xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision 9a234a2a)
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, since Intel uses
200  * MSR_IA32_SPEC_CTRL for SSBD.
201  */
202 void
203 x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl, bool setguest)
204 {
205 	u64 guestval, hostval;
206 	struct thread_info *ti = current_thread_info();
207 
208 	/*
209 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
210 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
211 	 */
212 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
213 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
214 		return;
215 
216 	/*
217 	 * If the host has SSBD mitigation enabled, force it in the host's
218 	 * virtual MSR value. If its not permanently enabled, evaluate
219 	 * current's TIF_SSBD thread flag.
220 	 */
221 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
222 		hostval = SPEC_CTRL_SSBD;
223 	else
224 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
225 
226 	/* Sanitize the guest value */
227 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
228 
229 	if (hostval != guestval) {
230 		unsigned long tif;
231 
232 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
233 				 ssbd_spec_ctrl_to_tif(hostval);
234 
235 		speculation_ctrl_update(tif);
236 	}
237 }
238 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
239 
240 static void x86_amd_ssb_disable(void)
241 {
242 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
243 
244 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
245 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
246 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
247 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
248 }
249 
250 #undef pr_fmt
251 #define pr_fmt(fmt)	"MDS: " fmt
252 
253 /* Default mitigation for MDS-affected CPUs */
254 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
255 static bool mds_nosmt __ro_after_init = false;
256 
257 static const char * const mds_strings[] = {
258 	[MDS_MITIGATION_OFF]	= "Vulnerable",
259 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
260 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
261 };
262 
263 static void __init mds_select_mitigation(void)
264 {
265 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
266 		mds_mitigation = MDS_MITIGATION_OFF;
267 		return;
268 	}
269 
270 	if (mds_mitigation == MDS_MITIGATION_FULL) {
271 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
272 			mds_mitigation = MDS_MITIGATION_VMWERV;
273 
274 		static_branch_enable(&mds_user_clear);
275 
276 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
277 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
278 			cpu_smt_disable(false);
279 	}
280 }
281 
282 static int __init mds_cmdline(char *str)
283 {
284 	if (!boot_cpu_has_bug(X86_BUG_MDS))
285 		return 0;
286 
287 	if (!str)
288 		return -EINVAL;
289 
290 	if (!strcmp(str, "off"))
291 		mds_mitigation = MDS_MITIGATION_OFF;
292 	else if (!strcmp(str, "full"))
293 		mds_mitigation = MDS_MITIGATION_FULL;
294 	else if (!strcmp(str, "full,nosmt")) {
295 		mds_mitigation = MDS_MITIGATION_FULL;
296 		mds_nosmt = true;
297 	}
298 
299 	return 0;
300 }
301 early_param("mds", mds_cmdline);
302 
303 #undef pr_fmt
304 #define pr_fmt(fmt)	"TAA: " fmt
305 
306 enum taa_mitigations {
307 	TAA_MITIGATION_OFF,
308 	TAA_MITIGATION_UCODE_NEEDED,
309 	TAA_MITIGATION_VERW,
310 	TAA_MITIGATION_TSX_DISABLED,
311 };
312 
313 /* Default mitigation for TAA-affected CPUs */
314 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
315 static bool taa_nosmt __ro_after_init;
316 
317 static const char * const taa_strings[] = {
318 	[TAA_MITIGATION_OFF]		= "Vulnerable",
319 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
320 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
321 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
322 };
323 
324 static void __init taa_select_mitigation(void)
325 {
326 	u64 ia32_cap;
327 
328 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
329 		taa_mitigation = TAA_MITIGATION_OFF;
330 		return;
331 	}
332 
333 	/* TSX previously disabled by tsx=off */
334 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
335 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
336 		return;
337 	}
338 
339 	if (cpu_mitigations_off()) {
340 		taa_mitigation = TAA_MITIGATION_OFF;
341 		return;
342 	}
343 
344 	/*
345 	 * TAA mitigation via VERW is turned off if both
346 	 * tsx_async_abort=off and mds=off are specified.
347 	 */
348 	if (taa_mitigation == TAA_MITIGATION_OFF &&
349 	    mds_mitigation == MDS_MITIGATION_OFF)
350 		return;
351 
352 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
353 		taa_mitigation = TAA_MITIGATION_VERW;
354 	else
355 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
356 
357 	/*
358 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
359 	 * A microcode update fixes this behavior to clear CPU buffers. It also
360 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
361 	 * ARCH_CAP_TSX_CTRL_MSR bit.
362 	 *
363 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
364 	 * update is required.
365 	 */
366 	ia32_cap = x86_read_arch_cap_msr();
367 	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
368 	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
369 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
370 
371 	/*
372 	 * TSX is enabled, select alternate mitigation for TAA which is
373 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
374 	 *
375 	 * For guests that can't determine whether the correct microcode is
376 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
377 	 */
378 	static_branch_enable(&mds_user_clear);
379 
380 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
381 		cpu_smt_disable(false);
382 }
383 
384 static int __init tsx_async_abort_parse_cmdline(char *str)
385 {
386 	if (!boot_cpu_has_bug(X86_BUG_TAA))
387 		return 0;
388 
389 	if (!str)
390 		return -EINVAL;
391 
392 	if (!strcmp(str, "off")) {
393 		taa_mitigation = TAA_MITIGATION_OFF;
394 	} else if (!strcmp(str, "full")) {
395 		taa_mitigation = TAA_MITIGATION_VERW;
396 	} else if (!strcmp(str, "full,nosmt")) {
397 		taa_mitigation = TAA_MITIGATION_VERW;
398 		taa_nosmt = true;
399 	}
400 
401 	return 0;
402 }
403 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
404 
405 #undef pr_fmt
406 #define pr_fmt(fmt)	"MMIO Stale Data: " fmt
407 
408 enum mmio_mitigations {
409 	MMIO_MITIGATION_OFF,
410 	MMIO_MITIGATION_UCODE_NEEDED,
411 	MMIO_MITIGATION_VERW,
412 };
413 
414 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
415 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
416 static bool mmio_nosmt __ro_after_init = false;
417 
418 static const char * const mmio_strings[] = {
419 	[MMIO_MITIGATION_OFF]		= "Vulnerable",
420 	[MMIO_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
421 	[MMIO_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
422 };
423 
424 static void __init mmio_select_mitigation(void)
425 {
426 	u64 ia32_cap;
427 
428 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
429 	     boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
430 	     cpu_mitigations_off()) {
431 		mmio_mitigation = MMIO_MITIGATION_OFF;
432 		return;
433 	}
434 
435 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
436 		return;
437 
438 	ia32_cap = x86_read_arch_cap_msr();
439 
440 	/*
441 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
442 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
443 	 */
444 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
445 					      boot_cpu_has(X86_FEATURE_RTM)))
446 		static_branch_enable(&mds_user_clear);
447 	else
448 		static_branch_enable(&mmio_stale_data_clear);
449 
450 	/*
451 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
452 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
453 	 * is required irrespective of SMT state.
454 	 */
455 	if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
456 		static_branch_enable(&mds_idle_clear);
457 
458 	/*
459 	 * Check if the system has the right microcode.
460 	 *
461 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
462 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
463 	 * affected systems.
464 	 */
465 	if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
466 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
467 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
468 	     !(ia32_cap & ARCH_CAP_MDS_NO)))
469 		mmio_mitigation = MMIO_MITIGATION_VERW;
470 	else
471 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
472 
473 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
474 		cpu_smt_disable(false);
475 }
476 
477 static int __init mmio_stale_data_parse_cmdline(char *str)
478 {
479 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
480 		return 0;
481 
482 	if (!str)
483 		return -EINVAL;
484 
485 	if (!strcmp(str, "off")) {
486 		mmio_mitigation = MMIO_MITIGATION_OFF;
487 	} else if (!strcmp(str, "full")) {
488 		mmio_mitigation = MMIO_MITIGATION_VERW;
489 	} else if (!strcmp(str, "full,nosmt")) {
490 		mmio_mitigation = MMIO_MITIGATION_VERW;
491 		mmio_nosmt = true;
492 	}
493 
494 	return 0;
495 }
496 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
497 
498 #undef pr_fmt
499 #define pr_fmt(fmt)     "" fmt
500 
501 static void __init md_clear_update_mitigation(void)
502 {
503 	if (cpu_mitigations_off())
504 		return;
505 
506 	if (!static_key_enabled(&mds_user_clear))
507 		goto out;
508 
509 	/*
510 	 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
511 	 * mitigation, if necessary.
