1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 1994 Linus Torvalds
4 *
5 * Cyrix stuff, June 1998 by:
6 * - Rafael R. Reilova (moved everything from head.S),
7 * <rreilova@ececs.uc.edu>
8 * - Channing Corn (tests & fixes),
9 * - Andrew D. Balsa (code cleanup).
10 */
11 #include <linux/init.h>
12 #include <linux/utsname.h>
13 #include <linux/cpu.h>
14 #include <linux/module.h>
15 #include <linux/nospec.h>
16 #include <linux/prctl.h>
17 #include <linux/sched/smt.h>
18 #include <linux/pgtable.h>
19 #include <linux/bpf.h>
20
21 #include <asm/spec-ctrl.h>
22 #include <asm/cmdline.h>
23 #include <asm/bugs.h>
24 #include <asm/processor.h>
25 #include <asm/processor-flags.h>
26 #include <asm/fpu/api.h>
27 #include <asm/msr.h>
28 #include <asm/vmx.h>
29 #include <asm/paravirt.h>
30 #include <asm/alternative.h>
31 #include <asm/set_memory.h>
32 #include <asm/intel-family.h>
33 #include <asm/e820/api.h>
34 #include <asm/hypervisor.h>
35 #include <asm/tlbflush.h>
36 #include <asm/cpu.h>
37
38 #include "cpu.h"
39
40 static void __init spectre_v1_select_mitigation(void);
41 static void __init spectre_v2_select_mitigation(void);
42 static void __init retbleed_select_mitigation(void);
43 static void __init spectre_v2_user_select_mitigation(void);
44 static void __init ssb_select_mitigation(void);
45 static void __init l1tf_select_mitigation(void);
46 static void __init mds_select_mitigation(void);
47 static void __init md_clear_update_mitigation(void);
48 static void __init md_clear_select_mitigation(void);
49 static void __init taa_select_mitigation(void);
50 static void __init mmio_select_mitigation(void);
51 static void __init srbds_select_mitigation(void);
52 static void __init l1d_flush_select_mitigation(void);
53
54 /* The base value of the SPEC_CTRL MSR without task-specific bits set */
55 u64 x86_spec_ctrl_base;
56 EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
57
58 /* The current value of the SPEC_CTRL MSR with task-specific bits set */
59 DEFINE_PER_CPU(u64, x86_spec_ctrl_current);
60 EXPORT_SYMBOL_GPL(x86_spec_ctrl_current);
61
62 static DEFINE_MUTEX(spec_ctrl_mutex);
63
64 /* Update SPEC_CTRL MSR and its cached copy unconditionally */
update_spec_ctrl(u64 val)65 static void update_spec_ctrl(u64 val)
66 {
67 this_cpu_write(x86_spec_ctrl_current, val);
68 wrmsrl(MSR_IA32_SPEC_CTRL, val);
69 }
70
71 /*
72 * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
73 * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
74 */
update_spec_ctrl_cond(u64 val)75 void update_spec_ctrl_cond(u64 val)
76 {
77 if (this_cpu_read(x86_spec_ctrl_current) == val)
78 return;
79
80 this_cpu_write(x86_spec_ctrl_current, val);
81
82 /*
83 * When KERNEL_IBRS this MSR is written on return-to-user, unless
84 * forced the update can be delayed until that time.
85 */
86 if (!cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
87 wrmsrl(MSR_IA32_SPEC_CTRL, val);
88 }
89
spec_ctrl_current(void)90 noinstr u64 spec_ctrl_current(void)
91 {
92 return this_cpu_read(x86_spec_ctrl_current);
93 }
94 EXPORT_SYMBOL_GPL(spec_ctrl_current);
95
96 /*
97 * AMD specific MSR info for Speculative Store Bypass control.
98 * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
99 */
100 u64 __ro_after_init x86_amd_ls_cfg_base;
101 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
102
103 /* Control conditional STIBP in switch_to() */
104 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
105 /* Control conditional IBPB in switch_mm() */
106 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
107 /* Control unconditional IBPB in switch_mm() */
108 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
109
110 /* Control MDS CPU buffer clear before returning to user space */
111 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
112 EXPORT_SYMBOL_GPL(mds_user_clear);
113 /* Control MDS CPU buffer clear before idling (halt, mwait) */
114 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
115 EXPORT_SYMBOL_GPL(mds_idle_clear);
116
117 /*
118 * Controls whether l1d flush based mitigations are enabled,
119 * based on hw features and admin setting via boot parameter
120 * defaults to false
121 */
122 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
123
124 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
125 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
126 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
127
check_bugs(void)128 void __init check_bugs(void)
129 {
130 identify_boot_cpu();
131
132 /*
133 * identify_boot_cpu() initialized SMT support information, let the
134 * core code know.
135 */
136 cpu_smt_check_topology();
137
138 if (!IS_ENABLED(CONFIG_SMP)) {
139 pr_info("CPU: ");
140 print_cpu_info(&boot_cpu_data);
141 }
142
143 /*
144 * Read the SPEC_CTRL MSR to account for reserved bits which may
145 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
146 * init code as it is not enumerated and depends on the family.
147 */
148 if (cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL)) {
149 rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
150
151 /*
152 * Previously running kernel (kexec), may have some controls
153 * turned ON. Clear them and let the mitigations setup below
154 * rediscover them based on configuration.
155 */
156 x86_spec_ctrl_base &= ~SPEC_CTRL_MITIGATIONS_MASK;
157 }
158
159 /* Select the proper CPU mitigations before patching alternatives: */
160 spectre_v1_select_mitigation();
161 spectre_v2_select_mitigation();
162 /*
163 * retbleed_select_mitigation() relies on the state set by
164 * spectre_v2_select_mitigation(); specifically it wants to know about
165 * spectre_v2=ibrs.
166 */
167 retbleed_select_mitigation();
168 /*
169 * spectre_v2_user_select_mitigation() relies on the state set by
170 * retbleed_select_mitigation(); specifically the STIBP selection is
171 * forced for UNRET or IBPB.
172 */
173 spectre_v2_user_select_mitigation();
174 ssb_select_mitigation();
175 l1tf_select_mitigation();
176 md_clear_select_mitigation();
177 srbds_select_mitigation();
178 l1d_flush_select_mitigation();
179
180 arch_smt_update();
181
182 #ifdef CONFIG_X86_32
183 /*
184 * Check whether we are able to run this kernel safely on SMP.
185 *
186 * - i386 is no longer supported.
187 * - In order to run on anything without a TSC, we need to be
188 * compiled for a i486.
189 */
190 if (boot_cpu_data.x86 < 4)
191 panic("Kernel requires i486+ for 'invlpg' and other features");
192
193 init_utsname()->machine[1] =
194 '0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
195 alternative_instructions();
196
197 fpu__init_check_bugs();
198 #else /* CONFIG_X86_64 */
199 alternative_instructions();
200
201 /*
202 * Make sure the first 2MB area is not mapped by huge pages
203 * There are typically fixed size MTRRs in there and overlapping
204 * MTRRs into large pages causes slow downs.
205 *
206 * Right now we don't do that with gbpages because there seems
207 * very little benefit for that case.
208 */
209 if (!direct_gbpages)
210 set_memory_4k((unsigned long)__va(0), 1);
211 #endif
212 }
213
214 /*
215 * NOTE: This function is *only* called for SVM, since Intel uses
216 * MSR_IA32_SPEC_CTRL for SSBD.
217 */
218 void
x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl,bool setguest)219 x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl, bool setguest)
220 {
221 u64 guestval, hostval;
222 struct thread_info *ti = current_thread_info();
223
224 /*
225 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
226 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
227 */
228 if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
229 !static_cpu_has(X86_FEATURE_VIRT_SSBD))
230 return;
231
232 /*
233 * If the host has SSBD mitigation enabled, force it in the host's
234 * virtual MSR value. If its not permanently enabled, evaluate
235 * current's TIF_SSBD thread flag.
236 */
237 if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
238 hostval = SPEC_CTRL_SSBD;
239 else
240 hostval = ssbd_tif_to_spec_ctrl(ti->flags);
241
242 /* Sanitize the guest value */
243 guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
244
245 if (hostval != guestval) {
246 unsigned long tif;
247
248 tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
249 ssbd_spec_ctrl_to_tif(hostval);
250
251 speculation_ctrl_update(tif);
252 }
253 }
254 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
255
x86_amd_ssb_disable(void)256 static void x86_amd_ssb_disable(void)
257 {
258 u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
259
260 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
261 wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
262 else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
263 wrmsrl(MSR_AMD64_LS_CFG, msrval);
264 }
265
266 #undef pr_fmt
267 #define pr_fmt(fmt) "MDS: " fmt
268
269 /* Default mitigation for MDS-affected CPUs */
270 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
271 static bool mds_nosmt __ro_after_init = false;
272
273 static const char * const mds_strings[] = {
274 [MDS_MITIGATION_OFF] = "Vulnerable",
275 [MDS_MITIGATION_FULL] = "Mitigation: Clear CPU buffers",
276 [MDS_MITIGATION_VMWERV] = "Vulnerable: Clear CPU buffers attempted, no microcode",
277 };
278
mds_select_mitigation(void)279 static void __init mds_select_mitigation(void)
280 {
281 if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
282 mds_mitigation = MDS_MITIGATION_OFF;
283 return;
284 }
285
286 if (mds_mitigation == MDS_MITIGATION_FULL) {
287 if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
288 mds_mitigation = MDS_MITIGATION_VMWERV;
289
290 static_branch_enable(&mds_user_clear);
291
292 if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
293 (mds_nosmt || cpu_mitigations_auto_nosmt()))
294 cpu_smt_disable(false);
295 }
296 }
297
mds_cmdline(char * str)298 static int __init mds_cmdline(char *str)
299 {
300 if (!boot_cpu_has_bug(X86_BUG_MDS))
301 return 0;
302
303 if (!str)
304 return -EINVAL;
305
306 if (!strcmp(str, "off"))
307 mds_mitigation = MDS_MITIGATION_OFF;
308 else if (!strcmp(str, "full"))
309 mds_mitigation = MDS_MITIGATION_FULL;
310 else if (!strcmp(str, "full,nosmt")) {
311 mds_mitigation = MDS_MITIGATION_FULL;
312 mds_nosmt = true;
313 }
314
315 return 0;
316 }
317 early_param("mds", mds_cmdline);
318
319 #undef pr_fmt
320 #define pr_fmt(fmt) "TAA: " fmt
321
322 enum taa_mitigations {
323 TAA_MITIGATION_OFF,
324 TAA_MITIGATION_UCODE_NEEDED,
325 TAA_MITIGATION_VERW,
326 TAA_MITIGATION_TSX_DISABLED,
327 };
328
329 /* Default mitigation for TAA-affected CPUs */
330 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
331 static bool taa_nosmt __ro_after_init;
332
333 static const char * const taa_strings[] = {
334 [TAA_MITIGATION_OFF] = "Vulnerable",
335 [TAA_MITIGATION_UCODE_NEEDED] = "Vulnerable: Clear CPU buffers attempted, no microcode",
336 [TAA_MITIGATION_VERW] = "Mitigation: Clear CPU buffers",
337 [TAA_MITIGATION_TSX_DISABLED] = "Mitigation: TSX disabled",
338 };
339
taa_select_mitigation(void)340 static void __init taa_select_mitigation(void)
341 {
342 u64 ia32_cap;
343
344 if (!boot_cpu_has_bug(X86_BUG_TAA)) {
345 taa_mitigation = TAA_MITIGATION_OFF;
346 return;
347 }
348
349 /* TSX previously disabled by tsx=off */
350 if (!boot_cpu_has(X86_FEATURE_RTM)) {
351 taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
352 return;
353 }
354
355 if (cpu_mitigations_off()) {
356 taa_mitigation = TAA_MITIGATION_OFF;
357 return;
358 }
359
360 /*
361 * TAA mitigation via VERW is turned off if both
362 * tsx_async_abort=off and mds=off are specified.
