xref: /linux/arch/x86/kernel/cpu/bugs.c (revision ebc733e54a1a79ea2dde2ba5121ae73a188e20d4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Copyright (C) 1994  Linus Torvalds
4  *
5  *  Cyrix stuff, June 1998 by:
6  *	- Rafael R. Reilova (moved everything from head.S),
7  *        <rreilova@ececs.uc.edu>
8  *	- Channing Corn (tests & fixes),
9  *	- Andrew D. Balsa (code cleanup).
10  */
11 #include <linux/init.h>
12 #include <linux/cpu.h>
13 #include <linux/module.h>
14 #include <linux/nospec.h>
15 #include <linux/prctl.h>
16 #include <linux/sched/smt.h>
17 #include <linux/pgtable.h>
18 #include <linux/bpf.h>
19 
20 #include <asm/spec-ctrl.h>
21 #include <asm/cmdline.h>
22 #include <asm/bugs.h>
23 #include <asm/processor.h>
24 #include <asm/processor-flags.h>
25 #include <asm/fpu/api.h>
26 #include <asm/msr.h>
27 #include <asm/vmx.h>
28 #include <asm/paravirt.h>
29 #include <asm/intel-family.h>
30 #include <asm/e820/api.h>
31 #include <asm/hypervisor.h>
32 #include <asm/tlbflush.h>
33 #include <asm/cpu.h>
34 
35 #include "cpu.h"
36 
37 static void __init spectre_v1_select_mitigation(void);
38 static void __init spectre_v2_select_mitigation(void);
39 static void __init retbleed_select_mitigation(void);
40 static void __init spectre_v2_user_select_mitigation(void);
41 static void __init ssb_select_mitigation(void);
42 static void __init l1tf_select_mitigation(void);
43 static void __init mds_select_mitigation(void);
44 static void __init md_clear_update_mitigation(void);
45 static void __init md_clear_select_mitigation(void);
46 static void __init taa_select_mitigation(void);
47 static void __init mmio_select_mitigation(void);
48 static void __init srbds_select_mitigation(void);
49 static void __init l1d_flush_select_mitigation(void);
50 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_SYMBOL_GPL(x86_spec_ctrl_current);
60 
61 u64 x86_pred_cmd __ro_after_init = PRED_CMD_IBPB;
62 EXPORT_SYMBOL_GPL(x86_pred_cmd);
63 
64 static DEFINE_MUTEX(spec_ctrl_mutex);
65 
66 void (*x86_return_thunk)(void) __ro_after_init = __x86_return_thunk;
67 
68 /* Update SPEC_CTRL MSR and its cached copy unconditionally */
69 static void update_spec_ctrl(u64 val)
70 {
71 	this_cpu_write(x86_spec_ctrl_current, val);
72 	wrmsrl(MSR_IA32_SPEC_CTRL, val);
73 }
74 
75 /*
76  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
77  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
78  */
79 void update_spec_ctrl_cond(u64 val)
80 {
81 	if (this_cpu_read(x86_spec_ctrl_current) == val)
82 		return;
83 
84 	this_cpu_write(x86_spec_ctrl_current, val);
85 
86 	/*
87 	 * When KERNEL_IBRS this MSR is written on return-to-user, unless
88 	 * forced the update can be delayed until that time.
89 	 */
90 	if (!cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
91 		wrmsrl(MSR_IA32_SPEC_CTRL, val);
92 }
93 
94 noinstr u64 spec_ctrl_current(void)
95 {
96 	return this_cpu_read(x86_spec_ctrl_current);
97 }
98 EXPORT_SYMBOL_GPL(spec_ctrl_current);
99 
100 /*
101  * AMD specific MSR info for Speculative Store Bypass control.
102  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
103  */
104 u64 __ro_after_init x86_amd_ls_cfg_base;
105 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
106 
107 /* Control conditional STIBP in switch_to() */
108 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
109 /* Control conditional IBPB in switch_mm() */
110 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
111 /* Control unconditional IBPB in switch_mm() */
112 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
113 
114 /* Control MDS CPU buffer clear before returning to user space */
115 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
116 EXPORT_SYMBOL_GPL(mds_user_clear);
117 /* Control MDS CPU buffer clear before idling (halt, mwait) */
118 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
119 EXPORT_SYMBOL_GPL(mds_idle_clear);
120 
121 /*
122  * Controls whether l1d flush based mitigations are enabled,
123  * based on hw features and admin setting via boot parameter
124  * defaults to false
125  */
126 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
127 
128 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
129 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
130 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
131 
132 void __init cpu_select_mitigations(void)
133 {
134 	/*
135 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
136 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
137 	 * init code as it is not enumerated and depends on the family.
138 	 */
139 	if (cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL)) {
140 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
141 
142 		/*
143 		 * Previously running kernel (kexec), may have some controls
144 		 * turned ON. Clear them and let the mitigations setup below
145 		 * rediscover them based on configuration.
146 		 */
147 		x86_spec_ctrl_base &= ~SPEC_CTRL_MITIGATIONS_MASK;
148 	}
149 
150 	/* Select the proper CPU mitigations before patching alternatives: */
151 	spectre_v1_select_mitigation();
152 	spectre_v2_select_mitigation();
153 	/*
154 	 * retbleed_select_mitigation() relies on the state set by
155 	 * spectre_v2_select_mitigation(); specifically it wants to know about
156 	 * spectre_v2=ibrs.
157 	 */
158 	retbleed_select_mitigation();
159 	/*
160 	 * spectre_v2_user_select_mitigation() relies on the state set by
161 	 * retbleed_select_mitigation(); specifically the STIBP selection is
162 	 * forced for UNRET or IBPB.
163 	 */
164 	spectre_v2_user_select_mitigation();
165 	ssb_select_mitigation();
166 	l1tf_select_mitigation();
167 	md_clear_select_mitigation();
168 	srbds_select_mitigation();
169 	l1d_flush_select_mitigation();
170 
171 	/*
172 	 * srso_select_mitigation() depends and must run after
173 	 * retbleed_select_mitigation().
174 	 */
175 	srso_select_mitigation();
176 	gds_select_mitigation();
177 }
178 
179 /*
180  * NOTE: This function is *only* called for SVM, since Intel uses
181  * MSR_IA32_SPEC_CTRL for SSBD.
182  */
183 void
184 x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl, bool setguest)
185 {
186 	u64 guestval, hostval;
187 	struct thread_info *ti = current_thread_info();
188 
189 	/*
190 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
191 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
192 	 */
193 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
194 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
195 		return;
196 
197 	/*
198 	 * If the host has SSBD mitigation enabled, force it in the host's
199 	 * virtual MSR value. If its not permanently enabled, evaluate
200 	 * current's TIF_SSBD thread flag.
201 	 */
202 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
203 		hostval = SPEC_CTRL_SSBD;
204 	else
205 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
206 
207 	/* Sanitize the guest value */
208 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
209 
210 	if (hostval != guestval) {
211 		unsigned long tif;
212 
213 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
214 				 ssbd_spec_ctrl_to_tif(hostval);
215 
216 		speculation_ctrl_update(tif);
217 	}
218 }
219 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
220 
221 static void x86_amd_ssb_disable(void)
222 {
223 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
224 
225 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
226 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
227 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
228 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
229 }
230 
231 #undef pr_fmt
232 #define pr_fmt(fmt)	"MDS: " fmt
233 
234 /* Default mitigation for MDS-affected CPUs */
235 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
236 static bool mds_nosmt __ro_after_init = false;
237 
238 static const char * const mds_strings[] = {
239 	[MDS_MITIGATION_OFF]	= "Vulnerable",
240 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
241 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
242 };
243 
244 static void __init mds_select_mitigation(void)
245 {
246 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
247 		mds_mitigation = MDS_MITIGATION_OFF;
248 		return;
249 	}
250 
251 	if (mds_mitigation == MDS_MITIGATION_FULL) {
252 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
253 			mds_mitigation = MDS_MITIGATION_VMWERV;
254 
255 		static_branch_enable(&mds_user_clear);
256 
257 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
258 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
259 			cpu_smt_disable(false);
260 	}
261 }
262 
263 static int __init mds_cmdline(char *str)
264 {
265 	if (!boot_cpu_has_bug(X86_BUG_MDS))
266 		return 0;
267 
268 	if (!str)
269 		return -EINVAL;
270 
271 	if (!strcmp(str, "off"))
272 		mds_mitigation = MDS_MITIGATION_OFF;
273 	else if (!strcmp(str, "full"))
274 		mds_mitigation = MDS_MITIGATION_FULL;
275 	else if (!strcmp(str, "full,nosmt")) {
276 		mds_mitigation = MDS_MITIGATION_FULL;
277 		mds_nosmt = true;
278 	}
279 
280 	return 0;
281 }
282 early_param("mds", mds_cmdline);
283 
284 #undef pr_fmt
285 #define pr_fmt(fmt)	"TAA: " fmt
286 
287 enum taa_mitigations {
288 	TAA_MITIGATION_OFF,
289 	TAA_MITIGATION_UCODE_NEEDED,
290 	TAA_MITIGATION_VERW,
291 	TAA_MITIGATION_TSX_DISABLED,
292 };
293 
294 /* Default mitigation for TAA-affected CPUs */
295 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
296 static bool taa_nosmt __ro_after_init;
297 
298 static const char * const taa_strings[] = {
299 	[TAA_MITIGATION_OFF]		= "Vulnerable",
300 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
301 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
302 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
303 };
304 
305 static void __init taa_select_mitigation(void)
306 {
307 	u64 ia32_cap;
308 
309 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
310 		taa_mitigation = TAA_MITIGATION_OFF;
311 		return;
312 	}
313 
314 	/* TSX previously disabled by tsx=off */
315 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
316 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
317 		return;
318 	}
319 
320 	if (cpu_mitigations_off()) {
321 		taa_mitigation = TAA_MITIGATION_OFF;
322 		return;
323 	}
324 
325 	/*
326 	 * TAA mitigation via VERW is turned off if both
327 	 * tsx_async_abort=off and mds=off are specified.
328 	 */
329 	if (taa_mitigation == TAA_MITIGATION_OFF &&
330 	    mds_mitigation == MDS_MITIGATION_OFF)
331 		return;
332 
333 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
334 		taa_mitigation = TAA_MITIGATION_VERW;
335 	else
336 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
337 
338 	/*
339 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
340 	 * A microcode update fixes this behavior to clear CPU buffers. It also
341 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
342 	 * ARCH_CAP_TSX_CTRL_MSR bit.
343 	 *
344 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
345 	 * update is required.
