xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision c09cc6e5)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Copyright (C) 1994  Linus Torvalds
4  *
5  *  Cyrix stuff, June 1998 by:
6  *	- Rafael R. Reilova (moved everything from head.S),
7  *        <rreilova@ececs.uc.edu>
8  *	- Channing Corn (tests & fixes),
9  *	- Andrew D. Balsa (code cleanup).
10  */
11 #include <linux/init.h>
12 #include <linux/utsname.h>
13 #include <linux/cpu.h>
14 #include <linux/module.h>
15 #include <linux/nospec.h>
16 #include <linux/prctl.h>
17 #include <linux/sched/smt.h>
18 #include <linux/pgtable.h>
19 #include <linux/bpf.h>
20 
21 #include <asm/spec-ctrl.h>
22 #include <asm/cmdline.h>
23 #include <asm/bugs.h>
24 #include <asm/processor.h>
25 #include <asm/processor-flags.h>
26 #include <asm/fpu/api.h>
27 #include <asm/msr.h>
28 #include <asm/vmx.h>
29 #include <asm/paravirt.h>
30 #include <asm/alternative.h>
31 #include <asm/set_memory.h>
32 #include <asm/intel-family.h>
33 #include <asm/e820/api.h>
34 #include <asm/hypervisor.h>
35 #include <asm/tlbflush.h>
36 
37 #include "cpu.h"
38 
39 static void __init spectre_v1_select_mitigation(void);
40 static void __init spectre_v2_select_mitigation(void);
41 static void __init retbleed_select_mitigation(void);
42 static void __init spectre_v2_user_select_mitigation(void);
43 static void __init ssb_select_mitigation(void);
44 static void __init l1tf_select_mitigation(void);
45 static void __init mds_select_mitigation(void);
46 static void __init md_clear_update_mitigation(void);
47 static void __init md_clear_select_mitigation(void);
48 static void __init taa_select_mitigation(void);
49 static void __init mmio_select_mitigation(void);
50 static void __init srbds_select_mitigation(void);
51 static void __init l1d_flush_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 static DEFINE_MUTEX(spec_ctrl_mutex);
62 
63 /*
64  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
65  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
66  */
67 void write_spec_ctrl_current(u64 val, bool force)
68 {
69 	if (this_cpu_read(x86_spec_ctrl_current) == val)
70 		return;
71 
72 	this_cpu_write(x86_spec_ctrl_current, val);
73 
74 	/*
75 	 * When KERNEL_IBRS this MSR is written on return-to-user, unless
76 	 * forced the update can be delayed until that time.
77 	 */
78 	if (force || !cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
79 		wrmsrl(MSR_IA32_SPEC_CTRL, val);
80 }
81 
82 u64 spec_ctrl_current(void)
83 {
84 	return this_cpu_read(x86_spec_ctrl_current);
85 }
86 EXPORT_SYMBOL_GPL(spec_ctrl_current);
87 
88 /*
89  * AMD specific MSR info for Speculative Store Bypass control.
90  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
91  */
92 u64 __ro_after_init x86_amd_ls_cfg_base;
93 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
94 
95 /* Control conditional STIBP in switch_to() */
96 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
97 /* Control conditional IBPB in switch_mm() */
98 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
99 /* Control unconditional IBPB in switch_mm() */
100 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
101 
102 /* Control MDS CPU buffer clear before returning to user space */
103 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
104 EXPORT_SYMBOL_GPL(mds_user_clear);
105 /* Control MDS CPU buffer clear before idling (halt, mwait) */
106 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
107 EXPORT_SYMBOL_GPL(mds_idle_clear);
108 
109 /*
110  * Controls whether l1d flush based mitigations are enabled,
111  * based on hw features and admin setting via boot parameter
112  * defaults to false
113  */
114 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
115 
116 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
117 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
118 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
119 
120 void __init check_bugs(void)
121 {
122 	identify_boot_cpu();
123 
124 	/*
125 	 * identify_boot_cpu() initialized SMT support information, let the
126 	 * core code know.
127 	 */
128 	cpu_smt_check_topology();
129 
130 	if (!IS_ENABLED(CONFIG_SMP)) {
131 		pr_info("CPU: ");
132 		print_cpu_info(&boot_cpu_data);
133 	}
134 
135 	/*
136 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
137 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
138 	 * init code as it is not enumerated and depends on the family.
139 	 */
140 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
141 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
142 
143 	/* Select the proper CPU mitigations before patching alternatives: */
144 	spectre_v1_select_mitigation();
145 	spectre_v2_select_mitigation();
146 	/*
147 	 * retbleed_select_mitigation() relies on the state set by
148 	 * spectre_v2_select_mitigation(); specifically it wants to know about
149 	 * spectre_v2=ibrs.
150 	 */
151 	retbleed_select_mitigation();
152 	/*
153 	 * spectre_v2_user_select_mitigation() relies on the state set by
154 	 * retbleed_select_mitigation(); specifically the STIBP selection is
155 	 * forced for UNRET or IBPB.
156 	 */
157 	spectre_v2_user_select_mitigation();
158 	ssb_select_mitigation();
159 	l1tf_select_mitigation();
160 	md_clear_select_mitigation();
161 	srbds_select_mitigation();
162 	l1d_flush_select_mitigation();
163 
164 	arch_smt_update();
165 
166 #ifdef CONFIG_X86_32
167 	/*
168 	 * Check whether we are able to run this kernel safely on SMP.
169 	 *
170 	 * - i386 is no longer supported.
171 	 * - In order to run on anything without a TSC, we need to be
172 	 *   compiled for a i486.
173 	 */
174 	if (boot_cpu_data.x86 < 4)
175 		panic("Kernel requires i486+ for 'invlpg' and other features");
176 
177 	init_utsname()->machine[1] =
178 		'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
179 	alternative_instructions();
180 
181 	fpu__init_check_bugs();
182 #else /* CONFIG_X86_64 */
183 	alternative_instructions();
184 
185 	/*
186 	 * Make sure the first 2MB area is not mapped by huge pages
187 	 * There are typically fixed size MTRRs in there and overlapping
188 	 * MTRRs into large pages causes slow downs.
189 	 *
190 	 * Right now we don't do that with gbpages because there seems
191 	 * very little benefit for that case.
192 	 */
193 	if (!direct_gbpages)
194 		set_memory_4k((unsigned long)__va(0), 1);
195 #endif
196 }
197 
198 /*
199  * NOTE: This function is *only* called for SVM.  VMX spec_ctrl handling is
200  * done in vmenter.S.
201  */
202 void
203 x86_virt_spec_ctrl(u64 guest_spec_ctrl, u64 guest_virt_spec_ctrl, bool setguest)
204 {
205 	u64 msrval, guestval = guest_spec_ctrl, hostval = spec_ctrl_current();
206 	struct thread_info *ti = current_thread_info();
207 
208 	if (static_cpu_has(X86_FEATURE_MSR_SPEC_CTRL)) {
209 		if (hostval != guestval) {
210 			msrval = setguest ? guestval : hostval;
211 			wrmsrl(MSR_IA32_SPEC_CTRL, msrval);
212 		}
213 	}
214 
215 	/*
216 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
217 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
218 	 */
219 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
220 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
221 		return;
222 
223 	/*
224 	 * If the host has SSBD mitigation enabled, force it in the host's
225 	 * virtual MSR value. If its not permanently enabled, evaluate
226 	 * current's TIF_SSBD thread flag.
