xref: /linux/arch/x86/kernel/cpu/bugs.c (revision f694f30e81c4ade358eb8c75273bac1a48f0cb8f)
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/cpu_device_id.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 static void __init srso_select_mitigation(void);
51 static void __init gds_select_mitigation(void);
52 
53 /* The base value of the SPEC_CTRL MSR without task-specific bits set */
54 u64 x86_spec_ctrl_base;
55 EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
56 
57 /* The current value of the SPEC_CTRL MSR with task-specific bits set */
58 DEFINE_PER_CPU(u64, x86_spec_ctrl_current);
59 EXPORT_PER_CPU_SYMBOL_GPL(x86_spec_ctrl_current);
60 
61 u64 x86_pred_cmd __ro_after_init = PRED_CMD_IBPB;
62 
63 static u64 __ro_after_init x86_arch_cap_msr;
64 
65 static DEFINE_MUTEX(spec_ctrl_mutex);
66 
67 void (*x86_return_thunk)(void) __ro_after_init = __x86_return_thunk;
68 
69 /* Update SPEC_CTRL MSR and its cached copy unconditionally */
70 static void update_spec_ctrl(u64 val)
71 {
72 	this_cpu_write(x86_spec_ctrl_current, val);
73 	wrmsrl(MSR_IA32_SPEC_CTRL, val);
74 }
75 
76 /*
77  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
78  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
79  */
80 void update_spec_ctrl_cond(u64 val)
81 {
82 	if (this_cpu_read(x86_spec_ctrl_current) == val)
83 		return;
84 
85 	this_cpu_write(x86_spec_ctrl_current, val);
86 
87 	/*
88 	 * When KERNEL_IBRS this MSR is written on return-to-user, unless
89 	 * forced the update can be delayed until that time.
90 	 */
91 	if (!cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
92 		wrmsrl(MSR_IA32_SPEC_CTRL, val);
93 }
94 
95 noinstr u64 spec_ctrl_current(void)
96 {
97 	return this_cpu_read(x86_spec_ctrl_current);
98 }
99 EXPORT_SYMBOL_GPL(spec_ctrl_current);
100 
101 /*
102  * AMD specific MSR info for Speculative Store Bypass control.
103  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
104  */
105 u64 __ro_after_init x86_amd_ls_cfg_base;
106 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
107 
108 /* Control conditional STIBP in switch_to() */
109 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
110 /* Control conditional IBPB in switch_mm() */
111 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
112 /* Control unconditional IBPB in switch_mm() */
113 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
114 
115 /* Control IBPB on vCPU load */
116 DEFINE_STATIC_KEY_FALSE(switch_vcpu_ibpb);
117 EXPORT_SYMBOL_GPL(switch_vcpu_ibpb);
118 
119 /* Control MDS CPU buffer clear before idling (halt, mwait) */
120 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
121 EXPORT_SYMBOL_GPL(mds_idle_clear);
122 
123 /*
124  * Controls whether l1d flush based mitigations are enabled,
125  * based on hw features and admin setting via boot parameter
126  * defaults to false
127  */
128 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
129 
130 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
131 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
132 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
133 
134 void __init cpu_select_mitigations(void)
135 {
136 	/*
137 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
138 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
139 	 * init code as it is not enumerated and depends on the family.
140 	 */
141 	if (cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL)) {
142 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
143 
144 		/*
145 		 * Previously running kernel (kexec), may have some controls
146 		 * turned ON. Clear them and let the mitigations setup below
147 		 * rediscover them based on configuration.
148 		 */
149 		x86_spec_ctrl_base &= ~SPEC_CTRL_MITIGATIONS_MASK;
150 	}
151 
152 	x86_arch_cap_msr = x86_read_arch_cap_msr();
153 
154 	/* Select the proper CPU mitigations before patching alternatives: */
155 	spectre_v1_select_mitigation();
156 	spectre_v2_select_mitigation();
157 	/*
158 	 * retbleed_select_mitigation() relies on the state set by
159 	 * spectre_v2_select_mitigation(); specifically it wants to know about
160 	 * spectre_v2=ibrs.
161 	 */
162 	retbleed_select_mitigation();
163 	/*
164 	 * spectre_v2_user_select_mitigation() relies on the state set by
165 	 * retbleed_select_mitigation(); specifically the STIBP selection is
166 	 * forced for UNRET or IBPB.
167 	 */
168 	spectre_v2_user_select_mitigation();
169 	ssb_select_mitigation();
170 	l1tf_select_mitigation();
171 	md_clear_select_mitigation();
172 	srbds_select_mitigation();
173 	l1d_flush_select_mitigation();
174 
175 	/*
176 	 * srso_select_mitigation() depends and must run after
177 	 * retbleed_select_mitigation().
178 	 */
179 	srso_select_mitigation();
180 	gds_select_mitigation();
181 }
182 
183 /*
184  * NOTE: This function is *only* called for SVM, since Intel uses
185  * MSR_IA32_SPEC_CTRL for SSBD.
186  */
187 void
188 x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl, bool setguest)
189 {
190 	u64 guestval, hostval;
191 	struct thread_info *ti = current_thread_info();
192 
193 	/*
194 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
195 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
196 	 */
197 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
198 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
199 		return;
200 
201 	/*
202 	 * If the host has SSBD mitigation enabled, force it in the host's
203 	 * virtual MSR value. If its not permanently enabled, evaluate
204 	 * current's TIF_SSBD thread flag.
205 	 */
206 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
207 		hostval = SPEC_CTRL_SSBD;
208 	else
209 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
210 
211 	/* Sanitize the guest value */
212 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
213 
214 	if (hostval != guestval) {
215 		unsigned long tif;
216 
217 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
218 				 ssbd_spec_ctrl_to_tif(hostval);
219 
220 		speculation_ctrl_update(tif);
221 	}
222 }
223 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
224 
225 static void x86_amd_ssb_disable(void)
226 {
227 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
228 
229 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
230 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
231 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
232 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
233 }
234 
235 #undef pr_fmt
236 #define pr_fmt(fmt)	"MDS: " fmt
237 
238 /* Default mitigation for MDS-affected CPUs */
239 static enum mds_mitigations mds_mitigation __ro_after_init =
240 	IS_ENABLED(CONFIG_MITIGATION_MDS) ? MDS_MITIGATION_AUTO : MDS_MITIGATION_OFF;
241 static bool mds_nosmt __ro_after_init = false;
242 
243 static const char * const mds_strings[] = {
244 	[MDS_MITIGATION_OFF]	= "Vulnerable",
245 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
246 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
247 };
248 
249 enum taa_mitigations {
250 	TAA_MITIGATION_OFF,
251 	TAA_MITIGATION_AUTO,
252 	TAA_MITIGATION_UCODE_NEEDED,
253 	TAA_MITIGATION_VERW,
254 	TAA_MITIGATION_TSX_DISABLED,
255 };
256 
257 /* Default mitigation for TAA-affected CPUs */
258 static enum taa_mitigations taa_mitigation __ro_after_init =
259 	IS_ENABLED(CONFIG_MITIGATION_TAA) ? TAA_MITIGATION_AUTO : TAA_MITIGATION_OFF;
260 
261 enum mmio_mitigations {
262 	MMIO_MITIGATION_OFF,
263 	MMIO_MITIGATION_AUTO,
264 	MMIO_MITIGATION_UCODE_NEEDED,
265 	MMIO_MITIGATION_VERW,
266 };
267 
268 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
269 static enum mmio_mitigations mmio_mitigation __ro_after_init =
270 	IS_ENABLED(CONFIG_MITIGATION_MMIO_STALE_DATA) ?	MMIO_MITIGATION_AUTO : MMIO_MITIGATION_OFF;
271 
272 enum rfds_mitigations {
273 	RFDS_MITIGATION_OFF,
274 	RFDS_MITIGATION_AUTO,
275 	RFDS_MITIGATION_VERW,
276 	RFDS_MITIGATION_UCODE_NEEDED,
277 };
278 
279 /* Default mitigation for Register File Data Sampling */
280 static enum rfds_mitigations rfds_mitigation __ro_after_init =
281 	IS_ENABLED(CONFIG_MITIGATION_RFDS) ? RFDS_MITIGATION_AUTO : RFDS_MITIGATION_OFF;
282 
283 static void __init mds_select_mitigation(void)
284 {
285 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
286 		mds_mitigation = MDS_MITIGATION_OFF;
287 		return;
288 	}
289 
290 	if (mds_mitigation == MDS_MITIGATION_AUTO)
291 		mds_mitigation = MDS_MITIGATION_FULL;
292 
293 	if (mds_mitigation == MDS_MITIGATION_FULL) {
294 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
295 			mds_mitigation = MDS_MITIGATION_VMWERV;
296 
297 		setup_force_cpu_cap(X86_FEATURE_CLEAR_CPU_BUF);
298 
299 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
300 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
301 			cpu_smt_disable(false);
302 	}
303 }
304 
305 static int __init mds_cmdline(char *str)
306 {
307 	if (!boot_cpu_has_bug(X86_BUG_MDS))
308 		return 0;
309 
310 	if (!str)
311 		return -EINVAL;
312 
313 	if (!strcmp(str, "off"))
314 		mds_mitigation = MDS_MITIGATION_OFF;
315 	else if (!strcmp(str, "full"))
316 		mds_mitigation = MDS_MITIGATION_FULL;
317 	else if (!strcmp(str, "full,nosmt")) {
318 		mds_mitigation = MDS_MITIGATION_FULL;
319 		mds_nosmt = true;
320 	}
321 
322 	return 0;
323 }
324 early_param("mds", mds_cmdline);
325 
326 #undef pr_fmt
327 #define pr_fmt(fmt)	"TAA: " fmt
328 
329 static bool taa_nosmt __ro_after_init;
330 
331 static const char * const taa_strings[] = {
332 	[TAA_MITIGATION_OFF]		= "Vulnerable",
333 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
334 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
335 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
336 };
337 
338 static void __init taa_select_mitigation(void)
339 {
340 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
341 		taa_mitigation = TAA_MITIGATION_OFF;
342 		return;
343 	}
344 
345 	/* TSX previously disabled by tsx=off */
346 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
347 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
348 		return;
349 	}
350 
351 	if (cpu_mitigations_off()) {
352 		taa_mitigation = TAA_MITIGATION_OFF;
353 		return;
354 	}
355 
356 	/*
357 	 * TAA mitigation via VERW is turned off if both
358 	 * tsx_async_abort=off and mds=off are specified.
359 	 */
360 	if (taa_mitigation == TAA_MITIGATION_OFF &&
361 	    mds_mitigation == MDS_MITIGATION_OFF)
362 		return;
363 
364 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
365 		taa_mitigation = TAA_MITIGATION_VERW;
366 	else
367 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
368 
369 	/*
370 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
371 	 * A microcode update fixes this behavior to clear CPU buffers. It also
372 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
373 	 * ARCH_CAP_TSX_CTRL_MSR bit.
374 	 *
375 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
376 	 * update is required.
377 	 */
378 	if ( (x86_arch_cap_msr & ARCH_CAP_MDS_NO) &&
379 	    !(x86_arch_cap_msr & ARCH_CAP_TSX_CTRL_MSR))
380 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
381 
382 	/*
383 	 * TSX is enabled, select alternate mitigation for TAA which is
384 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
385 	 *
386 	 * For guests that can't determine whether the correct microcode is
387 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
388 	 */
389 	setup_force_cpu_cap(X86_FEATURE_CLEAR_CPU_BUF);
390 
391 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
392 		cpu_smt_disable(false);
393 }
394 
395 static int __init tsx_async_abort_parse_cmdline(char *str)
396 {
397 	if (!boot_cpu_has_bug(X86_BUG_TAA))
398 		return 0;
399 
400 	if (!str)
401 		return -EINVAL;
402 
403 	if (!strcmp(str, "off")) {
404 		taa_mitigation = TAA_MITIGATION_OFF;
405 	} else if (!strcmp(str, "full")) {
406 		taa_mitigation = TAA_MITIGATION_VERW;
407 	} else if (!strcmp(str, "full,nosmt")) {
408 		taa_mitigation = TAA_MITIGATION_VERW;
409 		taa_nosmt = true;
410 	}
411 
412 	return 0;
413 }
414 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
415 
416 #undef pr_fmt
417 #define pr_fmt(fmt)	"MMIO Stale Data: " fmt
418 
419 static bool mmio_nosmt __ro_after_init = false;
420 
421 static const char * const mmio_strings[] = {
422 	[MMIO_MITIGATION_OFF]		= "Vulnerable",
423 	[MMIO_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
424 	[MMIO_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
425 };
426 
427 static void __init mmio_select_mitigation(void)
428 {
429 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
430 	     boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
431 	     cpu_mitigations_off()) {
432 		mmio_mitigation = MMIO_MITIGATION_OFF;
433 		return;
434 	}
435 
436 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
437 		return;
438 
439 	/*
440 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
441 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
442 	 */
443 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
444 					      boot_cpu_has(X86_FEATURE_RTM)))
445 		setup_force_cpu_cap(X86_FEATURE_CLEAR_CPU_BUF);
446 
447 	/*
448 	 * X86_FEATURE_CLEAR_CPU_BUF could be enabled by other VERW based
449 	 * mitigations, disable KVM-only mitigation in that case.
