xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision 2aa71b4b)
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/cpu.h>
13 #include <linux/module.h>
14 #include <linux/nospec.h>
15 #include <linux/prctl.h>
16 #include <linux/sched/smt.h>
17 #include <linux/pgtable.h>
18 #include <linux/bpf.h>
19 
20 #include <asm/spec-ctrl.h>
21 #include <asm/cmdline.h>
22 #include <asm/bugs.h>
23 #include <asm/processor.h>
24 #include <asm/processor-flags.h>
25 #include <asm/fpu/api.h>
26 #include <asm/msr.h>
27 #include <asm/vmx.h>
28 #include <asm/paravirt.h>
29 #include <asm/intel-family.h>
30 #include <asm/e820/api.h>
31 #include <asm/hypervisor.h>
32 #include <asm/tlbflush.h>
33 #include <asm/cpu.h>
34 
35 #include "cpu.h"
36 
37 static void __init spectre_v1_select_mitigation(void);
38 static void __init spectre_v2_select_mitigation(void);
39 static void __init retbleed_select_mitigation(void);
40 static void __init spectre_v2_user_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 md_clear_update_mitigation(void);
45 static void __init md_clear_select_mitigation(void);
46 static void __init taa_select_mitigation(void);
47 static void __init mmio_select_mitigation(void);
48 static void __init srbds_select_mitigation(void);
49 static void __init l1d_flush_select_mitigation(void);
50 
51 /* The base value of the SPEC_CTRL MSR without task-specific bits set */
52 u64 x86_spec_ctrl_base;
53 EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
54 
55 /* The current value of the SPEC_CTRL MSR with task-specific bits set */
56 DEFINE_PER_CPU(u64, x86_spec_ctrl_current);
57 EXPORT_SYMBOL_GPL(x86_spec_ctrl_current);
58 
59 static DEFINE_MUTEX(spec_ctrl_mutex);
60 
61 /* Update SPEC_CTRL MSR and its cached copy unconditionally */
62 static void update_spec_ctrl(u64 val)
63 {
64 	this_cpu_write(x86_spec_ctrl_current, val);
65 	wrmsrl(MSR_IA32_SPEC_CTRL, val);
66 }
67 
68 /*
69  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
70  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
71  */
72 void update_spec_ctrl_cond(u64 val)
73 {
74 	if (this_cpu_read(x86_spec_ctrl_current) == val)
75 		return;
76 
77 	this_cpu_write(x86_spec_ctrl_current, val);
78 
79 	/*
80 	 * When KERNEL_IBRS this MSR is written on return-to-user, unless
81 	 * forced the update can be delayed until that time.
82 	 */
83 	if (!cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
84 		wrmsrl(MSR_IA32_SPEC_CTRL, val);
85 }
86 
87 noinstr u64 spec_ctrl_current(void)
88 {
89 	return this_cpu_read(x86_spec_ctrl_current);
90 }
91 EXPORT_SYMBOL_GPL(spec_ctrl_current);
92 
93 /*
94  * AMD specific MSR info for Speculative Store Bypass control.
95  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
96  */
97 u64 __ro_after_init x86_amd_ls_cfg_base;
98 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
99 
100 /* Control conditional STIBP in switch_to() */
101 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
102 /* Control conditional IBPB in switch_mm() */
103 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
104 /* Control unconditional IBPB in switch_mm() */
105 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
106 
107 /* Control MDS CPU buffer clear before returning to user space */
108 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
109 EXPORT_SYMBOL_GPL(mds_user_clear);
110 /* Control MDS CPU buffer clear before idling (halt, mwait) */
111 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
112 EXPORT_SYMBOL_GPL(mds_idle_clear);
113 
114 /*
115  * Controls whether l1d flush based mitigations are enabled,
116  * based on hw features and admin setting via boot parameter
117  * defaults to false
118  */
119 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
120 
121 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
122 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
123 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
124 
125 void __init cpu_select_mitigations(void)
126 {
127 	/*
128 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
129 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
130 	 * init code as it is not enumerated and depends on the family.
131 	 */
132 	if (cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL)) {
133 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
134 
135 		/*
136 		 * Previously running kernel (kexec), may have some controls
137 		 * turned ON. Clear them and let the mitigations setup below
138 		 * rediscover them based on configuration.
139 		 */
140 		x86_spec_ctrl_base &= ~SPEC_CTRL_MITIGATIONS_MASK;
141 	}
142 
143 	/* Select the proper CPU mitigations before patching alternatives: */
144 	spectre_v1_select_mitigation();
145 	spectre_v2_select_mitigation();
146 	/*
147 	 * retbleed_select_mitigation() relies on the state set by
148 	 * spectre_v2_select_mitigation(); specifically it wants to know about
149 	 * spectre_v2=ibrs.
150 	 */
151 	retbleed_select_mitigation();
152 	/*
153 	 * spectre_v2_user_select_mitigation() relies on the state set by
154 	 * retbleed_select_mitigation(); specifically the STIBP selection is
155 	 * forced for UNRET or IBPB.
156 	 */
157 	spectre_v2_user_select_mitigation();
158 	ssb_select_mitigation();
159 	l1tf_select_mitigation();
160 	md_clear_select_mitigation();
161 	srbds_select_mitigation();
162 	l1d_flush_select_mitigation();
163 }
164 
165 /*
166  * NOTE: This function is *only* called for SVM, since Intel uses
167  * MSR_IA32_SPEC_CTRL for SSBD.
168  */
169 void
170 x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl, bool setguest)
171 {
172 	u64 guestval, hostval;
173 	struct thread_info *ti = current_thread_info();
174 
175 	/*
176 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
177 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
178 	 */
179 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
180 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
181 		return;
182 
183 	/*
184 	 * If the host has SSBD mitigation enabled, force it in the host's
185 	 * virtual MSR value. If its not permanently enabled, evaluate
186 	 * current's TIF_SSBD thread flag.
187 	 */
188 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
189 		hostval = SPEC_CTRL_SSBD;
190 	else
191 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
192 
193 	/* Sanitize the guest value */
194 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
195 
196 	if (hostval != guestval) {
197 		unsigned long tif;
198 
199 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
200 				 ssbd_spec_ctrl_to_tif(hostval);
201 
202 		speculation_ctrl_update(tif);
203 	}
204 }
205 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
206 
207 static void x86_amd_ssb_disable(void)
208 {
209 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
210 
211 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
212 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
213 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
214 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
215 }
216 
217 #undef pr_fmt
218 #define pr_fmt(fmt)	"MDS: " fmt
219 
220 /* Default mitigation for MDS-affected CPUs */
221 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
222 static bool mds_nosmt __ro_after_init = false;
223 
224 static const char * const mds_strings[] = {
225 	[MDS_MITIGATION_OFF]	= "Vulnerable",
226 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
227 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
228 };
229 
230 static void __init mds_select_mitigation(void)
231 {
232 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
233 		mds_mitigation = MDS_MITIGATION_OFF;
234 		return;
235 	}
236 
237 	if (mds_mitigation == MDS_MITIGATION_FULL) {
238 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
239 			mds_mitigation = MDS_MITIGATION_VMWERV;
240 
241 		static_branch_enable(&mds_user_clear);
242 
243 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
244 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
245 			cpu_smt_disable(false);
246 	}
247 }
248 
249 static int __init mds_cmdline(char *str)
250 {
251 	if (!boot_cpu_has_bug(X86_BUG_MDS))
252 		return 0;
253 
254 	if (!str)
255 		return -EINVAL;
256 
257 	if (!strcmp(str, "off"))
258 		mds_mitigation = MDS_MITIGATION_OFF;
259 	else if (!strcmp(str, "full"))
260 		mds_mitigation = MDS_MITIGATION_FULL;
261 	else if (!strcmp(str, "full,nosmt")) {
262 		mds_mitigation = MDS_MITIGATION_FULL;
263 		mds_nosmt = true;
264 	}
265 
266 	return 0;
267 }
268 early_param("mds", mds_cmdline);
269 
270 #undef pr_fmt
271 #define pr_fmt(fmt)	"TAA: " fmt
272 
273 enum taa_mitigations {
274 	TAA_MITIGATION_OFF,
275 	TAA_MITIGATION_UCODE_NEEDED,
276 	TAA_MITIGATION_VERW,
277 	TAA_MITIGATION_TSX_DISABLED,
278 };
279 
280 /* Default mitigation for TAA-affected CPUs */
281 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
282 static bool taa_nosmt __ro_after_init;
283 
284 static const char * const taa_strings[] = {
285 	[TAA_MITIGATION_OFF]		= "Vulnerable",
286 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
287 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
288 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
289 };
290 
291 static void __init taa_select_mitigation(void)
292 {
293 	u64 ia32_cap;
294 
295 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
296 		taa_mitigation = TAA_MITIGATION_OFF;
297 		return;
298 	}
299 
300 	/* TSX previously disabled by tsx=off */
301 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
302 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
303 		return;
304 	}
305 
306 	if (cpu_mitigations_off()) {
307 		taa_mitigation = TAA_MITIGATION_OFF;
308 		return;
309 	}
310 
311 	/*
312 	 * TAA mitigation via VERW is turned off if both
313 	 * tsx_async_abort=off and mds=off are specified.