512 	 */
513 	if (mds_mitigation == MDS_MITIGATION_OFF &&
514 	    boot_cpu_has_bug(X86_BUG_MDS)) {
515 		mds_mitigation = MDS_MITIGATION_FULL;
516 		mds_select_mitigation();
517 	}
518 	if (taa_mitigation == TAA_MITIGATION_OFF &&
519 	    boot_cpu_has_bug(X86_BUG_TAA)) {
520 		taa_mitigation = TAA_MITIGATION_VERW;
521 		taa_select_mitigation();
522 	}
523 	if (mmio_mitigation == MMIO_MITIGATION_OFF &&
524 	    boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
525 		mmio_mitigation = MMIO_MITIGATION_VERW;
526 		mmio_select_mitigation();
527 	}
528 out:
529 	if (boot_cpu_has_bug(X86_BUG_MDS))
530 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
531 	if (boot_cpu_has_bug(X86_BUG_TAA))
532 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
533 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
534 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
535 	else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
536 		pr_info("MMIO Stale Data: Unknown: No mitigations\n");
537 }
538 
539 static void __init md_clear_select_mitigation(void)
540 {
541 	mds_select_mitigation();
542 	taa_select_mitigation();
543 	mmio_select_mitigation();
544 
545 	/*
546 	 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
547 	 * and print their mitigation after MDS, TAA and MMIO Stale Data
548 	 * mitigation selection is done.
549 	 */
550 	md_clear_update_mitigation();
551 }
552 
553 #undef pr_fmt
554 #define pr_fmt(fmt)	"SRBDS: " fmt
555 
556 enum srbds_mitigations {
557 	SRBDS_MITIGATION_OFF,
558 	SRBDS_MITIGATION_UCODE_NEEDED,
559 	SRBDS_MITIGATION_FULL,
560 	SRBDS_MITIGATION_TSX_OFF,
561 	SRBDS_MITIGATION_HYPERVISOR,
562 };
563 
564 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
565 
566 static const char * const srbds_strings[] = {
567 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
568 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
569 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
570 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
571 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
572 };
573 
574 static bool srbds_off;
575 
576 void update_srbds_msr(void)
577 {
578 	u64 mcu_ctrl;
579 
580 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
581 		return;
582 
583 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
584 		return;
585 
586 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
587 		return;
588 
589 	/*
590 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
591 	 * being disabled and it hasn't received the SRBDS MSR microcode.
592 	 */
593 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
594 		return;
595 
596 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
597 
598 	switch (srbds_mitigation) {
599 	case SRBDS_MITIGATION_OFF:
600 	case SRBDS_MITIGATION_TSX_OFF:
601 		mcu_ctrl |= RNGDS_MITG_DIS;
602 		break;
603 	case SRBDS_MITIGATION_FULL:
604 		mcu_ctrl &= ~RNGDS_MITG_DIS;
605 		break;
606 	default:
607 		break;
608 	}
609 
610 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
611 }
612 
613 static void __init srbds_select_mitigation(void)
614 {
615 	u64 ia32_cap;
616 
617 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
618 		return;
619 
620 	/*
621 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
622 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
623 	 * by Processor MMIO Stale Data vulnerability.
624 	 */
625 	ia32_cap = x86_read_arch_cap_msr();
626 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
627 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
628 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
629 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
630 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
631 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
632 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
633 	else if (cpu_mitigations_off() || srbds_off)
634 		srbds_mitigation = SRBDS_MITIGATION_OFF;
635 
636 	update_srbds_msr();
637 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
638 }
639 
640 static int __init srbds_parse_cmdline(char *str)
641 {
642 	if (!str)
643 		return -EINVAL;
644 
645 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
646 		return 0;
647 
648 	srbds_off = !strcmp(str, "off");
649 	return 0;
650 }
651 early_param("srbds", srbds_parse_cmdline);
652 
653 #undef pr_fmt
654 #define pr_fmt(fmt)     "L1D Flush : " fmt
655 
656 enum l1d_flush_mitigations {
657 	L1D_FLUSH_OFF = 0,
658 	L1D_FLUSH_ON,
659 };
660 
661 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
662 
663 static void __init l1d_flush_select_mitigation(void)
664 {
665 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
666 		return;
667 
668 	static_branch_enable(&switch_mm_cond_l1d_flush);
669 	pr_info("Conditional flush on switch_mm() enabled\n");
670 }
671 
672 static int __init l1d_flush_parse_cmdline(char *str)
673 {
674 	if (!strcmp(str, "on"))
675 		l1d_flush_mitigation = L1D_FLUSH_ON;
676 
677 	return 0;
678 }
679 early_param("l1d_flush", l1d_flush_parse_cmdline);
680 
681 #undef pr_fmt
682 #define pr_fmt(fmt)     "Spectre V1 : " fmt
683 
684 enum spectre_v1_mitigation {
685 	SPECTRE_V1_MITIGATION_NONE,
686 	SPECTRE_V1_MITIGATION_AUTO,
687 };
688 
689 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
690 	SPECTRE_V1_MITIGATION_AUTO;
691 
692 static const char * const spectre_v1_strings[] = {
693 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
694 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
695 };
696 
697 /*
698  * Does SMAP provide full mitigation against speculative kernel access to
699  * userspace?
700  */
701 static bool smap_works_speculatively(void)
702 {
703 	if (!boot_cpu_has(X86_FEATURE_SMAP))
704 		return false;
705 
706 	/*
707 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
708 	 * prevent speculative access to user data in the L1 cache.
709 	 * Consider SMAP to be non-functional as a mitigation on these
710 	 * CPUs.
711 	 */
712 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
713 		return false;
714 
715 	return true;
716 }
717 
718 static void __init spectre_v1_select_mitigation(void)
719 {
720 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
721 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
722 		return;
723 	}
724 
725 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
726 		/*
727 		 * With Spectre v1, a user can speculatively control either
728 		 * path of a conditional swapgs with a user-controlled GS
729 		 * value.  The mitigation is to add lfences to both code paths.
730 		 *
731 		 * If FSGSBASE is enabled, the user can put a kernel address in
732 		 * GS, in which case SMAP provides no protection.
733 		 *
734 		 * If FSGSBASE is disabled, the user can only put a user space
735 		 * address in GS.  That makes an attack harder, but still
736 		 * possible if there's no SMAP protection.
737 		 */
738 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
739 		    !smap_works_speculatively()) {
740 			/*
741 			 * Mitigation can be provided from SWAPGS itself or
742 			 * PTI as the CR3 write in the Meltdown mitigation
743 			 * is serializing.
744 			 *
745 			 * If neither is there, mitigate with an LFENCE to
746 			 * stop speculation through swapgs.
747 			 */
748 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
749 			    !boot_cpu_has(X86_FEATURE_PTI))
750 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
751 
752 			/*
753 			 * Enable lfences in the kernel entry (non-swapgs)
754 			 * paths, to prevent user entry from speculatively
755 			 * skipping swapgs.
