xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision 83146efc)
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 	default:
1023 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1024 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
1025 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
1026 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
1027 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
1028 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
1029 		}
1030 
1031 		/*
1032 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
1033 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
1034 		 * be set accordingly below.
1035 		 */
1036 
1037 		break;
1038 	}
1039 
1040 	switch (retbleed_mitigation) {
1041 	case RETBLEED_MITIGATION_UNRET:
1042 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
1043 		setup_force_cpu_cap(X86_FEATURE_UNRET);
1044 
1045 		if (IS_ENABLED(CONFIG_RETHUNK))
1046 			x86_return_thunk = retbleed_return_thunk;
1047 
1048 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
1049 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
1050 			pr_err(RETBLEED_UNTRAIN_MSG);
1051 
1052 		mitigate_smt = true;
1053 		break;
1054 
1055 	case RETBLEED_MITIGATION_IBPB:
1056 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
1057 		setup_force_cpu_cap(X86_FEATURE_IBPB_ON_VMEXIT);
1058 		mitigate_smt = true;
1059 		break;
1060 
1061 	case RETBLEED_MITIGATION_STUFF:
1062 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
1063 		setup_force_cpu_cap(X86_FEATURE_CALL_DEPTH);
1064 		x86_set_skl_return_thunk();
1065 		break;
1066 
1067 	default:
1068 		break;
1069 	}
1070 
1071 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
1072 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
1073 		cpu_smt_disable(false);
1074 
1075 	/*
1076 	 * Let IBRS trump all on Intel without affecting the effects of the
1077 	 * retbleed= cmdline option except for call depth based stuffing
1078 	 */
1079 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1080 		switch (spectre_v2_enabled) {
1081 		case SPECTRE_V2_IBRS:
1082 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
1083 			break;
1084 		case SPECTRE_V2_EIBRS:
1085 		case SPECTRE_V2_EIBRS_RETPOLINE:
1086 		case SPECTRE_V2_EIBRS_LFENCE:
1087 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
1088 			break;
1089 		default:
1090 			if (retbleed_mitigation != RETBLEED_MITIGATION_STUFF)
1091 				pr_err(RETBLEED_INTEL_MSG);
1092 		}
1093 	}
1094 
1095 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
1096 }
1097 
1098 #undef pr_fmt
1099 #define pr_fmt(fmt)     "Spectre V2 : " fmt
1100 
1101 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
1102 	SPECTRE_V2_USER_NONE;
1103 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
1104 	SPECTRE_V2_USER_NONE;
1105 
1106 #ifdef CONFIG_RETPOLINE
1107 static bool spectre_v2_bad_module;
1108 
1109 bool retpoline_module_ok(bool has_retpoline)
1110 {
1111 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
1112 		return true;
1113 
1114 	pr_err("System may be vulnerable to spectre v2\n");
1115 	spectre_v2_bad_module = true;
1116 	return false;
1117 }
1118 
1119 static inline const char *spectre_v2_module_string(void)
1120 {
1121 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
1122 }
1123 #else
1124 static inline const char *spectre_v2_module_string(void) { return ""; }
1125 #endif
1126 
1127 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
1128 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
1129 #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"
1130 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
1131 
1132 #ifdef CONFIG_BPF_SYSCALL
1133 void unpriv_ebpf_notify(int new_state)
1134 {
1135 	if (new_state)
1136 		return;
1137 
1138 	/* Unprivileged eBPF is enabled */
1139 
1140 	switch (spectre_v2_enabled) {
1141 	case SPECTRE_V2_EIBRS:
1142 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1143 		break;
1144 	case SPECTRE_V2_EIBRS_LFENCE:
1145 		if (sched_smt_active())
1146 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1147 		break;
1148 	default:
1149 		break;
1150 	}
1151 }
1152 #endif
1153 
1154 static inline bool match_option(const char *arg, int arglen, const char *opt)
1155 {
1156 	int len = strlen(opt);
1157 
1158 	return len == arglen && !strncmp(arg, opt, len);
1159 }
1160 
1161 /* The kernel command line selection for spectre v2 */
1162 enum spectre_v2_mitigation_cmd {
1163 	SPECTRE_V2_CMD_NONE,
1164 	SPECTRE_V2_CMD_AUTO,
1165 	SPECTRE_V2_CMD_FORCE,
1166 	SPECTRE_V2_CMD_RETPOLINE,
1167 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1168 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1169 	SPECTRE_V2_CMD_EIBRS,
1170 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1171 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1172 	SPECTRE_V2_CMD_IBRS,
1173 };
1174 
1175 enum spectre_v2_user_cmd {
1176 	SPECTRE_V2_USER_CMD_NONE,
1177 	SPECTRE_V2_USER_CMD_AUTO,
1178 	SPECTRE_V2_USER_CMD_FORCE,
1179 	SPECTRE_V2_USER_CMD_PRCTL,
1180 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1181 	SPECTRE_V2_USER_CMD_SECCOMP,
1182 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1183 };
1184 
1185 static const char * const spectre_v2_user_strings[] = {
1186 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1187 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1188 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1189 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1190 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1191 };
1192 
1193 static const struct {
1194 	const char			*option;
1195 	enum spectre_v2_user_cmd	cmd;
1196 	bool				secure;
1197 } v2_user_options[] __initconst = {
1198 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1199 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1200 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1201 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1202 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1203 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1204 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1205 };
1206 
1207 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1208 {
1209 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1210 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1211 }
1212 
1213 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1214 
1215 static enum spectre_v2_user_cmd __init
1216 spectre_v2_parse_user_cmdline(void)
1217 {
1218 	char arg[20];
1219 	int ret, i;
1220 
1221 	switch (spectre_v2_cmd) {
1222 	case SPECTRE_V2_CMD_NONE:
1223 		return SPECTRE_V2_USER_CMD_NONE;
1224 	case SPECTRE_V2_CMD_FORCE:
1225 		return SPECTRE_V2_USER_CMD_FORCE;
1226 	default:
1227 		break;
1228 	}
1229 
1230 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1231 				  arg, sizeof(arg));
1232 	if (ret < 0)
1233 		return SPECTRE_V2_USER_CMD_AUTO;
1234 
1235 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1236 		if (match_option(arg, ret, v2_user_options[i].option)) {
1237 			spec_v2_user_print_cond(v2_user_options[i].option,
