsigned long exit_code)442 void do_vc_no_ghcb(struct pt_regs *regs, unsigned long exit_code)
443 {
444 	unsigned int subfn = lower_bits(regs->cx, 32);
445 	unsigned int fn = lower_bits(regs->ax, 32);
446 	u16 opcode = *(unsigned short *)regs->ip;
447 	struct cpuid_leaf leaf;
448 	int ret;
449 
450 	/* Only CPUID is supported via MSR protocol */
451 	if (exit_code != SVM_EXIT_CPUID)
452 		goto fail;
453 
454 	/* Is it really a CPUID insn? */
455 	if (opcode != 0xa20f)
456 		goto fail;
457 
458 	leaf.fn = fn;
459 	leaf.subfn = subfn;
460 
461 	/*
462 	 * If SNP is active, then snp_cpuid() uses the CPUID table to obtain the
463 	 * CPUID values (with possible HV interaction during post-processing of
464 	 * the values). But if SNP is not active (no CPUID table present), then
465 	 * snp_cpuid() returns -EOPNOTSUPP so that an SEV-ES guest can call the
466 	 * HV to obtain the CPUID information.
467 	 */
468 	ret = snp_cpuid(snp_cpuid_hv_msr, NULL, &leaf);
469 	if (!ret)
470 		goto cpuid_done;
471 
472 	if (ret != -EOPNOTSUPP)
473 		goto fail;
474 
475 	/*
476 	 * This is reached by a SEV-ES guest and needs to invoke the HV for
477 	 * the CPUID data.
478 	 */
479 	if (__sev_cpuid_hv_msr(&leaf))
480 		goto fail;
481 
482 cpuid_done:
483 	regs->ax = leaf.eax;
484 	regs->bx = leaf.ebx;
485 	regs->cx = leaf.ecx;
486 	regs->dx = leaf.edx;
487 
488 	/*
489 	 * This is a VC handler and the #VC is only raised when SEV-ES is
490 	 * active, which means SEV must be active too. Do sanity checks on the
491 	 * CPUID results to make sure the hypervisor does not trick the kernel
492 	 * into the no-sev path. This could map sensitive data unencrypted and
493 	 * make it accessible to the hypervisor.
494 	 *
495 	 * In particular, check for:
496 	 *	- Availability of CPUID leaf 0x8000001f
497 	 *	- SEV CPUID bit.
498 	 *
499 	 * The hypervisor might still report the wrong C-bit position, but this
500 	 * can't be checked here.
501 	 */
502 
503 	if (fn == 0x80000000 && (regs->ax < 0x8000001f))
504 		/* SEV leaf check */
505 		goto fail;
506 	else if ((fn == 0x8000001f && !(regs->ax & BIT(1))))
507 		/* SEV bit */
508 		goto fail;
509 
510 	/* Skip over the CPUID two-byte opcode */
511 	regs->ip += 2;
512 
513 	return;
514 
515 fail:
516 	/* Terminate the guest */
517 	sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
518 }
519 
520 struct cc_setup_data {
521 	struct setup_data header;
522 	u32 cc_blob_address;
523 };
524 
525 /*
526  * Search for a Confidential Computing blob passed in as a setup_data entry
527  * via the Linux Boot Protocol.
528  */
529 static __init
find_cc_blob_setup_data(struct boot_params * bp)530 struct cc_blob_sev_info *find_cc_blob_setup_data(struct boot_params *bp)
531 {
532 	struct cc_setup_data *sd = NULL;
533 	struct setup_data *hdr;
534 
535 	hdr = (struct setup_data *)bp->hdr.setup_data;
536 
537 	while (hdr) {
538 		if (hdr->type == SETUP_CC_BLOB) {
539 			sd = (struct cc_setup_data *)hdr;
540 			return (struct cc_blob_sev_info *)(unsigned long)sd->cc_blob_address;
541 		}
542 		hdr = (struct setup_data *)hdr->next;
543 	}
544 
545 	return NULL;
546 }
547 
548 /*
549  * Initialize the kernel's copy of the SNP CPUID table, and set up the
550  * pointer that will be used to access it.
551  *
552  * Maintaining a direct mapping of the SNP CPUID table used by firmware would
553  * be possible as an alternative, but the approach is brittle since the
554  * mapping needs to be updated in sync with all the changes to virtual memory
555  * layout and related mapping facilities throughout the boot process.