227 	 */
228 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
229 		hostval = SPEC_CTRL_SSBD;
230 	else
231 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
232 
233 	/* Sanitize the guest value */
234 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
235 
236 	if (hostval != guestval) {
237 		unsigned long tif;
238 
239 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
240 				 ssbd_spec_ctrl_to_tif(hostval);
241 
242 		speculation_ctrl_update(tif);
243 	}
244 }
245 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
246 
247 static void x86_amd_ssb_disable(void)
248 {
249 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
250 
251 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
252 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
253 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
254 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
255 }
256 
257 #undef pr_fmt
258 #define pr_fmt(fmt)	"MDS: " fmt
259 
260 /* Default mitigation for MDS-affected CPUs */
261 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
262 static bool mds_nosmt __ro_after_init = false;
263 
264 static const char * const mds_strings[] = {
265 	[MDS_MITIGATION_OFF]	= "Vulnerable",
266 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
267 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
268 };
269 
270 static void __init mds_select_mitigation(void)
271 {
272 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
273 		mds_mitigation = MDS_MITIGATION_OFF;
274 		return;
275 	}
276 
277 	if (mds_mitigation == MDS_MITIGATION_FULL) {
278 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
279 			mds_mitigation = MDS_MITIGATION_VMWERV;
280 
281 		static_branch_enable(&mds_user_clear);
282 
283 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
284 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
285 			cpu_smt_disable(false);
286 	}
287 }
288 
289 static int __init mds_cmdline(char *str)
290 {
291 	if (!boot_cpu_has_bug(X86_BUG_MDS))
292 		return 0;
293 
294 	if (!str)
295 		return -EINVAL;
296 
297 	if (!strcmp(str, "off"))
298 		mds_mitigation = MDS_MITIGATION_OFF;
299 	else if (!strcmp(str, "full"))
300 		mds_mitigation = MDS_MITIGATION_FULL;
301 	else if (!strcmp(str, "full,nosmt")) {
302 		mds_mitigation = MDS_MITIGATION_FULL;
303 		mds_nosmt = true;
304 	}
305 
306 	return 0;
307 }
308 early_param("mds", mds_cmdline);
309 
310 #undef pr_fmt
311 #define pr_fmt(fmt)	"TAA: " fmt
312 
313 enum taa_mitigations {
314 	TAA_MITIGATION_OFF,
315 	TAA_MITIGATION_UCODE_NEEDED,
316 	TAA_MITIGATION_VERW,
317 	TAA_MITIGATION_TSX_DISABLED,
318 };
319 
320 /* Default mitigation for TAA-affected CPUs */
321 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
322 static bool taa_nosmt __ro_after_init;
323 
324 static const char * const taa_strings[] = {
325 	[TAA_MITIGATION_OFF]		= "Vulnerable",
326 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
327 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
328 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
329 };
330 
331 static void __init taa_select_mitigation(void)
332 {
333 	u64 ia32_cap;
334 
335 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
336 		taa_mitigation = TAA_MITIGATION_OFF;
337 		return;
338 	}
339 
340 	/* TSX previously disabled by tsx=off */
341 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
342 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
343 		return;
344 	}
345 
346 	if (cpu_mitigations_off()) {
347 		taa_mitigation = TAA_MITIGATION_OFF;
348 		return;
349 	}
350 
351 	/*
352 	 * TAA mitigation via VERW is turned off if both
353 	 * tsx_async_abort=off and mds=off are specified.
354 	 */
355 	if (taa_mitigation == TAA_MITIGATION_OFF &&
356 	    mds_mitigation == MDS_MITIGATION_OFF)
357 		return;
358 
359 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
360 		taa_mitigation = TAA_MITIGATION_VERW;
361 	else
362 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
363 
364 	/*
365 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
366 	 * A microcode update fixes this behavior to clear CPU buffers. It also
367 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
368 	 * ARCH_CAP_TSX_CTRL_MSR bit.
369 	 *
370 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
371 	 * update is required.
372 	 */
373 	ia32_cap = x86_read_arch_cap_msr();
374 	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
375 	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
376 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
377 
378 	/*
379 	 * TSX is enabled, select alternate mitigation for TAA which is
380 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
381 	 *
382 	 * For guests that can't determine whether the correct microcode is
383 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
384 	 */
385 	static_branch_enable(&mds_user_clear);
386 
387 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
388 		cpu_smt_disable(false);
389 }
390 
391 static int __init tsx_async_abort_parse_cmdline(char *str)
392 {
393 	if (!boot_cpu_has_bug(X86_BUG_TAA))
394 		return 0;
395 
396 	if (!str)
397 		return -EINVAL;
398 
399 	if (!strcmp(str, "off")) {
400 		taa_mitigation = TAA_MITIGATION_OFF;
401 	} else if (!strcmp(str, "full")) {
402 		taa_mitigation = TAA_MITIGATION_VERW;
403 	} else if (!strcmp(str, "full,nosmt")) {
404 		taa_mitigation = TAA_MITIGATION_VERW;
405 		taa_nosmt = true;
406 	}
407 
408 	return 0;
409 }
410 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
411 
412 #undef pr_fmt
413 #define pr_fmt(fmt)	"MMIO Stale Data: " fmt
414 
415 enum mmio_mitigations {
416 	MMIO_MITIGATION_OFF,
417 	MMIO_MITIGATION_UCODE_NEEDED,
418 	MMIO_MITIGATION_VERW,
419 };
420 
421 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
422 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
423 static bool mmio_nosmt __ro_after_init = false;
424 
425 static const char * const mmio_strings[] = {
426 	[MMIO_MITIGATION_OFF]		= "Vulnerable",
427 	[MMIO_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
428 	[MMIO_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
429 };
430 
431 static void __init mmio_select_mitigation(void)
432 {
433 	u64 ia32_cap;
434 
435 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
436 	    cpu_mitigations_off()) {
437 		mmio_mitigation = MMIO_MITIGATION_OFF;
438 		return;
439 	}
440 
441 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
442 		return;
443 
444 	ia32_cap = x86_read_arch_cap_msr();
445 
446 	/*
447 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
448 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
449 	 */
450 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
451 					      boot_cpu_has(X86_FEATURE_RTM)))
452 		static_branch_enable(&mds_user_clear);
453 	else
454 		static_branch_enable(&mmio_stale_data_clear);
455 
456 	/*
457 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
458 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
459 	 * is required irrespective of SMT state.
460 	 */
461 	if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
462 		static_branch_enable(&mds_idle_clear);
463 
464 	/*
465 	 * Check if the system has the right microcode.
466 	 *
467 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
468 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
469 	 * affected systems.
470 	 */
471 	if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
472 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
473 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
474 	     !(ia32_cap & ARCH_CAP_MDS_NO)))
475 		mmio_mitigation = MMIO_MITIGATION_VERW;
476 	else
477 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
478 
479 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
480 		cpu_smt_disable(false);
481 }
482 
483 static int __init mmio_stale_data_parse_cmdline(char *str)
484 {
485 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
486 		return 0;
487 
488 	if (!str)
489 		return -EINVAL;
490 
491 	if (!strcmp(str, "off")) {
492 		mmio_mitigation = MMIO_MITIGATION_OFF;
493 	} else if (!strcmp(str, "full")) {
494 		mmio_mitigation = MMIO_MITIGATION_VERW;
495 	} else if (!strcmp(str, "full,nosmt")) {
496 		mmio_mitigation = MMIO_MITIGATION_VERW;
497 		mmio_nosmt = true;
498 	}
499 
500 	return 0;
501 }
502 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
503 
504 #undef pr_fmt
505 #define pr_fmt(fmt)     "" fmt
506 
507 static void __init md_clear_update_mitigation(void)
508 {
509 	if (cpu_mitigations_off())
510 		return;
511 
512 	if (!static_key_enabled(&mds_user_clear))
513 		goto out;
514 
515 	/*
516 	 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
517 	 * mitigation, if necessary.
518 	 */
519 	if (mds_mitigation == MDS_MITIGATION_OFF &&
520 	    boot_cpu_has_bug(X86_BUG_MDS)) {
521 		mds_mitigation = MDS_MITIGATION_FULL;
522 		mds_select_mitigation();
523 	}
524 	if (taa_mitigation == TAA_MITIGATION_OFF &&
525 	    boot_cpu_has_bug(X86_BUG_TAA)) {
526 		taa_mitigation = TAA_MITIGATION_VERW;
527 		taa_select_mitigation();
528 	}
529 	if (mmio_mitigation == MMIO_MITIGATION_OFF &&
530 	    boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
531 		mmio_mitigation = MMIO_MITIGATION_VERW;
532 		mmio_select_mitigation();
533 	}
534 out:
535 	if (boot_cpu_has_bug(X86_BUG_MDS))
536 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
537 	if (boot_cpu_has_bug(X86_BUG_TAA))
538 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
539 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
540 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
541 }
542 
543 static void __init md_clear_select_mitigation(void)
544 {
545 	mds_select_mitigation();
546 	taa_select_mitigation();
547 	mmio_select_mitigation();
548 
549 	/*
550 	 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
551 	 * and print their mitigation after MDS, TAA and MMIO Stale Data
552 	 * mitigation selection is done.
553 	 */
554 	md_clear_update_mitigation();
555 }
556 
557 #undef pr_fmt
558 #define pr_fmt(fmt)	"SRBDS: " fmt
559 
560 enum srbds_mitigations {
561 	SRBDS_MITIGATION_OFF,
562 	SRBDS_MITIGATION_UCODE_NEEDED,
563 	SRBDS_MITIGATION_FULL,
564 	SRBDS_MITIGATION_TSX_OFF,
565 	SRBDS_MITIGATION_HYPERVISOR,
566 };
567 
568 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
569 
570 static const char * const srbds_strings[] = {
571 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
572 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
573 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
574 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
575 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
576 };
577 
578 static bool srbds_off;
579 
580 void update_srbds_msr(void)
581 {
582 	u64 mcu_ctrl;
583 
584 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
585 		return;
586 
587 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
588 		return;
589 
590 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
591 		return;
592 
593 	/*
594 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
595 	 * being disabled and it hasn't received the SRBDS MSR microcode.