346 	 */
347 	ia32_cap = x86_read_arch_cap_msr();
348 	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
349 	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
350 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
351 
352 	/*
353 	 * TSX is enabled, select alternate mitigation for TAA which is
354 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
355 	 *
356 	 * For guests that can't determine whether the correct microcode is
357 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
358 	 */
359 	static_branch_enable(&mds_user_clear);
360 
361 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
362 		cpu_smt_disable(false);
363 }
364 
365 static int __init tsx_async_abort_parse_cmdline(char *str)
366 {
367 	if (!boot_cpu_has_bug(X86_BUG_TAA))
368 		return 0;
369 
370 	if (!str)
371 		return -EINVAL;
372 
373 	if (!strcmp(str, "off")) {
374 		taa_mitigation = TAA_MITIGATION_OFF;
375 	} else if (!strcmp(str, "full")) {
376 		taa_mitigation = TAA_MITIGATION_VERW;
377 	} else if (!strcmp(str, "full,nosmt")) {
378 		taa_mitigation = TAA_MITIGATION_VERW;
379 		taa_nosmt = true;
380 	}
381 
382 	return 0;
383 }
384 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
385 
386 #undef pr_fmt
387 #define pr_fmt(fmt)	"MMIO Stale Data: " fmt
388 
389 enum mmio_mitigations {
390 	MMIO_MITIGATION_OFF,
391 	MMIO_MITIGATION_UCODE_NEEDED,
392 	MMIO_MITIGATION_VERW,
393 };
394 
395 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
396 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
397 static bool mmio_nosmt __ro_after_init = false;
398 
399 static const char * const mmio_strings[] = {
400 	[MMIO_MITIGATION_OFF]		= "Vulnerable",
401 	[MMIO_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
402 	[MMIO_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
403 };
404 
405 static void __init mmio_select_mitigation(void)
406 {
407 	u64 ia32_cap;
408 
409 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
410 	     boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
411 	     cpu_mitigations_off()) {
412 		mmio_mitigation = MMIO_MITIGATION_OFF;
413 		return;
414 	}
415 
416 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
417 		return;
418 
419 	ia32_cap = x86_read_arch_cap_msr();
420 
421 	/*
422 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
423 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
424 	 */
425 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
426 					      boot_cpu_has(X86_FEATURE_RTM)))
427 		static_branch_enable(&mds_user_clear);
428 	else
429 		static_branch_enable(&mmio_stale_data_clear);
430 
431 	/*
432 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
433 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
434 	 * is required irrespective of SMT state.
435 	 */
436 	if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
437 		static_branch_enable(&mds_idle_clear);
438 
439 	/*
440 	 * Check if the system has the right microcode.
441 	 *
442 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
443 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
444 	 * affected systems.
445 	 */
446 	if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
447 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
448 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
449 	     !(ia32_cap & ARCH_CAP_MDS_NO)))
450 		mmio_mitigation = MMIO_MITIGATION_VERW;
451 	else
452 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
453 
454 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
455 		cpu_smt_disable(false);
456 }
457 
458 static int __init mmio_stale_data_parse_cmdline(char *str)
459 {
460 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
461 		return 0;
462 
463 	if (!str)
464 		return -EINVAL;
465 
466 	if (!strcmp(str, "off")) {
467 		mmio_mitigation = MMIO_MITIGATION_OFF;
468 	} else if (!strcmp(str, "full")) {
469 		mmio_mitigation = MMIO_MITIGATION_VERW;
470 	} else if (!strcmp(str, "full,nosmt")) {
471 		mmio_mitigation = MMIO_MITIGATION_VERW;
472 		mmio_nosmt = true;
473 	}
474 
475 	return 0;
476 }
477 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
478 
479 #undef pr_fmt
480 #define pr_fmt(fmt)     "" fmt
481 
482 static void __init md_clear_update_mitigation(void)
483 {
484 	if (cpu_mitigations_off())
485 		return;
486 
487 	if (!static_key_enabled(&mds_user_clear))
488 		goto out;
489 
490 	/*
491 	 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
492 	 * mitigation, if necessary.
493 	 */
494 	if (mds_mitigation == MDS_MITIGATION_OFF &&
495 	    boot_cpu_has_bug(X86_BUG_MDS)) {
496 		mds_mitigation = MDS_MITIGATION_FULL;
497 		mds_select_mitigation();
498 	}
499 	if (taa_mitigation == TAA_MITIGATION_OFF &&
500 	    boot_cpu_has_bug(X86_BUG_TAA)) {
501 		taa_mitigation = TAA_MITIGATION_VERW;
502 		taa_select_mitigation();
503 	}
504 	if (mmio_mitigation == MMIO_MITIGATION_OFF &&
505 	    boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
506 		mmio_mitigation = MMIO_MITIGATION_VERW;
507 		mmio_select_mitigation();
508 	}
509 out:
510 	if (boot_cpu_has_bug(X86_BUG_MDS))
511 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
512 	if (boot_cpu_has_bug(X86_BUG_TAA))
513 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
514 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
515 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
516 	else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
517 		pr_info("MMIO Stale Data: Unknown: No mitigations\n");
518 }
519 
520 static void __init md_clear_select_mitigation(void)
521 {
522 	mds_select_mitigation();
523 	taa_select_mitigation();
524 	mmio_select_mitigation();
525 
526 	/*
527 	 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
528 	 * and print their mitigation after MDS, TAA and MMIO Stale Data
529 	 * mitigation selection is done.
530 	 */
531 	md_clear_update_mitigation();
532 }
533 
534 #undef pr_fmt
535 #define pr_fmt(fmt)	"SRBDS: " fmt
536 
537 enum srbds_mitigations {
538 	SRBDS_MITIGATION_OFF,
539 	SRBDS_MITIGATION_UCODE_NEEDED,
540 	SRBDS_MITIGATION_FULL,
541 	SRBDS_MITIGATION_TSX_OFF,
542 	SRBDS_MITIGATION_HYPERVISOR,
543 };
544 
545 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
546 
547 static const char * const srbds_strings[] = {
548 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
549 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
550 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
551 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
552 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
553 };
554 
555 static bool srbds_off;
556 
557 void update_srbds_msr(void)
558 {
559 	u64 mcu_ctrl;
560 
561 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
562 		return;
563 
564 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
565 		return;
566 
567 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
568 		return;
569 
570 	/*
571 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
572 	 * being disabled and it hasn't received the SRBDS MSR microcode.
573 	 */
574 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
575 		return;
576 
577 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
578 
579 	switch (srbds_mitigation) {
580 	case SRBDS_MITIGATION_OFF:
581 	case SRBDS_MITIGATION_TSX_OFF:
582 		mcu_ctrl |= RNGDS_MITG_DIS;
583 		break;
584 	case SRBDS_MITIGATION_FULL:
585 		mcu_ctrl &= ~RNGDS_MITG_DIS;
586 		break;
587 	default:
588 		break;
589 	}
590 
591 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
592 }
593 
594 static void __init srbds_select_mitigation(void)
595 {
596 	u64 ia32_cap;
597 
598 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
599 		return;
600 
601 	/*
602 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
603 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
604 	 * by Processor MMIO Stale Data vulnerability.
605 	 */
606 	ia32_cap = x86_read_arch_cap_msr();
607 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
608 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
609 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
610 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
611 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
612 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
613 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
614 	else if (cpu_mitigations_off() || srbds_off)
615 		srbds_mitigation = SRBDS_MITIGATION_OFF;
616 
617 	update_srbds_msr();
618 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
619 }
620 
621 static int __init srbds_parse_cmdline(char *str)
622 {
623 	if (!str)
624 		return -EINVAL;
625 
626 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
627 		return 0;
628 
629 	srbds_off = !strcmp(str, "off");
630 	return 0;
631 }
632 early_param("srbds", srbds_parse_cmdline);
633 
634 #undef pr_fmt
635 #define pr_fmt(fmt)     "L1D Flush : " fmt
636 
637 enum l1d_flush_mitigations {
638 	L1D_FLUSH_OFF = 0,
639 	L1D_FLUSH_ON,
640 };
641 
642 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
643 
644 static void __init l1d_flush_select_mitigation(void)
645 {
646 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
647 		return;
648 
649 	static_branch_enable(&switch_mm_cond_l1d_flush);
650 	pr_info("Conditional flush on switch_mm() enabled\n");
651 }
652 
653 static int __init l1d_flush_parse_cmdline(char *str)
654 {
655 	if (!strcmp(str, "on"))
656 		l1d_flush_mitigation = L1D_FLUSH_ON;
657 
658 	return 0;
659 }
660 early_param("l1d_flush", l1d_flush_parse_cmdline);
661 
662 #undef pr_fmt
663 #define pr_fmt(fmt)	"GDS: " fmt
664 
665 enum gds_mitigations {
666 	GDS_MITIGATION_OFF,
667 	GDS_MITIGATION_UCODE_NEEDED,
668 	GDS_MITIGATION_FORCE,
669 	GDS_MITIGATION_FULL,
670 	GDS_MITIGATION_FULL_LOCKED,
671 	GDS_MITIGATION_HYPERVISOR,
672 };
673 
674 #if IS_ENABLED(CONFIG_GDS_FORCE_MITIGATION)
675 static enum gds_mitigations gds_mitigation __ro_after_init = GDS_MITIGATION_FORCE;
676 #else
677 static enum gds_mitigations gds_mitigation __ro_after_init = GDS_MITIGATION_FULL;
678 #endif
679 
680 static const char * const gds_strings[] = {
681 	[GDS_MITIGATION_OFF]		= "Vulnerable",
682 	[GDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
683 	[GDS_MITIGATION_FORCE]		= "Mitigation: AVX disabled, no microcode",
684 	[GDS_MITIGATION_FULL]		= "Mitigation: Microcode",
685 	[GDS_MITIGATION_FULL_LOCKED]	= "Mitigation: Microcode (locked)",
686 	[GDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
687 };
688 
689 bool gds_ucode_mitigated(void)
690 {
691 	return (gds_mitigation == GDS_MITIGATION_FULL ||
692 		gds_mitigation == GDS_MITIGATION_FULL_LOCKED);
693 }
694 EXPORT_SYMBOL_GPL(gds_ucode_mitigated);
695 
696 void update_gds_msr(void)
697 {
698 	u64 mcu_ctrl_after;
699 	u64 mcu_ctrl;
700 
701 	switch (gds_mitigation) {
702 	case GDS_MITIGATION_OFF:
703 		rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
704 		mcu_ctrl |= GDS_MITG_DIS;
705 		break;
706 	case GDS_MITIGATION_FULL_LOCKED:
707 		/*
708 		 * The LOCKED state comes from the boot CPU. APs might not have
709 		 * the same state. Make sure the mitigation is enabled on all
710 		 * CPUs.
711 		 */
712 	case GDS_MITIGATION_FULL:
713 		rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
714 		mcu_ctrl &= ~GDS_MITG_DIS;
715 		break;
716 	case GDS_MITIGATION_FORCE:
717 	case GDS_MITIGATION_UCODE_NEEDED:
718 	case GDS_MITIGATION_HYPERVISOR:
719 		return;
720 	}
721 
722 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
723 
724 	/*
725 	 * Check to make sure that the WRMSR value was not ignored. Writes to
726 	 * GDS_MITG_DIS will be ignored if this processor is locked but the boot
727 	 * processor was not.
