xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision aadaeca4)
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 
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 ssb_select_mitigation(void);
42 static void __init l1tf_select_mitigation(void);
43 static void __init mds_select_mitigation(void);
44 static void __init mds_print_mitigation(void);
45 static void __init taa_select_mitigation(void);
46 static void __init srbds_select_mitigation(void);
47 static void __init l1d_flush_select_mitigation(void);
48 
49 /* The base value of the SPEC_CTRL MSR that always has to be preserved. */
50 u64 x86_spec_ctrl_base;
51 EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
52 static DEFINE_MUTEX(spec_ctrl_mutex);
53 
54 /*
55  * The vendor and possibly platform specific bits which can be modified in
56  * x86_spec_ctrl_base.
57  */
58 static u64 __ro_after_init x86_spec_ctrl_mask = SPEC_CTRL_IBRS;
59 
60 /*
61  * AMD specific MSR info for Speculative Store Bypass control.
62  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
63  */
64 u64 __ro_after_init x86_amd_ls_cfg_base;
65 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
66 
67 /* Control conditional STIBP in switch_to() */
68 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
69 /* Control conditional IBPB in switch_mm() */
70 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
71 /* Control unconditional IBPB in switch_mm() */
72 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
73 
74 /* Control MDS CPU buffer clear before returning to user space */
75 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
76 EXPORT_SYMBOL_GPL(mds_user_clear);
77 /* Control MDS CPU buffer clear before idling (halt, mwait) */
78 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
79 EXPORT_SYMBOL_GPL(mds_idle_clear);
80 
81 /*
82  * Controls whether l1d flush based mitigations are enabled,
83  * based on hw features and admin setting via boot parameter
84  * defaults to false
85  */
86 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
87 
88 void __init check_bugs(void)
89 {
90 	identify_boot_cpu();
91 
92 	/*
93 	 * identify_boot_cpu() initialized SMT support information, let the
94 	 * core code know.
95 	 */
96 	cpu_smt_check_topology();
97 
98 	if (!IS_ENABLED(CONFIG_SMP)) {
99 		pr_info("CPU: ");
100 		print_cpu_info(&boot_cpu_data);
101 	}
102 
103 	/*
104 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
105 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
106 	 * init code as it is not enumerated and depends on the family.
107 	 */
108 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
109 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
110 
111 	/* Allow STIBP in MSR_SPEC_CTRL if supported */
112 	if (boot_cpu_has(X86_FEATURE_STIBP))
113 		x86_spec_ctrl_mask |= SPEC_CTRL_STIBP;
114 
115 	/* Select the proper CPU mitigations before patching alternatives: */
116 	spectre_v1_select_mitigation();
117 	spectre_v2_select_mitigation();
118 	ssb_select_mitigation();
119 	l1tf_select_mitigation();
120 	mds_select_mitigation();
121 	taa_select_mitigation();
122 	srbds_select_mitigation();
123 	l1d_flush_select_mitigation();
124 
125 	/*
126 	 * As MDS and TAA mitigations are inter-related, print MDS
127 	 * mitigation until after TAA mitigation selection is done.
128 	 */
129 	mds_print_mitigation();
130 
131 	arch_smt_update();
132 
133 #ifdef CONFIG_X86_32
134 	/*
135 	 * Check whether we are able to run this kernel safely on SMP.
136 	 *
137 	 * - i386 is no longer supported.
138 	 * - In order to run on anything without a TSC, we need to be
139 	 *   compiled for a i486.
140 	 */
141 	if (boot_cpu_data.x86 < 4)
142 		panic("Kernel requires i486+ for 'invlpg' and other features");
143 
144 	init_utsname()->machine[1] =
145 		'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
146 	alternative_instructions();
147 
148 	fpu__init_check_bugs();
149 #else /* CONFIG_X86_64 */
150 	alternative_instructions();
151 
152 	/*
153 	 * Make sure the first 2MB area is not mapped by huge pages
154 	 * There are typically fixed size MTRRs in there and overlapping
155 	 * MTRRs into large pages causes slow downs.
156 	 *
157 	 * Right now we don't do that with gbpages because there seems
158 	 * very little benefit for that case.
159 	 */
160 	if (!direct_gbpages)
161 		set_memory_4k((unsigned long)__va(0), 1);
162 #endif
163 }
164 
165 void
166 x86_virt_spec_ctrl(u64 guest_spec_ctrl, u64 guest_virt_spec_ctrl, bool setguest)
167 {
168 	u64 msrval, guestval, hostval = x86_spec_ctrl_base;
169 	struct thread_info *ti = current_thread_info();
170 
171 	/* Is MSR_SPEC_CTRL implemented ? */
172 	if (static_cpu_has(X86_FEATURE_MSR_SPEC_CTRL)) {
173 		/*
174 		 * Restrict guest_spec_ctrl to supported values. Clear the
175 		 * modifiable bits in the host base value and or the
176 		 * modifiable bits from the guest value.
177 		 */
178 		guestval = hostval & ~x86_spec_ctrl_mask;
179 		guestval |= guest_spec_ctrl & x86_spec_ctrl_mask;
180 
181 		/* SSBD controlled in MSR_SPEC_CTRL */
182 		if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
183 		    static_cpu_has(X86_FEATURE_AMD_SSBD))
184 			hostval |= ssbd_tif_to_spec_ctrl(ti->flags);
185 
186 		/* Conditional STIBP enabled? */
187 		if (static_branch_unlikely(&switch_to_cond_stibp))
188 			hostval |= stibp_tif_to_spec_ctrl(ti->flags);
189 
190 		if (hostval != guestval) {
191 			msrval = setguest ? guestval : hostval;
192 			wrmsrl(MSR_IA32_SPEC_CTRL, msrval);
193 		}
194 	}
195 
196 	/*
197 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
198 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
199 	 */
200 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
201 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
202 		return;
203 
204 	/*
205 	 * If the host has SSBD mitigation enabled, force it in the host's
206 	 * virtual MSR value. If its not permanently enabled, evaluate
207 	 * current's TIF_SSBD thread flag.
