xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision 2a12187d)
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 /* Update SPEC_CTRL MSR and its cached copy unconditionally */
64 static void update_spec_ctrl(u64 val)
65 {
66 	this_cpu_write(x86_spec_ctrl_current, val);
67 	wrmsrl(MSR_IA32_SPEC_CTRL, val);
68 }
69 
70 /*
71  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
72  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
73  */
74 void update_spec_ctrl_cond(u64 val)
75 {
76 	if (this_cpu_read(x86_spec_ctrl_current) == val)
77 		return;
78 
79 	this_cpu_write(x86_spec_ctrl_current, val);
80 
81 	/*
82 	 * When KERNEL_IBRS this MSR is written on return-to-user, unless
83 	 * forced the update can be delayed until that time.
84 	 */
85 	if (!cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
86 		wrmsrl(MSR_IA32_SPEC_CTRL, val);
87 }
88 
89 u64 spec_ctrl_current(void)
90 {
91 	return this_cpu_read(x86_spec_ctrl_current);
92 }
93 EXPORT_SYMBOL_GPL(spec_ctrl_current);
94 
95 /*
96  * AMD specific MSR info for Speculative Store Bypass control.
97  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
98  */
99 u64 __ro_after_init x86_amd_ls_cfg_base;
100 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
101 
102 /* Control conditional STIBP in switch_to() */
103 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
104 /* Control conditional IBPB in switch_mm() */
105 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
106 /* Control unconditional IBPB in switch_mm() */
107 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
108 
109 /* Control MDS CPU buffer clear before returning to user space */
110 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
111 EXPORT_SYMBOL_GPL(mds_user_clear);
112 /* Control MDS CPU buffer clear before idling (halt, mwait) */
113 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
114 EXPORT_SYMBOL_GPL(mds_idle_clear);
115 
116 /*
117  * Controls whether l1d flush based mitigations are enabled,
118  * based on hw features and admin setting via boot parameter
119  * defaults to false
120  */
121 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
122 
123 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
124 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
125 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
126 
127 void __init check_bugs(void)
128 {
129 	identify_boot_cpu();
130 
131 	/*
132 	 * identify_boot_cpu() initialized SMT support information, let the
133 	 * core code know.
134 	 */
135 	cpu_smt_check_topology();
136 
137 	if (!IS_ENABLED(CONFIG_SMP)) {
138 		pr_info("CPU: ");
139 		print_cpu_info(&boot_cpu_data);
140 	}
141 
142 	/*
143 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
144 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
145 	 * init code as it is not enumerated and depends on the family.
146 	 */
147 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
148 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
149 
150 	/* Select the proper CPU mitigations before patching alternatives: */
151 	spectre_v1_select_mitigation();
152 	spectre_v2_select_mitigation();
153 	/*
154 	 * retbleed_select_mitigation() relies on the state set by
155 	 * spectre_v2_select_mitigation(); specifically it wants to know about
156 	 * spectre_v2=ibrs.
157 	 */
158 	retbleed_select_mitigation();
159 	/*
160 	 * spectre_v2_user_select_mitigation() relies on the state set by
161 	 * retbleed_select_mitigation(); specifically the STIBP selection is
162 	 * forced for UNRET or IBPB.
163 	 */
164 	spectre_v2_user_select_mitigation();
165 	ssb_select_mitigation();
166 	l1tf_select_mitigation();
167 	md_clear_select_mitigation();
168 	srbds_select_mitigation();
169 	l1d_flush_select_mitigation();
170 
171 	arch_smt_update();
172 
173 #ifdef CONFIG_X86_32
174 	/*
175 	 * Check whether we are able to run this kernel safely on SMP.
176 	 *
177 	 * - i386 is no longer supported.
178 	 * - In order to run on anything without a TSC, we need to be
179 	 *   compiled for a i486.
180 	 */
181 	if (boot_cpu_data.x86 < 4)
182 		panic("Kernel requires i486+ for 'invlpg' and other features");
183 
184 	init_utsname()->machine[1] =
185 		'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
186 	alternative_instructions();
187 
188 	fpu__init_check_bugs();
189 #else /* CONFIG_X86_64 */
190 	alternative_instructions();
191 
192 	/*
193 	 * Make sure the first 2MB area is not mapped by huge pages
194 	 * There are typically fixed size MTRRs in there and overlapping
195 	 * MTRRs into large pages causes slow downs.
196 	 *
197 	 * Right now we don't do that with gbpages because there seems
198 	 * very little benefit for that case.
199 	 */
200 	if (!direct_gbpages)
201 		set_memory_4k((unsigned long)__va(0), 1);
202 #endif
203 }
204 
205 /*
206  * NOTE: This function is *only* called for SVM, since Intel uses
207  * MSR_IA32_SPEC_CTRL for SSBD.
208  */
209 void
210 x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl, bool setguest)
211 {
212 	u64 guestval, hostval;
213 	struct thread_info *ti = current_thread_info();
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 	     boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
437 	     cpu_mitigations_off()) {
438 		mmio_mitigation = MMIO_MITIGATION_OFF;
439 		return;
440 	}
441 
442 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
443 		return;
444 
445 	ia32_cap = x86_read_arch_cap_msr();
446 
447 	/*
448 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
449 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
450 	 */
451 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
452 					      boot_cpu_has(X86_FEATURE_RTM)))
453 		static_branch_enable(&mds_user_clear);
454 	else
455 		static_branch_enable(&mmio_stale_data_clear);
456 
457 	/*
458 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
459 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
460 	 * is required irrespective of SMT state.
461 	 */
462 	if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
463 		static_branch_enable(&mds_idle_clear);
464 
465 	/*
466 	 * Check if the system has the right microcode.
467 	 *
468 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
469 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
470 	 * affected systems.
471 	 */
472 	if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
473 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
474 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
475 	     !(ia32_cap & ARCH_CAP_MDS_NO)))
476 		mmio_mitigation = MMIO_MITIGATION_VERW;
477 	else
478 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
479 
480 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
481 		cpu_smt_disable(false);
482 }
483 
484 static int __init mmio_stale_data_parse_cmdline(char *str)
485 {
486 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
487 		return 0;
488 
489 	if (!str)
490 		return -EINVAL;
491 
492 	if (!strcmp(str, "off")) {
493 		mmio_mitigation = MMIO_MITIGATION_OFF;
494 	} else if (!strcmp(str, "full")) {
495 		mmio_mitigation = MMIO_MITIGATION_VERW;
496 	} else if (!strcmp(str, "full,nosmt")) {
497 		mmio_mitigation = MMIO_MITIGATION_VERW;
498 		mmio_nosmt = true;
499 	}
500 
501 	return 0;
502 }
503 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
504 
505 #undef pr_fmt
506 #define pr_fmt(fmt)     "" fmt
507 
508 static void __init md_clear_update_mitigation(void)
509 {
510 	if (cpu_mitigations_off())
511 		return;
512 
513 	if (!static_key_enabled(&mds_user_clear))
514 		goto out;
515 
516 	/*
517 	 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
518 	 * mitigation, if necessary.
519 	 */
520 	if (mds_mitigation == MDS_MITIGATION_OFF &&
521 	    boot_cpu_has_bug(X86_BUG_MDS)) {
522 		mds_mitigation = MDS_MITIGATION_FULL;
523 		mds_select_mitigation();
524 	}
525 	if (taa_mitigation == TAA_MITIGATION_OFF &&
526 	    boot_cpu_has_bug(X86_BUG_TAA)) {
527 		taa_mitigation = TAA_MITIGATION_VERW;
528 		taa_select_mitigation();
529 	}
530 	if (mmio_mitigation == MMIO_MITIGATION_OFF &&
531 	    boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
532 		mmio_mitigation = MMIO_MITIGATION_VERW;
533 		mmio_select_mitigation();
534 	}
535 out:
536 	if (boot_cpu_has_bug(X86_BUG_MDS))
537 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
538 	if (boot_cpu_has_bug(X86_BUG_TAA))
539 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
540 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
541 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
542 	else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
543 		pr_info("MMIO Stale Data: Unknown: No mitigations\n");
544 }
545 
546 static void __init md_clear_select_mitigation(void)
547 {
548 	mds_select_mitigation();
549 	taa_select_mitigation();
550 	mmio_select_mitigation();
551 
552 	/*
553 	 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
554 	 * and print their mitigation after MDS, TAA and MMIO Stale Data
555 	 * mitigation selection is done.
