xref: /openbmc/linux/arch/x86/kernel/sev-shared.c (revision 2bc7d3e0)
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 #ifndef __BOOT_COMPRESSED
13 #define error(v)	pr_err(v)
14 #define has_cpuflag(f)	boot_cpu_has(f)
15 #endif
16 
17 /* I/O parameters for CPUID-related helpers */
18 struct cpuid_leaf {
19 	u32 fn;
20 	u32 subfn;
21 	u32 eax;
22 	u32 ebx;
23 	u32 ecx;
24 	u32 edx;
25 };
26 
27 /*
28  * Individual entries of the SNP CPUID table, as defined by the SNP
29  * Firmware ABI, Revision 0.9, Section 7.1, Table 14.
30  */
31 struct snp_cpuid_fn {
32 	u32 eax_in;
33 	u32 ecx_in;
34 	u64 xcr0_in;
35 	u64 xss_in;
36 	u32 eax;
37 	u32 ebx;
38 	u32 ecx;
39 	u32 edx;
40 	u64 __reserved;
41 } __packed;
42 
43 /*
44  * SNP CPUID table, as defined by the SNP Firmware ABI, Revision 0.9,
45  * Section 8.14.2.6. Also noted there is the SNP firmware-enforced limit
46  * of 64 entries per CPUID table.
47  */
48 #define SNP_CPUID_COUNT_MAX 64
49 
50 struct snp_cpuid_table {
51 	u32 count;
52 	u32 __reserved1;
53 	u64 __reserved2;
54 	struct snp_cpuid_fn fn[SNP_CPUID_COUNT_MAX];
55 } __packed;
56 
57 /*
58  * Since feature negotiation related variables are set early in the boot
59  * process they must reside in the .data section so as not to be zeroed
60  * out when the .bss section is later cleared.
61  *
62  * GHCB protocol version negotiated with the hypervisor.
63  */
64 static u16 ghcb_version __ro_after_init;
65 
66 /* Copy of the SNP firmware's CPUID page. */
67 static struct snp_cpuid_table cpuid_table_copy __ro_after_init;
68 
69 /*
70  * These will be initialized based on CPUID table so that non-present
71  * all-zero leaves (for sparse tables) can be differentiated from
72  * invalid/out-of-range leaves. This is needed since all-zero leaves
73  * still need to be post-processed.
74  */
75 static u32 cpuid_std_range_max __ro_after_init;
76 static u32 cpuid_hyp_range_max __ro_after_init;
77 static u32 cpuid_ext_range_max __ro_after_init;
78 
79 static bool __init sev_es_check_cpu_features(void)
80 {
81 	if (!has_cpuflag(X86_FEATURE_RDRAND)) {
82 		error("RDRAND instruction not supported - no trusted source of randomness available\n");
83 		return false;
84 	}
85 
86 	return true;
87 }
88 
89 static void __noreturn sev_es_terminate(unsigned int set, unsigned int reason)
90 {
91 	u64 val = GHCB_MSR_TERM_REQ;
92 
93 	/* Tell the hypervisor what went wrong. */
94 	val |= GHCB_SEV_TERM_REASON(set, reason);
95 
96 	/* Request Guest Termination from Hypvervisor */
97 	sev_es_wr_ghcb_msr(val);
98 	VMGEXIT();
99 
100 	while (true)
101 		asm volatile("hlt\n" : : : "memory");
102 }
103 
104 /*
105  * The hypervisor features are available from GHCB version 2 onward.