208 	 */
209 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
210 		hostval = SPEC_CTRL_SSBD;
211 	else
212 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
213 
214 	/* Sanitize the guest value */
215 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
216 
217 	if (hostval != guestval) {
218 		unsigned long tif;
219 
220 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
221 				 ssbd_spec_ctrl_to_tif(hostval);
222 
223 		speculation_ctrl_update(tif);
224 	}
225 }
226 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
227 
228 static void x86_amd_ssb_disable(void)
229 {
230 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
231 
232 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
233 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
234 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
235 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
236 }
237 
238 #undef pr_fmt
239 #define pr_fmt(fmt)	"MDS: " fmt
240 
241 /* Default mitigation for MDS-affected CPUs */
242 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
243 static bool mds_nosmt __ro_after_init = false;
244 
245 static const char * const mds_strings[] = {
246 	[MDS_MITIGATION_OFF]	= "Vulnerable",
247 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
248 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
249 };
250 
251 static void __init mds_select_mitigation(void)
252 {
253 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
254 		mds_mitigation = MDS_MITIGATION_OFF;
255 		return;
256 	}
257 
258 	if (mds_mitigation == MDS_MITIGATION_FULL) {
259 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
260 			mds_mitigation = MDS_MITIGATION_VMWERV;
261 
262 		static_branch_enable(&mds_user_clear);
263 
264 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
265 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
266 			cpu_smt_disable(false);
267 	}
268 }
269 
270 static void __init mds_print_mitigation(void)
271 {
272 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off())
273 		return;
274 
275 	pr_info("%s\n", mds_strings[mds_mitigation]);
276 }
277 
278 static int __init mds_cmdline(char *str)
279 {
280 	if (!boot_cpu_has_bug(X86_BUG_MDS))
281 		return 0;
282 
283 	if (!str)
284 		return -EINVAL;
285 
286 	if (!strcmp(str, "off"))
287 		mds_mitigation = MDS_MITIGATION_OFF;
288 	else if (!strcmp(str, "full"))
289 		mds_mitigation = MDS_MITIGATION_FULL;
290 	else if (!strcmp(str, "full,nosmt")) {
291 		mds_mitigation = MDS_MITIGATION_FULL;
292 		mds_nosmt = true;
293 	}
294 
295 	return 0;
296 }
297 early_param("mds", mds_cmdline);
298 
299 #undef pr_fmt
300 #define pr_fmt(fmt)	"TAA: " fmt
301 
302 enum taa_mitigations {
303 	TAA_MITIGATION_OFF,
304 	TAA_MITIGATION_UCODE_NEEDED,
305 	TAA_MITIGATION_VERW,
306 	TAA_MITIGATION_TSX_DISABLED,
307 };
308 
309 /* Default mitigation for TAA-affected CPUs */
310 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
311 static bool taa_nosmt __ro_after_init;
312 
313 static const char * const taa_strings[] = {
314 	[TAA_MITIGATION_OFF]		= "Vulnerable",
315 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
316 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
317 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
318 };
319 
320 static void __init taa_select_mitigation(void)
321 {
322 	u64 ia32_cap;
323 
324 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
325 		taa_mitigation = TAA_MITIGATION_OFF;
326 		return;
327 	}
328 
329 	/* TSX previously disabled by tsx=off */
330 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
331 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
332 		goto out;
333 	}
334 
335 	if (cpu_mitigations_off()) {
336 		taa_mitigation = TAA_MITIGATION_OFF;
337 		return;
338 	}
339 
340 	/*
341 	 * TAA mitigation via VERW is turned off if both
342 	 * tsx_async_abort=off and mds=off are specified.
343 	 */
344 	if (taa_mitigation == TAA_MITIGATION_OFF &&
345 	    mds_mitigation == MDS_MITIGATION_OFF)
346 		goto out;
347 
348 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
349 		taa_mitigation = TAA_MITIGATION_VERW;
350 	else
351 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
352 
353 	/*
354 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
355 	 * A microcode update fixes this behavior to clear CPU buffers. It also
356 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
357 	 * ARCH_CAP_TSX_CTRL_MSR bit.
358 	 *
359 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
360 	 * update is required.
361 	 */
362 	ia32_cap = x86_read_arch_cap_msr();
363 	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
364 	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
365 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
366 
367 	/*
368 	 * TSX is enabled, select alternate mitigation for TAA which is
369 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
370 	 *
371 	 * For guests that can't determine whether the correct microcode is
372 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
373 	 */
374 	static_branch_enable(&mds_user_clear);
375 
376 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
377 		cpu_smt_disable(false);
378 
379 	/*
380 	 * Update MDS mitigation, if necessary, as the mds_user_clear is
381 	 * now enabled for TAA mitigation.
382 	 */
383 	if (mds_mitigation == MDS_MITIGATION_OFF &&
384 	    boot_cpu_has_bug(X86_BUG_MDS)) {
385 		mds_mitigation = MDS_MITIGATION_FULL;
386 		mds_select_mitigation();
387 	}
388 out:
389 	pr_info("%s\n", taa_strings[taa_mitigation]);
390 }
391 
392 static int __init tsx_async_abort_parse_cmdline(char *str)
393 {
394 	if (!boot_cpu_has_bug(X86_BUG_TAA))
395 		return 0;
396 
397 	if (!str)
398 		return -EINVAL;
399 
400 	if (!strcmp(str, "off")) {
401 		taa_mitigation = TAA_MITIGATION_OFF;
402 	} else if (!strcmp(str, "full")) {
403 		taa_mitigation = TAA_MITIGATION_VERW;
404 	} else if (!strcmp(str, "full,nosmt")) {
405 		taa_mitigation = TAA_MITIGATION_VERW;
406 		taa_nosmt = true;
407 	}
408 
409 	return 0;
410 }
411 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
412 
413 #undef pr_fmt
414 #define pr_fmt(fmt)	"SRBDS: " fmt
415 
416 enum srbds_mitigations {
417 	SRBDS_MITIGATION_OFF,
418 	SRBDS_MITIGATION_UCODE_NEEDED,
419 	SRBDS_MITIGATION_FULL,
420 	SRBDS_MITIGATION_TSX_OFF,
421 	SRBDS_MITIGATION_HYPERVISOR,
422 };
423 
424 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
425 
426 static const char * const srbds_strings[] = {
427 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
428 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
429 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
430 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
431 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
432 };
433 
434 static bool srbds_off;
435 
436 void update_srbds_msr(void)
437 {
438 	u64 mcu_ctrl;
439 
440 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
441 		return;
442 
443 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
444 		return;
445 
446 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
447 		return;
448 
449 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
450 
451 	switch (srbds_mitigation) {
452 	case SRBDS_MITIGATION_OFF:
453 	case SRBDS_MITIGATION_TSX_OFF:
454 		mcu_ctrl |= RNGDS_MITG_DIS;
455 		break;
456 	case SRBDS_MITIGATION_FULL:
457 		mcu_ctrl &= ~RNGDS_MITG_DIS;
458 		break;
459 	default:
460 		break;
461 	}
462 
463 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
464 }
465 
466 static void __init srbds_select_mitigation(void)
467 {
468 	u64 ia32_cap;
469 
470 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
471 		return;
472 
473 	/*
474 	 * Check to see if this is one of the MDS_NO systems supporting
475 	 * TSX that are only exposed to SRBDS when TSX is enabled.
