xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision af8f260a)
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 	/*
450 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
451 	 * being disabled and it hasn't received the SRBDS MSR microcode.
452 	 */
453 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
454 		return;
455 
456 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
457 
458 	switch (srbds_mitigation) {
459 	case SRBDS_MITIGATION_OFF:
460 	case SRBDS_MITIGATION_TSX_OFF:
461 		mcu_ctrl |= RNGDS_MITG_DIS;
462 		break;
463 	case SRBDS_MITIGATION_FULL:
464 		mcu_ctrl &= ~RNGDS_MITG_DIS;
465 		break;
466 	default:
467 		break;
468 	}
469 
470 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
471 }
472 
473 static void __init srbds_select_mitigation(void)
474 {
475 	u64 ia32_cap;
476 
477 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
478 		return;
479 
480 	/*
481 	 * Check to see if this is one of the MDS_NO systems supporting
482 	 * TSX that are only exposed to SRBDS when TSX is enabled.
483 	 */
484 	ia32_cap = x86_read_arch_cap_msr();
485 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM))
486 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
487 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
488 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
489 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
490 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
491 	else if (cpu_mitigations_off() || srbds_off)
492 		srbds_mitigation = SRBDS_MITIGATION_OFF;
493 
494 	update_srbds_msr();
495 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
496 }
497 
498 static int __init srbds_parse_cmdline(char *str)
499 {
500 	if (!str)
501 		return -EINVAL;
502 
503 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
504 		return 0;
505 
506 	srbds_off = !strcmp(str, "off");
507 	return 0;
508 }
509 early_param("srbds", srbds_parse_cmdline);
510 
511 #undef pr_fmt
512 #define pr_fmt(fmt)     "L1D Flush : " fmt
513 
514 enum l1d_flush_mitigations {
515 	L1D_FLUSH_OFF = 0,
516 	L1D_FLUSH_ON,
517 };
518 
519 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
520 
521 static void __init l1d_flush_select_mitigation(void)
522 {
523 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
524 		return;
525 
526 	static_branch_enable(&switch_mm_cond_l1d_flush);
527 	pr_info("Conditional flush on switch_mm() enabled\n");
528 }
529 
530 static int __init l1d_flush_parse_cmdline(char *str)
531 {
532 	if (!strcmp(str, "on"))
533 		l1d_flush_mitigation = L1D_FLUSH_ON;
534 
535 	return 0;
536 }
537 early_param("l1d_flush", l1d_flush_parse_cmdline);
538 
539 #undef pr_fmt
540 #define pr_fmt(fmt)     "Spectre V1 : " fmt
541 
542 enum spectre_v1_mitigation {
543 	SPECTRE_V1_MITIGATION_NONE,
544 	SPECTRE_V1_MITIGATION_AUTO,
545 };
546 
547 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
548 	SPECTRE_V1_MITIGATION_AUTO;
549 
550 static const char * const spectre_v1_strings[] = {
551 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
552 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
553 };
554 
555 /*
556  * Does SMAP provide full mitigation against speculative kernel access to
557  * userspace?
558  */
559 static bool smap_works_speculatively(void)
560 {
561 	if (!boot_cpu_has(X86_FEATURE_SMAP))
562 		return false;
563 
564 	/*
565 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
566 	 * prevent speculative access to user data in the L1 cache.
567 	 * Consider SMAP to be non-functional as a mitigation on these
568 	 * CPUs.
569 	 */
570 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
571 		return false;
572 
573 	return true;
574 }
575 
576 static void __init spectre_v1_select_mitigation(void)
577 {
578 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
579 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
580 		return;
581 	}
582 
583 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
584 		/*
585 		 * With Spectre v1, a user can speculatively control either
586 		 * path of a conditional swapgs with a user-controlled GS
587 		 * value.  The mitigation is to add lfences to both code paths.
588 		 *
589 		 * If FSGSBASE is enabled, the user can put a kernel address in
590 		 * GS, in which case SMAP provides no protection.
591 		 *
592 		 * If FSGSBASE is disabled, the user can only put a user space
593 		 * address in GS.  That makes an attack harder, but still
594 		 * possible if there's no SMAP protection.
595 		 */
596 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
597 		    !smap_works_speculatively()) {
598 			/*
599 			 * Mitigation can be provided from SWAPGS itself or
600 			 * PTI as the CR3 write in the Meltdown mitigation
601 			 * is serializing.
602 			 *
603 			 * If neither is there, mitigate with an LFENCE to
604 			 * stop speculation through swapgs.
605 			 */
606 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
607 			    !boot_cpu_has(X86_FEATURE_PTI))
608 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
609 
610 			/*
611 			 * Enable lfences in the kernel entry (non-swapgs)
612 			 * paths, to prevent user entry from speculatively
613 			 * skipping swapgs.