556 	 */
557 	md_clear_update_mitigation();
558 }
559 
560 #undef pr_fmt
561 #define pr_fmt(fmt)	"SRBDS: " fmt
562 
563 enum srbds_mitigations {
564 	SRBDS_MITIGATION_OFF,
565 	SRBDS_MITIGATION_UCODE_NEEDED,
566 	SRBDS_MITIGATION_FULL,
567 	SRBDS_MITIGATION_TSX_OFF,
568 	SRBDS_MITIGATION_HYPERVISOR,
569 };
570 
571 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
572 
573 static const char * const srbds_strings[] = {
574 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
575 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
576 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
577 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
578 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
579 };
580 
581 static bool srbds_off;
582 
583 void update_srbds_msr(void)
584 {
585 	u64 mcu_ctrl;
586 
587 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
588 		return;
589 
590 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
591 		return;
592 
593 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
594 		return;
595 
596 	/*
597 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
598 	 * being disabled and it hasn't received the SRBDS MSR microcode.
599 	 */
600 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
601 		return;
602 
603 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
604 
605 	switch (srbds_mitigation) {
606 	case SRBDS_MITIGATION_OFF:
607 	case SRBDS_MITIGATION_TSX_OFF:
608 		mcu_ctrl |= RNGDS_MITG_DIS;
609 		break;
610 	case SRBDS_MITIGATION_FULL:
611 		mcu_ctrl &= ~RNGDS_MITG_DIS;
612 		break;
613 	default:
614 		break;
615 	}
616 
617 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
618 }
619 
620 static void __init srbds_select_mitigation(void)
621 {
622 	u64 ia32_cap;
623 
624 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
625 		return;
626 
627 	/*
628 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
629 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
630 	 * by Processor MMIO Stale Data vulnerability.
631 	 */
632 	ia32_cap = x86_read_arch_cap_msr();
633 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
634 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
635 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
636 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
637 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
638 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
639 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
640 	else if (cpu_mitigations_off() || srbds_off)
641 		srbds_mitigation = SRBDS_MITIGATION_OFF;
642 
643 	update_srbds_msr();
644 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
645 }
646 
647 static int __init srbds_parse_cmdline(char *str)
648 {
649 	if (!str)
650 		return -EINVAL;
651 
652 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
653 		return 0;
654 
655 	srbds_off = !strcmp(str, "off");
656 	return 0;
657 }
658 early_param("srbds", srbds_parse_cmdline);
659 
660 #undef pr_fmt
661 #define pr_fmt(fmt)     "L1D Flush : " fmt
662 
663 enum l1d_flush_mitigations {
664 	L1D_FLUSH_OFF = 0,
665 	L1D_FLUSH_ON,
666 };
667 
668 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
669 
670 static void __init l1d_flush_select_mitigation(void)
671 {
672 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
673 		return;
674 
675 	static_branch_enable(&switch_mm_cond_l1d_flush);
676 	pr_info("Conditional flush on switch_mm() enabled\n");
677 }
678 
679 static int __init l1d_flush_parse_cmdline(char *str)
680 {
681 	if (!strcmp(str, "on"))
682 		l1d_flush_mitigation = L1D_FLUSH_ON;
683 
684 	return 0;
685 }
686 early_param("l1d_flush", l1d_flush_parse_cmdline);
687 
688 #undef pr_fmt
689 #define pr_fmt(fmt)     "Spectre V1 : " fmt
690 
691 enum spectre_v1_mitigation {
692 	SPECTRE_V1_MITIGATION_NONE,
693 	SPECTRE_V1_MITIGATION_AUTO,
694 };
695 
696 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
697 	SPECTRE_V1_MITIGATION_AUTO;
698 
699 static const char * const spectre_v1_strings[] = {
700 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
701 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
702 };
703 
704 /*
705  * Does SMAP provide full mitigation against speculative kernel access to
706  * userspace?
707  */
708 static bool smap_works_speculatively(void)
709 {
710 	if (!boot_cpu_has(X86_FEATURE_SMAP))
711 		return false;
712 
713 	/*
714 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
715 	 * prevent speculative access to user data in the L1 cache.
716 	 * Consider SMAP to be non-functional as a mitigation on these
717 	 * CPUs.
718 	 */
719 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
720 		return false;
721 
722 	return true;
723 }
724 
725 static void __init spectre_v1_select_mitigation(void)
726 {
727 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
728 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
729 		return;
730 	}
731 
732 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
733 		/*
734 		 * With Spectre v1, a user can speculatively control either
735 		 * path of a conditional swapgs with a user-controlled GS
736 		 * value.  The mitigation is to add lfences to both code paths.
737 		 *
738 		 * If FSGSBASE is enabled, the user can put a kernel address in
739 		 * GS, in which case SMAP provides no protection.
740 		 *
741 		 * If FSGSBASE is disabled, the user can only put a user space
742 		 * address in GS.  That makes an attack harder, but still
743 		 * possible if there's no SMAP protection.
744 		 */
745 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
746 		    !smap_works_speculatively()) {
747 			/*
748 			 * Mitigation can be provided from SWAPGS itself or
749 			 * PTI as the CR3 write in the Meltdown mitigation
750 			 * is serializing.
751 			 *
752 			 * If neither is there, mitigate with an LFENCE to
753 			 * stop speculation through swapgs.
754 			 */
755 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
756 			    !boot_cpu_has(X86_FEATURE_PTI))
757 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
758 
759 			/*
760 			 * Enable lfences in the kernel entry (non-swapgs)
761 			 * paths, to prevent user entry from speculatively
762 			 * skipping swapgs.