476 	 */
477 	ia32_cap = x86_read_arch_cap_msr();
478 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM))
479 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
480 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
481 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
482 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
483 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
484 	else if (cpu_mitigations_off() || srbds_off)
485 		srbds_mitigation = SRBDS_MITIGATION_OFF;
486 
487 	update_srbds_msr();
488 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
489 }
490 
491 static int __init srbds_parse_cmdline(char *str)
492 {
493 	if (!str)
494 		return -EINVAL;
495 
496 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
497 		return 0;
498 
499 	srbds_off = !strcmp(str, "off");
500 	return 0;
501 }
502 early_param("srbds", srbds_parse_cmdline);
503 
504 #undef pr_fmt
505 #define pr_fmt(fmt)     "L1D Flush : " fmt
506 
507 enum l1d_flush_mitigations {
508 	L1D_FLUSH_OFF = 0,
509 	L1D_FLUSH_ON,
510 };
511 
512 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
513 
514 static void __init l1d_flush_select_mitigation(void)
515 {
516 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
517 		return;
518 
519 	static_branch_enable(&switch_mm_cond_l1d_flush);
520 	pr_info("Conditional flush on switch_mm() enabled\n");
521 }
522 
523 static int __init l1d_flush_parse_cmdline(char *str)
524 {
525 	if (!strcmp(str, "on"))
526 		l1d_flush_mitigation = L1D_FLUSH_ON;
527 
528 	return 0;
529 }
530 early_param("l1d_flush", l1d_flush_parse_cmdline);
531 
532 #undef pr_fmt
533 #define pr_fmt(fmt)     "Spectre V1 : " fmt
534 
535 enum spectre_v1_mitigation {
536 	SPECTRE_V1_MITIGATION_NONE,
537 	SPECTRE_V1_MITIGATION_AUTO,
538 };
539 
540 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
541 	SPECTRE_V1_MITIGATION_AUTO;
542 
543 static const char * const spectre_v1_strings[] = {
544 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
545 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
546 };
547 
548 /*
549  * Does SMAP provide full mitigation against speculative kernel access to
550  * userspace?
551  */
552 static bool smap_works_speculatively(void)
553 {
554 	if (!boot_cpu_has(X86_FEATURE_SMAP))
555 		return false;
556 
557 	/*
558 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
559 	 * prevent speculative access to user data in the L1 cache.
560 	 * Consider SMAP to be non-functional as a mitigation on these
561 	 * CPUs.
562 	 */
563 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
564 		return false;
565 
566 	return true;
567 }
568 
569 static void __init spectre_v1_select_mitigation(void)
570 {
571 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
572 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
573 		return;
574 	}
575 
576 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
577 		/*
578 		 * With Spectre v1, a user can speculatively control either
579 		 * path of a conditional swapgs with a user-controlled GS
580 		 * value.  The mitigation is to add lfences to both code paths.
581 		 *
582 		 * If FSGSBASE is enabled, the user can put a kernel address in
583 		 * GS, in which case SMAP provides no protection.
584 		 *
585 		 * If FSGSBASE is disabled, the user can only put a user space
586 		 * address in GS.  That makes an attack harder, but still
587 		 * possible if there's no SMAP protection.
588 		 */
589 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
590 		    !smap_works_speculatively()) {
591 			/*
592 			 * Mitigation can be provided from SWAPGS itself or
593 			 * PTI as the CR3 write in the Meltdown mitigation
594 			 * is serializing.
595 			 *
596 			 * If neither is there, mitigate with an LFENCE to
597 			 * stop speculation through swapgs.
598 			 */
599 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
600 			    !boot_cpu_has(X86_FEATURE_PTI))
601 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
602 
603 			/*
604 			 * Enable lfences in the kernel entry (non-swapgs)
605 			 * paths, to prevent user entry from speculatively
606 			 * skipping swapgs.
607 			 */
608 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
609 		}
610 	}
611 
612 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
613 }
614 
615 static int __init nospectre_v1_cmdline(char *str)
616 {
617 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
618 	return 0;
619 }
620 early_param("nospectre_v1", nospectre_v1_cmdline);
621 
622 #undef pr_fmt
623 #define pr_fmt(fmt)     "Spectre V2 : " fmt
624 
625 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
626 	SPECTRE_V2_NONE;
627 
628 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
629 	SPECTRE_V2_USER_NONE;
630 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
631 	SPECTRE_V2_USER_NONE;
632 
633 #ifdef CONFIG_RETPOLINE
634 static bool spectre_v2_bad_module;
635 
636 bool retpoline_module_ok(bool has_retpoline)
637 {
638 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
639 		return true;
640 
641 	pr_err("System may be vulnerable to spectre v2\n");
642 	spectre_v2_bad_module = true;
643 	return false;
644 }
645 
646 static inline const char *spectre_v2_module_string(void)
647 {
648 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
649 }
650 #else
651 static inline const char *spectre_v2_module_string(void) { return ""; }
652 #endif
653 
654 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
655 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
656 #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"
657 
658 #ifdef CONFIG_BPF_SYSCALL
659 void unpriv_ebpf_notify(int new_state)
660 {
661 	if (new_state)
662 		return;
663 
664 	/* Unprivileged eBPF is enabled */
665 
666 	switch (spectre_v2_enabled) {
667 	case SPECTRE_V2_EIBRS:
668 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
669 		break;
670 	case SPECTRE_V2_EIBRS_LFENCE:
671 		if (sched_smt_active())
672 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
673 		break;
674 	default:
675 		break;
676 	}
677 }
678 #endif
679 
680 static inline bool match_option(const char *arg, int arglen, const char *opt)
681 {
682 	int len = strlen(opt);
683 
684 	return len == arglen && !strncmp(arg, opt, len);
685 }
686 
687 /* The kernel command line selection for spectre v2 */
688 enum spectre_v2_mitigation_cmd {
689 	SPECTRE_V2_CMD_NONE,
690 	SPECTRE_V2_CMD_AUTO,
691 	SPECTRE_V2_CMD_FORCE,
692 	SPECTRE_V2_CMD_RETPOLINE,
693 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
694 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
695 	SPECTRE_V2_CMD_EIBRS,
696 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
697 	SPECTRE_V2_CMD_EIBRS_LFENCE,
698 };
699 
700 enum spectre_v2_user_cmd {
701 	SPECTRE_V2_USER_CMD_NONE,
702 	SPECTRE_V2_USER_CMD_AUTO,
703 	SPECTRE_V2_USER_CMD_FORCE,
704 	SPECTRE_V2_USER_CMD_PRCTL,
705 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
706 	SPECTRE_V2_USER_CMD_SECCOMP,
707 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
708 };
709 
710 static const char * const spectre_v2_user_strings[] = {
711 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
712 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
713 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
714 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
715 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
716 };
717 
718 static const struct {
719 	const char			*option;
720 	enum spectre_v2_user_cmd	cmd;
721 	bool				secure;
722 } v2_user_options[] __initconst = {
723 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
724 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
725 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
726 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
727 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
728 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
729 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
730 };
731 
732 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
733 {
734 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