450 	 */
451 	if (boot_cpu_has(X86_FEATURE_CLEAR_CPU_BUF))
452 		static_branch_disable(&mmio_stale_data_clear);
453 	else
454 		static_branch_enable(&mmio_stale_data_clear);
455 
456 	/*
457 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
458 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
459 	 * is required irrespective of SMT state.
460 	 */
461 	if (!(x86_arch_cap_msr & ARCH_CAP_FBSDP_NO))
462 		static_branch_enable(&mds_idle_clear);
463 
464 	/*
465 	 * Check if the system has the right microcode.
466 	 *
467 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
468 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
469 	 * affected systems.
470 	 */
471 	if ((x86_arch_cap_msr & ARCH_CAP_FB_CLEAR) ||
472 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
473 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
474 	     !(x86_arch_cap_msr & ARCH_CAP_MDS_NO)))
475 		mmio_mitigation = MMIO_MITIGATION_VERW;
476 	else
477 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
478 
479 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
480 		cpu_smt_disable(false);
481 }
482 
483 static int __init mmio_stale_data_parse_cmdline(char *str)
484 {
485 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
486 		return 0;
487 
488 	if (!str)
489 		return -EINVAL;
490 
491 	if (!strcmp(str, "off")) {
492 		mmio_mitigation = MMIO_MITIGATION_OFF;
493 	} else if (!strcmp(str, "full")) {
494 		mmio_mitigation = MMIO_MITIGATION_VERW;
495 	} else if (!strcmp(str, "full,nosmt")) {
496 		mmio_mitigation = MMIO_MITIGATION_VERW;
497 		mmio_nosmt = true;
498 	}
499 
500 	return 0;
501 }
502 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
503 
504 #undef pr_fmt
505 #define pr_fmt(fmt)	"Register File Data Sampling: " fmt
506 
507 static const char * const rfds_strings[] = {
508 	[RFDS_MITIGATION_OFF]			= "Vulnerable",
509 	[RFDS_MITIGATION_VERW]			= "Mitigation: Clear Register File",
510 	[RFDS_MITIGATION_UCODE_NEEDED]		= "Vulnerable: No microcode",
511 };
512 
513 static void __init rfds_select_mitigation(void)
514 {
515 	if (!boot_cpu_has_bug(X86_BUG_RFDS) || cpu_mitigations_off()) {
516 		rfds_mitigation = RFDS_MITIGATION_OFF;
517 		return;
518 	}
519 	if (rfds_mitigation == RFDS_MITIGATION_OFF)
520 		return;
521 
522 	if (rfds_mitigation == RFDS_MITIGATION_AUTO)
523 		rfds_mitigation = RFDS_MITIGATION_VERW;
524 
525 	if (x86_arch_cap_msr & ARCH_CAP_RFDS_CLEAR)
526 		setup_force_cpu_cap(X86_FEATURE_CLEAR_CPU_BUF);
527 	else
528 		rfds_mitigation = RFDS_MITIGATION_UCODE_NEEDED;
529 }
530 
531 static __init int rfds_parse_cmdline(char *str)
532 {
533 	if (!str)
534 		return -EINVAL;
535 
536 	if (!boot_cpu_has_bug(X86_BUG_RFDS))
537 		return 0;
538 
539 	if (!strcmp(str, "off"))
540 		rfds_mitigation = RFDS_MITIGATION_OFF;
541 	else if (!strcmp(str, "on"))
542 		rfds_mitigation = RFDS_MITIGATION_VERW;
543 
544 	return 0;
545 }
546 early_param("reg_file_data_sampling", rfds_parse_cmdline);
547 
548 #undef pr_fmt
549 #define pr_fmt(fmt)     "" fmt
550 
551 static void __init md_clear_update_mitigation(void)
552 {
553 	if (cpu_mitigations_off())
554 		return;
555 
556 	if (!boot_cpu_has(X86_FEATURE_CLEAR_CPU_BUF))
557 		goto out;
558 
559 	/*
560 	 * X86_FEATURE_CLEAR_CPU_BUF is now enabled. Update MDS, TAA and MMIO
561 	 * Stale Data mitigation, if necessary.
562 	 */
563 	if (mds_mitigation == MDS_MITIGATION_OFF &&
564 	    boot_cpu_has_bug(X86_BUG_MDS)) {
565 		mds_mitigation = MDS_MITIGATION_FULL;
566 		mds_select_mitigation();
567 	}
568 	if (taa_mitigation == TAA_MITIGATION_OFF &&
569 	    boot_cpu_has_bug(X86_BUG_TAA)) {
570 		taa_mitigation = TAA_MITIGATION_VERW;
571 		taa_select_mitigation();
572 	}
573 	/*
574 	 * MMIO_MITIGATION_OFF is not checked here so that mmio_stale_data_clear
575 	 * gets updated correctly as per X86_FEATURE_CLEAR_CPU_BUF state.
576 	 */
577 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
578 		mmio_mitigation = MMIO_MITIGATION_VERW;
579 		mmio_select_mitigation();
580 	}
581 	if (rfds_mitigation == RFDS_MITIGATION_OFF &&
582 	    boot_cpu_has_bug(X86_BUG_RFDS)) {
583 		rfds_mitigation = RFDS_MITIGATION_VERW;
584 		rfds_select_mitigation();
585 	}
586 out:
587 	if (boot_cpu_has_bug(X86_BUG_MDS))
588 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
589 	if (boot_cpu_has_bug(X86_BUG_TAA))
590 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
591 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
592 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
593 	else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
594 		pr_info("MMIO Stale Data: Unknown: No mitigations\n");
595 	if (boot_cpu_has_bug(X86_BUG_RFDS))
596 		pr_info("Register File Data Sampling: %s\n", rfds_strings[rfds_mitigation]);
597 }
598 
599 static void __init md_clear_select_mitigation(void)
600 {
601 	mds_select_mitigation();
602 	taa_select_mitigation();
603 	mmio_select_mitigation();
604 	rfds_select_mitigation();
605 
606 	/*
607 	 * As these mitigations are inter-related and rely on VERW instruction
608 	 * to clear the microarchitural buffers, update and print their status
609 	 * after mitigation selection is done for each of these vulnerabilities.
610 	 */
611 	md_clear_update_mitigation();
612 }
613 
614 #undef pr_fmt
615 #define pr_fmt(fmt)	"SRBDS: " fmt
616 
617 enum srbds_mitigations {
618 	SRBDS_MITIGATION_OFF,
619 	SRBDS_MITIGATION_UCODE_NEEDED,
620 	SRBDS_MITIGATION_FULL,
621 	SRBDS_MITIGATION_TSX_OFF,
622 	SRBDS_MITIGATION_HYPERVISOR,
623 };
624 
625 static enum srbds_mitigations srbds_mitigation __ro_after_init =
626 	IS_ENABLED(CONFIG_MITIGATION_SRBDS) ? SRBDS_MITIGATION_FULL : SRBDS_MITIGATION_OFF;
627 
628 static const char * const srbds_strings[] = {
629 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
630 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
631 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
632 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
633 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
634 };
635 
636 static bool srbds_off;
637 
638 void update_srbds_msr(void)
639 {
640 	u64 mcu_ctrl;
641 
642 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
643 		return;
644 
645 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
646 		return;
647 
648 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
649 		return;
650 
651 	/*
652 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
653 	 * being disabled and it hasn't received the SRBDS MSR microcode.
654 	 */
655 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
656 		return;
657 
658 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
659 
660 	switch (srbds_mitigation) {
661 	case SRBDS_MITIGATION_OFF:
662 	case SRBDS_MITIGATION_TSX_OFF:
663 		mcu_ctrl |= RNGDS_MITG_DIS;
664 		break;
665 	case SRBDS_MITIGATION_FULL:
666 		mcu_ctrl &= ~RNGDS_MITG_DIS;
667 		break;
668 	default:
669 		break;
670 	}
671 
672 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
673 }
674 
675 static void __init srbds_select_mitigation(void)
676 {
677 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
678 		return;
679 
680 	/*
681 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
682 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
683 	 * by Processor MMIO Stale Data vulnerability.
684 	 */
685 	if ((x86_arch_cap_msr & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
686 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
687 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
688 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
689 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
690 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
691 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
692 	else if (cpu_mitigations_off() || srbds_off)
693 		srbds_mitigation = SRBDS_MITIGATION_OFF;
694 
695 	update_srbds_msr();
696 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
697 }
698 
699 static int __init srbds_parse_cmdline(char *str)
700 {
701 	if (!str)
702 		return -EINVAL;
703 
704 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
705 		return 0;
706 
707 	srbds_off = !strcmp(str, "off");
708 	return 0;
709 }
710 early_param("srbds", srbds_parse_cmdline);
711 
712 #undef pr_fmt
713 #define pr_fmt(fmt)     "L1D Flush : " fmt
714 
715 enum l1d_flush_mitigations {
716 	L1D_FLUSH_OFF = 0,
717 	L1D_FLUSH_ON,
718 };
719 
720 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
721 
722 static void __init l1d_flush_select_mitigation(void)
723 {
724 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
725 		return;
726 
727 	static_branch_enable(&switch_mm_cond_l1d_flush);
728 	pr_info("Conditional flush on switch_mm() enabled\n");
729 }
730 
731 static int __init l1d_flush_parse_cmdline(char *str)
732 {
733 	if (!strcmp(str, "on"))
734 		l1d_flush_mitigation = L1D_FLUSH_ON;
735 
736 	return 0;
737 }
738 early_param("l1d_flush", l1d_flush_parse_cmdline);
739 
740 #undef pr_fmt
741 #define pr_fmt(fmt)	"GDS: " fmt
742 
743 enum gds_mitigations {
744 	GDS_MITIGATION_OFF,
745 	GDS_MITIGATION_UCODE_NEEDED,
746 	GDS_MITIGATION_FORCE,
747 	GDS_MITIGATION_FULL,
748 	GDS_MITIGATION_FULL_LOCKED,
749 	GDS_MITIGATION_HYPERVISOR,
750 };
751 
752 static enum gds_mitigations gds_mitigation __ro_after_init =
753 	IS_ENABLED(CONFIG_MITIGATION_GDS) ? GDS_MITIGATION_FULL : GDS_MITIGATION_OFF;
754 
755 static const char * const gds_strings[] = {
756 	[GDS_MITIGATION_OFF]		= "Vulnerable",
757 	[GDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
758 	[GDS_MITIGATION_FORCE]		= "Mitigation: AVX disabled, no microcode",
759 	[GDS_MITIGATION_FULL]		= "Mitigation: Microcode",
760 	[GDS_MITIGATION_FULL_LOCKED]	= "Mitigation: Microcode (locked)",
761 	[GDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
762 };
763 
764 bool gds_ucode_mitigated(void)
765 {
766 	return (gds_mitigation == GDS_MITIGATION_FULL ||
767 		gds_mitigation == GDS_MITIGATION_FULL_LOCKED);
768 }
769 EXPORT_SYMBOL_GPL(gds_ucode_mitigated);
770 
771 void update_gds_msr(void)
772 {
773 	u64 mcu_ctrl_after;
774 	u64 mcu_ctrl;
775 
776 	switch (gds_mitigation) {
777 	case GDS_MITIGATION_OFF:
778 		rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
779 		mcu_ctrl |= GDS_MITG_DIS;
780 		break;
781 	case GDS_MITIGATION_FULL_LOCKED:
782 		/*
783 		 * The LOCKED state comes from the boot CPU. APs might not have
784 		 * the same state. Make sure the mitigation is enabled on all
785 		 * CPUs.
786 		 */
787 	case GDS_MITIGATION_FULL:
788 		rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
789 		mcu_ctrl &= ~GDS_MITG_DIS;
790 		break;
791 	case GDS_MITIGATION_FORCE:
792 	case GDS_MITIGATION_UCODE_NEEDED:
793 	case GDS_MITIGATION_HYPERVISOR:
794 		return;
795 	}
796 
797 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
798 
799 	/*
800 	 * Check to make sure that the WRMSR value was not ignored. Writes to
801 	 * GDS_MITG_DIS will be ignored if this processor is locked but the boot
802 	 * processor was not.