363 */
364 if (taa_mitigation == TAA_MITIGATION_OFF &&
365 mds_mitigation == MDS_MITIGATION_OFF)
366 return;
367
368 if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
369 taa_mitigation = TAA_MITIGATION_VERW;
370 else
371 taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
372
373 /*
374 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
375 * A microcode update fixes this behavior to clear CPU buffers. It also
376 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
377 * ARCH_CAP_TSX_CTRL_MSR bit.
378 *
379 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
380 * update is required.
381 */
382 ia32_cap = x86_read_arch_cap_msr();
383 if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
384 !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
385 taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
386
387 /*
388 * TSX is enabled, select alternate mitigation for TAA which is
389 * the same as MDS. Enable MDS static branch to clear CPU buffers.
390 *
391 * For guests that can't determine whether the correct microcode is
392 * present on host, enable the mitigation for UCODE_NEEDED as well.
393 */
394 static_branch_enable(&mds_user_clear);
395
396 if (taa_nosmt || cpu_mitigations_auto_nosmt())
397 cpu_smt_disable(false);
398 }
399
tsx_async_abort_parse_cmdline(char * str)400 static int __init tsx_async_abort_parse_cmdline(char *str)
401 {
402 if (!boot_cpu_has_bug(X86_BUG_TAA))
403 return 0;
404
405 if (!str)
406 return -EINVAL;
407
408 if (!strcmp(str, "off")) {
409 taa_mitigation = TAA_MITIGATION_OFF;
410 } else if (!strcmp(str, "full")) {
411 taa_mitigation = TAA_MITIGATION_VERW;
412 } else if (!strcmp(str, "full,nosmt")) {
413 taa_mitigation = TAA_MITIGATION_VERW;
414 taa_nosmt = true;
415 }
416
417 return 0;
418 }
419 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
420
421 #undef pr_fmt
422 #define pr_fmt(fmt) "MMIO Stale Data: " fmt
423
424 enum mmio_mitigations {
425 MMIO_MITIGATION_OFF,
426 MMIO_MITIGATION_UCODE_NEEDED,
427 MMIO_MITIGATION_VERW,
428 };
429
430 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
431 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
432 static bool mmio_nosmt __ro_after_init = false;
433
434 static const char * const mmio_strings[] = {
435 [MMIO_MITIGATION_OFF] = "Vulnerable",
436 [MMIO_MITIGATION_UCODE_NEEDED] = "Vulnerable: Clear CPU buffers attempted, no microcode",
437 [MMIO_MITIGATION_VERW] = "Mitigation: Clear CPU buffers",
438 };
439
mmio_select_mitigation(void)440 static void __init mmio_select_mitigation(void)
441 {
442 u64 ia32_cap;
443
444 if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
445 boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
446 cpu_mitigations_off()) {
447 mmio_mitigation = MMIO_MITIGATION_OFF;
448 return;
449 }
450
451 if (mmio_mitigation == MMIO_MITIGATION_OFF)
452 return;
453
454 ia32_cap = x86_read_arch_cap_msr();
455
456 /*
457 * Enable CPU buffer clear mitigation for host and VMM, if also affected
458 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
459 */
460 if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
461 boot_cpu_has(X86_FEATURE_RTM)))
462 static_branch_enable(&mds_user_clear);
463 else
464 static_branch_enable(&mmio_stale_data_clear);
465
466 /*
467 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
468 * be propagated to uncore buffers, clearing the Fill buffers on idle
469 * is required irrespective of SMT state.
470 */
471 if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
472 static_branch_enable(&mds_idle_clear);
473
474 /*
475 * Check if the system has the right microcode.
476 *
477 * CPU Fill buffer clear mitigation is enumerated by either an explicit
478 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
479 * affected systems.
480 */
481 if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
482 (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
483 boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
484 !(ia32_cap & ARCH_CAP_MDS_NO)))
485 mmio_mitigation = MMIO_MITIGATION_VERW;
486 else
487 mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
488
489 if (mmio_nosmt || cpu_mitigations_auto_nosmt())
490 cpu_smt_disable(false);
491 }
492
mmio_stale_data_parse_cmdline(char * str)493 static int __init mmio_stale_data_parse_cmdline(char *str)
494 {
495 if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
496 return 0;
497
498 if (!str)
499 return -EINVAL;
500
501 if (!strcmp(str, "off")) {
502 mmio_mitigation = MMIO_MITIGATION_OFF;
503 } else if (!strcmp(str, "full")) {
504 mmio_mitigation = MMIO_MITIGATION_VERW;
505 } else if (!strcmp(str, "full,nosmt")) {
506 mmio_mitigation = MMIO_MITIGATION_VERW;
507 mmio_nosmt = true;
508 }
509
510 return 0;
511 }
512 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
513
514 #undef pr_fmt
515 #define pr_fmt(fmt) "" fmt
516
md_clear_update_mitigation(void)517 static void __init md_clear_update_mitigation(void)
518 {
519 if (cpu_mitigations_off())
520 return;
521
522 if (!static_key_enabled(&mds_user_clear))
523 goto out;
524
525 /*
526 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
527 * mitigation, if necessary.
528 */
529 if (mds_mitigation == MDS_MITIGATION_OFF &&
530 boot_cpu_has_bug(X86_BUG_MDS)) {
531 mds_mitigation = MDS_MITIGATION_FULL;
532 mds_select_mitigation();
533 }
534 if (taa_mitigation == TAA_MITIGATION_OFF &&
535 boot_cpu_has_bug(X86_BUG_TAA)) {
536 taa_mitigation = TAA_MITIGATION_VERW;
537 taa_select_mitigation();
538 }
539 if (mmio_mitigation == MMIO_MITIGATION_OFF &&
540 boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
541 mmio_mitigation = MMIO_MITIGATION_VERW;
542 mmio_select_mitigation();
543 }
544 out:
545 if (boot_cpu_has_bug(X86_BUG_MDS))
546 pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
547 if (boot_cpu_has_bug(X86_BUG_TAA))
548 pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
549 if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
550 pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
551 else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
552 pr_info("MMIO Stale Data: Unknown: No mitigations\n");
553 }
554
md_clear_select_mitigation(void)555 static void __init md_clear_select_mitigation(void)
556 {
557 mds_select_mitigation();
558 taa_select_mitigation();
559 mmio_select_mitigation();
560
561 /*
562 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
563 * and print their mitigation after MDS, TAA and MMIO Stale Data
564 * mitigation selection is done.
565 */
566 md_clear_update_mitigation();
567 }
568
569 #undef pr_fmt
570 #define pr_fmt(fmt) "SRBDS: " fmt
571
572 enum srbds_mitigations {
573 SRBDS_MITIGATION_OFF,
574 SRBDS_MITIGATION_UCODE_NEEDED,
575 SRBDS_MITIGATION_FULL,
576 SRBDS_MITIGATION_TSX_OFF,
577 SRBDS_MITIGATION_HYPERVISOR,
578 };
579
580 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
581
582 static const char * const srbds_strings[] = {
583 [SRBDS_MITIGATION_OFF] = "Vulnerable",
584 [SRBDS_MITIGATION_UCODE_NEEDED] = "Vulnerable: No microcode",
585 [SRBDS_MITIGATION_FULL] = "Mitigation: Microcode",
586 [SRBDS_MITIGATION_TSX_OFF] = "Mitigation: TSX disabled",
587 [SRBDS_MITIGATION_HYPERVISOR] = "Unknown: Dependent on hypervisor status",
588 };
589
590 static bool srbds_off;
591
update_srbds_msr(void)592 void update_srbds_msr(void)
593 {
594 u64 mcu_ctrl;
595
596 if (!boot_cpu_has_bug(X86_BUG_SRBDS))
597 return;
598
599 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
600 return;
601
602 if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
603 return;
604
605 /*
606 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
607 * being disabled and it hasn't received the SRBDS MSR microcode.
608 */
609 if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
610 return;
611
612 rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
613
614 switch (srbds_mitigation) {
615 case SRBDS_MITIGATION_OFF:
616 case SRBDS_MITIGATION_TSX_OFF:
617 mcu_ctrl |= RNGDS_MITG_DIS;
618 break;
619 case SRBDS_MITIGATION_FULL:
620 mcu_ctrl &= ~RNGDS_MITG_DIS;
621 break;
622 default:
623 break;
624 }
625
626 wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
627 }
628
srbds_select_mitigation(void)629 static void __init srbds_select_mitigation(void)
630 {
631 u64 ia32_cap;
632
633 if (!boot_cpu_has_bug(X86_BUG_SRBDS))
634 return;
635
636 /*
637 * Check to see if this is one of the MDS_NO systems supporting TSX that
638 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
639 * by Processor MMIO Stale Data vulnerability.