735 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
736 }
737 
738 static enum spectre_v2_user_cmd __init
739 spectre_v2_parse_user_cmdline(enum spectre_v2_mitigation_cmd v2_cmd)
740 {
741 	char arg[20];
742 	int ret, i;
743 
744 	switch (v2_cmd) {
745 	case SPECTRE_V2_CMD_NONE:
746 		return SPECTRE_V2_USER_CMD_NONE;
747 	case SPECTRE_V2_CMD_FORCE:
748 		return SPECTRE_V2_USER_CMD_FORCE;
749 	default:
750 		break;
751 	}
752 
753 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
754 				  arg, sizeof(arg));
755 	if (ret < 0)
756 		return SPECTRE_V2_USER_CMD_AUTO;
757 
758 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
759 		if (match_option(arg, ret, v2_user_options[i].option)) {
760 			spec_v2_user_print_cond(v2_user_options[i].option,
761 						v2_user_options[i].secure);
762 			return v2_user_options[i].cmd;
763 		}
764 	}
765 
766 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
767 	return SPECTRE_V2_USER_CMD_AUTO;
768 }
769 
770 static inline bool spectre_v2_in_eibrs_mode(enum spectre_v2_mitigation mode)
771 {
772 	return (mode == SPECTRE_V2_EIBRS ||
773 		mode == SPECTRE_V2_EIBRS_RETPOLINE ||
774 		mode == SPECTRE_V2_EIBRS_LFENCE);
775 }
776 
777 static void __init
778 spectre_v2_user_select_mitigation(enum spectre_v2_mitigation_cmd v2_cmd)
779 {
780 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
781 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
782 	enum spectre_v2_user_cmd cmd;
783 
784 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
785 		return;
786 
787 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
788 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
789 		smt_possible = false;
790 
791 	cmd = spectre_v2_parse_user_cmdline(v2_cmd);
792 	switch (cmd) {
793 	case SPECTRE_V2_USER_CMD_NONE:
794 		goto set_mode;
795 	case SPECTRE_V2_USER_CMD_FORCE:
796 		mode = SPECTRE_V2_USER_STRICT;
797 		break;
798 	case SPECTRE_V2_USER_CMD_AUTO:
799 	case SPECTRE_V2_USER_CMD_PRCTL:
800 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
801 		mode = SPECTRE_V2_USER_PRCTL;
802 		break;
803 	case SPECTRE_V2_USER_CMD_SECCOMP:
804 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
805 		if (IS_ENABLED(CONFIG_SECCOMP))
806 			mode = SPECTRE_V2_USER_SECCOMP;
807 		else
808 			mode = SPECTRE_V2_USER_PRCTL;
809 		break;
810 	}
811 
812 	/* Initialize Indirect Branch Prediction Barrier */
813 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
814 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
815 
816 		spectre_v2_user_ibpb = mode;
817 		switch (cmd) {
818 		case SPECTRE_V2_USER_CMD_FORCE:
819 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
820 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
821 			static_branch_enable(&switch_mm_always_ibpb);
822 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
823 			break;
824 		case SPECTRE_V2_USER_CMD_PRCTL:
825 		case SPECTRE_V2_USER_CMD_AUTO:
826 		case SPECTRE_V2_USER_CMD_SECCOMP:
827 			static_branch_enable(&switch_mm_cond_ibpb);
828 			break;
829 		default:
830 			break;
831 		}
832 
833 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
834 			static_key_enabled(&switch_mm_always_ibpb) ?
835 			"always-on" : "conditional");
836 	}
837 
838 	/*
839 	 * If no STIBP, enhanced IBRS is enabled or SMT impossible, STIBP is not
840 	 * required.
841 	 */
842 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
843 	    !smt_possible ||
844 	    spectre_v2_in_eibrs_mode(spectre_v2_enabled))
845 		return;
846 
847 	/*
848 	 * At this point, an STIBP mode other than "off" has been set.
849 	 * If STIBP support is not being forced, check if STIBP always-on
850 	 * is preferred.
851 	 */
852 	if (mode != SPECTRE_V2_USER_STRICT &&
853 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
854 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
855 
856 	spectre_v2_user_stibp = mode;
857 
858 set_mode:
859 	pr_info("%s\n", spectre_v2_user_strings[mode]);
860 }
861 
862 static const char * const spectre_v2_strings[] = {
863 	[SPECTRE_V2_NONE]			= "Vulnerable",
864 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
865 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
866 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced IBRS",
867 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced IBRS + LFENCE",
868 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced IBRS + Retpolines",
869 };
870 
871 static const struct {
872 	const char *option;
873 	enum spectre_v2_mitigation_cmd cmd;
874 	bool secure;
875 } mitigation_options[] __initconst = {
876 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
877 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
878 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
879 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
880 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
881 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
882 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
883 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
884 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
885 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
886 };
887 
888 static void __init spec_v2_print_cond(const char *reason, bool secure)
889 {
890 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
891 		pr_info("%s selected on command line.\n", reason);
892 }
893 
894 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
895 {
896 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
897 	char arg[20];
898 	int ret, i;
899 
900 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
901 	    cpu_mitigations_off())
902 		return SPECTRE_V2_CMD_NONE;
903 
904 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
905 	if (ret < 0)
906 		return SPECTRE_V2_CMD_AUTO;
907 
908 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
909 		if (!match_option(arg, ret, mitigation_options[i].option))
910 			continue;
911 		cmd = mitigation_options[i].cmd;
912 		break;
913 	}
914 
915 	if (i >= ARRAY_SIZE(mitigation_options)) {
916 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
917 		return SPECTRE_V2_CMD_AUTO;
918 	}
919 
920 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
921 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
922 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
923 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
924 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
925 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
926 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
927 		       mitigation_options[i].option);
928 		return SPECTRE_V2_CMD_AUTO;
929 	}
930 
931 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
932 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
933 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
934 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
935 		pr_err("%s selected but CPU doesn't have eIBRS. Switching to AUTO select\n",
936 		       mitigation_options[i].option);
937 		return SPECTRE_V2_CMD_AUTO;
938 	}
939 
940 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
941 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
942 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
943 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
944 		       mitigation_options[i].option);
945 		return SPECTRE_V2_CMD_AUTO;
946 	}
947 
948 	spec_v2_print_cond(mitigation_options[i].option,
949 			   mitigation_options[i].secure);
950 	return cmd;
951 }
952 
953 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
954 {
955 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
956 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
957 		return SPECTRE_V2_NONE;
958 	}
959 
960 	return SPECTRE_V2_RETPOLINE;
961 }
962 
963 static void __init spectre_v2_select_mitigation(void)
964 {
965 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
966 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
967 
968 	/*
969 	 * If the CPU is not affected and the command line mode is NONE or AUTO
970 	 * then nothing to do.