728 	 */
729 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl_after);
730 	WARN_ON_ONCE(mcu_ctrl != mcu_ctrl_after);
731 }
732 
733 static void __init gds_select_mitigation(void)
734 {
735 	u64 mcu_ctrl;
736 
737 	if (!boot_cpu_has_bug(X86_BUG_GDS))
738 		return;
739 
740 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
741 		gds_mitigation = GDS_MITIGATION_HYPERVISOR;
742 		goto out;
743 	}
744 
745 	if (cpu_mitigations_off())
746 		gds_mitigation = GDS_MITIGATION_OFF;
747 	/* Will verify below that mitigation _can_ be disabled */
748 
749 	/* No microcode */
750 	if (!(x86_read_arch_cap_msr() & ARCH_CAP_GDS_CTRL)) {
751 		if (gds_mitigation == GDS_MITIGATION_FORCE) {
752 			/*
753 			 * This only needs to be done on the boot CPU so do it
754 			 * here rather than in update_gds_msr()
755 			 */
756 			setup_clear_cpu_cap(X86_FEATURE_AVX);
757 			pr_warn("Microcode update needed! Disabling AVX as mitigation.\n");
758 		} else {
759 			gds_mitigation = GDS_MITIGATION_UCODE_NEEDED;
760 		}
761 		goto out;
762 	}
763 
764 	/* Microcode has mitigation, use it */
765 	if (gds_mitigation == GDS_MITIGATION_FORCE)
766 		gds_mitigation = GDS_MITIGATION_FULL;
767 
768 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
769 	if (mcu_ctrl & GDS_MITG_LOCKED) {
770 		if (gds_mitigation == GDS_MITIGATION_OFF)
771 			pr_warn("Mitigation locked. Disable failed.\n");
772 
773 		/*
774 		 * The mitigation is selected from the boot CPU. All other CPUs
775 		 * _should_ have the same state. If the boot CPU isn't locked
776 		 * but others are then update_gds_msr() will WARN() of the state
777 		 * mismatch. If the boot CPU is locked update_gds_msr() will
778 		 * ensure the other CPUs have the mitigation enabled.
779 		 */
780 		gds_mitigation = GDS_MITIGATION_FULL_LOCKED;
781 	}
782 
783 	update_gds_msr();
784 out:
785 	pr_info("%s\n", gds_strings[gds_mitigation]);
786 }
787 
788 static int __init gds_parse_cmdline(char *str)
789 {
790 	if (!str)
791 		return -EINVAL;
792 
793 	if (!boot_cpu_has_bug(X86_BUG_GDS))
794 		return 0;
795 
796 	if (!strcmp(str, "off"))
797 		gds_mitigation = GDS_MITIGATION_OFF;
798 	else if (!strcmp(str, "force"))
799 		gds_mitigation = GDS_MITIGATION_FORCE;
800 
801 	return 0;
802 }
803 early_param("gather_data_sampling", gds_parse_cmdline);
804 
805 #undef pr_fmt
806 #define pr_fmt(fmt)     "Spectre V1 : " fmt
807 
808 enum spectre_v1_mitigation {
809 	SPECTRE_V1_MITIGATION_NONE,
810 	SPECTRE_V1_MITIGATION_AUTO,
811 };
812 
813 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
814 	SPECTRE_V1_MITIGATION_AUTO;
815 
816 static const char * const spectre_v1_strings[] = {
817 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
818 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
819 };
820 
821 /*
822  * Does SMAP provide full mitigation against speculative kernel access to
823  * userspace?
824  */
825 static bool smap_works_speculatively(void)
826 {
827 	if (!boot_cpu_has(X86_FEATURE_SMAP))
828 		return false;
829 
830 	/*
831 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
832 	 * prevent speculative access to user data in the L1 cache.
833 	 * Consider SMAP to be non-functional as a mitigation on these
834 	 * CPUs.
835 	 */
836 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
837 		return false;
838 
839 	return true;
840 }
841 
842 static void __init spectre_v1_select_mitigation(void)
843 {
844 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
845 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
846 		return;
847 	}
848 
849 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
850 		/*
851 		 * With Spectre v1, a user can speculatively control either
852 		 * path of a conditional swapgs with a user-controlled GS
853 		 * value.  The mitigation is to add lfences to both code paths.
854 		 *
855 		 * If FSGSBASE is enabled, the user can put a kernel address in
856 		 * GS, in which case SMAP provides no protection.
857 		 *
858 		 * If FSGSBASE is disabled, the user can only put a user space
859 		 * address in GS.  That makes an attack harder, but still
860 		 * possible if there's no SMAP protection.
861 		 */
862 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
863 		    !smap_works_speculatively()) {
864 			/*
865 			 * Mitigation can be provided from SWAPGS itself or
866 			 * PTI as the CR3 write in the Meltdown mitigation
867 			 * is serializing.
868 			 *
869 			 * If neither is there, mitigate with an LFENCE to
870 			 * stop speculation through swapgs.
871 			 */
872 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
873 			    !boot_cpu_has(X86_FEATURE_PTI))
874 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
875 
876 			/*
877 			 * Enable lfences in the kernel entry (non-swapgs)
878 			 * paths, to prevent user entry from speculatively
879 			 * skipping swapgs.
880 			 */
881 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
882 		}
883 	}
884 
885 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
886 }
887 
888 static int __init nospectre_v1_cmdline(char *str)
889 {
890 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
891 	return 0;
892 }
893 early_param("nospectre_v1", nospectre_v1_cmdline);
894 
895 enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init = SPECTRE_V2_NONE;
896 
897 #undef pr_fmt
898 #define pr_fmt(fmt)     "RETBleed: " fmt
899 
900 enum retbleed_mitigation {
901 	RETBLEED_MITIGATION_NONE,
902 	RETBLEED_MITIGATION_UNRET,
903 	RETBLEED_MITIGATION_IBPB,
904 	RETBLEED_MITIGATION_IBRS,
905 	RETBLEED_MITIGATION_EIBRS,
906 	RETBLEED_MITIGATION_STUFF,
907 };
908 
909 enum retbleed_mitigation_cmd {
910 	RETBLEED_CMD_OFF,
911 	RETBLEED_CMD_AUTO,
912 	RETBLEED_CMD_UNRET,
913 	RETBLEED_CMD_IBPB,
914 	RETBLEED_CMD_STUFF,
915 };
916 
917 static const char * const retbleed_strings[] = {
918 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
919 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
920 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
921 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
922 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
923 	[RETBLEED_MITIGATION_STUFF]	= "Mitigation: Stuffing",
924 };
925 
926 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
927 	RETBLEED_MITIGATION_NONE;
928 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
929 	RETBLEED_CMD_AUTO;
930 
931 static int __ro_after_init retbleed_nosmt = false;
932 
933 static int __init retbleed_parse_cmdline(char *str)
934 {
935 	if (!str)
936 		return -EINVAL;
937 
938 	while (str) {
939 		char *next = strchr(str, ',');
940 		if (next) {
941 			*next = 0;
942 			next++;
943 		}
944 
945 		if (!strcmp(str, "off")) {
946 			retbleed_cmd = RETBLEED_CMD_OFF;
947 		} else if (!strcmp(str, "auto")) {
948 			retbleed_cmd = RETBLEED_CMD_AUTO;
949 		} else if (!strcmp(str, "unret")) {
950 			retbleed_cmd = RETBLEED_CMD_UNRET;
951 		} else if (!strcmp(str, "ibpb")) {
952 			retbleed_cmd = RETBLEED_CMD_IBPB;
953 		} else if (!strcmp(str, "stuff")) {
954 			retbleed_cmd = RETBLEED_CMD_STUFF;
955 		} else if (!strcmp(str, "nosmt")) {
956 			retbleed_nosmt = true;
957 		} else if (!strcmp(str, "force")) {
958 			setup_force_cpu_bug(X86_BUG_RETBLEED);
959 		} else {
960 			pr_err("Ignoring unknown retbleed option (%s).", str);
961 		}
962 
963 		str = next;
964 	}
965 
966 	return 0;
967 }
968 early_param("retbleed", retbleed_parse_cmdline);
969 
970 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
971 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
972 
973 static void __init retbleed_select_mitigation(void)
974 {
975 	bool mitigate_smt = false;
976 
977 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
978 		return;
979 
980 	switch (retbleed_cmd) {
981 	case RETBLEED_CMD_OFF:
982 		return;
983 
984 	case RETBLEED_CMD_UNRET:
985 		if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
986 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
987 		} else {
988 			pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
989 			goto do_cmd_auto;
990 		}
991 		break;
992 
993 	case RETBLEED_CMD_IBPB:
994 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
995 			pr_err("WARNING: CPU does not support IBPB.\n");
996 			goto do_cmd_auto;
997 		} else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
998 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
999 		} else {
1000 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
1001 			goto do_cmd_auto;
1002 		}
1003 		break;
1004 
1005 	case RETBLEED_CMD_STUFF:
1006 		if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING) &&
1007 		    spectre_v2_enabled == SPECTRE_V2_RETPOLINE) {
1008 			retbleed_mitigation = RETBLEED_MITIGATION_STUFF;
1009 
1010 		} else {
1011 			if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING))
1012 				pr_err("WARNING: retbleed=stuff depends on spectre_v2=retpoline\n");
1013 			else
1014 				pr_err("WARNING: kernel not compiled with CALL_DEPTH_TRACKING.\n");
1015 
1016 			goto do_cmd_auto;
1017 		}
1018 		break;
1019 
1020 do_cmd_auto:
1021 	case RETBLEED_CMD_AUTO:
1022 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1023 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
1024 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
1025 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
1026 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
1027 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
1028 		}
1029 
1030 		/*
1031 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
1032 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
1033 		 * be set accordingly below.
