xref: /linux/arch/x86/boot/compressed/sev.c (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * AMD Encrypted Register State Support
4  *
5  * Author: Joerg Roedel <jroedel@suse.de>
6  */
7 
8 /*
9  * misc.h needs to be first because it knows how to include the other kernel
10  * headers in the pre-decompression code in a way that does not break
11  * compilation.
12  */
13 #include "misc.h"
14 
15 #include <asm/pgtable_types.h>
16 #include <asm/sev.h>
17 #include <asm/trapnr.h>
18 #include <asm/trap_pf.h>
19 #include <asm/msr-index.h>
20 #include <asm/fpu/xcr.h>
21 #include <asm/ptrace.h>
22 #include <asm/svm.h>
23 #include <asm/cpuid.h>
24 
25 #include "error.h"
26 #include "../msr.h"
27 
28 struct ghcb boot_ghcb_page __aligned(PAGE_SIZE);
29 struct ghcb *boot_ghcb;
30 
31 /*
32  * Copy a version of this function here - insn-eval.c can't be used in
33  * pre-decompression code.
34  */
35 static bool insn_has_rep_prefix(struct insn *insn)
36 {
37 	insn_byte_t p;
38 	int i;
39 
40 	insn_get_prefixes(insn);
41 
42 	for_each_insn_prefix(insn, i, p) {
43 		if (p == 0xf2 || p == 0xf3)
44 			return true;
45 	}
46 
47 	return false;
48 }
49 
50 /*
51  * Only a dummy for insn_get_seg_base() - Early boot-code is 64bit only and
52  * doesn't use segments.
53  */
54 static unsigned long insn_get_seg_base(struct pt_regs *regs, int seg_reg_idx)
55 {
56 	return 0UL;
57 }
58 
59 static inline u64 sev_es_rd_ghcb_msr(void)
60 {
61 	struct msr m;
62 
63 	boot_rdmsr(MSR_AMD64_SEV_ES_GHCB, &m);
64 
65 	return m.q;
66 }
67 
68 static inline void sev_es_wr_ghcb_msr(u64 val)
69 {
70 	struct msr m;
71 
72 	m.q = val;
73 	boot_wrmsr(MSR_AMD64_SEV_ES_GHCB, &m);
74 }
75 
76 static enum es_result vc_decode_insn(struct es_em_ctxt *ctxt)
77 {
78 	char buffer[MAX_INSN_SIZE];
79 	int ret;
80 
81 	memcpy(buffer, (unsigned char *)ctxt->regs->ip, MAX_INSN_SIZE);
82 
83 	ret = insn_decode(&ctxt->insn, buffer, MAX_INSN_SIZE, INSN_MODE_64);
84 	if (ret < 0)
85 		return ES_DECODE_FAILED;
86 
87 	return ES_OK;
88 }
89 
90 static enum es_result vc_write_mem(struct es_em_ctxt *ctxt,
91 				   void *dst, char *buf, size_t size)
92 {
93 	memcpy(dst, buf, size);
94 
95 	return ES_OK;
96 }
97 
98 static enum es_result vc_read_mem(struct es_em_ctxt *ctxt,
99 				  void *src, char *buf, size_t size)
100 {
101 	memcpy(buf, src, size);
102 
103 	return ES_OK;
104 }
105 
106 #undef __init
107 #undef __pa
108 #define __init
109 #define __pa(x)	((unsigned long)(x))
110 
111 #define __BOOT_COMPRESSED
112 
113 /* Basic instruction decoding support needed */
114 #include "../../lib/inat.c"
115 #include "../../lib/insn.c"
116 
117 /* Include code for early handlers */
118 #include "../../kernel/sev-shared.c"
119 
120 static inline bool sev_snp_enabled(void)
121 {
122 	return sev_status & MSR_AMD64_SEV_SNP_ENABLED;
123 }
124 
125 static void __page_state_change(unsigned long paddr, enum psc_op op)
126 {
127 	u64 val;
128 
129 	if (!sev_snp_enabled())
130 		return;
131 
132 	/*
133 	 * If private -> shared then invalidate the page before requesting the
134 	 * state change in the RMP table.