640 */
641 ia32_cap = x86_read_arch_cap_msr();
642 if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
643 !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
644 srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
645 else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
646 srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
647 else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
648 srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
649 else if (cpu_mitigations_off() || srbds_off)
650 srbds_mitigation = SRBDS_MITIGATION_OFF;
651
652 update_srbds_msr();
653 pr_info("%s\n", srbds_strings[srbds_mitigation]);
654 }
655
srbds_parse_cmdline(char * str)656 static int __init srbds_parse_cmdline(char *str)
657 {
658 if (!str)
659 return -EINVAL;
660
661 if (!boot_cpu_has_bug(X86_BUG_SRBDS))
662 return 0;
663
664 srbds_off = !strcmp(str, "off");
665 return 0;
666 }
667 early_param("srbds", srbds_parse_cmdline);
668
669 #undef pr_fmt
670 #define pr_fmt(fmt) "L1D Flush : " fmt
671
672 enum l1d_flush_mitigations {
673 L1D_FLUSH_OFF = 0,
674 L1D_FLUSH_ON,
675 };
676
677 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
678
l1d_flush_select_mitigation(void)679 static void __init l1d_flush_select_mitigation(void)
680 {
681 if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
682 return;
683
684 static_branch_enable(&switch_mm_cond_l1d_flush);
685 pr_info("Conditional flush on switch_mm() enabled\n");
686 }
687
l1d_flush_parse_cmdline(char * str)688 static int __init l1d_flush_parse_cmdline(char *str)
689 {
690 if (!strcmp(str, "on"))
691 l1d_flush_mitigation = L1D_FLUSH_ON;
692
693 return 0;
694 }
695 early_param("l1d_flush", l1d_flush_parse_cmdline);
696
697 #undef pr_fmt
698 #define pr_fmt(fmt) "Spectre V1 : " fmt
699
700 enum spectre_v1_mitigation {
701 SPECTRE_V1_MITIGATION_NONE,
702 SPECTRE_V1_MITIGATION_AUTO,
703 };
704
705 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
706 SPECTRE_V1_MITIGATION_AUTO;
707
708 static const char * const spectre_v1_strings[] = {
709 [SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
710 [SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
711 };
712
713 /*
714 * Does SMAP provide full mitigation against speculative kernel access to
715 * userspace?
716 */
smap_works_speculatively(void)717 static bool smap_works_speculatively(void)
718 {
719 if (!boot_cpu_has(X86_FEATURE_SMAP))
720 return false;
721
722 /*
723 * On CPUs which are vulnerable to Meltdown, SMAP does not
724 * prevent speculative access to user data in the L1 cache.
725 * Consider SMAP to be non-functional as a mitigation on these
726 * CPUs.
727 */
728 if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
729 return false;
730
731 return true;
732 }
733
spectre_v1_select_mitigation(void)734 static void __init spectre_v1_select_mitigation(void)
735 {
736 if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
737 spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
738 return;
739 }
740
741 if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
742 /*
743 * With Spectre v1, a user can speculatively control either
744 * path of a conditional swapgs with a user-controlled GS
745 * value. The mitigation is to add lfences to both code paths.
746 *
747 * If FSGSBASE is enabled, the user can put a kernel address in
748 * GS, in which case SMAP provides no protection.
749 *
750 * If FSGSBASE is disabled, the user can only put a user space
751 * address in GS. That makes an attack harder, but still
752 * possible if there's no SMAP protection.
753 */
754 if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
755 !smap_works_speculatively()) {
756 /*
757 * Mitigation can be provided from SWAPGS itself or
758 * PTI as the CR3 write in the Meltdown mitigation
759 * is serializing.
760 *
761 * If neither is there, mitigate with an LFENCE to
762 * stop speculation through swapgs.
763 */
764 if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
765 !boot_cpu_has(X86_FEATURE_PTI))
766 setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
767
768 /*
769 * Enable lfences in the kernel entry (non-swapgs)
770 * paths, to prevent user entry from speculatively
771 * skipping swapgs.
772 */
773 setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
774 }
775 }
776
777 pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
778 }
779
nospectre_v1_cmdline(char * str)780 static int __init nospectre_v1_cmdline(char *str)
781 {
782 spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
783 return 0;
784 }
785 early_param("nospectre_v1", nospectre_v1_cmdline);
786
787 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
788 SPECTRE_V2_NONE;
789
790 #undef pr_fmt
791 #define pr_fmt(fmt) "RETBleed: " fmt
792
793 enum retbleed_mitigation {
794 RETBLEED_MITIGATION_NONE,
795 RETBLEED_MITIGATION_UNRET,
796 RETBLEED_MITIGATION_IBPB,
797 RETBLEED_MITIGATION_IBRS,
798 RETBLEED_MITIGATION_EIBRS,
799 RETBLEED_MITIGATION_STUFF,
800 };
801
802 enum retbleed_mitigation_cmd {
803 RETBLEED_CMD_OFF,
804 RETBLEED_CMD_AUTO,
805 RETBLEED_CMD_UNRET,
806 RETBLEED_CMD_IBPB,
807 RETBLEED_CMD_STUFF,
808 };
809
810 static const char * const retbleed_strings[] = {
811 [RETBLEED_MITIGATION_NONE] = "Vulnerable",
812 [RETBLEED_MITIGATION_UNRET] = "Mitigation: untrained return thunk",
813 [RETBLEED_MITIGATION_IBPB] = "Mitigation: IBPB",
814 [RETBLEED_MITIGATION_IBRS] = "Mitigation: IBRS",
815 [RETBLEED_MITIGATION_EIBRS] = "Mitigation: Enhanced IBRS",
816 [RETBLEED_MITIGATION_STUFF] = "Mitigation: Stuffing",
817 };
818
819 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
820 RETBLEED_MITIGATION_NONE;
821 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
822 RETBLEED_CMD_AUTO;
823
824 static int __ro_after_init retbleed_nosmt = false;
825
retbleed_parse_cmdline(char * str)826 static int __init retbleed_parse_cmdline(char *str)
827 {
828 if (!str)
829 return -EINVAL;
830
831 while (str) {
832 char *next = strchr(str, ',');
833 if (next) {
834 *next = 0;
835 next++;
836 }
837
838 if (!strcmp(str, "off")) {
839 retbleed_cmd = RETBLEED_CMD_OFF;
840 } else if (!strcmp(str, "auto")) {
841 retbleed_cmd = RETBLEED_CMD_AUTO;
842 } else if (!strcmp(str, "unret")) {
843 retbleed_cmd = RETBLEED_CMD_UNRET;
844 } else if (!strcmp(str, "ibpb")) {
845 retbleed_cmd = RETBLEED_CMD_IBPB;
846 } else if (!strcmp(str, "stuff")) {
847 retbleed_cmd = RETBLEED_CMD_STUFF;
848 } else if (!strcmp(str, "nosmt")) {
849 retbleed_nosmt = true;
850 } else if (!strcmp(str, "force")) {
851 setup_force_cpu_bug(X86_BUG_RETBLEED);
852 } else {
853 pr_err("Ignoring unknown retbleed option (%s).", str);
854 }
855
856 str = next;
857 }
858
859 return 0;
860 }
861 early_param("retbleed", retbleed_parse_cmdline);
862
863 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
864 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
865
retbleed_select_mitigation(void)866 static void __init retbleed_select_mitigation(void)
867 {
868 bool mitigate_smt = false;
869
870 if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
871 return;
872
873 switch (retbleed_cmd) {
874 case RETBLEED_CMD_OFF:
875 return;
876
877 case RETBLEED_CMD_UNRET:
878 if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
879 retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
880 } else {
881 pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
882 goto do_cmd_auto;
883 }
884 break;
885
886 case RETBLEED_CMD_IBPB:
887 if (!boot_cpu_has(X86_FEATURE_IBPB)) {
888 pr_err("WARNING: CPU does not support IBPB.\n");
889 goto do_cmd_auto;
890 } else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
891 retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
892 } else {
893 pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
894 goto do_cmd_auto;
895 }
896 break;
897
898 case RETBLEED_CMD_STUFF:
899 if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING) &&
900 spectre_v2_enabled == SPECTRE_V2_RETPOLINE) {
901 retbleed_mitigation = RETBLEED_MITIGATION_STUFF;
902
903 } else {
904 if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING))
905 pr_err("WARNING: retbleed=stuff depends on spectre_v2=retpoline\n");
906 else
907 pr_err("WARNING: kernel not compiled with CALL_DEPTH_TRACKING.\n");
908
909 goto do_cmd_auto;
910 }
911 break;
912
913 do_cmd_auto:
914 case RETBLEED_CMD_AUTO:
915 default:
916 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
917 boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
918 if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
919 retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
920 else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
921 retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
922 }
923
924 /*
925 * The Intel mitigation (IBRS or eIBRS) was already selected in
926 * spectre_v2_select_mitigation(). 'retbleed_mitigation' will
927 * be set accordingly below.