314 	 */
315 	if (taa_mitigation == TAA_MITIGATION_OFF &&
316 	    mds_mitigation == MDS_MITIGATION_OFF)
317 		return;
318 
319 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
320 		taa_mitigation = TAA_MITIGATION_VERW;
321 	else
322 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
323 
324 	/*
325 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
326 	 * A microcode update fixes this behavior to clear CPU buffers. It also
327 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
328 	 * ARCH_CAP_TSX_CTRL_MSR bit.
329 	 *
330 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
331 	 * update is required.
332 	 */
333 	ia32_cap = x86_read_arch_cap_msr();
334 	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
335 	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
336 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
337 
338 	/*
339 	 * TSX is enabled, select alternate mitigation for TAA which is
340 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
341 	 *
342 	 * For guests that can't determine whether the correct microcode is
343 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
344 	 */
345 	static_branch_enable(&mds_user_clear);
346 
347 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
348 		cpu_smt_disable(false);
349 }
350 
351 static int __init tsx_async_abort_parse_cmdline(char *str)
352 {
353 	if (!boot_cpu_has_bug(X86_BUG_TAA))
354 		return 0;
355 
356 	if (!str)
357 		return -EINVAL;
358 
359 	if (!strcmp(str, "off")) {
360 		taa_mitigation = TAA_MITIGATION_OFF;
361 	} else if (!strcmp(str, "full")) {
362 		taa_mitigation = TAA_MITIGATION_VERW;
363 	} else if (!strcmp(str, "full,nosmt")) {
364 		taa_mitigation = TAA_MITIGATION_VERW;
365 		taa_nosmt = true;
366 	}
367 
368 	return 0;
369 }
370 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
371 
372 #undef pr_fmt
373 #define pr_fmt(fmt)	"MMIO Stale Data: " fmt
374 
375 enum mmio_mitigations {
376 	MMIO_MITIGATION_OFF,
377 	MMIO_MITIGATION_UCODE_NEEDED,
378 	MMIO_MITIGATION_VERW,
379 };
380 
381 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
382 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
383 static bool mmio_nosmt __ro_after_init = false;
384 
385 static const char * const mmio_strings[] = {
386 	[MMIO_MITIGATION_OFF]		= "Vulnerable",
387 	[MMIO_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
388 	[MMIO_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
389 };
390 
391 static void __init mmio_select_mitigation(void)
392 {
393 	u64 ia32_cap;
394 
395 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
396 	     boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
397 	     cpu_mitigations_off()) {
398 		mmio_mitigation = MMIO_MITIGATION_OFF;
399 		return;
400 	}
401 
402 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
403 		return;
404 
405 	ia32_cap = x86_read_arch_cap_msr();
406 
407 	/*
408 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
409 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
410 	 */
411 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
412 					      boot_cpu_has(X86_FEATURE_RTM)))
413 		static_branch_enable(&mds_user_clear);
414 	else
415 		static_branch_enable(&mmio_stale_data_clear);
416 
417 	/*
418 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
419 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
420 	 * is required irrespective of SMT state.
421 	 */
422 	if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
423 		static_branch_enable(&mds_idle_clear);
424 
425 	/*
426 	 * Check if the system has the right microcode.
427 	 *
428 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
429 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
430 	 * affected systems.
431 	 */
432 	if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
433 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
434 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
435 	     !(ia32_cap & ARCH_CAP_MDS_NO)))
436 		mmio_mitigation = MMIO_MITIGATION_VERW;
437 	else
438 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
439 
440 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
441 		cpu_smt_disable(false);
442 }
443 
444 static int __init mmio_stale_data_parse_cmdline(char *str)
445 {
446 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
447 		return 0;
448 
449 	if (!str)
450 		return -EINVAL;
451 
452 	if (!strcmp(str, "off")) {
453 		mmio_mitigation = MMIO_MITIGATION_OFF;
454 	} else if (!strcmp(str, "full")) {
455 		mmio_mitigation = MMIO_MITIGATION_VERW;
456 	} else if (!strcmp(str, "full,nosmt")) {
457 		mmio_mitigation = MMIO_MITIGATION_VERW;
458 		mmio_nosmt = true;
459 	}
460 
461 	return 0;
462 }
463 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
464 
465 #undef pr_fmt
466 #define pr_fmt(fmt)     "" fmt
467 
468 static void __init md_clear_update_mitigation(void)
469 {
470 	if (cpu_mitigations_off())
471 		return;
472 
473 	if (!static_key_enabled(&mds_user_clear))
474 		goto out;
475 
476 	/*
477 	 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
478 	 * mitigation, if necessary.
479 	 */
480 	if (mds_mitigation == MDS_MITIGATION_OFF &&
481 	    boot_cpu_has_bug(X86_BUG_MDS)) {
482 		mds_mitigation = MDS_MITIGATION_FULL;
483 		mds_select_mitigation();
484 	}
485 	if (taa_mitigation == TAA_MITIGATION_OFF &&
486 	    boot_cpu_has_bug(X86_BUG_TAA)) {
487 		taa_mitigation = TAA_MITIGATION_VERW;
488 		taa_select_mitigation();
489 	}
490 	if (mmio_mitigation == MMIO_MITIGATION_OFF &&
491 	    boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
492 		mmio_mitigation = MMIO_MITIGATION_VERW;
493 		mmio_select_mitigation();
494 	}
495 out:
496 	if (boot_cpu_has_bug(X86_BUG_MDS))
497 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
498 	if (boot_cpu_has_bug(X86_BUG_TAA))
499 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
500 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
501 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
502 	else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
503 		pr_info("MMIO Stale Data: Unknown: No mitigations\n");
504 }
505 
506 static void __init md_clear_select_mitigation(void)
507 {
508 	mds_select_mitigation();
509 	taa_select_mitigation();
510 	mmio_select_mitigation();
511 
512 	/*
513 	 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
514 	 * and print their mitigation after MDS, TAA and MMIO Stale Data
515 	 * mitigation selection is done.
516 	 */
517 	md_clear_update_mitigation();
518 }
519 
520 #undef pr_fmt
521 #define pr_fmt(fmt)	"SRBDS: " fmt
522 
523 enum srbds_mitigations {
524 	SRBDS_MITIGATION_OFF,
525 	SRBDS_MITIGATION_UCODE_NEEDED,
526 	SRBDS_MITIGATION_FULL,
527 	SRBDS_MITIGATION_TSX_OFF,
528 	SRBDS_MITIGATION_HYPERVISOR,
529 };
530 
531 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
532 
533 static const char * const srbds_strings[] = {
534 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
535 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
536 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
537 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
538 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
539 };
540 
541 static bool srbds_off;
542 
543 void update_srbds_msr(void)
544 {
545 	u64 mcu_ctrl;
546 
547 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
548 		return;
549 
550 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
551 		return;
552 
553 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
554 		return;
555 
556 	/*
557 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
558 	 * being disabled and it hasn't received the SRBDS MSR microcode.
559 	 */
560 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
561 		return;
562 
563 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
564 
565 	switch (srbds_mitigation) {
566 	case SRBDS_MITIGATION_OFF:
567 	case SRBDS_MITIGATION_TSX_OFF:
568 		mcu_ctrl |= RNGDS_MITG_DIS;
569 		break;
570 	case SRBDS_MITIGATION_FULL:
571 		mcu_ctrl &= ~RNGDS_MITG_DIS;
572 		break;
573 	default:
574 		break;
575 	}
576 
577 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
578 }
579 
580 static void __init srbds_select_mitigation(void)
581 {
582 	u64 ia32_cap;
583 
584 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
585 		return;
586 
587 	/*
588 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
589 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
590 	 * by Processor MMIO Stale Data vulnerability.
591 	 */
592 	ia32_cap = x86_read_arch_cap_msr();
593 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
594 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
595 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
596 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
597 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
598 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
599 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
600 	else if (cpu_mitigations_off() || srbds_off)
601 		srbds_mitigation = SRBDS_MITIGATION_OFF;
602 
603 	update_srbds_msr();
604 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
605 }
606 
607 static int __init srbds_parse_cmdline(char *str)
608 {
609 	if (!str)
610 		return -EINVAL;
611 
612 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
613 		return 0;
614 
615 	srbds_off = !strcmp(str, "off");
616 	return 0;
617 }
618 early_param("srbds", srbds_parse_cmdline);
619 
620 #undef pr_fmt
621 #define pr_fmt(fmt)     "L1D Flush : " fmt
622 
623 enum l1d_flush_mitigations {
624 	L1D_FLUSH_OFF = 0,
625 	L1D_FLUSH_ON,
626 };
627 
628 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
629 
630 static void __init l1d_flush_select_mitigation(void)
631 {
632 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
633 		return;
634 
635 	static_branch_enable(&switch_mm_cond_l1d_flush);
636 	pr_info("Conditional flush on switch_mm() enabled\n");
637 }
638 
639 static int __init l1d_flush_parse_cmdline(char *str)
640 {
641 	if (!strcmp(str, "on"))
642 		l1d_flush_mitigation = L1D_FLUSH_ON;
643 
644 	return 0;
645 }
646 early_param("l1d_flush", l1d_flush_parse_cmdline);
647 
648 #undef pr_fmt
649 #define pr_fmt(fmt)     "Spectre V1 : " fmt
650 
651 enum spectre_v1_mitigation {
652 	SPECTRE_V1_MITIGATION_NONE,
653 	SPECTRE_V1_MITIGATION_AUTO,
654 };
655 
656 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
657 	SPECTRE_V1_MITIGATION_AUTO;
658 
659 static const char * const spectre_v1_strings[] = {
660 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
661 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
662 };
663 
664 /*
665  * Does SMAP provide full mitigation against speculative kernel access to
666  * userspace?