135 	 */
136 	if (op == SNP_PAGE_STATE_SHARED && pvalidate(paddr, RMP_PG_SIZE_4K, 0))
137 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PVALIDATE);
138 
139 	/* Issue VMGEXIT to change the page state in RMP table. */
140 	sev_es_wr_ghcb_msr(GHCB_MSR_PSC_REQ_GFN(paddr >> PAGE_SHIFT, op));
141 	VMGEXIT();
142 
143 	/* Read the response of the VMGEXIT. */
144 	val = sev_es_rd_ghcb_msr();
145 	if ((GHCB_RESP_CODE(val) != GHCB_MSR_PSC_RESP) || GHCB_MSR_PSC_RESP_VAL(val))
146 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PSC);
147 
148 	/*
149 	 * Now that page state is changed in the RMP table, validate it so that it is
150 	 * consistent with the RMP entry.
151 	 */
152 	if (op == SNP_PAGE_STATE_PRIVATE && pvalidate(paddr, RMP_PG_SIZE_4K, 1))
153 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_PVALIDATE);
154 }
155 
156 void snp_set_page_private(unsigned long paddr)
157 {
158 	__page_state_change(paddr, SNP_PAGE_STATE_PRIVATE);
159 }
160 
161 void snp_set_page_shared(unsigned long paddr)
162 {
163 	__page_state_change(paddr, SNP_PAGE_STATE_SHARED);
164 }
165 
166 static bool early_setup_ghcb(void)
167 {
168 	if (set_page_decrypted((unsigned long)&boot_ghcb_page))
169 		return false;
170 
171 	/* Page is now mapped decrypted, clear it */
172 	memset(&boot_ghcb_page, 0, sizeof(boot_ghcb_page));
173 
174 	boot_ghcb = &boot_ghcb_page;
175 
176 	/* Initialize lookup tables for the instruction decoder */
177 	inat_init_tables();
178 
179 	/* SNP guest requires the GHCB GPA must be registered */
180 	if (sev_snp_enabled())
181 		snp_register_ghcb_early(__pa(&boot_ghcb_page));
182 
183 	return true;
184 }
185 
186 void sev_es_shutdown_ghcb(void)
187 {
188 	if (!boot_ghcb)
189 		return;
190 
191 	if (!sev_es_check_cpu_features())
192 		error("SEV-ES CPU Features missing.");
193 
194 	/*
195 	 * GHCB Page must be flushed from the cache and mapped encrypted again.
196 	 * Otherwise the running kernel will see strange cache effects when
197 	 * trying to use that page.
198 	 */
199 	if (set_page_encrypted((unsigned long)&boot_ghcb_page))
200 		error("Can't map GHCB page encrypted");
201 
202 	/*
203 	 * GHCB page is mapped encrypted again and flushed from the cache.
204 	 * Mark it non-present now to catch bugs when #VC exceptions trigger
205 	 * after this point.
206 	 */
207 	if (set_page_non_present((unsigned long)&boot_ghcb_page))
208 		error("Can't unmap GHCB page");
209 }
210 
211 bool sev_es_check_ghcb_fault(unsigned long address)
212 {
213 	/* Check whether the fault was on the GHCB page */
214 	return ((address & PAGE_MASK) == (unsigned long)&boot_ghcb_page);
215 }
216 
217 void do_boot_stage2_vc(struct pt_regs *regs, unsigned long exit_code)
218 {
219 	struct es_em_ctxt ctxt;
220 	enum es_result result;
221 
222 	if (!boot_ghcb && !early_setup_ghcb())
223 		sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
224 
225 	vc_ghcb_invalidate(boot_ghcb);
226 	result = vc_init_em_ctxt(&ctxt, regs, exit_code);
227 	if (result != ES_OK)
228 		goto finish;
229 
230 	switch (exit_code) {
231 	case SVM_EXIT_RDTSC:
232 	case SVM_EXIT_RDTSCP:
233 		result = vc_handle_rdtsc(boot_ghcb, &ctxt, exit_code);
234 		break;
235 	case SVM_EXIT_IOIO:
236 		result = vc_handle_ioio(boot_ghcb, &ctxt);
237 		break;
238 	case SVM_EXIT_CPUID:
239 		result = vc_handle_cpuid(boot_ghcb, &ctxt);
240 		break;
241 	default:
242 		result = ES_UNSUPPORTED;
243 		break;
244 	}
245 
246 finish:
247 	if (result == ES_OK)
248 		vc_finish_insn(&ctxt);
249 	else if (result != ES_RETRY)
250 		sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
251 }
252 
253 static void enforce_vmpl0(void)
254 {
255 	u64 attrs;
256 	int err;
257 
258 	/*
259 	 * RMPADJUST modifies RMP permissions of a lesser-privileged (numerically
260 	 * higher) privilege level. Here, clear the VMPL1 permission mask of the
261 	 * GHCB page. If the guest is not running at VMPL0, this will fail.