928 */
929
930 break;
931 }
932
933 switch (retbleed_mitigation) {
934 case RETBLEED_MITIGATION_UNRET:
935 setup_force_cpu_cap(X86_FEATURE_RETHUNK);
936 setup_force_cpu_cap(X86_FEATURE_UNRET);
937
938 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
939 boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
940 pr_err(RETBLEED_UNTRAIN_MSG);
941
942 mitigate_smt = true;
943 break;
944
945 case RETBLEED_MITIGATION_IBPB:
946 setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
947 mitigate_smt = true;
948 break;
949
950 case RETBLEED_MITIGATION_STUFF:
951 setup_force_cpu_cap(X86_FEATURE_RETHUNK);
952 setup_force_cpu_cap(X86_FEATURE_CALL_DEPTH);
953 x86_set_skl_return_thunk();
954 break;
955
956 default:
957 break;
958 }
959
960 if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
961 (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
962 cpu_smt_disable(false);
963
964 /*
965 * Let IBRS trump all on Intel without affecting the effects of the
966 * retbleed= cmdline option except for call depth based stuffing
967 */
968 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
969 switch (spectre_v2_enabled) {
970 case SPECTRE_V2_IBRS:
971 retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
972 break;
973 case SPECTRE_V2_EIBRS:
974 case SPECTRE_V2_EIBRS_RETPOLINE:
975 case SPECTRE_V2_EIBRS_LFENCE:
976 retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
977 break;
978 default:
979 if (retbleed_mitigation != RETBLEED_MITIGATION_STUFF)
980 pr_err(RETBLEED_INTEL_MSG);
981 }
982 }
983
984 pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
985 }
986
987 #undef pr_fmt
988 #define pr_fmt(fmt) "Spectre V2 : " fmt
989
990 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
991 SPECTRE_V2_USER_NONE;
992 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
993 SPECTRE_V2_USER_NONE;
994
995 #ifdef CONFIG_RETPOLINE
996 static bool spectre_v2_bad_module;
997
retpoline_module_ok(bool has_retpoline)998 bool retpoline_module_ok(bool has_retpoline)
999 {
1000 if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
1001 return true;
1002
1003 pr_err("System may be vulnerable to spectre v2\n");
1004 spectre_v2_bad_module = true;
1005 return false;
1006 }
1007
spectre_v2_module_string(void)1008 static inline const char *spectre_v2_module_string(void)
1009 {
1010 return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
1011 }
1012 #else
spectre_v2_module_string(void)1013 static inline const char *spectre_v2_module_string(void) { return ""; }
1014 #endif
1015
1016 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
1017 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
1018 #define SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS+LFENCE mitigation and SMT, data leaks possible via Spectre v2 BHB attacks!\n"
1019 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
1020
1021 #ifdef CONFIG_BPF_SYSCALL
unpriv_ebpf_notify(int new_state)1022 void unpriv_ebpf_notify(int new_state)
1023 {
1024 if (new_state)
1025 return;
1026
1027 /* Unprivileged eBPF is enabled */
1028
1029 switch (spectre_v2_enabled) {
1030 case SPECTRE_V2_EIBRS:
1031 pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1032 break;
1033 case SPECTRE_V2_EIBRS_LFENCE:
1034 if (sched_smt_active())
1035 pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1036 break;
1037 default:
1038 break;
1039 }
1040 }
1041 #endif
1042
match_option(const char * arg,int arglen,const char * opt)1043 static inline bool match_option(const char *arg, int arglen, const char *opt)
1044 {
1045 int len = strlen(opt);
1046
1047 return len == arglen && !strncmp(arg, opt, len);
1048 }
1049
1050 /* The kernel command line selection for spectre v2 */
1051 enum spectre_v2_mitigation_cmd {
1052 SPECTRE_V2_CMD_NONE,
1053 SPECTRE_V2_CMD_AUTO,
1054 SPECTRE_V2_CMD_FORCE,
1055 SPECTRE_V2_CMD_RETPOLINE,
1056 SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1057 SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1058 SPECTRE_V2_CMD_EIBRS,
1059 SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1060 SPECTRE_V2_CMD_EIBRS_LFENCE,
1061 SPECTRE_V2_CMD_IBRS,
1062 };
1063
1064 enum spectre_v2_user_cmd {
1065 SPECTRE_V2_USER_CMD_NONE,
1066 SPECTRE_V2_USER_CMD_AUTO,
1067 SPECTRE_V2_USER_CMD_FORCE,
1068 SPECTRE_V2_USER_CMD_PRCTL,
1069 SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1070 SPECTRE_V2_USER_CMD_SECCOMP,
1071 SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1072 };
1073
1074 static const char * const spectre_v2_user_strings[] = {
1075 [SPECTRE_V2_USER_NONE] = "User space: Vulnerable",
1076 [SPECTRE_V2_USER_STRICT] = "User space: Mitigation: STIBP protection",
1077 [SPECTRE_V2_USER_STRICT_PREFERRED] = "User space: Mitigation: STIBP always-on protection",
1078 [SPECTRE_V2_USER_PRCTL] = "User space: Mitigation: STIBP via prctl",
1079 [SPECTRE_V2_USER_SECCOMP] = "User space: Mitigation: STIBP via seccomp and prctl",
1080 };
1081
1082 static const struct {
1083 const char *option;
1084 enum spectre_v2_user_cmd cmd;
1085 bool secure;
1086 } v2_user_options[] __initconst = {
1087 { "auto", SPECTRE_V2_USER_CMD_AUTO, false },
1088 { "off", SPECTRE_V2_USER_CMD_NONE, false },
1089 { "on", SPECTRE_V2_USER_CMD_FORCE, true },
1090 { "prctl", SPECTRE_V2_USER_CMD_PRCTL, false },
1091 { "prctl,ibpb", SPECTRE_V2_USER_CMD_PRCTL_IBPB, false },
1092 { "seccomp", SPECTRE_V2_USER_CMD_SECCOMP, false },
1093 { "seccomp,ibpb", SPECTRE_V2_USER_CMD_SECCOMP_IBPB, false },
1094 };
1095
spec_v2_user_print_cond(const char * reason,bool secure)1096 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1097 {
1098 if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1099 pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1100 }
1101
1102 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1103
1104 static enum spectre_v2_user_cmd __init
spectre_v2_parse_user_cmdline(void)1105 spectre_v2_parse_user_cmdline(void)
1106 {
1107 char arg[20];
1108 int ret, i;
1109
1110 switch (spectre_v2_cmd) {
1111 case SPECTRE_V2_CMD_NONE:
1112 return SPECTRE_V2_USER_CMD_NONE;
1113 case SPECTRE_V2_CMD_FORCE:
1114 return SPECTRE_V2_USER_CMD_FORCE;
1115 default:
1116 break;
1117 }
1118
1119 ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1120 arg, sizeof(arg));
1121 if (ret < 0)
1122 return SPECTRE_V2_USER_CMD_AUTO;
1123
1124 for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1125 if (match_option(arg, ret, v2_user_options[i].option)) {
1126 spec_v2_user_print_cond(v2_user_options[i].option,
1127 v2_user_options[i].secure);
1128 return v2_user_options[i].cmd;
1129 }
1130 }
1131
1132 pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1133 return SPECTRE_V2_USER_CMD_AUTO;
1134 }
1135
spectre_v2_in_eibrs_mode(enum spectre_v2_mitigation mode)1136 static inline bool spectre_v2_in_eibrs_mode(enum spectre_v2_mitigation mode)
1137 {
1138 return mode == SPECTRE_V2_EIBRS ||
1139 mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1140 mode == SPECTRE_V2_EIBRS_LFENCE;
1141 }
1142
spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)1143 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1144 {
1145 return spectre_v2_in_eibrs_mode(mode) || mode == SPECTRE_V2_IBRS;
1146 }
1147
1148 static void __init
spectre_v2_user_select_mitigation(void)1149 spectre_v2_user_select_mitigation(void)
1150 {
1151 enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1152 bool smt_possible = IS_ENABLED(CONFIG_SMP);
1153 enum spectre_v2_user_cmd cmd;
1154
1155 if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1156 return;
1157
1158 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1159 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1160 smt_possible = false;
1161
1162 cmd = spectre_v2_parse_user_cmdline();
1163 switch (cmd) {
1164 case SPECTRE_V2_USER_CMD_NONE:
1165 goto set_mode;
1166 case SPECTRE_V2_USER_CMD_FORCE:
1167 mode = SPECTRE_V2_USER_STRICT;
1168 break;
1169 case SPECTRE_V2_USER_CMD_AUTO:
1170 case SPECTRE_V2_USER_CMD_PRCTL:
1171 case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1172 mode = SPECTRE_V2_USER_PRCTL;
1173 break;
1174 case SPECTRE_V2_USER_CMD_SECCOMP:
1175 case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1176 if (IS_ENABLED(CONFIG_SECCOMP))
1177 mode = SPECTRE_V2_USER_SECCOMP;
1178 else
1179 mode = SPECTRE_V2_USER_PRCTL;
1180 break;
1181 }
1182
1183 /* Initialize Indirect Branch Prediction Barrier */
1184 if (boot_cpu_has(X86_FEATURE_IBPB)) {
1185 setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1186
1187 spectre_v2_user_ibpb = mode;
1188 switch (cmd) {
1189 case SPECTRE_V2_USER_CMD_FORCE:
1190 case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1191 case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1192 static_branch_enable(&switch_mm_always_ibpb);
1193 spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1194 break;
1195 case SPECTRE_V2_USER_CMD_PRCTL:
1196 case SPECTRE_V2_USER_CMD_AUTO:
1197 case SPECTRE_V2_USER_CMD_SECCOMP:
1198 static_branch_enable(&switch_mm_cond_ibpb);
1199 break;
1200 default:
1201 break;
1202 }
1203
1204 pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1205 static_key_enabled(&switch_mm_always_ibpb) ?
1206 "always-on" : "conditional");
1207 }
1208
1209 /*
1210 * If no STIBP, enhanced IBRS is enabled, or SMT impossible, STIBP
1211 * is not required.
1212 *
1213 * Enhanced IBRS also protects against cross-thread branch target
1214 * injection in user-mode as the IBRS bit remains always set which
1215 * implicitly enables cross-thread protections. However, in legacy IBRS
1216 * mode, the IBRS bit is set only on kernel entry and cleared on return
1217 * to userspace. This disables the implicit cross-thread protection,
1218 * so allow for STIBP to be selected in that case.
1219 */
1220 if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1221 !smt_possible ||
1222 spectre_v2_in_eibrs_mode(spectre_v2_enabled))
1223 return;
1224
1225 /*
1226 * At this point, an STIBP mode other than "off" has been set.
1227 * If STIBP support is not being forced, check if STIBP always-on
1228 * is preferred.