971 	 */
972 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
973 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
974 		return;
975 
976 	switch (cmd) {
977 	case SPECTRE_V2_CMD_NONE:
978 		return;
979 
980 	case SPECTRE_V2_CMD_FORCE:
981 	case SPECTRE_V2_CMD_AUTO:
982 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
983 			mode = SPECTRE_V2_EIBRS;
984 			break;
985 		}
986 
987 		mode = spectre_v2_select_retpoline();
988 		break;
989 
990 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
991 		pr_err(SPECTRE_V2_LFENCE_MSG);
992 		mode = SPECTRE_V2_LFENCE;
993 		break;
994 
995 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
996 		mode = SPECTRE_V2_RETPOLINE;
997 		break;
998 
999 	case SPECTRE_V2_CMD_RETPOLINE:
1000 		mode = spectre_v2_select_retpoline();
1001 		break;
1002 
1003 	case SPECTRE_V2_CMD_EIBRS:
1004 		mode = SPECTRE_V2_EIBRS;
1005 		break;
1006 
1007 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1008 		mode = SPECTRE_V2_EIBRS_LFENCE;
1009 		break;
1010 
1011 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1012 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1013 		break;
1014 	}
1015 
1016 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1017 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1018 
1019 	if (spectre_v2_in_eibrs_mode(mode)) {
1020 		/* Force it so VMEXIT will restore correctly */
1021 		x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1022 		wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1023 	}
1024 
1025 	switch (mode) {
1026 	case SPECTRE_V2_NONE:
1027 	case SPECTRE_V2_EIBRS:
1028 		break;
1029 
1030 	case SPECTRE_V2_LFENCE:
1031 	case SPECTRE_V2_EIBRS_LFENCE:
1032 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1033 		fallthrough;
1034 
1035 	case SPECTRE_V2_RETPOLINE:
1036 	case SPECTRE_V2_EIBRS_RETPOLINE:
1037 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1038 		break;
1039 	}
1040 
1041 	spectre_v2_enabled = mode;
1042 	pr_info("%s\n", spectre_v2_strings[mode]);
1043 
1044 	/*
1045 	 * If spectre v2 protection has been enabled, unconditionally fill
1046 	 * RSB during a context switch; this protects against two independent
1047 	 * issues:
1048 	 *
1049 	 *	- RSB underflow (and switch to BTB) on Skylake+
1050 	 *	- SpectreRSB variant of spectre v2 on X86_BUG_SPECTRE_V2 CPUs
1051 	 */
1052 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1053 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1054 
1055 	/*
1056 	 * Retpoline means the kernel is safe because it has no indirect
1057 	 * branches. Enhanced IBRS protects firmware too, so, enable restricted
1058 	 * speculation around firmware calls only when Enhanced IBRS isn't
1059 	 * supported.
1060 	 *
1061 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1062 	 * the user might select retpoline on the kernel command line and if
1063 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1064 	 * enable IBRS around firmware calls.
1065 	 */
1066 	if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_eibrs_mode(mode)) {
1067 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1068 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1069 	}
1070 
1071 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1072 	spectre_v2_user_select_mitigation(cmd);
1073 }
1074 
1075 static void update_stibp_msr(void * __unused)
1076 {
1077 	wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1078 }
1079 
1080 /* Update x86_spec_ctrl_base in case SMT state changed. */
1081 static void update_stibp_strict(void)
1082 {
1083 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1084 
1085 	if (sched_smt_active())
1086 		mask |= SPEC_CTRL_STIBP;
1087 
1088 	if (mask == x86_spec_ctrl_base)
1089 		return;
1090 
1091 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1092 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1093 	x86_spec_ctrl_base = mask;
1094 	on_each_cpu(update_stibp_msr, NULL, 1);
1095 }
1096 
1097 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1098 static void update_indir_branch_cond(void)
1099 {
1100 	if (sched_smt_active())
1101 		static_branch_enable(&switch_to_cond_stibp);
1102 	else
1103 		static_branch_disable(&switch_to_cond_stibp);
1104 }
1105 
1106 #undef pr_fmt
1107 #define pr_fmt(fmt) fmt
1108 
1109 /* Update the static key controlling the MDS CPU buffer clear in idle */
1110 static void update_mds_branch_idle(void)
1111 {
1112 	/*
1113 	 * Enable the idle clearing if SMT is active on CPUs which are
1114 	 * affected only by MSBDS and not any other MDS variant.
1115 	 *
1116 	 * The other variants cannot be mitigated when SMT is enabled, so
1117 	 * clearing the buffers on idle just to prevent the Store Buffer
1118 	 * repartitioning leak would be a window dressing exercise.
1119 	 */
1120 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1121 		return;
1122 
1123 	if (sched_smt_active())
1124 		static_branch_enable(&mds_idle_clear);
1125 	else
1126 		static_branch_disable(&mds_idle_clear);
1127 }
1128 
1129 #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"
1130 #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"
1131 
1132 void cpu_bugs_smt_update(void)
1133 {
1134 	mutex_lock(&spec_ctrl_mutex);
1135 
1136 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1137 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1138 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1139 
1140 	switch (spectre_v2_user_stibp) {
1141 	case SPECTRE_V2_USER_NONE:
1142 		break;
1143 	case SPECTRE_V2_USER_STRICT:
1144 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1145 		update_stibp_strict();
1146 		break;
1147 	case SPECTRE_V2_USER_PRCTL:
1148 	case SPECTRE_V2_USER_SECCOMP:
1149 		update_indir_branch_cond();
1150 		break;
1151 	}
1152 
1153 	switch (mds_mitigation) {
1154 	case MDS_MITIGATION_FULL:
1155 	case MDS_MITIGATION_VMWERV:
1156 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1157 			pr_warn_once(MDS_MSG_SMT);
1158 		update_mds_branch_idle();
1159 		break;
1160 	case MDS_MITIGATION_OFF:
1161 		break;
1162 	}
1163 
1164 	switch (taa_mitigation) {
1165 	case TAA_MITIGATION_VERW:
1166 	case TAA_MITIGATION_UCODE_NEEDED:
1167 		if (sched_smt_active())
1168 			pr_warn_once(TAA_MSG_SMT);
1169 		break;
1170 	case TAA_MITIGATION_TSX_DISABLED:
1171 	case TAA_MITIGATION_OFF:
1172 		break;
1173 	}
1174 
1175 	mutex_unlock(&spec_ctrl_mutex);
1176 }
1177 
1178 #undef pr_fmt
1179 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1180 
1181 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1182 
1183 /* The kernel command line selection */
1184 enum ssb_mitigation_cmd {
1185 	SPEC_STORE_BYPASS_CMD_NONE,
1186 	SPEC_STORE_BYPASS_CMD_AUTO,
1187 	SPEC_STORE_BYPASS_CMD_ON,
1188 	SPEC_STORE_BYPASS_CMD_PRCTL,
1189 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1190 };
1191 
1192 static const char * const ssb_strings[] = {
1193 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1194 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1195 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1196 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1197 };
1198 
1199 static const struct {
1200 	const char *option;
1201 	enum ssb_mitigation_cmd cmd;
1202 } ssb_mitigation_options[]  __initconst = {
1203 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1204 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1205 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1206 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1207 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1208 };
1209 
1210 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1211 {
1212 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1213 	char arg[20];
1214 	int ret, i;
1215 
1216 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1217 	    cpu_mitigations_off()) {
1218 		return SPEC_STORE_BYPASS_CMD_NONE;
1219 	} else {
1220 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1221 					  arg, sizeof(arg));
1222 		if (ret < 0)
1223 			return SPEC_STORE_BYPASS_CMD_AUTO;
1224 
1225 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1226 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1227 				continue;
1228 
1229 			cmd = ssb_mitigation_options[i].cmd;
1230 			break;
1231 		}
1232 
1233 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1234 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1235 			return SPEC_STORE_BYPASS_CMD_AUTO;
1236 		}
1237 	}
1238 
1239 	return cmd;
1240 }
1241 
1242 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1243 {
1244 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1245 	enum ssb_mitigation_cmd cmd;
1246 
1247 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1248 		return mode;
1249 
1250 	cmd = ssb_parse_cmdline();
1251 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1252 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1253 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1254 		return mode;
1255 
1256 	switch (cmd) {
1257 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1258 		/*
1259 		 * Choose prctl+seccomp as the default mode if seccomp is
1260 		 * enabled.