262 	 *
263 	 * If the guest is running at VMPL0, it will succeed. Even if that operation
264 	 * modifies permission bits, it is still ok to do so currently because Linux
265 	 * SNP guests are supported only on VMPL0 so VMPL1 or higher permission masks
266 	 * changing is a don't-care.
267 	 */
268 	attrs = 1;
269 	if (rmpadjust((unsigned long)&boot_ghcb_page, RMP_PG_SIZE_4K, attrs))
270 		sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_NOT_VMPL0);
271 }
272 
273 void sev_enable(struct boot_params *bp)
274 {
275 	unsigned int eax, ebx, ecx, edx;
276 	struct msr m;
277 	bool snp;
278 
279 	/*
280 	 * bp->cc_blob_address should only be set by boot/compressed kernel.
281 	 * Initialize it to 0 to ensure that uninitialized values from
282 	 * buggy bootloaders aren't propagated.
283 	 */
284 	if (bp)
285 		bp->cc_blob_address = 0;
286 
287 	/*
288 	 * Setup/preliminary detection of SNP. This will be sanity-checked
289 	 * against CPUID/MSR values later.
290 	 */
291 	snp = snp_init(bp);
292 
293 	/* Check for the SME/SEV support leaf */
294 	eax = 0x80000000;
295 	ecx = 0;
296 	native_cpuid(&eax, &ebx, &ecx, &edx);
297 	if (eax < 0x8000001f)
298 		return;
299 
300 	/*
301 	 * Check for the SME/SEV feature:
302 	 *   CPUID Fn8000_001F[EAX]
303 	 *   - Bit 0 - Secure Memory Encryption support
304 	 *   - Bit 1 - Secure Encrypted Virtualization support
305 	 *   CPUID Fn8000_001F[EBX]
306 	 *   - Bits 5:0 - Pagetable bit position used to indicate encryption
307 	 */
308 	eax = 0x8000001f;
309 	ecx = 0;
310 	native_cpuid(&eax, &ebx, &ecx, &edx);
311 	/* Check whether SEV is supported */
312 	if (!(eax & BIT(1))) {
313 		if (snp)
314 			error("SEV-SNP support indicated by CC blob, but not CPUID.");
315 		return;
316 	}
317 
318 	/* Set the SME mask if this is an SEV guest. */
319 	boot_rdmsr(MSR_AMD64_SEV, &m);
320 	sev_status = m.q;
321 	if (!(sev_status & MSR_AMD64_SEV_ENABLED))
322 		return;
323 
324 	/* Negotiate the GHCB protocol version. */
325 	if (sev_status & MSR_AMD64_SEV_ES_ENABLED) {
326 		if (!sev_es_negotiate_protocol())
327 			sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_PROT_UNSUPPORTED);
328 	}
329 
330 	/*
331 	 * SNP is supported in v2 of the GHCB spec which mandates support for HV
332 	 * features.