106  */
107 static u64 get_hv_features(void)
108 {
109 	u64 val;
110 
111 	if (ghcb_version < 2)
112 		return 0;
113 
114 	sev_es_wr_ghcb_msr(GHCB_MSR_HV_FT_REQ);
115 	VMGEXIT();
116 
117 	val = sev_es_rd_ghcb_msr();
118 	if (GHCB_RESP_CODE(val) != GHCB_MSR_HV_FT_RESP)
119 		return 0;
120 
121 	return GHCB_MSR_HV_FT_RESP_VAL(val);
122 }
123 
124 static void snp_register_ghcb_early(unsigned long paddr)
125 {
126 	unsigned long pfn = paddr >> PAGE_SHIFT;
127 	u64 val;
128 
129 	sev_es_wr_ghcb_msr(GHCB_MSR_REG_GPA_REQ_VAL(pfn));
130 	VMGEXIT();
131 
132 	val = sev_es_rd_ghcb_msr();
133 
134 	/* If the response GPA is not ours then abort the guest */
135 	if ((GHCB_RESP_CODE(val) != GHCB_MSR_REG_GPA_RESP) ||
136 	    (GHCB_MSR_REG_GPA_RESP_VAL(val) != pfn))
137 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_REGISTER);
138 }
139 
140 static bool sev_es_negotiate_protocol(void)
141 {
142 	u64 val;
143 
144 	/* Do the GHCB protocol version negotiation */
145 	sev_es_wr_ghcb_msr(GHCB_MSR_SEV_INFO_REQ);
146 	VMGEXIT();
147 	val = sev_es_rd_ghcb_msr();
148 
149 	if (GHCB_MSR_INFO(val) != GHCB_MSR_SEV_INFO_RESP)
150 		return false;
151 
152 	if (GHCB_MSR_PROTO_MAX(val) < GHCB_PROTOCOL_MIN ||
153 	    GHCB_MSR_PROTO_MIN(val) > GHCB_PROTOCOL_MAX)
154 		return false;
155 
156 	ghcb_version = min_t(size_t, GHCB_MSR_PROTO_MAX(val), GHCB_PROTOCOL_MAX);
157 
158 	return true;
159 }
160 
161 static __always_inline void vc_ghcb_invalidate(struct ghcb *ghcb)
162 {
163 	ghcb->save.sw_exit_code = 0;
164 	__builtin_memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
165 }
166 
167 static bool vc_decoding_needed(unsigned long exit_code)
168 {
169 	/* Exceptions don't require to decode the instruction */
170 	return !(exit_code >= SVM_EXIT_EXCP_BASE &&
171 		 exit_code <= SVM_EXIT_LAST_EXCP);
172 }
173 
174 static enum es_result vc_init_em_ctxt(struct es_em_ctxt *ctxt,
175 				      struct pt_regs *regs,
176 				      unsigned long exit_code)
177 {
178 	enum es_result ret = ES_OK;
179 
180 	memset(ctxt, 0, sizeof(*ctxt));
181 	ctxt->regs = regs;
182 
183 	if (vc_decoding_needed(exit_code))
184 		ret = vc_decode_insn(ctxt);
185 
186 	return ret;
187 }
188 
189 static void vc_finish_insn(struct es_em_ctxt *ctxt)
190 {
191 	ctxt->regs->ip += ctxt->insn.length;
192 }
193 
194 static enum es_result verify_exception_info(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
195 {
196 	u32 ret;
197 
198 	ret = ghcb->save.sw_exit_info_1 & GENMASK_ULL(31, 0);
199 	if (!ret)
200 		return ES_OK;
201 
202 	if (ret == 1) {
203 		u64 info = ghcb->save.sw_exit_info_2;
204 		unsigned long v = info & SVM_EVTINJ_VEC_MASK;
205 
206 		/* Check if exception information from hypervisor is sane. */
207 		if ((info & SVM_EVTINJ_VALID) &&
208 		    ((v == X86_TRAP_GP) || (v == X86_TRAP_UD)) &&
209 		    ((info & SVM_EVTINJ_TYPE_MASK) == SVM_EVTINJ_TYPE_EXEPT)) {
210 			ctxt->fi.vector = v;
211 
212 			if (info & SVM_EVTINJ_VALID_ERR)
213 				ctxt->fi.error_code = info >> 32;
214 
215 			return ES_EXCEPTION;
216 		}
217 	}
218 
219 	return ES_VMM_ERROR;
220 }
221 
222 enum es_result sev_es_ghcb_hv_call(struct ghcb *ghcb, bool set_ghcb_msr,
223 				   struct es_em_ctxt *ctxt, u64 exit_code,
224 				   u64 exit_info_1, u64 exit_info_2)
225 {
226 	/* Fill in protocol and format specifiers */
227 	ghcb->protocol_version = ghcb_version;
228 	ghcb->ghcb_usage       = GHCB_DEFAULT_USAGE;
229 
230 	ghcb_set_sw_exit_code(ghcb, exit_code);
231 	ghcb_set_sw_exit_info_1(ghcb, exit_info_1);
232 	ghcb_set_sw_exit_info_2(ghcb, exit_info_2);
233 
234 	/*
235 	 * Hyper-V unenlightened guests use a paravisor for communicating and
236 	 * GHCB pages are being allocated and set up by that paravisor. Linux
237 	 * should not change the GHCB page's physical address.
238 	 */
239 	if (set_ghcb_msr)
240 		sev_es_wr_ghcb_msr(__pa(ghcb));
241 
242 	VMGEXIT();
243 
244 	return verify_exception_info(ghcb, ctxt);
245 }
246 
247 static int __sev_cpuid_hv(u32 fn, int reg_idx, u32 *reg)
248 {
249 	u64 val;
250 
251 	sev_es_wr_ghcb_msr(GHCB_CPUID_REQ(fn, reg_idx));
252 	VMGEXIT();
253 	val = sev_es_rd_ghcb_msr();
254 	if (GHCB_RESP_CODE(val) != GHCB_MSR_CPUID_RESP)
255 		return -EIO;
256 
257 	*reg = (val >> 32);
258 
259 	return 0;
260 }
261 
262 static int sev_cpuid_hv(struct cpuid_leaf *leaf)
263 {
264 	int ret;
265 
266 	/*
267 	 * MSR protocol does not support fetching non-zero subfunctions, but is
268 	 * sufficient to handle current early-boot cases. Should that change,
269 	 * make sure to report an error rather than ignoring the index and
270 	 * grabbing random values. If this issue arises in the future, handling
271 	 * can be added here to use GHCB-page protocol for cases that occur late
272 	 * enough in boot that GHCB page is available.