556  */
setup_cpuid_table(const struct cc_blob_sev_info * cc_info)557 static void __init setup_cpuid_table(const struct cc_blob_sev_info *cc_info)
558 {
559 	const struct snp_cpuid_table *cpuid_table_fw, *cpuid_table;
560 	int i;
561 
562 	if (!cc_info || !cc_info->cpuid_phys || cc_info->cpuid_len < PAGE_SIZE)
563 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID);
564 
565 	cpuid_table_fw = (const struct snp_cpuid_table *)cc_info->cpuid_phys;
566 	if (!cpuid_table_fw->count || cpuid_table_fw->count > SNP_CPUID_COUNT_MAX)
567 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID);
568 
569 	cpuid_table = snp_cpuid_get_table();
570 	memcpy((void *)cpuid_table, cpuid_table_fw, sizeof(*cpuid_table));
571 
572 	/* Initialize CPUID ranges for range-checking. */
573 	for (i = 0; i < cpuid_table->count; i++) {
574 		const struct snp_cpuid_fn *fn = &cpuid_table->fn[i];
575 
576 		if (fn->eax_in == 0x0)
577 			cpuid_std_range_max = fn->eax;
578 		else if (fn->eax_in == 0x40000000)
579 			cpuid_hyp_range_max = fn->eax;
580 		else if (fn->eax_in == 0x80000000)
581 			cpuid_ext_range_max = fn->eax;
582 	}
583 }
584 
svsm_call_msr_protocol(struct svsm_call * call)585 static int svsm_call_msr_protocol(struct svsm_call *call)
586 {
587 	int ret;
588 
589 	do {
590 		ret = svsm_perform_msr_protocol(call);
591 	} while (ret == -EAGAIN);
592 
593 	return ret;
594 }
595 
svsm_pval_4k_page(unsigned long paddr,bool validate,struct svsm_ca * caa,u64 caa_pa)596 static void svsm_pval_4k_page(unsigned long paddr, bool validate,
597 			      struct svsm_ca *caa, u64 caa_pa)
598 {
599 	struct svsm_pvalidate_call *pc;
600 	struct svsm_call call = {};
601 	unsigned long flags;
602 	u64 pc_pa;
603 
604 	/*
605 	 * This can be called very early in the boot, use native functions in
606 	 * order to avoid paravirt issues.
607 	 */
608 	flags = native_local_irq_save();
609 
610 	call.caa = caa;
611 
612 	pc = (struct svsm_pvalidate_call *)call.caa->svsm_buffer;
613 	pc_pa = caa_pa + offsetof(struct svsm_ca, svsm_buffer);
614 
615 	pc->num_entries = 1;
616 	pc->cur_index   = 0;
617 	pc->entry[0].page_size = RMP_PG_SIZE_4K;
618 	pc->entry[0].action    = validate;
619 	pc->entry[0].ignore_cf = 0;
620 	pc->entry[0].rsvd      = 0;
621 	pc->entry[0].pfn       = paddr >> PAGE_SHIFT;
622 
623 	/* Protocol 0, Call ID 1 */
624 	call.rax = SVSM_CORE_CALL(SVSM_CORE_PVALIDATE);
625 	call.rcx = pc_pa;
626 
627 	/*
628 	 * Use the MSR protocol exclusively, so that this code is usable in
629 	 * startup code where VA/PA translations of the GHCB page's address may
630 	 * be problematic.
631 	 */
632 	if (svsm_call_msr_protocol(&call))
633 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PVALIDATE);
634 
635 	native_local_irq_restore(flags);
636 }
637 
pvalidate_4k_page(unsigned long vaddr,unsigned long paddr,bool validate,struct svsm_ca * caa,u64 caa_pa)638 static void pvalidate_4k_page(unsigned long vaddr, unsigned long paddr,
639 			      bool validate, struct svsm_ca *caa, u64 caa_pa)
640 {
641 	int ret;
642 
643 	if (snp_vmpl) {
644 		svsm_pval_4k_page(paddr, validate, caa, caa_pa);
645 	} else {
646 		ret = pvalidate(vaddr, RMP_PG_SIZE_4K, validate);
647 		if (ret)
648 			sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PVALIDATE);
649 	}
650 
651 	/*
652 	 * If validating memory (making it private) and affected by the
653 	 * cache-coherency vulnerability, perform the cache eviction mitigation.
654 	 */
655 	if (validate && sev_snp_needs_sfw)
656 		sev_evict_cache((void *)vaddr, 1);
657 }
658 
__page_state_change(unsigned long vaddr,unsigned long paddr,const struct psc_desc * desc)659 static void __page_state_change(unsigned long vaddr, unsigned long paddr,
660 			        const struct psc_desc *desc)
661 {
662 	u64 val, msr;
663 
664 	/*
665 	 * If private -> shared then invalidate the page before requesting the
666 	 * state change in the RMP table.