1229 */
1230 if (mode != SPECTRE_V2_USER_STRICT &&
1231 boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1232 mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1233
1234 if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1235 retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1236 if (mode != SPECTRE_V2_USER_STRICT &&
1237 mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1238 pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1239 mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1240 }
1241
1242 spectre_v2_user_stibp = mode;
1243
1244 set_mode:
1245 pr_info("%s\n", spectre_v2_user_strings[mode]);
1246 }
1247
1248 static const char * const spectre_v2_strings[] = {
1249 [SPECTRE_V2_NONE] = "Vulnerable",
1250 [SPECTRE_V2_RETPOLINE] = "Mitigation: Retpolines",
1251 [SPECTRE_V2_LFENCE] = "Mitigation: LFENCE",
1252 [SPECTRE_V2_EIBRS] = "Mitigation: Enhanced / Automatic IBRS",
1253 [SPECTRE_V2_EIBRS_LFENCE] = "Mitigation: Enhanced / Automatic IBRS + LFENCE",
1254 [SPECTRE_V2_EIBRS_RETPOLINE] = "Mitigation: Enhanced / Automatic IBRS + Retpolines",
1255 [SPECTRE_V2_IBRS] = "Mitigation: IBRS",
1256 };
1257
1258 static const struct {
1259 const char *option;
1260 enum spectre_v2_mitigation_cmd cmd;
1261 bool secure;
1262 } mitigation_options[] __initconst = {
1263 { "off", SPECTRE_V2_CMD_NONE, false },
1264 { "on", SPECTRE_V2_CMD_FORCE, true },
1265 { "retpoline", SPECTRE_V2_CMD_RETPOLINE, false },
1266 { "retpoline,amd", SPECTRE_V2_CMD_RETPOLINE_LFENCE, false },
1267 { "retpoline,lfence", SPECTRE_V2_CMD_RETPOLINE_LFENCE, false },
1268 { "retpoline,generic", SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1269 { "eibrs", SPECTRE_V2_CMD_EIBRS, false },
1270 { "eibrs,lfence", SPECTRE_V2_CMD_EIBRS_LFENCE, false },
1271 { "eibrs,retpoline", SPECTRE_V2_CMD_EIBRS_RETPOLINE, false },
1272 { "auto", SPECTRE_V2_CMD_AUTO, false },
1273 { "ibrs", SPECTRE_V2_CMD_IBRS, false },
1274 };
1275
spec_v2_print_cond(const char * reason,bool secure)1276 static void __init spec_v2_print_cond(const char *reason, bool secure)
1277 {
1278 if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1279 pr_info("%s selected on command line.\n", reason);
1280 }
1281
spectre_v2_parse_cmdline(void)1282 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1283 {
1284 enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1285 char arg[20];
1286 int ret, i;
1287
1288 if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1289 cpu_mitigations_off())
1290 return SPECTRE_V2_CMD_NONE;
1291
1292 ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1293 if (ret < 0)
1294 return SPECTRE_V2_CMD_AUTO;
1295
1296 for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1297 if (!match_option(arg, ret, mitigation_options[i].option))
1298 continue;
1299 cmd = mitigation_options[i].cmd;
1300 break;
1301 }
1302
1303 if (i >= ARRAY_SIZE(mitigation_options)) {
1304 pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1305 return SPECTRE_V2_CMD_AUTO;
1306 }
1307
1308 if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1309 cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1310 cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1311 cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1312 cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1313 !IS_ENABLED(CONFIG_RETPOLINE)) {
1314 pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1315 mitigation_options[i].option);
1316 return SPECTRE_V2_CMD_AUTO;
1317 }
1318
1319 if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1320 cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1321 cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1322 !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1323 pr_err("%s selected but CPU doesn't have Enhanced or Automatic IBRS. Switching to AUTO select\n",
1324 mitigation_options[i].option);
1325 return SPECTRE_V2_CMD_AUTO;
1326 }
1327
1328 if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1329 cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1330 !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1331 pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1332 mitigation_options[i].option);
1333 return SPECTRE_V2_CMD_AUTO;
1334 }
1335
1336 if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1337 pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1338 mitigation_options[i].option);
1339 return SPECTRE_V2_CMD_AUTO;
1340 }
1341
1342 if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1343 pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1344 mitigation_options[i].option);
1345 return SPECTRE_V2_CMD_AUTO;
1346 }
1347
1348 if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1349 pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1350 mitigation_options[i].option);
1351 return SPECTRE_V2_CMD_AUTO;
1352 }
1353
1354 if (cmd == SPECTRE_V2_CMD_IBRS && cpu_feature_enabled(X86_FEATURE_XENPV)) {
1355 pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1356 mitigation_options[i].option);
1357 return SPECTRE_V2_CMD_AUTO;
1358 }
1359
1360 spec_v2_print_cond(mitigation_options[i].option,
1361 mitigation_options[i].secure);
1362 return cmd;
1363 }
1364
spectre_v2_select_retpoline(void)1365 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1366 {
1367 if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1368 pr_err("Kernel not compiled with retpoline; no mitigation available!");
1369 return SPECTRE_V2_NONE;
1370 }
1371
1372 return SPECTRE_V2_RETPOLINE;
1373 }
1374
1375 /* Disable in-kernel use of non-RSB RET predictors */
spec_ctrl_disable_kernel_rrsba(void)1376 static void __init spec_ctrl_disable_kernel_rrsba(void)
1377 {
1378 u64 ia32_cap;
1379
1380 if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1381 return;
1382
1383 ia32_cap = x86_read_arch_cap_msr();
1384
1385 if (ia32_cap & ARCH_CAP_RRSBA) {
1386 x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1387 update_spec_ctrl(x86_spec_ctrl_base);
1388 }
1389 }
1390
spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)1391 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1392 {
1393 /*
1394 * Similar to context switches, there are two types of RSB attacks
1395 * after VM exit:
1396 *
1397 * 1) RSB underflow
1398 *
1399 * 2) Poisoned RSB entry
1400 *
1401 * When retpoline is enabled, both are mitigated by filling/clearing
1402 * the RSB.
1403 *
1404 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1405 * prediction isolation protections, RSB still needs to be cleared
1406 * because of #2. Note that SMEP provides no protection here, unlike
1407 * user-space-poisoned RSB entries.
1408 *
1409 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1410 * bug is present then a LITE version of RSB protection is required,
1411 * just a single call needs to retire before a RET is executed.
1412 */
1413 switch (mode) {
1414 case SPECTRE_V2_NONE:
1415 return;
1416
1417 case SPECTRE_V2_EIBRS_LFENCE:
1418 case SPECTRE_V2_EIBRS:
1419 if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1420 setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1421 pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1422 }
1423 return;
1424
1425 case SPECTRE_V2_EIBRS_RETPOLINE:
1426 case SPECTRE_V2_RETPOLINE:
1427 case SPECTRE_V2_LFENCE:
1428 case SPECTRE_V2_IBRS:
1429 setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1430 pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1431 return;
1432 }
1433
1434 pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1435 dump_stack();
1436 }
1437
spectre_v2_select_mitigation(void)1438 static void __init spectre_v2_select_mitigation(void)
1439 {
1440 enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1441 enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1442
1443 /*
1444 * If the CPU is not affected and the command line mode is NONE or AUTO
1445 * then nothing to do.
1446 */
1447 if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1448 (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1449 return;
1450
1451 switch (cmd) {
1452 case SPECTRE_V2_CMD_NONE:
1453 return;
1454
1455 case SPECTRE_V2_CMD_FORCE:
1456 case SPECTRE_V2_CMD_AUTO:
1457 if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1458 mode = SPECTRE_V2_EIBRS;
1459 break;
1460 }
1461
1462 if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1463 boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1464 retbleed_cmd != RETBLEED_CMD_OFF &&
1465 retbleed_cmd != RETBLEED_CMD_STUFF &&
1466 boot_cpu_has(X86_FEATURE_IBRS) &&
1467 boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1468 mode = SPECTRE_V2_IBRS;
1469 break;
1470 }
1471
1472 mode = spectre_v2_select_retpoline();
1473 break;
1474
1475 case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1476 pr_err(SPECTRE_V2_LFENCE_MSG);
1477 mode = SPECTRE_V2_LFENCE;
1478 break;
1479
1480 case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1481 mode = SPECTRE_V2_RETPOLINE;
1482 break;
1483
1484 case SPECTRE_V2_CMD_RETPOLINE:
1485 mode = spectre_v2_select_retpoline();
1486 break;
1487
1488 case SPECTRE_V2_CMD_IBRS:
1489 mode = SPECTRE_V2_IBRS;
1490 break;
1491
1492 case SPECTRE_V2_CMD_EIBRS:
1493 mode = SPECTRE_V2_EIBRS;
1494 break;
1495
1496 case SPECTRE_V2_CMD_EIBRS_LFENCE:
1497 mode = SPECTRE_V2_EIBRS_LFENCE;
1498 break;
1499
1500 case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1501 mode = SPECTRE_V2_EIBRS_RETPOLINE;
1502 break;
1503 }
1504
1505 if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1506 pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1507
1508 if (spectre_v2_in_ibrs_mode(mode)) {
1509 if (boot_cpu_has(X86_FEATURE_AUTOIBRS)) {
1510 msr_set_bit(MSR_EFER, _EFER_AUTOIBRS);
1511 } else {
1512 x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1513 update_spec_ctrl(x86_spec_ctrl_base);
1514 }
1515 }
1516
1517 switch (mode) {
1518 case SPECTRE_V2_NONE:
1519 case SPECTRE_V2_EIBRS:
1520 break;
1521
1522 case SPECTRE_V2_IBRS:
1523 setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1524 if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1525 pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1526 break;
1527
1528 case SPECTRE_V2_LFENCE:
1529 case SPECTRE_V2_EIBRS_LFENCE:
1530 setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1531 fallthrough;
1532
1533 case SPECTRE_V2_RETPOLINE:
1534 case SPECTRE_V2_EIBRS_RETPOLINE:
1535 setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1536 break;
1537 }
1538
1539 /*
1540 * Disable alternate RSB predictions in kernel when indirect CALLs and
1541 * JMPs gets protection against BHI and Intramode-BTI, but RET
1542 * prediction from a non-RSB predictor is still a risk.
1543 */
1544 if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1545 mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1546 mode == SPECTRE_V2_RETPOLINE)
1547 spec_ctrl_disable_kernel_rrsba();
1548
1549 spectre_v2_enabled = mode;
1550 pr_info("%s\n", spectre_v2_strings[mode]);
1551
1552 /*
1553 * If Spectre v2 protection has been enabled, fill the RSB during a
1554 * context switch. In general there are two types of RSB attacks
1555 * across context switches, for which the CALLs/RETs may be unbalanced.
1556 *
1557 * 1) RSB underflow
1558 *
1559 * Some Intel parts have "bottomless RSB". When the RSB is empty,
1560 * speculated return targets may come from the branch predictor,
1561 * which could have a user-poisoned BTB or BHB entry.
1562 *
1563 * AMD has it even worse: *all* returns are speculated from the BTB,
1564 * regardless of the state of the RSB.
1565 *
1566 * When IBRS or eIBRS is enabled, the "user -> kernel" attack
1567 * scenario is mitigated by the IBRS branch prediction isolation
1568 * properties, so the RSB buffer filling wouldn't be necessary to
1569 * protect against this type of attack.
1570 *
1571 * The "user -> user" attack scenario is mitigated by RSB filling.
1572 *
1573 * 2) Poisoned RSB entry
1574 *
1575 * If the 'next' in-kernel return stack is shorter than 'prev',
1576 * 'next' could be tricked into speculating with a user-poisoned RSB
1577 * entry.
1578 *
1579 * The "user -> kernel" attack scenario is mitigated by SMEP and
1580 * eIBRS.
1581 *
1582 * The "user -> user" scenario, also known as SpectreBHB, requires
1583 * RSB clearing.
1584 *
1585 * So to mitigate all cases, unconditionally fill RSB on context
1586 * switches.
1587 *
1588 * FIXME: Is this pointless for retbleed-affected AMD?
1589 */
1590 setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1591 pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1592
1593 spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1594
1595 /*
1596 * Retpoline protects the kernel, but doesn't protect firmware. IBRS
1597 * and Enhanced IBRS protect firmware too, so enable IBRS around
1598 * firmware calls only when IBRS / Enhanced / Automatic IBRS aren't
1599 * otherwise enabled.
1600 *
1601 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1602 * the user might select retpoline on the kernel command line and if
1603 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1604 * enable IBRS around firmware calls.