273 	 */
274 	if (cpuid_function_is_indexed(leaf->fn) && leaf->subfn)
275 		return -EINVAL;
276 
277 	ret =         __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EAX, &leaf->eax);
278 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EBX, &leaf->ebx);
279 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_ECX, &leaf->ecx);
280 	ret = ret ? : __sev_cpuid_hv(leaf->fn, GHCB_CPUID_REQ_EDX, &leaf->edx);
281 
282 	return ret;
283 }
284 
285 /*
286  * This may be called early while still running on the initial identity
287  * mapping. Use RIP-relative addressing to obtain the correct address
288  * while running with the initial identity mapping as well as the
289  * switch-over to kernel virtual addresses later.
290  */
291 static const struct snp_cpuid_table *snp_cpuid_get_table(void)
292 {
293 	void *ptr;
294 
295 	asm ("lea cpuid_table_copy(%%rip), %0"
296 	     : "=r" (ptr)
297 	     : "p" (&cpuid_table_copy));
298 
299 	return ptr;
300 }
301 
302 /*
303  * The SNP Firmware ABI, Revision 0.9, Section 7.1, details the use of
304  * XCR0_IN and XSS_IN to encode multiple versions of 0xD subfunctions 0
305  * and 1 based on the corresponding features enabled by a particular
306  * combination of XCR0 and XSS registers so that a guest can look up the
307  * version corresponding to the features currently enabled in its XCR0/XSS
308  * registers. The only values that differ between these versions/table
309  * entries is the enabled XSAVE area size advertised via EBX.
310  *
311  * While hypervisors may choose to make use of this support, it is more
312  * robust/secure for a guest to simply find the entry corresponding to the
313  * base/legacy XSAVE area size (XCR0=1 or XCR0=3), and then calculate the
314  * XSAVE area size using subfunctions 2 through 64, as documented in APM
315  * Volume 3, Rev 3.31, Appendix E.3.8, which is what is done here.
316  *
317  * Since base/legacy XSAVE area size is documented as 0x240, use that value
318  * directly rather than relying on the base size in the CPUID table.
319  *
320  * Return: XSAVE area size on success, 0 otherwise.
321  */
322 static u32 snp_cpuid_calc_xsave_size(u64 xfeatures_en, bool compacted)
323 {
324 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
325 	u64 xfeatures_found = 0;
326 	u32 xsave_size = 0x240;
327 	int i;
328 
329 	for (i = 0; i < cpuid_table->count; i++) {
330 		const struct snp_cpuid_fn *e = &cpuid_table->fn[i];
331 
332 		if (!(e->eax_in == 0xD && e->ecx_in > 1 && e->ecx_in < 64))
333 			continue;
334 		if (!(xfeatures_en & (BIT_ULL(e->ecx_in))))
335 			continue;
336 		if (xfeatures_found & (BIT_ULL(e->ecx_in)))
337 			continue;
338 
339 		xfeatures_found |= (BIT_ULL(e->ecx_in));
340 
341 		if (compacted)
342 			xsave_size += e->eax;
343 		else
344 			xsave_size = max(xsave_size, e->eax + e->ebx);
345 	}
346 
347 	/*
348 	 * Either the guest set unsupported XCR0/XSS bits, or the corresponding
349 	 * entries in the CPUID table were not present. This is not a valid
350 	 * state to be in.
351 	 */
352 	if (xfeatures_found != (xfeatures_en & GENMASK_ULL(63, 2)))
353 		return 0;
354 
355 	return xsave_size;
356 }
357 
358 static bool
359 snp_cpuid_get_validated_func(struct cpuid_leaf *leaf)
360 {
361 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
362 	int i;
363 
364 	for (i = 0; i < cpuid_table->count; i++) {
365 		const struct snp_cpuid_fn *e = &cpuid_table->fn[i];
366 
367 		if (e->eax_in != leaf->fn)
368 			continue;
369 
370 		if (cpuid_function_is_indexed(leaf->fn) && e->ecx_in != leaf->subfn)
371 			continue;
372 
373 		/*
374 		 * For 0xD subfunctions 0 and 1, only use the entry corresponding
375 		 * to the base/legacy XSAVE area size (XCR0=1 or XCR0=3, XSS=0).
376 		 * See the comments above snp_cpuid_calc_xsave_size() for more
377 		 * details.