1034 		 */
1035 
1036 		break;
1037 	}
1038 
1039 	switch (retbleed_mitigation) {
1040 	case RETBLEED_MITIGATION_UNRET:
1041 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
1042 		setup_force_cpu_cap(X86_FEATURE_UNRET);
1043 
1044 		x86_return_thunk = retbleed_return_thunk;
1045 
1046 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
1047 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
1048 			pr_err(RETBLEED_UNTRAIN_MSG);
1049 
1050 		mitigate_smt = true;
1051 		break;
1052 
1053 	case RETBLEED_MITIGATION_IBPB:
1054 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
1055 		setup_force_cpu_cap(X86_FEATURE_IBPB_ON_VMEXIT);
1056 		mitigate_smt = true;
1057 		break;
1058 
1059 	case RETBLEED_MITIGATION_STUFF:
1060 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
1061 		setup_force_cpu_cap(X86_FEATURE_CALL_DEPTH);
1062 
1063 		x86_return_thunk = call_depth_return_thunk;
1064 		break;
1065 
1066 	default:
1067 		break;
1068 	}
1069 
1070 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
1071 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
1072 		cpu_smt_disable(false);
1073 
1074 	/*
1075 	 * Let IBRS trump all on Intel without affecting the effects of the
1076 	 * retbleed= cmdline option except for call depth based stuffing
1077 	 */
1078 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1079 		switch (spectre_v2_enabled) {
1080 		case SPECTRE_V2_IBRS:
1081 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
1082 			break;
1083 		case SPECTRE_V2_EIBRS:
1084 		case SPECTRE_V2_EIBRS_RETPOLINE:
1085 		case SPECTRE_V2_EIBRS_LFENCE:
1086 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
1087 			break;
1088 		default:
1089 			if (retbleed_mitigation != RETBLEED_MITIGATION_STUFF)
1090 				pr_err(RETBLEED_INTEL_MSG);
1091 		}
1092 	}
1093 
1094 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
1095 }
1096 
1097 #undef pr_fmt
1098 #define pr_fmt(fmt)     "Spectre V2 : " fmt
1099 
1100 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
1101 	SPECTRE_V2_USER_NONE;
1102 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
1103 	SPECTRE_V2_USER_NONE;
1104 
1105 #ifdef CONFIG_RETPOLINE
1106 static bool spectre_v2_bad_module;
1107 
1108 bool retpoline_module_ok(bool has_retpoline)
1109 {
1110 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
1111 		return true;
1112 
1113 	pr_err("System may be vulnerable to spectre v2\n");
1114 	spectre_v2_bad_module = true;
1115 	return false;
1116 }
1117 
1118 static inline const char *spectre_v2_module_string(void)
1119 {
1120 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
1121 }
1122 #else
1123 static inline const char *spectre_v2_module_string(void) { return ""; }
1124 #endif
1125 
1126 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
1127 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
1128 #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"
1129 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
1130 
1131 #ifdef CONFIG_BPF_SYSCALL
1132 void unpriv_ebpf_notify(int new_state)
1133 {
1134 	if (new_state)
1135 		return;
1136 
1137 	/* Unprivileged eBPF is enabled */
1138 
1139 	switch (spectre_v2_enabled) {
1140 	case SPECTRE_V2_EIBRS:
1141 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1142 		break;
1143 	case SPECTRE_V2_EIBRS_LFENCE:
1144 		if (sched_smt_active())
1145 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1146 		break;
1147 	default:
1148 		break;
1149 	}
1150 }
1151 #endif
1152 
1153 static inline bool match_option(const char *arg, int arglen, const char *opt)
1154 {
1155 	int len = strlen(opt);
1156 
1157 	return len == arglen && !strncmp(arg, opt, len);
1158 }
1159 
1160 /* The kernel command line selection for spectre v2 */
1161 enum spectre_v2_mitigation_cmd {
1162 	SPECTRE_V2_CMD_NONE,
1163 	SPECTRE_V2_CMD_AUTO,
1164 	SPECTRE_V2_CMD_FORCE,
1165 	SPECTRE_V2_CMD_RETPOLINE,
1166 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1167 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1168 	SPECTRE_V2_CMD_EIBRS,
1169 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1170 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1171 	SPECTRE_V2_CMD_IBRS,
1172 };
1173 
1174 enum spectre_v2_user_cmd {
1175 	SPECTRE_V2_USER_CMD_NONE,
1176 	SPECTRE_V2_USER_CMD_AUTO,
1177 	SPECTRE_V2_USER_CMD_FORCE,
1178 	SPECTRE_V2_USER_CMD_PRCTL,
1179 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1180 	SPECTRE_V2_USER_CMD_SECCOMP,
1181 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1182 };
1183 
1184 static const char * const spectre_v2_user_strings[] = {
1185 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1186 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1187 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1188 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1189 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1190 };
1191 
1192 static const struct {
1193 	const char			*option;
1194 	enum spectre_v2_user_cmd	cmd;
1195 	bool				secure;
1196 } v2_user_options[] __initconst = {
1197 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1198 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1199 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1200 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1201 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1202 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1203 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1204 };
1205 
1206 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1207 {
1208 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1209 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1210 }
1211 
1212 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1213 
1214 static enum spectre_v2_user_cmd __init
1215 spectre_v2_parse_user_cmdline(void)
1216 {
1217 	char arg[20];
1218 	int ret, i;
1219 
1220 	switch (spectre_v2_cmd) {
1221 	case SPECTRE_V2_CMD_NONE:
1222 		return SPECTRE_V2_USER_CMD_NONE;
1223 	case SPECTRE_V2_CMD_FORCE:
1224 		return SPECTRE_V2_USER_CMD_FORCE;
1225 	default:
1226 		break;
1227 	}
1228 
1229 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1230 				  arg, sizeof(arg));
1231 	if (ret < 0)
1232 		return SPECTRE_V2_USER_CMD_AUTO;
1233 
1234 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1235 		if (match_option(arg, ret, v2_user_options[i].option)) {
1236 			spec_v2_user_print_cond(v2_user_options[i].option,
1237 						v2_user_options[i].secure);
1238 			return v2_user_options[i].cmd;
1239 		}
1240 	}
1241 
1242 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1243 	return SPECTRE_V2_USER_CMD_AUTO;
1244 }
1245 
1246 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1247 {
1248 	return spectre_v2_in_eibrs_mode(mode) || mode == SPECTRE_V2_IBRS;
1249 }
1250 
1251 static void __init
1252 spectre_v2_user_select_mitigation(void)
1253 {
1254 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1255 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1256 	enum spectre_v2_user_cmd cmd;
1257 
1258 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1259 		return;
1260 
1261 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1262 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1263 		smt_possible = false;
1264 
1265 	cmd = spectre_v2_parse_user_cmdline();
1266 	switch (cmd) {
1267 	case SPECTRE_V2_USER_CMD_NONE:
1268 		goto set_mode;
1269 	case SPECTRE_V2_USER_CMD_FORCE:
1270 		mode = SPECTRE_V2_USER_STRICT;
1271 		break;
1272 	case SPECTRE_V2_USER_CMD_AUTO:
1273 	case SPECTRE_V2_USER_CMD_PRCTL:
1274 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1275 		mode = SPECTRE_V2_USER_PRCTL;
1276 		break;
1277 	case SPECTRE_V2_USER_CMD_SECCOMP:
1278 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1279 		if (IS_ENABLED(CONFIG_SECCOMP))
1280 			mode = SPECTRE_V2_USER_SECCOMP;
1281 		else
1282 			mode = SPECTRE_V2_USER_PRCTL;
1283 		break;
1284 	}
1285 
1286 	/* Initialize Indirect Branch Prediction Barrier */
1287 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1288 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1289 
1290 		spectre_v2_user_ibpb = mode;
1291 		switch (cmd) {
1292 		case SPECTRE_V2_USER_CMD_NONE:
1293 			break;
1294 		case SPECTRE_V2_USER_CMD_FORCE:
1295 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1296 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1297 			static_branch_enable(&switch_mm_always_ibpb);
1298 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1299 			break;
1300 		case SPECTRE_V2_USER_CMD_PRCTL:
1301 		case SPECTRE_V2_USER_CMD_AUTO:
1302 		case SPECTRE_V2_USER_CMD_SECCOMP:
1303 			static_branch_enable(&switch_mm_cond_ibpb);
1304 			break;
1305 		}
1306 
1307 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1308 			static_key_enabled(&switch_mm_always_ibpb) ?
1309 			"always-on" : "conditional");
1310 	}
1311 
1312 	/*
1313 	 * If no STIBP, Intel enhanced IBRS is enabled, or SMT impossible, STIBP
1314 	 * is not required.
1315 	 *
1316 	 * Intel's Enhanced IBRS also protects against cross-thread branch target
1317 	 * injection in user-mode as the IBRS bit remains always set which
1318 	 * implicitly enables cross-thread protections.  However, in legacy IBRS
1319 	 * mode, the IBRS bit is set only on kernel entry and cleared on return
1320 	 * to userspace.  AMD Automatic IBRS also does not protect userspace.
1321 	 * These modes therefore disable the implicit cross-thread protection,
1322 	 * so allow for STIBP to be selected in those cases.
1323 	 */
1324 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1325 	    !smt_possible ||
1326 	    (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
1327 	     !boot_cpu_has(X86_FEATURE_AUTOIBRS)))
1328 		return;
1329 
1330 	/*
1331 	 * At this point, an STIBP mode other than "off" has been set.
1332 	 * If STIBP support is not being forced, check if STIBP always-on
1333 	 * is preferred.
1334 	 */
1335 	if (mode != SPECTRE_V2_USER_STRICT &&
1336 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1337 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1338 
1339 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1340 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1341 		if (mode != SPECTRE_V2_USER_STRICT &&
1342 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1343 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1344 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1345 	}
1346 
1347 	spectre_v2_user_stibp = mode;
1348 
1349 set_mode:
1350 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1351 }
1352 
1353 static const char * const spectre_v2_strings[] = {
1354 	[SPECTRE_V2_NONE]			= "Vulnerable",
1355 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1356 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1357 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced / Automatic IBRS",
1358 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced / Automatic IBRS + LFENCE",
1359 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced / Automatic IBRS + Retpolines",
1360 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1361 };
1362 
1363 static const struct {
1364 	const char *option;
1365 	enum spectre_v2_mitigation_cmd cmd;
1366 	bool secure;
1367 } mitigation_options[] __initconst = {
1368 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1369 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1370 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1371 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1372 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1373 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1374 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1375 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1376 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1377 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1378 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1379 };
1380 
1381 static void __init spec_v2_print_cond(const char *reason, bool secure)
1382 {
1383 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1384 		pr_info("%s selected on command line.\n", reason);
1385 }
1386 
1387 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1388 {
1389 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1390 	char arg[20];
1391 	int ret, i;
1392 
1393 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1394 	    cpu_mitigations_off())
1395 		return SPECTRE_V2_CMD_NONE;
1396 
1397 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1398 	if (ret < 0)
1399 		return SPECTRE_V2_CMD_AUTO;
1400 
1401 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1402 		if (!match_option(arg, ret, mitigation_options[i].option))
1403 			continue;
1404 		cmd = mitigation_options[i].cmd;
1405 		break;
1406 	}
1407 
1408 	if (i >= ARRAY_SIZE(mitigation_options)) {
1409 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1410 		return SPECTRE_V2_CMD_AUTO;
1411 	}
1412 
1413 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1414 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1415 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1416 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1417 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1418 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1419 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1420 		       mitigation_options[i].option);
1421 		return SPECTRE_V2_CMD_AUTO;
1422 	}
1423 
1424 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1425 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1426 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1427 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1428 		pr_err("%s selected but CPU doesn't have Enhanced or Automatic IBRS. Switching to AUTO select\n",
1429 		       mitigation_options[i].option);
1430 		return SPECTRE_V2_CMD_AUTO;
1431 	}
1432 
1433 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1434 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1435 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1436 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1437 		       mitigation_options[i].option);
1438 		return SPECTRE_V2_CMD_AUTO;
1439 	}
1440 
1441 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1442 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1443 		       mitigation_options[i].option);
1444 		return SPECTRE_V2_CMD_AUTO;
1445 	}
1446 
1447 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1448 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1449 		       mitigation_options[i].option);
1450 		return SPECTRE_V2_CMD_AUTO;
1451 	}
1452 
1453 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1454 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1455 		       mitigation_options[i].option);
1456 		return SPECTRE_V2_CMD_AUTO;
1457 	}
1458 
1459 	if (cmd == SPECTRE_V2_CMD_IBRS && cpu_feature_enabled(X86_FEATURE_XENPV)) {
1460 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1461 		       mitigation_options[i].option);
1462 		return SPECTRE_V2_CMD_AUTO;
1463 	}
1464 
1465 	spec_v2_print_cond(mitigation_options[i].option,
1466 			   mitigation_options[i].secure);
1467 	return cmd;
1468 }
1469 
1470 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1471 {
1472 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1473 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1474 		return SPECTRE_V2_NONE;
1475 	}
1476 
1477 	return SPECTRE_V2_RETPOLINE;
1478 }
1479 
1480 /* Disable in-kernel use of non-RSB RET predictors */
1481 static void __init spec_ctrl_disable_kernel_rrsba(void)
1482 {
1483 	u64 ia32_cap;
1484 
1485 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1486 		return;
1487 
1488 	ia32_cap = x86_read_arch_cap_msr();
1489 
1490 	if (ia32_cap & ARCH_CAP_RRSBA) {
1491 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1492 		update_spec_ctrl(x86_spec_ctrl_base);
1493 	}
1494 }
1495 
1496 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1497 {
1498 	/*
1499 	 * Similar to context switches, there are two types of RSB attacks
1500 	 * after VM exit:
1501 	 *
1502 	 * 1) RSB underflow
1503 	 *
1504 	 * 2) Poisoned RSB entry
1505 	 *
1506 	 * When retpoline is enabled, both are mitigated by filling/clearing
1507 	 * the RSB.