756 			 */
757 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
758 		}
759 	}
760 
761 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
762 }
763 
764 static int __init nospectre_v1_cmdline(char *str)
765 {
766 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
767 	return 0;
768 }
769 early_param("nospectre_v1", nospectre_v1_cmdline);
770 
771 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
772 	SPECTRE_V2_NONE;
773 
774 #undef pr_fmt
775 #define pr_fmt(fmt)     "RETBleed: " fmt
776 
777 enum retbleed_mitigation {
778 	RETBLEED_MITIGATION_NONE,
779 	RETBLEED_MITIGATION_UNRET,
780 	RETBLEED_MITIGATION_IBPB,
781 	RETBLEED_MITIGATION_IBRS,
782 	RETBLEED_MITIGATION_EIBRS,
783 };
784 
785 enum retbleed_mitigation_cmd {
786 	RETBLEED_CMD_OFF,
787 	RETBLEED_CMD_AUTO,
788 	RETBLEED_CMD_UNRET,
789 	RETBLEED_CMD_IBPB,
790 };
791 
792 static const char * const retbleed_strings[] = {
793 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
794 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
795 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
796 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
797 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
798 };
799 
800 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
801 	RETBLEED_MITIGATION_NONE;
802 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
803 	RETBLEED_CMD_AUTO;
804 
805 static int __ro_after_init retbleed_nosmt = false;
806 
807 static int __init retbleed_parse_cmdline(char *str)
808 {
809 	if (!str)
810 		return -EINVAL;
811 
812 	while (str) {
813 		char *next = strchr(str, ',');
814 		if (next) {
815 			*next = 0;
816 			next++;
817 		}
818 
819 		if (!strcmp(str, "off")) {
820 			retbleed_cmd = RETBLEED_CMD_OFF;
821 		} else if (!strcmp(str, "auto")) {
822 			retbleed_cmd = RETBLEED_CMD_AUTO;
823 		} else if (!strcmp(str, "unret")) {
824 			retbleed_cmd = RETBLEED_CMD_UNRET;
825 		} else if (!strcmp(str, "ibpb")) {
826 			retbleed_cmd = RETBLEED_CMD_IBPB;
827 		} else if (!strcmp(str, "nosmt")) {
828 			retbleed_nosmt = true;
829 		} else {
830 			pr_err("Ignoring unknown retbleed option (%s).", str);
831 		}
832 
833 		str = next;
834 	}
835 
836 	return 0;
837 }
838 early_param("retbleed", retbleed_parse_cmdline);
839 
840 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
841 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
842 
843 static void __init retbleed_select_mitigation(void)
844 {
845 	bool mitigate_smt = false;
846 
847 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
848 		return;
849 
850 	switch (retbleed_cmd) {
851 	case RETBLEED_CMD_OFF:
852 		return;
853 
854 	case RETBLEED_CMD_UNRET:
855 		if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
856 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
857 		} else {
858 			pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
859 			goto do_cmd_auto;
860 		}
861 		break;
862 
863 	case RETBLEED_CMD_IBPB:
864 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
865 			pr_err("WARNING: CPU does not support IBPB.\n");
866 			goto do_cmd_auto;
867 		} else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
868 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
869 		} else {
870 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
871 			goto do_cmd_auto;
872 		}
873 		break;
874 
875 do_cmd_auto:
876 	case RETBLEED_CMD_AUTO:
877 	default:
878 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
879 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
880 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
881 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
882 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
883 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
884 		}
885 
886 		/*
887 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
888 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
889 		 * be set accordingly below.
890 		 */
891 
892 		break;
893 	}
894 
895 	switch (retbleed_mitigation) {
896 	case RETBLEED_MITIGATION_UNRET:
897 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
898 		setup_force_cpu_cap(X86_FEATURE_UNRET);
899 
900 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
901 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
902 			pr_err(RETBLEED_UNTRAIN_MSG);
903 
904 		mitigate_smt = true;
905 		break;
906 
907 	case RETBLEED_MITIGATION_IBPB:
908 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
909 		mitigate_smt = true;
910 		break;
911 
912 	default:
913 		break;
914 	}
915 
916 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
917 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
918 		cpu_smt_disable(false);
919 
920 	/*
921 	 * Let IBRS trump all on Intel without affecting the effects of the
922 	 * retbleed= cmdline option.
923 	 */
924 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
925 		switch (spectre_v2_enabled) {
926 		case SPECTRE_V2_IBRS:
927 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
928 			break;
929 		case SPECTRE_V2_EIBRS:
930 		case SPECTRE_V2_EIBRS_RETPOLINE:
931 		case SPECTRE_V2_EIBRS_LFENCE:
932 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
933 			break;
934 		default:
935 			pr_err(RETBLEED_INTEL_MSG);
936 		}
937 	}
938 
939 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
940 }
941 
942 #undef pr_fmt
943 #define pr_fmt(fmt)     "Spectre V2 : " fmt
944 
945 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
946 	SPECTRE_V2_USER_NONE;
947 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
948 	SPECTRE_V2_USER_NONE;
949 
950 #ifdef CONFIG_RETPOLINE
951 static bool spectre_v2_bad_module;
952 
953 bool retpoline_module_ok(bool has_retpoline)
954 {
955 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
956 		return true;
957 
958 	pr_err("System may be vulnerable to spectre v2\n");
959 	spectre_v2_bad_module = true;
960 	return false;
961 }
962 
963 static inline const char *spectre_v2_module_string(void)
964 {
965 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
966 }
967 #else
968 static inline const char *spectre_v2_module_string(void) { return ""; }
969 #endif
970 
971 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
972 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
973 #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"
974 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
975 
976 #ifdef CONFIG_BPF_SYSCALL
977 void unpriv_ebpf_notify(int new_state)
978 {
979 	if (new_state)
980 		return;
981 
982 	/* Unprivileged eBPF is enabled */
983 
984 	switch (spectre_v2_enabled) {
985 	case SPECTRE_V2_EIBRS:
986 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
987 		break;
988 	case SPECTRE_V2_EIBRS_LFENCE:
989 		if (sched_smt_active())
990 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
991 		break;
992 	default:
993 		break;
994 	}
995 }
996 #endif
997 
998 static inline bool match_option(const char *arg, int arglen, const char *opt)
999 {
1000 	int len = strlen(opt);
1001 
1002 	return len == arglen && !strncmp(arg, opt, len);
1003 }
1004 
1005 /* The kernel command line selection for spectre v2 */
1006 enum spectre_v2_mitigation_cmd {
1007 	SPECTRE_V2_CMD_NONE,
1008 	SPECTRE_V2_CMD_AUTO,
1009 	SPECTRE_V2_CMD_FORCE,
1010 	SPECTRE_V2_CMD_RETPOLINE,
1011 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1012 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1013 	SPECTRE_V2_CMD_EIBRS,
1014 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1015 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1016 	SPECTRE_V2_CMD_IBRS,
1017 };
1018 
1019 enum spectre_v2_user_cmd {
1020 	SPECTRE_V2_USER_CMD_NONE,
1021 	SPECTRE_V2_USER_CMD_AUTO,
1022 	SPECTRE_V2_USER_CMD_FORCE,
1023 	SPECTRE_V2_USER_CMD_PRCTL,
1024 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1025 	SPECTRE_V2_USER_CMD_SECCOMP,
1026 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1027 };
1028 
1029 static const char * const spectre_v2_user_strings[] = {
1030 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1031 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1032 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1033 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1034 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1035 };
1036 
1037 static const struct {
1038 	const char			*option;
1039 	enum spectre_v2_user_cmd	cmd;
1040 	bool				secure;
1041 } v2_user_options[] __initconst = {
1042 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1043 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1044 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1045 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1046 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1047 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1048 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1049 };
1050 
1051 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1052 {
1053 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1054 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1055 }
1056 
1057 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1058 
1059 static enum spectre_v2_user_cmd __init
1060 spectre_v2_parse_user_cmdline(void)
1061 {
1062 	char arg[20];
1063 	int ret, i;
1064 
1065 	switch (spectre_v2_cmd) {
1066 	case SPECTRE_V2_CMD_NONE:
1067 		return SPECTRE_V2_USER_CMD_NONE;
1068 	case SPECTRE_V2_CMD_FORCE:
1069 		return SPECTRE_V2_USER_CMD_FORCE;
1070 	default:
1071 		break;
1072 	}
1073 
1074 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1075 				  arg, sizeof(arg));
1076 	if (ret < 0)
1077 		return SPECTRE_V2_USER_CMD_AUTO;
1078 
1079 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1080 		if (match_option(arg, ret, v2_user_options[i].option)) {