667  */
668 static bool smap_works_speculatively(void)
669 {
670 	if (!boot_cpu_has(X86_FEATURE_SMAP))
671 		return false;
672 
673 	/*
674 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
675 	 * prevent speculative access to user data in the L1 cache.
676 	 * Consider SMAP to be non-functional as a mitigation on these
677 	 * CPUs.
678 	 */
679 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
680 		return false;
681 
682 	return true;
683 }
684 
685 static void __init spectre_v1_select_mitigation(void)
686 {
687 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
688 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
689 		return;
690 	}
691 
692 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
693 		/*
694 		 * With Spectre v1, a user can speculatively control either
695 		 * path of a conditional swapgs with a user-controlled GS
696 		 * value.  The mitigation is to add lfences to both code paths.
697 		 *
698 		 * If FSGSBASE is enabled, the user can put a kernel address in
699 		 * GS, in which case SMAP provides no protection.
700 		 *
701 		 * If FSGSBASE is disabled, the user can only put a user space
702 		 * address in GS.  That makes an attack harder, but still
703 		 * possible if there's no SMAP protection.
704 		 */
705 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
706 		    !smap_works_speculatively()) {
707 			/*
708 			 * Mitigation can be provided from SWAPGS itself or
709 			 * PTI as the CR3 write in the Meltdown mitigation
710 			 * is serializing.
711 			 *
712 			 * If neither is there, mitigate with an LFENCE to
713 			 * stop speculation through swapgs.
714 			 */
715 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
716 			    !boot_cpu_has(X86_FEATURE_PTI))
717 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
718 
719 			/*
720 			 * Enable lfences in the kernel entry (non-swapgs)
721 			 * paths, to prevent user entry from speculatively
722 			 * skipping swapgs.
723 			 */
724 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
725 		}
726 	}
727 
728 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
729 }
730 
731 static int __init nospectre_v1_cmdline(char *str)
732 {
733 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
734 	return 0;
735 }
736 early_param("nospectre_v1", nospectre_v1_cmdline);
737 
738 enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init = SPECTRE_V2_NONE;
739 
740 #undef pr_fmt
741 #define pr_fmt(fmt)     "RETBleed: " fmt
742 
743 enum retbleed_mitigation {
744 	RETBLEED_MITIGATION_NONE,
745 	RETBLEED_MITIGATION_UNRET,
746 	RETBLEED_MITIGATION_IBPB,
747 	RETBLEED_MITIGATION_IBRS,
748 	RETBLEED_MITIGATION_EIBRS,
749 	RETBLEED_MITIGATION_STUFF,
750 };
751 
752 enum retbleed_mitigation_cmd {
753 	RETBLEED_CMD_OFF,
754 	RETBLEED_CMD_AUTO,
755 	RETBLEED_CMD_UNRET,
756 	RETBLEED_CMD_IBPB,
757 	RETBLEED_CMD_STUFF,
758 };
759 
760 static const char * const retbleed_strings[] = {
761 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
762 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
763 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
764 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
765 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
766 	[RETBLEED_MITIGATION_STUFF]	= "Mitigation: Stuffing",
767 };
768 
769 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
770 	RETBLEED_MITIGATION_NONE;
771 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
772 	RETBLEED_CMD_AUTO;
773 
774 static int __ro_after_init retbleed_nosmt = false;
775 
776 static int __init retbleed_parse_cmdline(char *str)
777 {
778 	if (!str)
779 		return -EINVAL;
780 
781 	while (str) {
782 		char *next = strchr(str, ',');
783 		if (next) {
784 			*next = 0;
785 			next++;
786 		}
787 
788 		if (!strcmp(str, "off")) {
789 			retbleed_cmd = RETBLEED_CMD_OFF;
790 		} else if (!strcmp(str, "auto")) {
791 			retbleed_cmd = RETBLEED_CMD_AUTO;
792 		} else if (!strcmp(str, "unret")) {
793 			retbleed_cmd = RETBLEED_CMD_UNRET;
794 		} else if (!strcmp(str, "ibpb")) {
795 			retbleed_cmd = RETBLEED_CMD_IBPB;
796 		} else if (!strcmp(str, "stuff")) {
797 			retbleed_cmd = RETBLEED_CMD_STUFF;
798 		} else if (!strcmp(str, "nosmt")) {
799 			retbleed_nosmt = true;
800 		} else if (!strcmp(str, "force")) {
801 			setup_force_cpu_bug(X86_BUG_RETBLEED);
802 		} else {
803 			pr_err("Ignoring unknown retbleed option (%s).", str);
804 		}
805 
806 		str = next;
807 	}
808 
809 	return 0;
810 }
811 early_param("retbleed", retbleed_parse_cmdline);
812 
813 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
814 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
815 
816 static void __init retbleed_select_mitigation(void)
817 {
818 	bool mitigate_smt = false;
819 
820 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
821 		return;
822 
823 	switch (retbleed_cmd) {
824 	case RETBLEED_CMD_OFF:
825 		return;
826 
827 	case RETBLEED_CMD_UNRET:
828 		if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
829 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
830 		} else {
831 			pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
832 			goto do_cmd_auto;
833 		}
834 		break;
835 
836 	case RETBLEED_CMD_IBPB:
837 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
838 			pr_err("WARNING: CPU does not support IBPB.\n");
839 			goto do_cmd_auto;
840 		} else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
841 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
842 		} else {
843 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
844 			goto do_cmd_auto;
845 		}
846 		break;
847 
848 	case RETBLEED_CMD_STUFF:
849 		if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING) &&
850 		    spectre_v2_enabled == SPECTRE_V2_RETPOLINE) {
851 			retbleed_mitigation = RETBLEED_MITIGATION_STUFF;
852 
853 		} else {
854 			if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING))
855 				pr_err("WARNING: retbleed=stuff depends on spectre_v2=retpoline\n");
856 			else
857 				pr_err("WARNING: kernel not compiled with CALL_DEPTH_TRACKING.\n");
858 
859 			goto do_cmd_auto;
860 		}
861 		break;
862 
863 do_cmd_auto:
864 	case RETBLEED_CMD_AUTO:
865 	default:
866 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
867 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
868 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
869 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
870 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
871 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
872 		}
873 
874 		/*
875 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
876 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
877 		 * be set accordingly below.