596 	 */
597 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
598 		return;
599 
600 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
601 
602 	switch (srbds_mitigation) {
603 	case SRBDS_MITIGATION_OFF:
604 	case SRBDS_MITIGATION_TSX_OFF:
605 		mcu_ctrl |= RNGDS_MITG_DIS;
606 		break;
607 	case SRBDS_MITIGATION_FULL:
608 		mcu_ctrl &= ~RNGDS_MITG_DIS;
609 		break;
610 	default:
611 		break;
612 	}
613 
614 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
615 }
616 
617 static void __init srbds_select_mitigation(void)
618 {
619 	u64 ia32_cap;
620 
621 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
622 		return;
623 
624 	/*
625 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
626 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
627 	 * by Processor MMIO Stale Data vulnerability.
628 	 */
629 	ia32_cap = x86_read_arch_cap_msr();
630 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
631 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
632 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
633 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
634 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
635 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
636 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
637 	else if (cpu_mitigations_off() || srbds_off)
638 		srbds_mitigation = SRBDS_MITIGATION_OFF;
639 
640 	update_srbds_msr();
641 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
642 }
643 
644 static int __init srbds_parse_cmdline(char *str)
645 {
646 	if (!str)
647 		return -EINVAL;
648 
649 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
650 		return 0;
651 
652 	srbds_off = !strcmp(str, "off");
653 	return 0;
654 }
655 early_param("srbds", srbds_parse_cmdline);
656 
657 #undef pr_fmt
658 #define pr_fmt(fmt)     "L1D Flush : " fmt
659 
660 enum l1d_flush_mitigations {
661 	L1D_FLUSH_OFF = 0,
662 	L1D_FLUSH_ON,
663 };
664 
665 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
666 
667 static void __init l1d_flush_select_mitigation(void)
668 {
669 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
670 		return;
671 
672 	static_branch_enable(&switch_mm_cond_l1d_flush);
673 	pr_info("Conditional flush on switch_mm() enabled\n");
674 }
675 
676 static int __init l1d_flush_parse_cmdline(char *str)
677 {
678 	if (!strcmp(str, "on"))
679 		l1d_flush_mitigation = L1D_FLUSH_ON;
680 
681 	return 0;
682 }
683 early_param("l1d_flush", l1d_flush_parse_cmdline);
684 
685 #undef pr_fmt
686 #define pr_fmt(fmt)     "Spectre V1 : " fmt
687 
688 enum spectre_v1_mitigation {
689 	SPECTRE_V1_MITIGATION_NONE,
690 	SPECTRE_V1_MITIGATION_AUTO,
691 };
692 
693 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
694 	SPECTRE_V1_MITIGATION_AUTO;
695 
696 static const char * const spectre_v1_strings[] = {
697 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
698 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
699 };
700 
701 /*
702  * Does SMAP provide full mitigation against speculative kernel access to
703  * userspace?
704  */
705 static bool smap_works_speculatively(void)
706 {
707 	if (!boot_cpu_has(X86_FEATURE_SMAP))
708 		return false;
709 
710 	/*
711 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
712 	 * prevent speculative access to user data in the L1 cache.
713 	 * Consider SMAP to be non-functional as a mitigation on these
714 	 * CPUs.
715 	 */
716 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
717 		return false;
718 
719 	return true;
720 }
721 
722 static void __init spectre_v1_select_mitigation(void)
723 {
724 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
725 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
726 		return;
727 	}
728 
729 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
730 		/*
731 		 * With Spectre v1, a user can speculatively control either
732 		 * path of a conditional swapgs with a user-controlled GS
733 		 * value.  The mitigation is to add lfences to both code paths.
734 		 *
735 		 * If FSGSBASE is enabled, the user can put a kernel address in
736 		 * GS, in which case SMAP provides no protection.
737 		 *
738 		 * If FSGSBASE is disabled, the user can only put a user space
739 		 * address in GS.  That makes an attack harder, but still
740 		 * possible if there's no SMAP protection.
741 		 */
742 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
743 		    !smap_works_speculatively()) {
744 			/*
745 			 * Mitigation can be provided from SWAPGS itself or
746 			 * PTI as the CR3 write in the Meltdown mitigation
747 			 * is serializing.
748 			 *
749 			 * If neither is there, mitigate with an LFENCE to
750 			 * stop speculation through swapgs.
751 			 */
752 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
753 			    !boot_cpu_has(X86_FEATURE_PTI))
754 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
755 
756 			/*
757 			 * Enable lfences in the kernel entry (non-swapgs)
758 			 * paths, to prevent user entry from speculatively
759 			 * skipping swapgs.
760 			 */
761 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
762 		}
763 	}
764 
765 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
766 }
767 
768 static int __init nospectre_v1_cmdline(char *str)
769 {
770 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
771 	return 0;
772 }
773 early_param("nospectre_v1", nospectre_v1_cmdline);
774 
775 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
776 	SPECTRE_V2_NONE;
777 
778 #undef pr_fmt
779 #define pr_fmt(fmt)     "RETBleed: " fmt
780 
781 enum retbleed_mitigation {
782 	RETBLEED_MITIGATION_NONE,
783 	RETBLEED_MITIGATION_UNRET,
784 	RETBLEED_MITIGATION_IBPB,
785 	RETBLEED_MITIGATION_IBRS,
786 	RETBLEED_MITIGATION_EIBRS,
787 };
788 
789 enum retbleed_mitigation_cmd {
790 	RETBLEED_CMD_OFF,
791 	RETBLEED_CMD_AUTO,
792 	RETBLEED_CMD_UNRET,
793 	RETBLEED_CMD_IBPB,
794 };
795 
796 static const char * const retbleed_strings[] = {
797 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
798 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
799 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
800 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
801 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
802 };
803 
804 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
805 	RETBLEED_MITIGATION_NONE;
806 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
807 	RETBLEED_CMD_AUTO;
808 
809 static int __ro_after_init retbleed_nosmt = false;
810 
811 static int __init retbleed_parse_cmdline(char *str)
812 {
813 	if (!str)
814 		return -EINVAL;
815 
816 	while (str) {
817 		char *next = strchr(str, ',');
818 		if (next) {
819 			*next = 0;
820 			next++;
821 		}
822 
823 		if (!strcmp(str, "off")) {
824 			retbleed_cmd = RETBLEED_CMD_OFF;
825 		} else if (!strcmp(str, "auto")) {
826 			retbleed_cmd = RETBLEED_CMD_AUTO;
827 		} else if (!strcmp(str, "unret")) {
828 			retbleed_cmd = RETBLEED_CMD_UNRET;
829 		} else if (!strcmp(str, "ibpb")) {
830 			retbleed_cmd = RETBLEED_CMD_IBPB;
831 		} else if (!strcmp(str, "nosmt")) {
832 			retbleed_nosmt = true;
833 		} else {
834 			pr_err("Ignoring unknown retbleed option (%s).", str);
835 		}
836 
837 		str = next;
838 	}
839 
840 	return 0;
841 }
842 early_param("retbleed", retbleed_parse_cmdline);
843 
844 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
845 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
846 
847 static void __init retbleed_select_mitigation(void)
848 {
849 	bool mitigate_smt = false;
850 
851 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
852 		return;
853 
854 	switch (retbleed_cmd) {
855 	case RETBLEED_CMD_OFF:
856 		return;
857 
858 	case RETBLEED_CMD_UNRET:
859 		if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
860 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
861 		} else {
862 			pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
863 			goto do_cmd_auto;
864 		}
865 		break;
866 
867 	case RETBLEED_CMD_IBPB:
868 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
869 			pr_err("WARNING: CPU does not support IBPB.\n");
870 			goto do_cmd_auto;
871 		} else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
872 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
873 		} else {
874 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
875 			goto do_cmd_auto;
876 		}
877 		break;
878 
879 do_cmd_auto:
880 	case RETBLEED_CMD_AUTO:
881 	default:
882 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
883 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
884 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
885 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
886 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
887 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
888 		}
889 
890 		/*
891 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
892 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
893 		 * be set accordingly below.
894 		 */
895 
896 		break;
897 	}
898 
899 	switch (retbleed_mitigation) {
900 	case RETBLEED_MITIGATION_UNRET:
901 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
902 		setup_force_cpu_cap(X86_FEATURE_UNRET);
903 
904 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
905 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
906 			pr_err(RETBLEED_UNTRAIN_MSG);
907 
908 		mitigate_smt = true;
909 		break;
910 
911 	case RETBLEED_MITIGATION_IBPB:
912 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
913 		mitigate_smt = true;
914 		break;
915 
916 	default:
917 		break;
918 	}
919 
920 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
921 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
922 		cpu_smt_disable(false);
923 
924 	/*
925 	 * Let IBRS trump all on Intel without affecting the effects of the
926 	 * retbleed= cmdline option.