763 			 */
764 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
765 		}
766 	}
767 
768 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
769 }
770 
771 static int __init nospectre_v1_cmdline(char *str)
772 {
773 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
774 	return 0;
775 }
776 early_param("nospectre_v1", nospectre_v1_cmdline);
777 
778 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
779 	SPECTRE_V2_NONE;
780 
781 #undef pr_fmt
782 #define pr_fmt(fmt)     "RETBleed: " fmt
783 
784 enum retbleed_mitigation {
785 	RETBLEED_MITIGATION_NONE,
786 	RETBLEED_MITIGATION_UNRET,
787 	RETBLEED_MITIGATION_IBPB,
788 	RETBLEED_MITIGATION_IBRS,
789 	RETBLEED_MITIGATION_EIBRS,
790 	RETBLEED_MITIGATION_STUFF,
791 };
792 
793 enum retbleed_mitigation_cmd {
794 	RETBLEED_CMD_OFF,
795 	RETBLEED_CMD_AUTO,
796 	RETBLEED_CMD_UNRET,
797 	RETBLEED_CMD_IBPB,
798 	RETBLEED_CMD_STUFF,
799 };
800 
801 static const char * const retbleed_strings[] = {
802 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
803 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
804 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
805 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
806 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
807 	[RETBLEED_MITIGATION_STUFF]	= "Mitigation: Stuffing",
808 };
809 
810 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
811 	RETBLEED_MITIGATION_NONE;
812 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
813 	RETBLEED_CMD_AUTO;
814 
815 static int __ro_after_init retbleed_nosmt = false;
816 
817 static int __init retbleed_parse_cmdline(char *str)
818 {
819 	if (!str)
820 		return -EINVAL;
821 
822 	while (str) {
823 		char *next = strchr(str, ',');
824 		if (next) {
825 			*next = 0;
826 			next++;
827 		}
828 
829 		if (!strcmp(str, "off")) {
830 			retbleed_cmd = RETBLEED_CMD_OFF;
831 		} else if (!strcmp(str, "auto")) {
832 			retbleed_cmd = RETBLEED_CMD_AUTO;
833 		} else if (!strcmp(str, "unret")) {
834 			retbleed_cmd = RETBLEED_CMD_UNRET;
835 		} else if (!strcmp(str, "ibpb")) {
836 			retbleed_cmd = RETBLEED_CMD_IBPB;
837 		} else if (!strcmp(str, "stuff")) {
838 			retbleed_cmd = RETBLEED_CMD_STUFF;
839 		} else if (!strcmp(str, "nosmt")) {
840 			retbleed_nosmt = true;
841 		} else if (!strcmp(str, "force")) {
842 			setup_force_cpu_bug(X86_BUG_RETBLEED);
843 		} else {
844 			pr_err("Ignoring unknown retbleed option (%s).", str);
845 		}
846 
847 		str = next;
848 	}
849 
850 	return 0;
851 }
852 early_param("retbleed", retbleed_parse_cmdline);
853 
854 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
855 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
856 
857 static void __init retbleed_select_mitigation(void)
858 {
859 	bool mitigate_smt = false;
860 
861 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
862 		return;
863 
864 	switch (retbleed_cmd) {
865 	case RETBLEED_CMD_OFF:
866 		return;
867 
868 	case RETBLEED_CMD_UNRET:
869 		if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
870 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
871 		} else {
872 			pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
873 			goto do_cmd_auto;
874 		}
875 		break;
876 
877 	case RETBLEED_CMD_IBPB:
878 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
879 			pr_err("WARNING: CPU does not support IBPB.\n");
880 			goto do_cmd_auto;
881 		} else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
882 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
883 		} else {
884 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
885 			goto do_cmd_auto;
886 		}
887 		break;
888 
889 	case RETBLEED_CMD_STUFF:
890 		if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING) &&
891 		    spectre_v2_enabled == SPECTRE_V2_RETPOLINE) {
892 			retbleed_mitigation = RETBLEED_MITIGATION_STUFF;
893 
894 		} else {
895 			if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING))
896 				pr_err("WARNING: retbleed=stuff depends on spectre_v2=retpoline\n");
897 			else
898 				pr_err("WARNING: kernel not compiled with CALL_DEPTH_TRACKING.\n");
899 
900 			goto do_cmd_auto;
901 		}
902 		break;
903 
904 do_cmd_auto:
905 	case RETBLEED_CMD_AUTO:
906 	default:
907 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
908 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
909 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
910 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
911 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
912 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
913 		}
914 
915 		/*
916 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
917 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
918 		 * be set accordingly below.
919 		 */
920 
921 		break;
922 	}
923 
924 	switch (retbleed_mitigation) {
925 	case RETBLEED_MITIGATION_UNRET:
926 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
927 		setup_force_cpu_cap(X86_FEATURE_UNRET);
928 
929 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
930 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
931 			pr_err(RETBLEED_UNTRAIN_MSG);
932 
933 		mitigate_smt = true;
934 		break;
935 
936 	case RETBLEED_MITIGATION_IBPB:
937 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
938 		mitigate_smt = true;
939 		break;
940 
941 	case RETBLEED_MITIGATION_STUFF:
942 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
943 		setup_force_cpu_cap(X86_FEATURE_CALL_DEPTH);
944 		x86_set_skl_return_thunk();
945 		break;
946 
947 	default:
948 		break;
949 	}
950 
951 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
952 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
953 		cpu_smt_disable(false);
954 
955 	/*
956 	 * Let IBRS trump all on Intel without affecting the effects of the
957 	 * retbleed= cmdline option except for call depth based stuffing
958 	 */
959 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
960 		switch (spectre_v2_enabled) {
961 		case SPECTRE_V2_IBRS:
962 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
963 			break;
964 		case SPECTRE_V2_EIBRS:
965 		case SPECTRE_V2_EIBRS_RETPOLINE:
966 		case SPECTRE_V2_EIBRS_LFENCE:
967 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
968 			break;
969 		default:
970 			if (retbleed_mitigation != RETBLEED_MITIGATION_STUFF)
971 				pr_err(RETBLEED_INTEL_MSG);
972 		}
973 	}
974 
975 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
976 }
977 
978 #undef pr_fmt
979 #define pr_fmt(fmt)     "Spectre V2 : " fmt
980 
981 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
982 	SPECTRE_V2_USER_NONE;
983 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
984 	SPECTRE_V2_USER_NONE;
985 
986 #ifdef CONFIG_RETPOLINE
987 static bool spectre_v2_bad_module;
988 
989 bool retpoline_module_ok(bool has_retpoline)
990 {
991 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
992 		return true;
993 
994 	pr_err("System may be vulnerable to spectre v2\n");
995 	spectre_v2_bad_module = true;
996 	return false;
997 }
998 
999 static inline const char *spectre_v2_module_string(void)
1000 {
1001 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
1002 }
1003 #else
1004 static inline const char *spectre_v2_module_string(void) { return ""; }
1005 #endif
1006 
1007 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
1008 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
1009 #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"
1010 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
1011 
1012 #ifdef CONFIG_BPF_SYSCALL
1013 void unpriv_ebpf_notify(int new_state)
1014 {
1015 	if (new_state)
1016 		return;
1017 
1018 	/* Unprivileged eBPF is enabled */
1019 
1020 	switch (spectre_v2_enabled) {
1021 	case SPECTRE_V2_EIBRS:
1022 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1023 		break;
1024 	case SPECTRE_V2_EIBRS_LFENCE:
1025 		if (sched_smt_active())
1026 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1027 		break;
1028 	default:
1029 		break;
1030 	}
1031 }
1032 #endif
1033 
1034 static inline bool match_option(const char *arg, int arglen, const char *opt)
1035 {
1036 	int len = strlen(opt);
1037 
1038 	return len == arglen && !strncmp(arg, opt, len);
1039 }
1040 
1041 /* The kernel command line selection for spectre v2 */
1042 enum spectre_v2_mitigation_cmd {
1043 	SPECTRE_V2_CMD_NONE,
1044 	SPECTRE_V2_CMD_AUTO,
1045 	SPECTRE_V2_CMD_FORCE,
1046 	SPECTRE_V2_CMD_RETPOLINE,
1047 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1048 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1049 	SPECTRE_V2_CMD_EIBRS,
1050 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1051 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1052 	SPECTRE_V2_CMD_IBRS,
1053 };
1054 
1055 enum spectre_v2_user_cmd {
1056 	SPECTRE_V2_USER_CMD_NONE,
1057 	SPECTRE_V2_USER_CMD_AUTO,
1058 	SPECTRE_V2_USER_CMD_FORCE,
1059 	SPECTRE_V2_USER_CMD_PRCTL,
1060 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1061 	SPECTRE_V2_USER_CMD_SECCOMP,
1062 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1063 };
1064 
1065 static const char * const spectre_v2_user_strings[] = {
1066 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1067 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1068 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1069 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1070 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1071 };
1072 
1073 static const struct {
1074 	const char			*option;
1075 	enum spectre_v2_user_cmd	cmd;
1076 	bool				secure;
1077 } v2_user_options[] __initconst = {
1078 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1079 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1080 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1081 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1082 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1083 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1084 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1085 };
1086 
1087 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1088 {
1089 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1090 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1091 }
1092 
1093 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1094 
1095 static enum spectre_v2_user_cmd __init
1096 spectre_v2_parse_user_cmdline(void)
1097 {
1098 	char arg[20];
1099 	int ret, i;
1100 
1101 	switch (spectre_v2_cmd) {
1102 	case SPECTRE_V2_CMD_NONE:
1103 		return SPECTRE_V2_USER_CMD_NONE;
1104 	case SPECTRE_V2_CMD_FORCE:
1105 		return SPECTRE_V2_USER_CMD_FORCE;
1106 	default:
1107 		break;
1108 	}
1109 
1110 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1111 				  arg, sizeof(arg));
1112 	if (ret < 0)
1113 		return SPECTRE_V2_USER_CMD_AUTO;
1114 
1115 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1116 		if (match_option(arg, ret, v2_user_options[i].option)) {
1117 			spec_v2_user_print_cond(v2_user_options[i].option,
1118 						v2_user_options[i].secure);
1119 			return v2_user_options[i].cmd;
1120 		}
1121 	}
1122 
1123 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1124 	return SPECTRE_V2_USER_CMD_AUTO;
1125 }
1126 
1127 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1128 {
1129 	return mode == SPECTRE_V2_IBRS ||
1130 	       mode == SPECTRE_V2_EIBRS ||
1131 	       mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1132 	       mode == SPECTRE_V2_EIBRS_LFENCE;
1133 }
1134 
1135 static void __init
1136 spectre_v2_user_select_mitigation(void)
1137 {
1138 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1139 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1140 	enum spectre_v2_user_cmd cmd;
1141 
1142 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1143 		return;
1144 
1145 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1146 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1147 		smt_possible = false;
1148 
1149 	cmd = spectre_v2_parse_user_cmdline();
1150 	switch (cmd) {
1151 	case SPECTRE_V2_USER_CMD_NONE:
1152 		goto set_mode;
1153 	case SPECTRE_V2_USER_CMD_FORCE:
1154 		mode = SPECTRE_V2_USER_STRICT;
1155 		break;
1156 	case SPECTRE_V2_USER_CMD_AUTO:
1157 	case SPECTRE_V2_USER_CMD_PRCTL:
1158 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1159 		mode = SPECTRE_V2_USER_PRCTL;
1160 		break;
1161 	case SPECTRE_V2_USER_CMD_SECCOMP:
1162 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1163 		if (IS_ENABLED(CONFIG_SECCOMP))
1164 			mode = SPECTRE_V2_USER_SECCOMP;
1165 		else
1166 			mode = SPECTRE_V2_USER_PRCTL;
1167 		break;
1168 	}
1169 
1170 	/* Initialize Indirect Branch Prediction Barrier */
1171 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1172 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1173 
1174 		spectre_v2_user_ibpb = mode;
1175 		switch (cmd) {
1176 		case SPECTRE_V2_USER_CMD_FORCE:
1177 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1178 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1179 			static_branch_enable(&switch_mm_always_ibpb);
1180 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1181 			break;
1182 		case SPECTRE_V2_USER_CMD_PRCTL:
1183 		case SPECTRE_V2_USER_CMD_AUTO:
1184 		case SPECTRE_V2_USER_CMD_SECCOMP:
1185 			static_branch_enable(&switch_mm_cond_ibpb);
1186 			break;
1187 		default:
1188 			break;
1189 		}
1190 
1191 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1192 			static_key_enabled(&switch_mm_always_ibpb) ?