378 		 */
379 		if (e->eax_in == 0xD && (e->ecx_in == 0 || e->ecx_in == 1))
380 			if (!(e->xcr0_in == 1 || e->xcr0_in == 3) || e->xss_in)
381 				continue;
382 
383 		leaf->eax = e->eax;
384 		leaf->ebx = e->ebx;
385 		leaf->ecx = e->ecx;
386 		leaf->edx = e->edx;
387 
388 		return true;
389 	}
390 
391 	return false;
392 }
393 
394 static void snp_cpuid_hv(struct cpuid_leaf *leaf)
395 {
396 	if (sev_cpuid_hv(leaf))
397 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID_HV);
398 }
399 
400 static int snp_cpuid_postprocess(struct cpuid_leaf *leaf)
401 {
402 	struct cpuid_leaf leaf_hv = *leaf;
403 
404 	switch (leaf->fn) {
405 	case 0x1:
406 		snp_cpuid_hv(&leaf_hv);
407 
408 		/* initial APIC ID */
409 		leaf->ebx = (leaf_hv.ebx & GENMASK(31, 24)) | (leaf->ebx & GENMASK(23, 0));
410 		/* APIC enabled bit */
411 		leaf->edx = (leaf_hv.edx & BIT(9)) | (leaf->edx & ~BIT(9));
412 
413 		/* OSXSAVE enabled bit */
414 		if (native_read_cr4() & X86_CR4_OSXSAVE)
415 			leaf->ecx |= BIT(27);
416 		break;
417 	case 0x7:
418 		/* OSPKE enabled bit */
419 		leaf->ecx &= ~BIT(4);
420 		if (native_read_cr4() & X86_CR4_PKE)
421 			leaf->ecx |= BIT(4);
422 		break;
423 	case 0xB:
424 		leaf_hv.subfn = 0;
425 		snp_cpuid_hv(&leaf_hv);
426 
427 		/* extended APIC ID */
428 		leaf->edx = leaf_hv.edx;
429 		break;
430 	case 0xD: {
431 		bool compacted = false;
432 		u64 xcr0 = 1, xss = 0;
433 		u32 xsave_size;
434 
435 		if (leaf->subfn != 0 && leaf->subfn != 1)
436 			return 0;
437 
438 		if (native_read_cr4() & X86_CR4_OSXSAVE)
439 			xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
440 		if (leaf->subfn == 1) {
441 			/* Get XSS value if XSAVES is enabled. */
442 			if (leaf->eax & BIT(3)) {
443 				unsigned long lo, hi;
444 
445 				asm volatile("rdmsr" : "=a" (lo), "=d" (hi)
446 						     : "c" (MSR_IA32_XSS));
447 				xss = (hi << 32) | lo;
448 			}
449 
450 			/*
451 			 * The PPR and APM aren't clear on what size should be
452 			 * encoded in 0xD:0x1:EBX when compaction is not enabled
453 			 * by either XSAVEC (feature bit 1) or XSAVES (feature
454 			 * bit 3) since SNP-capable hardware has these feature
455 			 * bits fixed as 1. KVM sets it to 0 in this case, but
456 			 * to avoid this becoming an issue it's safer to simply
457 			 * treat this as unsupported for SNP guests.
458 			 */
459 			if (!(leaf->eax & (BIT(1) | BIT(3))))
460 				return -EINVAL;
461 
462 			compacted = true;
463 		}
464 
465 		xsave_size = snp_cpuid_calc_xsave_size(xcr0 | xss, compacted);
466 		if (!xsave_size)
467 			return -EINVAL;
468 
469 		leaf->ebx = xsave_size;
470 		}
471 		break;
472 	case 0x8000001E:
473 		snp_cpuid_hv(&leaf_hv);
474 
475 		/* extended APIC ID */
476 		leaf->eax = leaf_hv.eax;
477 		/* compute ID */
478 		leaf->ebx = (leaf->ebx & GENMASK(31, 8)) | (leaf_hv.ebx & GENMASK(7, 0));
479 		/* node ID */
480 		leaf->ecx = (leaf->ecx & GENMASK(31, 8)) | (leaf_hv.ecx & GENMASK(7, 0));
481 		break;
482 	default:
483 		/* No fix-ups needed, use values as-is. */
484 		break;
485 	}
486 
487 	return 0;
488 }
489 
490 /*
491  * Returns -EOPNOTSUPP if feature not enabled. Any other non-zero return value
492  * should be treated as fatal by caller.
493  */
494 static int snp_cpuid(struct cpuid_leaf *leaf)
495 {
496 	const struct snp_cpuid_table *cpuid_table = snp_cpuid_get_table();
497 
498 	if (!cpuid_table->count)
499 		return -EOPNOTSUPP;
500 
501 	if (!snp_cpuid_get_validated_func(leaf)) {
502 		/*
503 		 * Some hypervisors will avoid keeping track of CPUID entries
504 		 * where all values are zero, since they can be handled the
505 		 * same as out-of-range values (all-zero). This is useful here
506 		 * as well as it allows virtually all guest configurations to
507 		 * work using a single SNP CPUID table.