927 	 */
928 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
929 		switch (spectre_v2_enabled) {
930 		case SPECTRE_V2_IBRS:
931 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
932 			break;
933 		case SPECTRE_V2_EIBRS:
934 		case SPECTRE_V2_EIBRS_RETPOLINE:
935 		case SPECTRE_V2_EIBRS_LFENCE:
936 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
937 			break;
938 		default:
939 			pr_err(RETBLEED_INTEL_MSG);
940 		}
941 	}
942 
943 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
944 }
945 
946 #undef pr_fmt
947 #define pr_fmt(fmt)     "Spectre V2 : " fmt
948 
949 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
950 	SPECTRE_V2_USER_NONE;
951 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
952 	SPECTRE_V2_USER_NONE;
953 
954 #ifdef CONFIG_RETPOLINE
955 static bool spectre_v2_bad_module;
956 
957 bool retpoline_module_ok(bool has_retpoline)
958 {
959 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
960 		return true;
961 
962 	pr_err("System may be vulnerable to spectre v2\n");
963 	spectre_v2_bad_module = true;
964 	return false;
965 }
966 
967 static inline const char *spectre_v2_module_string(void)
968 {
969 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
970 }
971 #else
972 static inline const char *spectre_v2_module_string(void) { return ""; }
973 #endif
974 
975 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
976 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
977 #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"
978 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
979 
980 #ifdef CONFIG_BPF_SYSCALL
981 void unpriv_ebpf_notify(int new_state)
982 {
983 	if (new_state)
984 		return;
985 
986 	/* Unprivileged eBPF is enabled */
987 
988 	switch (spectre_v2_enabled) {
989 	case SPECTRE_V2_EIBRS:
990 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
991 		break;
992 	case SPECTRE_V2_EIBRS_LFENCE:
993 		if (sched_smt_active())
994 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
995 		break;
996 	default:
997 		break;
998 	}
999 }
1000 #endif
1001 
1002 static inline bool match_option(const char *arg, int arglen, const char *opt)
1003 {
1004 	int len = strlen(opt);
1005 
1006 	return len == arglen && !strncmp(arg, opt, len);
1007 }
1008 
1009 /* The kernel command line selection for spectre v2 */
1010 enum spectre_v2_mitigation_cmd {
1011 	SPECTRE_V2_CMD_NONE,
1012 	SPECTRE_V2_CMD_AUTO,
1013 	SPECTRE_V2_CMD_FORCE,
1014 	SPECTRE_V2_CMD_RETPOLINE,
1015 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1016 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1017 	SPECTRE_V2_CMD_EIBRS,
1018 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1019 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1020 	SPECTRE_V2_CMD_IBRS,
1021 };
1022 
1023 enum spectre_v2_user_cmd {
1024 	SPECTRE_V2_USER_CMD_NONE,
1025 	SPECTRE_V2_USER_CMD_AUTO,
1026 	SPECTRE_V2_USER_CMD_FORCE,
1027 	SPECTRE_V2_USER_CMD_PRCTL,
1028 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1029 	SPECTRE_V2_USER_CMD_SECCOMP,
1030 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1031 };
1032 
1033 static const char * const spectre_v2_user_strings[] = {
1034 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1035 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1036 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1037 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1038 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1039 };
1040 
1041 static const struct {
1042 	const char			*option;
1043 	enum spectre_v2_user_cmd	cmd;
1044 	bool				secure;
1045 } v2_user_options[] __initconst = {
1046 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1047 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1048 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1049 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1050 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1051 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1052 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1053 };
1054 
1055 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1056 {
1057 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1058 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1059 }
1060 
1061 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1062 
1063 static enum spectre_v2_user_cmd __init
1064 spectre_v2_parse_user_cmdline(void)
1065 {
1066 	char arg[20];
1067 	int ret, i;
1068 
1069 	switch (spectre_v2_cmd) {
1070 	case SPECTRE_V2_CMD_NONE:
1071 		return SPECTRE_V2_USER_CMD_NONE;
1072 	case SPECTRE_V2_CMD_FORCE:
1073 		return SPECTRE_V2_USER_CMD_FORCE;
1074 	default:
1075 		break;
1076 	}
1077 
1078 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1079 				  arg, sizeof(arg));
1080 	if (ret < 0)
1081 		return SPECTRE_V2_USER_CMD_AUTO;
1082 
1083 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1084 		if (match_option(arg, ret, v2_user_options[i].option)) {
1085 			spec_v2_user_print_cond(v2_user_options[i].option,
1086 						v2_user_options[i].secure);
1087 			return v2_user_options[i].cmd;
1088 		}
1089 	}
1090 
1091 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1092 	return SPECTRE_V2_USER_CMD_AUTO;
1093 }
1094 
1095 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1096 {
1097 	return mode == SPECTRE_V2_IBRS ||
1098 	       mode == SPECTRE_V2_EIBRS ||
1099 	       mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1100 	       mode == SPECTRE_V2_EIBRS_LFENCE;
1101 }
1102 
1103 static void __init
1104 spectre_v2_user_select_mitigation(void)
1105 {
1106 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1107 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1108 	enum spectre_v2_user_cmd cmd;
1109 
1110 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1111 		return;
1112 
1113 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1114 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1115 		smt_possible = false;
1116 
1117 	cmd = spectre_v2_parse_user_cmdline();
1118 	switch (cmd) {
1119 	case SPECTRE_V2_USER_CMD_NONE:
1120 		goto set_mode;
1121 	case SPECTRE_V2_USER_CMD_FORCE:
1122 		mode = SPECTRE_V2_USER_STRICT;
1123 		break;
1124 	case SPECTRE_V2_USER_CMD_AUTO:
1125 	case SPECTRE_V2_USER_CMD_PRCTL:
1126 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1127 		mode = SPECTRE_V2_USER_PRCTL;
1128 		break;
1129 	case SPECTRE_V2_USER_CMD_SECCOMP:
1130 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1131 		if (IS_ENABLED(CONFIG_SECCOMP))
1132 			mode = SPECTRE_V2_USER_SECCOMP;
1133 		else
1134 			mode = SPECTRE_V2_USER_PRCTL;
1135 		break;
1136 	}
1137 
1138 	/* Initialize Indirect Branch Prediction Barrier */
1139 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1140 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1141 
1142 		spectre_v2_user_ibpb = mode;
1143 		switch (cmd) {
1144 		case SPECTRE_V2_USER_CMD_FORCE:
1145 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1146 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1147 			static_branch_enable(&switch_mm_always_ibpb);
1148 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1149 			break;
1150 		case SPECTRE_V2_USER_CMD_PRCTL:
1151 		case SPECTRE_V2_USER_CMD_AUTO:
1152 		case SPECTRE_V2_USER_CMD_SECCOMP:
1153 			static_branch_enable(&switch_mm_cond_ibpb);
1154 			break;
1155 		default:
1156 			break;
1157 		}
1158 
1159 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1160 			static_key_enabled(&switch_mm_always_ibpb) ?
1161 			"always-on" : "conditional");
1162 	}
1163 
1164 	/*
1165 	 * If no STIBP, IBRS or enhanced IBRS is enabled, or SMT impossible,
1166 	 * STIBP is not required.
1167 	 */
1168 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1169 	    !smt_possible ||
1170 	    spectre_v2_in_ibrs_mode(spectre_v2_enabled))
1171 		return;
1172 
1173 	/*
1174 	 * At this point, an STIBP mode other than "off" has been set.
1175 	 * If STIBP support is not being forced, check if STIBP always-on
1176 	 * is preferred.