1238 						v2_user_options[i].secure);
1239 			return v2_user_options[i].cmd;
1240 		}
1241 	}
1242 
1243 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1244 	return SPECTRE_V2_USER_CMD_AUTO;
1245 }
1246 
1247 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1248 {
1249 	return spectre_v2_in_eibrs_mode(mode) || mode == SPECTRE_V2_IBRS;
1250 }
1251 
1252 static void __init
1253 spectre_v2_user_select_mitigation(void)
1254 {
1255 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1256 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1257 	enum spectre_v2_user_cmd cmd;
1258 
1259 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1260 		return;
1261 
1262 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1263 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1264 		smt_possible = false;
1265 
1266 	cmd = spectre_v2_parse_user_cmdline();
1267 	switch (cmd) {
1268 	case SPECTRE_V2_USER_CMD_NONE:
1269 		goto set_mode;
1270 	case SPECTRE_V2_USER_CMD_FORCE:
1271 		mode = SPECTRE_V2_USER_STRICT;
1272 		break;
1273 	case SPECTRE_V2_USER_CMD_AUTO:
1274 	case SPECTRE_V2_USER_CMD_PRCTL:
1275 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1276 		mode = SPECTRE_V2_USER_PRCTL;
1277 		break;
1278 	case SPECTRE_V2_USER_CMD_SECCOMP:
1279 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1280 		if (IS_ENABLED(CONFIG_SECCOMP))
1281 			mode = SPECTRE_V2_USER_SECCOMP;
1282 		else
1283 			mode = SPECTRE_V2_USER_PRCTL;
1284 		break;
1285 	}
1286 
1287 	/* Initialize Indirect Branch Prediction Barrier */
1288 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1289 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1290 
1291 		spectre_v2_user_ibpb = mode;
1292 		switch (cmd) {
1293 		case SPECTRE_V2_USER_CMD_FORCE:
1294 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1295 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1296 			static_branch_enable(&switch_mm_always_ibpb);
1297 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1298 			break;
1299 		case SPECTRE_V2_USER_CMD_PRCTL:
1300 		case SPECTRE_V2_USER_CMD_AUTO:
1301 		case SPECTRE_V2_USER_CMD_SECCOMP:
1302 			static_branch_enable(&switch_mm_cond_ibpb);
1303 			break;
1304 		default:
1305 			break;
1306 		}
1307 
1308 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1309 			static_key_enabled(&switch_mm_always_ibpb) ?
1310 			"always-on" : "conditional");
1311 	}
1312 
1313 	/*
1314 	 * If no STIBP, Intel enhanced IBRS is enabled, or SMT impossible, STIBP
1315 	 * is not required.
1316 	 *
1317 	 * Intel's Enhanced IBRS also protects against cross-thread branch target
1318 	 * injection in user-mode as the IBRS bit remains always set which
1319 	 * implicitly enables cross-thread protections.  However, in legacy IBRS
1320 	 * mode, the IBRS bit is set only on kernel entry and cleared on return
1321 	 * to userspace.  AMD Automatic IBRS also does not protect userspace.
1322 	 * These modes therefore disable the implicit cross-thread protection,
1323 	 * so allow for STIBP to be selected in those cases.
1324 	 */
1325 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1326 	    !smt_possible ||
1327 	    (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
1328 	     !boot_cpu_has(X86_FEATURE_AUTOIBRS)))
1329 		return;
1330 
1331 	/*
1332 	 * At this point, an STIBP mode other than "off" has been set.
1333 	 * If STIBP support is not being forced, check if STIBP always-on
1334 	 * is preferred.
1335 	 */
1336 	if (mode != SPECTRE_V2_USER_STRICT &&
1337 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1338 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1339 
1340 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1341 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1342 		if (mode != SPECTRE_V2_USER_STRICT &&
1343 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1344 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1345 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1346 	}
1347 
1348 	spectre_v2_user_stibp = mode;
1349 
1350 set_mode:
1351 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1352 }
1353 
1354 static const char * const spectre_v2_strings[] = {
1355 	[SPECTRE_V2_NONE]			= "Vulnerable",
1356 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1357 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1358 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced / Automatic IBRS",
1359 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced / Automatic IBRS + LFENCE",
1360 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced / Automatic IBRS + Retpolines",
1361 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1362 };
1363 
1364 static const struct {
1365 	const char *option;
1366 	enum spectre_v2_mitigation_cmd cmd;
1367 	bool secure;
1368 } mitigation_options[] __initconst = {
1369 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1370 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1371 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1372 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1373 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1374 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1375 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1376 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1377 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1378 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1379 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1380 };
1381 
1382 static void __init spec_v2_print_cond(const char *reason, bool secure)
1383 {
1384 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1385 		pr_info("%s selected on command line.\n", reason);
1386 }
1387 
1388 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1389 {
1390 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1391 	char arg[20];
1392 	int ret, i;
1393 
1394 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1395 	    cpu_mitigations_off())
1396 		return SPECTRE_V2_CMD_NONE;
1397 
1398 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1399 	if (ret < 0)
1400 		return SPECTRE_V2_CMD_AUTO;
1401 
1402 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1403 		if (!match_option(arg, ret, mitigation_options[i].option))
1404 			continue;
1405 		cmd = mitigation_options[i].cmd;
1406 		break;
1407 	}
1408 
1409 	if (i >= ARRAY_SIZE(mitigation_options)) {
1410 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1411 		return SPECTRE_V2_CMD_AUTO;
1412 	}
1413 
1414 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1415 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1416 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1417 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1418 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1419 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1420 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1421 		       mitigation_options[i].option);
1422 		return SPECTRE_V2_CMD_AUTO;
1423 	}
1424 