1193 			"always-on" : "conditional");
1194 	}
1195 
1196 	/*
1197 	 * If no STIBP, IBRS or enhanced IBRS is enabled, or SMT impossible,
1198 	 * STIBP is not required.
1199 	 */
1200 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1201 	    !smt_possible ||
1202 	    spectre_v2_in_ibrs_mode(spectre_v2_enabled))
1203 		return;
1204 
1205 	/*
1206 	 * At this point, an STIBP mode other than "off" has been set.
1207 	 * If STIBP support is not being forced, check if STIBP always-on
1208 	 * is preferred.
1209 	 */
1210 	if (mode != SPECTRE_V2_USER_STRICT &&
1211 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1212 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1213 
1214 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1215 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1216 		if (mode != SPECTRE_V2_USER_STRICT &&
1217 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1218 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1219 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1220 	}
1221 
1222 	spectre_v2_user_stibp = mode;
1223 
1224 set_mode:
1225 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1226 }
1227 
1228 static const char * const spectre_v2_strings[] = {
1229 	[SPECTRE_V2_NONE]			= "Vulnerable",
1230 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1231 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1232 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced IBRS",
1233 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced IBRS + LFENCE",
1234 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced IBRS + Retpolines",
1235 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1236 };
1237 
1238 static const struct {
1239 	const char *option;
1240 	enum spectre_v2_mitigation_cmd cmd;
1241 	bool secure;
1242 } mitigation_options[] __initconst = {
1243 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1244 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1245 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1246 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1247 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1248 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1249 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1250 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1251 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1252 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1253 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1254 };
1255 
1256 static void __init spec_v2_print_cond(const char *reason, bool secure)
1257 {
1258 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1259 		pr_info("%s selected on command line.\n", reason);
1260 }
1261 
1262 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1263 {
1264 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1265 	char arg[20];
1266 	int ret, i;
1267 
1268 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1269 	    cpu_mitigations_off())
1270 		return SPECTRE_V2_CMD_NONE;
1271 
1272 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1273 	if (ret < 0)
1274 		return SPECTRE_V2_CMD_AUTO;
1275 
1276 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1277 		if (!match_option(arg, ret, mitigation_options[i].option))
1278 			continue;
1279 		cmd = mitigation_options[i].cmd;
1280 		break;
1281 	}
1282 
1283 	if (i >= ARRAY_SIZE(mitigation_options)) {
1284 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1285 		return SPECTRE_V2_CMD_AUTO;
1286 	}
1287 
1288 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1289 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1290 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1291 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1292 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1293 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1294 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1295 		       mitigation_options[i].option);
1296 		return SPECTRE_V2_CMD_AUTO;
1297 	}
1298 
1299 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1300 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1301 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1302 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1303 		pr_err("%s selected but CPU doesn't have eIBRS. Switching to AUTO select\n",
1304 		       mitigation_options[i].option);
1305 		return SPECTRE_V2_CMD_AUTO;
1306 	}
1307 
1308 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1309 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1310 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1311 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1312 		       mitigation_options[i].option);
1313 		return SPECTRE_V2_CMD_AUTO;
1314 	}
1315 
1316 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1317 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1318 		       mitigation_options[i].option);
1319 		return SPECTRE_V2_CMD_AUTO;
1320 	}
1321 
1322 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1323 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1324 		       mitigation_options[i].option);
1325 		return SPECTRE_V2_CMD_AUTO;
1326 	}
1327 
1328 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1329 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1330 		       mitigation_options[i].option);
1331 		return SPECTRE_V2_CMD_AUTO;
1332 	}
1333 
1334 	if (cmd == SPECTRE_V2_CMD_IBRS && cpu_feature_enabled(X86_FEATURE_XENPV)) {
1335 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1336 		       mitigation_options[i].option);
1337 		return SPECTRE_V2_CMD_AUTO;
1338 	}
1339 
1340 	spec_v2_print_cond(mitigation_options[i].option,
1341 			   mitigation_options[i].secure);
1342 	return cmd;
1343 }
1344 
1345 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1346 {
1347 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1348 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1349 		return SPECTRE_V2_NONE;
1350 	}
1351 
1352 	return SPECTRE_V2_RETPOLINE;
1353 }
1354 
1355 /* Disable in-kernel use of non-RSB RET predictors */
1356 static void __init spec_ctrl_disable_kernel_rrsba(void)
1357 {
1358 	u64 ia32_cap;
1359 
1360 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1361 		return;
1362 
1363 	ia32_cap = x86_read_arch_cap_msr();
1364 
1365 	if (ia32_cap & ARCH_CAP_RRSBA) {
1366 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1367 		update_spec_ctrl(x86_spec_ctrl_base);
1368 	}
1369 }
1370 
1371 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1372 {
1373 	/*
1374 	 * Similar to context switches, there are two types of RSB attacks
1375 	 * after VM exit:
1376 	 *
1377 	 * 1) RSB underflow
1378 	 *
1379 	 * 2) Poisoned RSB entry
1380 	 *
1381 	 * When retpoline is enabled, both are mitigated by filling/clearing
1382 	 * the RSB.
1383 	 *
1384 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1385 	 * prediction isolation protections, RSB still needs to be cleared
1386 	 * because of #2.  Note that SMEP provides no protection here, unlike
1387 	 * user-space-poisoned RSB entries.
1388 	 *
1389 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1390 	 * bug is present then a LITE version of RSB protection is required,
1391 	 * just a single call needs to retire before a RET is executed.