508 		 *
509 		 * To allow for this, there is a need to distinguish between
510 		 * out-of-range entries and in-range zero entries, since the
511 		 * CPUID table entries are only a template that may need to be
512 		 * augmented with additional values for things like
513 		 * CPU-specific information during post-processing. So if it's
514 		 * not in the table, set the values to zero. Then, if they are
515 		 * within a valid CPUID range, proceed with post-processing
516 		 * using zeros as the initial values. Otherwise, skip
517 		 * post-processing and just return zeros immediately.
518 		 */
519 		leaf->eax = leaf->ebx = leaf->ecx = leaf->edx = 0;
520 
521 		/* Skip post-processing for out-of-range zero leafs. */
522 		if (!(leaf->fn <= cpuid_std_range_max ||
523 		      (leaf->fn >= 0x40000000 && leaf->fn <= cpuid_hyp_range_max) ||
524 		      (leaf->fn >= 0x80000000 && leaf->fn <= cpuid_ext_range_max)))
525 			return 0;
526 	}
527 
528 	return snp_cpuid_postprocess(leaf);
529 }
530 
531 /*
532  * Boot VC Handler - This is the first VC handler during boot, there is no GHCB
533  * page yet, so it only supports the MSR based communication with the
534  * hypervisor and only the CPUID exit-code.
535  */
536 void __init do_vc_no_ghcb(struct pt_regs *regs, unsigned long exit_code)
537 {
538 	unsigned int subfn = lower_bits(regs->cx, 32);
539 	unsigned int fn = lower_bits(regs->ax, 32);
540 	struct cpuid_leaf leaf;
541 	int ret;
542 
543 	/* Only CPUID is supported via MSR protocol */
544 	if (exit_code != SVM_EXIT_CPUID)
545 		goto fail;
546 
547 	leaf.fn = fn;
548 	leaf.subfn = subfn;
549 
550 	ret = snp_cpuid(&leaf);
551 	if (!ret)
552 		goto cpuid_done;
553 
554 	if (ret != -EOPNOTSUPP)
555 		goto fail;
556 
557 	if (sev_cpuid_hv(&leaf))
558 		goto fail;
559 
560 cpuid_done:
561 	regs->ax = leaf.eax;
562 	regs->bx = leaf.ebx;
563 	regs->cx = leaf.ecx;
564 	regs->dx = leaf.edx;
565 
566 	/*
567 	 * This is a VC handler and the #VC is only raised when SEV-ES is
568 	 * active, which means SEV must be active too. Do sanity checks on the
569 	 * CPUID results to make sure the hypervisor does not trick the kernel
570 	 * into the no-sev path. This could map sensitive data unencrypted and
571 	 * make it accessible to the hypervisor.
572 	 *
573 	 * In particular, check for:
574 	 *	- Availability of CPUID leaf 0x8000001f
575 	 *	- SEV CPUID bit.
576 	 *
577 	 * The hypervisor might still report the wrong C-bit position, but this
578 	 * can't be checked here.
579 	 */
580 
581 	if (fn == 0x80000000 && (regs->ax < 0x8000001f))
582 		/* SEV leaf check */
583 		goto fail;
584 	else if ((fn == 0x8000001f && !(regs->ax & BIT(1))))
585 		/* SEV bit */
586 		goto fail;
587 
588 	/* Skip over the CPUID two-byte opcode */
589 	regs->ip += 2;
590 
591 	return;
592 
593 fail:
594 	/* Terminate the guest */
595 	sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
596 }
597 
598 static enum es_result vc_insn_string_read(struct es_em_ctxt *ctxt,
599 					  void *src, char *buf,
600 					  unsigned int data_size,
601 					  unsigned int count,
602 					  bool backwards)
603 {
604 	int i, b = backwards ? -1 : 1;
605 	enum es_result ret = ES_OK;
606 
607 	for (i = 0; i < count; i++) {
608 		void *s = src + (i * data_size * b);
609 		char *d = buf + (i * data_size);
610 
611 		ret = vc_read_mem(ctxt, s, d, data_size);
612 		if (ret != ES_OK)
613 			break;
614 	}
615 
616 	return ret;
617 }
618 
619 static enum es_result vc_insn_string_write(struct es_em_ctxt *ctxt,
620 					   void *dst, char *buf,
621 					   unsigned int data_size,
622 					   unsigned int count,
623 					   bool backwards)
624 {
625 	int i, s = backwards ? -1 : 1;