803 	 */
804 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl_after);
805 	WARN_ON_ONCE(mcu_ctrl != mcu_ctrl_after);
806 }
807 
808 static void __init gds_select_mitigation(void)
809 {
810 	u64 mcu_ctrl;
811 
812 	if (!boot_cpu_has_bug(X86_BUG_GDS))
813 		return;
814 
815 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
816 		gds_mitigation = GDS_MITIGATION_HYPERVISOR;
817 		goto out;
818 	}
819 
820 	if (cpu_mitigations_off())
821 		gds_mitigation = GDS_MITIGATION_OFF;
822 	/* Will verify below that mitigation _can_ be disabled */
823 
824 	/* No microcode */
825 	if (!(x86_arch_cap_msr & ARCH_CAP_GDS_CTRL)) {
826 		if (gds_mitigation == GDS_MITIGATION_FORCE) {
827 			/*
828 			 * This only needs to be done on the boot CPU so do it
829 			 * here rather than in update_gds_msr()
830 			 */
831 			setup_clear_cpu_cap(X86_FEATURE_AVX);
832 			pr_warn("Microcode update needed! Disabling AVX as mitigation.\n");
833 		} else {
834 			gds_mitigation = GDS_MITIGATION_UCODE_NEEDED;
835 		}
836 		goto out;
837 	}
838 
839 	/* Microcode has mitigation, use it */
840 	if (gds_mitigation == GDS_MITIGATION_FORCE)
841 		gds_mitigation = GDS_MITIGATION_FULL;
842 
843 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
844 	if (mcu_ctrl & GDS_MITG_LOCKED) {
845 		if (gds_mitigation == GDS_MITIGATION_OFF)
846 			pr_warn("Mitigation locked. Disable failed.\n");
847 
848 		/*
849 		 * The mitigation is selected from the boot CPU. All other CPUs
850 		 * _should_ have the same state. If the boot CPU isn't locked
851 		 * but others are then update_gds_msr() will WARN() of the state
852 		 * mismatch. If the boot CPU is locked update_gds_msr() will
853 		 * ensure the other CPUs have the mitigation enabled.
854 		 */
855 		gds_mitigation = GDS_MITIGATION_FULL_LOCKED;
856 	}
857 
858 	update_gds_msr();
859 out:
860 	pr_info("%s\n", gds_strings[gds_mitigation]);
861 }
862 
863 static int __init gds_parse_cmdline(char *str)
864 {
865 	if (!str)
866 		return -EINVAL;
867 
868 	if (!boot_cpu_has_bug(X86_BUG_GDS))
869 		return 0;
870 
871 	if (!strcmp(str, "off"))
872 		gds_mitigation = GDS_MITIGATION_OFF;
873 	else if (!strcmp(str, "force"))
874 		gds_mitigation = GDS_MITIGATION_FORCE;
875 
876 	return 0;
877 }
878 early_param("gather_data_sampling", gds_parse_cmdline);
879 
880 #undef pr_fmt
881 #define pr_fmt(fmt)     "Spectre V1 : " fmt
882 
883 enum spectre_v1_mitigation {
884 	SPECTRE_V1_MITIGATION_NONE,
885 	SPECTRE_V1_MITIGATION_AUTO,
886 };
887 
888 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
889 	IS_ENABLED(CONFIG_MITIGATION_SPECTRE_V1) ?
890 		SPECTRE_V1_MITIGATION_AUTO : SPECTRE_V1_MITIGATION_NONE;
891 
892 static const char * const spectre_v1_strings[] = {
893 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
894 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
895 };
896 
897 /*
898  * Does SMAP provide full mitigation against speculative kernel access to
899  * userspace?
900  */
901 static bool smap_works_speculatively(void)
902 {
903 	if (!boot_cpu_has(X86_FEATURE_SMAP))
904 		return false;
905 
906 	/*
907 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
908 	 * prevent speculative access to user data in the L1 cache.
909 	 * Consider SMAP to be non-functional as a mitigation on these
910 	 * CPUs.
911 	 */
912 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
913 		return false;
914 
915 	return true;
916 }
917 
918 static void __init spectre_v1_select_mitigation(void)
919 {
920 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
921 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
922 		return;
923 	}
924 
925 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
926 		/*
927 		 * With Spectre v1, a user can speculatively control either
928 		 * path of a conditional swapgs with a user-controlled GS
929 		 * value.  The mitigation is to add lfences to both code paths.
930 		 *
931 		 * If FSGSBASE is enabled, the user can put a kernel address in
932 		 * GS, in which case SMAP provides no protection.
933 		 *
934 		 * If FSGSBASE is disabled, the user can only put a user space
935 		 * address in GS.  That makes an attack harder, but still
936 		 * possible if there's no SMAP protection.
937 		 */
938 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
939 		    !smap_works_speculatively()) {
940 			/*
941 			 * Mitigation can be provided from SWAPGS itself or
942 			 * PTI as the CR3 write in the Meltdown mitigation
943 			 * is serializing.
944 			 *
945 			 * If neither is there, mitigate with an LFENCE to
946 			 * stop speculation through swapgs.
947 			 */
948 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
949 			    !boot_cpu_has(X86_FEATURE_PTI))
950 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
951 
952 			/*
953 			 * Enable lfences in the kernel entry (non-swapgs)
954 			 * paths, to prevent user entry from speculatively
955 			 * skipping swapgs.
956 			 */
957 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
958 		}
959 	}
960 
961 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
962 }
963 
964 static int __init nospectre_v1_cmdline(char *str)
965 {
966 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
967 	return 0;
968 }
969 early_param("nospectre_v1", nospectre_v1_cmdline);
970 
971 enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init = SPECTRE_V2_NONE;
972 
973 #undef pr_fmt
974 #define pr_fmt(fmt)     "RETBleed: " fmt
975 
976 enum retbleed_mitigation {
977 	RETBLEED_MITIGATION_NONE,
978 	RETBLEED_MITIGATION_UNRET,
979 	RETBLEED_MITIGATION_IBPB,
980 	RETBLEED_MITIGATION_IBRS,
981 	RETBLEED_MITIGATION_EIBRS,
982 	RETBLEED_MITIGATION_STUFF,
983 };
984 
985 enum retbleed_mitigation_cmd {
986 	RETBLEED_CMD_OFF,
987 	RETBLEED_CMD_AUTO,
988 	RETBLEED_CMD_UNRET,
989 	RETBLEED_CMD_IBPB,
990 	RETBLEED_CMD_STUFF,
991 };
992 
993 static const char * const retbleed_strings[] = {
994 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
995 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
996 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
997 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
998 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
999 	[RETBLEED_MITIGATION_STUFF]	= "Mitigation: Stuffing",
1000 };
1001 
1002 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
1003 	RETBLEED_MITIGATION_NONE;
1004 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
1005 	IS_ENABLED(CONFIG_MITIGATION_RETBLEED) ? RETBLEED_CMD_AUTO : RETBLEED_CMD_OFF;
1006 
1007 static int __ro_after_init retbleed_nosmt = false;
1008 
1009 static int __init retbleed_parse_cmdline(char *str)
1010 {
1011 	if (!str)
1012 		return -EINVAL;
1013 
1014 	while (str) {
1015 		char *next = strchr(str, ',');
1016 		if (next) {
1017 			*next = 0;
1018 			next++;
1019 		}
1020 
1021 		if (!strcmp(str, "off")) {
1022 			retbleed_cmd = RETBLEED_CMD_OFF;
1023 		} else if (!strcmp(str, "auto")) {
1024 			retbleed_cmd = RETBLEED_CMD_AUTO;
1025 		} else if (!strcmp(str, "unret")) {
1026 			retbleed_cmd = RETBLEED_CMD_UNRET;
1027 		} else if (!strcmp(str, "ibpb")) {
1028 			retbleed_cmd = RETBLEED_CMD_IBPB;
1029 		} else if (!strcmp(str, "stuff")) {
1030 			retbleed_cmd = RETBLEED_CMD_STUFF;
1031 		} else if (!strcmp(str, "nosmt")) {
1032 			retbleed_nosmt = true;
1033 		} else if (!strcmp(str, "force")) {
1034 			setup_force_cpu_bug(X86_BUG_RETBLEED);
1035 		} else {
1036 			pr_err("Ignoring unknown retbleed option (%s).", str);
1037 		}
1038 
1039 		str = next;
1040 	}
1041 
1042 	return 0;
1043 }
1044 early_param("retbleed", retbleed_parse_cmdline);
1045 
1046 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
1047 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
1048 
1049 static void __init retbleed_select_mitigation(void)
1050 {
1051 	bool mitigate_smt = false;
1052 
1053 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
1054 		return;
1055 
1056 	switch (retbleed_cmd) {
1057 	case RETBLEED_CMD_OFF:
1058 		return;
1059 
1060 	case RETBLEED_CMD_UNRET:
1061 		if (IS_ENABLED(CONFIG_MITIGATION_UNRET_ENTRY)) {
1062 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
1063 		} else {
1064 			pr_err("WARNING: kernel not compiled with MITIGATION_UNRET_ENTRY.\n");
1065 			goto do_cmd_auto;
1066 		}
1067 		break;
1068 
1069 	case RETBLEED_CMD_IBPB:
1070 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
1071 			pr_err("WARNING: CPU does not support IBPB.\n");
1072 			goto do_cmd_auto;
1073 		} else if (IS_ENABLED(CONFIG_MITIGATION_IBPB_ENTRY)) {
1074 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
1075 		} else {
1076 			pr_err("WARNING: kernel not compiled with MITIGATION_IBPB_ENTRY.\n");
1077 			goto do_cmd_auto;
1078 		}
1079 		break;
1080 
1081 	case RETBLEED_CMD_STUFF:
1082 		if (IS_ENABLED(CONFIG_MITIGATION_CALL_DEPTH_TRACKING) &&
1083 		    spectre_v2_enabled == SPECTRE_V2_RETPOLINE) {
1084 			retbleed_mitigation = RETBLEED_MITIGATION_STUFF;
1085 
1086 		} else {
1087 			if (IS_ENABLED(CONFIG_MITIGATION_CALL_DEPTH_TRACKING))
1088 				pr_err("WARNING: retbleed=stuff depends on spectre_v2=retpoline\n");
1089 			else
1090 				pr_err("WARNING: kernel not compiled with MITIGATION_CALL_DEPTH_TRACKING.\n");
1091 
1092 			goto do_cmd_auto;
1093 		}
1094 		break;
1095 
1096 do_cmd_auto:
1097 	case RETBLEED_CMD_AUTO:
1098 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1099 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
1100 			if (IS_ENABLED(CONFIG_MITIGATION_UNRET_ENTRY))
1101 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
1102 			else if (IS_ENABLED(CONFIG_MITIGATION_IBPB_ENTRY) &&
1103 				 boot_cpu_has(X86_FEATURE_IBPB))
1104 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
1105 		}
1106 
1107 		/*
1108 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
1109 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
1110 		 * be set accordingly below.
1111 		 */
1112 
1113 		break;
1114 	}
1115 
1116 	switch (retbleed_mitigation) {
1117 	case RETBLEED_MITIGATION_UNRET:
1118 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
1119 		setup_force_cpu_cap(X86_FEATURE_UNRET);
1120 
1121 		x86_return_thunk = retbleed_return_thunk;
1122 
1123 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
1124 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
1125 			pr_err(RETBLEED_UNTRAIN_MSG);
1126 
1127 		mitigate_smt = true;
1128 		break;
1129 
1130 	case RETBLEED_MITIGATION_IBPB:
1131 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
1132 		setup_force_cpu_cap(X86_FEATURE_IBPB_ON_VMEXIT);
1133 		mitigate_smt = true;
1134 
1135 		/*
1136 		 * IBPB on entry already obviates the need for
1137 		 * software-based untraining so clear those in case some
1138 		 * other mitigation like SRSO has selected them.
1139 		 */
1140 		setup_clear_cpu_cap(X86_FEATURE_UNRET);
1141 		setup_clear_cpu_cap(X86_FEATURE_RETHUNK);
1142 
1143 		/*
1144 		 * There is no need for RSB filling: write_ibpb() ensures
1145 		 * all predictions, including the RSB, are invalidated,
1146 		 * regardless of IBPB implementation.