1392 	 */
1393 	switch (mode) {
1394 	case SPECTRE_V2_NONE:
1395 		return;
1396 
1397 	case SPECTRE_V2_EIBRS_LFENCE:
1398 	case SPECTRE_V2_EIBRS:
1399 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1400 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1401 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1402 		}
1403 		return;
1404 
1405 	case SPECTRE_V2_EIBRS_RETPOLINE:
1406 	case SPECTRE_V2_RETPOLINE:
1407 	case SPECTRE_V2_LFENCE:
1408 	case SPECTRE_V2_IBRS:
1409 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1410 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1411 		return;
1412 	}
1413 
1414 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1415 	dump_stack();
1416 }
1417 
1418 static void __init spectre_v2_select_mitigation(void)
1419 {
1420 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1421 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1422 
1423 	/*
1424 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1425 	 * then nothing to do.
1426 	 */
1427 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1428 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1429 		return;
1430 
1431 	switch (cmd) {
1432 	case SPECTRE_V2_CMD_NONE:
1433 		return;
1434 
1435 	case SPECTRE_V2_CMD_FORCE:
1436 	case SPECTRE_V2_CMD_AUTO:
1437 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1438 			mode = SPECTRE_V2_EIBRS;
1439 			break;
1440 		}
1441 
1442 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1443 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1444 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1445 		    retbleed_cmd != RETBLEED_CMD_STUFF &&
1446 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1447 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1448 			mode = SPECTRE_V2_IBRS;
1449 			break;
1450 		}
1451 
1452 		mode = spectre_v2_select_retpoline();
1453 		break;
1454 
1455 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1456 		pr_err(SPECTRE_V2_LFENCE_MSG);
1457 		mode = SPECTRE_V2_LFENCE;
1458 		break;
1459 
1460 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1461 		mode = SPECTRE_V2_RETPOLINE;
1462 		break;
1463 
1464 	case SPECTRE_V2_CMD_RETPOLINE:
1465 		mode = spectre_v2_select_retpoline();
1466 		break;
1467 
1468 	case SPECTRE_V2_CMD_IBRS:
1469 		mode = SPECTRE_V2_IBRS;
1470 		break;
1471 
1472 	case SPECTRE_V2_CMD_EIBRS:
1473 		mode = SPECTRE_V2_EIBRS;
1474 		break;
1475 
1476 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1477 		mode = SPECTRE_V2_EIBRS_LFENCE;
1478 		break;
1479 
1480 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1481 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1482 		break;
1483 	}
1484 
1485 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1486 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1487 
1488 	if (spectre_v2_in_ibrs_mode(mode)) {
1489 		x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1490 		update_spec_ctrl(x86_spec_ctrl_base);
1491 	}
1492 
1493 	switch (mode) {
1494 	case SPECTRE_V2_NONE:
1495 	case SPECTRE_V2_EIBRS:
1496 		break;
1497 
1498 	case SPECTRE_V2_IBRS:
1499 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1500 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1501 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1502 		break;
1503 
1504 	case SPECTRE_V2_LFENCE:
1505 	case SPECTRE_V2_EIBRS_LFENCE:
1506 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1507 		fallthrough;
1508 
1509 	case SPECTRE_V2_RETPOLINE:
1510 	case SPECTRE_V2_EIBRS_RETPOLINE:
1511 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1512 		break;
1513 	}
1514 
1515 	/*
1516 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1517 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1518 	 * prediction from a non-RSB predictor is still a risk.
1519 	 */
1520 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1521 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1522 	    mode == SPECTRE_V2_RETPOLINE)
1523 		spec_ctrl_disable_kernel_rrsba();
1524 
1525 	spectre_v2_enabled = mode;
1526 	pr_info("%s\n", spectre_v2_strings[mode]);
1527 
1528 	/*
1529 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1530 	 * context switch.  In general there are two types of RSB attacks
1531 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1532 	 *
1533 	 * 1) RSB underflow
1534 	 *
1535 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1536 	 *    speculated return targets may come from the branch predictor,
1537 	 *    which could have a user-poisoned BTB or BHB entry.
1538 	 *
1539 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1540 	 *    regardless of the state of the RSB.
1541 	 *
1542 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1543 	 *    scenario is mitigated by the IBRS branch prediction isolation
1544 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1545 	 *    protect against this type of attack.
1546 	 *
1547 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1548 	 *
1549 	 * 2) Poisoned RSB entry
1550 	 *
1551 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1552 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1553 	 *    entry.
1554 	 *
1555 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1556 	 *    eIBRS.
1557 	 *
1558 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1559 	 *    RSB clearing.
1560 	 *
1561 	 * So to mitigate all cases, unconditionally fill RSB on context
1562 	 * switches.
1563 	 *
1564 	 * FIXME: Is this pointless for retbleed-affected AMD?
1565 	 */
1566 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1567 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1568 
1569 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1570 
1571 	/*
1572 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1573 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1574 	 * firmware calls only when IBRS / Enhanced IBRS aren't otherwise
1575 	 * enabled.
1576 	 *
1577 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1578 	 * the user might select retpoline on the kernel command line and if
1579 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1580 	 * enable IBRS around firmware calls.
1581 	 */
1582 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1583 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1584 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1585 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1586 
1587 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1588 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1589 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1590 		}
1591 
1592 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1593 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1594 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1595 	}
1596 
1597 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1598 	spectre_v2_cmd = cmd;
1599 }
1600 
1601 static void update_stibp_msr(void * __unused)
1602 {
1603 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1604 	update_spec_ctrl(val);
1605 }
1606 
1607 /* Update x86_spec_ctrl_base in case SMT state changed. */
1608 static void update_stibp_strict(void)
1609 {
1610 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1611 
1612 	if (sched_smt_active())
1613 		mask |= SPEC_CTRL_STIBP;
1614 
1615 	if (mask == x86_spec_ctrl_base)
1616 		return;
1617 
1618 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1619 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1620 	x86_spec_ctrl_base = mask;
1621 	on_each_cpu(update_stibp_msr, NULL, 1);
1622 }
1623 
1624 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1625 static void update_indir_branch_cond(void)
1626 {
1627 	if (sched_smt_active())
1628 		static_branch_enable(&switch_to_cond_stibp);
1629 	else
1630 		static_branch_disable(&switch_to_cond_stibp);
1631 }
1632 
1633 #undef pr_fmt
1634 #define pr_fmt(fmt) fmt
1635 
1636 /* Update the static key controlling the MDS CPU buffer clear in idle */
1637 static void update_mds_branch_idle(void)
1638 {
1639 	u64 ia32_cap = x86_read_arch_cap_msr();
1640 
1641 	/*
1642 	 * Enable the idle clearing if SMT is active on CPUs which are
1643 	 * affected only by MSBDS and not any other MDS variant.
1644 	 *
1645 	 * The other variants cannot be mitigated when SMT is enabled, so
1646 	 * clearing the buffers on idle just to prevent the Store Buffer
1647 	 * repartitioning leak would be a window dressing exercise.