1177 	 */
1178 	if (mode != SPECTRE_V2_USER_STRICT &&
1179 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1180 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1181 
1182 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1183 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1184 		if (mode != SPECTRE_V2_USER_STRICT &&
1185 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1186 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1187 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1188 	}
1189 
1190 	spectre_v2_user_stibp = mode;
1191 
1192 set_mode:
1193 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1194 }
1195 
1196 static const char * const spectre_v2_strings[] = {
1197 	[SPECTRE_V2_NONE]			= "Vulnerable",
1198 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1199 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1200 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced IBRS",
1201 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced IBRS + LFENCE",
1202 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced IBRS + Retpolines",
1203 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1204 };
1205 
1206 static const struct {
1207 	const char *option;
1208 	enum spectre_v2_mitigation_cmd cmd;
1209 	bool secure;
1210 } mitigation_options[] __initconst = {
1211 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1212 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1213 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1214 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1215 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1216 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1217 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1218 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1219 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1220 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1221 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1222 };
1223 
1224 static void __init spec_v2_print_cond(const char *reason, bool secure)
1225 {
1226 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1227 		pr_info("%s selected on command line.\n", reason);
1228 }
1229 
1230 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1231 {
1232 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1233 	char arg[20];
1234 	int ret, i;
1235 
1236 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1237 	    cpu_mitigations_off())
1238 		return SPECTRE_V2_CMD_NONE;
1239 
1240 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1241 	if (ret < 0)
1242 		return SPECTRE_V2_CMD_AUTO;
1243 
1244 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1245 		if (!match_option(arg, ret, mitigation_options[i].option))
1246 			continue;
1247 		cmd = mitigation_options[i].cmd;
1248 		break;
1249 	}
1250 
1251 	if (i >= ARRAY_SIZE(mitigation_options)) {
1252 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1253 		return SPECTRE_V2_CMD_AUTO;
1254 	}
1255 
1256 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1257 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1258 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1259 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1260 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1261 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1262 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1263 		       mitigation_options[i].option);
1264 		return SPECTRE_V2_CMD_AUTO;
1265 	}
1266 
1267 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1268 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1269 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1270 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1271 		pr_err("%s selected but CPU doesn't have eIBRS. Switching to AUTO select\n",
1272 		       mitigation_options[i].option);
1273 		return SPECTRE_V2_CMD_AUTO;
1274 	}
1275 
1276 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1277 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1278 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1279 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1280 		       mitigation_options[i].option);
1281 		return SPECTRE_V2_CMD_AUTO;
1282 	}
1283 
1284 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1285 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1286 		       mitigation_options[i].option);
1287 		return SPECTRE_V2_CMD_AUTO;
1288 	}
1289 
1290 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1291 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1292 		       mitigation_options[i].option);
1293 		return SPECTRE_V2_CMD_AUTO;
1294 	}
1295 
1296 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1297 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1298 		       mitigation_options[i].option);
1299 		return SPECTRE_V2_CMD_AUTO;
1300 	}
1301 
1302 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_has(X86_FEATURE_XENPV)) {
1303 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1304 		       mitigation_options[i].option);
1305 		return SPECTRE_V2_CMD_AUTO;
1306 	}
1307 
1308 	spec_v2_print_cond(mitigation_options[i].option,
1309 			   mitigation_options[i].secure);
1310 	return cmd;
1311 }
1312 
1313 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1314 {
1315 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1316 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1317 		return SPECTRE_V2_NONE;
1318 	}
1319 
1320 	return SPECTRE_V2_RETPOLINE;
1321 }
1322 
1323 /* Disable in-kernel use of non-RSB RET predictors */
1324 static void __init spec_ctrl_disable_kernel_rrsba(void)
1325 {
1326 	u64 ia32_cap;
1327 
1328 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1329 		return;
1330 
1331 	ia32_cap = x86_read_arch_cap_msr();
1332 
1333 	if (ia32_cap & ARCH_CAP_RRSBA) {
1334 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1335 		write_spec_ctrl_current(x86_spec_ctrl_base, true);
1336 	}
1337 }
1338 
1339 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1340 {
1341 	/*
1342 	 * Similar to context switches, there are two types of RSB attacks
1343 	 * after VM exit:
1344 	 *
1345 	 * 1) RSB underflow
1346 	 *
1347 	 * 2) Poisoned RSB entry
1348 	 *
1349 	 * When retpoline is enabled, both are mitigated by filling/clearing
1350 	 * the RSB.
1351 	 *
1352 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1353 	 * prediction isolation protections, RSB still needs to be cleared
1354 	 * because of #2.  Note that SMEP provides no protection here, unlike
1355 	 * user-space-poisoned RSB entries.
1356 	 *
1357 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1358 	 * bug is present then a LITE version of RSB protection is required,
1359 	 * just a single call needs to retire before a RET is executed.
1360 	 */
1361 	switch (mode) {
1362 	case SPECTRE_V2_NONE:
1363 		return;
1364 
1365 	case SPECTRE_V2_EIBRS_LFENCE:
1366 	case SPECTRE_V2_EIBRS:
1367 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1368 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1369 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1370 		}
1371 		return;
1372 
1373 	case SPECTRE_V2_EIBRS_RETPOLINE:
1374 	case SPECTRE_V2_RETPOLINE:
1375 	case SPECTRE_V2_LFENCE:
1376 	case SPECTRE_V2_IBRS:
1377 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1378 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1379 		return;
1380 	}
1381 
1382 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1383 	dump_stack();
1384 }
1385 
1386 static void __init spectre_v2_select_mitigation(void)
1387 {
1388 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1389 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1390 
1391 	/*
1392 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1393 	 * then nothing to do.
1394 	 */
1395 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1396 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1397 		return;
1398 
1399 	switch (cmd) {
1400 	case SPECTRE_V2_CMD_NONE:
1401 		return;
1402 
1403 	case SPECTRE_V2_CMD_FORCE:
1404 	case SPECTRE_V2_CMD_AUTO:
1405 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1406 			mode = SPECTRE_V2_EIBRS;
1407 			break;
1408 		}
1409 
1410 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1411 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1412 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1413 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1414 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1415 			mode = SPECTRE_V2_IBRS;
1416 			break;
1417 		}
1418 
1419 		mode = spectre_v2_select_retpoline();
1420 		break;
1421 
1422 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1423 		pr_err(SPECTRE_V2_LFENCE_MSG);
1424 		mode = SPECTRE_V2_LFENCE;
1425 		break;
1426 
1427 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1428 		mode = SPECTRE_V2_RETPOLINE;
1429 		break;
1430 
1431 	case SPECTRE_V2_CMD_RETPOLINE:
1432 		mode = spectre_v2_select_retpoline();
1433 		break;
1434 
1435 	case SPECTRE_V2_CMD_IBRS:
1436 		mode = SPECTRE_V2_IBRS;
1437 		break;
1438 
1439 	case SPECTRE_V2_CMD_EIBRS:
1440 		mode = SPECTRE_V2_EIBRS;
1441 		break;
1442 
1443 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1444 		mode = SPECTRE_V2_EIBRS_LFENCE;
1445 		break;
1446 
1447 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1448 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1449 		break;
1450 	}
1451 
1452 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1453 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1454 
1455 	if (spectre_v2_in_ibrs_mode(mode)) {
1456 		x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1457 		write_spec_ctrl_current(x86_spec_ctrl_base, true);
1458 	}
1459 
1460 	switch (mode) {
1461 	case SPECTRE_V2_NONE:
1462 	case SPECTRE_V2_EIBRS:
1463 		break;
1464 
1465 	case SPECTRE_V2_IBRS:
1466 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1467 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1468 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1469 		break;
1470 
1471 	case SPECTRE_V2_LFENCE:
1472 	case SPECTRE_V2_EIBRS_LFENCE:
1473 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1474 		fallthrough;
1475 
1476 	case SPECTRE_V2_RETPOLINE:
1477 	case SPECTRE_V2_EIBRS_RETPOLINE:
1478 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1479 		break;
1480 	}
1481 
1482 	/*
1483 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1484 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1485 	 * prediction from a non-RSB predictor is still a risk.
1486 	 */
1487 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1488 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1489 	    mode == SPECTRE_V2_RETPOLINE)
1490 		spec_ctrl_disable_kernel_rrsba();
1491 
1492 	spectre_v2_enabled = mode;
1493 	pr_info("%s\n", spectre_v2_strings[mode]);
1494 
1495 	/*
1496 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1497 	 * context switch.  In general there are two types of RSB attacks
1498 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1499 	 *
1500 	 * 1) RSB underflow
1501 	 *
1502 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1503 	 *    speculated return targets may come from the branch predictor,
1504 	 *    which could have a user-poisoned BTB or BHB entry.
1505 	 *
1506 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1507 	 *    regardless of the state of the RSB.
1508 	 *
1509 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1510 	 *    scenario is mitigated by the IBRS branch prediction isolation
1511 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1512 	 *    protect against this type of attack.
1513 	 *
1514 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1515 	 *
1516 	 * 2) Poisoned RSB entry
1517 	 *
1518 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1519 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1520 	 *    entry.
1521 	 *
1522 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1523 	 *    eIBRS.
1524 	 *
1525 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1526 	 *    RSB clearing.
1527 	 *
1528 	 * So to mitigate all cases, unconditionally fill RSB on context
1529 	 * switches.
1530 	 *
1531 	 * FIXME: Is this pointless for retbleed-affected AMD?
1532 	 */
1533 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1534 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1535 
1536 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1537 
1538 	/*
1539 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1540 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1541 	 * firmware calls only when IBRS / Enhanced IBRS aren't otherwise
1542 	 * enabled.
1543 	 *
1544 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1545 	 * the user might select retpoline on the kernel command line and if
1546 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1547 	 * enable IBRS around firmware calls.