1147 		 */
1148 		setup_clear_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1149 
1150 		break;
1151 
1152 	case RETBLEED_MITIGATION_STUFF:
1153 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
1154 		setup_force_cpu_cap(X86_FEATURE_CALL_DEPTH);
1155 
1156 		x86_return_thunk = call_depth_return_thunk;
1157 		break;
1158 
1159 	default:
1160 		break;
1161 	}
1162 
1163 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
1164 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
1165 		cpu_smt_disable(false);
1166 
1167 	/*
1168 	 * Let IBRS trump all on Intel without affecting the effects of the
1169 	 * retbleed= cmdline option except for call depth based stuffing
1170 	 */
1171 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1172 		switch (spectre_v2_enabled) {
1173 		case SPECTRE_V2_IBRS:
1174 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
1175 			break;
1176 		case SPECTRE_V2_EIBRS:
1177 		case SPECTRE_V2_EIBRS_RETPOLINE:
1178 		case SPECTRE_V2_EIBRS_LFENCE:
1179 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
1180 			break;
1181 		default:
1182 			if (retbleed_mitigation != RETBLEED_MITIGATION_STUFF)
1183 				pr_err(RETBLEED_INTEL_MSG);
1184 		}
1185 	}
1186 
1187 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
1188 }
1189 
1190 #undef pr_fmt
1191 #define pr_fmt(fmt)     "Spectre V2 : " fmt
1192 
1193 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
1194 	SPECTRE_V2_USER_NONE;
1195 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
1196 	SPECTRE_V2_USER_NONE;
1197 
1198 #ifdef CONFIG_MITIGATION_RETPOLINE
1199 static bool spectre_v2_bad_module;
1200 
1201 bool retpoline_module_ok(bool has_retpoline)
1202 {
1203 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
1204 		return true;
1205 
1206 	pr_err("System may be vulnerable to spectre v2\n");
1207 	spectre_v2_bad_module = true;
1208 	return false;
1209 }
1210 
1211 static inline const char *spectre_v2_module_string(void)
1212 {
1213 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
1214 }
1215 #else
1216 static inline const char *spectre_v2_module_string(void) { return ""; }
1217 #endif
1218 
1219 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
1220 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
1221 #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"
1222 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
1223 
1224 #ifdef CONFIG_BPF_SYSCALL
1225 void unpriv_ebpf_notify(int new_state)
1226 {
1227 	if (new_state)
1228 		return;
1229 
1230 	/* Unprivileged eBPF is enabled */
1231 
1232 	switch (spectre_v2_enabled) {
1233 	case SPECTRE_V2_EIBRS:
1234 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1235 		break;
1236 	case SPECTRE_V2_EIBRS_LFENCE:
1237 		if (sched_smt_active())
1238 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1239 		break;
1240 	default:
1241 		break;
1242 	}
1243 }
1244 #endif
1245 
1246 static inline bool match_option(const char *arg, int arglen, const char *opt)
1247 {
1248 	int len = strlen(opt);
1249 
1250 	return len == arglen && !strncmp(arg, opt, len);
1251 }
1252 
1253 /* The kernel command line selection for spectre v2 */
1254 enum spectre_v2_mitigation_cmd {
1255 	SPECTRE_V2_CMD_NONE,
1256 	SPECTRE_V2_CMD_AUTO,
1257 	SPECTRE_V2_CMD_FORCE,
1258 	SPECTRE_V2_CMD_RETPOLINE,
1259 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1260 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1261 	SPECTRE_V2_CMD_EIBRS,
1262 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1263 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1264 	SPECTRE_V2_CMD_IBRS,
1265 };
1266 
1267 enum spectre_v2_user_cmd {
1268 	SPECTRE_V2_USER_CMD_NONE,
1269 	SPECTRE_V2_USER_CMD_AUTO,
1270 	SPECTRE_V2_USER_CMD_FORCE,
1271 	SPECTRE_V2_USER_CMD_PRCTL,
1272 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1273 	SPECTRE_V2_USER_CMD_SECCOMP,
1274 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1275 };
1276 
1277 static const char * const spectre_v2_user_strings[] = {
1278 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1279 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1280 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1281 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1282 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1283 };
1284 
1285 static const struct {
1286 	const char			*option;
1287 	enum spectre_v2_user_cmd	cmd;
1288 	bool				secure;
1289 } v2_user_options[] __initconst = {
1290 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1291 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1292 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1293 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1294 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1295 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1296 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1297 };
1298 
1299 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1300 {
1301 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1302 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1303 }
1304 
1305 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1306 
1307 static enum spectre_v2_user_cmd __init
1308 spectre_v2_parse_user_cmdline(void)
1309 {
1310 	enum spectre_v2_user_cmd mode;
1311 	char arg[20];
1312 	int ret, i;
1313 
1314 	mode = IS_ENABLED(CONFIG_MITIGATION_SPECTRE_V2) ?
1315 		SPECTRE_V2_USER_CMD_AUTO : SPECTRE_V2_USER_CMD_NONE;
1316 
1317 	switch (spectre_v2_cmd) {
1318 	case SPECTRE_V2_CMD_NONE:
1319 		return SPECTRE_V2_USER_CMD_NONE;
1320 	case SPECTRE_V2_CMD_FORCE:
1321 		return SPECTRE_V2_USER_CMD_FORCE;
1322 	default:
1323 		break;
1324 	}
1325 
1326 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1327 				  arg, sizeof(arg));
1328 	if (ret < 0)
1329 		return mode;
1330 
1331 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1332 		if (match_option(arg, ret, v2_user_options[i].option)) {
1333 			spec_v2_user_print_cond(v2_user_options[i].option,
1334 						v2_user_options[i].secure);
1335 			return v2_user_options[i].cmd;
1336 		}
1337 	}
1338 
1339 	pr_err("Unknown user space protection option (%s). Switching to default\n", arg);
1340 	return mode;
1341 }
1342 
1343 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1344 {
1345 	return spectre_v2_in_eibrs_mode(mode) || mode == SPECTRE_V2_IBRS;
1346 }
1347 
1348 static void __init
1349 spectre_v2_user_select_mitigation(void)
1350 {
1351 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1352 	enum spectre_v2_user_cmd cmd;
1353 
1354 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1355 		return;
1356 
1357 	cmd = spectre_v2_parse_user_cmdline();
1358 	switch (cmd) {
1359 	case SPECTRE_V2_USER_CMD_NONE:
1360 		goto set_mode;
1361 	case SPECTRE_V2_USER_CMD_FORCE:
1362 		mode = SPECTRE_V2_USER_STRICT;
1363 		break;
1364 	case SPECTRE_V2_USER_CMD_AUTO:
1365 	case SPECTRE_V2_USER_CMD_PRCTL:
1366 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1367 		mode = SPECTRE_V2_USER_PRCTL;
1368 		break;
1369 	case SPECTRE_V2_USER_CMD_SECCOMP:
1370 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1371 		if (IS_ENABLED(CONFIG_SECCOMP))
1372 			mode = SPECTRE_V2_USER_SECCOMP;
1373 		else
1374 			mode = SPECTRE_V2_USER_PRCTL;
1375 		break;
1376 	}
1377 
1378 	/* Initialize Indirect Branch Prediction Barrier */
1379 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1380 		static_branch_enable(&switch_vcpu_ibpb);
1381 
1382 		spectre_v2_user_ibpb = mode;
1383 		switch (cmd) {
1384 		case SPECTRE_V2_USER_CMD_NONE:
1385 			break;
1386 		case SPECTRE_V2_USER_CMD_FORCE:
1387 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1388 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1389 			static_branch_enable(&switch_mm_always_ibpb);
1390 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1391 			break;
1392 		case SPECTRE_V2_USER_CMD_PRCTL:
1393 		case SPECTRE_V2_USER_CMD_AUTO:
1394 		case SPECTRE_V2_USER_CMD_SECCOMP:
1395 			static_branch_enable(&switch_mm_cond_ibpb);
1396 			break;
1397 		}
1398 
1399 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1400 			static_key_enabled(&switch_mm_always_ibpb) ?
1401 			"always-on" : "conditional");
1402 	}
1403 
1404 	/*
1405 	 * If no STIBP, Intel enhanced IBRS is enabled, or SMT impossible, STIBP
1406 	 * is not required.
1407 	 *
1408 	 * Intel's Enhanced IBRS also protects against cross-thread branch target
1409 	 * injection in user-mode as the IBRS bit remains always set which
1410 	 * implicitly enables cross-thread protections.  However, in legacy IBRS
1411 	 * mode, the IBRS bit is set only on kernel entry and cleared on return
1412 	 * to userspace.  AMD Automatic IBRS also does not protect userspace.
1413 	 * These modes therefore disable the implicit cross-thread protection,
1414 	 * so allow for STIBP to be selected in those cases.
1415 	 */
1416 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1417 	    !cpu_smt_possible() ||
1418 	    (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
1419 	     !boot_cpu_has(X86_FEATURE_AUTOIBRS)))
1420 		return;
1421 
1422 	/*
1423 	 * At this point, an STIBP mode other than "off" has been set.
1424 	 * If STIBP support is not being forced, check if STIBP always-on
1425 	 * is preferred.
1426 	 */
1427 	if (mode != SPECTRE_V2_USER_STRICT &&
1428 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1429 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1430 
1431 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1432 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1433 		if (mode != SPECTRE_V2_USER_STRICT &&
1434 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1435 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1436 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1437 	}
1438 
1439 	spectre_v2_user_stibp = mode;
1440 
1441 set_mode:
1442 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1443 }
1444 
1445 static const char * const spectre_v2_strings[] = {
1446 	[SPECTRE_V2_NONE]			= "Vulnerable",
1447 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1448 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1449 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced / Automatic IBRS",
1450 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced / Automatic IBRS + LFENCE",
1451 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced / Automatic IBRS + Retpolines",
1452 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1453 };
1454 
1455 static const struct {
1456 	const char *option;
1457 	enum spectre_v2_mitigation_cmd cmd;
1458 	bool secure;
1459 } mitigation_options[] __initconst = {
1460 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1461 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1462 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1463 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1464 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1465 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1466 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1467 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1468 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1469 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1470 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1471 };
1472 
1473 static void __init spec_v2_print_cond(const char *reason, bool secure)
1474 {
1475 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1476 		pr_info("%s selected on command line.\n", reason);
1477 }
1478 
1479 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1480 {
1481 	enum spectre_v2_mitigation_cmd cmd;
1482 	char arg[20];
1483 	int ret, i;
1484 
1485 	cmd = IS_ENABLED(CONFIG_MITIGATION_SPECTRE_V2) ?  SPECTRE_V2_CMD_AUTO : SPECTRE_V2_CMD_NONE;
1486 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1487 	    cpu_mitigations_off())
1488 		return SPECTRE_V2_CMD_NONE;
1489 
1490 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1491 	if (ret < 0)
1492 		return cmd;
1493 
1494 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1495 		if (!match_option(arg, ret, mitigation_options[i].option))
1496 			continue;
1497 		cmd = mitigation_options[i].cmd;
1498 		break;
1499 	}
1500 
1501 	if (i >= ARRAY_SIZE(mitigation_options)) {
1502 		pr_err("unknown option (%s). Switching to default mode\n", arg);
1503 		return cmd;
1504 	}
1505 
1506 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1507 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1508 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1509 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1510 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1511 	    !IS_ENABLED(CONFIG_MITIGATION_RETPOLINE)) {
1512 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1513 		       mitigation_options[i].option);
1514 		return SPECTRE_V2_CMD_AUTO;
1515 	}
1516 
1517 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1518 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1519 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1520 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1521 		pr_err("%s selected but CPU doesn't have Enhanced or Automatic IBRS. Switching to AUTO select\n",
1522 		       mitigation_options[i].option);
1523 		return SPECTRE_V2_CMD_AUTO;
1524 	}
1525 
1526 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1527 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1528 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1529 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1530 		       mitigation_options[i].option);
1531 		return SPECTRE_V2_CMD_AUTO;
1532 	}
1533 
1534 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_MITIGATION_IBRS_ENTRY)) {
1535 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1536 		       mitigation_options[i].option);
1537 		return SPECTRE_V2_CMD_AUTO;
1538 	}
1539 
1540 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1541 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1542 		       mitigation_options[i].option);
1543 		return SPECTRE_V2_CMD_AUTO;
1544 	}
1545 
1546 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1547 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1548 		       mitigation_options[i].option);
1549 		return SPECTRE_V2_CMD_AUTO;
1550 	}
1551 
1552 	if (cmd == SPECTRE_V2_CMD_IBRS && cpu_feature_enabled(X86_FEATURE_XENPV)) {
1553 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1554 		       mitigation_options[i].option);
1555 		return SPECTRE_V2_CMD_AUTO;
1556 	}
1557 
1558 	spec_v2_print_cond(mitigation_options[i].option,
1559 			   mitigation_options[i].secure);
1560 	return cmd;
1561 }
1562 
1563 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1564 {
1565 	if (!IS_ENABLED(CONFIG_MITIGATION_RETPOLINE)) {
1566 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1567 		return SPECTRE_V2_NONE;
1568 	}
1569 
1570 	return SPECTRE_V2_RETPOLINE;
1571 }
1572 
1573 static bool __ro_after_init rrsba_disabled;
1574 
1575 /* Disable in-kernel use of non-RSB RET predictors */
1576 static void __init spec_ctrl_disable_kernel_rrsba(void)
1577 {
1578 	if (rrsba_disabled)
1579 		return;
1580 
1581 	if (!(x86_arch_cap_msr & ARCH_CAP_RRSBA)) {
1582 		rrsba_disabled = true;
1583 		return;
1584 	}
1585 
1586 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1587 		return;
1588 
1589 	x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1590 	update_spec_ctrl(x86_spec_ctrl_base);
1591 	rrsba_disabled = true;
1592 }
1593 
1594 static void __init spectre_v2_select_rsb_mitigation(enum spectre_v2_mitigation mode)
1595 {
1596 	/*
1597 	 * WARNING! There are many subtleties to consider when changing *any*
1598 	 * code related to RSB-related mitigations.  Before doing so, carefully
1599 	 * read the following document, and update if necessary:
1600 	 *
1601 	 *   Documentation/admin-guide/hw-vuln/rsb.rst
1602 	 *
1603 	 * In an overly simplified nutshell:
1604 	 *
1605 	 *   - User->user RSB attacks are conditionally mitigated during
1606 	 *     context switches by cond_mitigation -> write_ibpb().