1261 		 */
1262 		if (IS_ENABLED(CONFIG_SECCOMP))
1263 			mode = SPEC_STORE_BYPASS_SECCOMP;
1264 		else
1265 			mode = SPEC_STORE_BYPASS_PRCTL;
1266 		break;
1267 	case SPEC_STORE_BYPASS_CMD_ON:
1268 		mode = SPEC_STORE_BYPASS_DISABLE;
1269 		break;
1270 	case SPEC_STORE_BYPASS_CMD_AUTO:
1271 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1272 		mode = SPEC_STORE_BYPASS_PRCTL;
1273 		break;
1274 	case SPEC_STORE_BYPASS_CMD_NONE:
1275 		break;
1276 	}
1277 
1278 	/*
1279 	 * If SSBD is controlled by the SPEC_CTRL MSR, then set the proper
1280 	 * bit in the mask to allow guests to use the mitigation even in the
1281 	 * case where the host does not enable it.
1282 	 */
1283 	if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
1284 	    static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1285 		x86_spec_ctrl_mask |= SPEC_CTRL_SSBD;
1286 	}
1287 
1288 	/*
1289 	 * We have three CPU feature flags that are in play here:
1290 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1291 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1292 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1293 	 */
1294 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1295 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1296 		/*
1297 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1298 		 * use a completely different MSR and bit dependent on family.
1299 		 */
1300 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1301 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1302 			x86_amd_ssb_disable();
1303 		} else {
1304 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1305 			wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1306 		}
1307 	}
1308 
1309 	return mode;
1310 }
1311 
1312 static void ssb_select_mitigation(void)
1313 {
1314 	ssb_mode = __ssb_select_mitigation();
1315 
1316 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1317 		pr_info("%s\n", ssb_strings[ssb_mode]);
1318 }
1319 
1320 #undef pr_fmt
1321 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1322 
1323 static void task_update_spec_tif(struct task_struct *tsk)
1324 {
1325 	/* Force the update of the real TIF bits */
1326 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1327 
1328 	/*
1329 	 * Immediately update the speculation control MSRs for the current
1330 	 * task, but for a non-current task delay setting the CPU
1331 	 * mitigation until it is scheduled next.
1332 	 *
1333 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1334 	 * always the current task.
1335 	 */
1336 	if (tsk == current)
1337 		speculation_ctrl_update_current();
1338 }
1339 
1340 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1341 {
1342 
1343 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1344 		return -EPERM;
1345 
1346 	switch (ctrl) {
1347 	case PR_SPEC_ENABLE:
1348 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1349 		return 0;
1350 	case PR_SPEC_DISABLE:
1351 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1352 		return 0;
1353 	default:
1354 		return -ERANGE;
1355 	}
1356 }
1357 
1358 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1359 {
1360 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1361 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1362 		return -ENXIO;
1363 
1364 	switch (ctrl) {
1365 	case PR_SPEC_ENABLE:
1366 		/* If speculation is force disabled, enable is not allowed */
1367 		if (task_spec_ssb_force_disable(task))
1368 			return -EPERM;
1369 		task_clear_spec_ssb_disable(task);
1370 		task_clear_spec_ssb_noexec(task);
1371 		task_update_spec_tif(task);
1372 		break;
1373 	case PR_SPEC_DISABLE:
1374 		task_set_spec_ssb_disable(task);
1375 		task_clear_spec_ssb_noexec(task);
1376 		task_update_spec_tif(task);
1377 		break;
1378 	case PR_SPEC_FORCE_DISABLE:
1379 		task_set_spec_ssb_disable(task);
1380 		task_set_spec_ssb_force_disable(task);
1381 		task_clear_spec_ssb_noexec(task);
1382 		task_update_spec_tif(task);
1383 		break;
1384 	case PR_SPEC_DISABLE_NOEXEC:
1385 		if (task_spec_ssb_force_disable(task))
1386 			return -EPERM;
1387 		task_set_spec_ssb_disable(task);
1388 		task_set_spec_ssb_noexec(task);
1389 		task_update_spec_tif(task);
1390 		break;
1391 	default:
1392 		return -ERANGE;
1393 	}
1394 	return 0;
1395 }
1396 
1397 static bool is_spec_ib_user_controlled(void)
1398 {
1399 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1400 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1401 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1402 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1403 }
1404 
1405 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1406 {
1407 	switch (ctrl) {
1408 	case PR_SPEC_ENABLE:
1409 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1410 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1411 			return 0;
1412 
1413 		/*
1414 		 * With strict mode for both IBPB and STIBP, the instruction
1415 		 * code paths avoid checking this task flag and instead,
1416 		 * unconditionally run the instruction. However, STIBP and IBPB
1417 		 * are independent and either can be set to conditionally
1418 		 * enabled regardless of the mode of the other.
1419 		 *
1420 		 * If either is set to conditional, allow the task flag to be
1421 		 * updated, unless it was force-disabled by a previous prctl
1422 		 * call. Currently, this is possible on an AMD CPU which has the
1423 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1424 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1425 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1426 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1427 		 */
1428 		if (!is_spec_ib_user_controlled() ||
1429 		    task_spec_ib_force_disable(task))
1430 			return -EPERM;
1431 
1432 		task_clear_spec_ib_disable(task);
1433 		task_update_spec_tif(task);
1434 		break;
1435 	case PR_SPEC_DISABLE:
1436 	case PR_SPEC_FORCE_DISABLE:
1437 		/*
1438 		 * Indirect branch speculation is always allowed when
1439 		 * mitigation is force disabled.