1648 	 */
1649 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1650 		return;
1651 
1652 	if (sched_smt_active()) {
1653 		static_branch_enable(&mds_idle_clear);
1654 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1655 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1656 		static_branch_disable(&mds_idle_clear);
1657 	}
1658 }
1659 
1660 #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"
1661 #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"
1662 #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"
1663 
1664 void cpu_bugs_smt_update(void)
1665 {
1666 	mutex_lock(&spec_ctrl_mutex);
1667 
1668 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1669 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1670 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1671 
1672 	switch (spectre_v2_user_stibp) {
1673 	case SPECTRE_V2_USER_NONE:
1674 		break;
1675 	case SPECTRE_V2_USER_STRICT:
1676 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1677 		update_stibp_strict();
1678 		break;
1679 	case SPECTRE_V2_USER_PRCTL:
1680 	case SPECTRE_V2_USER_SECCOMP:
1681 		update_indir_branch_cond();
1682 		break;
1683 	}
1684 
1685 	switch (mds_mitigation) {
1686 	case MDS_MITIGATION_FULL:
1687 	case MDS_MITIGATION_VMWERV:
1688 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1689 			pr_warn_once(MDS_MSG_SMT);
1690 		update_mds_branch_idle();
1691 		break;
1692 	case MDS_MITIGATION_OFF:
1693 		break;
1694 	}
1695 
1696 	switch (taa_mitigation) {
1697 	case TAA_MITIGATION_VERW:
1698 	case TAA_MITIGATION_UCODE_NEEDED:
1699 		if (sched_smt_active())
1700 			pr_warn_once(TAA_MSG_SMT);
1701 		break;
1702 	case TAA_MITIGATION_TSX_DISABLED:
1703 	case TAA_MITIGATION_OFF:
1704 		break;
1705 	}
1706 
1707 	switch (mmio_mitigation) {
1708 	case MMIO_MITIGATION_VERW:
1709 	case MMIO_MITIGATION_UCODE_NEEDED:
1710 		if (sched_smt_active())
1711 			pr_warn_once(MMIO_MSG_SMT);
1712 		break;
1713 	case MMIO_MITIGATION_OFF:
1714 		break;
1715 	}
1716 
1717 	mutex_unlock(&spec_ctrl_mutex);
1718 }
1719 
1720 #undef pr_fmt
1721 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1722 
1723 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1724 
1725 /* The kernel command line selection */
1726 enum ssb_mitigation_cmd {
1727 	SPEC_STORE_BYPASS_CMD_NONE,
1728 	SPEC_STORE_BYPASS_CMD_AUTO,
1729 	SPEC_STORE_BYPASS_CMD_ON,
1730 	SPEC_STORE_BYPASS_CMD_PRCTL,
1731 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1732 };
1733 
1734 static const char * const ssb_strings[] = {
1735 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1736 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1737 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1738 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1739 };
1740 
1741 static const struct {
1742 	const char *option;
1743 	enum ssb_mitigation_cmd cmd;
1744 } ssb_mitigation_options[]  __initconst = {
1745 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1746 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1747 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1748 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1749 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1750 };
1751 
1752 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1753 {
1754 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1755 	char arg[20];
1756 	int ret, i;
1757 
1758 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1759 	    cpu_mitigations_off()) {
1760 		return SPEC_STORE_BYPASS_CMD_NONE;
1761 	} else {
1762 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1763 					  arg, sizeof(arg));
1764 		if (ret < 0)
1765 			return SPEC_STORE_BYPASS_CMD_AUTO;
1766 
1767 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1768 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1769 				continue;
1770 
1771 			cmd = ssb_mitigation_options[i].cmd;
1772 			break;
1773 		}
1774 
1775 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1776 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1777 			return SPEC_STORE_BYPASS_CMD_AUTO;
1778 		}
1779 	}
1780 
1781 	return cmd;
1782 }
1783 
1784 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1785 {
1786 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1787 	enum ssb_mitigation_cmd cmd;
1788 
1789 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1790 		return mode;
1791 
1792 	cmd = ssb_parse_cmdline();
1793 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1794 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1795 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1796 		return mode;
1797 
1798 	switch (cmd) {
1799 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1800 		/*
1801 		 * Choose prctl+seccomp as the default mode if seccomp is
1802 		 * enabled.
1803 		 */
1804 		if (IS_ENABLED(CONFIG_SECCOMP))
1805 			mode = SPEC_STORE_BYPASS_SECCOMP;
1806 		else
1807 			mode = SPEC_STORE_BYPASS_PRCTL;
1808 		break;
1809 	case SPEC_STORE_BYPASS_CMD_ON:
1810 		mode = SPEC_STORE_BYPASS_DISABLE;
1811 		break;
1812 	case SPEC_STORE_BYPASS_CMD_AUTO:
1813 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1814 		mode = SPEC_STORE_BYPASS_PRCTL;
1815 		break;
1816 	case SPEC_STORE_BYPASS_CMD_NONE:
1817 		break;
1818 	}
1819 
1820 	/*
1821 	 * We have three CPU feature flags that are in play here:
1822 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1823 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1824 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1825 	 */
1826 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1827 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1828 		/*
1829 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1830 		 * use a completely different MSR and bit dependent on family.
1831 		 */
1832 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1833 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1834 			x86_amd_ssb_disable();
1835 		} else {
1836 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1837 			update_spec_ctrl(x86_spec_ctrl_base);
1838 		}
1839 	}
1840 
1841 	return mode;
1842 }
1843 
1844 static void ssb_select_mitigation(void)
1845 {
1846 	ssb_mode = __ssb_select_mitigation();
1847 
1848 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1849 		pr_info("%s\n", ssb_strings[ssb_mode]);
1850 }
1851 
1852 #undef pr_fmt
1853 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1854 
1855 static void task_update_spec_tif(struct task_struct *tsk)
1856 {
1857 	/* Force the update of the real TIF bits */
1858 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1859 
1860 	/*
1861 	 * Immediately update the speculation control MSRs for the current
1862 	 * task, but for a non-current task delay setting the CPU
1863 	 * mitigation until it is scheduled next.
1864 	 *
1865 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1866 	 * always the current task.
1867 	 */
1868 	if (tsk == current)
1869 		speculation_ctrl_update_current();
1870 }
1871 
1872 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1873 {
1874 
1875 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1876 		return -EPERM;
1877 
1878 	switch (ctrl) {
1879 	case PR_SPEC_ENABLE:
1880 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1881 		return 0;
1882 	case PR_SPEC_DISABLE:
1883 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1884 		return 0;
1885 	default:
1886 		return -ERANGE;
1887 	}
1888 }
1889 
1890 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1891 {
1892 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1893 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1894 		return -ENXIO;
1895 
1896 	switch (ctrl) {
1897 	case PR_SPEC_ENABLE:
1898 		/* If speculation is force disabled, enable is not allowed */
1899 		if (task_spec_ssb_force_disable(task))
1900 			return -EPERM;
1901 		task_clear_spec_ssb_disable(task);
1902 		task_clear_spec_ssb_noexec(task);
1903 		task_update_spec_tif(task);
1904 		break;
1905 	case PR_SPEC_DISABLE:
1906 		task_set_spec_ssb_disable(task);
1907 		task_clear_spec_ssb_noexec(task);
1908 		task_update_spec_tif(task);
1909 		break;
1910 	case PR_SPEC_FORCE_DISABLE:
1911 		task_set_spec_ssb_disable(task);
1912 		task_set_spec_ssb_force_disable(task);
1913 		task_clear_spec_ssb_noexec(task);
1914 		task_update_spec_tif(task);
1915 		break;
1916 	case PR_SPEC_DISABLE_NOEXEC:
1917 		if (task_spec_ssb_force_disable(task))
1918 			return -EPERM;
1919 		task_set_spec_ssb_disable(task);
1920 		task_set_spec_ssb_noexec(task);
1921 		task_update_spec_tif(task);
1922 		break;
1923 	default:
1924 		return -ERANGE;
1925 	}
1926 	return 0;
1927 }
1928 
1929 static bool is_spec_ib_user_controlled(void)
1930 {
1931 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1932 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1933 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1934 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1935 }
1936 
1937 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1938 {
1939 	switch (ctrl) {
1940 	case PR_SPEC_ENABLE:
1941 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1942 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1943 			return 0;
1944 
1945 		/*
1946 		 * With strict mode for both IBPB and STIBP, the instruction
1947 		 * code paths avoid checking this task flag and instead,
1948 		 * unconditionally run the instruction. However, STIBP and IBPB
1949 		 * are independent and either can be set to conditionally
1950 		 * enabled regardless of the mode of the other.
1951 		 *
1952 		 * If either is set to conditional, allow the task flag to be
1953 		 * updated, unless it was force-disabled by a previous prctl
1954 		 * call. Currently, this is possible on an AMD CPU which has the
1955 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1956 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1957 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1958 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1959 		 */
1960 		if (!is_spec_ib_user_controlled() ||
1961 		    task_spec_ib_force_disable(task))
1962 			return -EPERM;
1963 
1964 		task_clear_spec_ib_disable(task);
1965 		task_update_spec_tif(task);
1966 		break;
1967 	case PR_SPEC_DISABLE:
1968 	case PR_SPEC_FORCE_DISABLE:
1969 		/*
1970 		 * Indirect branch speculation is always allowed when
1971 		 * mitigation is force disabled.