878 		 */
879 
880 		break;
881 	}
882 
883 	switch (retbleed_mitigation) {
884 	case RETBLEED_MITIGATION_UNRET:
885 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
886 		setup_force_cpu_cap(X86_FEATURE_UNRET);
887 
888 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
889 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
890 			pr_err(RETBLEED_UNTRAIN_MSG);
891 
892 		mitigate_smt = true;
893 		break;
894 
895 	case RETBLEED_MITIGATION_IBPB:
896 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
897 		mitigate_smt = true;
898 		break;
899 
900 	case RETBLEED_MITIGATION_STUFF:
901 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
902 		setup_force_cpu_cap(X86_FEATURE_CALL_DEPTH);
903 		x86_set_skl_return_thunk();
904 		break;
905 
906 	default:
907 		break;
908 	}
909 
910 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
911 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
912 		cpu_smt_disable(false);
913 
914 	/*
915 	 * Let IBRS trump all on Intel without affecting the effects of the
916 	 * retbleed= cmdline option except for call depth based stuffing
917 	 */
918 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
919 		switch (spectre_v2_enabled) {
920 		case SPECTRE_V2_IBRS:
921 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
922 			break;
923 		case SPECTRE_V2_EIBRS:
924 		case SPECTRE_V2_EIBRS_RETPOLINE:
925 		case SPECTRE_V2_EIBRS_LFENCE:
926 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
927 			break;
928 		default:
929 			if (retbleed_mitigation != RETBLEED_MITIGATION_STUFF)
930 				pr_err(RETBLEED_INTEL_MSG);
931 		}
932 	}
933 
934 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
935 }
936 
937 #undef pr_fmt
938 #define pr_fmt(fmt)     "Spectre V2 : " fmt
939 
940 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
941 	SPECTRE_V2_USER_NONE;
942 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
943 	SPECTRE_V2_USER_NONE;
944 
945 #ifdef CONFIG_RETPOLINE
946 static bool spectre_v2_bad_module;
947 
948 bool retpoline_module_ok(bool has_retpoline)
949 {
950 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
951 		return true;
952 
953 	pr_err("System may be vulnerable to spectre v2\n");
954 	spectre_v2_bad_module = true;
955 	return false;
956 }
957 
958 static inline const char *spectre_v2_module_string(void)
959 {
960 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
961 }
962 #else
963 static inline const char *spectre_v2_module_string(void) { return ""; }
964 #endif
965 
966 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
967 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
968 #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"
969 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
970 
971 #ifdef CONFIG_BPF_SYSCALL
972 void unpriv_ebpf_notify(int new_state)
973 {
974 	if (new_state)
975 		return;
976 
977 	/* Unprivileged eBPF is enabled */
978 
979 	switch (spectre_v2_enabled) {
980 	case SPECTRE_V2_EIBRS:
981 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
982 		break;
983 	case SPECTRE_V2_EIBRS_LFENCE:
984 		if (sched_smt_active())
985 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
986 		break;
987 	default:
988 		break;
989 	}
990 }
991 #endif
992 
993 static inline bool match_option(const char *arg, int arglen, const char *opt)
994 {
995 	int len = strlen(opt);
996 
997 	return len == arglen && !strncmp(arg, opt, len);
998 }
999 
1000 /* The kernel command line selection for spectre v2 */
1001 enum spectre_v2_mitigation_cmd {
1002 	SPECTRE_V2_CMD_NONE,
1003 	SPECTRE_V2_CMD_AUTO,
1004 	SPECTRE_V2_CMD_FORCE,
1005 	SPECTRE_V2_CMD_RETPOLINE,
1006 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1007 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1008 	SPECTRE_V2_CMD_EIBRS,
1009 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1010 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1011 	SPECTRE_V2_CMD_IBRS,
1012 };
1013 
1014 enum spectre_v2_user_cmd {
1015 	SPECTRE_V2_USER_CMD_NONE,
1016 	SPECTRE_V2_USER_CMD_AUTO,
1017 	SPECTRE_V2_USER_CMD_FORCE,
1018 	SPECTRE_V2_USER_CMD_PRCTL,
1019 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1020 	SPECTRE_V2_USER_CMD_SECCOMP,
1021 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1022 };
1023 
1024 static const char * const spectre_v2_user_strings[] = {
1025 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1026 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1027 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1028 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1029 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1030 };
1031 
1032 static const struct {
1033 	const char			*option;
1034 	enum spectre_v2_user_cmd	cmd;
1035 	bool				secure;
1036 } v2_user_options[] __initconst = {
1037 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1038 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1039 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1040 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1041 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1042 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1043 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1044 };
1045 
1046 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1047 {
1048 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1049 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1050 }
1051 
1052 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1053 
1054 static enum spectre_v2_user_cmd __init
1055 spectre_v2_parse_user_cmdline(void)
1056 {
1057 	char arg[20];
1058 	int ret, i;
1059 
1060 	switch (spectre_v2_cmd) {
1061 	case SPECTRE_V2_CMD_NONE:
1062 		return SPECTRE_V2_USER_CMD_NONE;
1063 	case SPECTRE_V2_CMD_FORCE:
1064 		return SPECTRE_V2_USER_CMD_FORCE;
1065 	default:
1066 		break;
1067 	}
1068 
1069 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1070 				  arg, sizeof(arg));
1071 	if (ret < 0)
1072 		return SPECTRE_V2_USER_CMD_AUTO;
1073 
1074 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1075 		if (match_option(arg, ret, v2_user_options[i].option)) {
1076 			spec_v2_user_print_cond(v2_user_options[i].option,
1077 						v2_user_options[i].secure);
1078 			return v2_user_options[i].cmd;
1079 		}
1080 	}
1081 
1082 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1083 	return SPECTRE_V2_USER_CMD_AUTO;
1084 }
1085 
1086 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1087 {
1088 	return spectre_v2_in_eibrs_mode(mode) || mode == SPECTRE_V2_IBRS;
1089 }
1090 
1091 static void __init
1092 spectre_v2_user_select_mitigation(void)
1093 {
1094 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1095 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1096 	enum spectre_v2_user_cmd cmd;
1097 
1098 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1099 		return;
1100 
1101 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1102 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1103 		smt_possible = false;
1104 
1105 	cmd = spectre_v2_parse_user_cmdline();
1106 	switch (cmd) {
1107 	case SPECTRE_V2_USER_CMD_NONE:
1108 		goto set_mode;
1109 	case SPECTRE_V2_USER_CMD_FORCE:
1110 		mode = SPECTRE_V2_USER_STRICT;
1111 		break;
1112 	case SPECTRE_V2_USER_CMD_AUTO:
1113 	case SPECTRE_V2_USER_CMD_PRCTL:
1114 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1115 		mode = SPECTRE_V2_USER_PRCTL;
1116 		break;
1117 	case SPECTRE_V2_USER_CMD_SECCOMP:
1118 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1119 		if (IS_ENABLED(CONFIG_SECCOMP))
1120 			mode = SPECTRE_V2_USER_SECCOMP;
1121 		else
1122 			mode = SPECTRE_V2_USER_PRCTL;
1123 		break;
1124 	}
1125 
1126 	/* Initialize Indirect Branch Prediction Barrier */
1127 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1128 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1129 
1130 		spectre_v2_user_ibpb = mode;
1131 		switch (cmd) {
1132 		case SPECTRE_V2_USER_CMD_FORCE:
1133 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1134 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1135 			static_branch_enable(&switch_mm_always_ibpb);
1136 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1137 			break;
1138 		case SPECTRE_V2_USER_CMD_PRCTL:
1139 		case SPECTRE_V2_USER_CMD_AUTO:
1140 		case SPECTRE_V2_USER_CMD_SECCOMP:
1141 			static_branch_enable(&switch_mm_cond_ibpb);
1142 			break;
1143 		default:
1144 			break;
1145 		}
1146 
1147 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1148 			static_key_enabled(&switch_mm_always_ibpb) ?
1149 			"always-on" : "conditional");
1150 	}
1151 
1152 	/*
1153 	 * If no STIBP, Intel enhanced IBRS is enabled, or SMT impossible, STIBP
1154 	 * is not required.
1155 	 *
1156 	 * Intel's Enhanced IBRS also protects against cross-thread branch target
1157 	 * injection in user-mode as the IBRS bit remains always set which
1158 	 * implicitly enables cross-thread protections.  However, in legacy IBRS
1159 	 * mode, the IBRS bit is set only on kernel entry and cleared on return
1160 	 * to userspace.  AMD Automatic IBRS also does not protect userspace.
1161 	 * These modes therefore disable the implicit cross-thread protection,
1162 	 * so allow for STIBP to be selected in those cases.
1163 	 */
1164 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1165 	    !smt_possible ||
1166 	    (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
1167 	     !boot_cpu_has(X86_FEATURE_AUTOIBRS)))
1168 		return;
1169 
1170 	/*
1171 	 * At this point, an STIBP mode other than "off" has been set.
1172 	 * If STIBP support is not being forced, check if STIBP always-on
1173 	 * is preferred.