1548 	 */
1549 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1550 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1551 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1552 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1553 
1554 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1555 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1556 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1557 		}
1558 
1559 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1560 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1561 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1562 	}
1563 
1564 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1565 	spectre_v2_cmd = cmd;
1566 }
1567 
1568 static void update_stibp_msr(void * __unused)
1569 {
1570 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1571 	write_spec_ctrl_current(val, true);
1572 }
1573 
1574 /* Update x86_spec_ctrl_base in case SMT state changed. */
1575 static void update_stibp_strict(void)
1576 {
1577 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1578 
1579 	if (sched_smt_active())
1580 		mask |= SPEC_CTRL_STIBP;
1581 
1582 	if (mask == x86_spec_ctrl_base)
1583 		return;
1584 
1585 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1586 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1587 	x86_spec_ctrl_base = mask;
1588 	on_each_cpu(update_stibp_msr, NULL, 1);
1589 }
1590 
1591 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1592 static void update_indir_branch_cond(void)
1593 {
1594 	if (sched_smt_active())
1595 		static_branch_enable(&switch_to_cond_stibp);
1596 	else
1597 		static_branch_disable(&switch_to_cond_stibp);
1598 }
1599 
1600 #undef pr_fmt
1601 #define pr_fmt(fmt) fmt
1602 
1603 /* Update the static key controlling the MDS CPU buffer clear in idle */
1604 static void update_mds_branch_idle(void)
1605 {
1606 	u64 ia32_cap = x86_read_arch_cap_msr();
1607 
1608 	/*
1609 	 * Enable the idle clearing if SMT is active on CPUs which are
1610 	 * affected only by MSBDS and not any other MDS variant.
1611 	 *
1612 	 * The other variants cannot be mitigated when SMT is enabled, so
1613 	 * clearing the buffers on idle just to prevent the Store Buffer
1614 	 * repartitioning leak would be a window dressing exercise.
1615 	 */
1616 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1617 		return;
1618 
1619 	if (sched_smt_active()) {
1620 		static_branch_enable(&mds_idle_clear);
1621 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1622 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1623 		static_branch_disable(&mds_idle_clear);
1624 	}
1625 }
1626 
1627 #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"
1628 #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"
1629 #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"
1630 
1631 void cpu_bugs_smt_update(void)
1632 {
1633 	mutex_lock(&spec_ctrl_mutex);
1634 
1635 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1636 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1637 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1638 
1639 	switch (spectre_v2_user_stibp) {
1640 	case SPECTRE_V2_USER_NONE:
1641 		break;
1642 	case SPECTRE_V2_USER_STRICT:
1643 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1644 		update_stibp_strict();
1645 		break;
1646 	case SPECTRE_V2_USER_PRCTL:
1647 	case SPECTRE_V2_USER_SECCOMP:
1648 		update_indir_branch_cond();
1649 		break;
1650 	}
1651 
1652 	switch (mds_mitigation) {
1653 	case MDS_MITIGATION_FULL:
1654 	case MDS_MITIGATION_VMWERV:
1655 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1656 			pr_warn_once(MDS_MSG_SMT);
1657 		update_mds_branch_idle();
1658 		break;
1659 	case MDS_MITIGATION_OFF:
1660 		break;
1661 	}
1662 
1663 	switch (taa_mitigation) {
1664 	case TAA_MITIGATION_VERW:
1665 	case TAA_MITIGATION_UCODE_NEEDED:
1666 		if (sched_smt_active())
1667 			pr_warn_once(TAA_MSG_SMT);
1668 		break;
1669 	case TAA_MITIGATION_TSX_DISABLED:
1670 	case TAA_MITIGATION_OFF:
1671 		break;
1672 	}
1673 
1674 	switch (mmio_mitigation) {
1675 	case MMIO_MITIGATION_VERW:
1676 	case MMIO_MITIGATION_UCODE_NEEDED:
1677 		if (sched_smt_active())
1678 			pr_warn_once(MMIO_MSG_SMT);
1679 		break;
1680 	case MMIO_MITIGATION_OFF:
1681 		break;
1682 	}
1683 
1684 	mutex_unlock(&spec_ctrl_mutex);
1685 }
1686 
1687 #undef pr_fmt
1688 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1689 
1690 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1691 
1692 /* The kernel command line selection */
1693 enum ssb_mitigation_cmd {
1694 	SPEC_STORE_BYPASS_CMD_NONE,
1695 	SPEC_STORE_BYPASS_CMD_AUTO,
1696 	SPEC_STORE_BYPASS_CMD_ON,
1697 	SPEC_STORE_BYPASS_CMD_PRCTL,
1698 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1699 };
1700 
1701 static const char * const ssb_strings[] = {
1702 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1703 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1704 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1705 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1706 };
1707 
1708 static const struct {
1709 	const char *option;
1710 	enum ssb_mitigation_cmd cmd;
1711 } ssb_mitigation_options[]  __initconst = {
1712 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1713 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1714 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1715 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1716 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1717 };
1718 
1719 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1720 {
1721 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1722 	char arg[20];
1723 	int ret, i;
1724 
1725 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1726 	    cpu_mitigations_off()) {
1727 		return SPEC_STORE_BYPASS_CMD_NONE;
1728 	} else {
1729 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1730 					  arg, sizeof(arg));
1731 		if (ret < 0)
1732 			return SPEC_STORE_BYPASS_CMD_AUTO;
1733 
1734 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1735 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1736 				continue;
1737 
1738 			cmd = ssb_mitigation_options[i].cmd;
1739 			break;
1740 		}
1741 
1742 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1743 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1744 			return SPEC_STORE_BYPASS_CMD_AUTO;
1745 		}
1746 	}
1747 
1748 	return cmd;
1749 }
1750 
1751 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1752 {
1753 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1754 	enum ssb_mitigation_cmd cmd;
1755 
1756 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1757 		return mode;
1758 
1759 	cmd = ssb_parse_cmdline();
1760 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1761 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1762 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1763 		return mode;
1764 
1765 	switch (cmd) {
1766 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1767 		/*
1768 		 * Choose prctl+seccomp as the default mode if seccomp is
1769 		 * enabled.
1770 		 */
1771 		if (IS_ENABLED(CONFIG_SECCOMP))
1772 			mode = SPEC_STORE_BYPASS_SECCOMP;
1773 		else
1774 			mode = SPEC_STORE_BYPASS_PRCTL;
1775 		break;
1776 	case SPEC_STORE_BYPASS_CMD_ON:
1777 		mode = SPEC_STORE_BYPASS_DISABLE;
1778 		break;
1779 	case SPEC_STORE_BYPASS_CMD_AUTO:
1780 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1781 		mode = SPEC_STORE_BYPASS_PRCTL;
1782 		break;
1783 	case SPEC_STORE_BYPASS_CMD_NONE:
1784 		break;
1785 	}
1786 
1787 	/*
1788 	 * We have three CPU feature flags that are in play here:
1789 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1790 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1791 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1792 	 */
1793 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1794 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1795 		/*
1796 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1797 		 * use a completely different MSR and bit dependent on family.
1798 		 */
1799 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1800 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1801 			x86_amd_ssb_disable();
1802 		} else {
1803 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1804 			write_spec_ctrl_current(x86_spec_ctrl_base, true);
1805 		}
1806 	}
1807 
1808 	return mode;
1809 }
1810 
1811 static void ssb_select_mitigation(void)
1812 {
1813 	ssb_mode = __ssb_select_mitigation();
1814 
1815 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1816 		pr_info("%s\n", ssb_strings[ssb_mode]);
1817 }
1818 
1819 #undef pr_fmt
1820 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1821 
1822 static void task_update_spec_tif(struct task_struct *tsk)
1823 {
1824 	/* Force the update of the real TIF bits */
1825 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1826 
1827 	/*
1828 	 * Immediately update the speculation control MSRs for the current
1829 	 * task, but for a non-current task delay setting the CPU
1830 	 * mitigation until it is scheduled next.
1831 	 *
1832 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1833 	 * always the current task.
1834 	 */
1835 	if (tsk == current)
1836 		speculation_ctrl_update_current();
1837 }
1838 
1839 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1840 {
1841 
1842 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1843 		return -EPERM;
1844 
1845 	switch (ctrl) {
1846 	case PR_SPEC_ENABLE:
1847 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1848 		return 0;
1849 	case PR_SPEC_DISABLE:
1850 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1851 		return 0;
1852 	default:
1853 		return -ERANGE;
1854 	}
1855 }
1856 
1857 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1858 {
1859 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1860 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1861 		return -ENXIO;
1862 
1863 	switch (ctrl) {
1864 	case PR_SPEC_ENABLE:
1865 		/* If speculation is force disabled, enable is not allowed */
1866 		if (task_spec_ssb_force_disable(task))
1867 			return -EPERM;
1868 		task_clear_spec_ssb_disable(task);
1869 		task_clear_spec_ssb_noexec(task);
1870 		task_update_spec_tif(task);
1871 		break;
1872 	case PR_SPEC_DISABLE:
1873 		task_set_spec_ssb_disable(task);
1874 		task_clear_spec_ssb_noexec(task);
1875 		task_update_spec_tif(task);
1876 		break;
1877 	case PR_SPEC_FORCE_DISABLE:
1878 		task_set_spec_ssb_disable(task);
1879 		task_set_spec_ssb_force_disable(task);
1880 		task_clear_spec_ssb_noexec(task);
1881 		task_update_spec_tif(task);
1882 		break;
1883 	case PR_SPEC_DISABLE_NOEXEC:
1884 		if (task_spec_ssb_force_disable(task))
1885 			return -EPERM;
1886 		task_set_spec_ssb_disable(task);
1887 		task_set_spec_ssb_noexec(task);
1888 		task_update_spec_tif(task);
1889 		break;
1890 	default:
1891 		return -ERANGE;
1892 	}
1893 	return 0;
1894 }
1895 
1896 static bool is_spec_ib_user_controlled(void)
1897 {
1898 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1899 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1900 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1901 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1902 }
1903 
1904 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1905 {
1906 	switch (ctrl) {
1907 	case PR_SPEC_ENABLE:
1908 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1909 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1910 			return 0;
1911 
1912 		/*
1913 		 * With strict mode for both IBPB and STIBP, the instruction
1914 		 * code paths avoid checking this task flag and instead,
1915 		 * unconditionally run the instruction. However, STIBP and IBPB
1916 		 * are independent and either can be set to conditionally
1917 		 * enabled regardless of the mode of the other.