1081 			spec_v2_user_print_cond(v2_user_options[i].option,
1082 						v2_user_options[i].secure);
1083 			return v2_user_options[i].cmd;
1084 		}
1085 	}
1086 
1087 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1088 	return SPECTRE_V2_USER_CMD_AUTO;
1089 }
1090 
1091 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1092 {
1093 	return mode == SPECTRE_V2_IBRS ||
1094 	       mode == SPECTRE_V2_EIBRS ||
1095 	       mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1096 	       mode == SPECTRE_V2_EIBRS_LFENCE;
1097 }
1098 
1099 static void __init
1100 spectre_v2_user_select_mitigation(void)
1101 {
1102 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1103 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1104 	enum spectre_v2_user_cmd cmd;
1105 
1106 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1107 		return;
1108 
1109 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1110 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1111 		smt_possible = false;
1112 
1113 	cmd = spectre_v2_parse_user_cmdline();
1114 	switch (cmd) {
1115 	case SPECTRE_V2_USER_CMD_NONE:
1116 		goto set_mode;
1117 	case SPECTRE_V2_USER_CMD_FORCE:
1118 		mode = SPECTRE_V2_USER_STRICT;
1119 		break;
1120 	case SPECTRE_V2_USER_CMD_AUTO:
1121 	case SPECTRE_V2_USER_CMD_PRCTL:
1122 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1123 		mode = SPECTRE_V2_USER_PRCTL;
1124 		break;
1125 	case SPECTRE_V2_USER_CMD_SECCOMP:
1126 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1127 		if (IS_ENABLED(CONFIG_SECCOMP))
1128 			mode = SPECTRE_V2_USER_SECCOMP;
1129 		else
1130 			mode = SPECTRE_V2_USER_PRCTL;
1131 		break;
1132 	}
1133 
1134 	/* Initialize Indirect Branch Prediction Barrier */
1135 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1136 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1137 
1138 		spectre_v2_user_ibpb = mode;
1139 		switch (cmd) {
1140 		case SPECTRE_V2_USER_CMD_FORCE:
1141 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1142 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1143 			static_branch_enable(&switch_mm_always_ibpb);
1144 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1145 			break;
1146 		case SPECTRE_V2_USER_CMD_PRCTL:
1147 		case SPECTRE_V2_USER_CMD_AUTO:
1148 		case SPECTRE_V2_USER_CMD_SECCOMP:
1149 			static_branch_enable(&switch_mm_cond_ibpb);
1150 			break;
1151 		default:
1152 			break;
1153 		}
1154 
1155 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1156 			static_key_enabled(&switch_mm_always_ibpb) ?
1157 			"always-on" : "conditional");
1158 	}
1159 
1160 	/*
1161 	 * If no STIBP, IBRS or enhanced IBRS is enabled, or SMT impossible,
1162 	 * STIBP is not required.
1163 	 */
1164 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1165 	    !smt_possible ||
1166 	    spectre_v2_in_ibrs_mode(spectre_v2_enabled))
1167 		return;
1168 
1169 	/*
1170 	 * At this point, an STIBP mode other than "off" has been set.
1171 	 * If STIBP support is not being forced, check if STIBP always-on
1172 	 * is preferred.
1173 	 */
1174 	if (mode != SPECTRE_V2_USER_STRICT &&
1175 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1176 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1177 
1178 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1179 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1180 		if (mode != SPECTRE_V2_USER_STRICT &&
1181 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1182 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1183 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1184 	}
1185 
1186 	spectre_v2_user_stibp = mode;
1187 
1188 set_mode:
1189 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1190 }
1191 
1192 static const char * const spectre_v2_strings[] = {
1193 	[SPECTRE_V2_NONE]			= "Vulnerable",
1194 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1195 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1196 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced IBRS",
1197 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced IBRS + LFENCE",
1198 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced IBRS + Retpolines",
1199 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1200 };
1201 
1202 static const struct {
1203 	const char *option;
1204 	enum spectre_v2_mitigation_cmd cmd;
1205 	bool secure;
1206 } mitigation_options[] __initconst = {
1207 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1208 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1209 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1210 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1211 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1212 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1213 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1214 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1215 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1216 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1217 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1218 };
1219 
1220 static void __init spec_v2_print_cond(const char *reason, bool secure)
1221 {
1222 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1223 		pr_info("%s selected on command line.\n", reason);
1224 }
1225 
1226 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1227 {
1228 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1229 	char arg[20];
1230 	int ret, i;
1231 
1232 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1233 	    cpu_mitigations_off())
1234 		return SPECTRE_V2_CMD_NONE;
1235 
1236 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1237 	if (ret < 0)
1238 		return SPECTRE_V2_CMD_AUTO;
1239 
1240 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1241 		if (!match_option(arg, ret, mitigation_options[i].option))
1242 			continue;
1243 		cmd = mitigation_options[i].cmd;
1244 		break;
1245 	}
1246 
1247 	if (i >= ARRAY_SIZE(mitigation_options)) {
1248 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1249 		return SPECTRE_V2_CMD_AUTO;
1250 	}
1251 
1252 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1253 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1254 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1255 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1256 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1257 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1258 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1259 		       mitigation_options[i].option);
1260 		return SPECTRE_V2_CMD_AUTO;
1261 	}
1262 
1263 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1264 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1265 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1266 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1267 		pr_err("%s selected but CPU doesn't have eIBRS. Switching to AUTO select\n",
1268 		       mitigation_options[i].option);
1269 		return SPECTRE_V2_CMD_AUTO;
1270 	}
1271 
1272 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1273 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1274 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1275 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1276 		       mitigation_options[i].option);
1277 		return SPECTRE_V2_CMD_AUTO;
1278 	}
1279 
1280 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1281 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1282 		       mitigation_options[i].option);
1283 		return SPECTRE_V2_CMD_AUTO;
1284 	}
1285 
1286 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1287 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1288 		       mitigation_options[i].option);
1289 		return SPECTRE_V2_CMD_AUTO;
1290 	}
1291 
1292 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1293 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1294 		       mitigation_options[i].option);
1295 		return SPECTRE_V2_CMD_AUTO;
1296 	}
1297 
1298 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_has(X86_FEATURE_XENPV)) {
1299 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1300 		       mitigation_options[i].option);
1301 		return SPECTRE_V2_CMD_AUTO;
1302 	}
1303 
1304 	spec_v2_print_cond(mitigation_options[i].option,
1305 			   mitigation_options[i].secure);
1306 	return cmd;
1307 }
1308 
1309 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1310 {
1311 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1312 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1313 		return SPECTRE_V2_NONE;
1314 	}
1315 
1316 	return SPECTRE_V2_RETPOLINE;
1317 }
1318 
1319 /* Disable in-kernel use of non-RSB RET predictors */
1320 static void __init spec_ctrl_disable_kernel_rrsba(void)
1321 {
1322 	u64 ia32_cap;
1323 
1324 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1325 		return;
1326 
1327 	ia32_cap = x86_read_arch_cap_msr();
1328 
1329 	if (ia32_cap & ARCH_CAP_RRSBA) {
1330 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1331 		write_spec_ctrl_current(x86_spec_ctrl_base, true);
1332 	}
1333 }
1334 
1335 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1336 {
1337 	/*
1338 	 * Similar to context switches, there are two types of RSB attacks
1339 	 * after VM exit:
1340 	 *
1341 	 * 1) RSB underflow
1342 	 *
1343 	 * 2) Poisoned RSB entry
1344 	 *
1345 	 * When retpoline is enabled, both are mitigated by filling/clearing
1346 	 * the RSB.
1347 	 *
1348 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1349 	 * prediction isolation protections, RSB still needs to be cleared
1350 	 * because of #2.  Note that SMEP provides no protection here, unlike
1351 	 * user-space-poisoned RSB entries.