1174 	 */
1175 	if (mode != SPECTRE_V2_USER_STRICT &&
1176 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1177 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1178 
1179 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1180 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1181 		if (mode != SPECTRE_V2_USER_STRICT &&
1182 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1183 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1184 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1185 	}
1186 
1187 	spectre_v2_user_stibp = mode;
1188 
1189 set_mode:
1190 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1191 }
1192 
1193 static const char * const spectre_v2_strings[] = {
1194 	[SPECTRE_V2_NONE]			= "Vulnerable",
1195 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1196 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1197 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced / Automatic IBRS",
1198 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced / Automatic IBRS + LFENCE",
1199 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced / Automatic IBRS + Retpolines",
1200 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1201 };
1202 
1203 static const struct {
1204 	const char *option;
1205 	enum spectre_v2_mitigation_cmd cmd;
1206 	bool secure;
1207 } mitigation_options[] __initconst = {
1208 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1209 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1210 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1211 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1212 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1213 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1214 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1215 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1216 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1217 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1218 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1219 };
1220 
1221 static void __init spec_v2_print_cond(const char *reason, bool secure)
1222 {
1223 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1224 		pr_info("%s selected on command line.\n", reason);
1225 }
1226 
1227 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1228 {
1229 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1230 	char arg[20];
1231 	int ret, i;
1232 
1233 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1234 	    cpu_mitigations_off())
1235 		return SPECTRE_V2_CMD_NONE;
1236 
1237 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1238 	if (ret < 0)
1239 		return SPECTRE_V2_CMD_AUTO;
1240 
1241 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1242 		if (!match_option(arg, ret, mitigation_options[i].option))
1243 			continue;
1244 		cmd = mitigation_options[i].cmd;
1245 		break;
1246 	}
1247 
1248 	if (i >= ARRAY_SIZE(mitigation_options)) {
1249 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1250 		return SPECTRE_V2_CMD_AUTO;
1251 	}
1252 
1253 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1254 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1255 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1256 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1257 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1258 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1259 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1260 		       mitigation_options[i].option);
1261 		return SPECTRE_V2_CMD_AUTO;
1262 	}
1263 
1264 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1265 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1266 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1267 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1268 		pr_err("%s selected but CPU doesn't have Enhanced or Automatic IBRS. Switching to AUTO select\n",
1269 		       mitigation_options[i].option);
1270 		return SPECTRE_V2_CMD_AUTO;
1271 	}
1272 
1273 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1274 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1275 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1276 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1277 		       mitigation_options[i].option);
1278 		return SPECTRE_V2_CMD_AUTO;
1279 	}
1280 
1281 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1282 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1283 		       mitigation_options[i].option);
1284 		return SPECTRE_V2_CMD_AUTO;
1285 	}
1286 
1287 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1288 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1289 		       mitigation_options[i].option);
1290 		return SPECTRE_V2_CMD_AUTO;
1291 	}
1292 
1293 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1294 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1295 		       mitigation_options[i].option);
1296 		return SPECTRE_V2_CMD_AUTO;
1297 	}
1298 
1299 	if (cmd == SPECTRE_V2_CMD_IBRS && cpu_feature_enabled(X86_FEATURE_XENPV)) {
1300 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1301 		       mitigation_options[i].option);
1302 		return SPECTRE_V2_CMD_AUTO;
1303 	}
1304 
1305 	spec_v2_print_cond(mitigation_options[i].option,
1306 			   mitigation_options[i].secure);
1307 	return cmd;
1308 }
1309 
1310 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1311 {
1312 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1313 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1314 		return SPECTRE_V2_NONE;
1315 	}
1316 
1317 	return SPECTRE_V2_RETPOLINE;
1318 }
1319 
1320 /* Disable in-kernel use of non-RSB RET predictors */
1321 static void __init spec_ctrl_disable_kernel_rrsba(void)
1322 {
1323 	u64 ia32_cap;
1324 
1325 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1326 		return;
1327 
1328 	ia32_cap = x86_read_arch_cap_msr();
1329 
1330 	if (ia32_cap & ARCH_CAP_RRSBA) {
1331 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1332 		update_spec_ctrl(x86_spec_ctrl_base);
1333 	}
1334 }
1335 
1336 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1337 {
1338 	/*
1339 	 * Similar to context switches, there are two types of RSB attacks
1340 	 * after VM exit:
1341 	 *
1342 	 * 1) RSB underflow
1343 	 *
1344 	 * 2) Poisoned RSB entry
1345 	 *
1346 	 * When retpoline is enabled, both are mitigated by filling/clearing
1347 	 * the RSB.
1348 	 *
1349 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1350 	 * prediction isolation protections, RSB still needs to be cleared
1351 	 * because of #2.  Note that SMEP provides no protection here, unlike
1352 	 * user-space-poisoned RSB entries.
1353 	 *
1354 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1355 	 * bug is present then a LITE version of RSB protection is required,
1356 	 * just a single call needs to retire before a RET is executed.
1357 	 */
1358 	switch (mode) {
1359 	case SPECTRE_V2_NONE:
1360 		return;
1361 
1362 	case SPECTRE_V2_EIBRS_LFENCE:
1363 	case SPECTRE_V2_EIBRS:
1364 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1365 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1366 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1367 		}
1368 		return;
1369 
1370 	case SPECTRE_V2_EIBRS_RETPOLINE:
1371 	case SPECTRE_V2_RETPOLINE:
1372 	case SPECTRE_V2_LFENCE:
1373 	case SPECTRE_V2_IBRS:
1374 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1375 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1376 		return;
1377 	}
1378 
1379 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1380 	dump_stack();
1381 }
1382 
1383 static void __init spectre_v2_select_mitigation(void)
1384 {
1385 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1386 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1387 
1388 	/*
1389 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1390 	 * then nothing to do.
1391 	 */
1392 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1393 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1394 		return;
1395 
1396 	switch (cmd) {
1397 	case SPECTRE_V2_CMD_NONE:
1398 		return;
1399 
1400 	case SPECTRE_V2_CMD_FORCE:
1401 	case SPECTRE_V2_CMD_AUTO:
1402 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1403 			mode = SPECTRE_V2_EIBRS;
1404 			break;
1405 		}
1406 
1407 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1408 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1409 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1410 		    retbleed_cmd != RETBLEED_CMD_STUFF &&
1411 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1412 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1413 			mode = SPECTRE_V2_IBRS;
1414 			break;
1415 		}
1416 
1417 		mode = spectre_v2_select_retpoline();
1418 		break;
1419 
1420 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1421 		pr_err(SPECTRE_V2_LFENCE_MSG);
1422 		mode = SPECTRE_V2_LFENCE;
1423 		break;
1424 
1425 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1426 		mode = SPECTRE_V2_RETPOLINE;
1427 		break;
1428 
1429 	case SPECTRE_V2_CMD_RETPOLINE:
1430 		mode = spectre_v2_select_retpoline();
1431 		break;
1432 
1433 	case SPECTRE_V2_CMD_IBRS:
1434 		mode = SPECTRE_V2_IBRS;
1435 		break;
1436 
1437 	case SPECTRE_V2_CMD_EIBRS:
1438 		mode = SPECTRE_V2_EIBRS;
1439 		break;
1440 
1441 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1442 		mode = SPECTRE_V2_EIBRS_LFENCE;
1443 		break;
1444 
1445 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1446 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1447 		break;
1448 	}
1449 
1450 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1451 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1452 
1453 	if (spectre_v2_in_ibrs_mode(mode)) {
1454 		if (boot_cpu_has(X86_FEATURE_AUTOIBRS)) {
1455 			msr_set_bit(MSR_EFER, _EFER_AUTOIBRS);
1456 		} else {
1457 			x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1458 			update_spec_ctrl(x86_spec_ctrl_base);
1459 		}
1460 	}
1461 
1462 	switch (mode) {
1463 	case SPECTRE_V2_NONE:
1464 	case SPECTRE_V2_EIBRS:
1465 		break;
1466 
1467 	case SPECTRE_V2_IBRS:
1468 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1469 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1470 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1471 		break;
1472 
1473 	case SPECTRE_V2_LFENCE:
1474 	case SPECTRE_V2_EIBRS_LFENCE:
1475 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1476 		fallthrough;
1477 
1478 	case SPECTRE_V2_RETPOLINE:
1479 	case SPECTRE_V2_EIBRS_RETPOLINE:
1480 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1481 		break;
1482 	}
1483 
1484 	/*
1485 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1486 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1487 	 * prediction from a non-RSB predictor is still a risk.
1488 	 */
1489 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1490 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1491 	    mode == SPECTRE_V2_RETPOLINE)
1492 		spec_ctrl_disable_kernel_rrsba();
1493 
1494 	spectre_v2_enabled = mode;
1495 	pr_info("%s\n", spectre_v2_strings[mode]);
1496 
1497 	/*
1498 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1499 	 * context switch.  In general there are two types of RSB attacks
1500 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1501 	 *
1502 	 * 1) RSB underflow
1503 	 *
1504 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1505 	 *    speculated return targets may come from the branch predictor,
1506 	 *    which could have a user-poisoned BTB or BHB entry.
1507 	 *
1508 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1509 	 *    regardless of the state of the RSB.
1510 	 *
1511 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1512 	 *    scenario is mitigated by the IBRS branch prediction isolation
1513 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1514 	 *    protect against this type of attack.
1515 	 *
1516 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1517 	 *
1518 	 * 2) Poisoned RSB entry
1519 	 *
1520 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1521 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1522 	 *    entry.
1523 	 *
1524 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1525 	 *    eIBRS.
1526 	 *
1527 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1528 	 *    RSB clearing.
1529 	 *
1530 	 * So to mitigate all cases, unconditionally fill RSB on context
1531 	 * switches.
1532 	 *
1533 	 * FIXME: Is this pointless for retbleed-affected AMD?
1534 	 */
1535 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1536 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1537 
1538 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1539 
1540 	/*
1541 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1542 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1543 	 * firmware calls only when IBRS / Enhanced / Automatic IBRS aren't
1544 	 * otherwise enabled.
1545 	 *
1546 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1547 	 * the user might select retpoline on the kernel command line and if
1548 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1549 	 * enable IBRS around firmware calls.