1918 		 *
1919 		 * If either is set to conditional, allow the task flag to be
1920 		 * updated, unless it was force-disabled by a previous prctl
1921 		 * call. Currently, this is possible on an AMD CPU which has the
1922 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1923 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1924 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1925 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1926 		 */
1927 		if (!is_spec_ib_user_controlled() ||
1928 		    task_spec_ib_force_disable(task))
1929 			return -EPERM;
1930 
1931 		task_clear_spec_ib_disable(task);
1932 		task_update_spec_tif(task);
1933 		break;
1934 	case PR_SPEC_DISABLE:
1935 	case PR_SPEC_FORCE_DISABLE:
1936 		/*
1937 		 * Indirect branch speculation is always allowed when
1938 		 * mitigation is force disabled.
1939 		 */
1940 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1941 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1942 			return -EPERM;
1943 
1944 		if (!is_spec_ib_user_controlled())
1945 			return 0;
1946 
1947 		task_set_spec_ib_disable(task);
1948 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1949 			task_set_spec_ib_force_disable(task);
1950 		task_update_spec_tif(task);
1951 		break;
1952 	default:
1953 		return -ERANGE;
1954 	}
1955 	return 0;
1956 }
1957 
1958 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1959 			     unsigned long ctrl)
1960 {
1961 	switch (which) {
1962 	case PR_SPEC_STORE_BYPASS:
1963 		return ssb_prctl_set(task, ctrl);
1964 	case PR_SPEC_INDIRECT_BRANCH:
1965 		return ib_prctl_set(task, ctrl);
1966 	case PR_SPEC_L1D_FLUSH:
1967 		return l1d_flush_prctl_set(task, ctrl);
1968 	default:
1969 		return -ENODEV;
1970 	}
1971 }
1972 
1973 #ifdef CONFIG_SECCOMP
1974 void arch_seccomp_spec_mitigate(struct task_struct *task)
1975 {
1976 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1977 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1978 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1979 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1980 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1981 }
1982 #endif
1983 
1984 static int l1d_flush_prctl_get(struct task_struct *task)
1985 {
1986 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1987 		return PR_SPEC_FORCE_DISABLE;
1988 
1989 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
1990 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1991 	else
1992 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1993 }
1994 
1995 static int ssb_prctl_get(struct task_struct *task)
1996 {
1997 	switch (ssb_mode) {
1998 	case SPEC_STORE_BYPASS_DISABLE:
1999 		return PR_SPEC_DISABLE;
2000 	case SPEC_STORE_BYPASS_SECCOMP:
2001 	case SPEC_STORE_BYPASS_PRCTL:
2002 		if (task_spec_ssb_force_disable(task))
2003 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2004 		if (task_spec_ssb_noexec(task))
2005 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2006 		if (task_spec_ssb_disable(task))
2007 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2008 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2009 	default:
2010 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2011 			return PR_SPEC_ENABLE;
2012 		return PR_SPEC_NOT_AFFECTED;
2013 	}
2014 }
2015 
2016 static int ib_prctl_get(struct task_struct *task)
2017 {
2018 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2019 		return PR_SPEC_NOT_AFFECTED;
2020 
2021 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2022 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2023 		return PR_SPEC_ENABLE;
2024 	else if (is_spec_ib_user_controlled()) {
2025 		if (task_spec_ib_force_disable(task))
2026 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2027 		if (task_spec_ib_disable(task))
2028 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2029 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2030 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2031 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2032 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2033 		return PR_SPEC_DISABLE;
2034 	else
2035 		return PR_SPEC_NOT_AFFECTED;
2036 }
2037 
2038 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2039 {
2040 	switch (which) {
2041 	case PR_SPEC_STORE_BYPASS:
2042 		return ssb_prctl_get(task);
2043 	case PR_SPEC_INDIRECT_BRANCH:
2044 		return ib_prctl_get(task);
2045 	case PR_SPEC_L1D_FLUSH:
2046 		return l1d_flush_prctl_get(task);
2047 	default:
2048 		return -ENODEV;
2049 	}
2050 }
2051 
2052 void x86_spec_ctrl_setup_ap(void)
2053 {
2054 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2055 		write_spec_ctrl_current(x86_spec_ctrl_base, true);
2056 
2057 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2058 		x86_amd_ssb_disable();
2059 }
2060 
2061 bool itlb_multihit_kvm_mitigation;
2062 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2063 
2064 #undef pr_fmt
2065 #define pr_fmt(fmt)	"L1TF: " fmt
2066 
2067 /* Default mitigation for L1TF-affected CPUs */
2068 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2069 #if IS_ENABLED(CONFIG_KVM_INTEL)
2070 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2071 #endif
2072 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2073 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2074 
2075 /*
2076  * These CPUs all support 44bits physical address space internally in the
2077  * cache but CPUID can report a smaller number of physical address bits.
2078  *
2079  * The L1TF mitigation uses the top most address bit for the inversion of
2080  * non present PTEs. When the installed memory reaches into the top most
2081  * address bit due to memory holes, which has been observed on machines
2082  * which report 36bits physical address bits and have 32G RAM installed,
2083  * then the mitigation range check in l1tf_select_mitigation() triggers.
2084  * This is a false positive because the mitigation is still possible due to
2085  * the fact that the cache uses 44bit internally. Use the cache bits
2086  * instead of the reported physical bits and adjust them on the affected
2087  * machines to 44bit if the reported bits are less than 44.