1352 	 *
1353 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1354 	 * bug is present then a LITE version of RSB protection is required,
1355 	 * just a single call needs to retire before a RET is executed.
1356 	 */
1357 	switch (mode) {
1358 	case SPECTRE_V2_NONE:
1359 		return;
1360 
1361 	case SPECTRE_V2_EIBRS_LFENCE:
1362 	case SPECTRE_V2_EIBRS:
1363 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1364 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1365 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1366 		}
1367 		return;
1368 
1369 	case SPECTRE_V2_EIBRS_RETPOLINE:
1370 	case SPECTRE_V2_RETPOLINE:
1371 	case SPECTRE_V2_LFENCE:
1372 	case SPECTRE_V2_IBRS:
1373 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1374 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1375 		return;
1376 	}
1377 
1378 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1379 	dump_stack();
1380 }
1381 
1382 static void __init spectre_v2_select_mitigation(void)
1383 {
1384 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1385 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1386 
1387 	/*
1388 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1389 	 * then nothing to do.
1390 	 */
1391 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1392 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1393 		return;
1394 
1395 	switch (cmd) {
1396 	case SPECTRE_V2_CMD_NONE:
1397 		return;
1398 
1399 	case SPECTRE_V2_CMD_FORCE:
1400 	case SPECTRE_V2_CMD_AUTO:
1401 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1402 			mode = SPECTRE_V2_EIBRS;
1403 			break;
1404 		}
1405 
1406 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1407 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1408 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1409 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1410 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1411 			mode = SPECTRE_V2_IBRS;
1412 			break;
1413 		}
1414 
1415 		mode = spectre_v2_select_retpoline();
1416 		break;
1417 
1418 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1419 		pr_err(SPECTRE_V2_LFENCE_MSG);
1420 		mode = SPECTRE_V2_LFENCE;
1421 		break;
1422 
1423 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1424 		mode = SPECTRE_V2_RETPOLINE;
1425 		break;
1426 
1427 	case SPECTRE_V2_CMD_RETPOLINE:
1428 		mode = spectre_v2_select_retpoline();
1429 		break;
1430 
1431 	case SPECTRE_V2_CMD_IBRS:
1432 		mode = SPECTRE_V2_IBRS;
1433 		break;
1434 
1435 	case SPECTRE_V2_CMD_EIBRS:
1436 		mode = SPECTRE_V2_EIBRS;
1437 		break;
1438 
1439 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1440 		mode = SPECTRE_V2_EIBRS_LFENCE;
1441 		break;
1442 
1443 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1444 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1445 		break;
1446 	}
1447 
1448 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1449 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1450 
1451 	if (spectre_v2_in_ibrs_mode(mode)) {
1452 		x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1453 		write_spec_ctrl_current(x86_spec_ctrl_base, true);
1454 	}
1455 
1456 	switch (mode) {
1457 	case SPECTRE_V2_NONE:
1458 	case SPECTRE_V2_EIBRS:
1459 		break;
1460 
1461 	case SPECTRE_V2_IBRS:
1462 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1463 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1464 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1465 		break;
1466 
1467 	case SPECTRE_V2_LFENCE:
1468 	case SPECTRE_V2_EIBRS_LFENCE:
1469 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1470 		fallthrough;
1471 
1472 	case SPECTRE_V2_RETPOLINE:
1473 	case SPECTRE_V2_EIBRS_RETPOLINE:
1474 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1475 		break;
1476 	}
1477 
1478 	/*
1479 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1480 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1481 	 * prediction from a non-RSB predictor is still a risk.
1482 	 */
1483 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1484 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1485 	    mode == SPECTRE_V2_RETPOLINE)
1486 		spec_ctrl_disable_kernel_rrsba();
1487 
1488 	spectre_v2_enabled = mode;
1489 	pr_info("%s\n", spectre_v2_strings[mode]);
1490 
1491 	/*
1492 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1493 	 * context switch.  In general there are two types of RSB attacks
1494 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1495 	 *
1496 	 * 1) RSB underflow
1497 	 *
1498 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1499 	 *    speculated return targets may come from the branch predictor,
1500 	 *    which could have a user-poisoned BTB or BHB entry.
1501 	 *
1502 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1503 	 *    regardless of the state of the RSB.
1504 	 *
1505 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1506 	 *    scenario is mitigated by the IBRS branch prediction isolation
1507 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1508 	 *    protect against this type of attack.
1509 	 *
1510 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1511 	 *
1512 	 * 2) Poisoned RSB entry
1513 	 *
1514 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1515 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1516 	 *    entry.
1517 	 *
1518 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1519 	 *    eIBRS.
1520 	 *
1521 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1522 	 *    RSB clearing.
1523 	 *
1524 	 * So to mitigate all cases, unconditionally fill RSB on context
1525 	 * switches.
1526 	 *
1527 	 * FIXME: Is this pointless for retbleed-affected AMD?
1528 	 */
1529 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1530 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1531 
1532 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1533 
1534 	/*
1535 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1536 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1537 	 * firmware calls only when IBRS / Enhanced IBRS aren't otherwise
1538 	 * enabled.
1539 	 *
1540 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1541 	 * the user might select retpoline on the kernel command line and if
1542 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1543 	 * enable IBRS around firmware calls.
1544 	 */
1545 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1546 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1547 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1548 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1549 
1550 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1551 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1552 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1553 		}
1554 
1555 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1556 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1557 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1558 	}
1559 
1560 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1561 	spectre_v2_cmd = cmd;
1562 }
1563 
1564 static void update_stibp_msr(void * __unused)
1565 {
1566 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1567 	write_spec_ctrl_current(val, true);
1568 }
1569 
1570 /* Update x86_spec_ctrl_base in case SMT state changed. */
1571 static void update_stibp_strict(void)
1572 {
1573 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1574 
1575 	if (sched_smt_active())
1576 		mask |= SPEC_CTRL_STIBP;
1577 
1578 	if (mask == x86_spec_ctrl_base)
1579 		return;
1580 
1581 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1582 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1583 	x86_spec_ctrl_base = mask;
1584 	on_each_cpu(update_stibp_msr, NULL, 1);
1585 }
1586 
1587 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1588 static void update_indir_branch_cond(void)
1589 {
1590 	if (sched_smt_active())
1591 		static_branch_enable(&switch_to_cond_stibp);
1592 	else
1593 		static_branch_disable(&switch_to_cond_stibp);
1594 }
1595 
1596 #undef pr_fmt
1597 #define pr_fmt(fmt) fmt
1598 
1599 /* Update the static key controlling the MDS CPU buffer clear in idle */
1600 static void update_mds_branch_idle(void)
1601 {
1602 	u64 ia32_cap = x86_read_arch_cap_msr();
1603 
1604 	/*
1605 	 * Enable the idle clearing if SMT is active on CPUs which are
1606 	 * affected only by MSBDS and not any other MDS variant.
1607 	 *
1608 	 * The other variants cannot be mitigated when SMT is enabled, so
1609 	 * clearing the buffers on idle just to prevent the Store Buffer
1610 	 * repartitioning leak would be a window dressing exercise.