1508 	 *
1509 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1510 	 * prediction isolation protections, RSB still needs to be cleared
1511 	 * because of #2.  Note that SMEP provides no protection here, unlike
1512 	 * user-space-poisoned RSB entries.
1513 	 *
1514 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1515 	 * bug is present then a LITE version of RSB protection is required,
1516 	 * just a single call needs to retire before a RET is executed.
1517 	 */
1518 	switch (mode) {
1519 	case SPECTRE_V2_NONE:
1520 		return;
1521 
1522 	case SPECTRE_V2_EIBRS_LFENCE:
1523 	case SPECTRE_V2_EIBRS:
1524 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1525 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1526 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1527 		}
1528 		return;
1529 
1530 	case SPECTRE_V2_EIBRS_RETPOLINE:
1531 	case SPECTRE_V2_RETPOLINE:
1532 	case SPECTRE_V2_LFENCE:
1533 	case SPECTRE_V2_IBRS:
1534 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1535 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1536 		return;
1537 	}
1538 
1539 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1540 	dump_stack();
1541 }
1542 
1543 static void __init spectre_v2_select_mitigation(void)
1544 {
1545 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1546 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1547 
1548 	/*
1549 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1550 	 * then nothing to do.
1551 	 */
1552 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1553 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1554 		return;
1555 
1556 	switch (cmd) {
1557 	case SPECTRE_V2_CMD_NONE:
1558 		return;
1559 
1560 	case SPECTRE_V2_CMD_FORCE:
1561 	case SPECTRE_V2_CMD_AUTO:
1562 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1563 			mode = SPECTRE_V2_EIBRS;
1564 			break;
1565 		}
1566 
1567 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1568 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1569 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1570 		    retbleed_cmd != RETBLEED_CMD_STUFF &&
1571 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1572 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1573 			mode = SPECTRE_V2_IBRS;
1574 			break;
1575 		}
1576 
1577 		mode = spectre_v2_select_retpoline();
1578 		break;
1579 
1580 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1581 		pr_err(SPECTRE_V2_LFENCE_MSG);
1582 		mode = SPECTRE_V2_LFENCE;
1583 		break;
1584 
1585 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1586 		mode = SPECTRE_V2_RETPOLINE;
1587 		break;
1588 
1589 	case SPECTRE_V2_CMD_RETPOLINE:
1590 		mode = spectre_v2_select_retpoline();
1591 		break;
1592 
1593 	case SPECTRE_V2_CMD_IBRS:
1594 		mode = SPECTRE_V2_IBRS;
1595 		break;
1596 
1597 	case SPECTRE_V2_CMD_EIBRS:
1598 		mode = SPECTRE_V2_EIBRS;
1599 		break;
1600 
1601 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1602 		mode = SPECTRE_V2_EIBRS_LFENCE;
1603 		break;
1604 
1605 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1606 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1607 		break;
1608 	}
1609 
1610 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1611 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1612 
1613 	if (spectre_v2_in_ibrs_mode(mode)) {
1614 		if (boot_cpu_has(X86_FEATURE_AUTOIBRS)) {
1615 			msr_set_bit(MSR_EFER, _EFER_AUTOIBRS);
1616 		} else {
1617 			x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1618 			update_spec_ctrl(x86_spec_ctrl_base);
1619 		}
1620 	}
1621 
1622 	switch (mode) {
1623 	case SPECTRE_V2_NONE:
1624 	case SPECTRE_V2_EIBRS:
1625 		break;
1626 
1627 	case SPECTRE_V2_IBRS:
1628 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1629 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1630 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1631 		break;
1632 
1633 	case SPECTRE_V2_LFENCE:
1634 	case SPECTRE_V2_EIBRS_LFENCE:
1635 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1636 		fallthrough;
1637 
1638 	case SPECTRE_V2_RETPOLINE:
1639 	case SPECTRE_V2_EIBRS_RETPOLINE:
1640 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1641 		break;
1642 	}
1643 
1644 	/*
1645 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1646 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1647 	 * prediction from a non-RSB predictor is still a risk.
1648 	 */
1649 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1650 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1651 	    mode == SPECTRE_V2_RETPOLINE)
1652 		spec_ctrl_disable_kernel_rrsba();
1653 
1654 	spectre_v2_enabled = mode;
1655 	pr_info("%s\n", spectre_v2_strings[mode]);
1656 
1657 	/*
1658 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1659 	 * context switch.  In general there are two types of RSB attacks
1660 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1661 	 *
1662 	 * 1) RSB underflow
1663 	 *
1664 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1665 	 *    speculated return targets may come from the branch predictor,
1666 	 *    which could have a user-poisoned BTB or BHB entry.
1667 	 *
1668 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1669 	 *    regardless of the state of the RSB.
1670 	 *
1671 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1672 	 *    scenario is mitigated by the IBRS branch prediction isolation
1673 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1674 	 *    protect against this type of attack.
1675 	 *
1676 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1677 	 *
1678 	 * 2) Poisoned RSB entry
1679 	 *
1680 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1681 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1682 	 *    entry.
1683 	 *
1684 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1685 	 *    eIBRS.
1686 	 *
1687 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1688 	 *    RSB clearing.
1689 	 *
1690 	 * So to mitigate all cases, unconditionally fill RSB on context
1691 	 * switches.
1692 	 *
1693 	 * FIXME: Is this pointless for retbleed-affected AMD?
1694 	 */
1695 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1696 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1697 
1698 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1699 
1700 	/*
1701 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1702 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1703 	 * firmware calls only when IBRS / Enhanced / Automatic IBRS aren't
1704 	 * otherwise enabled.
1705 	 *
1706 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1707 	 * the user might select retpoline on the kernel command line and if
1708 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1709 	 * enable IBRS around firmware calls.
1710 	 */
1711 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1712 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1713 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1714 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1715 
1716 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1717 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1718 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1719 		}
1720 
1721 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1722 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1723 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1724 	}
1725 
1726 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1727 	spectre_v2_cmd = cmd;
1728 }
1729 
1730 static void update_stibp_msr(void * __unused)
1731 {
1732 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1733 	update_spec_ctrl(val);
1734 }
1735 
1736 /* Update x86_spec_ctrl_base in case SMT state changed. */
1737 static void update_stibp_strict(void)
1738 {
1739 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1740 
1741 	if (sched_smt_active())
1742 		mask |= SPEC_CTRL_STIBP;
1743 
1744 	if (mask == x86_spec_ctrl_base)
1745 		return;
1746 
1747 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1748 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1749 	x86_spec_ctrl_base = mask;
1750 	on_each_cpu(update_stibp_msr, NULL, 1);
1751 }
1752 
1753 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1754 static void update_indir_branch_cond(void)
1755 {
1756 	if (sched_smt_active())
1757 		static_branch_enable(&switch_to_cond_stibp);
1758 	else
1759 		static_branch_disable(&switch_to_cond_stibp);
1760 }
1761 
1762 #undef pr_fmt
1763 #define pr_fmt(fmt) fmt
1764 
1765 /* Update the static key controlling the MDS CPU buffer clear in idle */
1766 static void update_mds_branch_idle(void)
1767 {
1768 	u64 ia32_cap = x86_read_arch_cap_msr();
1769 
1770 	/*
1771 	 * Enable the idle clearing if SMT is active on CPUs which are
1772 	 * affected only by MSBDS and not any other MDS variant.
1773 	 *
1774 	 * The other variants cannot be mitigated when SMT is enabled, so
1775 	 * clearing the buffers on idle just to prevent the Store Buffer
1776 	 * repartitioning leak would be a window dressing exercise.