626 	enum es_result ret = ES_OK;
627 
628 	for (i = 0; i < count; i++) {
629 		void *d = dst + (i * data_size * s);
630 		char *b = buf + (i * data_size);
631 
632 		ret = vc_write_mem(ctxt, d, b, data_size);
633 		if (ret != ES_OK)
634 			break;
635 	}
636 
637 	return ret;
638 }
639 
640 #define IOIO_TYPE_STR  BIT(2)
641 #define IOIO_TYPE_IN   1
642 #define IOIO_TYPE_INS  (IOIO_TYPE_IN | IOIO_TYPE_STR)
643 #define IOIO_TYPE_OUT  0
644 #define IOIO_TYPE_OUTS (IOIO_TYPE_OUT | IOIO_TYPE_STR)
645 
646 #define IOIO_REP       BIT(3)
647 
648 #define IOIO_ADDR_64   BIT(9)
649 #define IOIO_ADDR_32   BIT(8)
650 #define IOIO_ADDR_16   BIT(7)
651 
652 #define IOIO_DATA_32   BIT(6)
653 #define IOIO_DATA_16   BIT(5)
654 #define IOIO_DATA_8    BIT(4)
655 
656 #define IOIO_SEG_ES    (0 << 10)
657 #define IOIO_SEG_DS    (3 << 10)
658 
659 static enum es_result vc_ioio_exitinfo(struct es_em_ctxt *ctxt, u64 *exitinfo)
660 {
661 	struct insn *insn = &ctxt->insn;
662 	*exitinfo = 0;
663 
664 	switch (insn->opcode.bytes[0]) {
665 	/* INS opcodes */
666 	case 0x6c:
667 	case 0x6d:
668 		*exitinfo |= IOIO_TYPE_INS;
669 		*exitinfo |= IOIO_SEG_ES;
670 		*exitinfo |= (ctxt->regs->dx & 0xffff) << 16;
671 		break;
672 
673 	/* OUTS opcodes */
674 	case 0x6e:
675 	case 0x6f:
676 		*exitinfo |= IOIO_TYPE_OUTS;
677 		*exitinfo |= IOIO_SEG_DS;
678 		*exitinfo |= (ctxt->regs->dx & 0xffff) << 16;
679 		break;
680 
681 	/* IN immediate opcodes */
682 	case 0xe4:
683 	case 0xe5:
684 		*exitinfo |= IOIO_TYPE_IN;
685 		*exitinfo |= (u8)insn->immediate.value << 16;
686 		break;
687 
688 	/* OUT immediate opcodes */
689 	case 0xe6:
690 	case 0xe7:
691 		*exitinfo |= IOIO_TYPE_OUT;
692 		*exitinfo |= (u8)insn->immediate.value << 16;
693 		break;
694 
695 	/* IN register opcodes */
696 	case 0xec:
697 	case 0xed:
698 		*exitinfo |= IOIO_TYPE_IN;
699 		*exitinfo |= (ctxt->regs->dx & 0xffff) << 16;
700 		break;
701 
702 	/* OUT register opcodes */
703 	case 0xee:
704 	case 0xef:
705 		*exitinfo |= IOIO_TYPE_OUT;
706 		*exitinfo |= (ctxt->regs->dx & 0xffff) << 16;
707 		break;
708 
709 	default:
710 		return ES_DECODE_FAILED;
711 	}
712 
713 	switch (insn->opcode.bytes[0]) {
714 	case 0x6c:
715 	case 0x6e:
716 	case 0xe4:
717 	case 0xe6:
718 	case 0xec:
719 	case 0xee:
720 		/* Single byte opcodes */
721 		*exitinfo |= IOIO_DATA_8;
722 		break;
723 	default:
724 		/* Length determined by instruction parsing */
725 		*exitinfo |= (insn->opnd_bytes == 2) ? IOIO_DATA_16
726 						     : IOIO_DATA_32;
727 	}
728 	switch (insn->addr_bytes) {
729 	case 2:
730 		*exitinfo |= IOIO_ADDR_16;
731 		break;
732 	case 4:
733 		*exitinfo |= IOIO_ADDR_32;
734 		break;
735 	case 8:
736 		*exitinfo |= IOIO_ADDR_64;
737 		break;
738 	}
739 
740 	if (insn_has_rep_prefix(insn))
741 		*exitinfo |= IOIO_REP;
742 
743 	return ES_OK;
744 }
745 
746 static enum es_result vc_handle_ioio(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
747 {
748 	struct pt_regs *regs = ctxt->regs;
749 	u64 exit_info_1, exit_info_2;
750 	enum es_result ret;
751 
752 	ret = vc_ioio_exitinfo(ctxt, &exit_info_1);
753 	if (ret != ES_OK)
754 		return ret;
755 
756 	if (exit_info_1 & IOIO_TYPE_STR) {
757 
758 		/* (REP) INS/OUTS */
759 
760 		bool df = ((regs->flags & X86_EFLAGS_DF) == X86_EFLAGS_DF);
761 		unsigned int io_bytes, exit_bytes;
762 		unsigned int ghcb_count, op_count;
763 		unsigned long es_base;
764 		u64 sw_scratch;
765 
766 		/*
767 		 * For the string variants with rep prefix the amount of in/out
768 		 * operations per #VC exception is limited so that the kernel
769 		 * has a chance to take interrupts and re-schedule while the
770 		 * instruction is emulated.