614 			 */
615 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
616 		}
617 	}
618 
619 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
620 }
621 
622 static int __init nospectre_v1_cmdline(char *str)
623 {
624 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
625 	return 0;
626 }
627 early_param("nospectre_v1", nospectre_v1_cmdline);
628 
629 #undef pr_fmt
630 #define pr_fmt(fmt)     "Spectre V2 : " fmt
631 
632 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
633 	SPECTRE_V2_NONE;
634 
635 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
636 	SPECTRE_V2_USER_NONE;
637 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
638 	SPECTRE_V2_USER_NONE;
639 
640 #ifdef CONFIG_RETPOLINE
641 static bool spectre_v2_bad_module;
642 
643 bool retpoline_module_ok(bool has_retpoline)
644 {
645 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
646 		return true;
647 
648 	pr_err("System may be vulnerable to spectre v2\n");
649 	spectre_v2_bad_module = true;
650 	return false;
651 }
652 
653 static inline const char *spectre_v2_module_string(void)
654 {
655 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
656 }
657 #else
658 static inline const char *spectre_v2_module_string(void) { return ""; }
659 #endif
660 
661 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
662 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
663 #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"
664 
665 #ifdef CONFIG_BPF_SYSCALL
666 void unpriv_ebpf_notify(int new_state)
667 {
668 	if (new_state)
669 		return;
670 
671 	/* Unprivileged eBPF is enabled */
672 
673 	switch (spectre_v2_enabled) {
674 	case SPECTRE_V2_EIBRS:
675 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
676 		break;
677 	case SPECTRE_V2_EIBRS_LFENCE:
678 		if (sched_smt_active())
679 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
680 		break;
681 	default:
682 		break;
683 	}
684 }
685 #endif
686 
687 static inline bool match_option(const char *arg, int arglen, const char *opt)
688 {
689 	int len = strlen(opt);
690 
691 	return len == arglen && !strncmp(arg, opt, len);
692 }
693 
694 /* The kernel command line selection for spectre v2 */
695 enum spectre_v2_mitigation_cmd {
696 	SPECTRE_V2_CMD_NONE,
697 	SPECTRE_V2_CMD_AUTO,
698 	SPECTRE_V2_CMD_FORCE,
699 	SPECTRE_V2_CMD_RETPOLINE,
700 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
701 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
702 	SPECTRE_V2_CMD_EIBRS,
703 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
704 	SPECTRE_V2_CMD_EIBRS_LFENCE,
705 };
706 
707 enum spectre_v2_user_cmd {
708 	SPECTRE_V2_USER_CMD_NONE,
709 	SPECTRE_V2_USER_CMD_AUTO,
710 	SPECTRE_V2_USER_CMD_FORCE,
711 	SPECTRE_V2_USER_CMD_PRCTL,
712 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
713 	SPECTRE_V2_USER_CMD_SECCOMP,
714 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
715 };
716 
717 static const char * const spectre_v2_user_strings[] = {
718 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
719 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
720 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
721 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
722 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
723 };
724 
725 static const struct {
726 	const char			*option;
727 	enum spectre_v2_user_cmd	cmd;
728 	bool				secure;
729 } v2_user_options[] __initconst = {
730 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
731 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
732 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
733 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
734 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
735 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
736 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
737 };
738 
739 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
740 {
741 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
742 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
743 }
744 
745 static enum spectre_v2_user_cmd __init
746 spectre_v2_parse_user_cmdline(enum spectre_v2_mitigation_cmd v2_cmd)
747 {
748 	char arg[20];
749 	int ret, i;
750 
751 	switch (v2_cmd) {
752 	case SPECTRE_V2_CMD_NONE:
753 		return SPECTRE_V2_USER_CMD_NONE;
754 	case SPECTRE_V2_CMD_FORCE:
755 		return SPECTRE_V2_USER_CMD_FORCE;
756 	default:
757 		break;
758 	}
759 
760 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
761 				  arg, sizeof(arg));
762 	if (ret < 0)
763 		return SPECTRE_V2_USER_CMD_AUTO;
764 
765 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
766 		if (match_option(arg, ret, v2_user_options[i].option)) {
767 			spec_v2_user_print_cond(v2_user_options[i].option,
768 						v2_user_options[i].secure);
769 			return v2_user_options[i].cmd;
770 		}
771 	}
772 
773 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
774 	return SPECTRE_V2_USER_CMD_AUTO;
775 }
776 
777 static inline bool spectre_v2_in_eibrs_mode(enum spectre_v2_mitigation mode)
778 {
779 	return (mode == SPECTRE_V2_EIBRS ||
780 		mode == SPECTRE_V2_EIBRS_RETPOLINE ||
781 		mode == SPECTRE_V2_EIBRS_LFENCE);
782 }
783 
784 static void __init
785 spectre_v2_user_select_mitigation(enum spectre_v2_mitigation_cmd v2_cmd)
786 {
787 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
788 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
789 	enum spectre_v2_user_cmd cmd;
790 
791 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
792 		return;
793 
794 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
795 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
796 		smt_possible = false;
797 
798 	cmd = spectre_v2_parse_user_cmdline(v2_cmd);
799 	switch (cmd) {
800 	case SPECTRE_V2_USER_CMD_NONE:
801 		goto set_mode;
802 	case SPECTRE_V2_USER_CMD_FORCE:
803 		mode = SPECTRE_V2_USER_STRICT;
804 		break;
805 	case SPECTRE_V2_USER_CMD_AUTO:
806 	case SPECTRE_V2_USER_CMD_PRCTL:
807 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
808 		mode = SPECTRE_V2_USER_PRCTL;
809 		break;
810 	case SPECTRE_V2_USER_CMD_SECCOMP:
811 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
812 		if (IS_ENABLED(CONFIG_SECCOMP))
813 			mode = SPECTRE_V2_USER_SECCOMP;
814 		else
815 			mode = SPECTRE_V2_USER_PRCTL;
816 		break;
817 	}
818 
819 	/* Initialize Indirect Branch Prediction Barrier */
820 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
821 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
822 
823 		spectre_v2_user_ibpb = mode;
824 		switch (cmd) {
825 		case SPECTRE_V2_USER_CMD_FORCE:
826 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
827 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
828 			static_branch_enable(&switch_mm_always_ibpb);
829 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
830 			break;
831 		case SPECTRE_V2_USER_CMD_PRCTL:
832 		case SPECTRE_V2_USER_CMD_AUTO:
833 		case SPECTRE_V2_USER_CMD_SECCOMP:
834 			static_branch_enable(&switch_mm_cond_ibpb);
835 			break;
836 		default:
837 			break;
838 		}
839 
840 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
841 			static_key_enabled(&switch_mm_always_ibpb) ?