1611 	 */
1612 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1613 		return;
1614 
1615 	if (sched_smt_active()) {
1616 		static_branch_enable(&mds_idle_clear);
1617 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1618 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1619 		static_branch_disable(&mds_idle_clear);
1620 	}
1621 }
1622 
1623 #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"
1624 #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"
1625 #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"
1626 
1627 void cpu_bugs_smt_update(void)
1628 {
1629 	mutex_lock(&spec_ctrl_mutex);
1630 
1631 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1632 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1633 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1634 
1635 	switch (spectre_v2_user_stibp) {
1636 	case SPECTRE_V2_USER_NONE:
1637 		break;
1638 	case SPECTRE_V2_USER_STRICT:
1639 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1640 		update_stibp_strict();
1641 		break;
1642 	case SPECTRE_V2_USER_PRCTL:
1643 	case SPECTRE_V2_USER_SECCOMP:
1644 		update_indir_branch_cond();
1645 		break;
1646 	}
1647 
1648 	switch (mds_mitigation) {
1649 	case MDS_MITIGATION_FULL:
1650 	case MDS_MITIGATION_VMWERV:
1651 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1652 			pr_warn_once(MDS_MSG_SMT);
1653 		update_mds_branch_idle();
1654 		break;
1655 	case MDS_MITIGATION_OFF:
1656 		break;
1657 	}
1658 
1659 	switch (taa_mitigation) {
1660 	case TAA_MITIGATION_VERW:
1661 	case TAA_MITIGATION_UCODE_NEEDED:
1662 		if (sched_smt_active())
1663 			pr_warn_once(TAA_MSG_SMT);
1664 		break;
1665 	case TAA_MITIGATION_TSX_DISABLED:
1666 	case TAA_MITIGATION_OFF:
1667 		break;
1668 	}
1669 
1670 	switch (mmio_mitigation) {
1671 	case MMIO_MITIGATION_VERW:
1672 	case MMIO_MITIGATION_UCODE_NEEDED:
1673 		if (sched_smt_active())
1674 			pr_warn_once(MMIO_MSG_SMT);
1675 		break;
1676 	case MMIO_MITIGATION_OFF:
1677 		break;
1678 	}
1679 
1680 	mutex_unlock(&spec_ctrl_mutex);
1681 }
1682 
1683 #undef pr_fmt
1684 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1685 
1686 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1687 
1688 /* The kernel command line selection */
1689 enum ssb_mitigation_cmd {
1690 	SPEC_STORE_BYPASS_CMD_NONE,
1691 	SPEC_STORE_BYPASS_CMD_AUTO,
1692 	SPEC_STORE_BYPASS_CMD_ON,
1693 	SPEC_STORE_BYPASS_CMD_PRCTL,
1694 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1695 };
1696 
1697 static const char * const ssb_strings[] = {
1698 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1699 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1700 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1701 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1702 };
1703 
1704 static const struct {
1705 	const char *option;
1706 	enum ssb_mitigation_cmd cmd;
1707 } ssb_mitigation_options[]  __initconst = {
1708 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1709 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1710 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1711 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1712 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1713 };
1714 
1715 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1716 {
1717 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1718 	char arg[20];
1719 	int ret, i;
1720 
1721 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1722 	    cpu_mitigations_off()) {
1723 		return SPEC_STORE_BYPASS_CMD_NONE;
1724 	} else {
1725 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1726 					  arg, sizeof(arg));
1727 		if (ret < 0)
1728 			return SPEC_STORE_BYPASS_CMD_AUTO;
1729 
1730 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1731 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1732 				continue;
1733 
1734 			cmd = ssb_mitigation_options[i].cmd;
1735 			break;
1736 		}
1737 
1738 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1739 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1740 			return SPEC_STORE_BYPASS_CMD_AUTO;
1741 		}
1742 	}
1743 
1744 	return cmd;
1745 }
1746 
1747 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1748 {
1749 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1750 	enum ssb_mitigation_cmd cmd;
1751 
1752 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1753 		return mode;
1754 
1755 	cmd = ssb_parse_cmdline();
1756 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1757 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1758 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1759 		return mode;
1760 
1761 	switch (cmd) {
1762 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1763 		/*
1764 		 * Choose prctl+seccomp as the default mode if seccomp is
1765 		 * enabled.
1766 		 */
1767 		if (IS_ENABLED(CONFIG_SECCOMP))
1768 			mode = SPEC_STORE_BYPASS_SECCOMP;
1769 		else
1770 			mode = SPEC_STORE_BYPASS_PRCTL;
1771 		break;
1772 	case SPEC_STORE_BYPASS_CMD_ON:
1773 		mode = SPEC_STORE_BYPASS_DISABLE;
1774 		break;
1775 	case SPEC_STORE_BYPASS_CMD_AUTO:
1776 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1777 		mode = SPEC_STORE_BYPASS_PRCTL;
1778 		break;
1779 	case SPEC_STORE_BYPASS_CMD_NONE:
1780 		break;
1781 	}
1782 
1783 	/*
1784 	 * We have three CPU feature flags that are in play here:
1785 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1786 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1787 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1788 	 */
1789 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1790 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1791 		/*
1792 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1793 		 * use a completely different MSR and bit dependent on family.
1794 		 */
1795 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1796 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1797 			x86_amd_ssb_disable();
1798 		} else {
1799 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1800 			write_spec_ctrl_current(x86_spec_ctrl_base, true);
1801 		}
1802 	}
1803 
1804 	return mode;
1805 }
1806 
1807 static void ssb_select_mitigation(void)
1808 {
1809 	ssb_mode = __ssb_select_mitigation();
1810 
1811 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1812 		pr_info("%s\n", ssb_strings[ssb_mode]);
1813 }
1814 
1815 #undef pr_fmt
1816 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1817 
1818 static void task_update_spec_tif(struct task_struct *tsk)
1819 {
1820 	/* Force the update of the real TIF bits */
1821 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1822 
1823 	/*
1824 	 * Immediately update the speculation control MSRs for the current
1825 	 * task, but for a non-current task delay setting the CPU
1826 	 * mitigation until it is scheduled next.
1827 	 *
1828 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1829 	 * always the current task.
1830 	 */
1831 	if (tsk == current)
1832 		speculation_ctrl_update_current();
1833 }
1834 
1835 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1836 {
1837 
1838 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1839 		return -EPERM;
1840 
1841 	switch (ctrl) {
1842 	case PR_SPEC_ENABLE:
1843 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1844 		return 0;
1845 	case PR_SPEC_DISABLE:
1846 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1847 		return 0;
1848 	default:
1849 		return -ERANGE;
1850 	}
1851 }
1852 
1853 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1854 {
1855 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1856 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1857 		return -ENXIO;
1858 
1859 	switch (ctrl) {
1860 	case PR_SPEC_ENABLE:
1861 		/* If speculation is force disabled, enable is not allowed */
1862 		if (task_spec_ssb_force_disable(task))
1863 			return -EPERM;
1864 		task_clear_spec_ssb_disable(task);
1865 		task_clear_spec_ssb_noexec(task);
1866 		task_update_spec_tif(task);
1867 		break;
1868 	case PR_SPEC_DISABLE:
1869 		task_set_spec_ssb_disable(task);
1870 		task_clear_spec_ssb_noexec(task);
1871 		task_update_spec_tif(task);
1872 		break;
1873 	case PR_SPEC_FORCE_DISABLE:
1874 		task_set_spec_ssb_disable(task);
1875 		task_set_spec_ssb_force_disable(task);
1876 		task_clear_spec_ssb_noexec(task);
1877 		task_update_spec_tif(task);
1878 		break;
1879 	case PR_SPEC_DISABLE_NOEXEC:
1880 		if (task_spec_ssb_force_disable(task))
1881 			return -EPERM;
1882 		task_set_spec_ssb_disable(task);
1883 		task_set_spec_ssb_noexec(task);
1884 		task_update_spec_tif(task);
1885 		break;
1886 	default:
1887 		return -ERANGE;
1888 	}
1889 	return 0;
1890 }
1891 
1892 static bool is_spec_ib_user_controlled(void)
1893 {
1894 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1895 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1896 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1897 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1898 }
1899 
1900 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1901 {
1902 	switch (ctrl) {
1903 	case PR_SPEC_ENABLE:
1904 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1905 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1906 			return 0;
1907 
1908 		/*
1909 		 * With strict mode for both IBPB and STIBP, the instruction
1910 		 * code paths avoid checking this task flag and instead,
1911 		 * unconditionally run the instruction. However, STIBP and IBPB
1912 		 * are independent and either can be set to conditionally
1913 		 * enabled regardless of the mode of the other.