1425 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1426 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1427 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1428 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1429 		pr_err("%s selected but CPU doesn't have Enhanced or Automatic IBRS. Switching to AUTO select\n",
1430 		       mitigation_options[i].option);
1431 		return SPECTRE_V2_CMD_AUTO;
1432 	}
1433 
1434 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1435 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1436 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1437 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1438 		       mitigation_options[i].option);
1439 		return SPECTRE_V2_CMD_AUTO;
1440 	}
1441 
1442 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1443 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1444 		       mitigation_options[i].option);
1445 		return SPECTRE_V2_CMD_AUTO;
1446 	}
1447 
1448 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1449 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1450 		       mitigation_options[i].option);
1451 		return SPECTRE_V2_CMD_AUTO;
1452 	}
1453 
1454 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1455 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1456 		       mitigation_options[i].option);
1457 		return SPECTRE_V2_CMD_AUTO;
1458 	}
1459 
1460 	if (cmd == SPECTRE_V2_CMD_IBRS && cpu_feature_enabled(X86_FEATURE_XENPV)) {
1461 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1462 		       mitigation_options[i].option);
1463 		return SPECTRE_V2_CMD_AUTO;
1464 	}
1465 
1466 	spec_v2_print_cond(mitigation_options[i].option,
1467 			   mitigation_options[i].secure);
1468 	return cmd;
1469 }
1470 
1471 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1472 {
1473 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1474 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1475 		return SPECTRE_V2_NONE;
1476 	}
1477 
1478 	return SPECTRE_V2_RETPOLINE;
1479 }
1480 
1481 /* Disable in-kernel use of non-RSB RET predictors */
1482 static void __init spec_ctrl_disable_kernel_rrsba(void)
1483 {
1484 	u64 ia32_cap;
1485 
1486 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1487 		return;
1488 
1489 	ia32_cap = x86_read_arch_cap_msr();
1490 
1491 	if (ia32_cap & ARCH_CAP_RRSBA) {
1492 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1493 		update_spec_ctrl(x86_spec_ctrl_base);
1494 	}
1495 }
1496 
1497 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1498 {
1499 	/*
1500 	 * Similar to context switches, there are two types of RSB attacks
1501 	 * after VM exit:
1502 	 *
1503 	 * 1) RSB underflow
1504 	 *
1505 	 * 2) Poisoned RSB entry
1506 	 *
1507 	 * When retpoline is enabled, both are mitigated by filling/clearing
1508 	 * the RSB.
1509 	 *
1510 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1511 	 * prediction isolation protections, RSB still needs to be cleared
1512 	 * because of #2.  Note that SMEP provides no protection here, unlike
1513 	 * user-space-poisoned RSB entries.
1514 	 *
1515 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1516 	 * bug is present then a LITE version of RSB protection is required,
1517 	 * just a single call needs to retire before a RET is executed.
1518 	 */
1519 	switch (mode) {
1520 	case SPECTRE_V2_NONE:
1521 		return;
1522 
1523 	case SPECTRE_V2_EIBRS_LFENCE:
1524 	case SPECTRE_V2_EIBRS:
1525 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1526 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1527 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1528 		}
1529 		return;
1530 
1531 	case SPECTRE_V2_EIBRS_RETPOLINE:
1532 	case SPECTRE_V2_RETPOLINE:
1533 	case SPECTRE_V2_LFENCE:
1534 	case SPECTRE_V2_IBRS:
1535 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1536 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1537 		return;
1538 	}
1539 
1540 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1541 	dump_stack();
1542 }
1543 
1544 static void __init spectre_v2_select_mitigation(void)
1545 {
1546 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1547 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1548 
1549 	/*
1550 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1551 	 * then nothing to do.
1552 	 */
1553 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1554 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1555 		return;
1556 
1557 	switch (cmd) {
1558 	case SPECTRE_V2_CMD_NONE:
1559 		return;
1560 
1561 	case SPECTRE_V2_CMD_FORCE:
1562 	case SPECTRE_V2_CMD_AUTO:
1563 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1564 			mode = SPECTRE_V2_EIBRS;
1565 			break;
1566 		}
1567 
1568 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1569 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1570 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1571 		    retbleed_cmd != RETBLEED_CMD_STUFF &&
1572 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1573 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1574 			mode = SPECTRE_V2_IBRS;
1575 			break;
1576 		}
1577 
1578 		mode = spectre_v2_select_retpoline();
1579 		break;
1580 
1581 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1582 		pr_err(SPECTRE_V2_LFENCE_MSG);
1583 		mode = SPECTRE_V2_LFENCE;
1584 		break;
1585 
1586 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1587 		mode = SPECTRE_V2_RETPOLINE;
1588 		break;
1589 
1590 	case SPECTRE_V2_CMD_RETPOLINE:
1591 		mode = spectre_v2_select_retpoline();
1592 		break;
1593 
1594 	case SPECTRE_V2_CMD_IBRS:
1595 		mode = SPECTRE_V2_IBRS;
1596 		break;
1597 
1598 	case SPECTRE_V2_CMD_EIBRS:
1599 		mode = SPECTRE_V2_EIBRS;
1600 		break;
1601 
1602 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1603 		mode = SPECTRE_V2_EIBRS_LFENCE;
1604 		break;
1605 
1606 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1607 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1608 		break;
1609 	}
1610 
1611 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1612 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1613 
1614 	if (spectre_v2_in_ibrs_mode(mode)) {
1615 		if (boot_cpu_has(X86_FEATURE_AUTOIBRS)) {
1616 			msr_set_bit(MSR_EFER, _EFER_AUTOIBRS);
1617 		} else {
1618 			x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1619 			update_spec_ctrl(x86_spec_ctrl_base);
1620 		}
1621 	}
1622 
1623 	switch (mode) {
1624 	case SPECTRE_V2_NONE:
1625 	case SPECTRE_V2_EIBRS:
1626 		break;
1627 
1628 	case SPECTRE_V2_IBRS:
1629 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1630 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1631 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1632 		break;
1633 
1634 	case SPECTRE_V2_LFENCE:
1635 	case SPECTRE_V2_EIBRS_LFENCE:
1636 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1637 		fallthrough;
1638 
1639 	case SPECTRE_V2_RETPOLINE:
1640 	case SPECTRE_V2_EIBRS_RETPOLINE:
1641 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1642 		break;
1643 	}
1644 
1645 	/*
1646 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1647 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1648 	 * prediction from a non-RSB predictor is still a risk.