1550 	 */
1551 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1552 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1553 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1554 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1555 
1556 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1557 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1558 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1559 		}
1560 
1561 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1562 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1563 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1564 	}
1565 
1566 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1567 	spectre_v2_cmd = cmd;
1568 }
1569 
1570 static void update_stibp_msr(void * __unused)
1571 {
1572 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1573 	update_spec_ctrl(val);
1574 }
1575 
1576 /* Update x86_spec_ctrl_base in case SMT state changed. */
1577 static void update_stibp_strict(void)
1578 {
1579 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1580 
1581 	if (sched_smt_active())
1582 		mask |= SPEC_CTRL_STIBP;
1583 
1584 	if (mask == x86_spec_ctrl_base)
1585 		return;
1586 
1587 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1588 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1589 	x86_spec_ctrl_base = mask;
1590 	on_each_cpu(update_stibp_msr, NULL, 1);
1591 }
1592 
1593 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1594 static void update_indir_branch_cond(void)
1595 {
1596 	if (sched_smt_active())
1597 		static_branch_enable(&switch_to_cond_stibp);
1598 	else
1599 		static_branch_disable(&switch_to_cond_stibp);
1600 }
1601 
1602 #undef pr_fmt
1603 #define pr_fmt(fmt) fmt
1604 
1605 /* Update the static key controlling the MDS CPU buffer clear in idle */
1606 static void update_mds_branch_idle(void)
1607 {
1608 	u64 ia32_cap = x86_read_arch_cap_msr();
1609 
1610 	/*
1611 	 * Enable the idle clearing if SMT is active on CPUs which are
1612 	 * affected only by MSBDS and not any other MDS variant.
1613 	 *
1614 	 * The other variants cannot be mitigated when SMT is enabled, so
1615 	 * clearing the buffers on idle just to prevent the Store Buffer
1616 	 * repartitioning leak would be a window dressing exercise.
1617 	 */
1618 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1619 		return;
1620 
1621 	if (sched_smt_active()) {
1622 		static_branch_enable(&mds_idle_clear);
1623 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1624 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1625 		static_branch_disable(&mds_idle_clear);
1626 	}
1627 }
1628 
1629 #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"
1630 #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"
1631 #define MMIO_MSG_SMT "MMIO Stale Data CPU bug present and SMT on, data leak possible. See https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/processor_mmio_stale_data.html for more details.\n"
1632 
1633 void cpu_bugs_smt_update(void)
1634 {
1635 	mutex_lock(&spec_ctrl_mutex);
1636 
1637 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1638 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1639 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1640 
1641 	switch (spectre_v2_user_stibp) {
1642 	case SPECTRE_V2_USER_NONE:
1643 		break;
1644 	case SPECTRE_V2_USER_STRICT:
1645 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1646 		update_stibp_strict();
1647 		break;
1648 	case SPECTRE_V2_USER_PRCTL:
1649 	case SPECTRE_V2_USER_SECCOMP:
1650 		update_indir_branch_cond();
1651 		break;
1652 	}
1653 
1654 	switch (mds_mitigation) {
1655 	case MDS_MITIGATION_FULL:
1656 	case MDS_MITIGATION_VMWERV:
1657 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1658 			pr_warn_once(MDS_MSG_SMT);
1659 		update_mds_branch_idle();
1660 		break;
1661 	case MDS_MITIGATION_OFF:
1662 		break;
1663 	}
1664 
1665 	switch (taa_mitigation) {
1666 	case TAA_MITIGATION_VERW:
1667 	case TAA_MITIGATION_UCODE_NEEDED:
1668 		if (sched_smt_active())
1669 			pr_warn_once(TAA_MSG_SMT);
1670 		break;
1671 	case TAA_MITIGATION_TSX_DISABLED:
1672 	case TAA_MITIGATION_OFF:
1673 		break;
1674 	}
1675 
1676 	switch (mmio_mitigation) {
1677 	case MMIO_MITIGATION_VERW:
1678 	case MMIO_MITIGATION_UCODE_NEEDED:
1679 		if (sched_smt_active())
1680 			pr_warn_once(MMIO_MSG_SMT);
1681 		break;
1682 	case MMIO_MITIGATION_OFF:
1683 		break;
1684 	}
1685 
1686 	mutex_unlock(&spec_ctrl_mutex);
1687 }
1688 
1689 #undef pr_fmt
1690 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1691 
1692 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1693 
1694 /* The kernel command line selection */
1695 enum ssb_mitigation_cmd {
1696 	SPEC_STORE_BYPASS_CMD_NONE,
1697 	SPEC_STORE_BYPASS_CMD_AUTO,
1698 	SPEC_STORE_BYPASS_CMD_ON,
1699 	SPEC_STORE_BYPASS_CMD_PRCTL,
1700 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1701 };
1702 
1703 static const char * const ssb_strings[] = {
1704 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1705 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1706 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1707 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1708 };
1709 
1710 static const struct {
1711 	const char *option;
1712 	enum ssb_mitigation_cmd cmd;
1713 } ssb_mitigation_options[]  __initconst = {
1714 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1715 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1716 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1717 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1718 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1719 };
1720 
1721 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1722 {
1723 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1724 	char arg[20];
1725 	int ret, i;
1726 
1727 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1728 	    cpu_mitigations_off()) {
1729 		return SPEC_STORE_BYPASS_CMD_NONE;
1730 	} else {
1731 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1732 					  arg, sizeof(arg));
1733 		if (ret < 0)
1734 			return SPEC_STORE_BYPASS_CMD_AUTO;
1735 
1736 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1737 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1738 				continue;
1739 
1740 			cmd = ssb_mitigation_options[i].cmd;
1741 			break;
1742 		}
1743 
1744 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1745 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1746 			return SPEC_STORE_BYPASS_CMD_AUTO;
1747 		}
1748 	}
1749 
1750 	return cmd;
1751 }
1752 
1753 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1754 {
1755 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1756 	enum ssb_mitigation_cmd cmd;
1757 
1758 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1759 		return mode;
1760 
1761 	cmd = ssb_parse_cmdline();
1762 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1763 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1764 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1765 		return mode;
1766 
1767 	switch (cmd) {
1768 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1769 		/*
1770 		 * Choose prctl+seccomp as the default mode if seccomp is
1771 		 * enabled.
1772 		 */
1773 		if (IS_ENABLED(CONFIG_SECCOMP))
1774 			mode = SPEC_STORE_BYPASS_SECCOMP;
1775 		else
1776 			mode = SPEC_STORE_BYPASS_PRCTL;
1777 		break;
1778 	case SPEC_STORE_BYPASS_CMD_ON:
1779 		mode = SPEC_STORE_BYPASS_DISABLE;
1780 		break;
1781 	case SPEC_STORE_BYPASS_CMD_AUTO:
1782 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1783 		mode = SPEC_STORE_BYPASS_PRCTL;
1784 		break;
1785 	case SPEC_STORE_BYPASS_CMD_NONE:
1786 		break;
1787 	}
1788 
1789 	/*
1790 	 * We have three CPU feature flags that are in play here:
1791 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1792 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1793 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1794 	 */
1795 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1796 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1797 		/*
1798 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1799 		 * use a completely different MSR and bit dependent on family.
1800 		 */
1801 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1802 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1803 			x86_amd_ssb_disable();
1804 		} else {
1805 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1806 			update_spec_ctrl(x86_spec_ctrl_base);
1807 		}
1808 	}
1809 
1810 	return mode;
1811 }
1812 
1813 static void ssb_select_mitigation(void)
1814 {
1815 	ssb_mode = __ssb_select_mitigation();
1816 
1817 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1818 		pr_info("%s\n", ssb_strings[ssb_mode]);
1819 }
1820 
1821 #undef pr_fmt
1822 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1823 
1824 static void task_update_spec_tif(struct task_struct *tsk)
1825 {
1826 	/* Force the update of the real TIF bits */
1827 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1828 
1829 	/*
1830 	 * Immediately update the speculation control MSRs for the current
1831 	 * task, but for a non-current task delay setting the CPU
1832 	 * mitigation until it is scheduled next.
1833 	 *
1834 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1835 	 * always the current task.
1836 	 */
1837 	if (tsk == current)
1838 		speculation_ctrl_update_current();
1839 }
1840 
1841 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1842 {
1843 
1844 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1845 		return -EPERM;
1846 
1847 	switch (ctrl) {
1848 	case PR_SPEC_ENABLE:
1849 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1850 		return 0;
1851 	case PR_SPEC_DISABLE:
1852 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1853 		return 0;
1854 	default:
1855 		return -ERANGE;
1856 	}
1857 }
1858 
1859 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1860 {
1861 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1862 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1863 		return -ENXIO;
1864 
1865 	switch (ctrl) {
1866 	case PR_SPEC_ENABLE:
1867 		/* If speculation is force disabled, enable is not allowed */
1868 		if (task_spec_ssb_force_disable(task))
1869 			return -EPERM;
1870 		task_clear_spec_ssb_disable(task);
1871 		task_clear_spec_ssb_noexec(task);
1872 		task_update_spec_tif(task);
1873 		break;
1874 	case PR_SPEC_DISABLE:
1875 		task_set_spec_ssb_disable(task);
1876 		task_clear_spec_ssb_noexec(task);
1877 		task_update_spec_tif(task);
1878 		break;
1879 	case PR_SPEC_FORCE_DISABLE:
1880 		task_set_spec_ssb_disable(task);
1881 		task_set_spec_ssb_force_disable(task);
1882 		task_clear_spec_ssb_noexec(task);
1883 		task_update_spec_tif(task);
1884 		break;
1885 	case PR_SPEC_DISABLE_NOEXEC:
1886 		if (task_spec_ssb_force_disable(task))
1887 			return -EPERM;
1888 		task_set_spec_ssb_disable(task);
1889 		task_set_spec_ssb_noexec(task);
1890 		task_update_spec_tif(task);
1891 		break;
1892 	default:
1893 		return -ERANGE;
1894 	}
1895 	return 0;
1896 }
1897 
1898 static bool is_spec_ib_user_controlled(void)
1899 {
1900 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1901 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1902 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1903 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1904 }
1905 
1906 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1907 {
1908 	switch (ctrl) {
1909 	case PR_SPEC_ENABLE:
1910 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1911 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1912 			return 0;
1913 
1914 		/*
1915 		 * With strict mode for both IBPB and STIBP, the instruction
1916 		 * code paths avoid checking this task flag and instead,
1917 		 * unconditionally run the instruction. However, STIBP and IBPB
1918 		 * are independent and either can be set to conditionally
1919 		 * enabled regardless of the mode of the other.