333 	 */
334 	if (sev_status & MSR_AMD64_SEV_SNP_ENABLED) {
335 		if (!(get_hv_features() & GHCB_HV_FT_SNP))
336 			sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
337 
338 		enforce_vmpl0();
339 	}
340 
341 	if (snp && !(sev_status & MSR_AMD64_SEV_SNP_ENABLED))
342 		error("SEV-SNP supported indicated by CC blob, but not SEV status MSR.");
343 
344 	sme_me_mask = BIT_ULL(ebx & 0x3f);
345 }
346 
347 /* Search for Confidential Computing blob in the EFI config table. */
348 static struct cc_blob_sev_info *find_cc_blob_efi(struct boot_params *bp)
349 {
350 	unsigned long cfg_table_pa;
351 	unsigned int cfg_table_len;
352 	int ret;
353 
354 	ret = efi_get_conf_table(bp, &cfg_table_pa, &cfg_table_len);
355 	if (ret)
356 		return NULL;
357 
358 	return (struct cc_blob_sev_info *)efi_find_vendor_table(bp, cfg_table_pa,
359 								cfg_table_len,
360 								EFI_CC_BLOB_GUID);
361 }
362 
363 /*
364  * Initial set up of SNP relies on information provided by the
365  * Confidential Computing blob, which can be passed to the boot kernel
366  * by firmware/bootloader in the following ways:
367  *
368  * - via an entry in the EFI config table
369  * - via a setup_data structure, as defined by the Linux Boot Protocol
370  *
371  * Scan for the blob in that order.
372  */
373 static struct cc_blob_sev_info *find_cc_blob(struct boot_params *bp)
374 {
375 	struct cc_blob_sev_info *cc_info;
376 
377 	cc_info = find_cc_blob_efi(bp);
378 	if (cc_info)
379 		goto found_cc_info;
380 
381 	cc_info = find_cc_blob_setup_data(bp);
382 	if (!cc_info)
383 		return NULL;
384 
385 found_cc_info:
386 	if (cc_info->magic != CC_BLOB_SEV_HDR_MAGIC)
387 		sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
388 
389 	return cc_info;
390 }
391 
392 /*
393  * Indicate SNP based on presence of SNP-specific CC blob. Subsequent checks
394  * will verify the SNP CPUID/MSR bits.
395  */
396 bool snp_init(struct boot_params *bp)
397 {
398 	struct cc_blob_sev_info *cc_info;
399 
400 	if (!bp)
401 		return false;
402 
403 	cc_info = find_cc_blob(bp);
404 	if (!cc_info)
405 		return false;
406 
407 	/*
408 	 * If a SNP-specific Confidential Computing blob is present, then
409 	 * firmware/bootloader have indicated SNP support. Verifying this
410 	 * involves CPUID checks which will be more reliable if the SNP
411 	 * CPUID table is used. See comments over snp_setup_cpuid_table() for
412 	 * more details.
413 	 */
414 	setup_cpuid_table(cc_info);
415 
416 	/*
417 	 * Pass run-time kernel a pointer to CC info via boot_params so EFI
418 	 * config table doesn't need to be searched again during early startup
419 	 * phase.
420 	 */
421 	bp->cc_blob_address = (u32)(unsigned long)cc_info;
422 
423 	return true;
424 }
425 
426 void sev_prep_identity_maps(unsigned long top_level_pgt)
427 {
428 	/*
429 	 * The Confidential Computing blob is used very early in uncompressed
430 	 * kernel to find the in-memory CPUID table to handle CPUID
431 	 * instructions. Make sure an identity-mapping exists so it can be
432 	 * accessed after switchover.
433 	 */
434 	if (sev_snp_enabled()) {
435 		unsigned long cc_info_pa = boot_params->cc_blob_address;
436 		struct cc_blob_sev_info *cc_info;
437 
438 		kernel_add_identity_map(cc_info_pa, cc_info_pa + sizeof(*cc_info));
439 
440 		cc_info = (struct cc_blob_sev_info *)cc_info_pa;
441 		kernel_add_identity_map(cc_info->cpuid_phys, cc_info->cpuid_phys + cc_info->cpuid_len);
442 	}
443 
444 	sev_verify_cbit(top_level_pgt);
445 }
446