1440 		 */
1441 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1442 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1443 			return -EPERM;
1444 
1445 		if (!is_spec_ib_user_controlled())
1446 			return 0;
1447 
1448 		task_set_spec_ib_disable(task);
1449 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1450 			task_set_spec_ib_force_disable(task);
1451 		task_update_spec_tif(task);
1452 		break;
1453 	default:
1454 		return -ERANGE;
1455 	}
1456 	return 0;
1457 }
1458 
1459 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1460 			     unsigned long ctrl)
1461 {
1462 	switch (which) {
1463 	case PR_SPEC_STORE_BYPASS:
1464 		return ssb_prctl_set(task, ctrl);
1465 	case PR_SPEC_INDIRECT_BRANCH:
1466 		return ib_prctl_set(task, ctrl);
1467 	case PR_SPEC_L1D_FLUSH:
1468 		return l1d_flush_prctl_set(task, ctrl);
1469 	default:
1470 		return -ENODEV;
1471 	}
1472 }
1473 
1474 #ifdef CONFIG_SECCOMP
1475 void arch_seccomp_spec_mitigate(struct task_struct *task)
1476 {
1477 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1478 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1479 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1480 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1481 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1482 }
1483 #endif
1484 
1485 static int l1d_flush_prctl_get(struct task_struct *task)
1486 {
1487 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1488 		return PR_SPEC_FORCE_DISABLE;
1489 
1490 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
1491 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1492 	else
1493 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1494 }
1495 
1496 static int ssb_prctl_get(struct task_struct *task)
1497 {
1498 	switch (ssb_mode) {
1499 	case SPEC_STORE_BYPASS_DISABLE:
1500 		return PR_SPEC_DISABLE;
1501 	case SPEC_STORE_BYPASS_SECCOMP:
1502 	case SPEC_STORE_BYPASS_PRCTL:
1503 		if (task_spec_ssb_force_disable(task))
1504 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1505 		if (task_spec_ssb_noexec(task))
1506 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
1507 		if (task_spec_ssb_disable(task))
1508 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1509 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1510 	default:
1511 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1512 			return PR_SPEC_ENABLE;
1513 		return PR_SPEC_NOT_AFFECTED;
1514 	}
1515 }
1516 
1517 static int ib_prctl_get(struct task_struct *task)
1518 {
1519 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
1520 		return PR_SPEC_NOT_AFFECTED;
1521 
1522 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1523 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1524 		return PR_SPEC_ENABLE;
1525 	else if (is_spec_ib_user_controlled()) {
1526 		if (task_spec_ib_force_disable(task))
1527 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1528 		if (task_spec_ib_disable(task))
1529 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1530 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1531 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
1532 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
1533 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
1534 		return PR_SPEC_DISABLE;
1535 	else
1536 		return PR_SPEC_NOT_AFFECTED;
1537 }
1538 
1539 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
1540 {
1541 	switch (which) {
1542 	case PR_SPEC_STORE_BYPASS:
1543 		return ssb_prctl_get(task);
1544 	case PR_SPEC_INDIRECT_BRANCH:
1545 		return ib_prctl_get(task);
1546 	case PR_SPEC_L1D_FLUSH:
1547 		return l1d_flush_prctl_get(task);
1548 	default:
1549 		return -ENODEV;
1550 	}
1551 }
1552 
1553 void x86_spec_ctrl_setup_ap(void)
1554 {
1555 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
1556 		wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1557 
1558 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
1559 		x86_amd_ssb_disable();
1560 }
1561 
1562 bool itlb_multihit_kvm_mitigation;
1563 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
1564 
1565 #undef pr_fmt
1566 #define pr_fmt(fmt)	"L1TF: " fmt
1567 
1568 /* Default mitigation for L1TF-affected CPUs */
1569 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
1570 #if IS_ENABLED(CONFIG_KVM_INTEL)
1571 EXPORT_SYMBOL_GPL(l1tf_mitigation);
1572 #endif
1573 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
1574 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
1575 
1576 /*
1577  * These CPUs all support 44bits physical address space internally in the
1578  * cache but CPUID can report a smaller number of physical address bits.
1579  *
1580  * The L1TF mitigation uses the top most address bit for the inversion of
1581  * non present PTEs. When the installed memory reaches into the top most
1582  * address bit due to memory holes, which has been observed on machines
1583  * which report 36bits physical address bits and have 32G RAM installed,
1584  * then the mitigation range check in l1tf_select_mitigation() triggers.
1585  * This is a false positive because the mitigation is still possible due to
1586  * the fact that the cache uses 44bit internally. Use the cache bits
1587  * instead of the reported physical bits and adjust them on the affected
1588  * machines to 44bit if the reported bits are less than 44.
1589  */
1590 static void override_cache_bits(struct cpuinfo_x86 *c)
1591 {
1592 	if (c->x86 != 6)
1593 		return;
1594 
1595 	switch (c->x86_model) {
1596 	case INTEL_FAM6_NEHALEM:
1597 	case INTEL_FAM6_WESTMERE:
1598 	case INTEL_FAM6_SANDYBRIDGE:
1599 	case INTEL_FAM6_IVYBRIDGE:
1600 	case INTEL_FAM6_HASWELL:
1601 	case INTEL_FAM6_HASWELL_L:
1602 	case INTEL_FAM6_HASWELL_G:
1603 	case INTEL_FAM6_BROADWELL:
1604 	case INTEL_FAM6_BROADWELL_G:
1605 	case INTEL_FAM6_SKYLAKE_L:
1606 	case INTEL_FAM6_SKYLAKE:
1607 	case INTEL_FAM6_KABYLAKE_L:
1608 	case INTEL_FAM6_KABYLAKE:
1609 		if (c->x86_cache_bits < 44)
1610 			c->x86_cache_bits = 44;
1611 		break;
1612 	}
1613 }
1614 
1615 static void __init l1tf_select_mitigation(void)
1616 {
1617 	u64 half_pa;
1618 
1619 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
1620 		return;
1621 
1622 	if (cpu_mitigations_off())
1623 		l1tf_mitigation = L1TF_MITIGATION_OFF;
1624 	else if (cpu_mitigations_auto_nosmt())
1625 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
1626 
1627 	override_cache_bits(&boot_cpu_data);
1628 
1629 	switch (l1tf_mitigation) {
1630 	case L1TF_MITIGATION_OFF:
1631 	case L1TF_MITIGATION_FLUSH_NOWARN:
1632 	case L1TF_MITIGATION_FLUSH:
1633 		break;
1634 	case L1TF_MITIGATION_FLUSH_NOSMT:
1635 	case L1TF_MITIGATION_FULL:
1636 		cpu_smt_disable(false);
1637 		break;
1638 	case L1TF_MITIGATION_FULL_FORCE:
1639 		cpu_smt_disable(true);
1640 		break;
1641 	}
1642 
1643 #if CONFIG_PGTABLE_LEVELS == 2
1644 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
1645 	return;
1646 #endif
1647 
1648 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
1649 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
1650 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
1651 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
1652 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
1653 				half_pa);
1654 		pr_info("However, doing so will make a part of your RAM unusable.\n");
1655 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
1656 		return;
1657 	}
1658 
1659 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
1660 }
1661 
1662 static int __init l1tf_cmdline(char *str)
1663 {
1664 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
1665 		return 0;
1666 
1667 	if (!str)
1668 		return -EINVAL;
1669 
1670 	if (!strcmp(str, "off"))
1671 		l1tf_mitigation = L1TF_MITIGATION_OFF;
1672 	else if (!strcmp(str, "flush,nowarn"))
1673 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
1674 	else if (!strcmp(str, "flush"))
1675 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
1676 	else if (!strcmp(str, "flush,nosmt"))
1677 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
1678 	else if (!strcmp(str, "full"))
1679 		l1tf_mitigation = L1TF_MITIGATION_FULL;