1777 	 */
1778 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1779 		return;
1780 
1781 	if (sched_smt_active()) {
1782 		static_branch_enable(&mds_idle_clear);
1783 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1784 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1785 		static_branch_disable(&mds_idle_clear);
1786 	}
1787 }
1788 
1789 #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"
1790 #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"
1791 #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"
1792 
1793 void cpu_bugs_smt_update(void)
1794 {
1795 	mutex_lock(&spec_ctrl_mutex);
1796 
1797 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1798 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1799 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1800 
1801 	switch (spectre_v2_user_stibp) {
1802 	case SPECTRE_V2_USER_NONE:
1803 		break;
1804 	case SPECTRE_V2_USER_STRICT:
1805 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1806 		update_stibp_strict();
1807 		break;
1808 	case SPECTRE_V2_USER_PRCTL:
1809 	case SPECTRE_V2_USER_SECCOMP:
1810 		update_indir_branch_cond();
1811 		break;
1812 	}
1813 
1814 	switch (mds_mitigation) {
1815 	case MDS_MITIGATION_FULL:
1816 	case MDS_MITIGATION_VMWERV:
1817 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1818 			pr_warn_once(MDS_MSG_SMT);
1819 		update_mds_branch_idle();
1820 		break;
1821 	case MDS_MITIGATION_OFF:
1822 		break;
1823 	}
1824 
1825 	switch (taa_mitigation) {
1826 	case TAA_MITIGATION_VERW:
1827 	case TAA_MITIGATION_UCODE_NEEDED:
1828 		if (sched_smt_active())
1829 			pr_warn_once(TAA_MSG_SMT);
1830 		break;
1831 	case TAA_MITIGATION_TSX_DISABLED:
1832 	case TAA_MITIGATION_OFF:
1833 		break;
1834 	}
1835 
1836 	switch (mmio_mitigation) {
1837 	case MMIO_MITIGATION_VERW:
1838 	case MMIO_MITIGATION_UCODE_NEEDED:
1839 		if (sched_smt_active())
1840 			pr_warn_once(MMIO_MSG_SMT);
1841 		break;
1842 	case MMIO_MITIGATION_OFF:
1843 		break;
1844 	}
1845 
1846 	mutex_unlock(&spec_ctrl_mutex);
1847 }
1848 
1849 #undef pr_fmt
1850 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1851 
1852 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1853 
1854 /* The kernel command line selection */
1855 enum ssb_mitigation_cmd {
1856 	SPEC_STORE_BYPASS_CMD_NONE,
1857 	SPEC_STORE_BYPASS_CMD_AUTO,
1858 	SPEC_STORE_BYPASS_CMD_ON,
1859 	SPEC_STORE_BYPASS_CMD_PRCTL,
1860 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1861 };
1862 
1863 static const char * const ssb_strings[] = {
1864 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1865 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1866 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1867 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1868 };
1869 
1870 static const struct {
1871 	const char *option;
1872 	enum ssb_mitigation_cmd cmd;
1873 } ssb_mitigation_options[]  __initconst = {
1874 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1875 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1876 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1877 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1878 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1879 };
1880 
1881 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1882 {
1883 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1884 	char arg[20];
1885 	int ret, i;
1886 
1887 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1888 	    cpu_mitigations_off()) {
1889 		return SPEC_STORE_BYPASS_CMD_NONE;
1890 	} else {
1891 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1892 					  arg, sizeof(arg));
1893 		if (ret < 0)
1894 			return SPEC_STORE_BYPASS_CMD_AUTO;
1895 
1896 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1897 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1898 				continue;
1899 
1900 			cmd = ssb_mitigation_options[i].cmd;
1901 			break;
1902 		}
1903 
1904 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1905 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1906 			return SPEC_STORE_BYPASS_CMD_AUTO;
1907 		}
1908 	}
1909 
1910 	return cmd;
1911 }
1912 
1913 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1914 {
1915 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1916 	enum ssb_mitigation_cmd cmd;
1917 
1918 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1919 		return mode;
1920 
1921 	cmd = ssb_parse_cmdline();
1922 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1923 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1924 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1925 		return mode;
1926 
1927 	switch (cmd) {
1928 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1929 		/*
1930 		 * Choose prctl+seccomp as the default mode if seccomp is
1931 		 * enabled.
1932 		 */
1933 		if (IS_ENABLED(CONFIG_SECCOMP))
1934 			mode = SPEC_STORE_BYPASS_SECCOMP;
1935 		else
1936 			mode = SPEC_STORE_BYPASS_PRCTL;
1937 		break;
1938 	case SPEC_STORE_BYPASS_CMD_ON:
1939 		mode = SPEC_STORE_BYPASS_DISABLE;
1940 		break;
1941 	case SPEC_STORE_BYPASS_CMD_AUTO:
1942 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1943 		mode = SPEC_STORE_BYPASS_PRCTL;
1944 		break;
1945 	case SPEC_STORE_BYPASS_CMD_NONE:
1946 		break;
1947 	}
1948 
1949 	/*
1950 	 * We have three CPU feature flags that are in play here:
1951 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1952 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1953 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1954 	 */
1955 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1956 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1957 		/*
1958 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1959 		 * use a completely different MSR and bit dependent on family.
1960 		 */
1961 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1962 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1963 			x86_amd_ssb_disable();
1964 		} else {
1965 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1966 			update_spec_ctrl(x86_spec_ctrl_base);
1967 		}
1968 	}
1969 
1970 	return mode;
1971 }
1972 
1973 static void ssb_select_mitigation(void)
1974 {
1975 	ssb_mode = __ssb_select_mitigation();
1976 
1977 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1978 		pr_info("%s\n", ssb_strings[ssb_mode]);
1979 }
1980 
1981 #undef pr_fmt
1982 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1983 
1984 static void task_update_spec_tif(struct task_struct *tsk)
1985 {
1986 	/* Force the update of the real TIF bits */
1987 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1988 
1989 	/*
1990 	 * Immediately update the speculation control MSRs for the current
1991 	 * task, but for a non-current task delay setting the CPU
1992 	 * mitigation until it is scheduled next.
1993 	 *
1994 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1995 	 * always the current task.
1996 	 */
1997 	if (tsk == current)
1998 		speculation_ctrl_update_current();
1999 }
2000 
2001 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
2002 {
2003 
2004 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2005 		return -EPERM;
2006 
2007 	switch (ctrl) {
2008 	case PR_SPEC_ENABLE:
2009 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
2010 		return 0;
2011 	case PR_SPEC_DISABLE:
2012 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
2013 		return 0;
2014 	default:
2015 		return -ERANGE;
2016 	}
2017 }
2018 
2019 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
2020 {
2021 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
2022 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
2023 		return -ENXIO;
2024 
2025 	switch (ctrl) {
2026 	case PR_SPEC_ENABLE:
2027 		/* If speculation is force disabled, enable is not allowed */
2028 		if (task_spec_ssb_force_disable(task))
2029 			return -EPERM;
2030 		task_clear_spec_ssb_disable(task);
2031 		task_clear_spec_ssb_noexec(task);
2032 		task_update_spec_tif(task);
2033 		break;
2034 	case PR_SPEC_DISABLE:
2035 		task_set_spec_ssb_disable(task);
2036 		task_clear_spec_ssb_noexec(task);
2037 		task_update_spec_tif(task);
2038 		break;
2039 	case PR_SPEC_FORCE_DISABLE:
2040 		task_set_spec_ssb_disable(task);
2041 		task_set_spec_ssb_force_disable(task);
2042 		task_clear_spec_ssb_noexec(task);
2043 		task_update_spec_tif(task);
2044 		break;
2045 	case PR_SPEC_DISABLE_NOEXEC:
2046 		if (task_spec_ssb_force_disable(task))
2047 			return -EPERM;
2048 		task_set_spec_ssb_disable(task);
2049 		task_set_spec_ssb_noexec(task);
2050 		task_update_spec_tif(task);
2051 		break;
2052 	default:
2053 		return -ERANGE;
2054 	}
2055 	return 0;
2056 }
2057 
2058 static bool is_spec_ib_user_controlled(void)
2059 {
2060 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
2061 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2062 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
2063 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
2064 }
2065 
2066 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
2067 {
2068 	switch (ctrl) {
2069 	case PR_SPEC_ENABLE:
2070 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2071 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2072 			return 0;
2073 
2074 		/*
2075 		 * With strict mode for both IBPB and STIBP, the instruction
2076 		 * code paths avoid checking this task flag and instead,
2077 		 * unconditionally run the instruction. However, STIBP and IBPB
2078 		 * are independent and either can be set to conditionally
2079 		 * enabled regardless of the mode of the other.
2080 		 *
2081 		 * If either is set to conditional, allow the task flag to be
2082 		 * updated, unless it was force-disabled by a previous prctl
2083 		 * call. Currently, this is possible on an AMD CPU which has the
2084 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
2085 		 * kernel is booted with 'spectre_v2_user=seccomp', then
2086 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
2087 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
2088 		 */
2089 		if (!is_spec_ib_user_controlled() ||
2090 		    task_spec_ib_force_disable(task))
2091 			return -EPERM;
2092 
2093 		task_clear_spec_ib_disable(task);
2094 		task_update_spec_tif(task);
2095 		break;
2096 	case PR_SPEC_DISABLE:
2097 	case PR_SPEC_FORCE_DISABLE:
2098 		/*
2099 		 * Indirect branch speculation is always allowed when
2100 		 * mitigation is force disabled.
2101 		 */
2102 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2103 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2104 			return -EPERM;
2105 
2106 		if (!is_spec_ib_user_controlled())
2107 			return 0;
2108 
2109 		task_set_spec_ib_disable(task);
2110 		if (ctrl == PR_SPEC_FORCE_DISABLE)
2111 			task_set_spec_ib_force_disable(task);
2112 		task_update_spec_tif(task);
2113 		if (task == current)
2114 			indirect_branch_prediction_barrier();
2115 		break;
2116 	default:
2117 		return -ERANGE;
2118 	}
2119 	return 0;
2120 }
2121 
2122 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
2123 			     unsigned long ctrl)
2124 {
2125 	switch (which) {
2126 	case PR_SPEC_STORE_BYPASS:
2127 		return ssb_prctl_set(task, ctrl);
2128 	case PR_SPEC_INDIRECT_BRANCH:
2129 		return ib_prctl_set(task, ctrl);
2130 	case PR_SPEC_L1D_FLUSH:
2131 		return l1d_flush_prctl_set(task, ctrl);
2132 	default:
2133 		return -ENODEV;
2134 	}
2135 }
2136 
2137 #ifdef CONFIG_SECCOMP
2138 void arch_seccomp_spec_mitigate(struct task_struct *task)
2139 {
2140 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
2141 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2142 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2143 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
2144 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2145 }
2146 #endif
2147 
2148 static int l1d_flush_prctl_get(struct task_struct *task)
2149 {
2150 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2151 		return PR_SPEC_FORCE_DISABLE;
2152 
2153 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
2154 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2155 	else
2156 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2157 }
2158 
2159 static int ssb_prctl_get(struct task_struct *task)
2160 {
2161 	switch (ssb_mode) {
2162 	case SPEC_STORE_BYPASS_NONE:
2163 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2164 			return PR_SPEC_ENABLE;
2165 		return PR_SPEC_NOT_AFFECTED;
2166 	case SPEC_STORE_BYPASS_DISABLE:
2167 		return PR_SPEC_DISABLE;
2168 	case SPEC_STORE_BYPASS_SECCOMP:
2169 	case SPEC_STORE_BYPASS_PRCTL:
2170 		if (task_spec_ssb_force_disable(task))
2171 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2172 		if (task_spec_ssb_noexec(task))
2173 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2174 		if (task_spec_ssb_disable(task))
2175 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2176 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2177 	}
2178 	BUG();
2179 }
2180 
2181 static int ib_prctl_get(struct task_struct *task)
2182 {
2183 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2184 		return PR_SPEC_NOT_AFFECTED;
2185 
2186 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2187 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2188 		return PR_SPEC_ENABLE;
2189 	else if (is_spec_ib_user_controlled()) {
2190 		if (task_spec_ib_force_disable(task))
2191 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2192 		if (task_spec_ib_disable(task))
2193 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2194 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2195 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2196 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2197 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2198 		return PR_SPEC_DISABLE;
2199 	else
2200 		return PR_SPEC_NOT_AFFECTED;
2201 }
2202 
2203 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2204 {
2205 	switch (which) {
2206 	case PR_SPEC_STORE_BYPASS:
2207 		return ssb_prctl_get(task);
2208 	case PR_SPEC_INDIRECT_BRANCH:
2209 		return ib_prctl_get(task);
2210 	case PR_SPEC_L1D_FLUSH:
2211 		return l1d_flush_prctl_get(task);
2212 	default:
2213 		return -ENODEV;
2214 	}
2215 }
2216 
2217 void x86_spec_ctrl_setup_ap(void)
2218 {
2219 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2220 		update_spec_ctrl(x86_spec_ctrl_base);
2221 
2222 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2223 		x86_amd_ssb_disable();
2224 }
2225 
2226 bool itlb_multihit_kvm_mitigation;
2227 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2228 
2229 #undef pr_fmt
2230 #define pr_fmt(fmt)	"L1TF: " fmt
2231 
2232 /* Default mitigation for L1TF-affected CPUs */
2233 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2234 #if IS_ENABLED(CONFIG_KVM_INTEL)
2235 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2236 #endif
2237 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2238 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2239 
2240 /*
2241  * These CPUs all support 44bits physical address space internally in the
2242  * cache but CPUID can report a smaller number of physical address bits.