1607 	 *
1608 	 *   - User->kernel and guest->host attacks are mitigated by eIBRS or
1609 	 *     RSB filling.
1610 	 *
1611 	 *     Though, depending on config, note that other alternative
1612 	 *     mitigations may end up getting used instead, e.g., IBPB on
1613 	 *     entry/vmexit, call depth tracking, or return thunks.
1614 	 */
1615 
1616 	switch (mode) {
1617 	case SPECTRE_V2_NONE:
1618 		break;
1619 
1620 	case SPECTRE_V2_EIBRS:
1621 	case SPECTRE_V2_EIBRS_LFENCE:
1622 	case SPECTRE_V2_EIBRS_RETPOLINE:
1623 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1624 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1625 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1626 		}
1627 		break;
1628 
1629 	case SPECTRE_V2_RETPOLINE:
1630 	case SPECTRE_V2_LFENCE:
1631 	case SPECTRE_V2_IBRS:
1632 		pr_info("Spectre v2 / SpectreRSB: Filling RSB on context switch and VMEXIT\n");
1633 		setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1634 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1635 		break;
1636 
1637 	default:
1638 		pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation\n");
1639 		dump_stack();
1640 		break;
1641 	}
1642 }
1643 
1644 /*
1645  * Set BHI_DIS_S to prevent indirect branches in kernel to be influenced by
1646  * branch history in userspace. Not needed if BHI_NO is set.
1647  */
1648 static bool __init spec_ctrl_bhi_dis(void)
1649 {
1650 	if (!boot_cpu_has(X86_FEATURE_BHI_CTRL))
1651 		return false;
1652 
1653 	x86_spec_ctrl_base |= SPEC_CTRL_BHI_DIS_S;
1654 	update_spec_ctrl(x86_spec_ctrl_base);
1655 	setup_force_cpu_cap(X86_FEATURE_CLEAR_BHB_HW);
1656 
1657 	return true;
1658 }
1659 
1660 enum bhi_mitigations {
1661 	BHI_MITIGATION_OFF,
1662 	BHI_MITIGATION_ON,
1663 	BHI_MITIGATION_VMEXIT_ONLY,
1664 };
1665 
1666 static enum bhi_mitigations bhi_mitigation __ro_after_init =
1667 	IS_ENABLED(CONFIG_MITIGATION_SPECTRE_BHI) ? BHI_MITIGATION_ON : BHI_MITIGATION_OFF;
1668 
1669 static int __init spectre_bhi_parse_cmdline(char *str)
1670 {
1671 	if (!str)
1672 		return -EINVAL;
1673 
1674 	if (!strcmp(str, "off"))
1675 		bhi_mitigation = BHI_MITIGATION_OFF;
1676 	else if (!strcmp(str, "on"))
1677 		bhi_mitigation = BHI_MITIGATION_ON;
1678 	else if (!strcmp(str, "vmexit"))
1679 		bhi_mitigation = BHI_MITIGATION_VMEXIT_ONLY;
1680 	else
1681 		pr_err("Ignoring unknown spectre_bhi option (%s)", str);
1682 
1683 	return 0;
1684 }
1685 early_param("spectre_bhi", spectre_bhi_parse_cmdline);
1686 
1687 static void __init bhi_select_mitigation(void)
1688 {
1689 	if (bhi_mitigation == BHI_MITIGATION_OFF)
1690 		return;
1691 
1692 	/* Retpoline mitigates against BHI unless the CPU has RRSBA behavior */
1693 	if (boot_cpu_has(X86_FEATURE_RETPOLINE) &&
1694 	    !boot_cpu_has(X86_FEATURE_RETPOLINE_LFENCE)) {
1695 		spec_ctrl_disable_kernel_rrsba();
1696 		if (rrsba_disabled)
1697 			return;
1698 	}
1699 
1700 	/* Mitigate in hardware if supported */
1701 	if (spec_ctrl_bhi_dis())
1702 		return;
1703 
1704 	if (!IS_ENABLED(CONFIG_X86_64))
1705 		return;
1706 
1707 	if (bhi_mitigation == BHI_MITIGATION_VMEXIT_ONLY) {
1708 		pr_info("Spectre BHI mitigation: SW BHB clearing on VM exit only\n");
1709 		setup_force_cpu_cap(X86_FEATURE_CLEAR_BHB_LOOP_ON_VMEXIT);
1710 		return;
1711 	}
1712 
1713 	pr_info("Spectre BHI mitigation: SW BHB clearing on syscall and VM exit\n");
1714 	setup_force_cpu_cap(X86_FEATURE_CLEAR_BHB_LOOP);
1715 	setup_force_cpu_cap(X86_FEATURE_CLEAR_BHB_LOOP_ON_VMEXIT);
1716 }
1717 
1718 static void __init spectre_v2_select_mitigation(void)
1719 {
1720 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1721 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1722 
1723 	/*
1724 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1725 	 * then nothing to do.
1726 	 */
1727 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1728 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1729 		return;
1730 
1731 	switch (cmd) {
1732 	case SPECTRE_V2_CMD_NONE:
1733 		return;
1734 
1735 	case SPECTRE_V2_CMD_FORCE:
1736 	case SPECTRE_V2_CMD_AUTO:
1737 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1738 			mode = SPECTRE_V2_EIBRS;
1739 			break;
1740 		}
1741 
1742 		if (IS_ENABLED(CONFIG_MITIGATION_IBRS_ENTRY) &&
1743 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1744 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1745 		    retbleed_cmd != RETBLEED_CMD_STUFF &&
1746 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1747 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1748 			mode = SPECTRE_V2_IBRS;
1749 			break;
1750 		}
1751 
1752 		mode = spectre_v2_select_retpoline();
1753 		break;
1754 
1755 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1756 		pr_err(SPECTRE_V2_LFENCE_MSG);
1757 		mode = SPECTRE_V2_LFENCE;
1758 		break;
1759 
1760 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1761 		mode = SPECTRE_V2_RETPOLINE;
1762 		break;
1763 
1764 	case SPECTRE_V2_CMD_RETPOLINE:
1765 		mode = spectre_v2_select_retpoline();
1766 		break;
1767 
1768 	case SPECTRE_V2_CMD_IBRS:
1769 		mode = SPECTRE_V2_IBRS;
1770 		break;
1771 
1772 	case SPECTRE_V2_CMD_EIBRS:
1773 		mode = SPECTRE_V2_EIBRS;
1774 		break;
1775 
1776 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1777 		mode = SPECTRE_V2_EIBRS_LFENCE;
1778 		break;
1779 
1780 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1781 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1782 		break;
1783 	}
1784 
1785 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1786 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1787 
1788 	if (spectre_v2_in_ibrs_mode(mode)) {
1789 		if (boot_cpu_has(X86_FEATURE_AUTOIBRS)) {
1790 			msr_set_bit(MSR_EFER, _EFER_AUTOIBRS);
1791 		} else {
1792 			x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1793 			update_spec_ctrl(x86_spec_ctrl_base);
1794 		}
1795 	}
1796 
1797 	switch (mode) {
1798 	case SPECTRE_V2_NONE:
1799 	case SPECTRE_V2_EIBRS:
1800 		break;
1801 
1802 	case SPECTRE_V2_IBRS:
1803 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1804 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1805 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1806 		break;
1807 
1808 	case SPECTRE_V2_LFENCE:
1809 	case SPECTRE_V2_EIBRS_LFENCE:
1810 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1811 		fallthrough;
1812 
1813 	case SPECTRE_V2_RETPOLINE:
1814 	case SPECTRE_V2_EIBRS_RETPOLINE:
1815 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1816 		break;
1817 	}
1818 
1819 	/*
1820 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1821 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1822 	 * prediction from a non-RSB predictor is still a risk.
1823 	 */
1824 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1825 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1826 	    mode == SPECTRE_V2_RETPOLINE)
1827 		spec_ctrl_disable_kernel_rrsba();
1828 
1829 	if (boot_cpu_has(X86_BUG_BHI))
1830 		bhi_select_mitigation();
1831 
1832 	spectre_v2_enabled = mode;
1833 	pr_info("%s\n", spectre_v2_strings[mode]);
1834 
1835 	spectre_v2_select_rsb_mitigation(mode);
1836 
1837 	/*
1838 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1839 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1840 	 * firmware calls only when IBRS / Enhanced / Automatic IBRS aren't
1841 	 * otherwise enabled.
1842 	 *
1843 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1844 	 * the user might select retpoline on the kernel command line and if
1845 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1846 	 * enable IBRS around firmware calls.
1847 	 */
1848 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1849 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1850 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1851 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1852 
1853 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1854 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1855 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1856 		}
1857 
1858 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1859 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1860 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1861 	}
1862 
1863 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1864 	spectre_v2_cmd = cmd;
1865 }
1866 
1867 static void update_stibp_msr(void * __unused)
1868 {
1869 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1870 	update_spec_ctrl(val);
1871 }
1872 
1873 /* Update x86_spec_ctrl_base in case SMT state changed. */
1874 static void update_stibp_strict(void)
1875 {
1876 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1877 
1878 	if (sched_smt_active())
1879 		mask |= SPEC_CTRL_STIBP;
1880 
1881 	if (mask == x86_spec_ctrl_base)
1882 		return;
1883 
1884 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1885 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1886 	x86_spec_ctrl_base = mask;
1887 	on_each_cpu(update_stibp_msr, NULL, 1);
1888 }
1889 
1890 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1891 static void update_indir_branch_cond(void)
1892 {
1893 	if (sched_smt_active())
1894 		static_branch_enable(&switch_to_cond_stibp);
1895 	else
1896 		static_branch_disable(&switch_to_cond_stibp);
1897 }
1898 
1899 #undef pr_fmt
1900 #define pr_fmt(fmt) fmt
1901 
1902 /* Update the static key controlling the MDS CPU buffer clear in idle */
1903 static void update_mds_branch_idle(void)
1904 {
1905 	/*
1906 	 * Enable the idle clearing if SMT is active on CPUs which are
1907 	 * affected only by MSBDS and not any other MDS variant.
1908 	 *
1909 	 * The other variants cannot be mitigated when SMT is enabled, so
1910 	 * clearing the buffers on idle just to prevent the Store Buffer
1911 	 * repartitioning leak would be a window dressing exercise.