771 		 */
772 		io_bytes   = (exit_info_1 >> 4) & 0x7;
773 		ghcb_count = sizeof(ghcb->shared_buffer) / io_bytes;
774 
775 		op_count    = (exit_info_1 & IOIO_REP) ? regs->cx : 1;
776 		exit_info_2 = min(op_count, ghcb_count);
777 		exit_bytes  = exit_info_2 * io_bytes;
778 
779 		es_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_ES);
780 
781 		/* Read bytes of OUTS into the shared buffer */
782 		if (!(exit_info_1 & IOIO_TYPE_IN)) {
783 			ret = vc_insn_string_read(ctxt,
784 					       (void *)(es_base + regs->si),
785 					       ghcb->shared_buffer, io_bytes,
786 					       exit_info_2, df);
787 			if (ret)
788 				return ret;
789 		}
790 
791 		/*
792 		 * Issue an VMGEXIT to the HV to consume the bytes from the
793 		 * shared buffer or to have it write them into the shared buffer
794 		 * depending on the instruction: OUTS or INS.
795 		 */
796 		sw_scratch = __pa(ghcb) + offsetof(struct ghcb, shared_buffer);
797 		ghcb_set_sw_scratch(ghcb, sw_scratch);
798 		ret = sev_es_ghcb_hv_call(ghcb, true, ctxt, SVM_EXIT_IOIO,
799 					  exit_info_1, exit_info_2);
800 		if (ret != ES_OK)
801 			return ret;
802 
803 		/* Read bytes from shared buffer into the guest's destination. */
804 		if (exit_info_1 & IOIO_TYPE_IN) {
805 			ret = vc_insn_string_write(ctxt,
806 						   (void *)(es_base + regs->di),
807 						   ghcb->shared_buffer, io_bytes,
808 						   exit_info_2, df);
809 			if (ret)
810 				return ret;
811 
812 			if (df)
813 				regs->di -= exit_bytes;
814 			else
815 				regs->di += exit_bytes;
816 		} else {
817 			if (df)
818 				regs->si -= exit_bytes;
819 			else
820 				regs->si += exit_bytes;
821 		}
822 
823 		if (exit_info_1 & IOIO_REP)
824 			regs->cx -= exit_info_2;
825 
826 		ret = regs->cx ? ES_RETRY : ES_OK;
827 
828 	} else {
829 
830 		/* IN/OUT into/from rAX */
831 
832 		int bits = (exit_info_1 & 0x70) >> 1;
833 		u64 rax = 0;
834 
835 		if (!(exit_info_1 & IOIO_TYPE_IN))
836 			rax = lower_bits(regs->ax, bits);
837 
838 		ghcb_set_rax(ghcb, rax);
839 
840 		ret = sev_es_ghcb_hv_call(ghcb, true, ctxt,
841 					  SVM_EXIT_IOIO, exit_info_1, 0);
842 		if (ret != ES_OK)
843 			return ret;
844 
845 		if (exit_info_1 & IOIO_TYPE_IN) {
846 			if (!ghcb_rax_is_valid(ghcb))
847 				return ES_VMM_ERROR;
848 			regs->ax = lower_bits(ghcb->save.rax, bits);
849 		}
850 	}
851 
852 	return ret;
853 }
854 
855 static int vc_handle_cpuid_snp(struct pt_regs *regs)
856 {
857 	struct cpuid_leaf leaf;
858 	int ret;
859 
860 	leaf.fn = regs->ax;
861 	leaf.subfn = regs->cx;
862 	ret = snp_cpuid(&leaf);
863 	if (!ret) {
864 		regs->ax = leaf.eax;
865 		regs->bx = leaf.ebx;
866 		regs->cx = leaf.ecx;
867 		regs->dx = leaf.edx;
868 	}
869 
870 	return ret;
871 }
872 
873 static enum es_result vc_handle_cpuid(struct ghcb *ghcb,
874 				      struct es_em_ctxt *ctxt)
875 {
876 	struct pt_regs *regs = ctxt->regs;
877 	u32 cr4 = native_read_cr4();
878 	enum es_result ret;
879 	int snp_cpuid_ret;
880 
881 	snp_cpuid_ret = vc_handle_cpuid_snp(regs);
882 	if (!snp_cpuid_ret)
883 		return ES_OK;
884 	if (snp_cpuid_ret != -EOPNOTSUPP)
885 		return ES_VMM_ERROR;
886 
887 	ghcb_set_rax(ghcb, regs->ax);
888 	ghcb_set_rcx(ghcb, regs->cx);
889 
890 	if (cr4 & X86_CR4_OSXSAVE)
891 		/* Safe to read xcr0 */
892 		ghcb_set_xcr0(ghcb, xgetbv(XCR_XFEATURE_ENABLED_MASK));
893 	else
894 		/* xgetbv will cause #GP - use reset value for xcr0 */