842 			"always-on" : "conditional");
843 	}
844 
845 	/*
846 	 * If no STIBP, enhanced IBRS is enabled or SMT impossible, STIBP is not
847 	 * required.
848 	 */
849 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
850 	    !smt_possible ||
851 	    spectre_v2_in_eibrs_mode(spectre_v2_enabled))
852 		return;
853 
854 	/*
855 	 * At this point, an STIBP mode other than "off" has been set.
856 	 * If STIBP support is not being forced, check if STIBP always-on
857 	 * is preferred.
858 	 */
859 	if (mode != SPECTRE_V2_USER_STRICT &&
860 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
861 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
862 
863 	spectre_v2_user_stibp = mode;
864 
865 set_mode:
866 	pr_info("%s\n", spectre_v2_user_strings[mode]);
867 }
868 
869 static const char * const spectre_v2_strings[] = {
870 	[SPECTRE_V2_NONE]			= "Vulnerable",
871 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
872 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
873 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced IBRS",
874 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced IBRS + LFENCE",
875 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced IBRS + Retpolines",
876 };
877 
878 static const struct {
879 	const char *option;
880 	enum spectre_v2_mitigation_cmd cmd;
881 	bool secure;
882 } mitigation_options[] __initconst = {
883 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
884 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
885 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
886 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
887 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
888 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
889 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
890 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
891 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
892 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
893 };
894 
895 static void __init spec_v2_print_cond(const char *reason, bool secure)
896 {
897 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
898 		pr_info("%s selected on command line.\n", reason);
899 }
900 
901 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
902 {
903 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
904 	char arg[20];
905 	int ret, i;
906 
907 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
908 	    cpu_mitigations_off())
909 		return SPECTRE_V2_CMD_NONE;
910 
911 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
912 	if (ret < 0)
913 		return SPECTRE_V2_CMD_AUTO;
914 
915 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
916 		if (!match_option(arg, ret, mitigation_options[i].option))
917 			continue;
918 		cmd = mitigation_options[i].cmd;
919 		break;
920 	}
921 
922 	if (i >= ARRAY_SIZE(mitigation_options)) {
923 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
924 		return SPECTRE_V2_CMD_AUTO;
925 	}
926 
927 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
928 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
929 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
930 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
931 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
932 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
933 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
934 		       mitigation_options[i].option);
935 		return SPECTRE_V2_CMD_AUTO;
936 	}
937 
938 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
939 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
940 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
941 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
942 		pr_err("%s selected but CPU doesn't have eIBRS. Switching to AUTO select\n",
943 		       mitigation_options[i].option);
944 		return SPECTRE_V2_CMD_AUTO;
945 	}
946 
947 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
948 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
949 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
950 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
951 		       mitigation_options[i].option);
952 		return SPECTRE_V2_CMD_AUTO;
953 	}
954 
955 	spec_v2_print_cond(mitigation_options[i].option,
956 			   mitigation_options[i].secure);
957 	return cmd;
958 }
959 
960 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
961 {
962 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
963 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
964 		return SPECTRE_V2_NONE;
965 	}
966 
967 	return SPECTRE_V2_RETPOLINE;
968 }
969 
970 static void __init spectre_v2_select_mitigation(void)
971 {
972 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
973 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
974 
975 	/*
976 	 * If the CPU is not affected and the command line mode is NONE or AUTO
977 	 * then nothing to do.
978 	 */
979 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
980 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
981 		return;
982 
983 	switch (cmd) {
984 	case SPECTRE_V2_CMD_NONE:
985 		return;
986 
987 	case SPECTRE_V2_CMD_FORCE:
988 	case SPECTRE_V2_CMD_AUTO:
989 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
990 			mode = SPECTRE_V2_EIBRS;
991 			break;
992 		}
993 
994 		mode = spectre_v2_select_retpoline();
995 		break;
996 
997 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
998 		pr_err(SPECTRE_V2_LFENCE_MSG);
999 		mode = SPECTRE_V2_LFENCE;
1000 		break;
1001 
1002 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1003 		mode = SPECTRE_V2_RETPOLINE;
1004 		break;
1005 
1006 	case SPECTRE_V2_CMD_RETPOLINE:
1007 		mode = spectre_v2_select_retpoline();
1008 		break;
1009 
1010 	case SPECTRE_V2_CMD_EIBRS:
1011 		mode = SPECTRE_V2_EIBRS;
1012 		break;
1013 
1014 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1015 		mode = SPECTRE_V2_EIBRS_LFENCE;
1016 		break;
1017 
1018 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1019 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1020 		break;
1021 	}
1022 
1023 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1024 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1025 
1026 	if (spectre_v2_in_eibrs_mode(mode)) {
1027 		/* Force it so VMEXIT will restore correctly */
1028 		x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1029 		wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1030 	}
1031 
1032 	switch (mode) {
1033 	case SPECTRE_V2_NONE:
1034 	case SPECTRE_V2_EIBRS:
1035 		break;
1036 
1037 	case SPECTRE_V2_LFENCE:
1038 	case SPECTRE_V2_EIBRS_LFENCE:
1039 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1040 		fallthrough;
1041 
1042 	case SPECTRE_V2_RETPOLINE:
1043 	case SPECTRE_V2_EIBRS_RETPOLINE:
1044 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1045 		break;
1046 	}
1047 
1048 	spectre_v2_enabled = mode;
1049 	pr_info("%s\n", spectre_v2_strings[mode]);
1050 
1051 	/*
1052 	 * If spectre v2 protection has been enabled, unconditionally fill
1053 	 * RSB during a context switch; this protects against two independent
1054 	 * issues:
1055 	 *
1056 	 *	- RSB underflow (and switch to BTB) on Skylake+
1057 	 *	- SpectreRSB variant of spectre v2 on X86_BUG_SPECTRE_V2 CPUs
1058 	 */
1059 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1060 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1061 
1062 	/*
1063 	 * Retpoline means the kernel is safe because it has no indirect
1064 	 * branches. Enhanced IBRS protects firmware too, so, enable restricted
1065 	 * speculation around firmware calls only when Enhanced IBRS isn't
1066 	 * supported.