1912 	 */
1913 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1914 		return;
1915 
1916 	if (sched_smt_active()) {
1917 		static_branch_enable(&mds_idle_clear);
1918 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1919 		   (x86_arch_cap_msr & ARCH_CAP_FBSDP_NO)) {
1920 		static_branch_disable(&mds_idle_clear);
1921 	}
1922 }
1923 
1924 #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"
1925 #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"
1926 #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"
1927 
1928 void cpu_bugs_smt_update(void)
1929 {
1930 	mutex_lock(&spec_ctrl_mutex);
1931 
1932 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1933 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1934 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1935 
1936 	switch (spectre_v2_user_stibp) {
1937 	case SPECTRE_V2_USER_NONE:
1938 		break;
1939 	case SPECTRE_V2_USER_STRICT:
1940 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1941 		update_stibp_strict();
1942 		break;
1943 	case SPECTRE_V2_USER_PRCTL:
1944 	case SPECTRE_V2_USER_SECCOMP:
1945 		update_indir_branch_cond();
1946 		break;
1947 	}
1948 
1949 	switch (mds_mitigation) {
1950 	case MDS_MITIGATION_FULL:
1951 	case MDS_MITIGATION_AUTO:
1952 	case MDS_MITIGATION_VMWERV:
1953 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1954 			pr_warn_once(MDS_MSG_SMT);
1955 		update_mds_branch_idle();
1956 		break;
1957 	case MDS_MITIGATION_OFF:
1958 		break;
1959 	}
1960 
1961 	switch (taa_mitigation) {
1962 	case TAA_MITIGATION_VERW:
1963 	case TAA_MITIGATION_AUTO:
1964 	case TAA_MITIGATION_UCODE_NEEDED:
1965 		if (sched_smt_active())
1966 			pr_warn_once(TAA_MSG_SMT);
1967 		break;
1968 	case TAA_MITIGATION_TSX_DISABLED:
1969 	case TAA_MITIGATION_OFF:
1970 		break;
1971 	}
1972 
1973 	switch (mmio_mitigation) {
1974 	case MMIO_MITIGATION_VERW:
1975 	case MMIO_MITIGATION_AUTO:
1976 	case MMIO_MITIGATION_UCODE_NEEDED:
1977 		if (sched_smt_active())
1978 			pr_warn_once(MMIO_MSG_SMT);
1979 		break;
1980 	case MMIO_MITIGATION_OFF:
1981 		break;
1982 	}
1983 
1984 	mutex_unlock(&spec_ctrl_mutex);
1985 }
1986 
1987 #undef pr_fmt
1988 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1989 
1990 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1991 
1992 /* The kernel command line selection */
1993 enum ssb_mitigation_cmd {
1994 	SPEC_STORE_BYPASS_CMD_NONE,
1995 	SPEC_STORE_BYPASS_CMD_AUTO,
1996 	SPEC_STORE_BYPASS_CMD_ON,
1997 	SPEC_STORE_BYPASS_CMD_PRCTL,
1998 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1999 };
2000 
2001 static const char * const ssb_strings[] = {
2002 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
2003 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
2004 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
2005 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
2006 };
2007 
2008 static const struct {
2009 	const char *option;
2010 	enum ssb_mitigation_cmd cmd;
2011 } ssb_mitigation_options[]  __initconst = {
2012 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
2013 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
2014 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
2015 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
2016 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
2017 };
2018 
2019 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
2020 {
2021 	enum ssb_mitigation_cmd cmd;
2022 	char arg[20];
2023 	int ret, i;
2024 
2025 	cmd = IS_ENABLED(CONFIG_MITIGATION_SSB) ?
2026 		SPEC_STORE_BYPASS_CMD_AUTO : SPEC_STORE_BYPASS_CMD_NONE;
2027 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
2028 	    cpu_mitigations_off()) {
2029 		return SPEC_STORE_BYPASS_CMD_NONE;
2030 	} else {
2031 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
2032 					  arg, sizeof(arg));
2033 		if (ret < 0)
2034 			return cmd;
2035 
2036 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
2037 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
2038 				continue;
2039 
2040 			cmd = ssb_mitigation_options[i].cmd;
2041 			break;
2042 		}
2043 
2044 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
2045 			pr_err("unknown option (%s). Switching to default mode\n", arg);
2046 			return cmd;
2047 		}
2048 	}
2049 
2050 	return cmd;
2051 }
2052 
2053 static enum ssb_mitigation __init __ssb_select_mitigation(void)
2054 {
2055 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
2056 	enum ssb_mitigation_cmd cmd;
2057 
2058 	if (!boot_cpu_has(X86_FEATURE_SSBD))
2059 		return mode;
2060 
2061 	cmd = ssb_parse_cmdline();
2062 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
2063 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
2064 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
2065 		return mode;
2066 
2067 	switch (cmd) {
2068 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
2069 		/*
2070 		 * Choose prctl+seccomp as the default mode if seccomp is
2071 		 * enabled.
2072 		 */
2073 		if (IS_ENABLED(CONFIG_SECCOMP))
2074 			mode = SPEC_STORE_BYPASS_SECCOMP;
2075 		else
2076 			mode = SPEC_STORE_BYPASS_PRCTL;
2077 		break;
2078 	case SPEC_STORE_BYPASS_CMD_ON:
2079 		mode = SPEC_STORE_BYPASS_DISABLE;
2080 		break;
2081 	case SPEC_STORE_BYPASS_CMD_AUTO:
2082 	case SPEC_STORE_BYPASS_CMD_PRCTL:
2083 		mode = SPEC_STORE_BYPASS_PRCTL;
2084 		break;
2085 	case SPEC_STORE_BYPASS_CMD_NONE:
2086 		break;
2087 	}
2088 
2089 	/*
2090 	 * We have three CPU feature flags that are in play here:
2091 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
2092 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
2093 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
2094 	 */
2095 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
2096 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
2097 		/*
2098 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
2099 		 * use a completely different MSR and bit dependent on family.
2100 		 */
2101 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
2102 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
2103 			x86_amd_ssb_disable();
2104 		} else {
2105 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
2106 			update_spec_ctrl(x86_spec_ctrl_base);
2107 		}
2108 	}
2109 
2110 	return mode;
2111 }
2112 
2113 static void ssb_select_mitigation(void)
2114 {
2115 	ssb_mode = __ssb_select_mitigation();
2116 
2117 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2118 		pr_info("%s\n", ssb_strings[ssb_mode]);
2119 }
2120 
2121 #undef pr_fmt
2122 #define pr_fmt(fmt)     "Speculation prctl: " fmt
2123 
2124 static void task_update_spec_tif(struct task_struct *tsk)
2125 {
2126 	/* Force the update of the real TIF bits */
2127 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
2128 
2129 	/*
2130 	 * Immediately update the speculation control MSRs for the current
2131 	 * task, but for a non-current task delay setting the CPU
2132 	 * mitigation until it is scheduled next.
2133 	 *
2134 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
2135 	 * always the current task.
2136 	 */
2137 	if (tsk == current)
2138 		speculation_ctrl_update_current();
2139 }
2140 
2141 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
2142 {
2143 
2144 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2145 		return -EPERM;
2146 
2147 	switch (ctrl) {
2148 	case PR_SPEC_ENABLE:
2149 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
2150 		return 0;
2151 	case PR_SPEC_DISABLE:
2152 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
2153 		return 0;
2154 	default:
2155 		return -ERANGE;
2156 	}
2157 }
2158 
2159 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
2160 {
2161 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
2162 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
2163 		return -ENXIO;
2164 
2165 	switch (ctrl) {
2166 	case PR_SPEC_ENABLE:
2167 		/* If speculation is force disabled, enable is not allowed */
2168 		if (task_spec_ssb_force_disable(task))
2169 			return -EPERM;
2170 		task_clear_spec_ssb_disable(task);
2171 		task_clear_spec_ssb_noexec(task);
2172 		task_update_spec_tif(task);
2173 		break;
2174 	case PR_SPEC_DISABLE:
2175 		task_set_spec_ssb_disable(task);
2176 		task_clear_spec_ssb_noexec(task);
2177 		task_update_spec_tif(task);
2178 		break;
2179 	case PR_SPEC_FORCE_DISABLE:
2180 		task_set_spec_ssb_disable(task);
2181 		task_set_spec_ssb_force_disable(task);
2182 		task_clear_spec_ssb_noexec(task);
2183 		task_update_spec_tif(task);
2184 		break;
2185 	case PR_SPEC_DISABLE_NOEXEC:
2186 		if (task_spec_ssb_force_disable(task))
2187 			return -EPERM;
2188 		task_set_spec_ssb_disable(task);
2189 		task_set_spec_ssb_noexec(task);
2190 		task_update_spec_tif(task);
2191 		break;
2192 	default:
2193 		return -ERANGE;
2194 	}
2195 	return 0;
2196 }
2197 
2198 static bool is_spec_ib_user_controlled(void)
2199 {
2200 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
2201 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2202 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
2203 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
2204 }
2205 
2206 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
2207 {
2208 	switch (ctrl) {
2209 	case PR_SPEC_ENABLE:
2210 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2211 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2212 			return 0;
2213 
2214 		/*
2215 		 * With strict mode for both IBPB and STIBP, the instruction
2216 		 * code paths avoid checking this task flag and instead,
2217 		 * unconditionally run the instruction. However, STIBP and IBPB
2218 		 * are independent and either can be set to conditionally
2219 		 * enabled regardless of the mode of the other.
2220 		 *
2221 		 * If either is set to conditional, allow the task flag to be
2222 		 * updated, unless it was force-disabled by a previous prctl
2223 		 * call. Currently, this is possible on an AMD CPU which has the
2224 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
2225 		 * kernel is booted with 'spectre_v2_user=seccomp', then
2226 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
2227 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
2228 		 */
2229 		if (!is_spec_ib_user_controlled() ||
2230 		    task_spec_ib_force_disable(task))
2231 			return -EPERM;
2232 
2233 		task_clear_spec_ib_disable(task);
2234 		task_update_spec_tif(task);
2235 		break;
2236 	case PR_SPEC_DISABLE:
2237 	case PR_SPEC_FORCE_DISABLE:
2238 		/*
2239 		 * Indirect branch speculation is always allowed when
2240 		 * mitigation is force disabled.
2241 		 */
2242 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2243 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2244 			return -EPERM;
2245 
2246 		if (!is_spec_ib_user_controlled())
2247 			return 0;
2248 
2249 		task_set_spec_ib_disable(task);
2250 		if (ctrl == PR_SPEC_FORCE_DISABLE)
2251 			task_set_spec_ib_force_disable(task);
2252 		task_update_spec_tif(task);
2253 		if (task == current)
2254 			indirect_branch_prediction_barrier();
2255 		break;
2256 	default:
2257 		return -ERANGE;
2258 	}
2259 	return 0;
2260 }
2261 
2262 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
2263 			     unsigned long ctrl)
2264 {
2265 	switch (which) {
2266 	case PR_SPEC_STORE_BYPASS:
2267 		return ssb_prctl_set(task, ctrl);
2268 	case PR_SPEC_INDIRECT_BRANCH:
2269 		return ib_prctl_set(task, ctrl);
2270 	case PR_SPEC_L1D_FLUSH:
2271 		return l1d_flush_prctl_set(task, ctrl);
2272 	default:
2273 		return -ENODEV;
2274 	}
2275 }
2276 
2277 #ifdef CONFIG_SECCOMP
2278 void arch_seccomp_spec_mitigate(struct task_struct *task)
2279 {
2280 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
2281 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2282 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2283 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
2284 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2285 }
2286 #endif
2287 
2288 static int l1d_flush_prctl_get(struct task_struct *task)
2289 {
2290 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2291 		return PR_SPEC_FORCE_DISABLE;
2292 
2293 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
2294 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2295 	else
2296 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2297 }
2298 
2299 static int ssb_prctl_get(struct task_struct *task)
2300 {
2301 	switch (ssb_mode) {
2302 	case SPEC_STORE_BYPASS_NONE:
2303 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2304 			return PR_SPEC_ENABLE;
2305 		return PR_SPEC_NOT_AFFECTED;
2306 	case SPEC_STORE_BYPASS_DISABLE:
2307 		return PR_SPEC_DISABLE;
2308 	case SPEC_STORE_BYPASS_SECCOMP:
2309 	case SPEC_STORE_BYPASS_PRCTL:
2310 		if (task_spec_ssb_force_disable(task))
2311 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2312 		if (task_spec_ssb_noexec(task))
2313 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2314 		if (task_spec_ssb_disable(task))
2315 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2316 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2317 	}
2318 	BUG();
2319 }
2320 
2321 static int ib_prctl_get(struct task_struct *task)
2322 {
2323 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2324 		return PR_SPEC_NOT_AFFECTED;
2325 
2326 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2327 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2328 		return PR_SPEC_ENABLE;
2329 	else if (is_spec_ib_user_controlled()) {
2330 		if (task_spec_ib_force_disable(task))
2331 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2332 		if (task_spec_ib_disable(task))
2333 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2334 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2335 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2336 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2337 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2338 		return PR_SPEC_DISABLE;
2339 	else
2340 		return PR_SPEC_NOT_AFFECTED;
2341 }
2342 
2343 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2344 {
2345 	switch (which) {
2346 	case PR_SPEC_STORE_BYPASS:
2347 		return ssb_prctl_get(task);
2348 	case PR_SPEC_INDIRECT_BRANCH:
2349 		return ib_prctl_get(task);
2350 	case PR_SPEC_L1D_FLUSH:
2351 		return l1d_flush_prctl_get(task);
2352 	default:
2353 		return -ENODEV;
2354 	}
2355 }
2356 
2357 void x86_spec_ctrl_setup_ap(void)
2358 {
2359 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2360 		update_spec_ctrl(x86_spec_ctrl_base);
2361 
2362 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2363 		x86_amd_ssb_disable();
2364 }
2365 
2366 bool itlb_multihit_kvm_mitigation;
2367 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2368 
2369 #undef pr_fmt
2370 #define pr_fmt(fmt)	"L1TF: " fmt
2371 
2372 /* Default mitigation for L1TF-affected CPUs */
2373 enum l1tf_mitigations l1tf_mitigation __ro_after_init =
2374 	IS_ENABLED(CONFIG_MITIGATION_L1TF) ? L1TF_MITIGATION_FLUSH : L1TF_MITIGATION_OFF;
2375 #if IS_ENABLED(CONFIG_KVM_INTEL)
2376 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2377 #endif
2378 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2379 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2380 
2381 /*
2382  * These CPUs all support 44bits physical address space internally in the
2383  * cache but CPUID can report a smaller number of physical address bits.