1914 		 *
1915 		 * If either is set to conditional, allow the task flag to be
1916 		 * updated, unless it was force-disabled by a previous prctl
1917 		 * call. Currently, this is possible on an AMD CPU which has the
1918 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1919 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1920 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1921 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1922 		 */
1923 		if (!is_spec_ib_user_controlled() ||
1924 		    task_spec_ib_force_disable(task))
1925 			return -EPERM;
1926 
1927 		task_clear_spec_ib_disable(task);
1928 		task_update_spec_tif(task);
1929 		break;
1930 	case PR_SPEC_DISABLE:
1931 	case PR_SPEC_FORCE_DISABLE:
1932 		/*
1933 		 * Indirect branch speculation is always allowed when
1934 		 * mitigation is force disabled.
1935 		 */
1936 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1937 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1938 			return -EPERM;
1939 
1940 		if (!is_spec_ib_user_controlled())
1941 			return 0;
1942 
1943 		task_set_spec_ib_disable(task);
1944 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1945 			task_set_spec_ib_force_disable(task);
1946 		task_update_spec_tif(task);
1947 		break;
1948 	default:
1949 		return -ERANGE;
1950 	}
1951 	return 0;
1952 }
1953 
1954 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1955 			     unsigned long ctrl)
1956 {
1957 	switch (which) {
1958 	case PR_SPEC_STORE_BYPASS:
1959 		return ssb_prctl_set(task, ctrl);
1960 	case PR_SPEC_INDIRECT_BRANCH:
1961 		return ib_prctl_set(task, ctrl);
1962 	case PR_SPEC_L1D_FLUSH:
1963 		return l1d_flush_prctl_set(task, ctrl);
1964 	default:
1965 		return -ENODEV;
1966 	}
1967 }
1968 
1969 #ifdef CONFIG_SECCOMP
1970 void arch_seccomp_spec_mitigate(struct task_struct *task)
1971 {
1972 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1973 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1974 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1975 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1976 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1977 }
1978 #endif
1979 
1980 static int l1d_flush_prctl_get(struct task_struct *task)
1981 {
1982 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1983 		return PR_SPEC_FORCE_DISABLE;
1984 
1985 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
1986 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1987 	else
1988 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1989 }
1990 
1991 static int ssb_prctl_get(struct task_struct *task)
1992 {
1993 	switch (ssb_mode) {
1994 	case SPEC_STORE_BYPASS_DISABLE:
1995 		return PR_SPEC_DISABLE;
1996 	case SPEC_STORE_BYPASS_SECCOMP:
1997 	case SPEC_STORE_BYPASS_PRCTL:
1998 		if (task_spec_ssb_force_disable(task))
1999 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2000 		if (task_spec_ssb_noexec(task))
2001 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2002 		if (task_spec_ssb_disable(task))
2003 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2004 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2005 	default:
2006 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2007 			return PR_SPEC_ENABLE;
2008 		return PR_SPEC_NOT_AFFECTED;
2009 	}
2010 }
2011 
2012 static int ib_prctl_get(struct task_struct *task)
2013 {
2014 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2015 		return PR_SPEC_NOT_AFFECTED;
2016 
2017 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2018 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2019 		return PR_SPEC_ENABLE;
2020 	else if (is_spec_ib_user_controlled()) {
2021 		if (task_spec_ib_force_disable(task))
2022 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2023 		if (task_spec_ib_disable(task))
2024 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2025 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2026 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2027 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2028 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2029 		return PR_SPEC_DISABLE;
2030 	else
2031 		return PR_SPEC_NOT_AFFECTED;
2032 }
2033 
2034 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2035 {
2036 	switch (which) {
2037 	case PR_SPEC_STORE_BYPASS:
2038 		return ssb_prctl_get(task);
2039 	case PR_SPEC_INDIRECT_BRANCH:
2040 		return ib_prctl_get(task);
2041 	case PR_SPEC_L1D_FLUSH:
2042 		return l1d_flush_prctl_get(task);
2043 	default:
2044 		return -ENODEV;
2045 	}
2046 }
2047 
2048 void x86_spec_ctrl_setup_ap(void)
2049 {
2050 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2051 		write_spec_ctrl_current(x86_spec_ctrl_base, true);
2052 
2053 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2054 		x86_amd_ssb_disable();
2055 }
2056 
2057 bool itlb_multihit_kvm_mitigation;
2058 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2059 
2060 #undef pr_fmt
2061 #define pr_fmt(fmt)	"L1TF: " fmt
2062 
2063 /* Default mitigation for L1TF-affected CPUs */
2064 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2065 #if IS_ENABLED(CONFIG_KVM_INTEL)
2066 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2067 #endif
2068 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2069 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2070 
2071 /*
2072  * These CPUs all support 44bits physical address space internally in the
2073  * cache but CPUID can report a smaller number of physical address bits.
2074  *
2075  * The L1TF mitigation uses the top most address bit for the inversion of
2076  * non present PTEs. When the installed memory reaches into the top most
2077  * address bit due to memory holes, which has been observed on machines
2078  * which report 36bits physical address bits and have 32G RAM installed,
2079  * then the mitigation range check in l1tf_select_mitigation() triggers.
2080  * This is a false positive because the mitigation is still possible due to
2081  * the fact that the cache uses 44bit internally. Use the cache bits
2082  * instead of the reported physical bits and adjust them on the affected
2083  * machines to 44bit if the reported bits are less than 44.
2084  */
2085 static void override_cache_bits(struct cpuinfo_x86 *c)
2086 {
2087 	if (c->x86 != 6)
2088 		return;
2089 
2090 	switch (c->x86_model) {
2091 	case INTEL_FAM6_NEHALEM:
2092 	case INTEL_FAM6_WESTMERE:
2093 	case INTEL_FAM6_SANDYBRIDGE:
2094 	case INTEL_FAM6_IVYBRIDGE:
2095 	case INTEL_FAM6_HASWELL:
2096 	case INTEL_FAM6_HASWELL_L:
2097 	case INTEL_FAM6_HASWELL_G:
2098 	case INTEL_FAM6_BROADWELL:
2099 	case INTEL_FAM6_BROADWELL_G:
2100 	case INTEL_FAM6_SKYLAKE_L:
2101 	case INTEL_FAM6_SKYLAKE:
2102 	case INTEL_FAM6_KABYLAKE_L:
2103 	case INTEL_FAM6_KABYLAKE:
2104 		if (c->x86_cache_bits < 44)
2105 			c->x86_cache_bits = 44;
2106 		break;
2107 	}
2108 }
2109 
2110 static void __init l1tf_select_mitigation(void)
2111 {
2112 	u64 half_pa;
2113 
2114 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2115 		return;
2116 
2117 	if (cpu_mitigations_off())
2118 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2119 	else if (cpu_mitigations_auto_nosmt())
2120 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2121 
2122 	override_cache_bits(&boot_cpu_data);
2123 
2124 	switch (l1tf_mitigation) {
2125 	case L1TF_MITIGATION_OFF:
2126 	case L1TF_MITIGATION_FLUSH_NOWARN:
2127 	case L1TF_MITIGATION_FLUSH:
2128 		break;
2129 	case L1TF_MITIGATION_FLUSH_NOSMT:
2130 	case L1TF_MITIGATION_FULL:
2131 		cpu_smt_disable(false);
2132 		break;
2133 	case L1TF_MITIGATION_FULL_FORCE:
2134 		cpu_smt_disable(true);
2135 		break;
2136 	}
2137 
2138 #if CONFIG_PGTABLE_LEVELS == 2
2139 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2140 	return;
2141 #endif
2142 
2143 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2144 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2145 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2146 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2147 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2148 				half_pa);
2149 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2150 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2151 		return;
2152 	}
2153 
2154 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2155 }
2156 
2157 static int __init l1tf_cmdline(char *str)
2158 {
2159 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2160 		return 0;
2161 
2162 	if (!str)
2163 		return -EINVAL;
2164 
2165 	if (!strcmp(str, "off"))
2166 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2167 	else if (!strcmp(str, "flush,nowarn"))
2168 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2169 	else if (!strcmp(str, "flush"))
2170 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2171 	else if (!strcmp(str, "flush,nosmt"))
2172 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2173 	else if (!strcmp(str, "full"))
2174 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2175 	else if (!strcmp(str, "full,force"))