1067 	 *
1068 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1069 	 * the user might select retpoline on the kernel command line and if
1070 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1071 	 * enable IBRS around firmware calls.
1072 	 */
1073 	if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_eibrs_mode(mode)) {
1074 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1075 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1076 	}
1077 
1078 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1079 	spectre_v2_user_select_mitigation(cmd);
1080 }
1081 
1082 static void update_stibp_msr(void * __unused)
1083 {
1084 	wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1085 }
1086 
1087 /* Update x86_spec_ctrl_base in case SMT state changed. */
1088 static void update_stibp_strict(void)
1089 {
1090 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1091 
1092 	if (sched_smt_active())
1093 		mask |= SPEC_CTRL_STIBP;
1094 
1095 	if (mask == x86_spec_ctrl_base)
1096 		return;
1097 
1098 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1099 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1100 	x86_spec_ctrl_base = mask;
1101 	on_each_cpu(update_stibp_msr, NULL, 1);
1102 }
1103 
1104 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1105 static void update_indir_branch_cond(void)
1106 {
1107 	if (sched_smt_active())
1108 		static_branch_enable(&switch_to_cond_stibp);
1109 	else
1110 		static_branch_disable(&switch_to_cond_stibp);
1111 }
1112 
1113 #undef pr_fmt
1114 #define pr_fmt(fmt) fmt
1115 
1116 /* Update the static key controlling the MDS CPU buffer clear in idle */
1117 static void update_mds_branch_idle(void)
1118 {
1119 	/*
1120 	 * Enable the idle clearing if SMT is active on CPUs which are
1121 	 * affected only by MSBDS and not any other MDS variant.
1122 	 *
1123 	 * The other variants cannot be mitigated when SMT is enabled, so
1124 	 * clearing the buffers on idle just to prevent the Store Buffer
1125 	 * repartitioning leak would be a window dressing exercise.
1126 	 */
1127 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1128 		return;
1129 
1130 	if (sched_smt_active())
1131 		static_branch_enable(&mds_idle_clear);
1132 	else
1133 		static_branch_disable(&mds_idle_clear);
1134 }
1135 
1136 #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"
1137 #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"
1138 
1139 void cpu_bugs_smt_update(void)
1140 {
1141 	mutex_lock(&spec_ctrl_mutex);
1142 
1143 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1144 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1145 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1146 
1147 	switch (spectre_v2_user_stibp) {
1148 	case SPECTRE_V2_USER_NONE:
1149 		break;
1150 	case SPECTRE_V2_USER_STRICT:
1151 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1152 		update_stibp_strict();
1153 		break;
1154 	case SPECTRE_V2_USER_PRCTL:
1155 	case SPECTRE_V2_USER_SECCOMP:
1156 		update_indir_branch_cond();
1157 		break;
1158 	}
1159 
1160 	switch (mds_mitigation) {
1161 	case MDS_MITIGATION_FULL:
1162 	case MDS_MITIGATION_VMWERV:
1163 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1164 			pr_warn_once(MDS_MSG_SMT);
1165 		update_mds_branch_idle();
1166 		break;
1167 	case MDS_MITIGATION_OFF:
1168 		break;
1169 	}
1170 
1171 	switch (taa_mitigation) {
1172 	case TAA_MITIGATION_VERW:
1173 	case TAA_MITIGATION_UCODE_NEEDED:
1174 		if (sched_smt_active())
1175 			pr_warn_once(TAA_MSG_SMT);
1176 		break;
1177 	case TAA_MITIGATION_TSX_DISABLED:
1178 	case TAA_MITIGATION_OFF:
1179 		break;
1180 	}
1181 
1182 	mutex_unlock(&spec_ctrl_mutex);
1183 }
1184 
1185 #undef pr_fmt
1186 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1187 
1188 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1189 
1190 /* The kernel command line selection */
1191 enum ssb_mitigation_cmd {
1192 	SPEC_STORE_BYPASS_CMD_NONE,
1193 	SPEC_STORE_BYPASS_CMD_AUTO,
1194 	SPEC_STORE_BYPASS_CMD_ON,
1195 	SPEC_STORE_BYPASS_CMD_PRCTL,
1196 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1197 };
1198 
1199 static const char * const ssb_strings[] = {
1200 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1201 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1202 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1203 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1204 };
1205 
1206 static const struct {
1207 	const char *option;
1208 	enum ssb_mitigation_cmd cmd;
1209 } ssb_mitigation_options[]  __initconst = {
1210 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1211 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1212 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1213 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1214 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1215 };
1216 
1217 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1218 {
1219 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1220 	char arg[20];
1221 	int ret, i;
1222 
1223 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1224 	    cpu_mitigations_off()) {
1225 		return SPEC_STORE_BYPASS_CMD_NONE;
1226 	} else {
1227 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1228 					  arg, sizeof(arg));
1229 		if (ret < 0)
1230 			return SPEC_STORE_BYPASS_CMD_AUTO;
1231 
1232 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1233 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1234 				continue;
1235 
1236 			cmd = ssb_mitigation_options[i].cmd;
1237 			break;
1238 		}
1239 
1240 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1241 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1242 			return SPEC_STORE_BYPASS_CMD_AUTO;
1243 		}
1244 	}
1245 
1246 	return cmd;
1247 }
1248 
1249 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1250 {
1251 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1252 	enum ssb_mitigation_cmd cmd;
1253 
1254 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1255 		return mode;
1256 
1257 	cmd = ssb_parse_cmdline();
1258 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1259 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1260 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1261 		return mode;
1262 
1263 	switch (cmd) {
1264 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1265 		/*
1266 		 * Choose prctl+seccomp as the default mode if seccomp is
1267 		 * enabled.