2243  *
2244  * The L1TF mitigation uses the top most address bit for the inversion of
2245  * non present PTEs. When the installed memory reaches into the top most
2246  * address bit due to memory holes, which has been observed on machines
2247  * which report 36bits physical address bits and have 32G RAM installed,
2248  * then the mitigation range check in l1tf_select_mitigation() triggers.
2249  * This is a false positive because the mitigation is still possible due to
2250  * the fact that the cache uses 44bit internally. Use the cache bits
2251  * instead of the reported physical bits and adjust them on the affected
2252  * machines to 44bit if the reported bits are less than 44.
2253  */
2254 static void override_cache_bits(struct cpuinfo_x86 *c)
2255 {
2256 	if (c->x86 != 6)
2257 		return;
2258 
2259 	switch (c->x86_model) {
2260 	case INTEL_FAM6_NEHALEM:
2261 	case INTEL_FAM6_WESTMERE:
2262 	case INTEL_FAM6_SANDYBRIDGE:
2263 	case INTEL_FAM6_IVYBRIDGE:
2264 	case INTEL_FAM6_HASWELL:
2265 	case INTEL_FAM6_HASWELL_L:
2266 	case INTEL_FAM6_HASWELL_G:
2267 	case INTEL_FAM6_BROADWELL:
2268 	case INTEL_FAM6_BROADWELL_G:
2269 	case INTEL_FAM6_SKYLAKE_L:
2270 	case INTEL_FAM6_SKYLAKE:
2271 	case INTEL_FAM6_KABYLAKE_L:
2272 	case INTEL_FAM6_KABYLAKE:
2273 		if (c->x86_cache_bits < 44)
2274 			c->x86_cache_bits = 44;
2275 		break;
2276 	}
2277 }
2278 
2279 static void __init l1tf_select_mitigation(void)
2280 {
2281 	u64 half_pa;
2282 
2283 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2284 		return;
2285 
2286 	if (cpu_mitigations_off())
2287 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2288 	else if (cpu_mitigations_auto_nosmt())
2289 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2290 
2291 	override_cache_bits(&boot_cpu_data);
2292 
2293 	switch (l1tf_mitigation) {
2294 	case L1TF_MITIGATION_OFF:
2295 	case L1TF_MITIGATION_FLUSH_NOWARN:
2296 	case L1TF_MITIGATION_FLUSH:
2297 		break;
2298 	case L1TF_MITIGATION_FLUSH_NOSMT:
2299 	case L1TF_MITIGATION_FULL:
2300 		cpu_smt_disable(false);
2301 		break;
2302 	case L1TF_MITIGATION_FULL_FORCE:
2303 		cpu_smt_disable(true);
2304 		break;
2305 	}
2306 
2307 #if CONFIG_PGTABLE_LEVELS == 2
2308 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2309 	return;
2310 #endif
2311 
2312 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2313 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2314 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2315 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2316 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2317 				half_pa);
2318 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2319 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2320 		return;
2321 	}
2322 
2323 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2324 }
2325 
2326 static int __init l1tf_cmdline(char *str)
2327 {
2328 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2329 		return 0;
2330 
2331 	if (!str)
2332 		return -EINVAL;
2333 
2334 	if (!strcmp(str, "off"))
2335 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2336 	else if (!strcmp(str, "flush,nowarn"))
2337 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2338 	else if (!strcmp(str, "flush"))
2339 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2340 	else if (!strcmp(str, "flush,nosmt"))
2341 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2342 	else if (!strcmp(str, "full"))
2343 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2344 	else if (!strcmp(str, "full,force"))
2345 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2346 
2347 	return 0;
2348 }
2349 early_param("l1tf", l1tf_cmdline);
2350 
2351 #undef pr_fmt
2352 #define pr_fmt(fmt)	"Speculative Return Stack Overflow: " fmt
2353 
2354 enum srso_mitigation {
2355 	SRSO_MITIGATION_NONE,
2356 	SRSO_MITIGATION_UCODE_NEEDED,
2357 	SRSO_MITIGATION_SAFE_RET_UCODE_NEEDED,
2358 	SRSO_MITIGATION_MICROCODE,
2359 	SRSO_MITIGATION_SAFE_RET,
2360 	SRSO_MITIGATION_IBPB,
2361 	SRSO_MITIGATION_IBPB_ON_VMEXIT,
2362 };
2363 
2364 enum srso_mitigation_cmd {
2365 	SRSO_CMD_OFF,
2366 	SRSO_CMD_MICROCODE,
2367 	SRSO_CMD_SAFE_RET,
2368 	SRSO_CMD_IBPB,
2369 	SRSO_CMD_IBPB_ON_VMEXIT,
2370 };
2371 
2372 static const char * const srso_strings[] = {
2373 	[SRSO_MITIGATION_NONE]			= "Vulnerable",
2374 	[SRSO_MITIGATION_UCODE_NEEDED]		= "Vulnerable: No microcode",
2375 	[SRSO_MITIGATION_SAFE_RET_UCODE_NEEDED]	= "Vulnerable: Safe RET, no microcode",
2376 	[SRSO_MITIGATION_MICROCODE]		= "Vulnerable: Microcode, no safe RET",
2377 	[SRSO_MITIGATION_SAFE_RET]		= "Mitigation: Safe RET",
2378 	[SRSO_MITIGATION_IBPB]			= "Mitigation: IBPB",
2379 	[SRSO_MITIGATION_IBPB_ON_VMEXIT]	= "Mitigation: IBPB on VMEXIT only"
2380 };
2381 
2382 static enum srso_mitigation srso_mitigation __ro_after_init = SRSO_MITIGATION_NONE;
2383 static enum srso_mitigation_cmd srso_cmd __ro_after_init = SRSO_CMD_SAFE_RET;
2384 
2385 static int __init srso_parse_cmdline(char *str)
2386 {
2387 	if (!str)
2388 		return -EINVAL;
2389 
2390 	if (!strcmp(str, "off"))
2391 		srso_cmd = SRSO_CMD_OFF;
2392 	else if (!strcmp(str, "microcode"))
2393 		srso_cmd = SRSO_CMD_MICROCODE;
2394 	else if (!strcmp(str, "safe-ret"))
2395 		srso_cmd = SRSO_CMD_SAFE_RET;
2396 	else if (!strcmp(str, "ibpb"))
2397 		srso_cmd = SRSO_CMD_IBPB;
2398 	else if (!strcmp(str, "ibpb-vmexit"))
2399 		srso_cmd = SRSO_CMD_IBPB_ON_VMEXIT;
2400 	else
2401 		pr_err("Ignoring unknown SRSO option (%s).", str);
2402 
2403 	return 0;
2404 }
2405 early_param("spec_rstack_overflow", srso_parse_cmdline);
2406 
2407 #define SRSO_NOTICE "WARNING: See https://kernel.org/doc/html/latest/admin-guide/hw-vuln/srso.html for mitigation options."
2408 
2409 static void __init srso_select_mitigation(void)
2410 {
2411 	bool has_microcode = boot_cpu_has(X86_FEATURE_IBPB_BRTYPE);
2412 
2413 	if (cpu_mitigations_off())
2414 		return;
2415 
2416 	if (!boot_cpu_has_bug(X86_BUG_SRSO)) {
2417 		if (boot_cpu_has(X86_FEATURE_SBPB))
2418 			x86_pred_cmd = PRED_CMD_SBPB;
2419 		return;
2420 	}
2421 
2422 	if (has_microcode) {
2423 		/*
2424 		 * Zen1/2 with SMT off aren't vulnerable after the right
2425 		 * IBPB microcode has been applied.
2426 		 *
2427 		 * Zen1/2 don't have SBPB, no need to try to enable it here.
2428 		 */
2429 		if (boot_cpu_data.x86 < 0x19 && !cpu_smt_possible()) {
2430 			setup_force_cpu_cap(X86_FEATURE_SRSO_NO);
2431 			return;
2432 		}
2433 
2434 		if (retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2435 			srso_mitigation = SRSO_MITIGATION_IBPB;
2436 			goto out;
2437 		}
2438 	} else {
2439 		pr_warn("IBPB-extending microcode not applied!\n");
2440 		pr_warn(SRSO_NOTICE);
2441 
2442 		/* may be overwritten by SRSO_CMD_SAFE_RET below */
2443 		srso_mitigation = SRSO_MITIGATION_UCODE_NEEDED;
2444 	}
2445 
2446 	switch (srso_cmd) {
2447 	case SRSO_CMD_OFF:
2448 		if (boot_cpu_has(X86_FEATURE_SBPB))
2449 			x86_pred_cmd = PRED_CMD_SBPB;
2450 		return;
2451 
2452 	case SRSO_CMD_MICROCODE:
2453 		if (has_microcode) {
2454 			srso_mitigation = SRSO_MITIGATION_MICROCODE;
2455 			pr_warn(SRSO_NOTICE);
2456 		}
2457 		break;
2458 
2459 	case SRSO_CMD_SAFE_RET:
2460 		if (IS_ENABLED(CONFIG_CPU_SRSO)) {
2461 			/*
2462 			 * Enable the return thunk for generated code
2463 			 * like ftrace, static_call, etc.