2384  *
2385  * The L1TF mitigation uses the top most address bit for the inversion of
2386  * non present PTEs. When the installed memory reaches into the top most
2387  * address bit due to memory holes, which has been observed on machines
2388  * which report 36bits physical address bits and have 32G RAM installed,
2389  * then the mitigation range check in l1tf_select_mitigation() triggers.
2390  * This is a false positive because the mitigation is still possible due to
2391  * the fact that the cache uses 44bit internally. Use the cache bits
2392  * instead of the reported physical bits and adjust them on the affected
2393  * machines to 44bit if the reported bits are less than 44.
2394  */
2395 static void override_cache_bits(struct cpuinfo_x86 *c)
2396 {
2397 	if (c->x86 != 6)
2398 		return;
2399 
2400 	switch (c->x86_vfm) {
2401 	case INTEL_NEHALEM:
2402 	case INTEL_WESTMERE:
2403 	case INTEL_SANDYBRIDGE:
2404 	case INTEL_IVYBRIDGE:
2405 	case INTEL_HASWELL:
2406 	case INTEL_HASWELL_L:
2407 	case INTEL_HASWELL_G:
2408 	case INTEL_BROADWELL:
2409 	case INTEL_BROADWELL_G:
2410 	case INTEL_SKYLAKE_L:
2411 	case INTEL_SKYLAKE:
2412 	case INTEL_KABYLAKE_L:
2413 	case INTEL_KABYLAKE:
2414 		if (c->x86_cache_bits < 44)
2415 			c->x86_cache_bits = 44;
2416 		break;
2417 	}
2418 }
2419 
2420 static void __init l1tf_select_mitigation(void)
2421 {
2422 	u64 half_pa;
2423 
2424 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2425 		return;
2426 
2427 	if (cpu_mitigations_off())
2428 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2429 	else if (cpu_mitigations_auto_nosmt())
2430 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2431 
2432 	override_cache_bits(&boot_cpu_data);
2433 
2434 	switch (l1tf_mitigation) {
2435 	case L1TF_MITIGATION_OFF:
2436 	case L1TF_MITIGATION_FLUSH_NOWARN:
2437 	case L1TF_MITIGATION_FLUSH:
2438 		break;
2439 	case L1TF_MITIGATION_FLUSH_NOSMT:
2440 	case L1TF_MITIGATION_FULL:
2441 		cpu_smt_disable(false);
2442 		break;
2443 	case L1TF_MITIGATION_FULL_FORCE:
2444 		cpu_smt_disable(true);
2445 		break;
2446 	}
2447 
2448 #if CONFIG_PGTABLE_LEVELS == 2
2449 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2450 	return;
2451 #endif
2452 
2453 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2454 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2455 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2456 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2457 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2458 				half_pa);
2459 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2460 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2461 		return;
2462 	}
2463 
2464 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2465 }
2466 
2467 static int __init l1tf_cmdline(char *str)
2468 {
2469 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2470 		return 0;
2471 
2472 	if (!str)
2473 		return -EINVAL;
2474 
2475 	if (!strcmp(str, "off"))
2476 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2477 	else if (!strcmp(str, "flush,nowarn"))
2478 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2479 	else if (!strcmp(str, "flush"))
2480 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2481 	else if (!strcmp(str, "flush,nosmt"))
2482 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2483 	else if (!strcmp(str, "full"))
2484 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2485 	else if (!strcmp(str, "full,force"))
2486 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2487 
2488 	return 0;
2489 }
2490 early_param("l1tf", l1tf_cmdline);
2491 
2492 #undef pr_fmt
2493 #define pr_fmt(fmt)	"Speculative Return Stack Overflow: " fmt
2494 
2495 enum srso_mitigation {
2496 	SRSO_MITIGATION_NONE,
2497 	SRSO_MITIGATION_UCODE_NEEDED,
2498 	SRSO_MITIGATION_SAFE_RET_UCODE_NEEDED,
2499 	SRSO_MITIGATION_MICROCODE,
2500 	SRSO_MITIGATION_SAFE_RET,
2501 	SRSO_MITIGATION_IBPB,
2502 	SRSO_MITIGATION_IBPB_ON_VMEXIT,
2503 	SRSO_MITIGATION_BP_SPEC_REDUCE,
2504 };
2505 
2506 enum srso_mitigation_cmd {
2507 	SRSO_CMD_OFF,
2508 	SRSO_CMD_MICROCODE,
2509 	SRSO_CMD_SAFE_RET,
2510 	SRSO_CMD_IBPB,
2511 	SRSO_CMD_IBPB_ON_VMEXIT,
2512 };
2513 
2514 static const char * const srso_strings[] = {
2515 	[SRSO_MITIGATION_NONE]			= "Vulnerable",
2516 	[SRSO_MITIGATION_UCODE_NEEDED]		= "Vulnerable: No microcode",
2517 	[SRSO_MITIGATION_SAFE_RET_UCODE_NEEDED]	= "Vulnerable: Safe RET, no microcode",
2518 	[SRSO_MITIGATION_MICROCODE]		= "Vulnerable: Microcode, no safe RET",
2519 	[SRSO_MITIGATION_SAFE_RET]		= "Mitigation: Safe RET",
2520 	[SRSO_MITIGATION_IBPB]			= "Mitigation: IBPB",
2521 	[SRSO_MITIGATION_IBPB_ON_VMEXIT]	= "Mitigation: IBPB on VMEXIT only",
2522 	[SRSO_MITIGATION_BP_SPEC_REDUCE]	= "Mitigation: Reduced Speculation"
2523 };
2524 
2525 static enum srso_mitigation srso_mitigation __ro_after_init = SRSO_MITIGATION_NONE;
2526 static enum srso_mitigation_cmd srso_cmd __ro_after_init = SRSO_CMD_SAFE_RET;
2527 
2528 static int __init srso_parse_cmdline(char *str)
2529 {
2530 	if (!str)
2531 		return -EINVAL;
2532 
2533 	if (!strcmp(str, "off"))
2534 		srso_cmd = SRSO_CMD_OFF;
2535 	else if (!strcmp(str, "microcode"))
2536 		srso_cmd = SRSO_CMD_MICROCODE;
2537 	else if (!strcmp(str, "safe-ret"))
2538 		srso_cmd = SRSO_CMD_SAFE_RET;
2539 	else if (!strcmp(str, "ibpb"))
2540 		srso_cmd = SRSO_CMD_IBPB;
2541 	else if (!strcmp(str, "ibpb-vmexit"))
2542 		srso_cmd = SRSO_CMD_IBPB_ON_VMEXIT;
2543 	else
2544 		pr_err("Ignoring unknown SRSO option (%s).", str);
2545 
2546 	return 0;
2547 }
2548 early_param("spec_rstack_overflow", srso_parse_cmdline);
2549 
2550 #define SRSO_NOTICE "WARNING: See https://kernel.org/doc/html/latest/admin-guide/hw-vuln/srso.html for mitigation options."
2551 
2552 static void __init srso_select_mitigation(void)
2553 {
2554 	bool has_microcode = boot_cpu_has(X86_FEATURE_IBPB_BRTYPE);
2555 
2556 	if (!boot_cpu_has_bug(X86_BUG_SRSO) ||
2557 	    cpu_mitigations_off() ||
2558 	    srso_cmd == SRSO_CMD_OFF) {
2559 		if (boot_cpu_has(X86_FEATURE_SBPB))
2560 			x86_pred_cmd = PRED_CMD_SBPB;
2561 		goto out;
2562 	}
2563 
2564 	if (has_microcode) {
2565 		/*
2566 		 * Zen1/2 with SMT off aren't vulnerable after the right
2567 		 * IBPB microcode has been applied.
2568 		 *
2569 		 * Zen1/2 don't have SBPB, no need to try to enable it here.
2570 		 */
2571 		if (boot_cpu_data.x86 < 0x19 && !cpu_smt_possible()) {
2572 			setup_force_cpu_cap(X86_FEATURE_SRSO_NO);
2573 			goto out;
2574 		}
2575 
2576 		if (retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2577 			srso_mitigation = SRSO_MITIGATION_IBPB;
2578 			goto out;
2579 		}
2580 	} else {
2581 		pr_warn("IBPB-extending microcode not applied!\n");
2582 		pr_warn(SRSO_NOTICE);
2583 
2584 		/* may be overwritten by SRSO_CMD_SAFE_RET below */
2585 		srso_mitigation = SRSO_MITIGATION_UCODE_NEEDED;
2586 	}
2587 
2588 	switch (srso_cmd) {
2589 	case SRSO_CMD_MICROCODE:
2590 		if (has_microcode) {
2591 			srso_mitigation = SRSO_MITIGATION_MICROCODE;
2592 			pr_warn(SRSO_NOTICE);
2593 		}
2594 		break;
2595 
2596 	case SRSO_CMD_SAFE_RET:
2597 		if (boot_cpu_has(X86_FEATURE_SRSO_USER_KERNEL_NO))
2598 			goto ibpb_on_vmexit;
2599 
2600 		if (IS_ENABLED(CONFIG_MITIGATION_SRSO)) {
2601 			/*
2602 			 * Enable the return thunk for generated code
2603 			 * like ftrace, static_call, etc.
2604 			 */
2605 			setup_force_cpu_cap(X86_FEATURE_RETHUNK);
2606 			setup_force_cpu_cap(X86_FEATURE_UNRET);
2607 
2608 			if (boot_cpu_data.x86 == 0x19) {
2609 				setup_force_cpu_cap(X86_FEATURE_SRSO_ALIAS);
2610 				x86_return_thunk = srso_alias_return_thunk;
2611 			} else {
2612 				setup_force_cpu_cap(X86_FEATURE_SRSO);
2613 				x86_return_thunk = srso_return_thunk;
2614 			}
2615 			if (has_microcode)
2616 				srso_mitigation = SRSO_MITIGATION_SAFE_RET;
2617 			else
2618 				srso_mitigation = SRSO_MITIGATION_SAFE_RET_UCODE_NEEDED;
2619 		} else {
2620 			pr_err("WARNING: kernel not compiled with MITIGATION_SRSO.\n");
2621 		}
2622 		break;
2623 
2624 	case SRSO_CMD_IBPB:
2625 		if (IS_ENABLED(CONFIG_MITIGATION_IBPB_ENTRY)) {
2626 			if (has_microcode) {
2627 				setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
2628 				setup_force_cpu_cap(X86_FEATURE_IBPB_ON_VMEXIT);
2629 				srso_mitigation = SRSO_MITIGATION_IBPB;
2630 
2631 				/*
2632 				 * IBPB on entry already obviates the need for
2633 				 * software-based untraining so clear those in case some
2634 				 * other mitigation like Retbleed has selected them.
2635 				 */
2636 				setup_clear_cpu_cap(X86_FEATURE_UNRET);
2637 				setup_clear_cpu_cap(X86_FEATURE_RETHUNK);
2638 
2639 				/*
2640 				 * There is no need for RSB filling: write_ibpb() ensures
2641 				 * all predictions, including the RSB, are invalidated,
2642 				 * regardless of IBPB implementation.
2643 				 */
2644 				setup_clear_cpu_cap(X86_FEATURE_RSB_VMEXIT);
2645 			}
2646 		} else {
2647 			pr_err("WARNING: kernel not compiled with MITIGATION_IBPB_ENTRY.\n");
2648 		}
2649 		break;
2650 
2651 ibpb_on_vmexit:
2652 	case SRSO_CMD_IBPB_ON_VMEXIT:
2653 		if (boot_cpu_has(X86_FEATURE_SRSO_BP_SPEC_REDUCE)) {
2654 			pr_notice("Reducing speculation to address VM/HV SRSO attack vector.\n");
2655 			srso_mitigation = SRSO_MITIGATION_BP_SPEC_REDUCE;
2656 			break;
2657 		}
2658 
2659 		if (IS_ENABLED(CONFIG_MITIGATION_IBPB_ENTRY)) {
2660 			if (has_microcode) {
2661 				setup_force_cpu_cap(X86_FEATURE_IBPB_ON_VMEXIT);
2662 				srso_mitigation = SRSO_MITIGATION_IBPB_ON_VMEXIT;
2663 
2664 				/*
2665 				 * There is no need for RSB filling: write_ibpb() ensures
2666 				 * all predictions, including the RSB, are invalidated,
2667 				 * regardless of IBPB implementation.