1649 	 */
1650 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1651 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1652 	    mode == SPECTRE_V2_RETPOLINE)
1653 		spec_ctrl_disable_kernel_rrsba();
1654 
1655 	spectre_v2_enabled = mode;
1656 	pr_info("%s\n", spectre_v2_strings[mode]);
1657 
1658 	/*
1659 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1660 	 * context switch.  In general there are two types of RSB attacks
1661 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1662 	 *
1663 	 * 1) RSB underflow
1664 	 *
1665 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1666 	 *    speculated return targets may come from the branch predictor,
1667 	 *    which could have a user-poisoned BTB or BHB entry.
1668 	 *
1669 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1670 	 *    regardless of the state of the RSB.
1671 	 *
1672 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1673 	 *    scenario is mitigated by the IBRS branch prediction isolation
1674 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1675 	 *    protect against this type of attack.
1676 	 *
1677 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1678 	 *
1679 	 * 2) Poisoned RSB entry
1680 	 *
1681 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1682 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1683 	 *    entry.
1684 	 *
1685 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1686 	 *    eIBRS.
1687 	 *
1688 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1689 	 *    RSB clearing.
1690 	 *
1691 	 * So to mitigate all cases, unconditionally fill RSB on context
1692 	 * switches.
1693 	 *
1694 	 * FIXME: Is this pointless for retbleed-affected AMD?
1695 	 */
1696 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1697 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1698 
1699 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1700 
1701 	/*
1702 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1703 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1704 	 * firmware calls only when IBRS / Enhanced / Automatic IBRS aren't
1705 	 * otherwise enabled.
1706 	 *
1707 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1708 	 * the user might select retpoline on the kernel command line and if
1709 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1710 	 * enable IBRS around firmware calls.
1711 	 */
1712 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1713 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1714 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1715 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1716 
1717 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1718 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1719 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1720 		}
1721 
1722 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1723 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1724 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1725 	}
1726 
1727 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1728 	spectre_v2_cmd = cmd;
1729 }
1730 
1731 static void update_stibp_msr(void * __unused)
1732 {
1733 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1734 	update_spec_ctrl(val);
1735 }
1736 
1737 /* Update x86_spec_ctrl_base in case SMT state changed. */
1738 static void update_stibp_strict(void)
1739 {
1740 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1741 
1742 	if (sched_smt_active())
1743 		mask |= SPEC_CTRL_STIBP;
1744 
1745 	if (mask == x86_spec_ctrl_base)
1746 		return;
1747 
1748 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1749 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1750 	x86_spec_ctrl_base = mask;
1751 	on_each_cpu(update_stibp_msr, NULL, 1);
1752 }
1753 
1754 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1755 static void update_indir_branch_cond(void)
1756 {
1757 	if (sched_smt_active())
1758 		static_branch_enable(&switch_to_cond_stibp);
1759 	else
1760 		static_branch_disable(&switch_to_cond_stibp);
1761 }
1762 
1763 #undef pr_fmt
1764 #define pr_fmt(fmt) fmt
1765 
1766 /* Update the static key controlling the MDS CPU buffer clear in idle */
1767 static void update_mds_branch_idle(void)
1768 {
1769 	u64 ia32_cap = x86_read_arch_cap_msr();
1770 
1771 	/*
1772 	 * Enable the idle clearing if SMT is active on CPUs which are
1773 	 * affected only by MSBDS and not any other MDS variant.
1774 	 *
1775 	 * The other variants cannot be mitigated when SMT is enabled, so
1776 	 * clearing the buffers on idle just to prevent the Store Buffer
1777 	 * repartitioning leak would be a window dressing exercise.
1778 	 */
1779 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1780 		return;
1781 
1782 	if (sched_smt_active()) {
1783 		static_branch_enable(&mds_idle_clear);
1784 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1785 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1786 		static_branch_disable(&mds_idle_clear);
1787 	}
1788 }
1789 
1790 #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"
1791 #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"
1792 #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"
1793 
1794 void cpu_bugs_smt_update(void)
1795 {
1796 	mutex_lock(&spec_ctrl_mutex);
1797 
1798 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1799 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1800 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1801 
1802 	switch (spectre_v2_user_stibp) {
1803 	case SPECTRE_V2_USER_NONE:
1804 		break;
1805 	case SPECTRE_V2_USER_STRICT:
1806 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1807 		update_stibp_strict();
1808 		break;
1809 	case SPECTRE_V2_USER_PRCTL:
1810 	case SPECTRE_V2_USER_SECCOMP:
1811 		update_indir_branch_cond();
1812 		break;
1813 	}
1814 
1815 	switch (mds_mitigation) {
1816 	case MDS_MITIGATION_FULL:
1817 	case MDS_MITIGATION_VMWERV:
1818 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1819 			pr_warn_once(MDS_MSG_SMT);
1820 		update_mds_branch_idle();
1821 		break;
1822 	case MDS_MITIGATION_OFF:
1823 		break;
1824 	}
1825 
1826 	switch (taa_mitigation) {
1827 	case TAA_MITIGATION_VERW:
1828 	case TAA_MITIGATION_UCODE_NEEDED:
1829 		if (sched_smt_active())
1830 			pr_warn_once(TAA_MSG_SMT);
1831 		break;
1832 	case TAA_MITIGATION_TSX_DISABLED:
1833 	case TAA_MITIGATION_OFF:
1834 		break;
1835 	}
1836 
1837 	switch (mmio_mitigation) {
1838 	case MMIO_MITIGATION_VERW:
1839 	case MMIO_MITIGATION_UCODE_NEEDED:
1840 		if (sched_smt_active())
1841 			pr_warn_once(MMIO_MSG_SMT);
1842 		break;
1843 	case MMIO_MITIGATION_OFF:
1844 		break;
1845 	}
1846 
1847 	mutex_unlock(&spec_ctrl_mutex);
1848 }
1849 
1850 #undef pr_fmt
1851 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1852 
1853 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1854 
1855 /* The kernel command line selection */
1856 enum ssb_mitigation_cmd {
1857 	SPEC_STORE_BYPASS_CMD_NONE,
1858 	SPEC_STORE_BYPASS_CMD_AUTO,
1859 	SPEC_STORE_BYPASS_CMD_ON,
1860 	SPEC_STORE_BYPASS_CMD_PRCTL,
1861 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1862 };