1680 	else if (!strcmp(str, "full,force"))
1681 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
1682 
1683 	return 0;
1684 }
1685 early_param("l1tf", l1tf_cmdline);
1686 
1687 #undef pr_fmt
1688 #define pr_fmt(fmt) fmt
1689 
1690 #ifdef CONFIG_SYSFS
1691 
1692 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
1693 
1694 #if IS_ENABLED(CONFIG_KVM_INTEL)
1695 static const char * const l1tf_vmx_states[] = {
1696 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
1697 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
1698 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
1699 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
1700 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
1701 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
1702 };
1703 
1704 static ssize_t l1tf_show_state(char *buf)
1705 {
1706 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
1707 		return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
1708 
1709 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
1710 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
1711 	     sched_smt_active())) {
1712 		return sprintf(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
1713 			       l1tf_vmx_states[l1tf_vmx_mitigation]);
1714 	}
1715 
1716 	return sprintf(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
1717 		       l1tf_vmx_states[l1tf_vmx_mitigation],
1718 		       sched_smt_active() ? "vulnerable" : "disabled");
1719 }
1720 
1721 static ssize_t itlb_multihit_show_state(char *buf)
1722 {
1723 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
1724 	    !boot_cpu_has(X86_FEATURE_VMX))
1725 		return sprintf(buf, "KVM: Mitigation: VMX unsupported\n");
1726 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
1727 		return sprintf(buf, "KVM: Mitigation: VMX disabled\n");
1728 	else if (itlb_multihit_kvm_mitigation)
1729 		return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
1730 	else
1731 		return sprintf(buf, "KVM: Vulnerable\n");
1732 }
1733 #else
1734 static ssize_t l1tf_show_state(char *buf)
1735 {
1736 	return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
1737 }
1738 
1739 static ssize_t itlb_multihit_show_state(char *buf)
1740 {
1741 	return sprintf(buf, "Processor vulnerable\n");
1742 }
1743 #endif
1744 
1745 static ssize_t mds_show_state(char *buf)
1746 {
1747 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1748 		return sprintf(buf, "%s; SMT Host state unknown\n",
1749 			       mds_strings[mds_mitigation]);
1750 	}
1751 
1752 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
1753 		return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
1754 			       (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
1755 			        sched_smt_active() ? "mitigated" : "disabled"));
1756 	}
1757 
1758 	return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
1759 		       sched_smt_active() ? "vulnerable" : "disabled");
1760 }
1761 
1762 static ssize_t tsx_async_abort_show_state(char *buf)
1763 {
1764 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
1765 	    (taa_mitigation == TAA_MITIGATION_OFF))
1766 		return sprintf(buf, "%s\n", taa_strings[taa_mitigation]);
1767 
1768 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1769 		return sprintf(buf, "%s; SMT Host state unknown\n",
1770 			       taa_strings[taa_mitigation]);
1771 	}
1772 
1773 	return sprintf(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
1774 		       sched_smt_active() ? "vulnerable" : "disabled");
1775 }
1776 
1777 static char *stibp_state(void)
1778 {
1779 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled))
1780 		return "";
1781 
1782 	switch (spectre_v2_user_stibp) {
1783 	case SPECTRE_V2_USER_NONE:
1784 		return ", STIBP: disabled";
1785 	case SPECTRE_V2_USER_STRICT:
1786 		return ", STIBP: forced";
1787 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1788 		return ", STIBP: always-on";
1789 	case SPECTRE_V2_USER_PRCTL:
1790 	case SPECTRE_V2_USER_SECCOMP:
1791 		if (static_key_enabled(&switch_to_cond_stibp))
1792 			return ", STIBP: conditional";
1793 	}
1794 	return "";
1795 }
1796 
1797 static char *ibpb_state(void)
1798 {
1799 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1800 		if (static_key_enabled(&switch_mm_always_ibpb))
1801 			return ", IBPB: always-on";
1802 		if (static_key_enabled(&switch_mm_cond_ibpb))
1803 			return ", IBPB: conditional";
1804 		return ", IBPB: disabled";
1805 	}
1806 	return "";
1807 }
1808 
1809 static ssize_t spectre_v2_show_state(char *buf)
1810 {
1811 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
1812 		return sprintf(buf, "Vulnerable: LFENCE\n");
1813 
1814 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1815 		return sprintf(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
1816 
1817 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1818 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1819 		return sprintf(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
1820 
1821 	return sprintf(buf, "%s%s%s%s%s%s\n",
1822 		       spectre_v2_strings[spectre_v2_enabled],
1823 		       ibpb_state(),
1824 		       boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
1825 		       stibp_state(),
1826 		       boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
1827 		       spectre_v2_module_string());
1828 }
1829 
1830 static ssize_t srbds_show_state(char *buf)
1831 {
1832 	return sprintf(buf, "%s\n", srbds_strings[srbds_mitigation]);
1833 }
1834 
1835 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
1836 			       char *buf, unsigned int bug)
1837 {
1838 	if (!boot_cpu_has_bug(bug))
1839 		return sprintf(buf, "Not affected\n");
1840 
1841 	switch (bug) {
1842 	case X86_BUG_CPU_MELTDOWN:
1843 		if (boot_cpu_has(X86_FEATURE_PTI))
1844 			return sprintf(buf, "Mitigation: PTI\n");
1845 
1846 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
1847 			return sprintf(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
1848 
1849 		break;
1850 
1851 	case X86_BUG_SPECTRE_V1:
1852 		return sprintf(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
1853 
1854 	case X86_BUG_SPECTRE_V2:
1855 		return spectre_v2_show_state(buf);
1856 
1857 	case X86_BUG_SPEC_STORE_BYPASS:
1858 		return sprintf(buf, "%s\n", ssb_strings[ssb_mode]);
1859 
1860 	case X86_BUG_L1TF:
1861 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
1862 			return l1tf_show_state(buf);
1863 		break;
1864 
1865 	case X86_BUG_MDS:
1866 		return mds_show_state(buf);
1867 
1868 	case X86_BUG_TAA:
1869 		return tsx_async_abort_show_state(buf);
1870 
1871 	case X86_BUG_ITLB_MULTIHIT:
1872 		return itlb_multihit_show_state(buf);
1873 
1874 	case X86_BUG_SRBDS:
1875 		return srbds_show_state(buf);
1876 
1877 	default:
1878 		break;
1879 	}
1880 
1881 	return sprintf(buf, "Vulnerable\n");
1882 }
1883 
1884 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
1885 {
1886 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
1887 }
1888 
1889 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
1890 {
1891 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
1892 }
1893 
1894 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
1895 {
1896 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
1897 }
1898 
1899 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
1900 {
1901 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
1902 }
1903 
1904 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
1905 {
1906 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
1907 }
1908 
1909 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
1910 {
1911 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
1912 }
1913 
1914 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
1915 {
1916 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
1917 }
1918 
1919 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
1920 {
1921 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
1922 }
1923 
1924 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
1925 {
1926 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
1927 }
1928 #endif
1929