1972 		 */
1973 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1974 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1975 			return -EPERM;
1976 
1977 		if (!is_spec_ib_user_controlled())
1978 			return 0;
1979 
1980 		task_set_spec_ib_disable(task);
1981 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1982 			task_set_spec_ib_force_disable(task);
1983 		task_update_spec_tif(task);
1984 		break;
1985 	default:
1986 		return -ERANGE;
1987 	}
1988 	return 0;
1989 }
1990 
1991 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1992 			     unsigned long ctrl)
1993 {
1994 	switch (which) {
1995 	case PR_SPEC_STORE_BYPASS:
1996 		return ssb_prctl_set(task, ctrl);
1997 	case PR_SPEC_INDIRECT_BRANCH:
1998 		return ib_prctl_set(task, ctrl);
1999 	case PR_SPEC_L1D_FLUSH:
2000 		return l1d_flush_prctl_set(task, ctrl);
2001 	default:
2002 		return -ENODEV;
2003 	}
2004 }
2005 
2006 #ifdef CONFIG_SECCOMP
2007 void arch_seccomp_spec_mitigate(struct task_struct *task)
2008 {
2009 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
2010 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2011 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2012 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
2013 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2014 }
2015 #endif
2016 
2017 static int l1d_flush_prctl_get(struct task_struct *task)
2018 {
2019 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2020 		return PR_SPEC_FORCE_DISABLE;
2021 
2022 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
2023 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2024 	else
2025 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2026 }
2027 
2028 static int ssb_prctl_get(struct task_struct *task)
2029 {
2030 	switch (ssb_mode) {
2031 	case SPEC_STORE_BYPASS_DISABLE:
2032 		return PR_SPEC_DISABLE;
2033 	case SPEC_STORE_BYPASS_SECCOMP:
2034 	case SPEC_STORE_BYPASS_PRCTL:
2035 		if (task_spec_ssb_force_disable(task))
2036 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2037 		if (task_spec_ssb_noexec(task))
2038 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2039 		if (task_spec_ssb_disable(task))
2040 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2041 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2042 	default:
2043 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2044 			return PR_SPEC_ENABLE;
2045 		return PR_SPEC_NOT_AFFECTED;
2046 	}
2047 }
2048 
2049 static int ib_prctl_get(struct task_struct *task)
2050 {
2051 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2052 		return PR_SPEC_NOT_AFFECTED;
2053 
2054 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2055 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2056 		return PR_SPEC_ENABLE;
2057 	else if (is_spec_ib_user_controlled()) {
2058 		if (task_spec_ib_force_disable(task))
2059 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2060 		if (task_spec_ib_disable(task))
2061 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2062 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2063 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2064 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2065 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2066 		return PR_SPEC_DISABLE;
2067 	else
2068 		return PR_SPEC_NOT_AFFECTED;
2069 }
2070 
2071 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2072 {
2073 	switch (which) {
2074 	case PR_SPEC_STORE_BYPASS:
2075 		return ssb_prctl_get(task);
2076 	case PR_SPEC_INDIRECT_BRANCH:
2077 		return ib_prctl_get(task);
2078 	case PR_SPEC_L1D_FLUSH:
2079 		return l1d_flush_prctl_get(task);
2080 	default:
2081 		return -ENODEV;
2082 	}
2083 }
2084 
2085 void x86_spec_ctrl_setup_ap(void)
2086 {
2087 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2088 		update_spec_ctrl(x86_spec_ctrl_base);
2089 
2090 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2091 		x86_amd_ssb_disable();
2092 }
2093 
2094 bool itlb_multihit_kvm_mitigation;
2095 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2096 
2097 #undef pr_fmt
2098 #define pr_fmt(fmt)	"L1TF: " fmt
2099 
2100 /* Default mitigation for L1TF-affected CPUs */
2101 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2102 #if IS_ENABLED(CONFIG_KVM_INTEL)
2103 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2104 #endif
2105 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2106 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2107 
2108 /*
2109  * These CPUs all support 44bits physical address space internally in the
2110  * cache but CPUID can report a smaller number of physical address bits.
2111  *
2112  * The L1TF mitigation uses the top most address bit for the inversion of
2113  * non present PTEs. When the installed memory reaches into the top most
2114  * address bit due to memory holes, which has been observed on machines
2115  * which report 36bits physical address bits and have 32G RAM installed,
2116  * then the mitigation range check in l1tf_select_mitigation() triggers.
2117  * This is a false positive because the mitigation is still possible due to
2118  * the fact that the cache uses 44bit internally. Use the cache bits
2119  * instead of the reported physical bits and adjust them on the affected
2120  * machines to 44bit if the reported bits are less than 44.
2121  */
2122 static void override_cache_bits(struct cpuinfo_x86 *c)
2123 {
2124 	if (c->x86 != 6)
2125 		return;
2126 
2127 	switch (c->x86_model) {
2128 	case INTEL_FAM6_NEHALEM:
2129 	case INTEL_FAM6_WESTMERE:
2130 	case INTEL_FAM6_SANDYBRIDGE:
2131 	case INTEL_FAM6_IVYBRIDGE:
2132 	case INTEL_FAM6_HASWELL:
2133 	case INTEL_FAM6_HASWELL_L:
2134 	case INTEL_FAM6_HASWELL_G:
2135 	case INTEL_FAM6_BROADWELL:
2136 	case INTEL_FAM6_BROADWELL_G:
2137 	case INTEL_FAM6_SKYLAKE_L:
2138 	case INTEL_FAM6_SKYLAKE:
2139 	case INTEL_FAM6_KABYLAKE_L:
2140 	case INTEL_FAM6_KABYLAKE:
2141 		if (c->x86_cache_bits < 44)
2142 			c->x86_cache_bits = 44;
2143 		break;
2144 	}
2145 }
2146 
2147 static void __init l1tf_select_mitigation(void)
2148 {
2149 	u64 half_pa;
2150 
2151 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2152 		return;
2153 
2154 	if (cpu_mitigations_off())
2155 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2156 	else if (cpu_mitigations_auto_nosmt())
2157 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2158 
2159 	override_cache_bits(&boot_cpu_data);
2160 
2161 	switch (l1tf_mitigation) {
2162 	case L1TF_MITIGATION_OFF:
2163 	case L1TF_MITIGATION_FLUSH_NOWARN:
2164 	case L1TF_MITIGATION_FLUSH:
2165 		break;
2166 	case L1TF_MITIGATION_FLUSH_NOSMT:
2167 	case L1TF_MITIGATION_FULL:
2168 		cpu_smt_disable(false);
2169 		break;
2170 	case L1TF_MITIGATION_FULL_FORCE:
2171 		cpu_smt_disable(true);
2172 		break;
2173 	}
2174 
2175 #if CONFIG_PGTABLE_LEVELS == 2
2176 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2177 	return;
2178 #endif
2179 
2180 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2181 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2182 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2183 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2184 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2185 				half_pa);
2186 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2187 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2188 		return;
2189 	}
2190 
2191 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2192 }
2193 
2194 static int __init l1tf_cmdline(char *str)
2195 {
2196 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2197 		return 0;
2198 
2199 	if (!str)
2200 		return -EINVAL;
2201 
2202 	if (!strcmp(str, "off"))
2203 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2204 	else if (!strcmp(str, "flush,nowarn"))
2205 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2206 	else if (!strcmp(str, "flush"))
2207 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2208 	else if (!strcmp(str, "flush,nosmt"))
2209 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2210 	else if (!strcmp(str, "full"))
2211 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2212 	else if (!strcmp(str, "full,force"))
2213 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2214 
2215 	return 0;
2216 }
2217 early_param("l1tf", l1tf_cmdline);
2218 
2219 #undef pr_fmt
2220 #define pr_fmt(fmt) fmt
2221 
2222 #ifdef CONFIG_SYSFS
2223 
2224 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2225 
2226 #if IS_ENABLED(CONFIG_KVM_INTEL)
2227 static const char * const l1tf_vmx_states[] = {
2228 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2229 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2230 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2231 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2232 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2233 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2234 };
2235 
2236 static ssize_t l1tf_show_state(char *buf)
2237 {
2238 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2239 		return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2240 
2241 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2242 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2243 	     sched_smt_active())) {
2244 		return sysfs_emit(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2245 				  l1tf_vmx_states[l1tf_vmx_mitigation]);
