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