1268 		 */
1269 		if (IS_ENABLED(CONFIG_SECCOMP))
1270 			mode = SPEC_STORE_BYPASS_SECCOMP;
1271 		else
1272 			mode = SPEC_STORE_BYPASS_PRCTL;
1273 		break;
1274 	case SPEC_STORE_BYPASS_CMD_ON:
1275 		mode = SPEC_STORE_BYPASS_DISABLE;
1276 		break;
1277 	case SPEC_STORE_BYPASS_CMD_AUTO:
1278 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1279 		mode = SPEC_STORE_BYPASS_PRCTL;
1280 		break;
1281 	case SPEC_STORE_BYPASS_CMD_NONE:
1282 		break;
1283 	}
1284 
1285 	/*
1286 	 * If SSBD is controlled by the SPEC_CTRL MSR, then set the proper
1287 	 * bit in the mask to allow guests to use the mitigation even in the
1288 	 * case where the host does not enable it.
1289 	 */
1290 	if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
1291 	    static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1292 		x86_spec_ctrl_mask |= SPEC_CTRL_SSBD;
1293 	}
1294 
1295 	/*
1296 	 * We have three CPU feature flags that are in play here:
1297 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1298 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1299 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1300 	 */
1301 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1302 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1303 		/*
1304 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1305 		 * use a completely different MSR and bit dependent on family.
1306 		 */
1307 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1308 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1309 			x86_amd_ssb_disable();
1310 		} else {
1311 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1312 			wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1313 		}
1314 	}
1315 
1316 	return mode;
1317 }
1318 
1319 static void ssb_select_mitigation(void)
1320 {
1321 	ssb_mode = __ssb_select_mitigation();
1322 
1323 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1324 		pr_info("%s\n", ssb_strings[ssb_mode]);
1325 }
1326 
1327 #undef pr_fmt
1328 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1329 
1330 static void task_update_spec_tif(struct task_struct *tsk)
1331 {
1332 	/* Force the update of the real TIF bits */
1333 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1334 
1335 	/*
1336 	 * Immediately update the speculation control MSRs for the current
1337 	 * task, but for a non-current task delay setting the CPU
1338 	 * mitigation until it is scheduled next.
1339 	 *
1340 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1341 	 * always the current task.
1342 	 */
1343 	if (tsk == current)
1344 		speculation_ctrl_update_current();
1345 }
1346 
1347 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1348 {
1349 
1350 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1351 		return -EPERM;
1352 
1353 	switch (ctrl) {
1354 	case PR_SPEC_ENABLE:
1355 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1356 		return 0;
1357 	case PR_SPEC_DISABLE:
1358 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1359 		return 0;
1360 	default:
1361 		return -ERANGE;
1362 	}
1363 }
1364 
1365 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1366 {
1367 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1368 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1369 		return -ENXIO;
1370 
1371 	switch (ctrl) {
1372 	case PR_SPEC_ENABLE:
1373 		/* If speculation is force disabled, enable is not allowed */
1374 		if (task_spec_ssb_force_disable(task))
1375 			return -EPERM;
1376 		task_clear_spec_ssb_disable(task);
1377 		task_clear_spec_ssb_noexec(task);
1378 		task_update_spec_tif(task);
1379 		break;
1380 	case PR_SPEC_DISABLE:
1381 		task_set_spec_ssb_disable(task);
1382 		task_clear_spec_ssb_noexec(task);
1383 		task_update_spec_tif(task);
1384 		break;
1385 	case PR_SPEC_FORCE_DISABLE:
1386 		task_set_spec_ssb_disable(task);
1387 		task_set_spec_ssb_force_disable(task);
1388 		task_clear_spec_ssb_noexec(task);
1389 		task_update_spec_tif(task);
1390 		break;
1391 	case PR_SPEC_DISABLE_NOEXEC:
1392 		if (task_spec_ssb_force_disable(task))
1393 			return -EPERM;
1394 		task_set_spec_ssb_disable(task);
1395 		task_set_spec_ssb_noexec(task);
1396 		task_update_spec_tif(task);
1397 		break;
1398 	default:
1399 		return -ERANGE;
1400 	}
1401 	return 0;
1402 }
1403 
1404 static bool is_spec_ib_user_controlled(void)
1405 {
1406 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1407 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1408 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1409 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1410 }
1411 
1412 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1413 {
1414 	switch (ctrl) {
1415 	case PR_SPEC_ENABLE:
1416 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1417 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1418 			return 0;
1419 
1420 		/*
1421 		 * With strict mode for both IBPB and STIBP, the instruction
1422 		 * code paths avoid checking this task flag and instead,
1423 		 * unconditionally run the instruction. However, STIBP and IBPB
1424 		 * are independent and either can be set to conditionally
1425 		 * enabled regardless of the mode of the other.