1920 		 *
1921 		 * If either is set to conditional, allow the task flag to be
1922 		 * updated, unless it was force-disabled by a previous prctl
1923 		 * call. Currently, this is possible on an AMD CPU which has the
1924 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1925 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1926 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1927 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1928 		 */
1929 		if (!is_spec_ib_user_controlled() ||
1930 		    task_spec_ib_force_disable(task))
1931 			return -EPERM;
1932 
1933 		task_clear_spec_ib_disable(task);
1934 		task_update_spec_tif(task);
1935 		break;
1936 	case PR_SPEC_DISABLE:
1937 	case PR_SPEC_FORCE_DISABLE:
1938 		/*
1939 		 * Indirect branch speculation is always allowed when
1940 		 * mitigation is force disabled.
1941 		 */
1942 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1943 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1944 			return -EPERM;
1945 
1946 		if (!is_spec_ib_user_controlled())
1947 			return 0;
1948 
1949 		task_set_spec_ib_disable(task);
1950 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1951 			task_set_spec_ib_force_disable(task);
1952 		task_update_spec_tif(task);
1953 		if (task == current)
1954 			indirect_branch_prediction_barrier();
1955 		break;
1956 	default:
1957 		return -ERANGE;
1958 	}
1959 	return 0;
1960 }
1961 
1962 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1963 			     unsigned long ctrl)
1964 {
1965 	switch (which) {
1966 	case PR_SPEC_STORE_BYPASS:
1967 		return ssb_prctl_set(task, ctrl);
1968 	case PR_SPEC_INDIRECT_BRANCH:
1969 		return ib_prctl_set(task, ctrl);
1970 	case PR_SPEC_L1D_FLUSH:
1971 		return l1d_flush_prctl_set(task, ctrl);
1972 	default:
1973 		return -ENODEV;
1974 	}
1975 }
1976 
1977 #ifdef CONFIG_SECCOMP
1978 void arch_seccomp_spec_mitigate(struct task_struct *task)
1979 {
1980 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1981 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1982 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1983 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1984 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1985 }
1986 #endif
1987 
1988 static int l1d_flush_prctl_get(struct task_struct *task)
1989 {
1990 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1991 		return PR_SPEC_FORCE_DISABLE;
1992 
1993 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
1994 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1995 	else
1996 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1997 }
1998 
1999 static int ssb_prctl_get(struct task_struct *task)
2000 {
2001 	switch (ssb_mode) {
2002 	case SPEC_STORE_BYPASS_DISABLE:
2003 		return PR_SPEC_DISABLE;
2004 	case SPEC_STORE_BYPASS_SECCOMP:
2005 	case SPEC_STORE_BYPASS_PRCTL:
2006 		if (task_spec_ssb_force_disable(task))
2007 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2008 		if (task_spec_ssb_noexec(task))
2009 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2010 		if (task_spec_ssb_disable(task))
2011 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2012 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2013 	default:
2014 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2015 			return PR_SPEC_ENABLE;
2016 		return PR_SPEC_NOT_AFFECTED;
2017 	}
2018 }
2019 
2020 static int ib_prctl_get(struct task_struct *task)
2021 {
2022 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2023 		return PR_SPEC_NOT_AFFECTED;
2024 
2025 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2026 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2027 		return PR_SPEC_ENABLE;
2028 	else if (is_spec_ib_user_controlled()) {
2029 		if (task_spec_ib_force_disable(task))
2030 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2031 		if (task_spec_ib_disable(task))
2032 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2033 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2034 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2035 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2036 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2037 		return PR_SPEC_DISABLE;
2038 	else
2039 		return PR_SPEC_NOT_AFFECTED;
2040 }
2041 
2042 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2043 {
2044 	switch (which) {
2045 	case PR_SPEC_STORE_BYPASS:
2046 		return ssb_prctl_get(task);
2047 	case PR_SPEC_INDIRECT_BRANCH:
2048 		return ib_prctl_get(task);
2049 	case PR_SPEC_L1D_FLUSH:
2050 		return l1d_flush_prctl_get(task);
2051 	default:
2052 		return -ENODEV;
2053 	}
2054 }
2055 
2056 void x86_spec_ctrl_setup_ap(void)
2057 {
2058 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2059 		update_spec_ctrl(x86_spec_ctrl_base);
2060 
2061 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2062 		x86_amd_ssb_disable();
2063 }
2064 
2065 bool itlb_multihit_kvm_mitigation;
2066 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2067 
2068 #undef pr_fmt
2069 #define pr_fmt(fmt)	"L1TF: " fmt
2070 
2071 /* Default mitigation for L1TF-affected CPUs */
2072 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2073 #if IS_ENABLED(CONFIG_KVM_INTEL)
2074 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2075 #endif
2076 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2077 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2078 
2079 /*
2080  * These CPUs all support 44bits physical address space internally in the
2081  * cache but CPUID can report a smaller number of physical address bits.
2082  *
2083  * The L1TF mitigation uses the top most address bit for the inversion of
2084  * non present PTEs. When the installed memory reaches into the top most
2085  * address bit due to memory holes, which has been observed on machines
2086  * which report 36bits physical address bits and have 32G RAM installed,
2087  * then the mitigation range check in l1tf_select_mitigation() triggers.
2088  * This is a false positive because the mitigation is still possible due to
2089  * the fact that the cache uses 44bit internally. Use the cache bits
2090  * instead of the reported physical bits and adjust them on the affected
2091  * machines to 44bit if the reported bits are less than 44.