1605 */
1606 if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1607 boot_cpu_has(X86_FEATURE_IBPB) &&
1608 (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1609 boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1610
1611 if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1612 setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1613 pr_info("Enabling Speculation Barrier for firmware calls\n");
1614 }
1615
1616 } else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1617 setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1618 pr_info("Enabling Restricted Speculation for firmware calls\n");
1619 }
1620
1621 /* Set up IBPB and STIBP depending on the general spectre V2 command */
1622 spectre_v2_cmd = cmd;
1623 }
1624
update_stibp_msr(void * __unused)1625 static void update_stibp_msr(void * __unused)
1626 {
1627 u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1628 update_spec_ctrl(val);
1629 }
1630
1631 /* Update x86_spec_ctrl_base in case SMT state changed. */
update_stibp_strict(void)1632 static void update_stibp_strict(void)
1633 {
1634 u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1635
1636 if (sched_smt_active())
1637 mask |= SPEC_CTRL_STIBP;
1638
1639 if (mask == x86_spec_ctrl_base)
1640 return;
1641
1642 pr_info("Update user space SMT mitigation: STIBP %s\n",
1643 mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1644 x86_spec_ctrl_base = mask;
1645 on_each_cpu(update_stibp_msr, NULL, 1);
1646 }
1647
1648 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
update_indir_branch_cond(void)1649 static void update_indir_branch_cond(void)
1650 {
1651 if (sched_smt_active())
1652 static_branch_enable(&switch_to_cond_stibp);
1653 else
1654 static_branch_disable(&switch_to_cond_stibp);
1655 }
1656
1657 #undef pr_fmt
1658 #define pr_fmt(fmt) fmt
1659
1660 /* Update the static key controlling the MDS CPU buffer clear in idle */
update_mds_branch_idle(void)1661 static void update_mds_branch_idle(void)
1662 {
1663 u64 ia32_cap = x86_read_arch_cap_msr();
1664
1665 /*
1666 * Enable the idle clearing if SMT is active on CPUs which are
1667 * affected only by MSBDS and not any other MDS variant.
1668 *
1669 * The other variants cannot be mitigated when SMT is enabled, so
1670 * clearing the buffers on idle just to prevent the Store Buffer
1671 * repartitioning leak would be a window dressing exercise.
1672 */
1673 if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1674 return;
1675
1676 if (sched_smt_active()) {
1677 static_branch_enable(&mds_idle_clear);
1678 } else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1679 (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1680 static_branch_disable(&mds_idle_clear);
1681 }
1682 }
1683
1684 #define MDS_MSG_SMT "MDS CPU bug present and SMT on, data leak possible. See https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/mds.html for more details.\n"
1685 #define TAA_MSG_SMT "TAA CPU bug present and SMT on, data leak possible. See https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/tsx_async_abort.html for more details.\n"
1686 #define MMIO_MSG_SMT "MMIO Stale Data CPU bug present and SMT on, data leak possible. See https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/processor_mmio_stale_data.html for more details.\n"
1687
cpu_bugs_smt_update(void)1688 void cpu_bugs_smt_update(void)
1689 {
1690 mutex_lock(&spec_ctrl_mutex);
1691
1692 if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1693 spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1694 pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1695
1696 switch (spectre_v2_user_stibp) {
1697 case SPECTRE_V2_USER_NONE:
1698 break;
1699 case SPECTRE_V2_USER_STRICT:
1700 case SPECTRE_V2_USER_STRICT_PREFERRED:
1701 update_stibp_strict();
1702 break;
1703 case SPECTRE_V2_USER_PRCTL:
1704 case SPECTRE_V2_USER_SECCOMP:
1705 update_indir_branch_cond();
1706 break;
1707 }
1708
1709 switch (mds_mitigation) {
1710 case MDS_MITIGATION_FULL:
1711 case MDS_MITIGATION_VMWERV:
1712 if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1713 pr_warn_once(MDS_MSG_SMT);
1714 update_mds_branch_idle();
1715 break;
1716 case MDS_MITIGATION_OFF:
1717 break;
1718 }
1719
1720 switch (taa_mitigation) {
1721 case TAA_MITIGATION_VERW:
1722 case TAA_MITIGATION_UCODE_NEEDED:
1723 if (sched_smt_active())
1724 pr_warn_once(TAA_MSG_SMT);
1725 break;
1726 case TAA_MITIGATION_TSX_DISABLED:
1727 case TAA_MITIGATION_OFF:
1728 break;
1729 }
1730
1731 switch (mmio_mitigation) {
1732 case MMIO_MITIGATION_VERW:
1733 case MMIO_MITIGATION_UCODE_NEEDED:
1734 if (sched_smt_active())
1735 pr_warn_once(MMIO_MSG_SMT);
1736 break;
1737 case MMIO_MITIGATION_OFF:
1738 break;
1739 }
1740
1741 mutex_unlock(&spec_ctrl_mutex);
1742 }
1743
1744 #undef pr_fmt
1745 #define pr_fmt(fmt) "Speculative Store Bypass: " fmt
1746
1747 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1748
1749 /* The kernel command line selection */
1750 enum ssb_mitigation_cmd {
1751 SPEC_STORE_BYPASS_CMD_NONE,
1752 SPEC_STORE_BYPASS_CMD_AUTO,
1753 SPEC_STORE_BYPASS_CMD_ON,
1754 SPEC_STORE_BYPASS_CMD_PRCTL,
1755 SPEC_STORE_BYPASS_CMD_SECCOMP,
1756 };
1757
1758 static const char * const ssb_strings[] = {
1759 [SPEC_STORE_BYPASS_NONE] = "Vulnerable",
1760 [SPEC_STORE_BYPASS_DISABLE] = "Mitigation: Speculative Store Bypass disabled",
1761 [SPEC_STORE_BYPASS_PRCTL] = "Mitigation: Speculative Store Bypass disabled via prctl",
1762 [SPEC_STORE_BYPASS_SECCOMP] = "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1763 };
1764
1765 static const struct {
1766 const char *option;
1767 enum ssb_mitigation_cmd cmd;
1768 } ssb_mitigation_options[] __initconst = {
1769 { "auto", SPEC_STORE_BYPASS_CMD_AUTO }, /* Platform decides */
1770 { "on", SPEC_STORE_BYPASS_CMD_ON }, /* Disable Speculative Store Bypass */
1771 { "off", SPEC_STORE_BYPASS_CMD_NONE }, /* Don't touch Speculative Store Bypass */
1772 { "prctl", SPEC_STORE_BYPASS_CMD_PRCTL }, /* Disable Speculative Store Bypass via prctl */
1773 { "seccomp", SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1774 };
1775
ssb_parse_cmdline(void)1776 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1777 {
1778 enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1779 char arg[20];
1780 int ret, i;
1781
1782 if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1783 cpu_mitigations_off()) {
1784 return SPEC_STORE_BYPASS_CMD_NONE;
1785 } else {
1786 ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1787 arg, sizeof(arg));
1788 if (ret < 0)
1789 return SPEC_STORE_BYPASS_CMD_AUTO;
1790
1791 for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1792 if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1793 continue;
1794
1795 cmd = ssb_mitigation_options[i].cmd;
1796 break;
1797 }
1798
1799 if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1800 pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1801 return SPEC_STORE_BYPASS_CMD_AUTO;
1802 }
1803 }
1804
1805 return cmd;
1806 }
1807
__ssb_select_mitigation(void)1808 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1809 {
1810 enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1811 enum ssb_mitigation_cmd cmd;
1812
1813 if (!boot_cpu_has(X86_FEATURE_SSBD))
1814 return mode;
1815
1816 cmd = ssb_parse_cmdline();
1817 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1818 (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1819 cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1820 return mode;
1821
1822 switch (cmd) {
1823 case SPEC_STORE_BYPASS_CMD_SECCOMP:
1824 /*
1825 * Choose prctl+seccomp as the default mode if seccomp is
1826 * enabled.
1827 */
1828 if (IS_ENABLED(CONFIG_SECCOMP))
1829 mode = SPEC_STORE_BYPASS_SECCOMP;
1830 else
1831 mode = SPEC_STORE_BYPASS_PRCTL;
1832 break;
1833 case SPEC_STORE_BYPASS_CMD_ON:
1834 mode = SPEC_STORE_BYPASS_DISABLE;
1835 break;
1836 case SPEC_STORE_BYPASS_CMD_AUTO:
1837 case SPEC_STORE_BYPASS_CMD_PRCTL:
1838 mode = SPEC_STORE_BYPASS_PRCTL;
1839 break;
1840 case SPEC_STORE_BYPASS_CMD_NONE:
1841 break;
1842 }
1843
1844 /*
1845 * We have three CPU feature flags that are in play here:
1846 * - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1847 * - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1848 * - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1849 */
1850 if (mode == SPEC_STORE_BYPASS_DISABLE) {
1851 setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1852 /*
1853 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1854 * use a completely different MSR and bit dependent on family.
1855 */
1856 if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1857 !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1858 x86_amd_ssb_disable();
1859 } else {
1860 x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1861 update_spec_ctrl(x86_spec_ctrl_base);
1862 }
1863 }
1864
1865 return mode;
1866 }
1867
ssb_select_mitigation(void)1868 static void ssb_select_mitigation(void)
1869 {
1870 ssb_mode = __ssb_select_mitigation();
1871
1872 if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1873 pr_info("%s\n", ssb_strings[ssb_mode]);
1874 }
1875
1876 #undef pr_fmt
1877 #define pr_fmt(fmt) "Speculation prctl: " fmt
1878
task_update_spec_tif(struct task_struct * tsk)1879 static void task_update_spec_tif(struct task_struct *tsk)
1880 {
1881 /* Force the update of the real TIF bits */
1882 set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1883
1884 /*
1885 * Immediately update the speculation control MSRs for the current
1886 * task, but for a non-current task delay setting the CPU
1887 * mitigation until it is scheduled next.
1888 *
1889 * This can only happen for SECCOMP mitigation. For PRCTL it's
1890 * always the current task.