1863 
1864 static const char * const ssb_strings[] = {
1865 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1866 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1867 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1868 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1869 };
1870 
1871 static const struct {
1872 	const char *option;
1873 	enum ssb_mitigation_cmd cmd;
1874 } ssb_mitigation_options[]  __initconst = {
1875 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1876 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1877 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1878 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1879 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1880 };
1881 
1882 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1883 {
1884 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1885 	char arg[20];
1886 	int ret, i;
1887 
1888 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1889 	    cpu_mitigations_off()) {
1890 		return SPEC_STORE_BYPASS_CMD_NONE;
1891 	} else {
1892 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1893 					  arg, sizeof(arg));
1894 		if (ret < 0)
1895 			return SPEC_STORE_BYPASS_CMD_AUTO;
1896 
1897 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1898 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1899 				continue;
1900 
1901 			cmd = ssb_mitigation_options[i].cmd;
1902 			break;
1903 		}
1904 
1905 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1906 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1907 			return SPEC_STORE_BYPASS_CMD_AUTO;
1908 		}
1909 	}
1910 
1911 	return cmd;
1912 }
1913 
1914 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1915 {
1916 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1917 	enum ssb_mitigation_cmd cmd;
1918 
1919 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1920 		return mode;
1921 
1922 	cmd = ssb_parse_cmdline();
1923 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1924 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1925 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1926 		return mode;
1927 
1928 	switch (cmd) {
1929 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1930 		/*
1931 		 * Choose prctl+seccomp as the default mode if seccomp is
1932 		 * enabled.
1933 		 */
1934 		if (IS_ENABLED(CONFIG_SECCOMP))
1935 			mode = SPEC_STORE_BYPASS_SECCOMP;
1936 		else
1937 			mode = SPEC_STORE_BYPASS_PRCTL;
1938 		break;
1939 	case SPEC_STORE_BYPASS_CMD_ON:
1940 		mode = SPEC_STORE_BYPASS_DISABLE;
1941 		break;
1942 	case SPEC_STORE_BYPASS_CMD_AUTO:
1943 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1944 		mode = SPEC_STORE_BYPASS_PRCTL;
1945 		break;
1946 	case SPEC_STORE_BYPASS_CMD_NONE:
1947 		break;
1948 	}
1949 
1950 	/*
1951 	 * We have three CPU feature flags that are in play here:
1952 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1953 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1954 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1955 	 */
1956 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1957 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1958 		/*
1959 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1960 		 * use a completely different MSR and bit dependent on family.
1961 		 */
1962 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1963 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1964 			x86_amd_ssb_disable();
1965 		} else {
1966 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1967 			update_spec_ctrl(x86_spec_ctrl_base);
1968 		}
1969 	}
1970 
1971 	return mode;
1972 }
1973 
1974 static void ssb_select_mitigation(void)
1975 {
1976 	ssb_mode = __ssb_select_mitigation();
1977 
1978 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1979 		pr_info("%s\n", ssb_strings[ssb_mode]);
1980 }
1981 
1982 #undef pr_fmt
1983 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1984 
1985 static void task_update_spec_tif(struct task_struct *tsk)
1986 {
1987 	/* Force the update of the real TIF bits */
1988 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1989 
1990 	/*
1991 	 * Immediately update the speculation control MSRs for the current
1992 	 * task, but for a non-current task delay setting the CPU
1993 	 * mitigation until it is scheduled next.
1994 	 *
1995 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1996 	 * always the current task.
1997 	 */
1998 	if (tsk == current)
1999 		speculation_ctrl_update_current();
2000 }
2001 
2002 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
2003 {
2004 
2005 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2006 		return -EPERM;
2007 
2008 	switch (ctrl) {
2009 	case PR_SPEC_ENABLE:
2010 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
2011 		return 0;
2012 	case PR_SPEC_DISABLE:
2013 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
2014 		return 0;
2015 	default:
2016 		return -ERANGE;
2017 	}
2018 }
2019 
2020 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
2021 {
2022 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
2023 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
2024 		return -ENXIO;
2025 
2026 	switch (ctrl) {
2027 	case PR_SPEC_ENABLE:
2028 		/* If speculation is force disabled, enable is not allowed */
2029 		if (task_spec_ssb_force_disable(task))
2030 			return -EPERM;
2031 		task_clear_spec_ssb_disable(task);
2032 		task_clear_spec_ssb_noexec(task);
2033 		task_update_spec_tif(task);
2034 		break;
2035 	case PR_SPEC_DISABLE:
2036 		task_set_spec_ssb_disable(task);
2037 		task_clear_spec_ssb_noexec(task);
2038 		task_update_spec_tif(task);
2039 		break;
2040 	case PR_SPEC_FORCE_DISABLE:
2041 		task_set_spec_ssb_disable(task);
2042 		task_set_spec_ssb_force_disable(task);
2043 		task_clear_spec_ssb_noexec(task);
2044 		task_update_spec_tif(task);
2045 		break;
2046 	case PR_SPEC_DISABLE_NOEXEC:
2047 		if (task_spec_ssb_force_disable(task))
2048 			return -EPERM;
2049 		task_set_spec_ssb_disable(task);
2050 		task_set_spec_ssb_noexec(task);
2051 		task_update_spec_tif(task);
2052 		break;
2053 	default:
2054 		return -ERANGE;
2055 	}
2056 	return 0;
2057 }
2058 
2059 static bool is_spec_ib_user_controlled(void)
2060 {
2061 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
2062 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2063 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
2064 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
2065 }
2066 
2067 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
2068 {
2069 	switch (ctrl) {
2070 	case PR_SPEC_ENABLE:
2071 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2072 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2073 			return 0;
2074 
2075 		/*
2076 		 * With strict mode for both IBPB and STIBP, the instruction
2077 		 * code paths avoid checking this task flag and instead,
2078 		 * unconditionally run the instruction. However, STIBP and IBPB
2079 		 * are independent and either can be set to conditionally
2080 		 * enabled regardless of the mode of the other.