2092  */
2093 static void override_cache_bits(struct cpuinfo_x86 *c)
2094 {
2095 	if (c->x86 != 6)
2096 		return;
2097 
2098 	switch (c->x86_model) {
2099 	case INTEL_FAM6_NEHALEM:
2100 	case INTEL_FAM6_WESTMERE:
2101 	case INTEL_FAM6_SANDYBRIDGE:
2102 	case INTEL_FAM6_IVYBRIDGE:
2103 	case INTEL_FAM6_HASWELL:
2104 	case INTEL_FAM6_HASWELL_L:
2105 	case INTEL_FAM6_HASWELL_G:
2106 	case INTEL_FAM6_BROADWELL:
2107 	case INTEL_FAM6_BROADWELL_G:
2108 	case INTEL_FAM6_SKYLAKE_L:
2109 	case INTEL_FAM6_SKYLAKE:
2110 	case INTEL_FAM6_KABYLAKE_L:
2111 	case INTEL_FAM6_KABYLAKE:
2112 		if (c->x86_cache_bits < 44)
2113 			c->x86_cache_bits = 44;
2114 		break;
2115 	}
2116 }
2117 
2118 static void __init l1tf_select_mitigation(void)
2119 {
2120 	u64 half_pa;
2121 
2122 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2123 		return;
2124 
2125 	if (cpu_mitigations_off())
2126 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2127 	else if (cpu_mitigations_auto_nosmt())
2128 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2129 
2130 	override_cache_bits(&boot_cpu_data);
2131 
2132 	switch (l1tf_mitigation) {
2133 	case L1TF_MITIGATION_OFF:
2134 	case L1TF_MITIGATION_FLUSH_NOWARN:
2135 	case L1TF_MITIGATION_FLUSH:
2136 		break;
2137 	case L1TF_MITIGATION_FLUSH_NOSMT:
2138 	case L1TF_MITIGATION_FULL:
2139 		cpu_smt_disable(false);
2140 		break;
2141 	case L1TF_MITIGATION_FULL_FORCE:
2142 		cpu_smt_disable(true);
2143 		break;
2144 	}
2145 
2146 #if CONFIG_PGTABLE_LEVELS == 2
2147 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2148 	return;
2149 #endif
2150 
2151 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2152 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2153 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2154 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2155 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2156 				half_pa);
2157 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2158 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2159 		return;
2160 	}
2161 
2162 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2163 }
2164 
2165 static int __init l1tf_cmdline(char *str)
2166 {
2167 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2168 		return 0;
2169 
2170 	if (!str)
2171 		return -EINVAL;
2172 
2173 	if (!strcmp(str, "off"))
2174 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2175 	else if (!strcmp(str, "flush,nowarn"))
2176 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2177 	else if (!strcmp(str, "flush"))
2178 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2179 	else if (!strcmp(str, "flush,nosmt"))
2180 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2181 	else if (!strcmp(str, "full"))
2182 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2183 	else if (!strcmp(str, "full,force"))
2184 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2185 
2186 	return 0;
2187 }
2188 early_param("l1tf", l1tf_cmdline);
2189 
2190 #undef pr_fmt
2191 #define pr_fmt(fmt) fmt
2192 
2193 #ifdef CONFIG_SYSFS
2194 
2195 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2196 
2197 #if IS_ENABLED(CONFIG_KVM_INTEL)
2198 static const char * const l1tf_vmx_states[] = {
2199 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2200 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2201 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2202 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2203 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2204 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2205 };
2206 
2207 static ssize_t l1tf_show_state(char *buf)
2208 {
2209 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2210 		return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2211 
2212 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2213 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2214 	     sched_smt_active())) {
2215 		return sysfs_emit(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2216 				  l1tf_vmx_states[l1tf_vmx_mitigation]);
2217 	}
2218 
2219 	return sysfs_emit(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2220 			  l1tf_vmx_states[l1tf_vmx_mitigation],
2221 			  sched_smt_active() ? "vulnerable" : "disabled");
2222 }
2223 
2224 static ssize_t itlb_multihit_show_state(char *buf)
2225 {
2226 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2227 	    !boot_cpu_has(X86_FEATURE_VMX))
2228 		return sysfs_emit(buf, "KVM: Mitigation: VMX unsupported\n");
2229 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2230 		return sysfs_emit(buf, "KVM: Mitigation: VMX disabled\n");
2231 	else if (itlb_multihit_kvm_mitigation)
2232 		return sysfs_emit(buf, "KVM: Mitigation: Split huge pages\n");
2233 	else
2234 		return sysfs_emit(buf, "KVM: Vulnerable\n");
2235 }
2236 #else
2237 static ssize_t l1tf_show_state(char *buf)
2238 {
2239 	return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2240 }
2241 
2242 static ssize_t itlb_multihit_show_state(char *buf)
2243 {
2244 	return sysfs_emit(buf, "Processor vulnerable\n");
2245 }
2246 #endif
2247 
2248 static ssize_t mds_show_state(char *buf)
2249 {
2250 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2251 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2252 				  mds_strings[mds_mitigation]);
2253 	}
2254 
2255 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2256 		return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2257 				  (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2258 				   sched_smt_active() ? "mitigated" : "disabled"));
2259 	}
2260 
2261 	return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2262 			  sched_smt_active() ? "vulnerable" : "disabled");
2263 }
2264 
2265 static ssize_t tsx_async_abort_show_state(char *buf)
2266 {
2267 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2268 	    (taa_mitigation == TAA_MITIGATION_OFF))
2269 		return sysfs_emit(buf, "%s\n", taa_strings[taa_mitigation]);
2270 
2271 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2272 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2273 				  taa_strings[taa_mitigation]);
2274 	}
2275 
2276 	return sysfs_emit(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2277 			  sched_smt_active() ? "vulnerable" : "disabled");
2278 }
2279 
2280 static ssize_t mmio_stale_data_show_state(char *buf)
2281 {
2282 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2283 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2284 
2285 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2286 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2287 
2288 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2289 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2290 				  mmio_strings[mmio_mitigation]);
2291 	}
2292 
2293 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2294 			  sched_smt_active() ? "vulnerable" : "disabled");
2295 }
2296 
2297 static char *stibp_state(void)
2298 {
2299 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
2300 	    !boot_cpu_has(X86_FEATURE_AUTOIBRS))
2301 		return "";
2302 
2303 	switch (spectre_v2_user_stibp) {
2304 	case SPECTRE_V2_USER_NONE:
2305 		return ", STIBP: disabled";
2306 	case SPECTRE_V2_USER_STRICT:
2307 		return ", STIBP: forced";
2308 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2309 		return ", STIBP: always-on";
2310 	case SPECTRE_V2_USER_PRCTL:
2311 	case SPECTRE_V2_USER_SECCOMP:
2312 		if (static_key_enabled(&switch_to_cond_stibp))
2313 			return ", STIBP: conditional";
2314 	}
2315 	return "";
2316 }
2317 
2318 static char *ibpb_state(void)
2319 {
2320 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2321 		if (static_key_enabled(&switch_mm_always_ibpb))
2322 			return ", IBPB: always-on";
2323 		if (static_key_enabled(&switch_mm_cond_ibpb))
2324 			return ", IBPB: conditional";
2325 		return ", IBPB: disabled";
2326 	}
2327 	return "";
2328 }
2329 
2330 static char *pbrsb_eibrs_state(void)
2331 {
2332 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2333 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2334 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2335 			return ", PBRSB-eIBRS: SW sequence";
2336 		else
2337 			return ", PBRSB-eIBRS: Vulnerable";
2338 	} else {
2339 		return ", PBRSB-eIBRS: Not affected";
2340 	}
2341 }
2342 
2343 static ssize_t spectre_v2_show_state(char *buf)
2344 {
2345 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2346 		return sysfs_emit(buf, "Vulnerable: LFENCE\n");
2347 
2348 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2349 		return sysfs_emit(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2350 
2351 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2352 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2353 		return sysfs_emit(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2354 
2355 	return sysfs_emit(buf, "%s%s%s%s%s%s%s\n",
2356 			  spectre_v2_strings[spectre_v2_enabled],
2357 			  ibpb_state(),
2358 			  boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2359 			  stibp_state(),
2360 			  boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2361 			  pbrsb_eibrs_state(),
2362 			  spectre_v2_module_string());
2363 }
2364 
2365 static ssize_t srbds_show_state(char *buf)
2366 {
2367 	return sysfs_emit(buf, "%s\n", srbds_strings[srbds_mitigation]);
2368 }
2369 
2370 static ssize_t retbleed_show_state(char *buf)
2371 {
2372 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2373 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2374 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2375 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2376 			return sysfs_emit(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2377 
2378 		return sysfs_emit(buf, "%s; SMT %s\n", retbleed_strings[retbleed_mitigation],
2379 				  !sched_smt_active() ? "disabled" :
2380 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2381 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2382 				  "enabled with STIBP protection" : "vulnerable");
2383 	}
2384 
2385 	return sysfs_emit(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2386 }
2387 
2388 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2389 			       char *buf, unsigned int bug)
2390 {
2391 	if (!boot_cpu_has_bug(bug))
2392 		return sysfs_emit(buf, "Not affected\n");
2393 
2394 	switch (bug) {
2395 	case X86_BUG_CPU_MELTDOWN:
2396 		if (boot_cpu_has(X86_FEATURE_PTI))
2397 			return sysfs_emit(buf, "Mitigation: PTI\n");
2398 
2399 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2400 			return sysfs_emit(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2401 
2402 		break;
2403 
2404 	case X86_BUG_SPECTRE_V1:
2405 		return sysfs_emit(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2406 
2407 	case X86_BUG_SPECTRE_V2:
2408 		return spectre_v2_show_state(buf);
2409 
2410 	case X86_BUG_SPEC_STORE_BYPASS:
2411 		return sysfs_emit(buf, "%s\n", ssb_strings[ssb_mode]);
2412 
2413 	case X86_BUG_L1TF:
2414 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2415 			return l1tf_show_state(buf);
2416 		break;
2417 
2418 	case X86_BUG_MDS:
2419 		return mds_show_state(buf);
2420 
2421 	case X86_BUG_TAA:
2422 		return tsx_async_abort_show_state(buf);
2423 
2424 	case X86_BUG_ITLB_MULTIHIT:
2425 		return itlb_multihit_show_state(buf);
2426 
2427 	case X86_BUG_SRBDS:
2428 		return srbds_show_state(buf);
2429 
2430 	case X86_BUG_MMIO_STALE_DATA:
2431 	case X86_BUG_MMIO_UNKNOWN:
2432 		return mmio_stale_data_show_state(buf);
2433 
2434 	case X86_BUG_RETBLEED:
2435 		return retbleed_show_state(buf);
2436 
2437 	default:
2438 		break;
2439 	}
2440 
2441 	return sysfs_emit(buf, "Vulnerable\n");
2442 }
2443 
2444 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2445 {
2446 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2447 }
2448 
2449 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2450 {
2451 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2452 }
2453 
2454 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2455 {
2456 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2457 }
2458 
2459 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2460 {
2461 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2462 }
2463 
2464 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2465 {
2466 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2467 }
2468 
2469 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2470 {
2471 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2472 }
2473 
2474 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2475 {
2476 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2477 }
2478 
2479 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2480 {
2481 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2482 }
2483 
2484 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2485 {
2486 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2487 }
2488 
2489 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2490 {
2491 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2492 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2493 	else
2494 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2495 }
2496 
2497 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2498 {
2499 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2500 }
2501 #endif
2502