1891 */
1892 if (tsk == current)
1893 speculation_ctrl_update_current();
1894 }
1895
l1d_flush_prctl_set(struct task_struct * task,unsigned long ctrl)1896 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1897 {
1898
1899 if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1900 return -EPERM;
1901
1902 switch (ctrl) {
1903 case PR_SPEC_ENABLE:
1904 set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1905 return 0;
1906 case PR_SPEC_DISABLE:
1907 clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1908 return 0;
1909 default:
1910 return -ERANGE;
1911 }
1912 }
1913
ssb_prctl_set(struct task_struct * task,unsigned long ctrl)1914 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1915 {
1916 if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1917 ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1918 return -ENXIO;
1919
1920 switch (ctrl) {
1921 case PR_SPEC_ENABLE:
1922 /* If speculation is force disabled, enable is not allowed */
1923 if (task_spec_ssb_force_disable(task))
1924 return -EPERM;
1925 task_clear_spec_ssb_disable(task);
1926 task_clear_spec_ssb_noexec(task);
1927 task_update_spec_tif(task);
1928 break;
1929 case PR_SPEC_DISABLE:
1930 task_set_spec_ssb_disable(task);
1931 task_clear_spec_ssb_noexec(task);
1932 task_update_spec_tif(task);
1933 break;
1934 case PR_SPEC_FORCE_DISABLE:
1935 task_set_spec_ssb_disable(task);
1936 task_set_spec_ssb_force_disable(task);
1937 task_clear_spec_ssb_noexec(task);
1938 task_update_spec_tif(task);
1939 break;
1940 case PR_SPEC_DISABLE_NOEXEC:
1941 if (task_spec_ssb_force_disable(task))
1942 return -EPERM;
1943 task_set_spec_ssb_disable(task);
1944 task_set_spec_ssb_noexec(task);
1945 task_update_spec_tif(task);
1946 break;
1947 default:
1948 return -ERANGE;
1949 }
1950 return 0;
1951 }
1952
is_spec_ib_user_controlled(void)1953 static bool is_spec_ib_user_controlled(void)
1954 {
1955 return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1956 spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1957 spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1958 spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1959 }
1960
ib_prctl_set(struct task_struct * task,unsigned long ctrl)1961 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1962 {
1963 switch (ctrl) {
1964 case PR_SPEC_ENABLE:
1965 if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1966 spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1967 return 0;
1968
1969 /*
1970 * With strict mode for both IBPB and STIBP, the instruction
1971 * code paths avoid checking this task flag and instead,
1972 * unconditionally run the instruction. However, STIBP and IBPB
1973 * are independent and either can be set to conditionally
1974 * enabled regardless of the mode of the other.
1975 *
1976 * If either is set to conditional, allow the task flag to be
1977 * updated, unless it was force-disabled by a previous prctl
1978 * call. Currently, this is possible on an AMD CPU which has the
1979 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1980 * kernel is booted with 'spectre_v2_user=seccomp', then
1981 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1982 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1983 */
1984 if (!is_spec_ib_user_controlled() ||
1985 task_spec_ib_force_disable(task))
1986 return -EPERM;
1987
1988 task_clear_spec_ib_disable(task);
1989 task_update_spec_tif(task);
1990 break;
1991 case PR_SPEC_DISABLE:
1992 case PR_SPEC_FORCE_DISABLE:
1993 /*
1994 * Indirect branch speculation is always allowed when
1995 * mitigation is force disabled.
1996 */
1997 if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1998 spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1999 return -EPERM;
2000
2001 if (!is_spec_ib_user_controlled())
2002 return 0;
2003
2004 task_set_spec_ib_disable(task);
2005 if (ctrl == PR_SPEC_FORCE_DISABLE)
2006 task_set_spec_ib_force_disable(task);
2007 task_update_spec_tif(task);
2008 if (task == current)
2009 indirect_branch_prediction_barrier();
2010 break;
2011 default:
2012 return -ERANGE;
2013 }
2014 return 0;
2015 }
2016
arch_prctl_spec_ctrl_set(struct task_struct * task,unsigned long which,unsigned long ctrl)2017 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
2018 unsigned long ctrl)
2019 {
2020 switch (which) {
2021 case PR_SPEC_STORE_BYPASS:
2022 return ssb_prctl_set(task, ctrl);
2023 case PR_SPEC_INDIRECT_BRANCH:
2024 return ib_prctl_set(task, ctrl);
2025 case PR_SPEC_L1D_FLUSH:
2026 return l1d_flush_prctl_set(task, ctrl);
2027 default:
2028 return -ENODEV;
2029 }
2030 }
2031
2032 #ifdef CONFIG_SECCOMP
arch_seccomp_spec_mitigate(struct task_struct * task)2033 void arch_seccomp_spec_mitigate(struct task_struct *task)
2034 {
2035 if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
2036 ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2037 if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2038 spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
2039 ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2040 }
2041 #endif
2042
l1d_flush_prctl_get(struct task_struct * task)2043 static int l1d_flush_prctl_get(struct task_struct *task)
2044 {
2045 if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2046 return PR_SPEC_FORCE_DISABLE;
2047
2048 if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
2049 return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2050 else
2051 return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2052 }
2053
ssb_prctl_get(struct task_struct * task)2054 static int ssb_prctl_get(struct task_struct *task)
2055 {
2056 switch (ssb_mode) {
2057 case SPEC_STORE_BYPASS_DISABLE:
2058 return PR_SPEC_DISABLE;
2059 case SPEC_STORE_BYPASS_SECCOMP:
2060 case SPEC_STORE_BYPASS_PRCTL:
2061 if (task_spec_ssb_force_disable(task))
2062 return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2063 if (task_spec_ssb_noexec(task))
2064 return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2065 if (task_spec_ssb_disable(task))
2066 return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2067 return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2068 default:
2069 if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2070 return PR_SPEC_ENABLE;
2071 return PR_SPEC_NOT_AFFECTED;
2072 }
2073 }
2074
ib_prctl_get(struct task_struct * task)2075 static int ib_prctl_get(struct task_struct *task)
2076 {
2077 if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2078 return PR_SPEC_NOT_AFFECTED;
2079
2080 if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2081 spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2082 return PR_SPEC_ENABLE;
2083 else if (is_spec_ib_user_controlled()) {
2084 if (task_spec_ib_force_disable(task))
2085 return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2086 if (task_spec_ib_disable(task))
2087 return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2088 return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2089 } else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2090 spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2091 spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2092 return PR_SPEC_DISABLE;
2093 else
2094 return PR_SPEC_NOT_AFFECTED;
2095 }
2096
arch_prctl_spec_ctrl_get(struct task_struct * task,unsigned long which)2097 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2098 {
2099 switch (which) {
2100 case PR_SPEC_STORE_BYPASS:
2101 return ssb_prctl_get(task);
2102 case PR_SPEC_INDIRECT_BRANCH:
2103 return ib_prctl_get(task);
2104 case PR_SPEC_L1D_FLUSH:
2105 return l1d_flush_prctl_get(task);
2106 default:
2107 return -ENODEV;
2108 }
2109 }
2110
x86_spec_ctrl_setup_ap(void)2111 void x86_spec_ctrl_setup_ap(void)
2112 {
2113 if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2114 update_spec_ctrl(x86_spec_ctrl_base);
2115
2116 if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2117 x86_amd_ssb_disable();
2118 }
2119
2120 bool itlb_multihit_kvm_mitigation;
2121 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2122
2123 #undef pr_fmt
2124 #define pr_fmt(fmt) "L1TF: " fmt
2125
2126 /* Default mitigation for L1TF-affected CPUs */
2127 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2128 #if IS_ENABLED(CONFIG_KVM_INTEL)
2129 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2130 #endif
2131 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2132 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2133
2134 /*
2135 * These CPUs all support 44bits physical address space internally in the
2136 * cache but CPUID can report a smaller number of physical address bits.
2137 *
2138 * The L1TF mitigation uses the top most address bit for the inversion of
2139 * non present PTEs. When the installed memory reaches into the top most
2140 * address bit due to memory holes, which has been observed on machines
2141 * which report 36bits physical address bits and have 32G RAM installed,
2142 * then the mitigation range check in l1tf_select_mitigation() triggers.
2143 * This is a false positive because the mitigation is still possible due to
2144 * the fact that the cache uses 44bit internally. Use the cache bits
2145 * instead of the reported physical bits and adjust them on the affected
2146 * machines to 44bit if the reported bits are less than 44.