2081 		 *
2082 		 * If either is set to conditional, allow the task flag to be
2083 		 * updated, unless it was force-disabled by a previous prctl
2084 		 * call. Currently, this is possible on an AMD CPU which has the
2085 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
2086 		 * kernel is booted with 'spectre_v2_user=seccomp', then
2087 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
2088 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
2089 		 */
2090 		if (!is_spec_ib_user_controlled() ||
2091 		    task_spec_ib_force_disable(task))
2092 			return -EPERM;
2093 
2094 		task_clear_spec_ib_disable(task);
2095 		task_update_spec_tif(task);
2096 		break;
2097 	case PR_SPEC_DISABLE:
2098 	case PR_SPEC_FORCE_DISABLE:
2099 		/*
2100 		 * Indirect branch speculation is always allowed when
2101 		 * mitigation is force disabled.
2102 		 */
2103 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2104 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2105 			return -EPERM;
2106 
2107 		if (!is_spec_ib_user_controlled())
2108 			return 0;
2109 
2110 		task_set_spec_ib_disable(task);
2111 		if (ctrl == PR_SPEC_FORCE_DISABLE)
2112 			task_set_spec_ib_force_disable(task);
2113 		task_update_spec_tif(task);
2114 		if (task == current)
2115 			indirect_branch_prediction_barrier();
2116 		break;
2117 	default:
2118 		return -ERANGE;
2119 	}
2120 	return 0;
2121 }
2122 
2123 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
2124 			     unsigned long ctrl)
2125 {
2126 	switch (which) {
2127 	case PR_SPEC_STORE_BYPASS:
2128 		return ssb_prctl_set(task, ctrl);
2129 	case PR_SPEC_INDIRECT_BRANCH:
2130 		return ib_prctl_set(task, ctrl);
2131 	case PR_SPEC_L1D_FLUSH:
2132 		return l1d_flush_prctl_set(task, ctrl);
2133 	default:
2134 		return -ENODEV;
2135 	}
2136 }
2137 
2138 #ifdef CONFIG_SECCOMP
2139 void arch_seccomp_spec_mitigate(struct task_struct *task)
2140 {
2141 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
2142 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2143 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2144 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
2145 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2146 }
2147 #endif
2148 
2149 static int l1d_flush_prctl_get(struct task_struct *task)
2150 {
2151 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2152 		return PR_SPEC_FORCE_DISABLE;
2153 
2154 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
2155 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2156 	else
2157 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2158 }
2159 
2160 static int ssb_prctl_get(struct task_struct *task)
2161 {
2162 	switch (ssb_mode) {
2163 	case SPEC_STORE_BYPASS_DISABLE:
2164 		return PR_SPEC_DISABLE;
2165 	case SPEC_STORE_BYPASS_SECCOMP:
2166 	case SPEC_STORE_BYPASS_PRCTL:
2167 		if (task_spec_ssb_force_disable(task))
2168 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2169 		if (task_spec_ssb_noexec(task))
2170 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2171 		if (task_spec_ssb_disable(task))
2172 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2173 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2174 	default:
2175 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2176 			return PR_SPEC_ENABLE;
2177 		return PR_SPEC_NOT_AFFECTED;
2178 	}
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 (!boot_cpu_has_bug(X86_BUG_SRSO) || cpu_mitigations_off())
2414 		goto pred_cmd;
2415 
2416 	if (has_microcode) {
2417 		/*
2418 		 * Zen1/2 with SMT off aren't vulnerable after the right
2419 		 * IBPB microcode has been applied.
2420 		 */
2421 		if (boot_cpu_data.x86 < 0x19 && !cpu_smt_possible()) {
2422 			setup_force_cpu_cap(X86_FEATURE_SRSO_NO);
2423 			return;
2424 		}
2425 
2426 		if (retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2427 			srso_mitigation = SRSO_MITIGATION_IBPB;
2428 			goto out;
2429 		}
2430 	} else {
2431 		pr_warn("IBPB-extending microcode not applied!\n");
2432 		pr_warn(SRSO_NOTICE);
2433 
2434 		/* may be overwritten by SRSO_CMD_SAFE_RET below */
2435 		srso_mitigation = SRSO_MITIGATION_UCODE_NEEDED;
2436 	}
2437 
2438 	switch (srso_cmd) {
2439 	case SRSO_CMD_OFF:
2440 		goto pred_cmd;
2441 
2442 	case SRSO_CMD_MICROCODE:
2443 		if (has_microcode) {
2444 			srso_mitigation = SRSO_MITIGATION_MICROCODE;
2445 			pr_warn(SRSO_NOTICE);
2446 		}
2447 		break;
2448 
2449 	case SRSO_CMD_SAFE_RET:
2450 		if (IS_ENABLED(CONFIG_CPU_SRSO)) {
2451 			/*
2452 			 * Enable the return thunk for generated code
2453 			 * like ftrace, static_call, etc.