2464 			 */
2465 			setup_force_cpu_cap(X86_FEATURE_RETHUNK);
2466 			setup_force_cpu_cap(X86_FEATURE_UNRET);
2467 
2468 			if (boot_cpu_data.x86 == 0x19) {
2469 				setup_force_cpu_cap(X86_FEATURE_SRSO_ALIAS);
2470 				x86_return_thunk = srso_alias_return_thunk;
2471 			} else {
2472 				setup_force_cpu_cap(X86_FEATURE_SRSO);
2473 				x86_return_thunk = srso_return_thunk;
2474 			}
2475 			if (has_microcode)
2476 				srso_mitigation = SRSO_MITIGATION_SAFE_RET;
2477 			else
2478 				srso_mitigation = SRSO_MITIGATION_SAFE_RET_UCODE_NEEDED;
2479 		} else {
2480 			pr_err("WARNING: kernel not compiled with CPU_SRSO.\n");
2481 		}
2482 		break;
2483 
2484 	case SRSO_CMD_IBPB:
2485 		if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
2486 			if (has_microcode) {
2487 				setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
2488 				srso_mitigation = SRSO_MITIGATION_IBPB;
2489 			}
2490 		} else {
2491 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
2492 		}
2493 		break;
2494 
2495 	case SRSO_CMD_IBPB_ON_VMEXIT:
2496 		if (IS_ENABLED(CONFIG_CPU_SRSO)) {
2497 			if (!boot_cpu_has(X86_FEATURE_ENTRY_IBPB) && has_microcode) {
2498 				setup_force_cpu_cap(X86_FEATURE_IBPB_ON_VMEXIT);
2499 				srso_mitigation = SRSO_MITIGATION_IBPB_ON_VMEXIT;
2500 			}
2501 		} else {
2502 			pr_err("WARNING: kernel not compiled with CPU_SRSO.\n");
2503                 }
2504 		break;
2505 	}
2506 
2507 out:
2508 	pr_info("%s\n", srso_strings[srso_mitigation]);
2509 }
2510 
2511 #undef pr_fmt
2512 #define pr_fmt(fmt) fmt
2513 
2514 #ifdef CONFIG_SYSFS
2515 
2516 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2517 
2518 #if IS_ENABLED(CONFIG_KVM_INTEL)
2519 static const char * const l1tf_vmx_states[] = {
2520 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2521 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2522 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2523 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2524 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2525 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2526 };
2527 
2528 static ssize_t l1tf_show_state(char *buf)
2529 {
2530 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2531 		return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2532 
2533 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2534 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2535 	     sched_smt_active())) {
2536 		return sysfs_emit(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2537 				  l1tf_vmx_states[l1tf_vmx_mitigation]);
2538 	}
2539 
2540 	return sysfs_emit(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2541 			  l1tf_vmx_states[l1tf_vmx_mitigation],
2542 			  sched_smt_active() ? "vulnerable" : "disabled");
2543 }
2544 
2545 static ssize_t itlb_multihit_show_state(char *buf)
2546 {
2547 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2548 	    !boot_cpu_has(X86_FEATURE_VMX))
2549 		return sysfs_emit(buf, "KVM: Mitigation: VMX unsupported\n");
2550 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2551 		return sysfs_emit(buf, "KVM: Mitigation: VMX disabled\n");
2552 	else if (itlb_multihit_kvm_mitigation)
2553 		return sysfs_emit(buf, "KVM: Mitigation: Split huge pages\n");
2554 	else
2555 		return sysfs_emit(buf, "KVM: Vulnerable\n");
2556 }
2557 #else
2558 static ssize_t l1tf_show_state(char *buf)
2559 {
2560 	return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2561 }
2562 
2563 static ssize_t itlb_multihit_show_state(char *buf)
2564 {
2565 	return sysfs_emit(buf, "Processor vulnerable\n");
2566 }
2567 #endif
2568 
2569 static ssize_t mds_show_state(char *buf)
2570 {
2571 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2572 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2573 				  mds_strings[mds_mitigation]);
2574 	}
2575 
2576 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2577 		return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2578 				  (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2579 				   sched_smt_active() ? "mitigated" : "disabled"));
2580 	}
2581 
2582 	return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2583 			  sched_smt_active() ? "vulnerable" : "disabled");
2584 }
2585 
2586 static ssize_t tsx_async_abort_show_state(char *buf)
2587 {
2588 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2589 	    (taa_mitigation == TAA_MITIGATION_OFF))
2590 		return sysfs_emit(buf, "%s\n", taa_strings[taa_mitigation]);
2591 
2592 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2593 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2594 				  taa_strings[taa_mitigation]);
2595 	}
2596 
2597 	return sysfs_emit(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2598 			  sched_smt_active() ? "vulnerable" : "disabled");
2599 }
2600 
2601 static ssize_t mmio_stale_data_show_state(char *buf)
2602 {
2603 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2604 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2605 
2606 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2607 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2608 
2609 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2610 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2611 				  mmio_strings[mmio_mitigation]);
2612 	}
2613 
2614 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2615 			  sched_smt_active() ? "vulnerable" : "disabled");
2616 }
2617 
2618 static char *stibp_state(void)
2619 {
2620 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
2621 	    !boot_cpu_has(X86_FEATURE_AUTOIBRS))
2622 		return "";
2623 
2624 	switch (spectre_v2_user_stibp) {
2625 	case SPECTRE_V2_USER_NONE:
2626 		return ", STIBP: disabled";
2627 	case SPECTRE_V2_USER_STRICT:
2628 		return ", STIBP: forced";
2629 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2630 		return ", STIBP: always-on";
2631 	case SPECTRE_V2_USER_PRCTL:
2632 	case SPECTRE_V2_USER_SECCOMP:
2633 		if (static_key_enabled(&switch_to_cond_stibp))
2634 			return ", STIBP: conditional";
2635 	}
2636 	return "";
2637 }
2638 
2639 static char *ibpb_state(void)
2640 {
2641 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2642 		if (static_key_enabled(&switch_mm_always_ibpb))
2643 			return ", IBPB: always-on";
2644 		if (static_key_enabled(&switch_mm_cond_ibpb))
2645 			return ", IBPB: conditional";
2646 		return ", IBPB: disabled";
2647 	}
2648 	return "";
2649 }
2650 
2651 static char *pbrsb_eibrs_state(void)
2652 {
2653 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2654 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2655 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2656 			return ", PBRSB-eIBRS: SW sequence";
2657 		else
2658 			return ", PBRSB-eIBRS: Vulnerable";
2659 	} else {
2660 		return ", PBRSB-eIBRS: Not affected";
2661 	}
2662 }
2663 
2664 static ssize_t spectre_v2_show_state(char *buf)
2665 {
2666 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2667 		return sysfs_emit(buf, "Vulnerable: LFENCE\n");
2668 
2669 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2670 		return sysfs_emit(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2671 
2672 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2673 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2674 		return sysfs_emit(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2675 
2676 	return sysfs_emit(buf, "%s%s%s%s%s%s%s\n",
2677 			  spectre_v2_strings[spectre_v2_enabled],
2678 			  ibpb_state(),
2679 			  boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2680 			  stibp_state(),
2681 			  boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2682 			  pbrsb_eibrs_state(),
2683 			  spectre_v2_module_string());
2684 }
2685 
2686 static ssize_t srbds_show_state(char *buf)
2687 {
2688 	return sysfs_emit(buf, "%s\n", srbds_strings[srbds_mitigation]);
2689 }
2690 
2691 static ssize_t retbleed_show_state(char *buf)
2692 {
2693 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2694 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2695 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2696 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2697 			return sysfs_emit(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2698 
2699 		return sysfs_emit(buf, "%s; SMT %s\n", retbleed_strings[retbleed_mitigation],
2700 				  !sched_smt_active() ? "disabled" :
2701 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2702 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2703 				  "enabled with STIBP protection" : "vulnerable");
2704 	}
2705 
2706 	return sysfs_emit(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2707 }
2708 
2709 static ssize_t srso_show_state(char *buf)
2710 {
2711 	if (boot_cpu_has(X86_FEATURE_SRSO_NO))
2712 		return sysfs_emit(buf, "Mitigation: SMT disabled\n");
2713 
2714 	return sysfs_emit(buf, "%s\n", srso_strings[srso_mitigation]);
2715 }
2716 
2717 static ssize_t gds_show_state(char *buf)
2718 {
2719 	return sysfs_emit(buf, "%s\n", gds_strings[gds_mitigation]);
2720 }
2721 
2722 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2723 			       char *buf, unsigned int bug)
2724 {
2725 	if (!boot_cpu_has_bug(bug))
2726 		return sysfs_emit(buf, "Not affected\n");
2727 
2728 	switch (bug) {
2729 	case X86_BUG_CPU_MELTDOWN:
2730 		if (boot_cpu_has(X86_FEATURE_PTI))
2731 			return sysfs_emit(buf, "Mitigation: PTI\n");
2732 
2733 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2734 			return sysfs_emit(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2735 
2736 		break;
2737 
2738 	case X86_BUG_SPECTRE_V1:
2739 		return sysfs_emit(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2740 
2741 	case X86_BUG_SPECTRE_V2:
2742 		return spectre_v2_show_state(buf);
2743 
2744 	case X86_BUG_SPEC_STORE_BYPASS:
2745 		return sysfs_emit(buf, "%s\n", ssb_strings[ssb_mode]);
2746 
2747 	case X86_BUG_L1TF:
2748 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2749 			return l1tf_show_state(buf);
2750 		break;
2751 
2752 	case X86_BUG_MDS:
2753 		return mds_show_state(buf);
2754 
2755 	case X86_BUG_TAA:
2756 		return tsx_async_abort_show_state(buf);
2757 
2758 	case X86_BUG_ITLB_MULTIHIT:
2759 		return itlb_multihit_show_state(buf);
2760 
2761 	case X86_BUG_SRBDS:
2762 		return srbds_show_state(buf);
2763 
2764 	case X86_BUG_MMIO_STALE_DATA:
2765 	case X86_BUG_MMIO_UNKNOWN:
2766 		return mmio_stale_data_show_state(buf);
2767 
2768 	case X86_BUG_RETBLEED:
2769 		return retbleed_show_state(buf);
2770 
2771 	case X86_BUG_SRSO:
2772 		return srso_show_state(buf);
2773 
2774 	case X86_BUG_GDS:
2775 		return gds_show_state(buf);
2776 
2777 	default:
2778 		break;
2779 	}
2780 
2781 	return sysfs_emit(buf, "Vulnerable\n");
2782 }
2783 
2784 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2785 {
2786 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2787 }
2788 
2789 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2790 {
2791 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2792 }
2793 
2794 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2795 {
2796 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2797 }
2798 
2799 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2800 {
2801 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2802 }
2803 
2804 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2805 {
2806 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2807 }
2808 
2809 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2810 {
2811 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2812 }
2813 
2814 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2815 {
2816 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2817 }
2818 
2819 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2820 {
2821 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2822 }
2823 
2824 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2825 {
2826 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2827 }
2828 
2829 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2830 {
2831 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2832 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2833 	else
2834 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2835 }
2836 
2837 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2838 {
2839 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2840 }
2841 
2842 ssize_t cpu_show_spec_rstack_overflow(struct device *dev, struct device_attribute *attr, char *buf)
2843 {
2844 	return cpu_show_common(dev, attr, buf, X86_BUG_SRSO);
2845 }
2846 
2847 ssize_t cpu_show_gds(struct device *dev, struct device_attribute *attr, char *buf)
2848 {
2849 	return cpu_show_common(dev, attr, buf, X86_BUG_GDS);
2850 }
2851 #endif
2852