1426 		 *
1427 		 * If either is set to conditional, allow the task flag to be
1428 		 * updated, unless it was force-disabled by a previous prctl
1429 		 * call. Currently, this is possible on an AMD CPU which has the
1430 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1431 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1432 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1433 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1434 		 */
1435 		if (!is_spec_ib_user_controlled() ||
1436 		    task_spec_ib_force_disable(task))
1437 			return -EPERM;
1438 
1439 		task_clear_spec_ib_disable(task);
1440 		task_update_spec_tif(task);
1441 		break;
1442 	case PR_SPEC_DISABLE:
1443 	case PR_SPEC_FORCE_DISABLE:
1444 		/*
1445 		 * Indirect branch speculation is always allowed when
1446 		 * mitigation is force disabled.
1447 		 */
1448 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1449 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1450 			return -EPERM;
1451 
1452 		if (!is_spec_ib_user_controlled())
1453 			return 0;
1454 
1455 		task_set_spec_ib_disable(task);
1456 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1457 			task_set_spec_ib_force_disable(task);
1458 		task_update_spec_tif(task);
1459 		break;
1460 	default:
1461 		return -ERANGE;
1462 	}
1463 	return 0;
1464 }
1465 
1466 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1467 			     unsigned long ctrl)
1468 {
1469 	switch (which) {
1470 	case PR_SPEC_STORE_BYPASS:
1471 		return ssb_prctl_set(task, ctrl);
1472 	case PR_SPEC_INDIRECT_BRANCH:
1473 		return ib_prctl_set(task, ctrl);
1474 	case PR_SPEC_L1D_FLUSH:
1475 		return l1d_flush_prctl_set(task, ctrl);
1476 	default:
1477 		return -ENODEV;
1478 	}
1479 }
1480 
1481 #ifdef CONFIG_SECCOMP
1482 void arch_seccomp_spec_mitigate(struct task_struct *task)
1483 {
1484 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1485 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1486 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1487 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1488 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1489 }
1490 #endif
1491 
1492 static int l1d_flush_prctl_get(struct task_struct *task)
1493 {
1494 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1495 		return PR_SPEC_FORCE_DISABLE;
1496 
1497 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
1498 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1499 	else
1500 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1501 }
1502 
1503 static int ssb_prctl_get(struct task_struct *task)
1504 {
1505 	switch (ssb_mode) {
1506 	case SPEC_STORE_BYPASS_DISABLE:
1507 		return PR_SPEC_DISABLE;
1508 	case SPEC_STORE_BYPASS_SECCOMP:
1509 	case SPEC_STORE_BYPASS_PRCTL:
1510 		if (task_spec_ssb_force_disable(task))
1511 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1512 		if (task_spec_ssb_noexec(task))
1513 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
1514 		if (task_spec_ssb_disable(task))
1515 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1516 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1517 	default:
1518 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1519 			return PR_SPEC_ENABLE;
1520 		return PR_SPEC_NOT_AFFECTED;
1521 	}
1522 }
1523 
1524 static int ib_prctl_get(struct task_struct *task)
1525 {
1526 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
1527 		return PR_SPEC_NOT_AFFECTED;
1528 
1529 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1530 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1531 		return PR_SPEC_ENABLE;
1532 	else if (is_spec_ib_user_controlled()) {
1533 		if (task_spec_ib_force_disable(task))
1534 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1535 		if (task_spec_ib_disable(task))
1536 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1537 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1538 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
1539 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
1540 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
1541 		return PR_SPEC_DISABLE;
1542 	else
1543 		return PR_SPEC_NOT_AFFECTED;
1544 }
1545 
1546 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
1547 {
1548 	switch (which) {
1549 	case PR_SPEC_STORE_BYPASS:
1550 		return ssb_prctl_get(task);
1551 	case PR_SPEC_INDIRECT_BRANCH:
1552 		return ib_prctl_get(task);
1553 	case PR_SPEC_L1D_FLUSH:
1554 		return l1d_flush_prctl_get(task);
1555 	default:
1556 		return -ENODEV;
1557 	}
1558 }
1559 
1560 void x86_spec_ctrl_setup_ap(void)
1561 {
1562 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
1563 		wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1564 
1565 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
1566 		x86_amd_ssb_disable();
1567 }
1568 
1569 bool itlb_multihit_kvm_mitigation;
1570 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
1571 
1572 #undef pr_fmt
1573 #define pr_fmt(fmt)	"L1TF: " fmt
1574 
1575 /* Default mitigation for L1TF-affected CPUs */
1576 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
1577 #if IS_ENABLED(CONFIG_KVM_INTEL)
1578 EXPORT_SYMBOL_GPL(l1tf_mitigation);
1579 #endif
1580 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
1581 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
1582 
1583 /*
1584  * These CPUs all support 44bits physical address space internally in the
1585  * cache but CPUID can report a smaller number of physical address bits.