2176 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2177 
2178 	return 0;
2179 }
2180 early_param("l1tf", l1tf_cmdline);
2181 
2182 #undef pr_fmt
2183 #define pr_fmt(fmt) fmt
2184 
2185 #ifdef CONFIG_SYSFS
2186 
2187 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2188 
2189 #if IS_ENABLED(CONFIG_KVM_INTEL)
2190 static const char * const l1tf_vmx_states[] = {
2191 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2192 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2193 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2194 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2195 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2196 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2197 };
2198 
2199 static ssize_t l1tf_show_state(char *buf)
2200 {
2201 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2202 		return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2203 
2204 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2205 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2206 	     sched_smt_active())) {
2207 		return sprintf(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2208 			       l1tf_vmx_states[l1tf_vmx_mitigation]);
2209 	}
2210 
2211 	return sprintf(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2212 		       l1tf_vmx_states[l1tf_vmx_mitigation],
2213 		       sched_smt_active() ? "vulnerable" : "disabled");
2214 }
2215 
2216 static ssize_t itlb_multihit_show_state(char *buf)
2217 {
2218 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2219 	    !boot_cpu_has(X86_FEATURE_VMX))
2220 		return sprintf(buf, "KVM: Mitigation: VMX unsupported\n");
2221 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2222 		return sprintf(buf, "KVM: Mitigation: VMX disabled\n");
2223 	else if (itlb_multihit_kvm_mitigation)
2224 		return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
2225 	else
2226 		return sprintf(buf, "KVM: Vulnerable\n");
2227 }
2228 #else
2229 static ssize_t l1tf_show_state(char *buf)
2230 {
2231 	return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2232 }
2233 
2234 static ssize_t itlb_multihit_show_state(char *buf)
2235 {
2236 	return sprintf(buf, "Processor vulnerable\n");
2237 }
2238 #endif
2239 
2240 static ssize_t mds_show_state(char *buf)
2241 {
2242 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2243 		return sprintf(buf, "%s; SMT Host state unknown\n",
2244 			       mds_strings[mds_mitigation]);
2245 	}
2246 
2247 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2248 		return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2249 			       (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2250 			        sched_smt_active() ? "mitigated" : "disabled"));
2251 	}
2252 
2253 	return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2254 		       sched_smt_active() ? "vulnerable" : "disabled");
2255 }
2256 
2257 static ssize_t tsx_async_abort_show_state(char *buf)
2258 {
2259 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2260 	    (taa_mitigation == TAA_MITIGATION_OFF))
2261 		return sprintf(buf, "%s\n", taa_strings[taa_mitigation]);
2262 
2263 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2264 		return sprintf(buf, "%s; SMT Host state unknown\n",
2265 			       taa_strings[taa_mitigation]);
2266 	}
2267 
2268 	return sprintf(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2269 		       sched_smt_active() ? "vulnerable" : "disabled");
2270 }
2271 
2272 static ssize_t mmio_stale_data_show_state(char *buf)
2273 {
2274 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2275 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2276 
2277 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2278 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2279 
2280 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2281 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2282 				  mmio_strings[mmio_mitigation]);
2283 	}
2284 
2285 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2286 			  sched_smt_active() ? "vulnerable" : "disabled");
2287 }
2288 
2289 static char *stibp_state(void)
2290 {
2291 	if (spectre_v2_in_ibrs_mode(spectre_v2_enabled))
2292 		return "";
2293 
2294 	switch (spectre_v2_user_stibp) {
2295 	case SPECTRE_V2_USER_NONE:
2296 		return ", STIBP: disabled";
2297 	case SPECTRE_V2_USER_STRICT:
2298 		return ", STIBP: forced";
2299 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2300 		return ", STIBP: always-on";
2301 	case SPECTRE_V2_USER_PRCTL:
2302 	case SPECTRE_V2_USER_SECCOMP:
2303 		if (static_key_enabled(&switch_to_cond_stibp))
2304 			return ", STIBP: conditional";
2305 	}
2306 	return "";
2307 }
2308 
2309 static char *ibpb_state(void)
2310 {
2311 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2312 		if (static_key_enabled(&switch_mm_always_ibpb))
2313 			return ", IBPB: always-on";
2314 		if (static_key_enabled(&switch_mm_cond_ibpb))
2315 			return ", IBPB: conditional";
2316 		return ", IBPB: disabled";
2317 	}
2318 	return "";
2319 }
2320 
2321 static char *pbrsb_eibrs_state(void)
2322 {
2323 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2324 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2325 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2326 			return ", PBRSB-eIBRS: SW sequence";
2327 		else
2328 			return ", PBRSB-eIBRS: Vulnerable";
2329 	} else {
2330 		return ", PBRSB-eIBRS: Not affected";
2331 	}
2332 }
2333 
2334 static ssize_t spectre_v2_show_state(char *buf)
2335 {
2336 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2337 		return sprintf(buf, "Vulnerable: LFENCE\n");
2338 
2339 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2340 		return sprintf(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2341 
2342 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2343 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2344 		return sprintf(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2345 
2346 	return sprintf(buf, "%s%s%s%s%s%s%s\n",
2347 		       spectre_v2_strings[spectre_v2_enabled],
2348 		       ibpb_state(),
2349 		       boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2350 		       stibp_state(),
2351 		       boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2352 		       pbrsb_eibrs_state(),
2353 		       spectre_v2_module_string());
2354 }
2355 
2356 static ssize_t srbds_show_state(char *buf)
2357 {
2358 	return sprintf(buf, "%s\n", srbds_strings[srbds_mitigation]);
2359 }
2360 
2361 static ssize_t retbleed_show_state(char *buf)
2362 {
2363 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2364 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2365 	    if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2366 		boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2367 		    return sprintf(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2368 
2369 	    return sprintf(buf, "%s; SMT %s\n",
2370 			   retbleed_strings[retbleed_mitigation],
2371 			   !sched_smt_active() ? "disabled" :
2372 			   spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2373 			   spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2374 			   "enabled with STIBP protection" : "vulnerable");
2375 	}
2376 
2377 	return sprintf(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2378 }
2379 
2380 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2381 			       char *buf, unsigned int bug)
2382 {
2383 	if (!boot_cpu_has_bug(bug))
2384 		return sprintf(buf, "Not affected\n");
2385 
2386 	switch (bug) {
2387 	case X86_BUG_CPU_MELTDOWN:
2388 		if (boot_cpu_has(X86_FEATURE_PTI))
2389 			return sprintf(buf, "Mitigation: PTI\n");
2390 
2391 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2392 			return sprintf(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2393 
2394 		break;
2395 
2396 	case X86_BUG_SPECTRE_V1:
2397 		return sprintf(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2398 
2399 	case X86_BUG_SPECTRE_V2:
2400 		return spectre_v2_show_state(buf);
2401 
2402 	case X86_BUG_SPEC_STORE_BYPASS:
2403 		return sprintf(buf, "%s\n", ssb_strings[ssb_mode]);
2404 
2405 	case X86_BUG_L1TF:
2406 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2407 			return l1tf_show_state(buf);
2408 		break;
2409 
2410 	case X86_BUG_MDS:
2411 		return mds_show_state(buf);
2412 
2413 	case X86_BUG_TAA:
2414 		return tsx_async_abort_show_state(buf);
2415 
2416 	case X86_BUG_ITLB_MULTIHIT:
2417 		return itlb_multihit_show_state(buf);
2418 
2419 	case X86_BUG_SRBDS:
2420 		return srbds_show_state(buf);
2421 
2422 	case X86_BUG_MMIO_STALE_DATA:
2423 	case X86_BUG_MMIO_UNKNOWN:
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 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2484 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2485 	else
2486 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2487 }
2488 
2489 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2490 {
2491 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2492 }
2493 #endif
2494