667 	 */
668 	if (desc->op == SNP_PAGE_STATE_SHARED)
669 		pvalidate_4k_page(vaddr, paddr, false, desc->ca, desc->caa_pa);
670 
671 	/* Save the current GHCB MSR value */
672 	msr = sev_es_rd_ghcb_msr();
673 
674 	/* Issue VMGEXIT to change the page state in RMP table. */
675 	sev_es_wr_ghcb_msr(GHCB_MSR_PSC_REQ_GFN(paddr >> PAGE_SHIFT, desc->op));
676 	VMGEXIT();
677 
678 	/* Read the response of the VMGEXIT. */
679 	val = sev_es_rd_ghcb_msr();
680 	if ((GHCB_RESP_CODE(val) != GHCB_MSR_PSC_RESP) || GHCB_MSR_PSC_RESP_VAL(val))
681 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PSC);
682 
683 	/* Restore the GHCB MSR value */
684 	sev_es_wr_ghcb_msr(msr);
685 
686 	/*
687 	 * Now that page state is changed in the RMP table, validate it so that it is
688 	 * consistent with the RMP entry.
689 	 */
690 	if (desc->op == SNP_PAGE_STATE_PRIVATE)
691 		pvalidate_4k_page(vaddr, paddr, true, desc->ca, desc->caa_pa);
692 }
693 
694 /*
695  * Maintain the GPA of the SVSM Calling Area (CA) in order to utilize the SVSM
696  * services needed when not running in VMPL0.
697  */
svsm_setup_ca(const struct cc_blob_sev_info * cc_info,void * page)698 static bool __init svsm_setup_ca(const struct cc_blob_sev_info *cc_info,
699 				 void *page)
700 {
701 	struct snp_secrets_page *secrets_page;
702 	struct snp_cpuid_table *cpuid_table;
703 	unsigned int i;
704 	u64 caa;
705 
706 	BUILD_BUG_ON(sizeof(*secrets_page) != PAGE_SIZE);
707 
708 	/*
709 	 * Check if running at VMPL0.
710 	 *
711 	 * Use RMPADJUST (see the rmpadjust() function for a description of what
712 	 * the instruction does) to update the VMPL1 permissions of a page. If
713 	 * the guest is running at VMPL0, this will succeed and implies there is
714 	 * no SVSM. If the guest is running at any other VMPL, this will fail.
715 	 * Linux SNP guests only ever run at a single VMPL level so permission mask
716 	 * changes of a lesser-privileged VMPL are a don't-care.
717 	 *
718 	 * Use a rip-relative reference to obtain the proper address, since this
719 	 * routine is running identity mapped when called, both by the decompressor
720 	 * code and the early kernel code.
721 	 */
722 	if (!rmpadjust((unsigned long)page, RMP_PG_SIZE_4K, 1))
723 		return false;
724 
725 	/*
726 	 * Not running at VMPL0, ensure everything has been properly supplied
727 	 * for running under an SVSM.
728 	 */
729 	if (!cc_info || !cc_info->secrets_phys || cc_info->secrets_len != PAGE_SIZE)
730 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SECRETS_PAGE);
731 
732 	secrets_page = (struct snp_secrets_page *)cc_info->secrets_phys;
733 	if (!secrets_page->svsm_size)
734 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_NO_SVSM);
735 
736 	if (!secrets_page->svsm_guest_vmpl)
737 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SVSM_VMPL0);
738 
739 	snp_vmpl = secrets_page->svsm_guest_vmpl;
740 
741 	caa = secrets_page->svsm_caa;
742 
743 	/*
744 	 * An open-coded PAGE_ALIGNED() in order to avoid including
745 	 * kernel-proper headers into the decompressor.
746 	 */
747 	if (caa & (PAGE_SIZE - 1))
748 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_SVSM_CAA);
749 
750 	boot_svsm_caa_pa = caa;
751 
752 	/* Advertise the SVSM presence via CPUID. */
753 	cpuid_table = (struct snp_cpuid_table *)snp_cpuid_get_table();
754 	for (i = 0; i < cpuid_table->count; i++) {
755 		struct snp_cpuid_fn *fn = &cpuid_table->fn[i];
756 
757 		if (fn->eax_in == 0x8000001f)
758 			fn->eax |= BIT(28);
759 	}
760 
761 	return true;
762 }
763