Merge tag 'x86_kdump_for_v5.19_rc1' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-block.git] / arch / x86 / boot / compressed / sev.c
CommitLineData
29dcc60f
JR
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
39336f4f 15#include <asm/pgtable_types.h>
e759959f 16#include <asm/sev.h>
597cfe48
JR
17#include <asm/trapnr.h>
18#include <asm/trap_pf.h>
29dcc60f 19#include <asm/msr-index.h>
a7de15d4 20#include <asm/fpu/xcr.h>
29dcc60f
JR
21#include <asm/ptrace.h>
22#include <asm/svm.h>
801baa69 23#include <asm/cpuid.h>
29dcc60f 24
597cfe48 25#include "error.h"
950d0055 26#include "../msr.h"
597cfe48
JR
27
28struct ghcb boot_ghcb_page __aligned(PAGE_SIZE);
29struct ghcb *boot_ghcb;
30
25189d08
TL
31/*
32 * Copy a version of this function here - insn-eval.c can't be used in
33 * pre-decompression code.
34 */
35static bool insn_has_rep_prefix(struct insn *insn)
36{
84da009f 37 insn_byte_t p;
25189d08
TL
38 int i;
39
40 insn_get_prefixes(insn);
41
84da009f 42 for_each_insn_prefix(insn, i, p) {
25189d08
TL
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 */
54static unsigned long insn_get_seg_base(struct pt_regs *regs, int seg_reg_idx)
55{
56 return 0UL;
57}
58
29dcc60f
JR
59static inline u64 sev_es_rd_ghcb_msr(void)
60{
950d0055 61 struct msr m;
29dcc60f 62
950d0055 63 boot_rdmsr(MSR_AMD64_SEV_ES_GHCB, &m);
29dcc60f 64
950d0055 65 return m.q;
29dcc60f
JR
66}
67
68static inline void sev_es_wr_ghcb_msr(u64 val)
69{
950d0055 70 struct msr m;
29dcc60f 71
950d0055
MR
72 m.q = val;
73 boot_wrmsr(MSR_AMD64_SEV_ES_GHCB, &m);
29dcc60f
JR
74}
75
597cfe48
JR
76static enum es_result vc_decode_insn(struct es_em_ctxt *ctxt)
77{
78 char buffer[MAX_INSN_SIZE];
514ef776 79 int ret;
597cfe48
JR
80
81 memcpy(buffer, (unsigned char *)ctxt->regs->ip, MAX_INSN_SIZE);
82
514ef776
BP
83 ret = insn_decode(&ctxt->insn, buffer, MAX_INSN_SIZE, INSN_MODE_64);
84 if (ret < 0)
85 return ES_DECODE_FAILED;
597cfe48 86
514ef776 87 return ES_OK;
597cfe48
JR
88}
89
90static 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
98static 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
29dcc60f 106#undef __init
597cfe48 107#undef __pa
29dcc60f 108#define __init
597cfe48
JR
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"
29dcc60f
JR
116
117/* Include code for early handlers */
e759959f 118#include "../../kernel/sev-shared.c"
597cfe48 119
4f9c403e
BS
120static inline bool sev_snp_enabled(void)
121{
122 return sev_status & MSR_AMD64_SEV_SNP_ENABLED;
123}
124
125static 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
156void snp_set_page_private(unsigned long paddr)
157{
158 __page_state_change(paddr, SNP_PAGE_STATE_PRIVATE);
159}
160
161void snp_set_page_shared(unsigned long paddr)
162{
163 __page_state_change(paddr, SNP_PAGE_STATE_SHARED);
164}
165
cbd3d4f7 166static bool early_setup_ghcb(void)
597cfe48 167{
597cfe48
JR
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
87294bdb
BS
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
597cfe48
JR
183 return true;
184}
185
186void sev_es_shutdown_ghcb(void)
187{
188 if (!boot_ghcb)
189 return;
190
f5ed7775
MR
191 if (!sev_es_check_cpu_features())
192 error("SEV-ES CPU Features missing.");
193
597cfe48
JR
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");
69add17a
JR
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
211bool 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);
597cfe48
JR
215}
216
217void 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
cbd3d4f7 222 if (!boot_ghcb && !early_setup_ghcb())
6c0f74d6 223 sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
597cfe48
JR
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) {
4711e7ac
TL
231 case SVM_EXIT_RDTSC:
232 case SVM_EXIT_RDTSCP:
233 result = vc_handle_rdtsc(boot_ghcb, &ctxt, exit_code);
234 break;
25189d08
TL
235 case SVM_EXIT_IOIO:
236 result = vc_handle_ioio(boot_ghcb, &ctxt);
237 break;
a7de15d4
TL
238 case SVM_EXIT_CPUID:
239 result = vc_handle_cpuid(boot_ghcb, &ctxt);
240 break;
597cfe48
JR
241 default:
242 result = ES_UNSUPPORTED;
243 break;
244 }
245
246finish:
f15a0a73 247 if (result == ES_OK)
597cfe48 248 vc_finish_insn(&ctxt);
f15a0a73 249 else if (result != ES_RETRY)
6c0f74d6 250 sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_GEN_REQ);
597cfe48 251}
ec1c66af 252
81cc3df9
BS
253static 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
ec1c66af
MR
273void sev_enable(struct boot_params *bp)
274{
275 unsigned int eax, ebx, ecx, edx;
276 struct msr m;
c01fce9c
MR
277 bool snp;
278
279 /*
280 * Setup/preliminary detection of SNP. This will be sanity-checked
281 * against CPUID/MSR values later.