2454 			 */
2455 			setup_force_cpu_cap(X86_FEATURE_RETHUNK);
2456 			setup_force_cpu_cap(X86_FEATURE_UNRET);
2457 
2458 			if (boot_cpu_data.x86 == 0x19) {
2459 				setup_force_cpu_cap(X86_FEATURE_SRSO_ALIAS);
2460 				x86_return_thunk = srso_alias_return_thunk;
2461 			} else {
2462 				setup_force_cpu_cap(X86_FEATURE_SRSO);
2463 				x86_return_thunk = srso_return_thunk;
2464 			}
2465 			if (has_microcode)
2466 				srso_mitigation = SRSO_MITIGATION_SAFE_RET;
2467 			else
2468 				srso_mitigation = SRSO_MITIGATION_SAFE_RET_UCODE_NEEDED;
2469 		} else {
2470 			pr_err("WARNING: kernel not compiled with CPU_SRSO.\n");
2471 			goto pred_cmd;
2472 		}
2473 		break;
2474 
2475 	case SRSO_CMD_IBPB:
2476 		if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
2477 			if (has_microcode) {
2478 				setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
2479 				srso_mitigation = SRSO_MITIGATION_IBPB;
2480 			}
2481 		} else {
2482 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
2483 			goto pred_cmd;
2484 		}
2485 		break;
2486 
2487 	case SRSO_CMD_IBPB_ON_VMEXIT:
2488 		if (IS_ENABLED(CONFIG_CPU_SRSO)) {
2489 			if (!boot_cpu_has(X86_FEATURE_ENTRY_IBPB) && has_microcode) {
2490 				setup_force_cpu_cap(X86_FEATURE_IBPB_ON_VMEXIT);
2491 				srso_mitigation = SRSO_MITIGATION_IBPB_ON_VMEXIT;
2492 			}
2493 		} else {
2494 			pr_err("WARNING: kernel not compiled with CPU_SRSO.\n");
2495 			goto pred_cmd;
2496                 }
2497 		break;
2498 
2499 	default:
2500 		break;
2501 	}
2502 
2503 out:
2504 	pr_info("%s\n", srso_strings[srso_mitigation]);
2505 
2506 pred_cmd:
2507 	if ((!boot_cpu_has_bug(X86_BUG_SRSO) || srso_cmd == SRSO_CMD_OFF) &&
2508 	     boot_cpu_has(X86_FEATURE_SBPB))
2509 		x86_pred_cmd = PRED_CMD_SBPB;
2510 }
2511 
2512 #undef pr_fmt
2513 #define pr_fmt(fmt) fmt
2514 
2515 #ifdef CONFIG_SYSFS
2516 
2517 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2518 
2519 #if IS_ENABLED(CONFIG_KVM_INTEL)
2520 static const char * const l1tf_vmx_states[] = {
2521 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2522 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2523 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2524 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2525 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2526 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2527 };
2528 
2529 static ssize_t l1tf_show_state(char *buf)
2530 {
2531 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2532 		return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2533 
2534 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2535 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2536 	     sched_smt_active())) {
2537 		return sysfs_emit(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2538 				  l1tf_vmx_states[l1tf_vmx_mitigation]);
2539 	}
2540 
2541 	return sysfs_emit(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2542 			  l1tf_vmx_states[l1tf_vmx_mitigation],
2543 			  sched_smt_active() ? "vulnerable" : "disabled");
2544 }
2545 
2546 static ssize_t itlb_multihit_show_state(char *buf)
2547 {
2548 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2549 	    !boot_cpu_has(X86_FEATURE_VMX))
2550 		return sysfs_emit(buf, "KVM: Mitigation: VMX unsupported\n");
2551 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2552 		return sysfs_emit(buf, "KVM: Mitigation: VMX disabled\n");
2553 	else if (itlb_multihit_kvm_mitigation)
2554 		return sysfs_emit(buf, "KVM: Mitigation: Split huge pages\n");
2555 	else
2556 		return sysfs_emit(buf, "KVM: Vulnerable\n");
2557 }
2558 #else
2559 static ssize_t l1tf_show_state(char *buf)
2560 {
2561 	return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2562 }
2563 
2564 static ssize_t itlb_multihit_show_state(char *buf)
2565 {
2566 	return sysfs_emit(buf, "Processor vulnerable\n");
2567 }
2568 #endif
2569 
2570 static ssize_t mds_show_state(char *buf)
2571 {
2572 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2573 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2574 				  mds_strings[mds_mitigation]);
2575 	}
2576 
2577 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2578 		return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2579 				  (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2580 				   sched_smt_active() ? "mitigated" : "disabled"));
2581 	}
2582 
2583 	return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2584 			  sched_smt_active() ? "vulnerable" : "disabled");
2585 }
2586 
2587 static ssize_t tsx_async_abort_show_state(char *buf)
2588 {
2589 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2590 	    (taa_mitigation == TAA_MITIGATION_OFF))
2591 		return sysfs_emit(buf, "%s\n", taa_strings[taa_mitigation]);
2592 
2593 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2594 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2595 				  taa_strings[taa_mitigation]);
2596 	}
2597 
2598 	return sysfs_emit(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2599 			  sched_smt_active() ? "vulnerable" : "disabled");
2600 }
2601 
2602 static ssize_t mmio_stale_data_show_state(char *buf)
2603 {
2604 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2605 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2606 
2607 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2608 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2609 
2610 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2611 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2612 				  mmio_strings[mmio_mitigation]);
2613 	}
2614 
2615 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2616 			  sched_smt_active() ? "vulnerable" : "disabled");
2617 }
2618 
2619 static char *stibp_state(void)
2620 {
2621 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled) &&
2622 	    !boot_cpu_has(X86_FEATURE_AUTOIBRS))
2623 		return "";
2624 
2625 	switch (spectre_v2_user_stibp) {
2626 	case SPECTRE_V2_USER_NONE:
2627 		return ", STIBP: disabled";
2628 	case SPECTRE_V2_USER_STRICT:
2629 		return ", STIBP: forced";
2630 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2631 		return ", STIBP: always-on";
2632 	case SPECTRE_V2_USER_PRCTL:
2633 	case SPECTRE_V2_USER_SECCOMP:
2634 		if (static_key_enabled(&switch_to_cond_stibp))
2635 			return ", STIBP: conditional";
2636 	}
2637 	return "";
2638 }
2639 
2640 static char *ibpb_state(void)
2641 {
2642 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2643 		if (static_key_enabled(&switch_mm_always_ibpb))