2088  */
2089 static void override_cache_bits(struct cpuinfo_x86 *c)
2090 {
2091 	if (c->x86 != 6)
2092 		return;
2093 
2094 	switch (c->x86_model) {
2095 	case INTEL_FAM6_NEHALEM:
2096 	case INTEL_FAM6_WESTMERE:
2097 	case INTEL_FAM6_SANDYBRIDGE:
2098 	case INTEL_FAM6_IVYBRIDGE:
2099 	case INTEL_FAM6_HASWELL:
2100 	case INTEL_FAM6_HASWELL_L:
2101 	case INTEL_FAM6_HASWELL_G:
2102 	case INTEL_FAM6_BROADWELL:
2103 	case INTEL_FAM6_BROADWELL_G:
2104 	case INTEL_FAM6_SKYLAKE_L:
2105 	case INTEL_FAM6_SKYLAKE:
2106 	case INTEL_FAM6_KABYLAKE_L:
2107 	case INTEL_FAM6_KABYLAKE:
2108 		if (c->x86_cache_bits < 44)
2109 			c->x86_cache_bits = 44;
2110 		break;
2111 	}
2112 }
2113 
2114 static void __init l1tf_select_mitigation(void)
2115 {
2116 	u64 half_pa;
2117 
2118 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2119 		return;
2120 
2121 	if (cpu_mitigations_off())
2122 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2123 	else if (cpu_mitigations_auto_nosmt())
2124 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2125 
2126 	override_cache_bits(&boot_cpu_data);
2127 
2128 	switch (l1tf_mitigation) {
2129 	case L1TF_MITIGATION_OFF:
2130 	case L1TF_MITIGATION_FLUSH_NOWARN:
2131 	case L1TF_MITIGATION_FLUSH:
2132 		break;
2133 	case L1TF_MITIGATION_FLUSH_NOSMT:
2134 	case L1TF_MITIGATION_FULL:
2135 		cpu_smt_disable(false);
2136 		break;
2137 	case L1TF_MITIGATION_FULL_FORCE:
2138 		cpu_smt_disable(true);
2139 		break;
2140 	}
2141 
2142 #if CONFIG_PGTABLE_LEVELS == 2
2143 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2144 	return;
2145 #endif
2146 
2147 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2148 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2149 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2150 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2151 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2152 				half_pa);
2153 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2154 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2155 		return;
2156 	}
2157 
2158 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2159 }
2160 
2161 static int __init l1tf_cmdline(char *str)
2162 {
2163 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2164 		return 0;
2165 
2166 	if (!str)
2167 		return -EINVAL;
2168 
2169 	if (!strcmp(str, "off"))
2170 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2171 	else if (!strcmp(str, "flush,nowarn"))
2172 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2173 	else if (!strcmp(str, "flush"))
2174 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2175 	else if (!strcmp(str, "flush,nosmt"))
2176 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2177 	else if (!strcmp(str, "full"))
2178 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2179 	else if (!strcmp(str, "full,force"))
2180 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2181 
2182 	return 0;
2183 }
2184 early_param("l1tf", l1tf_cmdline);
2185 
2186 #undef pr_fmt
2187 #define pr_fmt(fmt) fmt
2188 
2189 #ifdef CONFIG_SYSFS
2190 
2191 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2192 
2193 #if IS_ENABLED(CONFIG_KVM_INTEL)
2194 static const char * const l1tf_vmx_states[] = {
2195 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2196 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2197 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2198 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2199 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2200 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2201 };
2202 
2203 static ssize_t l1tf_show_state(char *buf)
2204 {
2205 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2206 		return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2207 
2208 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2209 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2210 	     sched_smt_active())) {
2211 		return sprintf(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2212 			       l1tf_vmx_states[l1tf_vmx_mitigation]);
2213 	}
2214 
2215 	return sprintf(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2216 		       l1tf_vmx_states[l1tf_vmx_mitigation],
2217 		       sched_smt_active() ? "vulnerable" : "disabled");
2218 }
2219 
2220 static ssize_t itlb_multihit_show_state(char *buf)
2221 {
2222 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2223 	    !boot_cpu_has(X86_FEATURE_VMX))
2224 		return sprintf(buf, "KVM: Mitigation: VMX unsupported\n");
2225 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2226 		return sprintf(buf, "KVM: Mitigation: VMX disabled\n");
2227 	else if (itlb_multihit_kvm_mitigation)
2228 		return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
2229 	else
2230 		return sprintf(buf, "KVM: Vulnerable\n");
2231 }
2232 #else
2233 static ssize_t l1tf_show_state(char *buf)
2234 {
2235 	return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2236 }
2237 
2238 static ssize_t itlb_multihit_show_state(char *buf)
2239 {
2240 	return sprintf(buf, "Processor vulnerable\n");
2241 }
2242 #endif
2243 
2244 static ssize_t mds_show_state(char *buf)
2245 {
2246 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2247 		return sprintf(buf, "%s; SMT Host state unknown\n",
2248 			       mds_strings[mds_mitigation]);
2249 	}
2250 
2251 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2252 		return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2253 			       (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2254 			        sched_smt_active() ? "mitigated" : "disabled"));
2255 	}
2256 
2257 	return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2258 		       sched_smt_active() ? "vulnerable" : "disabled");
2259 }
2260 
2261 static ssize_t tsx_async_abort_show_state(char *buf)
2262 {
2263 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2264 	    (taa_mitigation == TAA_MITIGATION_OFF))
2265 		return sprintf(buf, "%s\n", taa_strings[taa_mitigation]);
2266 
2267 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2268 		return sprintf(buf, "%s; SMT Host state unknown\n",
2269 			       taa_strings[taa_mitigation]);
2270 	}
2271 
2272 	return sprintf(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2273 		       sched_smt_active() ? "vulnerable" : "disabled");
2274 }
2275 
2276 static ssize_t mmio_stale_data_show_state(char *buf)
2277 {
2278 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2279 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2280 
2281 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2282 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2283 				  mmio_strings[mmio_mitigation]);
2284 	}
2285 
2286 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2287 			  sched_smt_active() ? "vulnerable" : "disabled");
2288 }
2289 
2290 static char *stibp_state(void)
2291 {
2292 	if (spectre_v2_in_ibrs_mode(spectre_v2_enabled))
2293 		return "";
2294 
2295 	switch (spectre_v2_user_stibp) {
2296 	case SPECTRE_V2_USER_NONE:
2297 		return ", STIBP: disabled";
2298 	case SPECTRE_V2_USER_STRICT:
2299 		return ", STIBP: forced";
2300 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2301 		return ", STIBP: always-on";
2302 	case SPECTRE_V2_USER_PRCTL:
2303 	case SPECTRE_V2_USER_SECCOMP:
2304 		if (static_key_enabled(&switch_to_cond_stibp))
2305 			return ", STIBP: conditional";
2306 	}
2307 	return "";
2308 }
2309 
2310 static char *ibpb_state(void)
2311 {
2312 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2313 		if (static_key_enabled(&switch_mm_always_ibpb))
2314 			return ", IBPB: always-on";
2315 		if (static_key_enabled(&switch_mm_cond_ibpb))
2316 			return ", IBPB: conditional";
2317 		return ", IBPB: disabled";
2318 	}
2319 	return "";
2320 }
2321 
2322 static char *pbrsb_eibrs_state(void)
2323 {
2324 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2325 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2326 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2327 			return ", PBRSB-eIBRS: SW sequence";
2328 		else
2329 			return ", PBRSB-eIBRS: Vulnerable";
2330 	} else {
2331 		return ", PBRSB-eIBRS: Not affected";
2332 	}
2333 }
2334 
2335 static ssize_t spectre_v2_show_state(char *buf)
2336 {
2337 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2338 		return sprintf(buf, "Vulnerable: LFENCE\n");
2339 
2340 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2341 		return sprintf(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2342 
2343 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2344 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2345 		return sprintf(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2346 
2347 	return sprintf(buf, "%s%s%s%s%s%s%s\n",
2348 		       spectre_v2_strings[spectre_v2_enabled],
2349 		       ibpb_state(),
2350 		       boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2351 		       stibp_state(),
2352 		       boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2353 		       pbrsb_eibrs_state(),
2354 		       spectre_v2_module_string());
2355 }
2356 
2357 static ssize_t srbds_show_state(char *buf)
2358 {
2359 	return sprintf(buf, "%s\n", srbds_strings[srbds_mitigation]);
2360 }
2361 
2362 static ssize_t retbleed_show_state(char *buf)
2363 {
2364 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2365 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2366 	    if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2367 		boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2368 		    return sprintf(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2369 
2370 	    return sprintf(buf, "%s; SMT %s\n",
2371 			   retbleed_strings[retbleed_mitigation],
2372 			   !sched_smt_active() ? "disabled" :
2373 			   spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2374 			   spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2375 			   "enabled with STIBP protection" : "vulnerable");
2376 	}
2377 
2378 	return sprintf(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2379 }
2380 
2381 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2382 			       char *buf, unsigned int bug)
2383 {
2384 	if (!boot_cpu_has_bug(bug))
2385 		return sprintf(buf, "Not affected\n");
2386 
2387 	switch (bug) {
2388 	case X86_BUG_CPU_MELTDOWN:
2389 		if (boot_cpu_has(X86_FEATURE_PTI))
2390 			return sprintf(buf, "Mitigation: PTI\n");
2391 
2392 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2393 			return sprintf(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2394 
2395 		break;
2396 
2397 	case X86_BUG_SPECTRE_V1:
2398 		return sprintf(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2399 
2400 	case X86_BUG_SPECTRE_V2:
2401 		return spectre_v2_show_state(buf);
2402 
2403 	case X86_BUG_SPEC_STORE_BYPASS:
2404 		return sprintf(buf, "%s\n", ssb_strings[ssb_mode]);
2405 
2406 	case X86_BUG_L1TF:
2407 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2408 			return l1tf_show_state(buf);
2409 		break;
2410 
2411 	case X86_BUG_MDS:
2412 		return mds_show_state(buf);
2413 
2414 	case X86_BUG_TAA:
2415 		return tsx_async_abort_show_state(buf);
2416 
2417 	case X86_BUG_ITLB_MULTIHIT:
2418 		return itlb_multihit_show_state(buf);
2419 
2420 	case X86_BUG_SRBDS:
2421 		return srbds_show_state(buf);
2422 
2423 	case X86_BUG_MMIO_STALE_DATA:
2424 		return mmio_stale_data_show_state(buf);
2425 
2426 	case X86_BUG_RETBLEED:
2427 		return retbleed_show_state(buf);
2428 
2429 	default:
2430 		break;
2431 	}
2432 
2433 	return sprintf(buf, "Vulnerable\n");
2434 }
2435 
2436 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2437 {
2438 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2439 }
2440 
2441 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2442 {
2443 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2444 }
2445 
2446 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2447 {
2448 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2449 }
2450 
2451 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2452 {
2453 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2454 }
2455 
2456 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2457 {
2458 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2459 }
2460 
2461 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2462 {
2463 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2464 }
2465 
2466 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2467 {
2468 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2469 }
2470 
2471 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2472 {
2473 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2474 }
2475 
2476 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2477 {
2478 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2479 }
2480 
2481 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2482 {
2483 	return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2484 }
2485 
2486 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2487 {
2488 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2489 }
2490 #endif
2491