282 */
283 snp = snp_init(bp);
ec1c66af
MR
284
285 /* Check for the SME/SEV support leaf */
286 eax = 0x80000000;
287 ecx = 0;
288 native_cpuid(&eax, &ebx, &ecx, &edx);
289 if (eax < 0x8000001f)
290 return;
291
292 /*
293 * Check for the SME/SEV feature:
294 * CPUID Fn8000_001F[EAX]
295 * - Bit 0 - Secure Memory Encryption support
296 * - Bit 1 - Secure Encrypted Virtualization support
297 * CPUID Fn8000_001F[EBX]
298 * - Bits 5:0 - Pagetable bit position used to indicate encryption
299 */
300 eax = 0x8000001f;
301 ecx = 0;
302 native_cpuid(&eax, &ebx, &ecx, &edx);
303 /* Check whether SEV is supported */
c01fce9c
MR
304 if (!(eax & BIT(1))) {
305 if (snp)
306 error("SEV-SNP support indicated by CC blob, but not CPUID.");
ec1c66af 307 return;
c01fce9c 308 }
ec1c66af
MR
309
310 /* Set the SME mask if this is an SEV guest. */
311 boot_rdmsr(MSR_AMD64_SEV, &m);
312 sev_status = m.q;
313 if (!(sev_status & MSR_AMD64_SEV_ENABLED))
314 return;
315
cbd3d4f7
BS
316 /* Negotiate the GHCB protocol version. */
317 if (sev_status & MSR_AMD64_SEV_ES_ENABLED) {
318 if (!sev_es_negotiate_protocol())
319 sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_PROT_UNSUPPORTED);
320 }
321
322 /*
323 * SNP is supported in v2 of the GHCB spec which mandates support for HV
324 * features.
325 */
81cc3df9
BS
326 if (sev_status & MSR_AMD64_SEV_SNP_ENABLED) {
327 if (!(get_hv_features() & GHCB_HV_FT_SNP))
328 sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
329
330 enforce_vmpl0();
331 }
cbd3d4f7 332
c01fce9c
MR
333 if (snp && !(sev_status & MSR_AMD64_SEV_SNP_ENABLED))
334 error("SEV-SNP supported indicated by CC blob, but not SEV status MSR.");
335
ec1c66af
MR
336 sme_me_mask = BIT_ULL(ebx & 0x3f);
337}
c01fce9c
MR
338
339/* Search for Confidential Computing blob in the EFI config table. */
340static struct cc_blob_sev_info *find_cc_blob_efi(struct boot_params *bp)
341{
342 unsigned long cfg_table_pa;
343 unsigned int cfg_table_len;
344 int ret;
345
346 ret = efi_get_conf_table(bp, &cfg_table_pa, &cfg_table_len);
347 if (ret)
348 return NULL;
349
350 return (struct cc_blob_sev_info *)efi_find_vendor_table(bp, cfg_table_pa,
351 cfg_table_len,
352 EFI_CC_BLOB_GUID);
353}
354
c01fce9c
MR
355/*
356 * Initial set up of SNP relies on information provided by the
357 * Confidential Computing blob, which can be passed to the boot kernel
358 * by firmware/bootloader in the following ways:
359 *
360 * - via an entry in the EFI config table
361 * - via a setup_data structure, as defined by the Linux Boot Protocol
362 *
363 * Scan for the blob in that order.
364 */
365static struct cc_blob_sev_info *find_cc_blob(struct boot_params *bp)
366{
367 struct cc_blob_sev_info *cc_info;
368
369 cc_info = find_cc_blob_efi(bp);
370 if (cc_info)
371 goto found_cc_info;
372
373 cc_info = find_cc_blob_setup_data(bp);
374 if (!cc_info)
375 return NULL;
376
377found_cc_info:
378 if (cc_info->magic != CC_BLOB_SEV_HDR_MAGIC)
379 sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
380
381 return cc_info;
382}
383
384/*
385 * Indicate SNP based on presence of SNP-specific CC blob. Subsequent checks
386 * will verify the SNP CPUID/MSR bits.
387 */
388bool snp_init(struct boot_params *bp)
389{
390 struct cc_blob_sev_info *cc_info;
391
392 if (!bp)
393 return false;
394
395 cc_info = find_cc_blob(bp);
396 if (!cc_info)
397 return false;
398
5f211f4f
MR
399 /*
400 * If a SNP-specific Confidential Computing blob is present, then
401 * firmware/bootloader have indicated SNP support. Verifying this
402 * involves CPUID checks which will be more reliable if the SNP
403 * CPUID table is used. See comments over snp_setup_cpuid_table() for
404 * more details.
405 */
406 setup_cpuid_table(cc_info);
407
c01fce9c
MR
408 /*
409 * Pass run-time kernel a pointer to CC info via boot_params so EFI
410 * config table doesn't need to be searched again during early startup
411 * phase.
412 */
413 bp->cc_blob_address = (u32)(unsigned long)cc_info;
414
415 return true;
416}
76f61e1e
MR
417
418void sev_prep_identity_maps(unsigned long top_level_pgt)
419{
420 /*
421 * The Confidential Computing blob is used very early in uncompressed
422 * kernel to find the in-memory CPUID table to handle CPUID
423 * instructions. Make sure an identity-mapping exists so it can be
424 * accessed after switchover.
425 */
426 if (sev_snp_enabled()) {
427 unsigned long cc_info_pa = boot_params->cc_blob_address;
428 struct cc_blob_sev_info *cc_info;
429
430 kernel_add_identity_map(cc_info_pa, cc_info_pa + sizeof(*cc_info));
431
432 cc_info = (struct cc_blob_sev_info *)cc_info_pa;
433 kernel_add_identity_map(cc_info->cpuid_phys, cc_info->cpuid_phys + cc_info->cpuid_len);
434 }
435
436 sev_verify_cbit(top_level_pgt);
437}