2668 				 */
2669 				setup_clear_cpu_cap(X86_FEATURE_RSB_VMEXIT);
2670 			}
2671 		} else {
2672 			pr_err("WARNING: kernel not compiled with MITIGATION_IBPB_ENTRY.\n");
2673 		}
2674 		break;
2675 	default:
2676 		break;
2677 	}
2678 
2679 out:
2680 	/*
2681 	 * Clear the feature flag if this mitigation is not selected as that
2682 	 * feature flag controls the BpSpecReduce MSR bit toggling in KVM.
2683 	 */
2684 	if (srso_mitigation != SRSO_MITIGATION_BP_SPEC_REDUCE)
2685 		setup_clear_cpu_cap(X86_FEATURE_SRSO_BP_SPEC_REDUCE);
2686 
2687 	if (srso_mitigation != SRSO_MITIGATION_NONE)
2688 		pr_info("%s\n", srso_strings[srso_mitigation]);
2689 }
2690 
2691 #undef pr_fmt
2692 #define pr_fmt(fmt) fmt
2693 
2694 #ifdef CONFIG_SYSFS
2695 
2696 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2697 
2698 #if IS_ENABLED(CONFIG_KVM_INTEL)
2699 static const char * const l1tf_vmx_states[] = {
2700 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2701 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2702 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2703 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2704 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2705 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2706 };
2707 
2708 static ssize_t l1tf_show_state(char *buf)
2709 {
2710 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2711 		return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2712 
2713 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2714 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2715 	     sched_smt_active())) {
2716 		return sysfs_emit(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2717 				  l1tf_vmx_states[l1tf_vmx_mitigation]);
2718 	}
2719 
2720 	return sysfs_emit(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2721 			  l1tf_vmx_states[l1tf_vmx_mitigation],
2722 			  sched_smt_active() ? "vulnerable" : "disabled");
2723 }
2724 
2725 static ssize_t itlb_multihit_show_state(char *buf)
2726 {
2727 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2728 	    !boot_cpu_has(X86_FEATURE_VMX))
2729 		return sysfs_emit(buf, "KVM: Mitigation: VMX unsupported\n");
2730 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2731 		return sysfs_emit(buf, "KVM: Mitigation: VMX disabled\n");
2732 	else if (itlb_multihit_kvm_mitigation)
2733 		return sysfs_emit(buf, "KVM: Mitigation: Split huge pages\n");
2734 	else
2735 		return sysfs_emit(buf, "KVM: Vulnerable\n");
2736 }
2737 #else
2738 static ssize_t l1tf_show_state(char *buf)
2739 {
2740 	return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2741 }
2742 
2743 static ssize_t itlb_multihit_show_state(char *buf)
2744 {
2745 	return sysfs_emit(buf, "Processor vulnerable\n");
2746 }
2747 #endif
2748 
2749 static ssize_t mds_show_state(char *buf)
2750 {
2751 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2752 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2753 				  mds_strings[mds_mitigation]);
2754 	}
2755 
2756 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2757 		return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2758 				  (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2759 				   sched_smt_active() ? "mitigated" : "disabled"));
2760 	}
2761 
2762 	return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2763 			  sched_smt_active() ? "vulnerable" : "disabled");
2764 }
2765 
2766 static ssize_t tsx_async_abort_show_state(char *buf)
2767 {
2768 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2769 	    (taa_mitigation == TAA_MITIGATION_OFF))
2770 		return sysfs_emit(buf, "%s\n", taa_strings[taa_mitigation]);
2771 
2772 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2773 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2774 				  taa_strings[taa_mitigation]);
2775 	}
2776 
2777 	return sysfs_emit(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2778 			  sched_smt_active() ? "vulnerable" : "disabled");
2779 }
2780 
2781 static ssize_t mmio_stale_data_show_state(char *buf)
2782 {
2783 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2784 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2785 
2786 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2787 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2788 
2789 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2790 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2791 				  mmio_strings[mmio_mitigation]);
2792 	}
2793 
2794 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2795 			  sched_smt_active() ? "vulnerable" : "disabled");
2796 }
2797 
2798 static ssize_t rfds_show_state(char *buf)
2799 {
2800 	return sysfs_emit(buf, "%s\n", rfds_strings[rfds_mitigation]);
2801 }
2802 
2803 static char *stibp_state(void)
2804 {
2805 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
2806 	    !boot_cpu_has(X86_FEATURE_AUTOIBRS))
2807 		return "";
2808 
2809 	switch (spectre_v2_user_stibp) {
2810 	case SPECTRE_V2_USER_NONE:
2811 		return "; STIBP: disabled";
2812 	case SPECTRE_V2_USER_STRICT:
2813 		return "; STIBP: forced";
2814 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2815 		return "; STIBP: always-on";
2816 	case SPECTRE_V2_USER_PRCTL:
2817 	case SPECTRE_V2_USER_SECCOMP:
2818 		if (static_key_enabled(&switch_to_cond_stibp))
2819 			return "; STIBP: conditional";
2820 	}
2821 	return "";
2822 }
2823 
2824 static char *ibpb_state(void)
2825 {
2826 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2827 		if (static_key_enabled(&switch_mm_always_ibpb))
2828 			return "; IBPB: always-on";
2829 		if (static_key_enabled(&switch_mm_cond_ibpb))
2830 			return "; IBPB: conditional";
2831 		return "; IBPB: disabled";
2832 	}
2833 	return "";
2834 }
2835 
2836 static char *pbrsb_eibrs_state(void)
2837 {
2838 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2839 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2840 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2841 			return "; PBRSB-eIBRS: SW sequence";
2842 		else
2843 			return "; PBRSB-eIBRS: Vulnerable";
2844 	} else {
2845 		return "; PBRSB-eIBRS: Not affected";
2846 	}
2847 }
2848 
2849 static const char *spectre_bhi_state(void)
2850 {
2851 	if (!boot_cpu_has_bug(X86_BUG_BHI))
2852 		return "; BHI: Not affected";
2853 	else if (boot_cpu_has(X86_FEATURE_CLEAR_BHB_HW))
2854 		return "; BHI: BHI_DIS_S";
2855 	else if (boot_cpu_has(X86_FEATURE_CLEAR_BHB_LOOP))
2856 		return "; BHI: SW loop, KVM: SW loop";
2857 	else if (boot_cpu_has(X86_FEATURE_RETPOLINE) &&
2858 		 !boot_cpu_has(X86_FEATURE_RETPOLINE_LFENCE) &&
2859 		 rrsba_disabled)
2860 		return "; BHI: Retpoline";
2861 	else if (boot_cpu_has(X86_FEATURE_CLEAR_BHB_LOOP_ON_VMEXIT))
2862 		return "; BHI: Vulnerable, KVM: SW loop";
2863 
2864 	return "; BHI: Vulnerable";
2865 }
2866 
2867 static ssize_t spectre_v2_show_state(char *buf)
2868 {
2869 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2870 		return sysfs_emit(buf, "Vulnerable: LFENCE\n");
2871 
2872 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2873 		return sysfs_emit(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2874 
2875 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2876 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2877 		return sysfs_emit(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2878 
2879 	return sysfs_emit(buf, "%s%s%s%s%s%s%s%s\n",
2880 			  spectre_v2_strings[spectre_v2_enabled],
2881 			  ibpb_state(),
2882 			  boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? "; IBRS_FW" : "",
2883 			  stibp_state(),
2884 			  boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? "; RSB filling" : "",
2885 			  pbrsb_eibrs_state(),
2886 			  spectre_bhi_state(),
2887 			  /* this should always be at the end */
2888 			  spectre_v2_module_string());
2889 }
2890 
2891 static ssize_t srbds_show_state(char *buf)
2892 {
2893 	return sysfs_emit(buf, "%s\n", srbds_strings[srbds_mitigation]);
2894 }
2895 
2896 static ssize_t retbleed_show_state(char *buf)
2897 {
2898 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2899 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2900 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2901 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2902 			return sysfs_emit(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2903 
2904 		return sysfs_emit(buf, "%s; SMT %s\n", retbleed_strings[retbleed_mitigation],
2905 				  !sched_smt_active() ? "disabled" :
2906 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2907 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2908 				  "enabled with STIBP protection" : "vulnerable");
2909 	}
2910 
2911 	return sysfs_emit(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2912 }
2913 
2914 static ssize_t srso_show_state(char *buf)
2915 {
2916 	if (boot_cpu_has(X86_FEATURE_SRSO_NO))
2917 		return sysfs_emit(buf, "Mitigation: SMT disabled\n");
2918 
2919 	return sysfs_emit(buf, "%s\n", srso_strings[srso_mitigation]);
2920 }
2921 
2922 static ssize_t gds_show_state(char *buf)
2923 {
2924 	return sysfs_emit(buf, "%s\n", gds_strings[gds_mitigation]);
2925 }
2926 
2927 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2928 			       char *buf, unsigned int bug)
2929 {
2930 	if (!boot_cpu_has_bug(bug))
2931 		return sysfs_emit(buf, "Not affected\n");
2932 
2933 	switch (bug) {
2934 	case X86_BUG_CPU_MELTDOWN:
2935 		if (boot_cpu_has(X86_FEATURE_PTI))
2936 			return sysfs_emit(buf, "Mitigation: PTI\n");
2937 
2938 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2939 			return sysfs_emit(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2940 
2941 		break;
2942 
2943 	case X86_BUG_SPECTRE_V1:
2944 		return sysfs_emit(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2945 
2946 	case X86_BUG_SPECTRE_V2:
2947 		return spectre_v2_show_state(buf);
2948 
2949 	case X86_BUG_SPEC_STORE_BYPASS:
2950 		return sysfs_emit(buf, "%s\n", ssb_strings[ssb_mode]);
2951 
2952 	case X86_BUG_L1TF:
2953 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2954 			return l1tf_show_state(buf);
2955 		break;
2956 
2957 	case X86_BUG_MDS:
2958 		return mds_show_state(buf);
2959 
2960 	case X86_BUG_TAA:
2961 		return tsx_async_abort_show_state(buf);
2962 
2963 	case X86_BUG_ITLB_MULTIHIT:
2964 		return itlb_multihit_show_state(buf);
2965 
2966 	case X86_BUG_SRBDS:
2967 		return srbds_show_state(buf);
2968 
2969 	case X86_BUG_MMIO_STALE_DATA:
2970 	case X86_BUG_MMIO_UNKNOWN:
2971 		return mmio_stale_data_show_state(buf);
2972 
2973 	case X86_BUG_RETBLEED:
2974 		return retbleed_show_state(buf);
2975 
2976 	case X86_BUG_SRSO:
2977 		return srso_show_state(buf);
2978 
2979 	case X86_BUG_GDS:
2980 		return gds_show_state(buf);
2981 
2982 	case X86_BUG_RFDS:
2983 		return rfds_show_state(buf);
2984 
2985 	default:
2986 		break;
2987 	}
2988 
2989 	return sysfs_emit(buf, "Vulnerable\n");
2990 }
2991 
2992 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2993 {
2994 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2995 }
2996 
2997 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2998 {
2999 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
3000 }
3001 
3002 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
3003 {
3004 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
3005 }
3006 
3007 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
3008 {
3009 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
3010 }
3011 
3012 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
3013 {
3014 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
3015 }
3016 
3017 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
3018 {
3019 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
3020 }
3021 
3022 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
3023 {
3024 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
3025 }
3026 
3027 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
3028 {
3029 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
3030 }
3031 
3032 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
3033 {
3034 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
3035 }
3036 
3037 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
3038 {
3039 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
3040 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
3041 	else
3042 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
3043 }
3044 
3045 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
3046 {
3047 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
3048 }
3049 
3050 ssize_t cpu_show_spec_rstack_overflow(struct device *dev, struct device_attribute *attr, char *buf)
3051 {
3052 	return cpu_show_common(dev, attr, buf, X86_BUG_SRSO);
3053 }
3054 
3055 ssize_t cpu_show_gds(struct device *dev, struct device_attribute *attr, char *buf)
3056 {
3057 	return cpu_show_common(dev, attr, buf, X86_BUG_GDS);
3058 }
3059 
3060 ssize_t cpu_show_reg_file_data_sampling(struct device *dev, struct device_attribute *attr, char *buf)
3061 {
3062 	return cpu_show_common(dev, attr, buf, X86_BUG_RFDS);
3063 }
3064 #endif
3065 
3066 void __warn_thunk(void)
3067 {
3068 	WARN_ONCE(1, "Unpatched return thunk in use. This should not happen!\n");
3069 }
3070