1586  *
1587  * The L1TF mitigation uses the top most address bit for the inversion of
1588  * non present PTEs. When the installed memory reaches into the top most
1589  * address bit due to memory holes, which has been observed on machines
1590  * which report 36bits physical address bits and have 32G RAM installed,
1591  * then the mitigation range check in l1tf_select_mitigation() triggers.
1592  * This is a false positive because the mitigation is still possible due to
1593  * the fact that the cache uses 44bit internally. Use the cache bits
1594  * instead of the reported physical bits and adjust them on the affected
1595  * machines to 44bit if the reported bits are less than 44.
1596  */
1597 static void override_cache_bits(struct cpuinfo_x86 *c)
1598 {
1599 	if (c->x86 != 6)
1600 		return;
1601 
1602 	switch (c->x86_model) {
1603 	case INTEL_FAM6_NEHALEM:
1604 	case INTEL_FAM6_WESTMERE:
1605 	case INTEL_FAM6_SANDYBRIDGE:
1606 	case INTEL_FAM6_IVYBRIDGE:
1607 	case INTEL_FAM6_HASWELL:
1608 	case INTEL_FAM6_HASWELL_L:
1609 	case INTEL_FAM6_HASWELL_G:
1610 	case INTEL_FAM6_BROADWELL:
1611 	case INTEL_FAM6_BROADWELL_G:
1612 	case INTEL_FAM6_SKYLAKE_L:
1613 	case INTEL_FAM6_SKYLAKE:
1614 	case INTEL_FAM6_KABYLAKE_L:
1615 	case INTEL_FAM6_KABYLAKE:
1616 		if (c->x86_cache_bits < 44)
1617 			c->x86_cache_bits = 44;
1618 		break;
1619 	}
1620 }
1621 
1622 static void __init l1tf_select_mitigation(void)
1623 {
1624 	u64 half_pa;
1625 
1626 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
1627 		return;
1628 
1629 	if (cpu_mitigations_off())
1630 		l1tf_mitigation = L1TF_MITIGATION_OFF;
1631 	else if (cpu_mitigations_auto_nosmt())
1632 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
1633 
1634 	override_cache_bits(&boot_cpu_data);
1635 
1636 	switch (l1tf_mitigation) {
1637 	case L1TF_MITIGATION_OFF:
1638 	case L1TF_MITIGATION_FLUSH_NOWARN:
1639 	case L1TF_MITIGATION_FLUSH:
1640 		break;
1641 	case L1TF_MITIGATION_FLUSH_NOSMT:
1642 	case L1TF_MITIGATION_FULL:
1643 		cpu_smt_disable(false);
1644 		break;
1645 	case L1TF_MITIGATION_FULL_FORCE:
1646 		cpu_smt_disable(true);
1647 		break;
1648 	}
1649 
1650 #if CONFIG_PGTABLE_LEVELS == 2
1651 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
1652 	return;
1653 #endif
1654 
1655 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
1656 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
1657 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
1658 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
1659 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
1660 				half_pa);
1661 		pr_info("However, doing so will make a part of your RAM unusable.\n");
1662 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
1663 		return;
1664 	}
1665 
1666 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
1667 }
1668 
1669 static int __init l1tf_cmdline(char *str)
1670 {
1671 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
1672 		return 0;
1673 
1674 	if (!str)
1675 		return -EINVAL;
1676 
1677 	if (!strcmp(str, "off"))
1678 		l1tf_mitigation = L1TF_MITIGATION_OFF;
1679 	else if (!strcmp(str, "flush,nowarn"))
1680 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
1681 	else if (!strcmp(str, "flush"))
1682 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
1683 	else if (!strcmp(str, "flush,nosmt"))
1684 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
1685 	else if (!strcmp(str, "full"))
1686 		l1tf_mitigation = L1TF_MITIGATION_FULL;
1687 	else if (!strcmp(str, "full,force"))
1688 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
1689 
1690 	return 0;
1691 }
1692 early_param("l1tf", l1tf_cmdline);
1693 
1694 #undef pr_fmt
1695 #define pr_fmt(fmt) fmt
1696 
1697 #ifdef CONFIG_SYSFS
1698 
1699 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
1700 
1701 #if IS_ENABLED(CONFIG_KVM_INTEL)
1702 static const char * const l1tf_vmx_states[] = {
1703 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
1704 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
1705 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
1706 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
1707 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
1708 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
1709 };
1710 
1711 static ssize_t l1tf_show_state(char *buf)
1712 {
1713 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
1714 		return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
1715 
1716 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
1717 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
1718 	     sched_smt_active())) {
1719 		return sprintf(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
1720 			       l1tf_vmx_states[l1tf_vmx_mitigation]);
1721 	}
1722 
1723 	return sprintf(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
1724 		       l1tf_vmx_states[l1tf_vmx_mitigation],
1725 		       sched_smt_active() ? "vulnerable" : "disabled");
1726 }
1727 
1728 static ssize_t itlb_multihit_show_state(char *buf)
1729 {
1730 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
1731 	    !boot_cpu_has(X86_FEATURE_VMX))
1732 		return sprintf(buf, "KVM: Mitigation: VMX unsupported\n");
1733 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
1734 		return sprintf(buf, "KVM: Mitigation: VMX disabled\n");
1735 	else if (itlb_multihit_kvm_mitigation)
1736 		return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
1737 	else
1738 		return sprintf(buf, "KVM: Vulnerable\n");
1739 }
1740 #else
1741 static ssize_t l1tf_show_state(char *buf)
1742 {
1743 	return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
1744 }
1745 
1746 static ssize_t itlb_multihit_show_state(char *buf)
1747 {
1748 	return sprintf(buf, "Processor vulnerable\n");
1749 }
1750 #endif
1751 