2147 */
override_cache_bits(struct cpuinfo_x86 * c)2148 static void override_cache_bits(struct cpuinfo_x86 *c)
2149 {
2150 if (c->x86 != 6)
2151 return;
2152
2153 switch (c->x86_model) {
2154 case INTEL_FAM6_NEHALEM:
2155 case INTEL_FAM6_WESTMERE:
2156 case INTEL_FAM6_SANDYBRIDGE:
2157 case INTEL_FAM6_IVYBRIDGE:
2158 case INTEL_FAM6_HASWELL:
2159 case INTEL_FAM6_HASWELL_L:
2160 case INTEL_FAM6_HASWELL_G:
2161 case INTEL_FAM6_BROADWELL:
2162 case INTEL_FAM6_BROADWELL_G:
2163 case INTEL_FAM6_SKYLAKE_L:
2164 case INTEL_FAM6_SKYLAKE:
2165 case INTEL_FAM6_KABYLAKE_L:
2166 case INTEL_FAM6_KABYLAKE:
2167 if (c->x86_cache_bits < 44)
2168 c->x86_cache_bits = 44;
2169 break;
2170 }
2171 }
2172
l1tf_select_mitigation(void)2173 static void __init l1tf_select_mitigation(void)
2174 {
2175 u64 half_pa;
2176
2177 if (!boot_cpu_has_bug(X86_BUG_L1TF))
2178 return;
2179
2180 if (cpu_mitigations_off())
2181 l1tf_mitigation = L1TF_MITIGATION_OFF;
2182 else if (cpu_mitigations_auto_nosmt())
2183 l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2184
2185 override_cache_bits(&boot_cpu_data);
2186
2187 switch (l1tf_mitigation) {
2188 case L1TF_MITIGATION_OFF:
2189 case L1TF_MITIGATION_FLUSH_NOWARN:
2190 case L1TF_MITIGATION_FLUSH:
2191 break;
2192 case L1TF_MITIGATION_FLUSH_NOSMT:
2193 case L1TF_MITIGATION_FULL:
2194 cpu_smt_disable(false);
2195 break;
2196 case L1TF_MITIGATION_FULL_FORCE:
2197 cpu_smt_disable(true);
2198 break;
2199 }
2200
2201 #if CONFIG_PGTABLE_LEVELS == 2
2202 pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2203 return;
2204 #endif
2205
2206 half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2207 if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2208 e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2209 pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2210 pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2211 half_pa);
2212 pr_info("However, doing so will make a part of your RAM unusable.\n");
2213 pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2214 return;
2215 }
2216
2217 setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2218 }
2219
l1tf_cmdline(char * str)2220 static int __init l1tf_cmdline(char *str)
2221 {
2222 if (!boot_cpu_has_bug(X86_BUG_L1TF))
2223 return 0;
2224
2225 if (!str)
2226 return -EINVAL;
2227
2228 if (!strcmp(str, "off"))
2229 l1tf_mitigation = L1TF_MITIGATION_OFF;
2230 else if (!strcmp(str, "flush,nowarn"))
2231 l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2232 else if (!strcmp(str, "flush"))
2233 l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2234 else if (!strcmp(str, "flush,nosmt"))
2235 l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2236 else if (!strcmp(str, "full"))
2237 l1tf_mitigation = L1TF_MITIGATION_FULL;
2238 else if (!strcmp(str, "full,force"))
2239 l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2240
2241 return 0;
2242 }
2243 early_param("l1tf", l1tf_cmdline);
2244
2245 #undef pr_fmt
2246 #define pr_fmt(fmt) fmt
2247
2248 #ifdef CONFIG_SYSFS
2249
2250 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2251
2252 #if IS_ENABLED(CONFIG_KVM_INTEL)
2253 static const char * const l1tf_vmx_states[] = {
2254 [VMENTER_L1D_FLUSH_AUTO] = "auto",
2255 [VMENTER_L1D_FLUSH_NEVER] = "vulnerable",
2256 [VMENTER_L1D_FLUSH_COND] = "conditional cache flushes",
2257 [VMENTER_L1D_FLUSH_ALWAYS] = "cache flushes",
2258 [VMENTER_L1D_FLUSH_EPT_DISABLED] = "EPT disabled",
2259 [VMENTER_L1D_FLUSH_NOT_REQUIRED] = "flush not necessary"
2260 };
2261
l1tf_show_state(char * buf)2262 static ssize_t l1tf_show_state(char *buf)
2263 {
2264 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2265 return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2266
2267 if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2268 (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2269 sched_smt_active())) {
2270 return sysfs_emit(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2271 l1tf_vmx_states[l1tf_vmx_mitigation]);
2272 }
2273
2274 return sysfs_emit(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2275 l1tf_vmx_states[l1tf_vmx_mitigation],
2276 sched_smt_active() ? "vulnerable" : "disabled");
2277 }
2278
itlb_multihit_show_state(char * buf)2279 static ssize_t itlb_multihit_show_state(char *buf)
2280 {
2281 if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2282 !boot_cpu_has(X86_FEATURE_VMX))
2283 return sysfs_emit(buf, "KVM: Mitigation: VMX unsupported\n");
2284 else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2285 return sysfs_emit(buf, "KVM: Mitigation: VMX disabled\n");
2286 else if (itlb_multihit_kvm_mitigation)
2287 return sysfs_emit(buf, "KVM: Mitigation: Split huge pages\n");
2288 else
2289 return sysfs_emit(buf, "KVM: Vulnerable\n");
2290 }
2291 #else
l1tf_show_state(char * buf)2292 static ssize_t l1tf_show_state(char *buf)
2293 {
2294 return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2295 }
2296
itlb_multihit_show_state(char * buf)2297 static ssize_t itlb_multihit_show_state(char *buf)
2298 {
2299 return sysfs_emit(buf, "Processor vulnerable\n");
2300 }
2301 #endif
2302
mds_show_state(char * buf)2303 static ssize_t mds_show_state(char *buf)
2304 {
2305 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2306 return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2307 mds_strings[mds_mitigation]);
2308 }
2309
2310 if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2311 return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2312 (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2313 sched_smt_active() ? "mitigated" : "disabled"));
2314 }
2315
2316 return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2317 sched_smt_active() ? "vulnerable" : "disabled");
2318 }
2319
tsx_async_abort_show_state(char * buf)2320 static ssize_t tsx_async_abort_show_state(char *buf)
2321 {
2322 if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2323 (taa_mitigation == TAA_MITIGATION_OFF))
2324 return sysfs_emit(buf, "%s\n", taa_strings[taa_mitigation]);
2325
2326 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2327 return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2328 taa_strings[taa_mitigation]);
2329 }
2330
2331 return sysfs_emit(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2332 sched_smt_active() ? "vulnerable" : "disabled");
2333 }
2334
mmio_stale_data_show_state(char * buf)2335 static ssize_t mmio_stale_data_show_state(char *buf)
2336 {
2337 if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2338 return sysfs_emit(buf, "Unknown: No mitigations\n");
2339
2340 if (mmio_mitigation == MMIO_MITIGATION_OFF)
2341 return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2342
2343 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2344 return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2345 mmio_strings[mmio_mitigation]);
2346 }
2347
2348 return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2349 sched_smt_active() ? "vulnerable" : "disabled");
2350 }
2351
stibp_state(void)2352 static char *stibp_state(void)
2353 {
2354 if (spectre_v2_in_eibrs_mode(spectre_v2_enabled))
2355 return "";
2356
2357 switch (spectre_v2_user_stibp) {
2358 case SPECTRE_V2_USER_NONE:
2359 return ", STIBP: disabled";
2360 case SPECTRE_V2_USER_STRICT:
2361 return ", STIBP: forced";
2362 case SPECTRE_V2_USER_STRICT_PREFERRED:
2363 return ", STIBP: always-on";
2364 case SPECTRE_V2_USER_PRCTL:
2365 case SPECTRE_V2_USER_SECCOMP:
2366 if (static_key_enabled(&switch_to_cond_stibp))
2367 return ", STIBP: conditional";
2368 }
2369 return "";
2370 }
2371
ibpb_state(void)2372 static char *ibpb_state(void)
2373 {
2374 if (boot_cpu_has(X86_FEATURE_IBPB)) {
2375 if (static_key_enabled(&switch_mm_always_ibpb))
2376 return ", IBPB: always-on";
2377 if (static_key_enabled(&switch_mm_cond_ibpb))
2378 return ", IBPB: conditional";
2379 return ", IBPB: disabled";
2380 }
2381 return "";
2382 }
2383
pbrsb_eibrs_state(void)2384 static char *pbrsb_eibrs_state(void)
2385 {
2386 if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2387 if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2388 boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2389 return ", PBRSB-eIBRS: SW sequence";
2390 else
2391 return ", PBRSB-eIBRS: Vulnerable";
2392 } else {
2393 return ", PBRSB-eIBRS: Not affected";
2394 }
2395 }
2396
spectre_v2_show_state(char * buf)2397 static ssize_t spectre_v2_show_state(char *buf)
2398 {
2399 if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2400 return sysfs_emit(buf, "Vulnerable: LFENCE\n");
2401
2402 if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2403 return sysfs_emit(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2404
2405 if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2406 spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2407 return sysfs_emit(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2408
2409 return sysfs_emit(buf, "%s%s%s%s%s%s%s\n",
2410 spectre_v2_strings[spectre_v2_enabled],
2411 ibpb_state(),
2412 boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2413 stibp_state(),
2414 boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2415 pbrsb_eibrs_state(),
2416 spectre_v2_module_string());
2417 }
2418
srbds_show_state(char * buf)2419 static ssize_t srbds_show_state(char *buf)
2420 {
2421 return sysfs_emit(buf, "%s\n", srbds_strings[srbds_mitigation]);
2422 }
2423
retbleed_show_state(char * buf)2424 static ssize_t retbleed_show_state(char *buf)
2425 {
2426 if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2427 retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2428 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2429 boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2430 return sysfs_emit(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2431
2432 return sysfs_emit(buf, "%s; SMT %s\n", retbleed_strings[retbleed_mitigation],
2433 !sched_smt_active() ? "disabled" :
2434 spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2435 spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2436 "enabled with STIBP protection" : "vulnerable");
2437 }
2438
2439 return sysfs_emit(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2440 }
2441
cpu_show_common(struct device * dev,struct device_attribute * attr,char * buf,unsigned int bug)2442 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2443 char *buf, unsigned int bug)
2444 {
2445 if (!boot_cpu_has_bug(bug))
2446 return sysfs_emit(buf, "Not affected\n");
2447
2448 switch (bug) {
2449 case X86_BUG_CPU_MELTDOWN:
2450 if (boot_cpu_has(X86_FEATURE_PTI))
2451 return sysfs_emit(buf, "Mitigation: PTI\n");
2452
2453 if (hypervisor_is_type(X86_HYPER_XEN_PV))
2454 return sysfs_emit(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2455
2456 break;
2457
2458 case X86_BUG_SPECTRE_V1:
2459 return sysfs_emit(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2460
2461 case X86_BUG_SPECTRE_V2:
2462 return spectre_v2_show_state(buf);
2463
2464 case X86_BUG_SPEC_STORE_BYPASS:
2465 return sysfs_emit(buf, "%s\n", ssb_strings[ssb_mode]);
2466
2467 case X86_BUG_L1TF:
2468 if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2469 return l1tf_show_state(buf);
2470 break;
2471
2472 case X86_BUG_MDS:
2473 return mds_show_state(buf);
2474
2475 case X86_BUG_TAA:
2476 return tsx_async_abort_show_state(buf);
2477
2478 case X86_BUG_ITLB_MULTIHIT:
2479 return itlb_multihit_show_state(buf);
2480
2481 case X86_BUG_SRBDS:
2482 return srbds_show_state(buf);
2483
2484 case X86_BUG_MMIO_STALE_DATA:
2485 case X86_BUG_MMIO_UNKNOWN:
2486 return mmio_stale_data_show_state(buf);
2487
2488 case X86_BUG_RETBLEED:
2489 return retbleed_show_state(buf);
2490
2491 default:
2492 break;
2493 }
2494
2495 return sysfs_emit(buf, "Vulnerable\n");
2496 }
2497
cpu_show_meltdown(struct device * dev,struct device_attribute * attr,char * buf)2498 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2499 {
2500 return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2501 }
2502
cpu_show_spectre_v1(struct device * dev,struct device_attribute * attr,char * buf)2503 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2504 {
2505 return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2506 }
2507
cpu_show_spectre_v2(struct device * dev,struct device_attribute * attr,char * buf)2508 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2509 {
2510 return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2511 }
2512
cpu_show_spec_store_bypass(struct device * dev,struct device_attribute * attr,char * buf)2513 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2514 {
2515 return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2516 }
2517
cpu_show_l1tf(struct device * dev,struct device_attribute * attr,char * buf)2518 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2519 {
2520 return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2521 }
2522
cpu_show_mds(struct device * dev,struct device_attribute * attr,char * buf)2523 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2524 {
2525 return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2526 }
2527
cpu_show_tsx_async_abort(struct device * dev,struct device_attribute * attr,char * buf)2528 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2529 {
2530 return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2531 }
2532
cpu_show_itlb_multihit(struct device * dev,struct device_attribute * attr,char * buf)2533 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2534 {
2535 return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2536 }
2537
cpu_show_srbds(struct device * dev,struct device_attribute * attr,char * buf)2538 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2539 {
2540 return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2541 }
2542
cpu_show_mmio_stale_data(struct device * dev,struct device_attribute * attr,char * buf)2543 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2544 {
2545 if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2546 return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2547 else
2548 return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2549 }
2550
cpu_show_retbleed(struct device * dev,struct device_attribute * attr,char * buf)2551 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2552 {
2553 return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2554 }
2555 #endif
2556