2644 			return ", IBPB: always-on";
2645 		if (static_key_enabled(&switch_mm_cond_ibpb))
2646 			return ", IBPB: conditional";
2647 		return ", IBPB: disabled";
2648 	}
2649 	return "";
2650 }
2651 
2652 static char *pbrsb_eibrs_state(void)
2653 {
2654 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2655 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2656 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2657 			return ", PBRSB-eIBRS: SW sequence";
2658 		else
2659 			return ", PBRSB-eIBRS: Vulnerable";
2660 	} else {
2661 		return ", PBRSB-eIBRS: Not affected";
2662 	}
2663 }
2664 
2665 static ssize_t spectre_v2_show_state(char *buf)
2666 {
2667 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2668 		return sysfs_emit(buf, "Vulnerable: LFENCE\n");
2669 
2670 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2671 		return sysfs_emit(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2672 
2673 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2674 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2675 		return sysfs_emit(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2676 
2677 	return sysfs_emit(buf, "%s%s%s%s%s%s%s\n",
2678 			  spectre_v2_strings[spectre_v2_enabled],
2679 			  ibpb_state(),
2680 			  boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2681 			  stibp_state(),
2682 			  boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2683 			  pbrsb_eibrs_state(),
2684 			  spectre_v2_module_string());
2685 }
2686 
2687 static ssize_t srbds_show_state(char *buf)
2688 {
2689 	return sysfs_emit(buf, "%s\n", srbds_strings[srbds_mitigation]);
2690 }
2691 
2692 static ssize_t retbleed_show_state(char *buf)
2693 {
2694 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2695 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2696 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2697 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2698 			return sysfs_emit(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2699 
2700 		return sysfs_emit(buf, "%s; SMT %s\n", retbleed_strings[retbleed_mitigation],
2701 				  !sched_smt_active() ? "disabled" :
2702 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2703 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2704 				  "enabled with STIBP protection" : "vulnerable");
2705 	}
2706 
2707 	return sysfs_emit(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2708 }
2709 
2710 static ssize_t srso_show_state(char *buf)
2711 {
2712 	if (boot_cpu_has(X86_FEATURE_SRSO_NO))
2713 		return sysfs_emit(buf, "Mitigation: SMT disabled\n");
2714 
2715 	return sysfs_emit(buf, "%s\n", srso_strings[srso_mitigation]);
2716 }
2717 
2718 static ssize_t gds_show_state(char *buf)
2719 {
2720 	return sysfs_emit(buf, "%s\n", gds_strings[gds_mitigation]);
2721 }
2722 
2723 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2724 			       char *buf, unsigned int bug)
2725 {
2726 	if (!boot_cpu_has_bug(bug))
2727 		return sysfs_emit(buf, "Not affected\n");
2728 
2729 	switch (bug) {
2730 	case X86_BUG_CPU_MELTDOWN:
2731 		if (boot_cpu_has(X86_FEATURE_PTI))
2732 			return sysfs_emit(buf, "Mitigation: PTI\n");
2733 
2734 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2735 			return sysfs_emit(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2736 
2737 		break;
2738 
2739 	case X86_BUG_SPECTRE_V1:
2740 		return sysfs_emit(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2741 
2742 	case X86_BUG_SPECTRE_V2:
2743 		return spectre_v2_show_state(buf);
2744 
2745 	case X86_BUG_SPEC_STORE_BYPASS:
2746 		return sysfs_emit(buf, "%s\n", ssb_strings[ssb_mode]);
2747 
2748 	case X86_BUG_L1TF:
2749 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2750 			return l1tf_show_state(buf);
2751 		break;
2752 
2753 	case X86_BUG_MDS:
2754 		return mds_show_state(buf);
2755 
2756 	case X86_BUG_TAA:
2757 		return tsx_async_abort_show_state(buf);
2758 
2759 	case X86_BUG_ITLB_MULTIHIT:
2760 		return itlb_multihit_show_state(buf);
2761 
2762 	case X86_BUG_SRBDS:
2763 		return srbds_show_state(buf);
2764 
2765 	case X86_BUG_MMIO_STALE_DATA:
2766 	case X86_BUG_MMIO_UNKNOWN:
2767 		return mmio_stale_data_show_state(buf);
2768 
2769 	case X86_BUG_RETBLEED:
2770 		return retbleed_show_state(buf);
2771 
2772 	case X86_BUG_SRSO:
2773 		return srso_show_state(buf);
2774 
2775 	case X86_BUG_GDS:
2776 		return gds_show_state(buf);
2777 
2778 	default:
2779 		break;
2780 	}
2781 
2782 	return sysfs_emit(buf, "Vulnerable\n");
2783 }
2784 
2785 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2786 {
2787 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2788 }
2789 
2790 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2791 {
2792 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2793 }
2794 
2795 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2796 {
2797 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2798 }
2799 
2800 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2801 {
2802 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2803 }
2804 
2805 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2806 {
2807 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2808 }
2809 
2810 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2811 {
2812 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2813 }
2814 
2815 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2816 {
2817 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2818 }
2819 
2820 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2821 {
2822 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2823 }
2824 
2825 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2826 {
2827 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2828 }
2829 
2830 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2831 {
2832 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2833 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2834 	else
2835 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2836 }
2837 
2838 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2839 {
2840 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2841 }
2842 
2843 ssize_t cpu_show_spec_rstack_overflow(struct device *dev, struct device_attribute *attr, char *buf)
2844 {
2845 	return cpu_show_common(dev, attr, buf, X86_BUG_SRSO);
2846 }
2847 
2848 ssize_t cpu_show_gds(struct device *dev, struct device_attribute *attr, char *buf)
2849 {
2850 	return cpu_show_common(dev, attr, buf, X86_BUG_GDS);
2851 }
2852 #endif
2853