1752 static ssize_t mds_show_state(char *buf)
1753 {
1754 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1755 		return sprintf(buf, "%s; SMT Host state unknown\n",
1756 			       mds_strings[mds_mitigation]);
1757 	}
1758 
1759 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
1760 		return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
1761 			       (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
1762 			        sched_smt_active() ? "mitigated" : "disabled"));
1763 	}
1764 
1765 	return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
1766 		       sched_smt_active() ? "vulnerable" : "disabled");
1767 }
1768 
1769 static ssize_t tsx_async_abort_show_state(char *buf)
1770 {
1771 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
1772 	    (taa_mitigation == TAA_MITIGATION_OFF))
1773 		return sprintf(buf, "%s\n", taa_strings[taa_mitigation]);
1774 
1775 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1776 		return sprintf(buf, "%s; SMT Host state unknown\n",
1777 			       taa_strings[taa_mitigation]);
1778 	}
1779 
1780 	return sprintf(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
1781 		       sched_smt_active() ? "vulnerable" : "disabled");
1782 }
1783 
1784 static char *stibp_state(void)
1785 {
1786 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled))
1787 		return "";
1788 
1789 	switch (spectre_v2_user_stibp) {
1790 	case SPECTRE_V2_USER_NONE:
1791 		return ", STIBP: disabled";
1792 	case SPECTRE_V2_USER_STRICT:
1793 		return ", STIBP: forced";
1794 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1795 		return ", STIBP: always-on";
1796 	case SPECTRE_V2_USER_PRCTL:
1797 	case SPECTRE_V2_USER_SECCOMP:
1798 		if (static_key_enabled(&switch_to_cond_stibp))
1799 			return ", STIBP: conditional";
1800 	}
1801 	return "";
1802 }
1803 
1804 static char *ibpb_state(void)
1805 {
1806 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1807 		if (static_key_enabled(&switch_mm_always_ibpb))
1808 			return ", IBPB: always-on";
1809 		if (static_key_enabled(&switch_mm_cond_ibpb))
1810 			return ", IBPB: conditional";
1811 		return ", IBPB: disabled";
1812 	}
1813 	return "";
1814 }
1815 
1816 static ssize_t spectre_v2_show_state(char *buf)
1817 {
1818 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
1819 		return sprintf(buf, "Vulnerable: LFENCE\n");
1820 
1821 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1822 		return sprintf(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
1823 
1824 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1825 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1826 		return sprintf(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
1827 
1828 	return sprintf(buf, "%s%s%s%s%s%s\n",
1829 		       spectre_v2_strings[spectre_v2_enabled],
1830 		       ibpb_state(),
1831 		       boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
1832 		       stibp_state(),
1833 		       boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
1834 		       spectre_v2_module_string());
1835 }
1836 
1837 static ssize_t srbds_show_state(char *buf)
1838 {
1839 	return sprintf(buf, "%s\n", srbds_strings[srbds_mitigation]);
1840 }
1841 
1842 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
1843 			       char *buf, unsigned int bug)
1844 {
1845 	if (!boot_cpu_has_bug(bug))
1846 		return sprintf(buf, "Not affected\n");
1847 
1848 	switch (bug) {
1849 	case X86_BUG_CPU_MELTDOWN:
1850 		if (boot_cpu_has(X86_FEATURE_PTI))
1851 			return sprintf(buf, "Mitigation: PTI\n");
1852 
1853 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
1854 			return sprintf(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
1855 
1856 		break;
1857 
1858 	case X86_BUG_SPECTRE_V1:
1859 		return sprintf(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
1860 
1861 	case X86_BUG_SPECTRE_V2:
1862 		return spectre_v2_show_state(buf);
1863 
1864 	case X86_BUG_SPEC_STORE_BYPASS:
1865 		return sprintf(buf, "%s\n", ssb_strings[ssb_mode]);
1866 
1867 	case X86_BUG_L1TF:
1868 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
1869 			return l1tf_show_state(buf);
1870 		break;
1871 
1872 	case X86_BUG_MDS:
1873 		return mds_show_state(buf);
1874 
1875 	case X86_BUG_TAA:
1876 		return tsx_async_abort_show_state(buf);
1877 
1878 	case X86_BUG_ITLB_MULTIHIT:
1879 		return itlb_multihit_show_state(buf);
1880 
1881 	case X86_BUG_SRBDS:
1882 		return srbds_show_state(buf);
1883 
1884 	default:
1885 		break;
1886 	}
1887 
1888 	return sprintf(buf, "Vulnerable\n");
1889 }
1890 
1891 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
1892 {
1893 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
1894 }
1895 
1896 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
1897 {
1898 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
1899 }
1900 
1901 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
1902 {
1903 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
1904 }
1905 
1906 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
1907 {
1908 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
1909 }
1910 
1911 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
1912 {
1913 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
1914 }
1915 
1916 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
1917 {
1918 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
1919 }
1920 
1921 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
1922 {
1923 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
1924 }
1925 
1926 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
1927 {
1928 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
1929 }
1930 
1931 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
1932 {
1933 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
1934 }
1935 #endif
1936