895 		ghcb_set_xcr0(ghcb, 1);
896 
897 	ret = sev_es_ghcb_hv_call(ghcb, true, ctxt, SVM_EXIT_CPUID, 0, 0);
898 	if (ret != ES_OK)
899 		return ret;
900 
901 	if (!(ghcb_rax_is_valid(ghcb) &&
902 	      ghcb_rbx_is_valid(ghcb) &&
903 	      ghcb_rcx_is_valid(ghcb) &&
904 	      ghcb_rdx_is_valid(ghcb)))
905 		return ES_VMM_ERROR;
906 
907 	regs->ax = ghcb->save.rax;
908 	regs->bx = ghcb->save.rbx;
909 	regs->cx = ghcb->save.rcx;
910 	regs->dx = ghcb->save.rdx;
911 
912 	return ES_OK;
913 }
914 
915 static enum es_result vc_handle_rdtsc(struct ghcb *ghcb,
916 				      struct es_em_ctxt *ctxt,
917 				      unsigned long exit_code)
918 {
919 	bool rdtscp = (exit_code == SVM_EXIT_RDTSCP);
920 	enum es_result ret;
921 
922 	ret = sev_es_ghcb_hv_call(ghcb, true, ctxt, exit_code, 0, 0);
923 	if (ret != ES_OK)
924 		return ret;
925 
926 	if (!(ghcb_rax_is_valid(ghcb) && ghcb_rdx_is_valid(ghcb) &&
927 	     (!rdtscp || ghcb_rcx_is_valid(ghcb))))
928 		return ES_VMM_ERROR;
929 
930 	ctxt->regs->ax = ghcb->save.rax;
931 	ctxt->regs->dx = ghcb->save.rdx;
932 	if (rdtscp)
933 		ctxt->regs->cx = ghcb->save.rcx;
934 
935 	return ES_OK;
936 }
937 
938 struct cc_setup_data {
939 	struct setup_data header;
940 	u32 cc_blob_address;
941 };
942 
943 /*
944  * Search for a Confidential Computing blob passed in as a setup_data entry
945  * via the Linux Boot Protocol.
946  */
947 static struct cc_blob_sev_info *find_cc_blob_setup_data(struct boot_params *bp)
948 {
949 	struct cc_setup_data *sd = NULL;
950 	struct setup_data *hdr;
951 
952 	hdr = (struct setup_data *)bp->hdr.setup_data;
953 
954 	while (hdr) {
955 		if (hdr->type == SETUP_CC_BLOB) {
956 			sd = (struct cc_setup_data *)hdr;
957 			return (struct cc_blob_sev_info *)(unsigned long)sd->cc_blob_address;
958 		}
959 		hdr = (struct setup_data *)hdr->next;
960 	}
961 
962 	return NULL;
963 }
964 
965 /*
966  * Initialize the kernel's copy of the SNP CPUID table, and set up the
967  * pointer that will be used to access it.
968  *
969  * Maintaining a direct mapping of the SNP CPUID table used by firmware would
970  * be possible as an alternative, but the approach is brittle since the
971  * mapping needs to be updated in sync with all the changes to virtual memory
972  * layout and related mapping facilities throughout the boot process.
973  */
974 static void __init setup_cpuid_table(const struct cc_blob_sev_info *cc_info)
975 {
976 	const struct snp_cpuid_table *cpuid_table_fw, *cpuid_table;
977 	int i;
978 
979 	if (!cc_info || !cc_info->cpuid_phys || cc_info->cpuid_len < PAGE_SIZE)
980 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID);
981 
982 	cpuid_table_fw = (const struct snp_cpuid_table *)cc_info->cpuid_phys;
983 	if (!cpuid_table_fw->count || cpuid_table_fw->count > SNP_CPUID_COUNT_MAX)
984 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_CPUID);
985 
986 	cpuid_table = snp_cpuid_get_table();
987 	memcpy((void *)cpuid_table, cpuid_table_fw, sizeof(*cpuid_table));
988 
989 	/* Initialize CPUID ranges for range-checking. */
990 	for (i = 0; i < cpuid_table->count; i++) {
991 		const struct snp_cpuid_fn *fn = &cpuid_table->fn[i];
992 
993 		if (fn->eax_in == 0x0)
994 			cpuid_std_range_max = fn->eax;
995 		else if (fn->eax_in == 0x40000000)
996 			cpuid_hyp_range_max = fn->eax;
997 		else if (fn->eax_in == 0x80000000)
998 			cpuid_ext_range_max = fn->eax;
999 	}
1000 }
1001