2246 	}
2247 
2248 	return sysfs_emit(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2249 			  l1tf_vmx_states[l1tf_vmx_mitigation],
2250 			  sched_smt_active() ? "vulnerable" : "disabled");
2251 }
2252 
2253 static ssize_t itlb_multihit_show_state(char *buf)
2254 {
2255 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2256 	    !boot_cpu_has(X86_FEATURE_VMX))
2257 		return sysfs_emit(buf, "KVM: Mitigation: VMX unsupported\n");
2258 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2259 		return sysfs_emit(buf, "KVM: Mitigation: VMX disabled\n");
2260 	else if (itlb_multihit_kvm_mitigation)
2261 		return sysfs_emit(buf, "KVM: Mitigation: Split huge pages\n");
2262 	else
2263 		return sysfs_emit(buf, "KVM: Vulnerable\n");
2264 }
2265 #else
2266 static ssize_t l1tf_show_state(char *buf)
2267 {
2268 	return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2269 }
2270 
2271 static ssize_t itlb_multihit_show_state(char *buf)
2272 {
2273 	return sysfs_emit(buf, "Processor vulnerable\n");
2274 }
2275 #endif
2276 
2277 static ssize_t mds_show_state(char *buf)
2278 {
2279 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2280 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2281 				  mds_strings[mds_mitigation]);
2282 	}
2283 
2284 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2285 		return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2286 				  (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2287 				   sched_smt_active() ? "mitigated" : "disabled"));
2288 	}
2289 
2290 	return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2291 			  sched_smt_active() ? "vulnerable" : "disabled");
2292 }
2293 
2294 static ssize_t tsx_async_abort_show_state(char *buf)
2295 {
2296 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2297 	    (taa_mitigation == TAA_MITIGATION_OFF))
2298 		return sysfs_emit(buf, "%s\n", taa_strings[taa_mitigation]);
2299 
2300 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2301 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2302 				  taa_strings[taa_mitigation]);
2303 	}
2304 
2305 	return sysfs_emit(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2306 			  sched_smt_active() ? "vulnerable" : "disabled");
2307 }
2308 
2309 static ssize_t mmio_stale_data_show_state(char *buf)
2310 {
2311 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2312 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2313 
2314 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2315 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2316 
2317 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2318 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2319 				  mmio_strings[mmio_mitigation]);
2320 	}
2321 
2322 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2323 			  sched_smt_active() ? "vulnerable" : "disabled");
2324 }
2325 
2326 static char *stibp_state(void)
2327 {
2328 	if (spectre_v2_in_ibrs_mode(spectre_v2_enabled))
2329 		return "";
2330 
2331 	switch (spectre_v2_user_stibp) {
2332 	case SPECTRE_V2_USER_NONE:
2333 		return ", STIBP: disabled";
2334 	case SPECTRE_V2_USER_STRICT:
2335 		return ", STIBP: forced";
2336 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2337 		return ", STIBP: always-on";
2338 	case SPECTRE_V2_USER_PRCTL:
2339 	case SPECTRE_V2_USER_SECCOMP:
2340 		if (static_key_enabled(&switch_to_cond_stibp))
2341 			return ", STIBP: conditional";
2342 	}
2343 	return "";
2344 }
2345 
2346 static char *ibpb_state(void)
2347 {
2348 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2349 		if (static_key_enabled(&switch_mm_always_ibpb))
2350 			return ", IBPB: always-on";
2351 		if (static_key_enabled(&switch_mm_cond_ibpb))
2352 			return ", IBPB: conditional";
2353 		return ", IBPB: disabled";
2354 	}
2355 	return "";
2356 }
2357 
2358 static char *pbrsb_eibrs_state(void)
2359 {
2360 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2361 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2362 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2363 			return ", PBRSB-eIBRS: SW sequence";
2364 		else
2365 			return ", PBRSB-eIBRS: Vulnerable";
2366 	} else {
2367 		return ", PBRSB-eIBRS: Not affected";
2368 	}
2369 }
2370 
2371 static ssize_t spectre_v2_show_state(char *buf)
2372 {
2373 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2374 		return sysfs_emit(buf, "Vulnerable: LFENCE\n");
2375 
2376 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2377 		return sysfs_emit(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2378 
2379 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2380 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2381 		return sysfs_emit(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2382 
2383 	return sysfs_emit(buf, "%s%s%s%s%s%s%s\n",
2384 			  spectre_v2_strings[spectre_v2_enabled],
2385 			  ibpb_state(),
2386 			  boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2387 			  stibp_state(),
2388 			  boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2389 			  pbrsb_eibrs_state(),
2390 			  spectre_v2_module_string());
2391 }
2392 
2393 static ssize_t srbds_show_state(char *buf)
2394 {
2395 	return sysfs_emit(buf, "%s\n", srbds_strings[srbds_mitigation]);
2396 }
2397 
2398 static ssize_t retbleed_show_state(char *buf)
2399 {
2400 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2401 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2402 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2403 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2404 			return sysfs_emit(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2405 
2406 		return sysfs_emit(buf, "%s; SMT %s\n", retbleed_strings[retbleed_mitigation],
2407 				  !sched_smt_active() ? "disabled" :
2408 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2409 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2410 				  "enabled with STIBP protection" : "vulnerable");
2411 	}
2412 
2413 	return sysfs_emit(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2414 }
2415 
2416 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2417 			       char *buf, unsigned int bug)
2418 {
2419 	if (!boot_cpu_has_bug(bug))
2420 		return sysfs_emit(buf, "Not affected\n");
2421 
2422 	switch (bug) {
2423 	case X86_BUG_CPU_MELTDOWN:
2424 		if (boot_cpu_has(X86_FEATURE_PTI))
2425 			return sysfs_emit(buf, "Mitigation: PTI\n");
2426 
2427 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2428 			return sysfs_emit(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2429 
2430 		break;
2431 
2432 	case X86_BUG_SPECTRE_V1:
2433 		return sysfs_emit(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2434 
2435 	case X86_BUG_SPECTRE_V2:
2436 		return spectre_v2_show_state(buf);
2437 
2438 	case X86_BUG_SPEC_STORE_BYPASS:
2439 		return sysfs_emit(buf, "%s\n", ssb_strings[ssb_mode]);
2440 
2441 	case X86_BUG_L1TF:
2442 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2443 			return l1tf_show_state(buf);
2444 		break;
2445 
2446 	case X86_BUG_MDS:
2447 		return mds_show_state(buf);
2448 
2449 	case X86_BUG_TAA:
2450 		return tsx_async_abort_show_state(buf);
2451 
2452 	case X86_BUG_ITLB_MULTIHIT:
2453 		return itlb_multihit_show_state(buf);
2454 
2455 	case X86_BUG_SRBDS:
2456 		return srbds_show_state(buf);
2457 
2458 	case X86_BUG_MMIO_STALE_DATA:
2459 	case X86_BUG_MMIO_UNKNOWN:
2460 		return mmio_stale_data_show_state(buf);
2461 
2462 	case X86_BUG_RETBLEED:
2463 		return retbleed_show_state(buf);
2464 
2465 	default:
2466 		break;
2467 	}
2468 
2469 	return sysfs_emit(buf, "Vulnerable\n");
2470 }
2471 
2472 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2473 {
2474 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2475 }
2476 
2477 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2478 {
2479 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2480 }
2481 
2482 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2483 {
2484 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2485 }
2486 
2487 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2488 {
2489 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2490 }
2491 
2492 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2493 {
2494 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2495 }
2496 
2497 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2498 {
2499 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2500 }
2501 
2502 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2503 {
2504 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2505 }
2506 
2507 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2508 {
2509 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2510 }
2511 
2512 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2513 {
2514 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2515 }
2516 
2517 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2518 {
2519 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2520 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2521 	else
2522 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2523 }
2524 
2525 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2526 {
2527 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2528 }
2529 #endif
2530